2019-06-03 13:44:50 +08:00
|
|
|
// SPDX-License-Identifier: GPL-2.0-only
|
2012-03-05 19:49:32 +08:00
|
|
|
/*
|
2018-10-05 20:31:10 +08:00
|
|
|
* ARMv8 PMUv3 Performance Events handling code.
|
2012-03-05 19:49:32 +08:00
|
|
|
*
|
|
|
|
* Copyright (C) 2012 ARM Limited
|
|
|
|
* Author: Will Deacon <will.deacon@arm.com>
|
|
|
|
*
|
|
|
|
* This code is based heavily on the ARMv7 perf event code.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <asm/irq_regs.h>
|
2016-03-25 00:01:16 +08:00
|
|
|
#include <asm/perf_event.h>
|
2016-01-26 01:31:13 +08:00
|
|
|
#include <asm/virt.h>
|
2012-03-05 19:49:32 +08:00
|
|
|
|
2020-07-16 13:11:27 +08:00
|
|
|
#include <clocksource/arm_arch_timer.h>
|
|
|
|
|
2016-09-15 06:32:29 +08:00
|
|
|
#include <linux/acpi.h>
|
arm64: perf: Add cap_user_time aarch64
It is useful to get the running time of a thread. Doing so in an
efficient manner can be important for performance of user applications.
Avoiding system calls in `clock_gettime` when handling
CLOCK_THREAD_CPUTIME_ID is important. Other clocks are handled in the
VDSO, but CLOCK_THREAD_CPUTIME_ID falls back on the system call.
CLOCK_THREAD_CPUTIME_ID is not handled in the VDSO since it would have
costs associated with maintaining updated user space accessible time
offsets. These offsets have to be updated everytime the a thread is
scheduled/descheduled. However, for programs regularly checking the
running time of a thread, this is a performance improvement.
This patch takes a middle ground, and adds support for cap_user_time an
optional feature of the perf_event API. This way costs are only
incurred when the perf_event api is enabled. This is done the same way
as it is in x86.
Ultimately this allows calculating the thread running time in userspace
on aarch64 as follows (adapted from perf_event_open manpage):
u32 seq, time_mult, time_shift;
u64 running, count, time_offset, quot, rem, delta;
struct perf_event_mmap_page *pc;
pc = buf; // buf is the perf event mmaped page as documented in the API.
if (pc->cap_usr_time) {
do {
seq = pc->lock;
barrier();
running = pc->time_running;
count = readCNTVCT_EL0(); // Read ARM hardware clock.
time_offset = pc->time_offset;
time_mult = pc->time_mult;
time_shift = pc->time_shift;
barrier();
} while (pc->lock != seq);
quot = (count >> time_shift);
rem = count & (((u64)1 << time_shift) - 1);
delta = time_offset + quot * time_mult +
((rem * time_mult) >> time_shift);
running += delta;
// running now has the current nanosecond level thread time.
}
Summary of changes in the patch:
For aarch64 systems, make arch_perf_update_userpage update the timing
information stored in the perf_event page. Requiring the following
calculations:
- Calculate the appropriate time_mult, and time_shift factors to convert
ticks to nano seconds for the current clock frequency.
- Adjust the mult and shift factors to avoid shift factors of 32 bits.
(possibly unnecessary)
- The time_offset userspace should apply when doing calculations:
negative the current sched time (now), because time_running and
time_enabled fields of the perf_event page have just been updated.
Toggle bits to appropriate values:
- Enable cap_user_time
Signed-off-by: Michael O'Farrell <micpof@gmail.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
2018-07-31 04:14:34 +08:00
|
|
|
#include <linux/clocksource.h>
|
2015-10-02 17:55:03 +08:00
|
|
|
#include <linux/of.h>
|
|
|
|
#include <linux/perf/arm_pmu.h>
|
2023-03-18 03:50:20 +08:00
|
|
|
#include <linux/perf/arm_pmuv3.h>
|
2015-10-02 17:55:03 +08:00
|
|
|
#include <linux/platform_device.h>
|
2020-07-16 13:11:26 +08:00
|
|
|
#include <linux/sched_clock.h>
|
2019-08-20 23:57:45 +08:00
|
|
|
#include <linux/smp.h>
|
2012-03-05 19:49:32 +08:00
|
|
|
|
2023-03-18 03:50:21 +08:00
|
|
|
#include <asm/arm_pmuv3.h>
|
|
|
|
|
2015-10-02 17:55:04 +08:00
|
|
|
/* ARMv8 Cortex-A53 specific event types. */
|
2016-04-21 20:58:41 +08:00
|
|
|
#define ARMV8_A53_PERFCTR_PREF_LINEFILL 0xC2
|
2015-10-02 17:55:04 +08:00
|
|
|
|
2016-02-19 00:50:11 +08:00
|
|
|
/* ARMv8 Cavium ThunderX specific event types. */
|
2016-04-21 20:58:41 +08:00
|
|
|
#define ARMV8_THUNDER_PERFCTR_L1D_CACHE_MISS_ST 0xE9
|
|
|
|
#define ARMV8_THUNDER_PERFCTR_L1D_CACHE_PREF_ACCESS 0xEA
|
|
|
|
#define ARMV8_THUNDER_PERFCTR_L1D_CACHE_PREF_MISS 0xEB
|
|
|
|
#define ARMV8_THUNDER_PERFCTR_L1I_CACHE_PREF_ACCESS 0xEC
|
|
|
|
#define ARMV8_THUNDER_PERFCTR_L1I_CACHE_PREF_MISS 0xED
|
2015-10-02 17:55:05 +08:00
|
|
|
|
2016-09-15 06:32:30 +08:00
|
|
|
/*
|
|
|
|
* ARMv8 Architectural defined events, not all of these may
|
2018-10-05 20:28:07 +08:00
|
|
|
* be supported on any given implementation. Unsupported events will
|
|
|
|
* be disabled at run-time based on the PMCEID registers.
|
2016-09-15 06:32:30 +08:00
|
|
|
*/
|
2012-03-05 19:49:32 +08:00
|
|
|
static const unsigned armv8_pmuv3_perf_map[PERF_COUNT_HW_MAX] = {
|
2015-07-21 18:36:39 +08:00
|
|
|
PERF_MAP_ALL_UNSUPPORTED,
|
2016-04-21 20:58:41 +08:00
|
|
|
[PERF_COUNT_HW_CPU_CYCLES] = ARMV8_PMUV3_PERFCTR_CPU_CYCLES,
|
|
|
|
[PERF_COUNT_HW_INSTRUCTIONS] = ARMV8_PMUV3_PERFCTR_INST_RETIRED,
|
|
|
|
[PERF_COUNT_HW_CACHE_REFERENCES] = ARMV8_PMUV3_PERFCTR_L1D_CACHE,
|
|
|
|
[PERF_COUNT_HW_CACHE_MISSES] = ARMV8_PMUV3_PERFCTR_L1D_CACHE_REFILL,
|
|
|
|
[PERF_COUNT_HW_BRANCH_MISSES] = ARMV8_PMUV3_PERFCTR_BR_MIS_PRED,
|
2016-09-15 06:32:30 +08:00
|
|
|
[PERF_COUNT_HW_BUS_CYCLES] = ARMV8_PMUV3_PERFCTR_BUS_CYCLES,
|
|
|
|
[PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] = ARMV8_PMUV3_PERFCTR_STALL_FRONTEND,
|
|
|
|
[PERF_COUNT_HW_STALLED_CYCLES_BACKEND] = ARMV8_PMUV3_PERFCTR_STALL_BACKEND,
|
2012-03-05 19:49:32 +08:00
|
|
|
};
|
|
|
|
|
|
|
|
static const unsigned armv8_pmuv3_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_OP_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
|
2015-07-21 18:36:39 +08:00
|
|
|
PERF_CACHE_MAP_ALL_UNSUPPORTED,
|
|
|
|
|
2016-04-21 20:58:41 +08:00
|
|
|
[C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_PMUV3_PERFCTR_L1D_CACHE,
|
|
|
|
[C(L1D)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_PMUV3_PERFCTR_L1D_CACHE_REFILL,
|
2015-07-21 18:36:39 +08:00
|
|
|
|
2016-09-15 06:32:30 +08:00
|
|
|
[C(L1I)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_PMUV3_PERFCTR_L1I_CACHE,
|
|
|
|
[C(L1I)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_PMUV3_PERFCTR_L1I_CACHE_REFILL,
|
|
|
|
|
|
|
|
[C(DTLB)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_PMUV3_PERFCTR_L1D_TLB_REFILL,
|
|
|
|
[C(DTLB)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_PMUV3_PERFCTR_L1D_TLB,
|
|
|
|
|
|
|
|
[C(ITLB)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_PMUV3_PERFCTR_L1I_TLB_REFILL,
|
|
|
|
[C(ITLB)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_PMUV3_PERFCTR_L1I_TLB,
|
|
|
|
|
2020-08-11 13:35:05 +08:00
|
|
|
[C(LL)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_PMUV3_PERFCTR_LL_CACHE_MISS_RD,
|
|
|
|
[C(LL)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_PMUV3_PERFCTR_LL_CACHE_RD,
|
|
|
|
|
2016-04-21 20:58:41 +08:00
|
|
|
[C(BPU)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_PMUV3_PERFCTR_BR_PRED,
|
|
|
|
[C(BPU)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_PMUV3_PERFCTR_BR_MIS_PRED,
|
2012-03-05 19:49:32 +08:00
|
|
|
};
|
|
|
|
|
2015-10-02 17:55:04 +08:00
|
|
|
static const unsigned armv8_a53_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_OP_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
|
|
|
|
PERF_CACHE_MAP_ALL_UNSUPPORTED,
|
|
|
|
|
2016-04-21 20:58:41 +08:00
|
|
|
[C(L1D)][C(OP_PREFETCH)][C(RESULT_MISS)] = ARMV8_A53_PERFCTR_PREF_LINEFILL,
|
2015-10-02 17:55:04 +08:00
|
|
|
|
2017-07-26 00:27:36 +08:00
|
|
|
[C(NODE)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_BUS_ACCESS_RD,
|
|
|
|
[C(NODE)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_BUS_ACCESS_WR,
|
2015-10-02 17:55:04 +08:00
|
|
|
};
|
|
|
|
|
2015-10-02 17:55:05 +08:00
|
|
|
static const unsigned armv8_a57_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_OP_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
|
|
|
|
PERF_CACHE_MAP_ALL_UNSUPPORTED,
|
|
|
|
|
2016-04-21 20:58:41 +08:00
|
|
|
[C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_RD,
|
|
|
|
[C(L1D)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_REFILL_RD,
|
|
|
|
[C(L1D)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_WR,
|
|
|
|
[C(L1D)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_REFILL_WR,
|
2015-10-02 17:55:05 +08:00
|
|
|
|
2016-04-21 20:58:41 +08:00
|
|
|
[C(DTLB)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_REFILL_RD,
|
|
|
|
[C(DTLB)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_REFILL_WR,
|
2015-10-02 17:55:05 +08:00
|
|
|
|
2017-07-26 00:27:36 +08:00
|
|
|
[C(NODE)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_BUS_ACCESS_RD,
|
|
|
|
[C(NODE)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_BUS_ACCESS_WR,
|
2015-10-02 17:55:05 +08:00
|
|
|
};
|
|
|
|
|
2017-08-10 00:46:38 +08:00
|
|
|
static const unsigned armv8_a73_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_OP_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
|
|
|
|
PERF_CACHE_MAP_ALL_UNSUPPORTED,
|
|
|
|
|
|
|
|
[C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_RD,
|
|
|
|
[C(L1D)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_WR,
|
|
|
|
};
|
|
|
|
|
2016-02-19 00:50:11 +08:00
|
|
|
static const unsigned armv8_thunder_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_OP_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
|
|
|
|
PERF_CACHE_MAP_ALL_UNSUPPORTED,
|
|
|
|
|
2016-04-21 20:58:41 +08:00
|
|
|
[C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_RD,
|
|
|
|
[C(L1D)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_REFILL_RD,
|
|
|
|
[C(L1D)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_WR,
|
|
|
|
[C(L1D)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV8_THUNDER_PERFCTR_L1D_CACHE_MISS_ST,
|
|
|
|
[C(L1D)][C(OP_PREFETCH)][C(RESULT_ACCESS)] = ARMV8_THUNDER_PERFCTR_L1D_CACHE_PREF_ACCESS,
|
|
|
|
[C(L1D)][C(OP_PREFETCH)][C(RESULT_MISS)] = ARMV8_THUNDER_PERFCTR_L1D_CACHE_PREF_MISS,
|
|
|
|
|
|
|
|
[C(L1I)][C(OP_PREFETCH)][C(RESULT_ACCESS)] = ARMV8_THUNDER_PERFCTR_L1I_CACHE_PREF_ACCESS,
|
|
|
|
[C(L1I)][C(OP_PREFETCH)][C(RESULT_MISS)] = ARMV8_THUNDER_PERFCTR_L1I_CACHE_PREF_MISS,
|
|
|
|
|
|
|
|
[C(DTLB)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_RD,
|
|
|
|
[C(DTLB)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_REFILL_RD,
|
|
|
|
[C(DTLB)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_WR,
|
|
|
|
[C(DTLB)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_REFILL_WR,
|
2015-10-02 17:55:05 +08:00
|
|
|
};
|
|
|
|
|
2016-04-21 20:58:45 +08:00
|
|
|
static const unsigned armv8_vulcan_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_OP_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
|
|
|
|
PERF_CACHE_MAP_ALL_UNSUPPORTED,
|
|
|
|
|
|
|
|
[C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_RD,
|
|
|
|
[C(L1D)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_REFILL_RD,
|
|
|
|
[C(L1D)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_WR,
|
|
|
|
