UbiquitousOS
/
OpenCloudOS-Kernel
mirror of https://gitee.com/OpenCloudOS/OpenCloudOS-Kernel.git
11
0
Fork 0
Code Issues Projects Releases Wiki Activity
OpenCloudOS-Kernel/arch/x86/kernel/cpu/perf_event_p4.c

948 lines
28 KiB
C
Raw Normal View History

perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
/*
* Netburst Perfomance Events (P4, old Xeon)
*
* Copyright (C) 2010 Parallels, Inc., Cyrill Gorcunov <gorcunov@openvz.org>
* Copyright (C) 2010 Intel Corporation, Lin Ming <ming.m.lin@intel.com>
*
* For licencing details see kernel-base/COPYING
*/
#ifdef CONFIG_CPU_SUP_INTEL
#include <asm/perf_event_p4.h>
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
#define P4_CNTR_LIMIT 3
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
/*
* array indices: 0,1 - HT threads, used with HT enabled cpu
*/
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
struct p4_event_bind {
unsigned int opcode; /* Event code and ESCR selector */
unsigned int escr_msr[2]; /* ESCR MSR for this event */
x86, perf: P4 PMU - fix counters management logic Jaswinder reported this #GP: | | Message from syslogd@ht at May 14 09:39:32 ... | kernel:[ 314.908612] EIP: [<c100ccca>] | x86_perf_event_set_period+0x19d/0x1b2 SS:ESP 0068:edac3d70 | Ming has narrowed it down to a comparision issue between arguments with different sizes and signs. As result event index reached a wrong value which in turn led to a GP fault. At the same time it was found that p4_next_cntr has broken logic and should return the counter index only if it was not yet borrowed for another event. Reported-by: Jaswinder Singh Rajput <jaswinderlinux@gmail.com> Reported-by: Lin Ming <ming.m.lin@intel.com> Bisected-by: Lin Ming <ming.m.lin@intel.com> Tested-by: Jaswinder Singh Rajput <jaswinderlinux@gmail.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> CC: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100514190815.GG13509@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-15 03:08:15 +08:00
char cntr[2][P4_CNTR_LIMIT]; /* counter index (offset), -1 on abscence */
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
};
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
struct p4_pebs_bind {
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
unsigned int metric_pebs;
unsigned int metric_vert;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
};
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
/* it sets P4_PEBS_ENABLE_UOP_TAG as well */
#define P4_GEN_PEBS_BIND(name, pebs, vert) \
[P4_PEBS_METRIC__##name] = { \
.metric_pebs = pebs | P4_PEBS_ENABLE_UOP_TAG, \
.metric_vert = vert, \
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
}
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
/*
* note we have P4_PEBS_ENABLE_UOP_TAG always set here
*
* it's needed for mapping P4_PEBS_CONFIG_METRIC_MASK bits of
* event configuration to find out which values are to be
* written into MSR_IA32_PEBS_ENABLE and MSR_P4_PEBS_MATRIX_VERT
* resgisters
*/
static struct p4_pebs_bind p4_pebs_bind_map[] = {
P4_GEN_PEBS_BIND(1stl_cache_load_miss_retired, 0x0000001, 0x0000001),
P4_GEN_PEBS_BIND(2ndl_cache_load_miss_retired, 0x0000002, 0x0000001),
P4_GEN_PEBS_BIND(dtlb_load_miss_retired, 0x0000004, 0x0000001),
P4_GEN_PEBS_BIND(dtlb_store_miss_retired, 0x0000004, 0x0000002),
P4_GEN_PEBS_BIND(dtlb_all_miss_retired, 0x0000004, 0x0000003),
P4_GEN_PEBS_BIND(tagged_mispred_branch, 0x0018000, 0x0000010),
P4_GEN_PEBS_BIND(mob_load_replay_retired, 0x0000200, 0x0000001),
P4_GEN_PEBS_BIND(split_load_retired, 0x0000400, 0x0000001),
P4_GEN_PEBS_BIND(split_store_retired, 0x0000400, 0x0000002),
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
};
/*
* Note that we don't use CCCR1 here, there is an
* exception for P4_BSQ_ALLOCATION but we just have
* no workaround
*
* consider this binding as resources which particular
* event may borrow, it doesn't contain EventMask,
* Tags and friends -- they are left to a caller
*/
static struct p4_event_bind p4_event_bind_map[] = {
[P4_EVENT_TC_DELIVER_MODE] = {
.opcode = P4_OPCODE(P4_EVENT_TC_DELIVER_MODE),
.escr_msr = { MSR_P4_TC_ESCR0, MSR_P4_TC_ESCR1 },
.cntr = { {4, 5, -1}, {6, 7, -1} },
},
[P4_EVENT_BPU_FETCH_REQUEST] = {
.opcode = P4_OPCODE(P4_EVENT_BPU_FETCH_REQUEST),
.escr_msr = { MSR_P4_BPU_ESCR0, MSR_P4_BPU_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_ITLB_REFERENCE] = {
.opcode = P4_OPCODE(P4_EVENT_ITLB_REFERENCE),
.escr_msr = { MSR_P4_ITLB_ESCR0, MSR_P4_ITLB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_MEMORY_CANCEL] = {
.opcode = P4_OPCODE(P4_EVENT_MEMORY_CANCEL),
.escr_msr = { MSR_P4_DAC_ESCR0, MSR_P4_DAC_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_MEMORY_COMPLETE] = {
.opcode = P4_OPCODE(P4_EVENT_MEMORY_COMPLETE),
.escr_msr = { MSR_P4_SAAT_ESCR0 , MSR_P4_SAAT_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_LOAD_PORT_REPLAY] = {
.opcode = P4_OPCODE(P4_EVENT_LOAD_PORT_REPLAY),
.escr_msr = { MSR_P4_SAAT_ESCR0, MSR_P4_SAAT_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_STORE_PORT_REPLAY] = {
.opcode = P4_OPCODE(P4_EVENT_STORE_PORT_REPLAY),
.escr_msr = { MSR_P4_SAAT_ESCR0 , MSR_P4_SAAT_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_MOB_LOAD_REPLAY] = {
.opcode = P4_OPCODE(P4_EVENT_MOB_LOAD_REPLAY),
.escr_msr = { MSR_P4_MOB_ESCR0, MSR_P4_MOB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_PAGE_WALK_TYPE] = {
.opcode = P4_OPCODE(P4_EVENT_PAGE_WALK_TYPE),
.escr_msr = { MSR_P4_PMH_ESCR0, MSR_P4_PMH_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_BSQ_CACHE_REFERENCE] = {
.opcode = P4_OPCODE(P4_EVENT_BSQ_CACHE_REFERENCE),
.escr_msr = { MSR_P4_BSU_ESCR0, MSR_P4_BSU_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_IOQ_ALLOCATION] = {
.opcode = P4_OPCODE(P4_EVENT_IOQ_ALLOCATION),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_IOQ_ACTIVE_ENTRIES] = { /* shared ESCR */
.opcode = P4_OPCODE(P4_EVENT_IOQ_ACTIVE_ENTRIES),
.escr_msr = { MSR_P4_FSB_ESCR1, MSR_P4_FSB_ESCR1 },
.cntr = { {2, -1, -1}, {3, -1, -1} },
},
[P4_EVENT_FSB_DATA_ACTIVITY] = {
.opcode = P4_OPCODE(P4_EVENT_FSB_DATA_ACTIVITY),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_BSQ_ALLOCATION] = { /* shared ESCR, broken CCCR1 */
.opcode = P4_OPCODE(P4_EVENT_BSQ_ALLOCATION),
.escr_msr = { MSR_P4_BSU_ESCR0, MSR_P4_BSU_ESCR0 },
.cntr = { {0, -1, -1}, {1, -1, -1} },
},
[P4_EVENT_BSQ_ACTIVE_ENTRIES] = { /* shared ESCR */
.opcode = P4_OPCODE(P4_EVENT_BSQ_ACTIVE_ENTRIES),
.escr_msr = { MSR_P4_BSU_ESCR1 , MSR_P4_BSU_ESCR1 },
.cntr = { {2, -1, -1}, {3, -1, -1} },
},
[P4_EVENT_SSE_INPUT_ASSIST] = {
.opcode = P4_OPCODE(P4_EVENT_SSE_INPUT_ASSIST),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_PACKED_SP_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_PACKED_SP_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_PACKED_DP_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_PACKED_DP_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_SCALAR_SP_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_SCALAR_SP_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_SCALAR_DP_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_SCALAR_DP_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_64BIT_MMX_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_64BIT_MMX_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_128BIT_MMX_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_128BIT_MMX_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_X87_FP_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_X87_FP_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_TC_MISC] = {
.opcode = P4_OPCODE(P4_EVENT_TC_MISC),
.escr_msr = { MSR_P4_TC_ESCR0, MSR_P4_TC_ESCR1 },
.cntr = { {4, 5, -1}, {6, 7, -1} },
},
[P4_EVENT_GLOBAL_POWER_EVENTS] = {
.opcode = P4_OPCODE(P4_EVENT_GLOBAL_POWER_EVENTS),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_TC_MS_XFER] = {
.opcode = P4_OPCODE(P4_EVENT_TC_MS_XFER),
.escr_msr = { MSR_P4_MS_ESCR0, MSR_P4_MS_ESCR1 },
.cntr = { {4, 5, -1}, {6, 7, -1} },
},
[P4_EVENT_UOP_QUEUE_WRITES] = {
.opcode = P4_OPCODE(P4_EVENT_UOP_QUEUE_WRITES),
.escr_msr = { MSR_P4_MS_ESCR0, MSR_P4_MS_ESCR1 },
.cntr = { {4, 5, -1}, {6, 7, -1} },
},
[P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE] = {
