OpenCloudOS-Kernel/tools/perf/util/callchain.c

844 lines
19 KiB
C
Raw Normal View History

/*
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
* Copyright (C) 2009-2011, Frederic Weisbecker <fweisbec@gmail.com>
*
* Handle the callchains from the stream in an ad-hoc radix tree and then
* sort them in an rbtree.
*
* Using a radix for code path provides a fast retrieval and factorizes
* memory use. Also that lets us use the paths in a hierarchical graph view.
*
*/
#include <stdlib.h>
#include <stdio.h>
#include <stdbool.h>
#include <errno.h>
#include <math.h>
#include "asm/bug.h"
#include "hist.h"
perf annotate: Use build-ids to find the right DSO We were still using the pathname found on the MMAP event, that could not be the one we used when recording, so use the build-id cache for that, only falling back to use the pathname in the MMAP event if no build-ids are available. With this we now also are able to do secure, seamless offline annotation. Example: [root@doppio linux-2.6-tip]# perf report -g none -v 2> /dev/null | head -10 8.12% Xorg /usr/lib64/libpixman-1.so.0.14.0 0x0000000000026d02 B [.] pixman_rasterize_edges 4.68% firefox /usr/lib64/xulrunner-1.9.1/libxul.so 0x00000000005dbdba B [.] 0x000000005dbdba 3.70% swapper /lib/modules/2.6.34-rc6/build/vmlinux 0xffffffff81022cea ! [k] read_hpet 2.96% init /lib/modules/2.6.34-rc6/build/vmlinux 0xffffffff81022cea ! [k] read_hpet 2.73% swapper /lib/modules/2.6.34-rc6/build/vmlinux 0xffffffff8100a738 ! [k] mwait_idle_with_hints [root@doppio linux-2.6-tip]# perf annotate -v pixman_rasterize_edges 2>&1 | grep Executing Executing: objdump --start-address=0x000000371ce26670 --stop-address=0x000000371ce2709f -dS /root/.debug/.build-id/bd/6ac5199137aaeb279f864717d8d061477466c1|grep -v /root/.debug/.build-id/bd/6ac5199137aaeb279f864717d8d061477466c1|expand [root@doppio linux-2.6-tip]# perf buildid-list | grep libpixman-1.so.0.14.0 bd6ac5199137aaeb279f864717d8d061477466c1 /usr/lib64/libpixman-1.so.0.14.0 [root@doppio linux-2.6-tip]# Reported-by: Stephane Eranian <eranian@google.com> Cc: Frédéric Weisbecker <fweisbec@gmail.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> Cc: Tom Zanussi <tzanussi@gmail.com> LKML-Reference: <new-submission> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2010-05-20 23:15:33 +08:00
#include "util.h"
#include "sort.h"
#include "machine.h"
#include "callchain.h"
__thread struct callchain_cursor callchain_cursor;
#ifdef HAVE_DWARF_UNWIND_SUPPORT
static int get_stack_size(const char *str, unsigned long *_size)
{
char *endptr;
unsigned long size;
unsigned long max_size = round_down(USHRT_MAX, sizeof(u64));
size = strtoul(str, &endptr, 0);
do {
if (*endptr)
break;
size = round_up(size, sizeof(u64));
if (!size || size > max_size)
break;
*_size = size;
return 0;
} while (0);
pr_err("callchain: Incorrect stack dump size (max %ld): %s\n",
max_size, str);
return -1;
}
#endif /* HAVE_DWARF_UNWIND_SUPPORT */
int parse_callchain_record_opt(const char *arg)
{
char *tok, *name, *saveptr = NULL;
char *buf;
int ret = -1;