[C(L1D)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_REFILL_WR,
|
|
|
|
|
|
|
|
[C(DTLB)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_RD,
|
|
|
|
[C(DTLB)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_WR,
|
|
|
|
[C(DTLB)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_REFILL_RD,
|
|
|
|
[C(DTLB)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_REFILL_WR,
|
|
|
|
|
|
|
|
[C(NODE)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_BUS_ACCESS_RD,
|
|
|
|
[C(NODE)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_BUS_ACCESS_WR,
|
|
|
|
};
|
2016-04-21 20:58:44 +08:00
|
|
|
|
|
|
|
static ssize_t
|
|
|
|
armv8pmu_events_sysfs_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *page)
|
|
|
|
{
|
|
|
|
struct perf_pmu_events_attr *pmu_attr;
|
|
|
|
|
|
|
|
pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr);
|
|
|
|
|
2020-06-18 21:35:44 +08:00
|
|
|
return sprintf(page, "event=0x%04llx\n", pmu_attr->id);
|
2016-04-21 20:58:44 +08:00
|
|
|
}
|
|
|
|
|
2019-10-30 11:46:17 +08:00
|
|
|
#define ARMV8_EVENT_ATTR(name, config) \
|
2021-06-09 14:41:03 +08:00
|
|
|
PMU_EVENT_ATTR_ID(name, armv8pmu_events_sysfs_show, config)
|
2015-10-22 22:07:32 +08:00
|
|
|
|
|
|
|
static struct attribute *armv8_pmuv3_event_attrs[] = {
|
2019-10-30 11:46:17 +08:00
|
|
|
ARMV8_EVENT_ATTR(sw_incr, ARMV8_PMUV3_PERFCTR_SW_INCR),
|
|
|
|
ARMV8_EVENT_ATTR(l1i_cache_refill, ARMV8_PMUV3_PERFCTR_L1I_CACHE_REFILL),
|
|
|
|
ARMV8_EVENT_ATTR(l1i_tlb_refill, ARMV8_PMUV3_PERFCTR_L1I_TLB_REFILL),
|
|
|
|
ARMV8_EVENT_ATTR(l1d_cache_refill, ARMV8_PMUV3_PERFCTR_L1D_CACHE_REFILL),
|
|
|
|
ARMV8_EVENT_ATTR(l1d_cache, ARMV8_PMUV3_PERFCTR_L1D_CACHE),
|
|
|
|
ARMV8_EVENT_ATTR(l1d_tlb_refill, ARMV8_PMUV3_PERFCTR_L1D_TLB_REFILL),
|
|
|
|
ARMV8_EVENT_ATTR(ld_retired, ARMV8_PMUV3_PERFCTR_LD_RETIRED),
|
|
|
|
ARMV8_EVENT_ATTR(st_retired, ARMV8_PMUV3_PERFCTR_ST_RETIRED),
|
|
|
|
ARMV8_EVENT_ATTR(inst_retired, ARMV8_PMUV3_PERFCTR_INST_RETIRED),
|
|
|
|
ARMV8_EVENT_ATTR(exc_taken, ARMV8_PMUV3_PERFCTR_EXC_TAKEN),
|
|
|
|
ARMV8_EVENT_ATTR(exc_return, ARMV8_PMUV3_PERFCTR_EXC_RETURN),
|
|
|
|
ARMV8_EVENT_ATTR(cid_write_retired, ARMV8_PMUV3_PERFCTR_CID_WRITE_RETIRED),
|
|
|
|
ARMV8_EVENT_ATTR(pc_write_retired, ARMV8_PMUV3_PERFCTR_PC_WRITE_RETIRED),
|
|
|
|
ARMV8_EVENT_ATTR(br_immed_retired, ARMV8_PMUV3_PERFCTR_BR_IMMED_RETIRED),
|
|
|
|
ARMV8_EVENT_ATTR(br_return_retired, ARMV8_PMUV3_PERFCTR_BR_RETURN_RETIRED),
|
|
|
|
ARMV8_EVENT_ATTR(unaligned_ldst_retired, ARMV8_PMUV3_PERFCTR_UNALIGNED_LDST_RETIRED),
|
|
|
|
ARMV8_EVENT_ATTR(br_mis_pred, ARMV8_PMUV3_PERFCTR_BR_MIS_PRED),
|
|
|
|
ARMV8_EVENT_ATTR(cpu_cycles, ARMV8_PMUV3_PERFCTR_CPU_CYCLES),
|
|
|
|
ARMV8_EVENT_ATTR(br_pred, ARMV8_PMUV3_PERFCTR_BR_PRED),
|
|
|
|
ARMV8_EVENT_ATTR(mem_access, ARMV8_PMUV3_PERFCTR_MEM_ACCESS),
|
|
|
|
ARMV8_EVENT_ATTR(l1i_cache, ARMV8_PMUV3_PERFCTR_L1I_CACHE),
|
|
|
|
ARMV8_EVENT_ATTR(l1d_cache_wb, ARMV8_PMUV3_PERFCTR_L1D_CACHE_WB),
|
|
|
|
ARMV8_EVENT_ATTR(l2d_cache, ARMV8_PMUV3_PERFCTR_L2D_CACHE),
|
|
|
|
ARMV8_EVENT_ATTR(l2d_cache_refill, ARMV8_PMUV3_PERFCTR_L2D_CACHE_REFILL),
|
|
|
|
ARMV8_EVENT_ATTR(l2d_cache_wb, ARMV8_PMUV3_PERFCTR_L2D_CACHE_WB),
|
|
|
|
ARMV8_EVENT_ATTR(bus_access, ARMV8_PMUV3_PERFCTR_BUS_ACCESS),
|
|
|
|
ARMV8_EVENT_ATTR(memory_error, ARMV8_PMUV3_PERFCTR_MEMORY_ERROR),
|
|
|
|
ARMV8_EVENT_ATTR(inst_spec, ARMV8_PMUV3_PERFCTR_INST_SPEC),
|
|
|
|
ARMV8_EVENT_ATTR(ttbr_write_retired, ARMV8_PMUV3_PERFCTR_TTBR_WRITE_RETIRED),
|
|
|
|
ARMV8_EVENT_ATTR(bus_cycles, ARMV8_PMUV3_PERFCTR_BUS_CYCLES),
|
|
|
|
/* Don't expose the chain event in /sys, since it's useless in isolation */
|
|
|
|
ARMV8_EVENT_ATTR(l1d_cache_allocate, ARMV8_PMUV3_PERFCTR_L1D_CACHE_ALLOCATE),
|
|
|
|
ARMV8_EVENT_ATTR(l2d_cache_allocate, ARMV8_PMUV3_PERFCTR_L2D_CACHE_ALLOCATE),
|
|
|
|
ARMV8_EVENT_ATTR(br_retired, ARMV8_PMUV3_PERFCTR_BR_RETIRED),
|
|
|
|
ARMV8_EVENT_ATTR(br_mis_pred_retired, ARMV8_PMUV3_PERFCTR_BR_MIS_PRED_RETIRED),
|
|
|
|
ARMV8_EVENT_ATTR(stall_frontend, ARMV8_PMUV3_PERFCTR_STALL_FRONTEND),
|
|
|
|
ARMV8_EVENT_ATTR(stall_backend, ARMV8_PMUV3_PERFCTR_STALL_BACKEND),
|
|
|
|
ARMV8_EVENT_ATTR(l1d_tlb, ARMV8_PMUV3_PERFCTR_L1D_TLB),
|
|
|
|
ARMV8_EVENT_ATTR(l1i_tlb, ARMV8_PMUV3_PERFCTR_L1I_TLB),
|
|
|
|
ARMV8_EVENT_ATTR(l2i_cache, ARMV8_PMUV3_PERFCTR_L2I_CACHE),
|
|
|
|
ARMV8_EVENT_ATTR(l2i_cache_refill, ARMV8_PMUV3_PERFCTR_L2I_CACHE_REFILL),
|
|
|
|
ARMV8_EVENT_ATTR(l3d_cache_allocate, ARMV8_PMUV3_PERFCTR_L3D_CACHE_ALLOCATE),
|
|
|
|
ARMV8_EVENT_ATTR(l3d_cache_refill, ARMV8_PMUV3_PERFCTR_L3D_CACHE_REFILL),
|
|
|
|
ARMV8_EVENT_ATTR(l3d_cache, ARMV8_PMUV3_PERFCTR_L3D_CACHE),
|
|
|
|
ARMV8_EVENT_ATTR(l3d_cache_wb, ARMV8_PMUV3_PERFCTR_L3D_CACHE_WB),
|
|
|
|
ARMV8_EVENT_ATTR(l2d_tlb_refill, ARMV8_PMUV3_PERFCTR_L2D_TLB_REFILL),
|
|
|
|
ARMV8_EVENT_ATTR(l2i_tlb_refill, ARMV8_PMUV3_PERFCTR_L2I_TLB_REFILL),
|
|
|
|
ARMV8_EVENT_ATTR(l2d_tlb, ARMV8_PMUV3_PERFCTR_L2D_TLB),
|
|
|
|
ARMV8_EVENT_ATTR(l2i_tlb, ARMV8_PMUV3_PERFCTR_L2I_TLB),
|
|
|
|
ARMV8_EVENT_ATTR(remote_access, ARMV8_PMUV3_PERFCTR_REMOTE_ACCESS),
|
|
|
|
ARMV8_EVENT_ATTR(ll_cache, ARMV8_PMUV3_PERFCTR_LL_CACHE),
|
|
|
|
ARMV8_EVENT_ATTR(ll_cache_miss, ARMV8_PMUV3_PERFCTR_LL_CACHE_MISS),
|
|
|
|
ARMV8_EVENT_ATTR(dtlb_walk, ARMV8_PMUV3_PERFCTR_DTLB_WALK),
|
|
|
|
ARMV8_EVENT_ATTR(itlb_walk, ARMV8_PMUV3_PERFCTR_ITLB_WALK),
|
|
|
|
ARMV8_EVENT_ATTR(ll_cache_rd, ARMV8_PMUV3_PERFCTR_LL_CACHE_RD),
|
|
|
|
ARMV8_EVENT_ATTR(ll_cache_miss_rd, ARMV8_PMUV3_PERFCTR_LL_CACHE_MISS_RD),
|
|
|
|
ARMV8_EVENT_ATTR(remote_access_rd, ARMV8_PMUV3_PERFCTR_REMOTE_ACCESS_RD),
|
2020-07-21 18:49:33 +08:00
|
|
|
ARMV8_EVENT_ATTR(l1d_cache_lmiss_rd, ARMV8_PMUV3_PERFCTR_L1D_CACHE_LMISS_RD),
|
|
|
|
ARMV8_EVENT_ATTR(op_retired, ARMV8_PMUV3_PERFCTR_OP_RETIRED),
|
|
|
|
ARMV8_EVENT_ATTR(op_spec, ARMV8_PMUV3_PERFCTR_OP_SPEC),
|
|
|
|
ARMV8_EVENT_ATTR(stall, ARMV8_PMUV3_PERFCTR_STALL),
|
|
|
|
ARMV8_EVENT_ATTR(stall_slot_backend, ARMV8_PMUV3_PERFCTR_STALL_SLOT_BACKEND),
|
|
|
|
ARMV8_EVENT_ATTR(stall_slot_frontend, ARMV8_PMUV3_PERFCTR_STALL_SLOT_FRONTEND),
|
|
|
|
ARMV8_EVENT_ATTR(stall_slot, ARMV8_PMUV3_PERFCTR_STALL_SLOT),
|
2019-10-30 11:46:17 +08:00
|
|
|
ARMV8_EVENT_ATTR(sample_pop, ARMV8_SPE_PERFCTR_SAMPLE_POP),
|
|
|
|
ARMV8_EVENT_ATTR(sample_feed, ARMV8_SPE_PERFCTR_SAMPLE_FEED),
|
|
|
|
ARMV8_EVENT_ATTR(sample_filtrate, ARMV8_SPE_PERFCTR_SAMPLE_FILTRATE),
|
|
|
|
ARMV8_EVENT_ATTR(sample_collision, ARMV8_SPE_PERFCTR_SAMPLE_COLLISION),
|
2020-07-21 18:49:33 +08:00
|
|
|
ARMV8_EVENT_ATTR(cnt_cycles, ARMV8_AMU_PERFCTR_CNT_CYCLES),
|
|
|
|
ARMV8_EVENT_ATTR(stall_backend_mem, ARMV8_AMU_PERFCTR_STALL_BACKEND_MEM),
|
|
|
|
ARMV8_EVENT_ATTR(l1i_cache_lmiss, ARMV8_PMUV3_PERFCTR_L1I_CACHE_LMISS),
|
|
|
|
ARMV8_EVENT_ATTR(l2d_cache_lmiss_rd, ARMV8_PMUV3_PERFCTR_L2D_CACHE_LMISS_RD),
|
|
|
|
ARMV8_EVENT_ATTR(l2i_cache_lmiss, ARMV8_PMUV3_PERFCTR_L2I_CACHE_LMISS),
|
|
|
|
ARMV8_EVENT_ATTR(l3d_cache_lmiss_rd, ARMV8_PMUV3_PERFCTR_L3D_CACHE_LMISS_RD),
|
2022-03-03 16:54:19 +08:00
|
|
|
ARMV8_EVENT_ATTR(trb_wrap, ARMV8_PMUV3_PERFCTR_TRB_WRAP),
|
|
|
|
ARMV8_EVENT_ATTR(trb_trig, ARMV8_PMUV3_PERFCTR_TRB_TRIG),
|
|
|
|
ARMV8_EVENT_ATTR(trcextout0, ARMV8_PMUV3_PERFCTR_TRCEXTOUT0),
|
|
|
|
ARMV8_EVENT_ATTR(trcextout1, ARMV8_PMUV3_PERFCTR_TRCEXTOUT1),
|
|
|
|
ARMV8_EVENT_ATTR(trcextout2, ARMV8_PMUV3_PERFCTR_TRCEXTOUT2),
|
|
|
|
ARMV8_EVENT_ATTR(trcextout3, ARMV8_PMUV3_PERFCTR_TRCEXTOUT3),
|
|
|
|
ARMV8_EVENT_ATTR(cti_trigout4, ARMV8_PMUV3_PERFCTR_CTI_TRIGOUT4),
|
|
|
|
ARMV8_EVENT_ATTR(cti_trigout5, ARMV8_PMUV3_PERFCTR_CTI_TRIGOUT5),
|
|
|
|
ARMV8_EVENT_ATTR(cti_trigout6, ARMV8_PMUV3_PERFCTR_CTI_TRIGOUT6),
|
|
|
|
ARMV8_EVENT_ATTR(cti_trigout7, ARMV8_PMUV3_PERFCTR_CTI_TRIGOUT7),
|
2020-07-21 18:49:33 +08:00
|
|
|
ARMV8_EVENT_ATTR(ldst_align_lat, ARMV8_PMUV3_PERFCTR_LDST_ALIGN_LAT),
|
|
|
|
ARMV8_EVENT_ATTR(ld_align_lat, ARMV8_PMUV3_PERFCTR_LD_ALIGN_LAT),
|
|
|
|
ARMV8_EVENT_ATTR(st_align_lat, ARMV8_PMUV3_PERFCTR_ST_ALIGN_LAT),
|
|
|
|
ARMV8_EVENT_ATTR(mem_access_checked, ARMV8_MTE_PERFCTR_MEM_ACCESS_CHECKED),
|
|
|
|
ARMV8_EVENT_ATTR(mem_access_checked_rd, ARMV8_MTE_PERFCTR_MEM_ACCESS_CHECKED_RD),
|
|
|
|
ARMV8_EVENT_ATTR(mem_access_checked_wr, ARMV8_MTE_PERFCTR_MEM_ACCESS_CHECKED_WR),
|
2015-12-22 22:42:57 +08:00
|
|
|
NULL,
|
2015-10-22 22:07:32 +08:00
|
|
|
};
|
|
|
|
|
2016-04-21 20:58:44 +08:00
|
|
|
static umode_t
|
|
|
|
armv8pmu_event_attr_is_visible(struct kobject *kobj,
|
|
|
|
struct attribute *attr, int unused)
|
|
|
|
{
|
|
|
|
struct device *dev = kobj_to_dev(kobj);
|
|
|
|
struct pmu *pmu = dev_get_drvdata(dev);
|
|
|
|
struct arm_pmu *cpu_pmu = container_of(pmu, struct arm_pmu, pmu);
|
|
|
|
struct perf_pmu_events_attr *pmu_attr;
|
|
|
|
|
|
|
|
pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr.attr);
|
|
|
|
|
2018-10-05 20:28:07 +08:00
|
|
|
if (pmu_attr->id < ARMV8_PMUV3_MAX_COMMON_EVENTS &&
|
|
|
|
test_bit(pmu_attr->id, cpu_pmu->pmceid_bitmap))
|
|
|
|
return attr->mode;
|
|
|
|
|
2020-06-18 21:35:44 +08:00
|
|
|
if (pmu_attr->id >= ARMV8_PMUV3_EXT_COMMON_EVENT_BASE) {
|
|
|
|
u64 id = pmu_attr->id - ARMV8_PMUV3_EXT_COMMON_EVENT_BASE;
|
|
|
|
|
|
|
|
if (id < ARMV8_PMUV3_MAX_COMMON_EVENTS &&
|
|
|
|
test_bit(id, cpu_pmu->pmceid_ext_bitmap))
|
|
|
|
return attr->mode;
|
|
|
|
}
|
2016-04-21 20:58:44 +08:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2021-01-31 22:36:15 +08:00
|
|
|
static const struct attribute_group armv8_pmuv3_events_attr_group = {
|
2015-10-22 22:07:32 +08:00
|
|
|
.name = "events",
|
|
|
|
.attrs = armv8_pmuv3_event_attrs,
|
2016-04-21 20:58:44 +08:00
|
|
|
.is_visible = armv8pmu_event_attr_is_visible,
|
2015-10-22 22:07:32 +08:00
|
|
|
};
|
|
|
|
|
2017-05-24 15:43:18 +08:00
|
|
|
PMU_FORMAT_ATTR(event, "config:0-15");
|
2018-07-10 16:58:04 +08:00
|
|
|
PMU_FORMAT_ATTR(long, "config1:0");
|
2021-12-09 04:11:23 +08:00
|
|
|
PMU_FORMAT_ATTR(rdpmc, "config1:1");
|
2018-07-10 16:58:04 +08:00
|
|
|
|
2021-12-09 04:11:22 +08:00
|
|
|
static int sysctl_perf_user_access __read_mostly;
|
|
|
|
|
2018-07-10 16:58:04 +08:00
|
|
|
static inline bool armv8pmu_event_is_64bit(struct perf_event *event)
|
|
|
|
{
|
|
|
|
return event->attr.config1 & 0x1;
|
|
|
|
}
|
2015-12-22 22:42:57 +08:00
|
|
|
|
2021-12-09 04:11:23 +08:00
|
|
|
static inline bool armv8pmu_event_want_user_access(struct perf_event *event)
|
|
|
|
{
|
|
|
|
return event->attr.config1 & 0x2;
|
|
|
|
}
|
|
|
|
|
2015-12-22 22:42:57 +08:00
|
|
|
static struct attribute *armv8_pmuv3_format_attrs[] = {
|
|
|
|
&format_attr_event.attr,
|
2018-07-10 16:58:04 +08:00
|
|
|
&format_attr_long.attr,
|
2021-12-09 04:11:23 +08:00
|
|
|
&format_attr_rdpmc.attr,
|
2015-12-22 22:42:57 +08:00
|
|
|
NULL,
|
|
|
|
};
|
|
|
|
|
2021-01-31 22:36:15 +08:00
|
|
|
static const struct attribute_group armv8_pmuv3_format_attr_group = {
|
2015-12-22 22:42:57 +08:00
|
|
|
.name = "format",
|
|
|
|
.attrs = armv8_pmuv3_format_attrs,
|
|
|
|
};
|
|
|
|
|
2020-09-22 13:53:45 +08:00
|
|
|
static ssize_t slots_show(struct device *dev, struct device_attribute *attr,
|
|
|
|
char *page)
|
|
|
|
{
|
|
|
|
struct pmu *pmu = dev_get_drvdata(dev);
|
|
|
|
struct arm_pmu *cpu_pmu = container_of(pmu, struct arm_pmu, pmu);