.opcode = P4_OPCODE(P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE),
.escr_msr = { MSR_P4_TBPU_ESCR0 , MSR_P4_TBPU_ESCR0 },
.cntr = { {4, 5, -1}, {6, 7, -1} },
},
[P4_EVENT_RETIRED_BRANCH_TYPE] = {
.opcode = P4_OPCODE(P4_EVENT_RETIRED_BRANCH_TYPE),
.escr_msr = { MSR_P4_TBPU_ESCR0 , MSR_P4_TBPU_ESCR1 },
.cntr = { {4, 5, -1}, {6, 7, -1} },
},
[P4_EVENT_RESOURCE_STALL] = {
.opcode = P4_OPCODE(P4_EVENT_RESOURCE_STALL),
.escr_msr = { MSR_P4_ALF_ESCR0, MSR_P4_ALF_ESCR1 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_WC_BUFFER] = {
.opcode = P4_OPCODE(P4_EVENT_WC_BUFFER),
.escr_msr = { MSR_P4_DAC_ESCR0, MSR_P4_DAC_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_B2B_CYCLES] = {
.opcode = P4_OPCODE(P4_EVENT_B2B_CYCLES),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_BNR] = {
.opcode = P4_OPCODE(P4_EVENT_BNR),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_SNOOP] = {
.opcode = P4_OPCODE(P4_EVENT_SNOOP),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_RESPONSE] = {
.opcode = P4_OPCODE(P4_EVENT_RESPONSE),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_FRONT_END_EVENT] = {
.opcode = P4_OPCODE(P4_EVENT_FRONT_END_EVENT),
.escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_EXECUTION_EVENT] = {
.opcode = P4_OPCODE(P4_EVENT_EXECUTION_EVENT),
.escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_REPLAY_EVENT] = {
.opcode = P4_OPCODE(P4_EVENT_REPLAY_EVENT),
.escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_INSTR_RETIRED] = {
.opcode = P4_OPCODE(P4_EVENT_INSTR_RETIRED),
.escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_UOPS_RETIRED] = {
.opcode = P4_OPCODE(P4_EVENT_UOPS_RETIRED),
.escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_UOP_TYPE] = {
.opcode = P4_OPCODE(P4_EVENT_UOP_TYPE),
.escr_msr = { MSR_P4_RAT_ESCR0, MSR_P4_RAT_ESCR1 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_BRANCH_RETIRED] = {
.opcode = P4_OPCODE(P4_EVENT_BRANCH_RETIRED),
.escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_MISPRED_BRANCH_RETIRED] = {
.opcode = P4_OPCODE(P4_EVENT_MISPRED_BRANCH_RETIRED),
.escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_X87_ASSIST] = {
.opcode = P4_OPCODE(P4_EVENT_X87_ASSIST),
.escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_MACHINE_CLEAR] = {
.opcode = P4_OPCODE(P4_EVENT_MACHINE_CLEAR),
.escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_INSTR_COMPLETED] = {
.opcode = P4_OPCODE(P4_EVENT_INSTR_COMPLETED),
.escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
};
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
#define P4_GEN_CACHE_EVENT(event, bit, metric) \
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
p4_config_pack_escr(P4_ESCR_EVENT(event) | \
P4_ESCR_EMASK_BIT(event, bit)) | \
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
p4_config_pack_cccr(metric | \
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
P4_CCCR_ESEL(P4_OPCODE_ESEL(P4_OPCODE(event))))
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
perf, x86: Fix __initconst vs const All variables that have __initconst should also be const. Suggested-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-29 19:09:53 +08:00
static __initconst const u64 p4_hw_cache_event_ids
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX] =
{
[ C(L1D ) ] = {
[ C(OP_READ) ] = {
[ C(RESULT_ACCESS) ] = 0x0,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
[ C(RESULT_MISS) ] = P4_GEN_CACHE_EVENT(P4_EVENT_REPLAY_EVENT, NBOGUS,
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
P4_PEBS_METRIC__1stl_cache_load_miss_retired),
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
},
},
[ C(LL ) ] = {
[ C(OP_READ) ] = {
[ C(RESULT_ACCESS) ] = 0x0,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
[ C(RESULT_MISS) ] = P4_GEN_CACHE_EVENT(P4_EVENT_REPLAY_EVENT, NBOGUS,
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
P4_PEBS_METRIC__2ndl_cache_load_miss_retired),
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
},
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
},
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
[ C(DTLB) ] = {
[ C(OP_READ) ] = {
[ C(RESULT_ACCESS) ] = 0x0,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
[ C(RESULT_MISS) ] = P4_GEN_CACHE_EVENT(P4_EVENT_REPLAY_EVENT, NBOGUS,
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
P4_PEBS_METRIC__dtlb_load_miss_retired),
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
},
[ C(OP_WRITE) ] = {
[ C(RESULT_ACCESS) ] = 0x0,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
[ C(RESULT_MISS) ] = P4_GEN_CACHE_EVENT(P4_EVENT_REPLAY_EVENT, NBOGUS,
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
P4_PEBS_METRIC__dtlb_store_miss_retired),
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
},
},
[ C(ITLB) ] = {
[ C(OP_READ) ] = {
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
[ C(RESULT_ACCESS) ] = P4_GEN_CACHE_EVENT(P4_EVENT_ITLB_REFERENCE, HIT,
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
P4_PEBS_METRIC__none),
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
[ C(RESULT_MISS) ] = P4_GEN_CACHE_EVENT(P4_EVENT_ITLB_REFERENCE, MISS,
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
P4_PEBS_METRIC__none),
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
},
[ C(OP_WRITE) ] = {
[ C(RESULT_ACCESS) ] = -1,
[ C(RESULT_MISS) ] = -1,
},
[ C(OP_PREFETCH) ] = {
[ C(RESULT_ACCESS) ] = -1,
[ C(RESULT_MISS) ] = -1,
},
},
};
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
static u64 p4_general_events[PERF_COUNT_HW_MAX] = {
/* non-halted CPU clocks */
[PERF_COUNT_HW_CPU_CYCLES] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_GLOBAL_POWER_EVENTS) |
P4_ESCR_EMASK_BIT(P4_EVENT_GLOBAL_POWER_EVENTS, RUNNING)),
/*
* retired instructions
* in a sake of simplicity we don't use the FSB tagging
*/
[PERF_COUNT_HW_INSTRUCTIONS] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_INSTR_RETIRED) |
P4_ESCR_EMASK_BIT(P4_EVENT_INSTR_RETIRED, NBOGUSNTAG) |
P4_ESCR_EMASK_BIT(P4_EVENT_INSTR_RETIRED, BOGUSNTAG)),
/* cache hits */
[PERF_COUNT_HW_CACHE_REFERENCES] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_BSQ_CACHE_REFERENCE) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_HITS) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_HITE) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_HITM) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_HITS) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_HITE) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_HITM)),
/* cache misses */
[PERF_COUNT_HW_CACHE_MISSES] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_BSQ_CACHE_REFERENCE) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_MISS) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_MISS) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, WR_2ndL_MISS)),
/* branch instructions retired */
[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_RETIRED_BRANCH_TYPE) |
P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_BRANCH_TYPE, CONDITIONAL) |
P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_BRANCH_TYPE, CALL) |
P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_BRANCH_TYPE, RETURN) |
P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_BRANCH_TYPE, INDIRECT)),
/* mispredicted branches retired */
[PERF_COUNT_HW_BRANCH_MISSES] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_MISPRED_BRANCH_RETIRED) |
P4_ESCR_EMASK_BIT(P4_EVENT_MISPRED_BRANCH_RETIRED, NBOGUS)),
/* bus ready clocks (cpu is driving #DRDY_DRV\#DRDY_OWN): */
[PERF_COUNT_HW_BUS_CYCLES] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_FSB_DATA_ACTIVITY) |
P4_ESCR_EMASK_BIT(P4_EVENT_FSB_DATA_ACTIVITY, DRDY_DRV) |
P4_ESCR_EMASK_BIT(P4_EVENT_FSB_DATA_ACTIVITY, DRDY_OWN)) |
p4_config_pack_cccr(P4_CCCR_EDGE | P4_CCCR_COMPARE),
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
};
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
static struct p4_event_bind *p4_config_get_bind(u64 config)
{
unsigned int evnt = p4_config_unpack_event(config);
struct p4_event_bind *bind = NULL;
if (evnt < ARRAY_SIZE(p4_event_bind_map))
bind = &p4_event_bind_map[evnt];
return bind;
}