/* We need buffer that we know we can write to. */
buf = malloc(strlen(arg) + 1);
if (!buf)
return -ENOMEM;
strcpy(buf, arg);
tok = strtok_r((char *)buf, ",", &saveptr);
name = tok ? : (char *)buf;
do {
/* Framepointer style */
if (!strncmp(name, "fp", sizeof("fp"))) {
if (!strtok_r(NULL, ",", &saveptr)) {
callchain_param.record_mode = CALLCHAIN_FP;
ret = 0;
} else
pr_err("callchain: No more arguments "
"needed for -g fp\n");
break;
#ifdef HAVE_DWARF_UNWIND_SUPPORT
/* Dwarf style */
} else if (!strncmp(name, "dwarf", sizeof("dwarf"))) {
const unsigned long default_stack_dump_size = 8192;
ret = 0;
callchain_param.record_mode = CALLCHAIN_DWARF;
callchain_param.dump_size = default_stack_dump_size;
tok = strtok_r(NULL, ",", &saveptr);
if (tok) {
unsigned long size = 0;
ret = get_stack_size(tok, &size);
callchain_param.dump_size = size;
}
#endif /* HAVE_DWARF_UNWIND_SUPPORT */
} else {
pr_err("callchain: Unknown --call-graph option "
"value: %s\n", arg);
break;
}
} while (0);
free(buf);
return ret;
}
static int parse_callchain_mode(const char *value)
{
if (!strncmp(value, "graph", strlen(value))) {
callchain_param.mode = CHAIN_GRAPH_ABS;
return 0;
}
if (!strncmp(value, "flat", strlen(value))) {
callchain_param.mode = CHAIN_FLAT;
return 0;
}
if (!strncmp(value, "fractal", strlen(value))) {
callchain_param.mode = CHAIN_GRAPH_REL;
return 0;
}
return -1;
}
static int parse_callchain_order(const char *value)
{
if (!strncmp(value, "caller", strlen(value))) {
callchain_param.order = ORDER_CALLER;
return 0;
}
if (!strncmp(value, "callee", strlen(value))) {
callchain_param.order = ORDER_CALLEE;
return 0;
}
return -1;
}
static int parse_callchain_sort_key(const char *value)
{
if (!strncmp(value, "function", strlen(value))) {
callchain_param.key = CCKEY_FUNCTION;
return 0;
}
if (!strncmp(value, "address", strlen(value))) {
callchain_param.key = CCKEY_ADDRESS;
return 0;
}
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
if (!strncmp(value, "branch", strlen(value))) {
callchain_param.branch_callstack = 1;
return 0;
}
return -1;
}
int
parse_callchain_report_opt(const char *arg)
{
char *tok;
char *endptr;
bool minpcnt_set = false;
symbol_conf.use_callchain = true;
if (!arg)
return 0;
while ((tok = strtok((char *)arg, ",")) != NULL) {
if (!strncmp(tok, "none", strlen(tok))) {
callchain_param.mode = CHAIN_NONE;
symbol_conf.use_callchain = false;
return 0;
}
if (!parse_callchain_mode(tok) ||
!parse_callchain_order(tok) ||
!parse_callchain_sort_key(tok)) {
/* parsing ok - move on to the next */
} else if (!minpcnt_set) {
/* try to get the min percent */
callchain_param.min_percent = strtod(tok, &endptr);
if (tok == endptr)
return -1;
minpcnt_set = true;
} else {
/* try print limit at last */
callchain_param.print_limit = strtoul(tok, &endptr, 0);
if (tok == endptr)
return -1;
}
arg = NULL;
}
if (callchain_register_param(&callchain_param) < 0) {
pr_err("Can't register callchain params\n");
return -1;
}
return 0;
}
int perf_callchain_config(const char *var, const char *value)
{
char *endptr;
if (prefixcmp(var, "call-graph."))