|
|
|
|
u32 slots = cpu_pmu->reg_pmmir & ARMV8_PMU_SLOTS_MASK;
|
|
|
|
|
2021-05-20 15:59:45 +08:00
|
|
|
return sysfs_emit(page, "0x%08x\n", slots);
|
2020-09-22 13:53:45 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static DEVICE_ATTR_RO(slots);
|
|
|
|
|
2021-06-03 15:15:02 +08:00
|
|
|
static ssize_t bus_slots_show(struct device *dev, struct device_attribute *attr,
|
|
|
|
char *page)
|
|
|
|
{
|
|
|
|
struct pmu *pmu = dev_get_drvdata(dev);
|
|
|
|
struct arm_pmu *cpu_pmu = container_of(pmu, struct arm_pmu, pmu);
|
|
|
|
u32 bus_slots = (cpu_pmu->reg_pmmir >> ARMV8_PMU_BUS_SLOTS_SHIFT)
|
|
|
|
& ARMV8_PMU_BUS_SLOTS_MASK;
|
|
|
|
|
|
|
|
return sysfs_emit(page, "0x%08x\n", bus_slots);
|
|
|
|
}
|
|
|
|
|
|
|
|
static DEVICE_ATTR_RO(bus_slots);
|
|
|
|
|
|
|
|
static ssize_t bus_width_show(struct device *dev, struct device_attribute *attr,
|
|
|
|
char *page)
|
|
|
|
{
|
|
|
|
struct pmu *pmu = dev_get_drvdata(dev);
|
|
|
|
struct arm_pmu *cpu_pmu = container_of(pmu, struct arm_pmu, pmu);
|
|
|
|
u32 bus_width = (cpu_pmu->reg_pmmir >> ARMV8_PMU_BUS_WIDTH_SHIFT)
|
|
|
|
& ARMV8_PMU_BUS_WIDTH_MASK;
|
|
|
|
u32 val = 0;
|
|
|
|
|
|
|
|
/* Encoded as Log2(number of bytes), plus one */
|
|
|
|
if (bus_width > 2 && bus_width < 13)
|
|
|
|
val = 1 << (bus_width - 1);
|
|
|
|
|
|
|
|
return sysfs_emit(page, "0x%08x\n", val);
|
|
|
|
}
|
|
|
|
|
|
|
|
static DEVICE_ATTR_RO(bus_width);
|
|
|
|
|
2020-09-22 13:53:45 +08:00
|
|
|
static struct attribute *armv8_pmuv3_caps_attrs[] = {
|
|
|
|
&dev_attr_slots.attr,
|
2021-06-03 15:15:02 +08:00
|
|
|
&dev_attr_bus_slots.attr,
|
|
|
|
&dev_attr_bus_width.attr,
|
2020-09-22 13:53:45 +08:00
|
|
|
NULL,
|
|
|
|
};
|
|
|
|
|
2021-01-31 22:36:15 +08:00
|
|
|
static const struct attribute_group armv8_pmuv3_caps_attr_group = {
|
2020-09-22 13:53:45 +08:00
|
|
|
.name = "caps",
|
|
|
|
.attrs = armv8_pmuv3_caps_attrs,
|
|
|
|
};
|
|
|
|
|
2012-03-05 19:49:32 +08:00
|
|
|
/*
|
|
|
|
* Perf Events' indices
|
|
|
|
*/
|
|
|
|
#define ARMV8_IDX_CYCLE_COUNTER 0
|
|
|
|
#define ARMV8_IDX_COUNTER0 1
|
2021-12-09 04:11:23 +08:00
|
|
|
#define ARMV8_IDX_CYCLE_COUNTER_USER 32
|
2020-03-03 02:17:52 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* We unconditionally enable ARMv8.5-PMU long event counter support
|
|
|
|
* (64-bit events) where supported. Indicate if this arm_pmu has long
|
|
|
|
* event counter support.
|
2023-03-18 03:50:26 +08:00
|
|
|
*
|
|
|
|
* On AArch32, long counters make no sense (you can't access the top
|
|
|
|
* bits), so we only enable this on AArch64.
|
2020-03-03 02:17:52 +08:00
|
|
|
*/
|
|
|
|
static bool armv8pmu_has_long_event(struct arm_pmu *cpu_pmu)
|
|
|
|
{
|
2023-03-18 03:50:26 +08:00
|
|
|
return (IS_ENABLED(CONFIG_ARM64) && is_pmuv3p5(cpu_pmu->pmuver));
|
2020-03-03 02:17:52 +08:00
|
|
|
}
|
|
|
|
|
2021-12-09 04:11:23 +08:00
|
|
|
static inline bool armv8pmu_event_has_user_read(struct perf_event *event)
|
|
|
|
{
|
|
|
|
return event->hw.flags & PERF_EVENT_FLAG_USER_READ_CNT;
|
|
|
|
}
|
|
|
|
|
2018-07-10 16:58:04 +08:00
|
|
|
/*
|
|
|
|
* We must chain two programmable counters for 64 bit events,
|
|
|
|
* except when we have allocated the 64bit cycle counter (for CPU
|
2021-12-09 04:11:23 +08:00
|
|
|
* cycles event) or when user space counter access is enabled.
|
2018-07-10 16:58:04 +08:00
|
|
|
*/
|
|
|
|
static inline bool armv8pmu_event_is_chained(struct perf_event *event)
|
|
|
|
{
|
|
|
|
int idx = event->hw.idx;
|
2020-03-03 02:17:52 +08:00
|
|
|
struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
|
2018-07-10 16:58:04 +08:00
|
|
|
|
2021-12-09 04:11:23 +08:00
|
|
|
return !armv8pmu_event_has_user_read(event) &&
|
2018-07-10 16:58:04 +08:00
|
|
|
armv8pmu_event_is_64bit(event) &&
|
2020-03-03 02:17:52 +08:00
|
|
|
!armv8pmu_has_long_event(cpu_pmu) &&
|
2018-07-10 16:58:04 +08:00
|
|
|
(idx != ARMV8_IDX_CYCLE_COUNTER);
|
|
|
|
}
|
|
|
|
|
2012-03-05 19:49:32 +08:00
|
|
|
/*
|
|
|
|
* ARMv8 low level PMU access
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Perf Event to low level counters mapping
|
|
|
|
*/
|
|
|
|
#define ARMV8_IDX_TO_COUNTER(x) \
|
2016-03-25 00:01:16 +08:00
|
|
|
(((x) - ARMV8_IDX_COUNTER0) & ARMV8_PMU_COUNTER_MASK)
|
2012-03-05 19:49:32 +08:00
|
|
|
|
|
|
|
static inline u32 armv8pmu_pmcr_read(void)
|
|
|
|
{
|
2023-03-18 03:50:21 +08:00
|
|
|
return read_pmcr();
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline void armv8pmu_pmcr_write(u32 val)
|
|
|
|
{
|
2016-03-25 00:01:16 +08:00
|
|
|
val &= ARMV8_PMU_PMCR_MASK;
|
2012-03-05 19:49:32 +08:00
|
|
|
isb();
|
2023-03-18 03:50:21 +08:00
|
|
|
write_pmcr(val);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline int armv8pmu_has_overflowed(u32 pmovsr)
|
|
|
|
{
|
2016-03-25 00:01:16 +08:00
|
|
|
return pmovsr & ARMV8_PMU_OVERFLOWED_MASK;
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline int armv8pmu_counter_has_overflowed(u32 pmnc, int idx)
|
|
|
|
{
|
2015-10-02 17:55:03 +08:00
|
|
|
return pmnc & BIT(ARMV8_IDX_TO_COUNTER(idx));
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2021-03-10 08:44:12 +08:00
|
|
|
static inline u64 armv8pmu_read_evcntr(int idx)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
2015-10-02 17:55:03 +08:00
|
|
|
u32 counter = ARMV8_IDX_TO_COUNTER(idx);
|
2012-03-05 19:49:32 +08:00
|
|
|
|
arm64: perf: Avoid PMXEV* indirection
Currently we access the counter registers and their respective type
registers indirectly. This requires us to write to PMSELR, issue an ISB,
then access the relevant PMXEV* registers.
This is unfortunate, because:
* Under virtualization, accessing one register requires two traps to
the hypervisor, even though we could access the register directly with
a single trap.
* We have to issue an ISB which we could otherwise avoid the cost of.
* When we use NMIs, the NMI handler will have to save/restore the select
register in case the code it preempted was attempting to access a
counter or its type register.
We can avoid these issues by directly accessing the relevant registers.
This patch adds helpers to do so.
In armv8pmu_enable_event() we still need the ISB to prevent the PE from
reordering the write to PMINTENSET_EL1 register. If the interrupt is
enabled before we disable the counter and the new event is configured,
we might get an interrupt triggered by the previously programmed event
overflowing, but which we wrongly attribute to the event that we are
enabling. Execute an ISB after we disable the counter.
In the process, remove the comment that refers to the ARMv7 PMU.
[Julien T.: Don't inline read/write functions to avoid big code-size
increase, remove unused read_pmevtypern function,
fix counter index issue.]
[Alexandru E.: Removed comment, removed trailing semicolons in macros,
added ISB]
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com>
Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox)
Cc: Julien Thierry <julien.thierry.kdev@gmail.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 19:07:01 +08:00
|
|
|
return read_pmevcntrn(counter);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2018-07-10 16:58:04 +08:00
|
|
|
static inline u64 armv8pmu_read_hw_counter(struct perf_event *event)
|
|
|
|
{
|
|
|
|
int idx = event->hw.idx;
|
2021-04-01 19:16:41 +08:00
|
|
|
u64 val = armv8pmu_read_evcntr(idx);
|
2018-07-10 16:58:04 +08:00
|
|
|
|
|
|
|
if (armv8pmu_event_is_chained(event))
|
|
|
|
val = (val << 32) | armv8pmu_read_evcntr(idx - 1);
|
|
|
|
return val;
|
|
|
|
}
|
|
|
|
|
2020-03-03 02:17:52 +08:00
|
|
|
/*
|
|
|
|
* The cycle counter is always a 64-bit counter. When ARMV8_PMU_PMCR_LP
|
|
|
|
* is set the event counters also become 64-bit counters. Unless the
|
|
|
|
* user has requested a long counter (attr.config1) then we want to
|
|
|
|
* interrupt upon 32-bit overflow - we achieve this by applying a bias.
|
|
|
|
*/
|
|
|
|
static bool armv8pmu_event_needs_bias(struct perf_event *event)
|
|
|
|
{
|
|
|
|
struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
|
|
|
|
struct hw_perf_event *hwc = &event->hw;
|
|
|
|
int idx = hwc->idx;
|
|
|
|
|
|
|
|
if (armv8pmu_event_is_64bit(event))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
if (armv8pmu_has_long_event(cpu_pmu) ||
|
|
|
|
idx == ARMV8_IDX_CYCLE_COUNTER)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
static u64 armv8pmu_bias_long_counter(struct perf_event *event, u64 value)
|
|
|
|
{
|
|
|
|
if (armv8pmu_event_needs_bias(event))
|
2023-03-18 03:50:24 +08:00
|
|
|
value |= GENMASK_ULL(63, 32);
|
2020-03-03 02:17:52 +08:00
|
|
|
|
|
|
|
return value;
|
|
|
|
}
|
|
|
|
|
|
|
|
static u64 armv8pmu_unbias_long_counter(struct perf_event *event, u64 value)
|
|
|
|
{
|
|
|
|
if (armv8pmu_event_needs_bias(event))
|
2023-03-18 03:50:24 +08:00
|
|
|
value &= ~GENMASK_ULL(63, 32);
|
2020-03-03 02:17:52 +08:00
|
|
|
|
|
|
|
return value;
|
|
|
|
}
|
|
|
|
|
2019-04-12 00:16:46 +08:00
|
|
|
static u64 armv8pmu_read_counter(struct perf_event *event)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
2015-10-02 17:55:03 +08:00
|
|
|
struct hw_perf_event *hwc = &event->hw;
|
|
|
|
int idx = hwc->idx;
|
2021-04-01 19:16:41 +08:00
|
|
|
u64 value;
|
2012-03-05 19:49:32 +08:00
|
|
|
|
2020-09-04 17:57:38 +08:00
|
|
|
if (idx == ARMV8_IDX_CYCLE_COUNTER)
|
2023-03-18 03:50:21 +08:00
|
|
|
value = read_pmccntr();
|
2018-07-10 16:58:02 +08:00
|
|
|
else
|
2018-07-10 16:58:04 +08:00
|
|
|
value = armv8pmu_read_hw_counter(event);
|
2012-03-05 19:49:32 +08:00
|
|
|
|
2020-03-03 02:17:52 +08:00
|
|
|
return armv8pmu_unbias_long_counter(event, value);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2020-03-03 02:17:52 +08:00
|
|
|
static inline void armv8pmu_write_evcntr(int idx, u64 value)
|
2018-07-10 16:58:02 +08:00
|
|
|
{
|
arm64: perf: Avoid PMXEV* indirection
Currently we access the counter registers and their respective type
registers indirectly. This requires us to write to PMSELR, issue an ISB,
then access the relevant PMXEV* registers.
This is unfortunate, because:
* Under virtualization, accessing one register requires two traps to
the hypervisor, even though we could access the register directly with
a single trap.
* We have to issue an ISB which we could otherwise avoid the cost of.
* When we use NMIs, the NMI handler will have to save/restore the select
register in case the code it preempted was attempting to access a
counter or its type register.
We can avoid these issues by directly accessing the relevant registers.
This patch adds helpers to do so.
In armv8pmu_enable_event() we still need the ISB to prevent the PE from
reordering the write to PMINTENSET_EL1 register. If the interrupt is
enabled before we disable the counter and the new event is configured,
we might get an interrupt triggered by the previously programmed event
overflowing, but which we wrongly attribute to the event that we are
enabling. Execute an ISB after we disable the counter.
In the process, remove the comment that refers to the ARMv7 PMU.
[Julien T.: Don't inline read/write functions to avoid big code-size
increase, remove unused read_pmevtypern function,
fix counter index issue.]