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
static u64 p4_pmu_event_map(int hw_event)
{
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
struct p4_event_bind *bind;
unsigned int esel;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
u64 config;
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
config = p4_general_events[hw_event];
bind = p4_config_get_bind(config);
esel = P4_OPCODE_ESEL(bind->opcode);
config |= p4_config_pack_cccr(P4_CCCR_ESEL(esel));
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
return config;
}
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
static int p4_validate_raw_event(struct perf_event *event)
{
unsigned int v;
/* user data may have out-of-bound event index */
v = p4_config_unpack_event(event->attr.config);
if (v >= ARRAY_SIZE(p4_event_bind_map)) {
pr_warning("P4 PMU: Unknown event code: %d\n", v);
return -EINVAL;
}
/*
* it may have some screwed PEBS bits
*/
if (p4_config_pebs_has(event->attr.config, P4_PEBS_CONFIG_ENABLE)) {
pr_warning("P4 PMU: PEBS are not supported yet\n");
return -EINVAL;
}
v = p4_config_unpack_metric(event->attr.config);
if (v >= ARRAY_SIZE(p4_pebs_bind_map)) {
pr_warning("P4 PMU: Unknown metric code: %d\n", v);
return -EINVAL;
}
return 0;
}
perf, x86: Fix up the ANY flag stuff Stephane noticed that the ANY flag was in generic arch code, and Cyrill reported that it broke the P4 code. Solve this by merging x86_pmu::raw_event into x86_pmu::hw_config and provide intel_pmu and amd_pmu specific versions of this callback. The intel_pmu one deals with the ANY flag, the amd_pmu adds the few extra event bits AMD64 has. Reported-by: Stephane Eranian <eranian@google.com> Reported-by: Cyrill Gorcunov <gorcunov@gmail.com> Acked-by: Robert Richter <robert.richter@amd.com> Acked-by: Cyrill Gorcunov <gorcunov@gmail.com> Acked-by: Stephane Eranian <eranian@google.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269968113.5258.442.camel@laptop> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 23:00:06 +08:00
static int p4_hw_config(struct perf_event *event)
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
{
x86, perf: P4 PMU -- protect sensible procedures from preemption Steven reported: | | I'm getting: | | Pid: 3477, comm: perf Not tainted 2.6.34-rc6 #2727 | Call Trace: | [<ffffffff811c7565>] debug_smp_processor_id+0xd5/0xf0 | [<ffffffff81019874>] p4_hw_config+0x2b/0x15c | [<ffffffff8107acbc>] ? trace_hardirqs_on_caller+0x12b/0x14f | [<ffffffff81019143>] hw_perf_event_init+0x468/0x7be | [<ffffffff810782fd>] ? debug_mutex_init+0x31/0x3c | [<ffffffff810c68b2>] T.850+0x273/0x42e | [<ffffffff810c6cab>] sys_perf_event_open+0x23e/0x3f1 | [<ffffffff81009e6a>] ? sysret_check+0x2e/0x69 | [<ffffffff81009e32>] system_call_fastpath+0x16/0x1b | | When running perf record in latest tip/perf/core | Due to the fact that p4 counters are shared between HT threads we synthetically divide the whole set of counters into two non-intersected subsets. And while we're "borrowing" counters from these subsets we should not be preempted (well, strictly speaking in p4_hw_config we just pre-set reference to the subset which allow to save some cycles in schedule routine if it happens on the same cpu). So use get_cpu/put_cpu pair. Also p4_pmu_schedule_events should use smp_processor_id rather than raw_ version. This allow us to catch up preemption issue (if there will ever be). Reported-by: Steven Rostedt <rostedt@goodmis.org> Tested-by: Steven Rostedt <rostedt@goodmis.org> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100508112716.963478928@openvz.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-08 19:25:52 +08:00
int cpu = get_cpu();
int rc = 0;
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
u32 escr, cccr;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
/*
* the reason we use cpu that early is that: if we get scheduled
* first time on the same cpu -- we will not need swap thread
* specific flags in config (and will save some cpu cycles)
*/
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
cccr = p4_default_cccr_conf(cpu);
perf, x86: Fix up the ANY flag stuff Stephane noticed that the ANY flag was in generic arch code, and Cyrill reported that it broke the P4 code. Solve this by merging x86_pmu::raw_event into x86_pmu::hw_config and provide intel_pmu and amd_pmu specific versions of this callback. The intel_pmu one deals with the ANY flag, the amd_pmu adds the few extra event bits AMD64 has. Reported-by: Stephane Eranian <eranian@google.com> Reported-by: Cyrill Gorcunov <gorcunov@gmail.com> Acked-by: Robert Richter <robert.richter@amd.com> Acked-by: Cyrill Gorcunov <gorcunov@gmail.com> Acked-by: Stephane Eranian <eranian@google.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269968113.5258.442.camel@laptop> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 23:00:06 +08:00
escr = p4_default_escr_conf(cpu, event->attr.exclude_kernel,
event->attr.exclude_user);
event->hw.config = p4_config_pack_escr(escr) |
p4_config_pack_cccr(cccr);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
if (p4_ht_active() && p4_ht_thread(cpu))
perf, x86: Fix up the ANY flag stuff Stephane noticed that the ANY flag was in generic arch code, and Cyrill reported that it broke the P4 code. Solve this by merging x86_pmu::raw_event into x86_pmu::hw_config and provide intel_pmu and amd_pmu specific versions of this callback. The intel_pmu one deals with the ANY flag, the amd_pmu adds the few extra event bits AMD64 has. Reported-by: Stephane Eranian <eranian@google.com> Reported-by: Cyrill Gorcunov <gorcunov@gmail.com> Acked-by: Robert Richter <robert.richter@amd.com> Acked-by: Cyrill Gorcunov <gorcunov@gmail.com> Acked-by: Stephane Eranian <eranian@google.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269968113.5258.442.camel@laptop> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 23:00:06 +08:00
event->hw.config = p4_set_ht_bit(event->hw.config);
x86, perf: P4 PMU -- configure predefined events If an event is not RAW we should not exit p4_hw_config early but call x86_setup_perfctr as well. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> Cc: Robert Richter <robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-08 19:39:52 +08:00
if (event->attr.type == PERF_TYPE_RAW) {
x86, perf: P4 PMU -- check for proper event index in RAW events RAW events are special and we should be ready for user passing in insane event index values. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100508112717.315897547@openvz.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-08 19:25:54 +08:00
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
rc = p4_validate_raw_event(event);
if (rc)
x86, perf: P4 PMU -- check for proper event index in RAW events RAW events are special and we should be ready for user passing in insane event index values. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100508112717.315897547@openvz.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-08 19:25:54 +08:00
goto out;
x86, perf: P4 PMU -- configure predefined events If an event is not RAW we should not exit p4_hw_config early but call x86_setup_perfctr as well. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> Cc: Robert Richter <robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-08 19:39:52 +08:00
/*
* We don't control raw events so it's up to the caller
* to pass sane values (and we don't count the thread number
* on HT machine but allow HT-compatible specifics to be
* passed on)
*
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
* Note that for RAW events we allow user to use P4_CCCR_RESERVED
* bits since we keep additional info here (for cache events and etc)
*
x86, perf: P4 PMU -- configure predefined events If an event is not RAW we should not exit p4_hw_config early but call x86_setup_perfctr as well. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> Cc: Robert Richter <robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-08 19:39:52 +08:00
* XXX: HT wide things should check perf_paranoid_cpu() &&
* CAP_SYS_ADMIN
*/
event->hw.config |= event->attr.config &
(p4_config_pack_escr(P4_ESCR_MASK_HT) |
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