return 0;
var += sizeof("call-graph.") - 1;
if (!strcmp(var, "record-mode"))
return parse_callchain_record_opt(value);
#ifdef HAVE_DWARF_UNWIND_SUPPORT
if (!strcmp(var, "dump-size")) {
unsigned long size = 0;
int ret;
ret = get_stack_size(value, &size);
callchain_param.dump_size = size;
return ret;
}
#endif
if (!strcmp(var, "print-type"))
return parse_callchain_mode(value);
if (!strcmp(var, "order"))
return parse_callchain_order(value);
if (!strcmp(var, "sort-key"))
return parse_callchain_sort_key(value);
if (!strcmp(var, "threshold")) {
callchain_param.min_percent = strtod(value, &endptr);
if (value == endptr)
return -1;
}
if (!strcmp(var, "print-limit")) {
callchain_param.print_limit = strtod(value, &endptr);
if (value == endptr)
return -1;
}
return 0;
}
static void
perf report: Add support for callchain graph output Currently, the printing of callchains is done in a single vertical level, this is the "flat" mode: 8.25% [k] copy_user_generic_string 4.19% copy_user_generic_string generic_file_aio_read do_sync_read vfs_read sys_pread64 system_call_fastpath pread64 This patch introduces a new "graph" mode which provides a hierarchical output of factorized paths recursively sorted: 8.25% [k] copy_user_generic_string | |--4.31%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--4.19%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --0.12%-- sys_read | system_call_fastpath | __read | |--3.24%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--3.14%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --0.10%-- sys_write [...] The command line has then changed. By providing the -c option, the callchain will output in the flat mode by default. But you can override it: perf report -c graph or perf report -c flat You can also pass the abreviated mode: perf report -c g or perf report -c gra will both make use of the graph mode. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246550301-8954-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-02 23:58:21 +08:00
rb_insert_callchain(struct rb_root *root, struct callchain_node *chain,
enum chain_mode mode)
{
struct rb_node **p = &root->rb_node;
struct rb_node *parent = NULL;
struct callchain_node *rnode;
u64 chain_cumul = callchain_cumul_hits(chain);
while (*p) {
u64 rnode_cumul;
parent = *p;
rnode = rb_entry(parent, struct callchain_node, rb_node);
rnode_cumul = callchain_cumul_hits(rnode);
perf report: Add support for callchain graph output Currently, the printing of callchains is done in a single vertical level, this is the "flat" mode: 8.25% [k] copy_user_generic_string 4.19% copy_user_generic_string generic_file_aio_read do_sync_read vfs_read sys_pread64 system_call_fastpath pread64 This patch introduces a new "graph" mode which provides a hierarchical output of factorized paths recursively sorted: 8.25% [k] copy_user_generic_string | |--4.31%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--4.19%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --0.12%-- sys_read | system_call_fastpath | __read | |--3.24%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--3.14%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --0.10%-- sys_write [...] The command line has then changed. By providing the -c option, the callchain will output in the flat mode by default. But you can override it: perf report -c graph or perf report -c flat You can also pass the abreviated mode: perf report -c g or perf report -c gra will both make use of the graph mode. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246550301-8954-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-02 23:58:21 +08:00
switch (mode) {
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
case CHAIN_FLAT:
perf report: Add support for callchain graph output Currently, the printing of callchains is done in a single vertical level, this is the "flat" mode: 8.25% [k] copy_user_generic_string 4.19% copy_user_generic_string generic_file_aio_read do_sync_read vfs_read sys_pread64 system_call_fastpath pread64 This patch introduces a new "graph" mode which provides a hierarchical output of factorized paths recursively sorted: 8.25% [k] copy_user_generic_string | |--4.31%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--4.19%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --0.12%-- sys_read | system_call_fastpath | __read | |--3.24%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--3.14%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --0.10%-- sys_write [...] The command line has then changed. By providing the -c option, the callchain will output in the flat mode by default. But you can override it: perf report -c graph or perf report -c flat You can also pass the abreviated mode: perf report -c g or perf report -c gra will both make use of the graph mode. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246550301-8954-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-02 23:58:21 +08:00
if (rnode->hit < chain->hit)
p = &(*p)->rb_left;
else
p = &(*p)->rb_right;
break;
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
case CHAIN_GRAPH_ABS: /* Falldown */
case CHAIN_GRAPH_REL:
if (rnode_cumul < chain_cumul)