[Alexandru E.: Removed comment, removed trailing semicolons in macros,
added ISB]
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com>
Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox)
Cc: Julien Thierry <julien.thierry.kdev@gmail.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 19:07:01 +08:00
|
|
|
u32 counter = ARMV8_IDX_TO_COUNTER(idx);
|
|
|
|
|
|
|
|
write_pmevcntrn(counter, value);
|
2018-07-10 16:58:02 +08:00
|
|
|
}
|
|
|
|
|
2018-07-10 16:58:04 +08:00
|
|
|
static inline void armv8pmu_write_hw_counter(struct perf_event *event,
|
|
|
|
u64 value)
|
|
|
|
{
|
|
|
|
int idx = event->hw.idx;
|
|
|
|
|
|
|
|
if (armv8pmu_event_is_chained(event)) {
|
|
|
|
armv8pmu_write_evcntr(idx, upper_32_bits(value));
|
|
|
|
armv8pmu_write_evcntr(idx - 1, lower_32_bits(value));
|
|
|
|
} else {
|
|
|
|
armv8pmu_write_evcntr(idx, value);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-04-12 00:16:46 +08:00
|
|
|
static void armv8pmu_write_counter(struct perf_event *event, u64 value)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
2015-10-02 17:55:03 +08:00
|
|
|
struct hw_perf_event *hwc = &event->hw;
|
|
|
|
int idx = hwc->idx;
|
|
|
|
|
2020-03-03 02:17:52 +08:00
|
|
|
value = armv8pmu_bias_long_counter(event, value);
|
|
|
|
|
2020-09-04 17:57:38 +08:00
|
|
|
if (idx == ARMV8_IDX_CYCLE_COUNTER)
|
2023-03-18 03:50:21 +08:00
|
|
|
write_pmccntr(value);
|
2020-03-03 02:17:52 +08:00
|
|
|
else
|
2018-07-10 16:58:04 +08:00
|
|
|
armv8pmu_write_hw_counter(event, value);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline void armv8pmu_write_evtype(int idx, u32 val)
|
|
|
|
{
|
arm64: perf: Avoid PMXEV* indirection
Currently we access the counter registers and their respective type
registers indirectly. This requires us to write to PMSELR, issue an ISB,
then access the relevant PMXEV* registers.
This is unfortunate, because:
* Under virtualization, accessing one register requires two traps to
the hypervisor, even though we could access the register directly with
a single trap.
* We have to issue an ISB which we could otherwise avoid the cost of.
* When we use NMIs, the NMI handler will have to save/restore the select
register in case the code it preempted was attempting to access a
counter or its type register.
We can avoid these issues by directly accessing the relevant registers.
This patch adds helpers to do so.
In armv8pmu_enable_event() we still need the ISB to prevent the PE from
reordering the write to PMINTENSET_EL1 register. If the interrupt is
enabled before we disable the counter and the new event is configured,
we might get an interrupt triggered by the previously programmed event
overflowing, but which we wrongly attribute to the event that we are
enabling. Execute an ISB after we disable the counter.
In the process, remove the comment that refers to the ARMv7 PMU.
[Julien T.: Don't inline read/write functions to avoid big code-size
increase, remove unused read_pmevtypern function,
fix counter index issue.]
[Alexandru E.: Removed comment, removed trailing semicolons in macros,
added ISB]
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com>
Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox)
Cc: Julien Thierry <julien.thierry.kdev@gmail.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 19:07:01 +08:00
|
|
|
u32 counter = ARMV8_IDX_TO_COUNTER(idx);
|
|
|
|
|
2018-07-10 16:58:02 +08:00
|
|
|
val &= ARMV8_PMU_EVTYPE_MASK;
|
arm64: perf: Avoid PMXEV* indirection
Currently we access the counter registers and their respective type
registers indirectly. This requires us to write to PMSELR, issue an ISB,
then access the relevant PMXEV* registers.
This is unfortunate, because:
* Under virtualization, accessing one register requires two traps to
the hypervisor, even though we could access the register directly with
a single trap.
* We have to issue an ISB which we could otherwise avoid the cost of.
* When we use NMIs, the NMI handler will have to save/restore the select
register in case the code it preempted was attempting to access a
counter or its type register.
We can avoid these issues by directly accessing the relevant registers.
This patch adds helpers to do so.
In armv8pmu_enable_event() we still need the ISB to prevent the PE from
reordering the write to PMINTENSET_EL1 register. If the interrupt is
enabled before we disable the counter and the new event is configured,
we might get an interrupt triggered by the previously programmed event
overflowing, but which we wrongly attribute to the event that we are
enabling. Execute an ISB after we disable the counter.
In the process, remove the comment that refers to the ARMv7 PMU.
[Julien T.: Don't inline read/write functions to avoid big code-size
increase, remove unused read_pmevtypern function,
fix counter index issue.]
[Alexandru E.: Removed comment, removed trailing semicolons in macros,
added ISB]
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com>
Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox)
Cc: Julien Thierry <julien.thierry.kdev@gmail.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 19:07:01 +08:00
|
|
|
write_pmevtypern(counter, val);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2018-07-10 16:58:04 +08:00
|
|
|
static inline void armv8pmu_write_event_type(struct perf_event *event)
|
|
|
|
{
|
|
|
|
struct hw_perf_event *hwc = &event->hw;
|
|
|
|
int idx = hwc->idx;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* For chained events, the low counter is programmed to count
|
|
|
|
* the event of interest and the high counter is programmed
|
|
|
|
* with CHAIN event code with filters set to count at all ELs.
|
|
|
|
*/
|
|
|
|
if (armv8pmu_event_is_chained(event)) {
|
|
|
|
u32 chain_evt = ARMV8_PMUV3_PERFCTR_CHAIN |
|
|
|
|
ARMV8_PMU_INCLUDE_EL2;
|
|
|
|
|
|
|
|
armv8pmu_write_evtype(idx - 1, hwc->config_base);
|
|
|
|
armv8pmu_write_evtype(idx, chain_evt);
|
|
|
|
} else {
|
arm64: perf: Avoid PMXEV* indirection
Currently we access the counter registers and their respective type
registers indirectly. This requires us to write to PMSELR, issue an ISB,
then access the relevant PMXEV* registers.
This is unfortunate, because:
* Under virtualization, accessing one register requires two traps to
the hypervisor, even though we could access the register directly with
a single trap.
* We have to issue an ISB which we could otherwise avoid the cost of.
* When we use NMIs, the NMI handler will have to save/restore the select
register in case the code it preempted was attempting to access a
counter or its type register.
We can avoid these issues by directly accessing the relevant registers.
This patch adds helpers to do so.
In armv8pmu_enable_event() we still need the ISB to prevent the PE from
reordering the write to PMINTENSET_EL1 register. If the interrupt is
enabled before we disable the counter and the new event is configured,
we might get an interrupt triggered by the previously programmed event
overflowing, but which we wrongly attribute to the event that we are
enabling. Execute an ISB after we disable the counter.
In the process, remove the comment that refers to the ARMv7 PMU.
[Julien T.: Don't inline read/write functions to avoid big code-size
increase, remove unused read_pmevtypern function,
fix counter index issue.]
[Alexandru E.: Removed comment, removed trailing semicolons in macros,
added ISB]
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com>
Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox)
Cc: Julien Thierry <julien.thierry.kdev@gmail.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 19:07:01 +08:00
|
|
|
if (idx == ARMV8_IDX_CYCLE_COUNTER)
|
2023-03-18 03:50:21 +08:00
|
|
|
write_pmccfiltr(hwc->config_base);
|
arm64: perf: Avoid PMXEV* indirection
Currently we access the counter registers and their respective type
registers indirectly. This requires us to write to PMSELR, issue an ISB,
then access the relevant PMXEV* registers.
This is unfortunate, because:
* Under virtualization, accessing one register requires two traps to
the hypervisor, even though we could access the register directly with
a single trap.
* We have to issue an ISB which we could otherwise avoid the cost of.
* When we use NMIs, the NMI handler will have to save/restore the select
register in case the code it preempted was attempting to access a
counter or its type register.
We can avoid these issues by directly accessing the relevant registers.
This patch adds helpers to do so.
In armv8pmu_enable_event() we still need the ISB to prevent the PE from
reordering the write to PMINTENSET_EL1 register. If the interrupt is
enabled before we disable the counter and the new event is configured,
we might get an interrupt triggered by the previously programmed event
overflowing, but which we wrongly attribute to the event that we are
enabling. Execute an ISB after we disable the counter.
In the process, remove the comment that refers to the ARMv7 PMU.
[Julien T.: Don't inline read/write functions to avoid big code-size
increase, remove unused read_pmevtypern function,
fix counter index issue.]
[Alexandru E.: Removed comment, removed trailing semicolons in macros,
added ISB]
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com>
Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox)
Cc: Julien Thierry <julien.thierry.kdev@gmail.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 19:07:01 +08:00
|
|
|
else
|
|
|
|
armv8pmu_write_evtype(idx, hwc->config_base);
|
2018-07-10 16:58:04 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-03-18 02:22:54 +08:00
|
|
|
static u32 armv8pmu_event_cnten_mask(struct perf_event *event)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
2020-03-18 02:22:54 +08:00
|
|
|
int counter = ARMV8_IDX_TO_COUNTER(event->hw.idx);
|
|
|
|
u32 mask = BIT(counter);
|
|
|
|
|
|
|
|
if (armv8pmu_event_is_chained(event))
|
|
|
|
mask |= BIT(counter - 1);
|
|
|
|
return mask;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void armv8pmu_enable_counter(u32 mask)
|
|
|
|
{
|
2020-09-24 19:07:00 +08:00
|
|
|
/*
|
|
|
|
* Make sure event configuration register writes are visible before we
|
|
|
|
* enable the counter.
|
|
|
|
* */
|
|
|
|
isb();
|
2023-03-18 03:50:21 +08:00
|
|
|
write_pmcntenset(mask);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2018-07-10 16:58:04 +08:00
|
|
|
static inline void armv8pmu_enable_event_counter(struct perf_event *event)
|
|
|
|
{
|
2019-04-10 03:22:13 +08:00
|
|
|
struct perf_event_attr *attr = &event->attr;
|
2020-03-18 02:22:54 +08:00
|
|
|
u32 mask = armv8pmu_event_cnten_mask(event);
|
2019-04-10 03:22:13 +08:00
|
|
|
|
2020-03-18 02:22:54 +08:00
|
|
|
kvm_set_pmu_events(mask, attr);
|
2019-04-10 03:22:13 +08:00
|
|
|
|
|
|
|
/* We rely on the hypervisor switch code to enable guest counters */
|
2020-03-18 02:22:54 +08:00
|
|
|
if (!kvm_pmu_counter_deferred(attr))
|
|
|
|
armv8pmu_enable_counter(mask);
|
2018-07-10 16:58:04 +08:00
|
|
|
}
|
|
|
|
|
2020-03-18 02:22:54 +08:00
|
|
|
static inline void armv8pmu_disable_counter(u32 mask)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
2023-03-18 03:50:21 +08:00
|
|
|
write_pmcntenclr(mask);
|
arm64: perf: Avoid PMXEV* indirection
Currently we access the counter registers and their respective type
registers indirectly. This requires us to write to PMSELR, issue an ISB,
then access the relevant PMXEV* registers.
This is unfortunate, because:
* Under virtualization, accessing one register requires two traps to
the hypervisor, even though we could access the register directly with
a single trap.
* We have to issue an ISB which we could otherwise avoid the cost of.
* When we use NMIs, the NMI handler will have to save/restore the select
register in case the code it preempted was attempting to access a
counter or its type register.
We can avoid these issues by directly accessing the relevant registers.
This patch adds helpers to do so.
In armv8pmu_enable_event() we still need the ISB to prevent the PE from
reordering the write to PMINTENSET_EL1 register. If the interrupt is
enabled before we disable the counter and the new event is configured,
we might get an interrupt triggered by the previously programmed event
overflowing, but which we wrongly attribute to the event that we are
enabling. Execute an ISB after we disable the counter.
In the process, remove the comment that refers to the ARMv7 PMU.
[Julien T.: Don't inline read/write functions to avoid big code-size
increase, remove unused read_pmevtypern function,
fix counter index issue.]
[Alexandru E.: Removed comment, removed trailing semicolons in macros,
added ISB]
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com>
Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox)
Cc: Julien Thierry <julien.thierry.kdev@gmail.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 19:07:01 +08:00
|
|
|
/*
|
|
|
|
* Make sure the effects of disabling the counter are visible before we
|
|
|
|
* start configuring the event.
|
|
|
|
*/
|
|
|
|
isb();
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2018-07-10 16:58:04 +08:00
|
|
|
static inline void armv8pmu_disable_event_counter(struct perf_event *event)
|
|
|
|
{
|
2019-04-10 03:22:13 +08:00
|
|
|
struct perf_event_attr *attr = &event->attr;
|
2020-03-18 02:22:54 +08:00
|
|
|
u32 mask = armv8pmu_event_cnten_mask(event);
|
2018-07-10 16:58:04 +08:00
|
|
|
|
2020-03-18 02:22:54 +08:00
|
|
|
kvm_clr_pmu_events(mask);
|
2019-04-10 03:22:13 +08:00
|
|
|
|
|
|
|
/* We rely on the hypervisor switch code to disable guest counters */
|
2020-03-18 02:22:54 +08:00
|
|
|
if (!kvm_pmu_counter_deferred(attr))
|
|
|
|
armv8pmu_disable_counter(mask);
|
2018-07-10 16:58:04 +08:00
|
|
|
}
|
|
|
|
|
2020-03-18 02:22:54 +08:00
|
|
|
static inline void armv8pmu_enable_intens(u32 mask)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
2023-03-18 03:50:21 +08:00
|
|
|
write_pmintenset(mask);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2020-03-18 02:22:54 +08:00
|
|
|
static inline void armv8pmu_enable_event_irq(struct perf_event *event)
|
2018-07-10 16:58:04 +08:00
|
|
|
{
|
2020-03-18 02:22:54 +08:00
|
|
|
u32 counter = ARMV8_IDX_TO_COUNTER(event->hw.idx);
|
|
|
|
armv8pmu_enable_intens(BIT(counter));
|
2018-07-10 16:58:04 +08:00
|
|
|
}
|
|
|
|
|
2020-03-18 02:22:54 +08:00
|
|
|
static inline void armv8pmu_disable_intens(u32 mask)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
2023-03-18 03:50:21 +08:00
|
|
|
write_pmintenclr(mask);
|
2012-03-05 19:49:32 +08:00
|
|
|
isb();
|
|
|
|
/* Clear the overflow flag in case an interrupt is pending. */
|
2023-03-18 03:50:21 +08:00
|
|
|
write_pmovsclr(mask);
|
2012-03-05 19:49:32 +08:00
|
|
|
isb();
|
|
|
|
}
|
|
|
|
|
2020-03-18 02:22:54 +08:00
|
|
|
static inline void armv8pmu_disable_event_irq(struct perf_event *event)
|
2018-07-10 16:58:04 +08:00
|
|
|
{
|
2020-03-18 02:22:54 +08:00
|
|
|
u32 counter = ARMV8_IDX_TO_COUNTER(event->hw.idx);
|
|
|
|
armv8pmu_disable_intens(BIT(counter));
|
2018-07-10 16:58:04 +08:00
|
|
|
}
|
|
|
|
|
2012-03-05 19:49:32 +08:00
|
|
|
static inline u32 armv8pmu_getreset_flags(void)
|
|
|
|
{
|
|
|
|
u32 value;
|
|
|
|
|
|
|
|
/* Read */
|
2023-03-18 03:50:21 +08:00
|
|
|
value = read_pmovsclr();
|
2012-03-05 19:49:32 +08:00
|
|
|
|
|
|
|
/* Write to clear flags */
|
2016-03-25 00:01:16 +08:00
|
|
|
value &= ARMV8_PMU_OVSR_MASK;
|
2023-03-18 03:50:21 +08:00
|
|
|
write_pmovsclr(value);
|
2012-03-05 19:49:32 +08:00
|
|
|
|
|
|
|
return value;
|
|
|
|
}
|
|
|
|
|
2021-12-09 04:11:23 +08:00
|
|
|
static void armv8pmu_disable_user_access(void)
|
|
|
|
{
|
2023-03-18 03:50:21 +08:00
|
|
|
write_pmuserenr(0);
|
2021-12-09 04:11:23 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void armv8pmu_enable_user_access(struct arm_pmu *cpu_pmu)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
struct pmu_hw_events *cpuc = this_cpu_ptr(cpu_pmu->hw_events);
|
|
|
|
|
|
|
|
/* Clear any unused counters to avoid leaking their contents */
|
|
|
|
for_each_clear_bit(i, cpuc->used_mask, cpu_pmu->num_events) {
|
|
|
|
if (i == ARMV8_IDX_CYCLE_COUNTER)
|
2023-03-18 03:50:21 +08:00
|
|
|
write_pmccntr(0);
|
2021-12-09 04:11:23 +08:00
|
|
|
else
|
|
|
|
armv8pmu_write_evcntr(i, 0);
|
|
|
|
}
|
|
|
|
|
2023-03-18 03:50:21 +08:00
|
|
|
write_pmuserenr(0);
|
|
|
|
write_pmuserenr(ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_CR);
|
2021-12-09 04:11:23 +08:00
|
|
|
}
|
|
|
|
|
2015-10-02 17:55:03 +08:00
|
|
|
static void armv8pmu_enable_event(struct perf_event *event)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
|
|
|
/*
|
|
|
|
* Enable counter and interrupt, and set the counter to count
|
|
|
|
* the event that we're interested in.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Disable counter
|
|
|
|
*/
|
2018-07-10 16:58:04 +08:00
|
|
|
armv8pmu_disable_event_counter(event);
|
2012-03-05 19:49:32 +08:00
|
|
|
|
|
|
|
/*
|
arm64: perf: Avoid PMXEV* indirection
Currently we access the counter registers and their respective type
registers indirectly. This requires us to write to PMSELR, issue an ISB,
then access the relevant PMXEV* registers.