p4_config_pack_cccr(P4_CCCR_MASK_HT | P4_CCCR_RESERVED));
perf, x86, Pentium4: Clear the P4_CCCR_FORCE_OVF flag If on Pentium4 CPUs the FORCE_OVF flag is set then an NMI happens on every event, which can generate a flood of NMIs. Clear it. Reported-by: Vince Weaver <vweaver1@eecs.utk.edu> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-08-26 05:06:32 +08:00
event->hw.config &= ~P4_CCCR_FORCE_OVF;
x86, perf: P4 PMU -- configure predefined events If an event is not RAW we should not exit p4_hw_config early but call x86_setup_perfctr as well. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> Cc: Robert Richter <robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-08 19:39:52 +08:00
}
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
x86, perf: P4 PMU -- protect sensible procedures from preemption Steven reported: | | I'm getting: | | Pid: 3477, comm: perf Not tainted 2.6.34-rc6 #2727 | Call Trace: | [<ffffffff811c7565>] debug_smp_processor_id+0xd5/0xf0 | [<ffffffff81019874>] p4_hw_config+0x2b/0x15c | [<ffffffff8107acbc>] ? trace_hardirqs_on_caller+0x12b/0x14f | [<ffffffff81019143>] hw_perf_event_init+0x468/0x7be | [<ffffffff810782fd>] ? debug_mutex_init+0x31/0x3c | [<ffffffff810c68b2>] T.850+0x273/0x42e | [<ffffffff810c6cab>] sys_perf_event_open+0x23e/0x3f1 | [<ffffffff81009e6a>] ? sysret_check+0x2e/0x69 | [<ffffffff81009e32>] system_call_fastpath+0x16/0x1b | | When running perf record in latest tip/perf/core | Due to the fact that p4 counters are shared between HT threads we synthetically divide the whole set of counters into two non-intersected subsets. And while we're "borrowing" counters from these subsets we should not be preempted (well, strictly speaking in p4_hw_config we just pre-set reference to the subset which allow to save some cycles in schedule routine if it happens on the same cpu). So use get_cpu/put_cpu pair. Also p4_pmu_schedule_events should use smp_processor_id rather than raw_ version. This allow us to catch up preemption issue (if there will ever be). Reported-by: Steven Rostedt <rostedt@goodmis.org> Tested-by: Steven Rostedt <rostedt@goodmis.org> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100508112716.963478928@openvz.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-08 19:25:52 +08:00
rc = x86_setup_perfctr(event);
x86, perf: P4 PMU -- check for proper event index in RAW events RAW events are special and we should be ready for user passing in insane event index values. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100508112717.315897547@openvz.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-08 19:25:54 +08:00
out:
x86, perf: P4 PMU -- protect sensible procedures from preemption Steven reported: | | I'm getting: | | Pid: 3477, comm: perf Not tainted 2.6.34-rc6 #2727 | Call Trace: | [<ffffffff811c7565>] debug_smp_processor_id+0xd5/0xf0 | [<ffffffff81019874>] p4_hw_config+0x2b/0x15c | [<ffffffff8107acbc>] ? trace_hardirqs_on_caller+0x12b/0x14f | [<ffffffff81019143>] hw_perf_event_init+0x468/0x7be | [<ffffffff810782fd>] ? debug_mutex_init+0x31/0x3c | [<ffffffff810c68b2>] T.850+0x273/0x42e | [<ffffffff810c6cab>] sys_perf_event_open+0x23e/0x3f1 | [<ffffffff81009e6a>] ? sysret_check+0x2e/0x69 | [<ffffffff81009e32>] system_call_fastpath+0x16/0x1b | | When running perf record in latest tip/perf/core | Due to the fact that p4 counters are shared between HT threads we synthetically divide the whole set of counters into two non-intersected subsets. And while we're "borrowing" counters from these subsets we should not be preempted (well, strictly speaking in p4_hw_config we just pre-set reference to the subset which allow to save some cycles in schedule routine if it happens on the same cpu). So use get_cpu/put_cpu pair. Also p4_pmu_schedule_events should use smp_processor_id rather than raw_ version. This allow us to catch up preemption issue (if there will ever be). Reported-by: Steven Rostedt <rostedt@goodmis.org> Tested-by: Steven Rostedt <rostedt@goodmis.org> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100508112716.963478928@openvz.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-08 19:25:52 +08:00
put_cpu();
return rc;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
perf, x86: P4 PMU -- handle unflagged events It might happen that an event can overflow without the proper overflow flag set. Check the sign bit in the raw counter value to solve this problem. Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: fweisbec@gmail.com Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1274083984.6540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-17 16:13:04 +08:00
static inline int p4_pmu_clear_cccr_ovf(struct hw_perf_event *hwc)
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
{
perf, x86: P4 PMU -- handle unflagged events It might happen that an event can overflow without the proper overflow flag set. Check the sign bit in the raw counter value to solve this problem. Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: fweisbec@gmail.com Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1274083984.6540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-17 16:13:04 +08:00
int overflow = 0;
u32 low, high;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
perf, x86: P4 PMU -- handle unflagged events It might happen that an event can overflow without the proper overflow flag set. Check the sign bit in the raw counter value to solve this problem. Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: fweisbec@gmail.com Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1274083984.6540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-17 16:13:04 +08:00
rdmsr(hwc->config_base + hwc->idx, low, high);
/* we need to check high bit for unflagged overflows */
perf, x86: P4 PMU -- fix typo in unflagged NMI handling Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Cyrill Gorcunov <gorcunov@gmail.com> LKML-Reference: <1274174954.22793.17.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-18 17:29:14 +08:00
if ((low & P4_CCCR_OVF) || !(high & (1 << 31))) {
perf, x86: P4 PMU -- handle unflagged events It might happen that an event can overflow without the proper overflow flag set. Check the sign bit in the raw counter value to solve this problem. Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: fweisbec@gmail.com Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1274083984.6540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-17 16:13:04 +08:00
overflow = 1;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
(void)checking_wrmsrl(hwc->config_base + hwc->idx,
perf, x86: P4 PMU -- handle unflagged events It might happen that an event can overflow without the proper overflow flag set. Check the sign bit in the raw counter value to solve this problem. Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: fweisbec@gmail.com Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1274083984.6540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-17 16:13:04 +08:00
((u64)low) & ~P4_CCCR_OVF);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
perf, x86: P4 PMU -- handle unflagged events It might happen that an event can overflow without the proper overflow flag set. Check the sign bit in the raw counter value to solve this problem. Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: fweisbec@gmail.com Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1274083984.6540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-17 16:13:04 +08:00
return overflow;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
static void p4_pmu_disable_pebs(void)
{
/*
* FIXME
*
* It's still allowed that two threads setup same cache
* events so we can't simply clear metrics until we knew
* noone is depending on us, so we need kind of counter
* for "ReplayEvent" users.
*
* What is more complex -- RAW events, if user (for some
* reason) will pass some cache event metric with improper
* event opcode -- it's fine from hardware point of view
* but completely nonsence from "meaning" of such action.
*
* So at moment let leave metrics turned on forever -- it's
* ok for now but need to be revisited!
*
* (void)checking_wrmsrl(MSR_IA32_PEBS_ENABLE, (u64)0);
* (void)checking_wrmsrl(MSR_P4_PEBS_MATRIX_VERT, (u64)0);
*/
}
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
static inline void p4_pmu_disable_event(struct perf_event *event)
{
struct hw_perf_event *hwc = &event->hw;
/*
* If event gets disabled while counter is in overflowed
* state we need to clear P4_CCCR_OVF, otherwise interrupt get
* asserted again and again
*/
(void)checking_wrmsrl(hwc->config_base + hwc->idx,
(u64)(p4_config_unpack_cccr(hwc->config)) &
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
~P4_CCCR_ENABLE & ~P4_CCCR_OVF & ~P4_CCCR_RESERVED);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
static void p4_pmu_disable_all(void)
{
struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
int idx;
perf, x86: Undo some some *_counter* -> *_event* renames The big rename: cdd6c48 perf: Do the big rename: Performance Counters -> Performance Events accidentally renamed some members of stucts that were named after registers in the spec. To avoid confusion this patch reverts some changes. The related specs are MSR descriptions in AMD's BKDGs and the ARCHITECTURAL PERFORMANCE MONITORING section in the Intel 64 and IA-32 Architectures Software Developer's Manuals. This patch does: $ sed -i -e 's:num_events:num_counters:g' \ arch/x86/include/asm/perf_event.h \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c \ arch/x86/oprofile/op_model_ppro.c $ sed -i -e 's:event_bits:cntval_bits:g' -e 's:event_mask:cntval_mask:g' \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c Signed-off-by: Robert Richter <robert.richter@amd.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269880612-25800-2-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 00:36:50 +08:00
for (idx = 0; idx < x86_pmu.num_counters; idx++) {
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
struct perf_event *event = cpuc->events[idx];
if (!test_bit(idx, cpuc->active_mask))
continue;
p4_pmu_disable_event(event);
}
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
p4_pmu_disable_pebs();
}
/* configuration must be valid */
static void p4_pmu_enable_pebs(u64 config)
{
struct p4_pebs_bind *bind;
unsigned int idx;
BUILD_BUG_ON(P4_PEBS_METRIC__max > P4_PEBS_CONFIG_METRIC_MASK);
idx = p4_config_unpack_metric(config);
if (idx == P4_PEBS_METRIC__none)
return;
bind = &p4_pebs_bind_map[idx];
(void)checking_wrmsrl(MSR_IA32_PEBS_ENABLE, (u64)bind->metric_pebs);
(void)checking_wrmsrl(MSR_P4_PEBS_MATRIX_VERT, (u64)bind->metric_vert);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
static void p4_pmu_enable_event(struct perf_event *event)
{
struct hw_perf_event *hwc = &event->hw;
int thread = p4_ht_config_thread(hwc->config);
u64 escr_conf = p4_config_unpack_escr(p4_clear_ht_bit(hwc->config));
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
unsigned int idx = p4_config_unpack_event(hwc->config);
struct p4_event_bind *bind;
u64 escr_addr, cccr;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
bind = &p4_event_bind_map[idx];
escr_addr = (u64)bind->escr_msr[thread];
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
/*
* - we dont support cascaded counters yet
* - and counter 1 is broken (erratum)
*/
WARN_ON_ONCE(p4_is_event_cascaded(hwc->config));
WARN_ON_ONCE(hwc->idx == 1);
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
/* we need a real Event value */
escr_conf &= ~P4_ESCR_EVENT_MASK;
escr_conf |= P4_ESCR_EVENT(P4_OPCODE_EVNT(bind->opcode));
cccr = p4_config_unpack_cccr(hwc->config);
/*
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
* it could be Cache event so we need to write metrics
* into additional MSRs
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
*/
perf, x86: P4 PMU -- redesign cache events To support cache events we have reserved the low 6 bits in hw_perf_event::config (which is a part of CCCR register configuration actually). These bits represent Replay Event mertic enumerated in enum P4_PEBS_METRIC. The caller should not care about which exact bits should be set and how -- the caller just chooses one P4_PEBS_METRIC entity and puts it into the config. The kernel will track it and set appropriate additional MSR registers (metrics) when needed. The reason for this redesign was the PEBS enable bit, which should not be set until DS (and PEBS sampling) support will be implemented properly. TODO ==== - PEBS sampling (note it's tricky and works with _one_ counter only so for HT machines it will be not that easy to handle both threads) - tracking of PEBS registers state, a user might need to turn PEBS off completely (ie no PEBS enable, no UOP_tag) but some other event may need it, such events clashes and should not run simultaneously, at moment we just don't support such events - eventually export user space bits in separate header which will allow user apps to configure raw events more conveniently. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <1278295769.9540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-07-05 10:09:29 +08:00
p4_pmu_enable_pebs(hwc->config);
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
(void)checking_wrmsrl(escr_addr, escr_conf);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
(void)checking_wrmsrl(hwc->config_base + hwc->idx,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
(cccr & ~P4_CCCR_RESERVED) | P4_CCCR_ENABLE);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
perf, x86: Add Nehelem PMU programming errata workaround Implement the workaround for Intel Errata AAK100 and AAP53. Also, remove the Core-i7 name for Nehalem events since there are also Westmere based i7 chips. Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> LKML-Reference: <1269608924.12097.147.camel@laptop> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-26 21:08:44 +08:00
static void p4_pmu_enable_all(int added)
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
{
struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
int idx;
perf, x86: Undo some some *_counter* -> *_event* renames The big rename: cdd6c48 perf: Do the big rename: Performance Counters -> Performance Events accidentally renamed some members of stucts that were named after registers in the spec. To avoid confusion this patch reverts some changes. The related specs are MSR descriptions in AMD's BKDGs and the ARCHITECTURAL PERFORMANCE MONITORING section in the Intel 64 and IA-32 Architectures Software Developer's Manuals. This patch does: $ sed -i -e 's:num_events:num_counters:g' \ arch/x86/include/asm/perf_event.h \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c \ arch/x86/oprofile/op_model_ppro.c $ sed -i -e 's:event_bits:cntval_bits:g' -e 's:event_mask:cntval_mask:g' \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c Signed-off-by: Robert Richter <robert.richter@amd.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269880612-25800-2-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 00:36:50 +08:00
for (idx = 0; idx < x86_pmu.num_counters; idx++) {
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
struct perf_event *event = cpuc->events[idx];
if (!test_bit(idx, cpuc->active_mask))
continue;
p4_pmu_enable_event(event);
}
}
static int p4_pmu_handle_irq(struct pt_regs *regs)
{
struct perf_sample_data data;
struct cpu_hw_events *cpuc;
struct perf_event *event;
struct hw_perf_event *hwc;
int idx, handled = 0;
u64 val;
data.addr = 0;
data.raw = NULL;
cpuc = &__get_cpu_var(cpu_hw_events);
perf, x86: Undo some some *_counter* -> *_event* renames The big rename: cdd6c48 perf: Do the big rename: Performance Counters -> Performance Events accidentally renamed some members of stucts that were named after registers in the spec. To avoid confusion this patch reverts some changes. The related specs are MSR descriptions in AMD's BKDGs and the ARCHITECTURAL PERFORMANCE MONITORING section in the Intel 64 and IA-32 Architectures Software Developer's Manuals. This patch does: $ sed -i -e 's:num_events:num_counters:g' \ arch/x86/include/asm/perf_event.h \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c \ arch/x86/oprofile/op_model_ppro.c $ sed -i -e 's:event_bits:cntval_bits:g' -e 's:event_mask:cntval_mask:g' \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c Signed-off-by: Robert Richter <robert.richter@amd.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269880612-25800-2-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 00:36:50 +08:00