This is unfortunate, because:
* Under virtualization, accessing one register requires two traps to
the hypervisor, even though we could access the register directly with
a single trap.
* We have to issue an ISB which we could otherwise avoid the cost of.
* When we use NMIs, the NMI handler will have to save/restore the select
register in case the code it preempted was attempting to access a
counter or its type register.
We can avoid these issues by directly accessing the relevant registers.
This patch adds helpers to do so.
In armv8pmu_enable_event() we still need the ISB to prevent the PE from
reordering the write to PMINTENSET_EL1 register. If the interrupt is
enabled before we disable the counter and the new event is configured,
we might get an interrupt triggered by the previously programmed event
overflowing, but which we wrongly attribute to the event that we are
enabling. Execute an ISB after we disable the counter.
In the process, remove the comment that refers to the ARMv7 PMU.
[Julien T.: Don't inline read/write functions to avoid big code-size
increase, remove unused read_pmevtypern function,
fix counter index issue.]
[Alexandru E.: Removed comment, removed trailing semicolons in macros,
added ISB]
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com>
Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox)
Cc: Julien Thierry <julien.thierry.kdev@gmail.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 19:07:01 +08:00
|
|
|
* Set event.
|
2012-03-05 19:49:32 +08:00
|
|
|
*/
|
2018-07-10 16:58:04 +08:00
|
|
|
armv8pmu_write_event_type(event);
|
2012-03-05 19:49:32 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Enable interrupt for this counter
|
|
|
|
*/
|
2018-07-10 16:58:04 +08:00
|
|
|
armv8pmu_enable_event_irq(event);
|
2012-03-05 19:49:32 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Enable counter
|
|
|
|
*/
|
2018-07-10 16:58:04 +08:00
|
|
|
armv8pmu_enable_event_counter(event);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2015-10-02 17:55:03 +08:00
|
|
|
static void armv8pmu_disable_event(struct perf_event *event)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
|
|
|
/*
|
|
|
|
* Disable counter
|
|
|
|
*/
|
2018-07-10 16:58:04 +08:00
|
|
|
armv8pmu_disable_event_counter(event);
|
2012-03-05 19:49:32 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Disable interrupt for this counter
|
|
|
|
*/
|
2018-07-10 16:58:04 +08:00
|
|
|
armv8pmu_disable_event_irq(event);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2018-07-10 16:58:03 +08:00
|
|
|
static void armv8pmu_start(struct arm_pmu *cpu_pmu)
|
|
|
|
{
|
perf: Rewrite core context handling
There have been various issues and limitations with the way perf uses
(task) contexts to track events. Most notable is the single hardware
PMU task context, which has resulted in a number of yucky things (both
proposed and merged).
Notably:
- HW breakpoint PMU
- ARM big.little PMU / Intel ADL PMU
- Intel Branch Monitoring PMU
- AMD IBS PMU
- S390 cpum_cf PMU
- PowerPC trace_imc PMU
*Current design:*
Currently we have a per task and per cpu perf_event_contexts:
task_struct::perf_events_ctxp[] <-> perf_event_context <-> perf_cpu_context
^ | ^ | ^
`---------------------------------' | `--> pmu ---'
v ^
perf_event ------'
Each task has an array of pointers to a perf_event_context. Each
perf_event_context has a direct relation to a PMU and a group of
events for that PMU. The task related perf_event_context's have a
pointer back to that task.
Each PMU has a per-cpu pointer to a per-cpu perf_cpu_context, which
includes a perf_event_context, which again has a direct relation to
that PMU, and a group of events for that PMU.
The perf_cpu_context also tracks which task context is currently
associated with that CPU and includes a few other things like the
hrtimer for rotation etc.
Each perf_event is then associated with its PMU and one
perf_event_context.
*Proposed design:*
New design proposed by this patch reduce to a single task context and
a single CPU context but adds some intermediate data-structures:
task_struct::perf_event_ctxp -> perf_event_context <- perf_cpu_context
^ | ^ ^
`---------------------------' | |
| | perf_cpu_pmu_context <--.
| `----. ^ |
| | | |
| v v |
| ,--> perf_event_pmu_context |
| | |
| | |
v v |
perf_event ---> pmu ----------------'
With the new design, perf_event_context will hold all events for all
pmus in the (respective pinned/flexible) rbtrees. This can be achieved
by adding pmu to rbtree key:
{cpu, pmu, cgroup, group_index}
Each perf_event_context carries a list of perf_event_pmu_context which
is used to hold per-pmu-per-context state. For example, it keeps track
of currently active events for that pmu, a pmu specific task_ctx_data,
a flag to tell whether rotation is required or not etc.
Additionally, perf_cpu_pmu_context is used to hold per-pmu-per-cpu
state like hrtimer details to drive the event rotation, a pointer to
perf_event_pmu_context of currently running task and some other
ancillary information.
Each perf_event is associated to it's pmu, perf_event_context and
perf_event_pmu_context.
Further optimizations to current implementation are possible. For
example, ctx_resched() can be optimized to reschedule only single pmu
events.
Much thanks to Ravi for picking this up and pushing it towards
completion.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Co-developed-by: Ravi Bangoria <ravi.bangoria@amd.com>
Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20221008062424.313-1-ravi.bangoria@amd.com
2022-10-08 14:24:24 +08:00
|
|
|
struct perf_event_context *ctx;
|
|
|
|
int nr_user = 0;
|
2021-12-09 04:11:23 +08:00
|
|
|
|
perf: Rewrite core context handling
There have been various issues and limitations with the way perf uses
(task) contexts to track events. Most notable is the single hardware
PMU task context, which has resulted in a number of yucky things (both
proposed and merged).
Notably:
- HW breakpoint PMU
- ARM big.little PMU / Intel ADL PMU
- Intel Branch Monitoring PMU
- AMD IBS PMU
- S390 cpum_cf PMU
- PowerPC trace_imc PMU
*Current design:*
Currently we have a per task and per cpu perf_event_contexts:
task_struct::perf_events_ctxp[] <-> perf_event_context <-> perf_cpu_context
^ | ^ | ^
`---------------------------------' | `--> pmu ---'
v ^
perf_event ------'
Each task has an array of pointers to a perf_event_context. Each
perf_event_context has a direct relation to a PMU and a group of
events for that PMU. The task related perf_event_context's have a
pointer back to that task.
Each PMU has a per-cpu pointer to a per-cpu perf_cpu_context, which
includes a perf_event_context, which again has a direct relation to
that PMU, and a group of events for that PMU.
The perf_cpu_context also tracks which task context is currently
associated with that CPU and includes a few other things like the
hrtimer for rotation etc.
Each perf_event is then associated with its PMU and one
perf_event_context.
*Proposed design:*
New design proposed by this patch reduce to a single task context and
a single CPU context but adds some intermediate data-structures:
task_struct::perf_event_ctxp -> perf_event_context <- perf_cpu_context
^ | ^ ^
`---------------------------' | |
| | perf_cpu_pmu_context <--.
| `----. ^ |
| | | |
| v v |
| ,--> perf_event_pmu_context |
| | |
| | |
v v |
perf_event ---> pmu ----------------'
With the new design, perf_event_context will hold all events for all
pmus in the (respective pinned/flexible) rbtrees. This can be achieved
by adding pmu to rbtree key:
{cpu, pmu, cgroup, group_index}
Each perf_event_context carries a list of perf_event_pmu_context which
is used to hold per-pmu-per-context state. For example, it keeps track
of currently active events for that pmu, a pmu specific task_ctx_data,
a flag to tell whether rotation is required or not etc.
Additionally, perf_cpu_pmu_context is used to hold per-pmu-per-cpu
state like hrtimer details to drive the event rotation, a pointer to
perf_event_pmu_context of currently running task and some other
ancillary information.
Each perf_event is associated to it's pmu, perf_event_context and
perf_event_pmu_context.
Further optimizations to current implementation are possible. For
example, ctx_resched() can be optimized to reschedule only single pmu
events.
Much thanks to Ravi for picking this up and pushing it towards
completion.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Co-developed-by: Ravi Bangoria <ravi.bangoria@amd.com>
Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20221008062424.313-1-ravi.bangoria@amd.com
2022-10-08 14:24:24 +08:00
|
|
|
ctx = perf_cpu_task_ctx();
|
|
|
|
if (ctx)
|
|
|
|
nr_user = ctx->nr_user;
|
|
|
|
|
|
|
|
if (sysctl_perf_user_access && nr_user)
|
2021-12-09 04:11:23 +08:00
|
|
|
armv8pmu_enable_user_access(cpu_pmu);
|
|
|
|
else
|
|
|
|
armv8pmu_disable_user_access();
|
|
|
|
|
2018-07-10 16:58:03 +08:00
|
|
|
/* Enable all counters */
|
|
|
|
armv8pmu_pmcr_write(armv8pmu_pmcr_read() | ARMV8_PMU_PMCR_E);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void armv8pmu_stop(struct arm_pmu *cpu_pmu)
|
|
|
|
{
|
|
|
|
/* Disable all counters */
|
|
|
|
armv8pmu_pmcr_write(armv8pmu_pmcr_read() & ~ARMV8_PMU_PMCR_E);
|
|
|
|
}
|
|
|
|
|
2018-05-10 18:35:15 +08:00
|
|
|
static irqreturn_t armv8pmu_handle_irq(struct arm_pmu *cpu_pmu)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
|
|
|
u32 pmovsr;
|
|
|
|
struct perf_sample_data data;
|
2015-10-02 17:55:03 +08:00
|
|
|
struct pmu_hw_events *cpuc = this_cpu_ptr(cpu_pmu->hw_events);
|
2012-03-05 19:49:32 +08:00
|
|
|
struct pt_regs *regs;
|
|
|
|
int idx;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Get and reset the IRQ flags
|
|
|
|
*/
|
|
|
|
pmovsr = armv8pmu_getreset_flags();
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Did an overflow occur?
|
|
|
|
*/
|
|
|
|
if (!armv8pmu_has_overflowed(pmovsr))
|
|
|
|
return IRQ_NONE;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Handle the counter(s) overflow(s)
|
|
|
|
*/
|
|
|
|
regs = get_irq_regs();
|
|
|
|
|
2018-07-10 16:58:03 +08:00
|
|
|
/*
|
|
|
|
* Stop the PMU while processing the counter overflows
|
|
|
|
* to prevent skews in group events.
|
|
|
|
*/
|
|
|
|
armv8pmu_stop(cpu_pmu);
|
2012-03-05 19:49:32 +08:00
|
|
|
for (idx = 0; idx < cpu_pmu->num_events; ++idx) {
|
|
|
|
struct perf_event *event = cpuc->events[idx];
|
|
|
|
struct hw_perf_event *hwc;
|
|
|
|
|
|
|
|
/* Ignore if we don't have an event. */
|
|
|
|
if (!event)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We have a single interrupt for all counters. Check that
|
|
|
|
* each counter has overflowed before we process it.
|
|
|
|
*/
|
|
|
|
if (!armv8pmu_counter_has_overflowed(pmovsr, idx))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
hwc = &event->hw;
|
2015-10-02 17:55:03 +08:00
|
|
|
armpmu_event_update(event);
|
2012-03-05 19:49:32 +08:00
|
|
|
perf_sample_data_init(&data, 0, hwc->last_period);
|
2015-10-02 17:55:03 +08:00
|
|
|
if (!armpmu_event_set_period(event))
|
2012-03-05 19:49:32 +08:00
|
|
|
continue;
|
|
|
|
|
2020-09-24 19:07:03 +08:00
|
|
|
/*
|
|
|
|
* Perf event overflow will queue the processing of the event as
|
|
|
|
* an irq_work which will be taken care of in the handling of
|
|
|
|
* IPI_IRQ_WORK.
|
|
|
|
*/
|
2012-03-05 19:49:32 +08:00
|
|
|
if (perf_event_overflow(event, &data, regs))
|
2015-10-02 17:55:03 +08:00
|
|
|
cpu_pmu->disable(event);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
2018-07-10 16:58:03 +08:00
|
|
|
armv8pmu_start(cpu_pmu);
|
2012-03-05 19:49:32 +08:00
|
|
|
|
|
|
|
return IRQ_HANDLED;
|
|
|
|
}
|
|
|
|
|
2018-07-10 16:58:04 +08:00
|
|
|
static int armv8pmu_get_single_idx(struct pmu_hw_events *cpuc,
|
|
|
|
struct arm_pmu *cpu_pmu)
|
|
|
|
{
|
|
|
|
int idx;
|
|
|
|
|
2021-02-04 09:43:49 +08:00
|
|
|
for (idx = ARMV8_IDX_COUNTER0; idx < cpu_pmu->num_events; idx++) {
|
2018-07-10 16:58:04 +08:00
|
|
|
if (!test_and_set_bit(idx, cpuc->used_mask))
|
|
|
|
return idx;
|
|
|
|
}
|
|
|
|
return -EAGAIN;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int armv8pmu_get_chain_idx(struct pmu_hw_events *cpuc,
|
|
|
|
struct arm_pmu *cpu_pmu)
|
|
|
|
{
|
|
|
|
int idx;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Chaining requires two consecutive event counters, where
|
|
|
|
* the lower idx must be even.