for (idx = 0; idx < x86_pmu.num_counters; idx++) {
perf, x86: P4 PMU -- update nmi irq statistics and unmask lvt entry properly In case if last active performance counter is not overflowed at moment of NMI being triggered by another counter, the irq statistics may miss an update stage. As a more serious consequence -- apic quirk may not be triggered so apic lvt entry stay masked. Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100805150917.GA6311@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-08-05 23:09:17 +08:00
int overflow;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
if (!test_bit(idx, cpuc->active_mask))
continue;
event = cpuc->events[idx];
hwc = &event->hw;
WARN_ON_ONCE(hwc->idx != idx);
perf, x86: P4 PMU -- handle unflagged events It might happen that an event can overflow without the proper overflow flag set. Check the sign bit in the raw counter value to solve this problem. Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: fweisbec@gmail.com Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1274083984.6540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-17 16:13:04 +08:00
/* it might be unflagged overflow */
perf, x86: P4 PMU -- update nmi irq statistics and unmask lvt entry properly In case if last active performance counter is not overflowed at moment of NMI being triggered by another counter, the irq statistics may miss an update stage. As a more serious consequence -- apic quirk may not be triggered so apic lvt entry stay masked. Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100805150917.GA6311@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-08-05 23:09:17 +08:00
overflow = p4_pmu_clear_cccr_ovf(hwc);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
val = x86_perf_event_update(event);
perf, x86: P4 PMU -- update nmi irq statistics and unmask lvt entry properly In case if last active performance counter is not overflowed at moment of NMI being triggered by another counter, the irq statistics may miss an update stage. As a more serious consequence -- apic quirk may not be triggered so apic lvt entry stay masked. Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100805150917.GA6311@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-08-05 23:09:17 +08:00
if (!overflow && (val & (1ULL << (x86_pmu.cntval_bits - 1))))
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
continue;
perf, x86: P4 PMU -- update nmi irq statistics and unmask lvt entry properly In case if last active performance counter is not overflowed at moment of NMI being triggered by another counter, the irq statistics may miss an update stage. As a more serious consequence -- apic quirk may not be triggered so apic lvt entry stay masked. Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100805150917.GA6311@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-08-05 23:09:17 +08:00
handled += overflow;
perf, x86: P4 PMU -- handle unflagged events It might happen that an event can overflow without the proper overflow flag set. Check the sign bit in the raw counter value to solve this problem. Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: fweisbec@gmail.com Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1274083984.6540.15.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-17 16:13:04 +08:00
/* event overflow for sure */
data.period = event->hw.last_period;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
if (!x86_perf_event_set_period(event))
continue;
if (perf_event_overflow(event, 1, &data, regs))
p4_pmu_disable_event(event);
}
if (handled) {
/* p4 quirk: unmask it again */
apic_write(APIC_LVTPC, apic_read(APIC_LVTPC) & ~APIC_LVT_MASKED);
inc_irq_stat(apic_perf_irqs);
}
perf, x86: P4 PMU -- update nmi irq statistics and unmask lvt entry properly In case if last active performance counter is not overflowed at moment of NMI being triggered by another counter, the irq statistics may miss an update stage. As a more serious consequence -- apic quirk may not be triggered so apic lvt entry stay masked. Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Stephane Eranian <eranian@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100805150917.GA6311@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-08-05 23:09:17 +08:00
return handled > 0;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
/*
* swap thread specific fields according to a thread
* we are going to run on
*/
static void p4_pmu_swap_config_ts(struct hw_perf_event *hwc, int cpu)
{
u32 escr, cccr;
/*
* we either lucky and continue on same cpu or no HT support
*/
if (!p4_should_swap_ts(hwc->config, cpu))
return;
/*
* the event is migrated from an another logical
* cpu, so we need to swap thread specific flags
*/
escr = p4_config_unpack_escr(hwc->config);
cccr = p4_config_unpack_cccr(hwc->config);
if (p4_ht_thread(cpu)) {
cccr &= ~P4_CCCR_OVF_PMI_T0;
cccr |= P4_CCCR_OVF_PMI_T1;
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
if (escr & P4_ESCR_T0_OS) {
escr &= ~P4_ESCR_T0_OS;
escr |= P4_ESCR_T1_OS;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
if (escr & P4_ESCR_T0_USR) {
escr &= ~P4_ESCR_T0_USR;
escr |= P4_ESCR_T1_USR;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
hwc->config = p4_config_pack_escr(escr);
hwc->config |= p4_config_pack_cccr(cccr);
hwc->config |= P4_CONFIG_HT;
} else {
cccr &= ~P4_CCCR_OVF_PMI_T1;
cccr |= P4_CCCR_OVF_PMI_T0;
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
if (escr & P4_ESCR_T1_OS) {
escr &= ~P4_ESCR_T1_OS;
escr |= P4_ESCR_T0_OS;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
if (escr & P4_ESCR_T1_USR) {
escr &= ~P4_ESCR_T1_USR;
escr |= P4_ESCR_T0_USR;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
hwc->config = p4_config_pack_escr(escr);
hwc->config |= p4_config_pack_cccr(cccr);
hwc->config &= ~P4_CONFIG_HT;
}
}
x86, perf: P4 PMU -- use hash for p4_get_escr_idx() Linear search over all p4 MSRs should be fine if only we would not use it in events scheduling routine which is pretty time critical. Lets use hashes. It should speed scheduling up significantly. v2: Steven proposed to use more gentle approach than issue BUG on error, so we use WARN_ONCE now Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100512174242.GA5190@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-13 01:42:42 +08:00
/*
* ESCR address hashing is tricky, ESCRs are not sequential
perf, x86: P4 PMU -- do a real check for ESCR address being in hash To prevent from clashes in future code modifications do a real check for ESCR address being in hash. At moment the callers are known to pass sane values but better to be on a safe side. And comment fix. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Lin Ming <ming.m.lin@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> CC: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100518212439.004503600@openvz.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-19 05:19:17 +08:00
* in memory but all starts from MSR_P4_BSU_ESCR0 (0x03a0) and
x86, perf: P4 PMU -- use hash for p4_get_escr_idx() Linear search over all p4 MSRs should be fine if only we would not use it in events scheduling routine which is pretty time critical. Lets use hashes. It should speed scheduling up significantly. v2: Steven proposed to use more gentle approach than issue BUG on error, so we use WARN_ONCE now Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100512174242.GA5190@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-13 01:42:42 +08:00
* the metric between any ESCRs is laid in range [0xa0,0xe1]
*
* so we make ~70% filled hashtable
*/
#define P4_ESCR_MSR_BASE 0x000003a0
#define P4_ESCR_MSR_MAX 0x000003e1
#define P4_ESCR_MSR_TABLE_SIZE (P4_ESCR_MSR_MAX - P4_ESCR_MSR_BASE + 1)
#define P4_ESCR_MSR_IDX(msr) (msr - P4_ESCR_MSR_BASE)
#define P4_ESCR_MSR_TABLE_ENTRY(msr) [P4_ESCR_MSR_IDX(msr)] = msr
static const unsigned int p4_escr_table[P4_ESCR_MSR_TABLE_SIZE] = {
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_ALF_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_ALF_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_BPU_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_BPU_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_BSU_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_BSU_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_CRU_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_CRU_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_CRU_ESCR2),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_CRU_ESCR3),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_CRU_ESCR4),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_CRU_ESCR5),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_DAC_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_DAC_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_FIRM_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_FIRM_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_FLAME_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_FLAME_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_FSB_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_FSB_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_IQ_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_IQ_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_IS_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_IS_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_ITLB_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_ITLB_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_IX_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_IX_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_MOB_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_MOB_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_MS_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_MS_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_PMH_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_PMH_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_RAT_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_RAT_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_SAAT_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_SAAT_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_SSU_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_SSU_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_TBPU_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_TBPU_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_TC_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_TC_ESCR1),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_U2L_ESCR0),
P4_ESCR_MSR_TABLE_ENTRY(MSR_P4_U2L_ESCR1),
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
};
static int p4_get_escr_idx(unsigned int addr)
{
x86, perf: P4 PMU -- use hash for p4_get_escr_idx() Linear search over all p4 MSRs should be fine if only we would not use it in events scheduling routine which is pretty time critical. Lets use hashes. It should speed scheduling up significantly. v2: Steven proposed to use more gentle approach than issue BUG on error, so we use WARN_ONCE now Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100512174242.GA5190@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-13 01:42:42 +08:00
unsigned int idx = P4_ESCR_MSR_IDX(addr);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
perf, x86: P4 PMU -- do a real check for ESCR address being in hash To prevent from clashes in future code modifications do a real check for ESCR address being in hash. At moment the callers are known to pass sane values but better to be on a safe side. And comment fix. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Lin Ming <ming.m.lin@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> CC: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100518212439.004503600@openvz.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-19 05:19:17 +08:00
if (unlikely(idx >= P4_ESCR_MSR_TABLE_SIZE ||
!p4_escr_table[idx] ||
p4_escr_table[idx] != addr)) {
x86, perf: P4 PMU -- use hash for p4_get_escr_idx() Linear search over all p4 MSRs should be fine if only we would not use it in events scheduling routine which is pretty time critical. Lets use hashes. It should speed scheduling up significantly. v2: Steven proposed to use more gentle approach than issue BUG on error, so we use WARN_ONCE now Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100512174242.GA5190@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-13 01:42:42 +08:00
WARN_ONCE(1, "P4 PMU: Wrong address passed: %x\n", addr);
return -1;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
x86, perf: P4 PMU -- use hash for p4_get_escr_idx() Linear search over all p4 MSRs should be fine if only we would not use it in events scheduling routine which is pretty time critical. Lets use hashes. It should speed scheduling up significantly. v2: Steven proposed to use more gentle approach than issue BUG on error, so we use WARN_ONCE now Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100512174242.GA5190@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-13 01:42:42 +08:00
return idx;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
static int p4_next_cntr(int thread, unsigned long *used_mask,
struct p4_event_bind *bind)
{
x86, perf: P4 PMU - fix counters management logic Jaswinder reported this #GP: | | Message from syslogd@ht at May 14 09:39:32 ... | kernel:[ 314.908612] EIP: [<c100ccca>] | x86_perf_event_set_period+0x19d/0x1b2 SS:ESP 0068:edac3d70 | Ming has narrowed it down to a comparision issue between arguments with different sizes and signs. As result event index reached a wrong value which in turn led to a GP fault. At the same time it was found that p4_next_cntr has broken logic and should return the counter index only if it was not yet borrowed for another event. Reported-by: Jaswinder Singh Rajput <jaswinderlinux@gmail.com> Reported-by: Lin Ming <ming.m.lin@intel.com> Bisected-by: Lin Ming <ming.m.lin@intel.com> Tested-by: Jaswinder Singh Rajput <jaswinderlinux@gmail.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> CC: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100514190815.GG13509@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-15 03:08:15 +08:00
int i, j;
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
for (i = 0; i < P4_CNTR_LIMIT; i++) {
x86, perf: P4 PMU - fix counters management logic Jaswinder reported this #GP: | | Message from syslogd@ht at May 14 09:39:32 ... | kernel:[ 314.908612] EIP: [<c100ccca>] | x86_perf_event_set_period+0x19d/0x1b2 SS:ESP 0068:edac3d70 | Ming has narrowed it down to a comparision issue between arguments with different sizes and signs. As result event index reached a wrong value which in turn led to a GP fault. At the same time it was found that p4_next_cntr has broken logic and should return the counter index only if it was not yet borrowed for another event. Reported-by: Jaswinder Singh Rajput <jaswinderlinux@gmail.com> Reported-by: Lin Ming <ming.m.lin@intel.com> Bisected-by: Lin Ming <ming.m.lin@intel.com> Tested-by: Jaswinder Singh Rajput <jaswinderlinux@gmail.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> CC: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100514190815.GG13509@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-15 03:08:15 +08:00
j = bind->cntr[thread][i];
if (j != -1 && !test_bit(j, used_mask))
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
return j;
}
return -1;
}
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
static int p4_pmu_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign)
{
unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
x86, perf: P4 PMU -- use hash for p4_get_escr_idx() Linear search over all p4 MSRs should be fine if only we would not use it in events scheduling routine which is pretty time critical. Lets use hashes. It should speed scheduling up significantly. v2: Steven proposed to use more gentle approach than issue BUG on error, so we use WARN_ONCE now Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100512174242.GA5190@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-13 01:42:42 +08:00