|
|
|
|
*/
|
|
|
|
for (idx = ARMV8_IDX_COUNTER0 + 1; idx < cpu_pmu->num_events; idx += 2) {
|
|
|
|
if (!test_and_set_bit(idx, cpuc->used_mask)) {
|
|
|
|
/* Check if the preceding even counter is available */
|
|
|
|
if (!test_and_set_bit(idx - 1, cpuc->used_mask))
|
|
|
|
return idx;
|
|
|
|
/* Release the Odd counter */
|
|
|
|
clear_bit(idx, cpuc->used_mask);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return -EAGAIN;
|
|
|
|
}
|
|
|
|
|
2012-03-05 19:49:32 +08:00
|
|
|
static int armv8pmu_get_event_idx(struct pmu_hw_events *cpuc,
|
2015-10-02 17:55:03 +08:00
|
|
|
struct perf_event *event)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
2015-10-02 17:55:03 +08:00
|
|
|
struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
|
|
|
|
struct hw_perf_event *hwc = &event->hw;
|
2016-03-25 00:01:16 +08:00
|
|
|
unsigned long evtype = hwc->config_base & ARMV8_PMU_EVTYPE_EVENT;
|
2012-03-05 19:49:32 +08:00
|
|
|
|
2017-07-01 14:33:35 +08:00
|
|
|
/* Always prefer to place a cycle counter into the cycle counter. */
|
2016-04-21 20:58:41 +08:00
|
|
|
if (evtype == ARMV8_PMUV3_PERFCTR_CPU_CYCLES) {
|
2017-07-01 14:33:35 +08:00
|
|
|
if (!test_and_set_bit(ARMV8_IDX_CYCLE_COUNTER, cpuc->used_mask))
|
|
|
|
return ARMV8_IDX_CYCLE_COUNTER;
|
2021-12-09 04:11:23 +08:00
|
|
|
else if (armv8pmu_event_is_64bit(event) &&
|
|
|
|
armv8pmu_event_want_user_access(event) &&
|
|
|
|
!armv8pmu_has_long_event(cpu_pmu))
|
|
|
|
return -EAGAIN;
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2017-07-01 14:33:35 +08:00
|
|
|
* Otherwise use events counters
|
2012-03-05 19:49:32 +08:00
|
|
|
*/
|
2021-12-09 04:11:23 +08:00
|
|
|
if (armv8pmu_event_is_chained(event))
|
2018-07-10 16:58:04 +08:00
|
|
|
return armv8pmu_get_chain_idx(cpuc, cpu_pmu);
|
|
|
|
else
|
|
|
|
return armv8pmu_get_single_idx(cpuc, cpu_pmu);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2018-07-10 16:58:01 +08:00
|
|
|
static void armv8pmu_clear_event_idx(struct pmu_hw_events *cpuc,
|
2018-07-10 16:58:04 +08:00
|
|
|
struct perf_event *event)
|
2018-07-10 16:58:01 +08:00
|
|
|
{
|
2018-07-10 16:58:04 +08:00
|
|
|
int idx = event->hw.idx;
|
|
|
|
|
|
|
|
clear_bit(idx, cpuc->used_mask);
|
|
|
|
if (armv8pmu_event_is_chained(event))
|
|
|
|
clear_bit(idx - 1, cpuc->used_mask);
|
2018-07-10 16:58:01 +08:00
|
|
|
}
|
|
|
|
|
2021-12-09 04:11:23 +08:00
|
|
|
static int armv8pmu_user_event_idx(struct perf_event *event)
|
|
|
|
{
|
|
|
|
if (!sysctl_perf_user_access || !armv8pmu_event_has_user_read(event))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We remap the cycle counter index to 32 to
|
|
|
|
* match the offset applied to the rest of
|
|
|
|
* the counter indices.
|
|
|
|
*/
|
|
|
|
if (event->hw.idx == ARMV8_IDX_CYCLE_COUNTER)
|
|
|
|
return ARMV8_IDX_CYCLE_COUNTER_USER;
|
|
|
|
|
|
|
|
return event->hw.idx;
|
|
|
|
}
|
|
|
|
|
2012-03-05 19:49:32 +08:00
|
|
|
/*
|
2019-01-18 22:02:27 +08:00
|
|
|
* Add an event filter to a given event.
|
2012-03-05 19:49:32 +08:00
|
|
|
*/
|
|
|
|
static int armv8pmu_set_event_filter(struct hw_perf_event *event,
|
|
|
|
struct perf_event_attr *attr)
|
|
|
|
{
|
|
|
|
unsigned long config_base = 0;
|
|
|
|
|
|
|
|
if (attr->exclude_idle)
|
|
|
|
return -EPERM;
|
2017-05-03 00:29:34 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* If we're running in hyp mode, then we *are* the hypervisor.
|
|
|
|
* Therefore we ignore exclude_hv in this configuration, since
|
|
|
|
* there's no hypervisor to sample anyway. This is consistent
|
|
|
|
* with other architectures (x86 and Power).
|
|
|
|
*/
|
|
|
|
if (is_kernel_in_hyp_mode()) {
|
2019-04-10 03:22:15 +08:00
|
|
|
if (!attr->exclude_kernel && !attr->exclude_host)
|
2017-05-03 00:29:34 +08:00
|
|
|
config_base |= ARMV8_PMU_INCLUDE_EL2;
|
2019-04-10 03:22:15 +08:00
|
|
|
if (attr->exclude_guest)
|
2017-05-03 00:29:34 +08:00
|
|
|
config_base |= ARMV8_PMU_EXCLUDE_EL1;
|
2019-04-10 03:22:15 +08:00
|
|
|
if (attr->exclude_host)
|
|
|
|
config_base |= ARMV8_PMU_EXCLUDE_EL0;
|
2017-05-03 00:29:34 +08:00
|
|
|
} else {
|
2019-04-10 03:22:13 +08:00
|
|
|
if (!attr->exclude_hv && !attr->exclude_host)
|
2017-05-03 00:29:34 +08:00
|
|
|
config_base |= ARMV8_PMU_INCLUDE_EL2;
|
|
|
|
}
|
2019-04-10 03:22:13 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Filter out !VHE kernels and guest kernels
|
|
|
|
*/
|
|
|
|
if (attr->exclude_kernel)
|
|
|
|
config_base |= ARMV8_PMU_EXCLUDE_EL1;
|
|
|
|
|
2012-03-05 19:49:32 +08:00
|
|
|
if (attr->exclude_user)
|
2016-03-25 00:01:16 +08:00
|
|
|
config_base |= ARMV8_PMU_EXCLUDE_EL0;
|
2012-03-05 19:49:32 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Install the filter into config_base as this is used to
|
|
|
|
* construct the event type.
|
|
|
|
*/
|
|
|
|
event->config_base = config_base;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void armv8pmu_reset(void *info)
|
|
|
|
{
|
2020-03-03 02:17:52 +08:00
|
|
|
struct arm_pmu *cpu_pmu = (struct arm_pmu *)info;
|
|
|
|
u32 pmcr;
|
|
|
|
|
2012-03-05 19:49:32 +08:00
|
|
|
/* The counter and interrupt enable registers are unknown at reset. */
|
2020-03-18 02:22:54 +08:00
|
|
|
armv8pmu_disable_counter(U32_MAX);
|
|
|
|
armv8pmu_disable_intens(U32_MAX);
|
2012-03-05 19:49:32 +08:00
|
|
|
|
2019-04-10 03:22:13 +08:00
|
|
|
/* Clear the counters we flip at guest entry/exit */
|
|
|
|
kvm_clr_pmu_events(U32_MAX);
|
|
|
|
|
2016-02-19 00:50:13 +08:00
|
|
|
/*
|
|
|
|
* Initialize & Reset PMNC. Request overflow interrupt for
|
|
|
|
* 64 bit cycle counter but cheat in armv8pmu_write_counter().
|
|
|
|
*/
|
2020-03-03 02:17:52 +08:00
|
|
|
pmcr = ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C | ARMV8_PMU_PMCR_LC;
|
|
|
|
|
|
|
|
/* Enable long event counter support where available */
|
|
|
|
if (armv8pmu_has_long_event(cpu_pmu))
|
|
|
|
pmcr |= ARMV8_PMU_PMCR_LP;
|
|
|
|
|
|
|
|
armv8pmu_pmcr_write(pmcr);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2023-04-11 17:38:09 +08:00
|
|
|
static int __armv8_pmuv3_map_event_id(struct arm_pmu *armpmu,
|
|
|
|
struct perf_event *event)
|
|
|
|
{
|
|
|
|
if (event->attr.type == PERF_TYPE_HARDWARE &&
|
|
|
|
event->attr.config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS) {
|
|
|
|
|
|
|
|
if (test_bit(ARMV8_PMUV3_PERFCTR_PC_WRITE_RETIRED,
|
|
|
|
armpmu->pmceid_bitmap))
|
|
|
|
return ARMV8_PMUV3_PERFCTR_PC_WRITE_RETIRED;
|
|
|
|
|
|
|
|
if (test_bit(ARMV8_PMUV3_PERFCTR_BR_RETIRED,
|
|
|
|
armpmu->pmceid_bitmap))
|
|
|
|
return ARMV8_PMUV3_PERFCTR_BR_RETIRED;
|
|
|
|
|
|
|
|
return HW_OP_UNSUPPORTED;
|
|
|
|
}
|
|
|
|
|
|
|
|
return armpmu_map_event(event, &armv8_pmuv3_perf_map,
|
|
|
|
&armv8_pmuv3_perf_cache_map,
|
|
|
|
ARMV8_PMU_EVTYPE_EVENT);
|
|
|
|
}
|
|
|
|
|
2017-08-08 23:58:33 +08:00
|
|
|
static int __armv8_pmuv3_map_event(struct perf_event *event,
|
|
|
|
const unsigned (*extra_event_map)
|
|
|
|
[PERF_COUNT_HW_MAX],
|
|
|
|
const unsigned (*extra_cache_map)
|
|
|
|
[PERF_COUNT_HW_CACHE_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_OP_MAX]
|
|
|
|
[PERF_COUNT_HW_CACHE_RESULT_MAX])
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
2016-09-15 06:32:30 +08:00
|
|
|
int hw_event_id;
|
|
|
|
struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
|
|
|
|
|
2023-04-11 17:38:09 +08:00
|
|
|
hw_event_id = __armv8_pmuv3_map_event_id(armpmu, event);
|
2016-09-15 06:32:30 +08:00
|
|
|
|
2023-02-16 22:12:39 +08:00
|
|
|
/*
|
|
|
|
* CHAIN events only work when paired with an adjacent counter, and it
|
|
|
|
* never makes sense for a user to open one in isolation, as they'll be
|
|
|
|
* rotated arbitrarily.
|
|
|
|
*/
|
|
|
|
if (hw_event_id == ARMV8_PMUV3_PERFCTR_CHAIN)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2018-07-10 16:58:04 +08:00
|
|
|
if (armv8pmu_event_is_64bit(event))
|
|
|
|
event->hw.flags |= ARMPMU_EVT_64BIT;
|
|
|
|
|
2021-12-09 04:11:23 +08:00
|
|
|
/*
|
|
|
|
* User events must be allocated into a single counter, and so
|
|
|
|
* must not be chained.
|
|
|
|
*
|
|
|
|
* Most 64-bit events require long counter support, but 64-bit
|
|
|
|
* CPU_CYCLES events can be placed into the dedicated cycle
|
|
|
|
* counter when this is free.
|
|
|
|
*/
|
|
|
|
if (armv8pmu_event_want_user_access(event)) {
|
|
|
|
if (!(event->attach_state & PERF_ATTACH_TASK))
|
|
|
|
return -EINVAL;
|
|
|
|
if (armv8pmu_event_is_64bit(event) &&
|
|
|
|
(hw_event_id != ARMV8_PMUV3_PERFCTR_CPU_CYCLES) &&
|
|
|
|
!armv8pmu_has_long_event(armpmu))
|
|
|
|
return -EOPNOTSUPP;
|
|
|
|
|
|
|
|
event->hw.flags |= PERF_EVENT_FLAG_USER_READ_CNT;
|
|
|
|
}
|
|
|
|
|
2018-10-06 15:57:38 +08:00
|
|
|
/* Only expose micro/arch events supported by this PMU */
|
2017-08-08 23:58:33 +08:00
|
|
|
if ((hw_event_id > 0) && (hw_event_id < ARMV8_PMUV3_MAX_COMMON_EVENTS)
|
|
|
|
&& test_bit(hw_event_id, armpmu->pmceid_bitmap)) {
|
|
|
|
return hw_event_id;
|
2016-09-15 06:32:30 +08:00
|
|
|
}
|
|
|
|
|
2017-08-08 23:58:33 +08:00
|
|
|
return armpmu_map_event(event, extra_event_map, extra_cache_map,
|
|
|
|
ARMV8_PMU_EVTYPE_EVENT);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int armv8_pmuv3_map_event(struct perf_event *event)
|
|
|
|
{
|
|
|
|
return __armv8_pmuv3_map_event(event, NULL, NULL);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2015-10-02 17:55:04 +08:00
|
|
|
static int armv8_a53_map_event(struct perf_event *event)
|
|
|
|
{
|
2017-08-09 00:11:27 +08:00
|
|
|
return __armv8_pmuv3_map_event(event, NULL, &armv8_a53_perf_cache_map);
|
2015-10-02 17:55:04 +08:00
|
|
|
}
|
|
|
|
|
2015-10-02 17:55:05 +08:00
|
|
|
static int armv8_a57_map_event(struct perf_event *event)
|
|
|
|
{
|
2017-08-09 00:11:27 +08:00
|
|
|
return __armv8_pmuv3_map_event(event, NULL, &armv8_a57_perf_cache_map);
|
2015-10-02 17:55:05 +08:00
|
|
|
}
|
|
|
|
|
2017-08-10 00:46:38 +08:00
|
|
|
static int armv8_a73_map_event(struct perf_event *event)
|
|
|
|
{
|
|
|
|
return __armv8_pmuv3_map_event(event, NULL, &armv8_a73_perf_cache_map);
|
|
|
|
}
|
|
|
|
|
2016-02-19 00:50:11 +08:00
|
|
|
static int armv8_thunder_map_event(struct perf_event *event)
|
|
|
|
{
|
2017-08-09 00:11:27 +08:00
|
|
|
return __armv8_pmuv3_map_event(event, NULL,
|
2017-08-08 23:58:33 +08:00
|
|
|
&armv8_thunder_perf_cache_map);
|
2016-02-19 00:50:11 +08:00
|
|
|
}
|
|
|
|
|
2016-04-21 20:58:45 +08:00
|
|
|
static int armv8_vulcan_map_event(struct perf_event *event)
|
|
|
|
{
|
2017-08-09 00:11:27 +08:00
|
|
|
return __armv8_pmuv3_map_event(event, NULL,
|
2017-08-08 23:58:33 +08:00
|
|
|
&armv8_vulcan_perf_cache_map);
|
2016-04-21 20:58:45 +08:00
|
|
|
}
|
|
|
|
|
2017-04-11 16:39:56 +08:00
|
|
|
struct armv8pmu_probe_info {
|
|
|
|
struct arm_pmu *pmu;
|
|
|
|
bool present;
|
|
|
|
};
|
|
|
|
|
2016-04-21 20:58:44 +08:00
|
|
|
static void __armv8pmu_probe_pmu(void *info)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
2017-04-11 16:39:56 +08:00
|
|
|
struct armv8pmu_probe_info *probe = info;
|
|
|
|
struct arm_pmu *cpu_pmu = probe->pmu;
|
2018-10-05 20:28:07 +08:00
|
|
|
u64 pmceid_raw[2];
|
2016-04-21 20:58:44 +08:00
|
|
|
u32 pmceid[2];
|
2017-04-25 19:08:50 +08:00
|
|
|
int pmuver;
|
2012-03-05 19:49:32 +08:00
|
|
|
|
2023-03-18 03:50:21 +08:00
|
|
|
pmuver = read_pmuver();
|
2023-03-18 03:50:22 +08:00
|
|
|
if (!pmuv3_implemented(pmuver))
|
2017-04-11 16:39:56 +08:00
|
|
|
return;
|
|
|
|
|
2020-03-03 02:17:52 +08:00
|
|
|
cpu_pmu->pmuver = pmuver;
|
2017-04-11 16:39:56 +08:00
|
|
|
probe->present = true;
|
|
|
|
|
2012-03-05 19:49:32 +08:00
|
|
|
/* Read the nb of CNTx counters supported from PMNC */
|
2016-04-21 20:58:44 +08:00
|
|
|
cpu_pmu->num_events = (armv8pmu_pmcr_read() >> ARMV8_PMU_PMCR_N_SHIFT)
|
|
|
|
& ARMV8_PMU_PMCR_N_MASK;
|
2012-03-05 19:49:32 +08:00
|
|
|
|
2015-10-02 17:55:03 +08:00
|
|
|
/* Add the CPU cycles counter */
|
2016-04-21 20:58:44 +08:00
|
|
|
cpu_pmu->num_events += 1;
|
|
|
|
|
2023-03-18 03:50:21 +08:00
|
|
|
pmceid[0] = pmceid_raw[0] = read_pmceid0();
|
|
|
|
pmceid[1] = pmceid_raw[1] = read_pmceid1();
|
2016-04-21 20:58:44 +08:00
|
|
|
|
2018-02-07 07:38:06 +08:00
|
|
|
bitmap_from_arr32(cpu_pmu->pmceid_bitmap,
|
|
|
|
pmceid, ARMV8_PMUV3_MAX_COMMON_EVENTS);
|
2018-10-05 20:28:07 +08:00
|
|
|
|
|
|
|
pmceid[0] = pmceid_raw[0] >> 32;
|
|
|
|
pmceid[1] = pmceid_raw[1] >> 32;
|
|
|
|
|
|
|
|
bitmap_from_arr32(cpu_pmu->pmceid_ext_bitmap,
|
|
|
|
pmceid, ARMV8_PMUV3_MAX_COMMON_EVENTS);
|
2020-09-22 13:53:45 +08:00
|
|
|
|
2023-03-18 03:50:21 +08:00
|
|
|
/* store PMMIR register for sysfs */
|
2023-03-18 03:50:22 +08:00
|
|
|
if (is_pmuv3p4(pmuver) && (pmceid_raw[1] & BIT(31)))
|
2023-03-18 03:50:21 +08:00
|
|
|
cpu_pmu->reg_pmmir = read_pmmir();
|
2020-09-22 13:53:45 +08:00
|
|
|
else
|
|
|
|
cpu_pmu->reg_pmmir = 0;
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2016-04-21 20:58:44 +08:00
|
|
|
static int armv8pmu_probe_pmu(struct arm_pmu *cpu_pmu)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
2017-04-11 16:39:56 +08:00
|
|
|
struct armv8pmu_probe_info probe = {
|
|
|
|
.pmu = cpu_pmu,
|
|
|
|
.present = false,
|
|
|
|
};
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = smp_call_function_any(&cpu_pmu->supported_cpus,
|
2016-04-21 20:58:44 +08:00
|
|
|
__armv8pmu_probe_pmu,
|
2017-04-11 16:39:56 +08:00
|
|
|
&probe, 1);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
return probe.present ? 0 : -ENODEV;
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2021-12-09 04:11:23 +08:00
|
|
|
static void armv8pmu_disable_user_access_ipi(void *unused)
|
|
|
|
{
|
|
|
|
armv8pmu_disable_user_access();
|
|
|
|
}
|
|
|
|
|
|
|
|
static int armv8pmu_proc_user_access_handler(struct ctl_table *table, int write,
|
|
|
|
void *buffer, size_t *lenp, loff_t *ppos)
|
|
|
|
{
|
|
|
|
int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
|
|
|
|
if (ret || !write || sysctl_perf_user_access)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
on_each_cpu(armv8pmu_disable_user_access_ipi, NULL, 1);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2021-12-09 04:11:22 +08:00
|
|
|
static struct ctl_table armv8_pmu_sysctl_table[] = {
|
|
|
|
{
|
|
|
|
.procname = "perf_user_access",
|
|
|
|
.data = &sysctl_perf_user_access,
|
|
|
|
.maxlen = sizeof(unsigned int),
|
|
|
|
.mode = 0644,
|
2021-12-09 04:11:23 +08:00
|
|
|
.proc_handler = armv8pmu_proc_user_access_handler,
|
2021-12-09 04:11:22 +08:00
|
|
|
.extra1 = SYSCTL_ZERO,
|
|
|
|
.extra2 = SYSCTL_ONE,
|
|
|
|
},
|
|
|
|
{ }
|
|
|
|
};
|
|
|
|
|
2022-01-04 22:57:14 +08:00
|
|
|
static void armv8_pmu_register_sysctl_table(void)
|
|
|
|
{
|
|
|
|
static u32 tbl_registered = 0;
|
|
|
|
|
|
|
|
if (!cmpxchg_relaxed(&tbl_registered, 0, 1))
|
|
|
|
register_sysctl("kernel", armv8_pmu_sysctl_table);
|
|
|
|
}
|
|
|
|
|
2020-02-22 03:35:31 +08:00
|
|
|
static int armv8_pmu_init(struct arm_pmu *cpu_pmu, char *name,
|
|
|
|
int (*map_event)(struct perf_event *event),
|
|
|
|
const struct attribute_group *events,
|
2020-09-22 13:53:45 +08:00
|
|
|
const struct attribute_group *format,
|
|
|
|
const struct attribute_group *caps)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
2017-04-11 16:39:56 +08:00
|
|
|
int ret = armv8pmu_probe_pmu(cpu_pmu);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
2018-10-05 20:26:21 +08:00
|
|
|
cpu_pmu->handle_irq = armv8pmu_handle_irq;
|
|
|
|
cpu_pmu->enable = armv8pmu_enable_event;
|
|
|
|
cpu_pmu->disable = armv8pmu_disable_event;
|
|
|
|
cpu_pmu->read_counter = armv8pmu_read_counter;
|
|
|
|
cpu_pmu->write_counter = armv8pmu_write_counter;
|
|
|
|
cpu_pmu->get_event_idx = armv8pmu_get_event_idx;
|
|
|
|
cpu_pmu->clear_event_idx = armv8pmu_clear_event_idx;
|
|
|
|
cpu_pmu->start = armv8pmu_start;
|
|
|
|
cpu_pmu->stop = armv8pmu_stop;
|
|
|
|
cpu_pmu->reset = armv8pmu_reset;
|
2015-10-02 17:55:04 +08:00
|
|
|
cpu_pmu->set_event_filter = armv8pmu_set_event_filter;
|
2017-04-11 16:39:56 +08:00
|
|
|
|
2021-12-09 04:11:23 +08:00
|
|
|
cpu_pmu->pmu.event_idx = armv8pmu_user_event_idx;
|
|
|
|
|
2020-02-22 03:35:31 +08:00
|
|
|
cpu_pmu->name = name;
|
|
|
|
cpu_pmu->map_event = map_event;
|
|
|
|
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] = events ?
|
|
|
|
events : &armv8_pmuv3_events_attr_group;
|
|
|
|
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] = format ?
|
|
|
|
format : &armv8_pmuv3_format_attr_group;
|
2020-09-22 13:53:45 +08:00
|
|
|
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_CAPS] = caps ?
|
|
|
|
caps : &armv8_pmuv3_caps_attr_group;
|
2020-02-22 03:35:31 +08:00
|
|
|
|
2022-01-04 22:57:14 +08:00
|
|
|
armv8_pmu_register_sysctl_table();
|
2017-04-11 16:39:56 +08:00
|
|
|
return 0;
|
2015-10-02 17:55:04 +08:00
|
|
|
}
|
|
|
|
|
2020-09-22 13:53:45 +08:00
|
|
|
static int armv8_pmu_init_nogroups(struct arm_pmu *cpu_pmu, char *name,
|
|
|
|
int (*map_event)(struct perf_event *event))
|
|
|
|
{
|
|
|
|
return armv8_pmu_init(cpu_pmu, name, map_event, NULL, NULL, NULL);
|
|
|
|
}
|
|
|
|
|
2021-12-14 22:16:14 +08:00
|
|
|
#define PMUV3_INIT_SIMPLE(name) \
|
|
|
|
static int name##_pmu_init(struct arm_pmu *cpu_pmu) \
|
|
|
|
{ \
|
|
|
|
return armv8_pmu_init_nogroups(cpu_pmu, #name, armv8_pmuv3_map_event);\
|
2015-10-02 17:55:04 +08:00
|
|
|
}
|
|
|
|
|
2021-12-14 22:16:14 +08:00
|
|
|
PMUV3_INIT_SIMPLE(armv8_pmuv3)
|
|
|
|
|
|
|
|
PMUV3_INIT_SIMPLE(armv8_cortex_a34)
|
|
|
|
PMUV3_INIT_SIMPLE(armv8_cortex_a55)
|
|
|
|
PMUV3_INIT_SIMPLE(armv8_cortex_a65)
|
|
|
|
PMUV3_INIT_SIMPLE(armv8_cortex_a75)
|
|
|
|
PMUV3_INIT_SIMPLE(armv8_cortex_a76)
|
|
|
|
PMUV3_INIT_SIMPLE(armv8_cortex_a77)
|
|
|
|
PMUV3_INIT_SIMPLE(armv8_cortex_a78)
|
2021-12-14 22:16:15 +08:00
|
|
|
PMUV3_INIT_SIMPLE(armv9_cortex_a510)
|
|
|
|
PMUV3_INIT_SIMPLE(armv9_cortex_a710)
|
|
|
|
PMUV3_INIT_SIMPLE(armv8_cortex_x1)
|
|
|
|
PMUV3_INIT_SIMPLE(armv9_cortex_x2)
|
2021-12-14 22:16:14 +08:00
|
|
|
PMUV3_INIT_SIMPLE(armv8_neoverse_e1)
|
|
|
|
PMUV3_INIT_SIMPLE(armv8_neoverse_n1)
|
2021-12-14 22:16:15 +08:00
|
|
|
PMUV3_INIT_SIMPLE(armv9_neoverse_n2)
|
|
|
|
PMUV3_INIT_SIMPLE(armv8_neoverse_v1)
|
2021-12-14 22:16:14 +08:00
|
|
|
|
|
|
|
PMUV3_INIT_SIMPLE(armv8_nvidia_carmel)
|
|
|
|
PMUV3_INIT_SIMPLE(armv8_nvidia_denver)
|
2020-02-22 03:35:32 +08:00
|
|
|
|
2017-08-10 00:46:39 +08:00
|
|
|
static int armv8_a35_pmu_init(struct arm_pmu *cpu_pmu)
|
|
|
|
{
|
2020-09-22 13:53:45 +08:00
|
|
|
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a35",
|
|
|
|
armv8_a53_map_event);
|
2017-08-10 00:46:39 +08:00
|
|
|
}
|
|
|
|
|
2015-10-02 17:55:04 +08:00
|
|
|
static int armv8_a53_pmu_init(struct arm_pmu *cpu_pmu)
|
|
|
|
{
|
2020-09-22 13:53:45 +08:00
|
|
|
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a53",
|
|
|
|
armv8_a53_map_event);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2015-10-02 17:55:05 +08:00
|
|
|
static int armv8_a57_pmu_init(struct arm_pmu *cpu_pmu)
|
|
|
|
{
|
2020-09-22 13:53:45 +08:00
|
|
|
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a57",
|
|
|
|
armv8_a57_map_event);
|
2015-10-02 17:55:05 +08:00
|
|
|
}
|
|
|
|
|
2015-12-22 22:45:35 +08:00
|
|
|
static int armv8_a72_pmu_init(struct arm_pmu *cpu_pmu)
|
|
|
|
{
|
2020-09-22 13:53:45 +08:00
|
|
|
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a72",
|
|
|
|
armv8_a57_map_event);
|
2015-12-22 22:45:35 +08:00
|
|
|
}
|
|
|
|
|
2017-08-10 00:46:38 +08:00
|
|
|
static int armv8_a73_pmu_init(struct arm_pmu *cpu_pmu)
|
|
|
|
{
|
2020-09-22 13:53:45 +08:00
|
|
|
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a73",
|
|
|
|
armv8_a73_map_event);
|
2017-08-10 00:46:38 +08:00
|
|
|
}
|
|
|
|
|
2016-02-19 00:50:11 +08:00
|
|
|
static int armv8_thunder_pmu_init(struct arm_pmu *cpu_pmu)
|
|
|
|
{
|
2020-09-22 13:53:45 +08:00
|
|
|
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cavium_thunder",
|
|
|
|
armv8_thunder_map_event);
|
2016-02-19 00:50:11 +08:00
|
|
|
}
|
|
|
|
|
2016-04-21 20:58:45 +08:00
|
|
|
static int armv8_vulcan_pmu_init(struct arm_pmu *cpu_pmu)
|
|
|
|
{
|
2020-09-22 13:53:45 +08:00
|
|
|
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_brcm_vulcan",
|
|
|
|
armv8_vulcan_map_event);
|
2016-04-21 20:58:45 +08:00
|
|
|
}
|
|
|
|
|
2015-10-02 17:55:03 +08:00
|
|
|
static const struct of_device_id armv8_pmu_of_device_ids[] = {
|
2021-12-14 22:16:14 +08:00
|
|
|
{.compatible = "arm,armv8-pmuv3", .data = armv8_pmuv3_pmu_init},
|
|
|
|
{.compatible = "arm,cortex-a34-pmu", .data = armv8_cortex_a34_pmu_init},
|
2017-08-10 00:46:39 +08:00
|
|
|
{.compatible = "arm,cortex-a35-pmu", .data = armv8_a35_pmu_init},
|
2015-10-02 17:55:04 +08:00
|
|
|
{.compatible = "arm,cortex-a53-pmu", .data = armv8_a53_pmu_init},
|
2021-12-14 22:16:14 +08:00
|
|
|
{.compatible = "arm,cortex-a55-pmu", .data = armv8_cortex_a55_pmu_init},
|
2015-10-02 17:55:05 +08:00
|
|
|
{.compatible = "arm,cortex-a57-pmu", .data = armv8_a57_pmu_init},
|
2021-12-14 22:16:14 +08:00
|
|
|
{.compatible = "arm,cortex-a65-pmu", .data = armv8_cortex_a65_pmu_init},
|
2015-12-22 22:45:35 +08:00
|
|
|
{.compatible = "arm,cortex-a72-pmu", .data = armv8_a72_pmu_init},
|
2017-08-10 00:46:38 +08:00
|
|
|
{.compatible = "arm,cortex-a73-pmu", .data = armv8_a73_pmu_init},
|
2021-12-14 22:16:14 +08:00
|
|
|
{.compatible = "arm,cortex-a75-pmu", .data = armv8_cortex_a75_pmu_init},
|
|
|
|
{.compatible = "arm,cortex-a76-pmu", .data = armv8_cortex_a76_pmu_init},
|
|
|
|
{.compatible = "arm,cortex-a77-pmu", .data = armv8_cortex_a77_pmu_init},
|
|
|
|
{.compatible = "arm,cortex-a78-pmu", .data = armv8_cortex_a78_pmu_init},
|
2021-12-14 22:16:15 +08:00
|
|
|
{.compatible = "arm,cortex-a510-pmu", .data = armv9_cortex_a510_pmu_init},
|
|
|
|
{.compatible = "arm,cortex-a710-pmu", .data = armv9_cortex_a710_pmu_init},
|
|
|
|
{.compatible = "arm,cortex-x1-pmu", .data = armv8_cortex_x1_pmu_init},
|
|
|
|
{.compatible = "arm,cortex-x2-pmu", .data = armv9_cortex_x2_pmu_init},
|
2021-12-14 22:16:14 +08:00
|
|
|
{.compatible = "arm,neoverse-e1-pmu", .data = armv8_neoverse_e1_pmu_init},
|
|
|
|
{.compatible = "arm,neoverse-n1-pmu", .data = armv8_neoverse_n1_pmu_init},
|
2021-12-14 22:16:15 +08:00
|
|
|
{.compatible = "arm,neoverse-n2-pmu", .data = armv9_neoverse_n2_pmu_init},
|
|
|
|
{.compatible = "arm,neoverse-v1-pmu", .data = armv8_neoverse_v1_pmu_init},
|
2016-02-19 00:50:11 +08:00
|
|
|
{.compatible = "cavium,thunder-pmu", .data = armv8_thunder_pmu_init},
|
2016-04-21 20:58:45 +08:00
|
|
|
{.compatible = "brcm,vulcan-pmu", .data = armv8_vulcan_pmu_init},
|
2021-12-14 22:16:14 +08:00
|
|
|
{.compatible = "nvidia,carmel-pmu", .data = armv8_nvidia_carmel_pmu_init},
|
|
|
|
{.compatible = "nvidia,denver-pmu", .data = armv8_nvidia_denver_pmu_init},
|
2012-03-05 19:49:32 +08:00
|
|
|
{},
|
|
|
|
};
|
|
|
|
|
2015-10-02 17:55:03 +08:00
|
|
|
static int armv8_pmu_device_probe(struct platform_device *pdev)
|
2012-03-05 19:49:32 +08:00
|
|
|
{
|
2017-04-11 16:39:57 +08:00
|
|
|
return arm_pmu_device_probe(pdev, armv8_pmu_of_device_ids, NULL);
|
2012-03-05 19:49:32 +08:00
|
|
|
}
|
|
|
|
|
2015-10-02 17:55:03 +08:00
|
|
|
static struct platform_driver armv8_pmu_driver = {
|
2012-03-05 19:49:32 +08:00
|
|
|
.driver = {
|
2016-09-15 06:32:31 +08:00
|
|
|
.name = ARMV8_PMU_PDEV_NAME,
|
2015-10-02 17:55:03 +08:00
|
|
|
.of_match_table = armv8_pmu_of_device_ids,
|
2018-10-17 23:26:22 +08:00
|
|
|
.suppress_bind_attrs = true,
|
2012-03-05 19:49:32 +08:00
|
|
|
},
|
2015-10-02 17:55:03 +08:00
|
|
|
.probe = armv8_pmu_device_probe,
|
2012-03-05 19:49:32 +08:00
|
|
|
};
|
|
|
|
|
2017-04-11 16:39:57 +08:00
|
|
|
static int __init armv8_pmu_driver_init(void)
|
|
|
|
{
|
|
|
|
if (acpi_disabled)
|
2021-01-13 06:18:55 +08:00
|
|
|
return platform_driver_register(&armv8_pmu_driver);
|
2017-04-11 16:39:57 +08:00
|
|
|
else
|
2021-12-14 22:16:14 +08:00
|
|
|
return arm_pmu_acpi_probe(armv8_pmuv3_pmu_init);
|
2017-04-11 16:39:57 +08:00
|
|
|
}
|
|
|
|
device_initcall(armv8_pmu_driver_init)
|
arm64: perf: Add cap_user_time aarch64
It is useful to get the running time of a thread. Doing so in an
efficient manner can be important for performance of user applications.