unsigned long escr_mask[BITS_TO_LONGS(P4_ESCR_MSR_TABLE_SIZE)];
perf, x86: P4_pmu_schedule_events -- use smp_processor_id instead of raw_ This snippet somehow escaped the commit: | commit 137351e0feeb9f25d99488ee1afc1c79f5499a9a | Author: Cyrill Gorcunov <gorcunov@openvz.org> | Date: Sat May 8 15:25:52 2010 +0400 | | x86, perf: P4 PMU -- protect sensible procedures from preemption so bring it eventually back. It helps to catch preemption issue (if there will be, rule of thumb -- don't use raw_ if you can). Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Lin Ming <ming.m.lin@intel.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100518212439.167259349@openvz.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-19 05:19:18 +08:00
int cpu = smp_processor_id();
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
struct hw_perf_event *hwc;
struct p4_event_bind *bind;
unsigned int i, thread, num;
int cntr_idx, escr_idx;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
bitmap_zero(used_mask, X86_PMC_IDX_MAX);
x86, perf: P4 PMU -- use hash for p4_get_escr_idx() Linear search over all p4 MSRs should be fine if only we would not use it in events scheduling routine which is pretty time critical. Lets use hashes. It should speed scheduling up significantly. v2: Steven proposed to use more gentle approach than issue BUG on error, so we use WARN_ONCE now Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100512174242.GA5190@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-13 01:42:42 +08:00
bitmap_zero(escr_mask, P4_ESCR_MSR_TABLE_SIZE);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
for (i = 0, num = n; i < n; i++, num--) {
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
hwc = &cpuc->event_list[i]->hw;
thread = p4_ht_thread(cpu);
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
bind = p4_config_get_bind(hwc->config);
escr_idx = p4_get_escr_idx(bind->escr_msr[thread]);
x86, perf: P4 PMU -- use hash for p4_get_escr_idx() Linear search over all p4 MSRs should be fine if only we would not use it in events scheduling routine which is pretty time critical. Lets use hashes. It should speed scheduling up significantly. v2: Steven proposed to use more gentle approach than issue BUG on error, so we use WARN_ONCE now Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100512174242.GA5190@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-13 01:42:42 +08:00
if (unlikely(escr_idx == -1))
goto done;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
if (hwc->idx != -1 && !p4_should_swap_ts(hwc->config, cpu)) {
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
cntr_idx = hwc->idx;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
if (assign)
assign[i] = hwc->idx;
goto reserve;
}
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
cntr_idx = p4_next_cntr(thread, used_mask, bind);
if (cntr_idx == -1 || test_bit(escr_idx, escr_mask))
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
goto done;
p4_pmu_swap_config_ts(hwc, cpu);
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
if (assign)
assign[i] = cntr_idx;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
reserve:
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
set_bit(cntr_idx, used_mask);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
set_bit(escr_idx, escr_mask);
}
done:
return num ? -ENOSPC : 0;
}
perf, x86: Fix __initconst vs const All variables that have __initconst should also be const. Suggested-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-29 19:09:53 +08:00
static __initconst const struct x86_pmu p4_pmu = {
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
.name = "Netburst P4/Xeon",
.handle_irq = p4_pmu_handle_irq,
.disable_all = p4_pmu_disable_all,
.enable_all = p4_pmu_enable_all,
.enable = p4_pmu_enable_event,
.disable = p4_pmu_disable_event,
.eventsel = MSR_P4_BPU_CCCR0,
.perfctr = MSR_P4_BPU_PERFCTR0,
.event_map = p4_pmu_event_map,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
.max_events = ARRAY_SIZE(p4_general_events),
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
.get_event_constraints = x86_get_event_constraints,
/*
* IF HT disabled we may need to use all
* ARCH_P4_MAX_CCCR counters simulaneously
* though leave it restricted at moment assuming
* HT is on
*/
perf, x86: Undo some some *_counter* -> *_event* renames The big rename: cdd6c48 perf: Do the big rename: Performance Counters -> Performance Events accidentally renamed some members of stucts that were named after registers in the spec. To avoid confusion this patch reverts some changes. The related specs are MSR descriptions in AMD's BKDGs and the ARCHITECTURAL PERFORMANCE MONITORING section in the Intel 64 and IA-32 Architectures Software Developer's Manuals. This patch does: $ sed -i -e 's:num_events:num_counters:g' \ arch/x86/include/asm/perf_event.h \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c \ arch/x86/oprofile/op_model_ppro.c $ sed -i -e 's:event_bits:cntval_bits:g' -e 's:event_mask:cntval_mask:g' \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c Signed-off-by: Robert Richter <robert.richter@amd.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269880612-25800-2-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 00:36:50 +08:00
.num_counters = ARCH_P4_MAX_CCCR,
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
.apic = 1,
perf, x86: Undo some some *_counter* -> *_event* renames The big rename: cdd6c48 perf: Do the big rename: Performance Counters -> Performance Events accidentally renamed some members of stucts that were named after registers in the spec. To avoid confusion this patch reverts some changes. The related specs are MSR descriptions in AMD's BKDGs and the ARCHITECTURAL PERFORMANCE MONITORING section in the Intel 64 and IA-32 Architectures Software Developer's Manuals. This patch does: $ sed -i -e 's:num_events:num_counters:g' \ arch/x86/include/asm/perf_event.h \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c \ arch/x86/oprofile/op_model_ppro.c $ sed -i -e 's:event_bits:cntval_bits:g' -e 's:event_mask:cntval_mask:g' \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c Signed-off-by: Robert Richter <robert.richter@amd.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269880612-25800-2-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 00:36:50 +08:00
.cntval_bits = 40,
.cntval_mask = (1ULL << 40) - 1,
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
.max_period = (1ULL << 39) - 1,
.hw_config = p4_hw_config,
.schedule_events = p4_pmu_schedule_events,
perf, x86: Make a second write to performance counter if needed On Netburst PMU we need a second write to a performance counter due to cpu erratum. A simple flag test instead of alternative instructions was choosen because wrmsrl is already a macro and if virtualization is turned on will need an additional wrapper call which is more expencise. nb: we should propably switch to jump-labels as only this facility reach the mainline. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100602212304.GC5264@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-06-03 05:23:04 +08:00
/*
* This handles erratum N15 in intel doc 249199-029,
* the counter may not be updated correctly on write
* so we need a second write operation to do the trick
* (the official workaround didn't work)
*
* the former idea is taken from OProfile code
*/
.perfctr_second_write = 1,
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
};
static __init int p4_pmu_init(void)
{
unsigned int low, high;
/* If we get stripped -- indexig fails */
BUILD_BUG_ON(ARCH_P4_MAX_CCCR > X86_PMC_MAX_GENERIC);
rdmsr(MSR_IA32_MISC_ENABLE, low, high);
if (!(low & (1 << 7))) {
pr_cont("unsupported Netburst CPU model %d ",
boot_cpu_data.x86_model);
return -ENODEV;
}
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
memcpy(hw_cache_event_ids, p4_hw_cache_event_ids,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
sizeof(hw_cache_event_ids));
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
pr_cont("Netburst events, ");
x86_pmu = p4_pmu;
return 0;
}
#endif /* CONFIG_CPU_SUP_INTEL */