Avoiding system calls in `clock_gettime` when handling
CLOCK_THREAD_CPUTIME_ID is important. Other clocks are handled in the
VDSO, but CLOCK_THREAD_CPUTIME_ID falls back on the system call.
CLOCK_THREAD_CPUTIME_ID is not handled in the VDSO since it would have
costs associated with maintaining updated user space accessible time
offsets. These offsets have to be updated everytime the a thread is
scheduled/descheduled. However, for programs regularly checking the
running time of a thread, this is a performance improvement.
This patch takes a middle ground, and adds support for cap_user_time an
optional feature of the perf_event API. This way costs are only
incurred when the perf_event api is enabled. This is done the same way
as it is in x86.
Ultimately this allows calculating the thread running time in userspace
on aarch64 as follows (adapted from perf_event_open manpage):
u32 seq, time_mult, time_shift;
u64 running, count, time_offset, quot, rem, delta;
struct perf_event_mmap_page *pc;
pc = buf; // buf is the perf event mmaped page as documented in the API.
if (pc->cap_usr_time) {
do {
seq = pc->lock;
barrier();
running = pc->time_running;
count = readCNTVCT_EL0(); // Read ARM hardware clock.
time_offset = pc->time_offset;
time_mult = pc->time_mult;
time_shift = pc->time_shift;
barrier();
} while (pc->lock != seq);
quot = (count >> time_shift);
rem = count & (((u64)1 << time_shift) - 1);
delta = time_offset + quot * time_mult +
((rem * time_mult) >> time_shift);
running += delta;
// running now has the current nanosecond level thread time.
}
Summary of changes in the patch:
For aarch64 systems, make arch_perf_update_userpage update the timing
information stored in the perf_event page. Requiring the following
calculations:
- Calculate the appropriate time_mult, and time_shift factors to convert
ticks to nano seconds for the current clock frequency.
- Adjust the mult and shift factors to avoid shift factors of 32 bits.
(possibly unnecessary)
- The time_offset userspace should apply when doing calculations:
negative the current sched time (now), because time_running and
time_enabled fields of the perf_event page have just been updated.
Toggle bits to appropriate values:
- Enable cap_user_time
Signed-off-by: Michael O'Farrell <micpof@gmail.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
2018-07-31 04:14:34 +08:00
|
|
|
|
|
|
|
void arch_perf_update_userpage(struct perf_event *event,
|
|
|
|
struct perf_event_mmap_page *userpg, u64 now)
|
|
|
|
{
|
2020-07-16 13:11:26 +08:00
|
|
|
struct clock_read_data *rd;
|
|
|
|
unsigned int seq;
|
|
|
|
u64 ns;
|
arm64: perf: Add cap_user_time aarch64
It is useful to get the running time of a thread. Doing so in an
efficient manner can be important for performance of user applications.
Avoiding system calls in `clock_gettime` when handling
CLOCK_THREAD_CPUTIME_ID is important. Other clocks are handled in the
VDSO, but CLOCK_THREAD_CPUTIME_ID falls back on the system call.
CLOCK_THREAD_CPUTIME_ID is not handled in the VDSO since it would have
costs associated with maintaining updated user space accessible time
offsets. These offsets have to be updated everytime the a thread is
scheduled/descheduled. However, for programs regularly checking the
running time of a thread, this is a performance improvement.
This patch takes a middle ground, and adds support for cap_user_time an
optional feature of the perf_event API. This way costs are only
incurred when the perf_event api is enabled. This is done the same way
as it is in x86.
Ultimately this allows calculating the thread running time in userspace
on aarch64 as follows (adapted from perf_event_open manpage):
u32 seq, time_mult, time_shift;
u64 running, count, time_offset, quot, rem, delta;
struct perf_event_mmap_page *pc;
pc = buf; // buf is the perf event mmaped page as documented in the API.
if (pc->cap_usr_time) {
do {
seq = pc->lock;
barrier();
running = pc->time_running;
count = readCNTVCT_EL0(); // Read ARM hardware clock.
time_offset = pc->time_offset;
time_mult = pc->time_mult;
time_shift = pc->time_shift;
barrier();
} while (pc->lock != seq);
quot = (count >> time_shift);
rem = count & (((u64)1 << time_shift) - 1);
delta = time_offset + quot * time_mult +
((rem * time_mult) >> time_shift);
running += delta;
// running now has the current nanosecond level thread time.
}
Summary of changes in the patch:
For aarch64 systems, make arch_perf_update_userpage update the timing
information stored in the perf_event page. Requiring the following
calculations:
- Calculate the appropriate time_mult, and time_shift factors to convert
ticks to nano seconds for the current clock frequency.
- Adjust the mult and shift factors to avoid shift factors of 32 bits.
(possibly unnecessary)
- The time_offset userspace should apply when doing calculations:
negative the current sched time (now), because time_running and
time_enabled fields of the perf_event page have just been updated.
Toggle bits to appropriate values:
- Enable cap_user_time
Signed-off-by: Michael O'Farrell <micpof@gmail.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
2018-07-31 04:14:34 +08:00
|
|
|
|
2020-07-16 13:11:27 +08:00
|
|
|
userpg->cap_user_time = 0;
|
|
|
|
userpg->cap_user_time_zero = 0;
|
2020-07-16 13:11:29 +08:00
|
|
|
userpg->cap_user_time_short = 0;
|
2021-12-09 04:11:23 +08:00
|
|
|
userpg->cap_user_rdpmc = armv8pmu_event_has_user_read(event);
|
|
|
|
|
|
|
|
if (userpg->cap_user_rdpmc) {
|
|
|
|
if (event->hw.flags & ARMPMU_EVT_64BIT)
|
|
|
|
userpg->pmc_width = 64;
|
|
|
|
else
|
|
|
|
userpg->pmc_width = 32;
|
|
|
|
}
|
2020-07-16 13:11:26 +08:00
|
|
|
|
|
|
|
do {
|
|
|
|
rd = sched_clock_read_begin(&seq);
|
|
|
|
|
2020-07-16 13:11:27 +08:00
|
|
|
if (rd->read_sched_clock != arch_timer_read_counter)
|
|
|
|
return;
|
|
|
|
|
2020-07-16 13:11:26 +08:00
|
|
|
userpg->time_mult = rd->mult;
|
|
|
|
userpg->time_shift = rd->shift;
|
|
|
|
userpg->time_zero = rd->epoch_ns;
|
2020-07-16 13:11:29 +08:00
|
|
|
userpg->time_cycles = rd->epoch_cyc;
|
|
|
|
userpg->time_mask = rd->sched_clock_mask;
|
2020-07-16 13:11:26 +08:00
|
|
|
|
|
|
|
/*
|
2020-07-16 13:11:29 +08:00
|
|
|
* Subtract the cycle base, such that software that
|
|
|
|
* doesn't know about cap_user_time_short still 'works'
|
|
|
|
* assuming no wraps.
|
2020-07-16 13:11:26 +08:00
|
|
|
*/
|
|
|
|
ns = mul_u64_u32_shr(rd->epoch_cyc, rd->mult, rd->shift);
|
|
|
|
userpg->time_zero -= ns;
|
|
|
|
|
|
|
|
} while (sched_clock_read_retry(seq));
|
|
|
|
|
|
|
|
userpg->time_offset = userpg->time_zero - now;
|
arm64: perf: Add cap_user_time aarch64
It is useful to get the running time of a thread. Doing so in an
efficient manner can be important for performance of user applications.
Avoiding system calls in `clock_gettime` when handling
CLOCK_THREAD_CPUTIME_ID is important. Other clocks are handled in the
VDSO, but CLOCK_THREAD_CPUTIME_ID falls back on the system call.
CLOCK_THREAD_CPUTIME_ID is not handled in the VDSO since it would have
costs associated with maintaining updated user space accessible time
offsets. These offsets have to be updated everytime the a thread is
scheduled/descheduled. However, for programs regularly checking the
running time of a thread, this is a performance improvement.
This patch takes a middle ground, and adds support for cap_user_time an
optional feature of the perf_event API. This way costs are only
incurred when the perf_event api is enabled. This is done the same way
as it is in x86.
Ultimately this allows calculating the thread running time in userspace
on aarch64 as follows (adapted from perf_event_open manpage):
u32 seq, time_mult, time_shift;
u64 running, count, time_offset, quot, rem, delta;
struct perf_event_mmap_page *pc;
pc = buf; // buf is the perf event mmaped page as documented in the API.
if (pc->cap_usr_time) {
do {
seq = pc->lock;
barrier();
running = pc->time_running;
count = readCNTVCT_EL0(); // Read ARM hardware clock.
time_offset = pc->time_offset;
time_mult = pc->time_mult;
time_shift = pc->time_shift;
barrier();
} while (pc->lock != seq);
quot = (count >> time_shift);
rem = count & (((u64)1 << time_shift) - 1);
delta = time_offset + quot * time_mult +
((rem * time_mult) >> time_shift);
running += delta;
// running now has the current nanosecond level thread time.
}
Summary of changes in the patch:
For aarch64 systems, make arch_perf_update_userpage update the timing
information stored in the perf_event page. Requiring the following
calculations:
- Calculate the appropriate time_mult, and time_shift factors to convert
ticks to nano seconds for the current clock frequency.
- Adjust the mult and shift factors to avoid shift factors of 32 bits.
(possibly unnecessary)
- The time_offset userspace should apply when doing calculations:
negative the current sched time (now), because time_running and
time_enabled fields of the perf_event page have just been updated.
Toggle bits to appropriate values:
- Enable cap_user_time
Signed-off-by: Michael O'Farrell <micpof@gmail.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
2018-07-31 04:14:34 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* time_shift is not expected to be greater than 31 due to
|
|
|
|
* the original published conversion algorithm shifting a
|
|
|
|
* 32-bit value (now specifies a 64-bit value) - refer
|
|
|
|
* perf_event_mmap_page documentation in perf_event.h.
|
|
|
|
*/
|
2020-07-16 13:11:26 +08:00
|
|
|
if (userpg->time_shift == 32) {
|
|
|
|
userpg->time_shift = 31;
|
arm64: perf: Add cap_user_time aarch64
It is useful to get the running time of a thread. Doing so in an
efficient manner can be important for performance of user applications.
Avoiding system calls in `clock_gettime` when handling
CLOCK_THREAD_CPUTIME_ID is important. Other clocks are handled in the
VDSO, but CLOCK_THREAD_CPUTIME_ID falls back on the system call.
CLOCK_THREAD_CPUTIME_ID is not handled in the VDSO since it would have
costs associated with maintaining updated user space accessible time
offsets. These offsets have to be updated everytime the a thread is
scheduled/descheduled. However, for programs regularly checking the
running time of a thread, this is a performance improvement.
This patch takes a middle ground, and adds support for cap_user_time an
optional feature of the perf_event API. This way costs are only
incurred when the perf_event api is enabled. This is done the same way
as it is in x86.
Ultimately this allows calculating the thread running time in userspace
on aarch64 as follows (adapted from perf_event_open manpage):
u32 seq, time_mult, time_shift;
u64 running, count, time_offset, quot, rem, delta;
struct perf_event_mmap_page *pc;
pc = buf; // buf is the perf event mmaped page as documented in the API.
if (pc->cap_usr_time) {
do {
seq = pc->lock;
barrier();
running = pc->time_running;
count = readCNTVCT_EL0(); // Read ARM hardware clock.
time_offset = pc->time_offset;
time_mult = pc->time_mult;
time_shift = pc->time_shift;
barrier();
} while (pc->lock != seq);
quot = (count >> time_shift);
rem = count & (((u64)1 << time_shift) - 1);
delta = time_offset + quot * time_mult +
((rem * time_mult) >> time_shift);
running += delta;
// running now has the current nanosecond level thread time.
}
Summary of changes in the patch:
For aarch64 systems, make arch_perf_update_userpage update the timing
information stored in the perf_event page. Requiring the following
calculations:
- Calculate the appropriate time_mult, and time_shift factors to convert
ticks to nano seconds for the current clock frequency.
- Adjust the mult and shift factors to avoid shift factors of 32 bits.
(possibly unnecessary)
- The time_offset userspace should apply when doing calculations:
negative the current sched time (now), because time_running and
time_enabled fields of the perf_event page have just been updated.
Toggle bits to appropriate values:
- Enable cap_user_time
Signed-off-by: Michael O'Farrell <micpof@gmail.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
2018-07-31 04:14:34 +08:00
|
|
|
userpg->time_mult >>= 1;
|
|
|
|
}
|
2020-07-16 13:11:26 +08:00
|
|
|
|
2020-07-16 13:11:27 +08:00
|
|
|
/*
|
|
|
|
* Internal timekeeping for enabled/running/stopped times
|
|
|
|
* is always computed with the sched_clock.
|
|
|
|
*/
|
|
|
|
userpg->cap_user_time = 1;
|
|
|
|
userpg->cap_user_time_zero = 1;
|
2020-07-16 13:11:29 +08:00
|
|
|
userpg->cap_user_time_short = 1;
|
arm64: perf: Add cap_user_time aarch64
It is useful to get the running time of a thread. Doing so in an
efficient manner can be important for performance of user applications.
Avoiding system calls in `clock_gettime` when handling
CLOCK_THREAD_CPUTIME_ID is important. Other clocks are handled in the
VDSO, but CLOCK_THREAD_CPUTIME_ID falls back on the system call.
CLOCK_THREAD_CPUTIME_ID is not handled in the VDSO since it would have
costs associated with maintaining updated user space accessible time
offsets. These offsets have to be updated everytime the a thread is
scheduled/descheduled. However, for programs regularly checking the
running time of a thread, this is a performance improvement.
This patch takes a middle ground, and adds support for cap_user_time an
optional feature of the perf_event API. This way costs are only
incurred when the perf_event api is enabled. This is done the same way
as it is in x86.
Ultimately this allows calculating the thread running time in userspace
on aarch64 as follows (adapted from perf_event_open manpage):
u32 seq, time_mult, time_shift;
u64 running, count, time_offset, quot, rem, delta;
struct perf_event_mmap_page *pc;
pc = buf; // buf is the perf event mmaped page as documented in the API.
if (pc->cap_usr_time) {
do {
seq = pc->lock;
barrier();
running = pc->time_running;
count = readCNTVCT_EL0(); // Read ARM hardware clock.
time_offset = pc->time_offset;
time_mult = pc->time_mult;
time_shift = pc->time_shift;
barrier();
} while (pc->lock != seq);
quot = (count >> time_shift);
rem = count & (((u64)1 << time_shift) - 1);
delta = time_offset + quot * time_mult +
((rem * time_mult) >> time_shift);
running += delta;
// running now has the current nanosecond level thread time.
}
Summary of changes in the patch:
For aarch64 systems, make arch_perf_update_userpage update the timing
information stored in the perf_event page. Requiring the following
calculations:
- Calculate the appropriate time_mult, and time_shift factors to convert
ticks to nano seconds for the current clock frequency.
- Adjust the mult and shift factors to avoid shift factors of 32 bits.
(possibly unnecessary)
- The time_offset userspace should apply when doing calculations:
negative the current sched time (now), because time_running and
time_enabled fields of the perf_event page have just been updated.
Toggle bits to appropriate values:
- Enable cap_user_time
Signed-off-by: Michael O'Farrell <micpof@gmail.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
2018-07-31 04:14:34 +08:00
|
|
|
}
|