perf report: Add support for callchain graph output Currently, the printing of callchains is done in a single vertical level, this is the "flat" mode: 8.25% [k] copy_user_generic_string 4.19% copy_user_generic_string generic_file_aio_read do_sync_read vfs_read sys_pread64 system_call_fastpath pread64 This patch introduces a new "graph" mode which provides a hierarchical output of factorized paths recursively sorted: 8.25% [k] copy_user_generic_string | |--4.31%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--4.19%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --0.12%-- sys_read | system_call_fastpath | __read | |--3.24%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--3.14%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --0.10%-- sys_write [...] The command line has then changed. By providing the -c option, the callchain will output in the flat mode by default. But you can override it: perf report -c graph or perf report -c flat You can also pass the abreviated mode: perf report -c g or perf report -c gra will both make use of the graph mode. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246550301-8954-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-02 23:58:21 +08:00
p = &(*p)->rb_left;
else
p = &(*p)->rb_right;
break;
perf: Enable more compiler warnings Related to a shadowed variable bug fix Valdis Kletnieks noticed that perf does not get built with -Wshadow, which could have helped us avoid the bug. So enable -Wshadow and also enable the following warnings on perf builds, in addition to the already enabled -Wall -Wextra -std=gnu99 warnings: -Wcast-align -Wformat=2 -Wshadow -Winit-self -Wpacked -Wredundant-decls -Wstack-protector -Wstrict-aliasing=3 -Wswitch-default -Wswitch-enum -Wno-system-headers -Wundef -Wvolatile-register-var -Wwrite-strings -Wbad-function-cast -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wold-style-definition -Wstrict-prototypes -Wdeclaration-after-statement And change/fix the perf code to build cleanly under GCC 4.3.2. The list of warnings enablement is rather arbitrary: it's based on my (quick) reading of the GCC manpages and trying them on perf. I categorized the warnings based on individually enabling them and looking whether they trigger something in the perf build. If i liked those warnings (i.e. if they trigger for something that arguably could be improved) i enabled the warning. If the warnings seemed to come from language laywers spamming the build with tons of nuisance warnings i generally kept them off. Most of the sign conversion related warnings were in this category. (A second patch enabling some of the sign warnings might be welcome - sign bugs can be nasty.) I also kept warnings that seem to make sense from their manpage description and which produced no actual warnings on our code base. These warnings might still be turned off if they end up being a nuisance. I also left out a few warnings that are not supported in older compilers. [ Note that these changes might break the build on older compilers i did not test, or on non-x86 architectures that produce different warnings, so more testing would be welcome. ] Reported-by: Valdis.Kletnieks@vt.edu Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-08-15 18:26:57 +08:00
case CHAIN_NONE:
perf report: Add support for callchain graph output Currently, the printing of callchains is done in a single vertical level, this is the "flat" mode: 8.25% [k] copy_user_generic_string 4.19% copy_user_generic_string generic_file_aio_read do_sync_read vfs_read sys_pread64 system_call_fastpath pread64 This patch introduces a new "graph" mode which provides a hierarchical output of factorized paths recursively sorted: 8.25% [k] copy_user_generic_string | |--4.31%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--4.19%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --0.12%-- sys_read | system_call_fastpath | __read | |--3.24%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--3.14%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --0.10%-- sys_write [...] The command line has then changed. By providing the -c option, the callchain will output in the flat mode by default. But you can override it: perf report -c graph or perf report -c flat You can also pass the abreviated mode: perf report -c g or perf report -c gra will both make use of the graph mode. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246550301-8954-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-02 23:58:21 +08:00
default:
break;
}
}
rb_link_node(&chain->rb_node, parent, p);
rb_insert_color(&chain->rb_node, root);
}
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
static void
__sort_chain_flat(struct rb_root *rb_root, struct callchain_node *node,
u64 min_hit)
{
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
struct rb_node *n;
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
struct callchain_node *child;
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
n = rb_first(&node->rb_root_in);
while (n) {
child = rb_entry(n, struct callchain_node, rb_node_in);
n = rb_next(n);
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
__sort_chain_flat(rb_root, child, min_hit);
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
}
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
if (node->hit && node->hit >= min_hit)
rb_insert_callchain(rb_root, node, CHAIN_FLAT);
}
/*
* Once we get every callchains from the stream, we can now
* sort them by hit
*/
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
static void
sort_chain_flat(struct rb_root *rb_root, struct callchain_root *root,
perf tools: Use __maybe_used for unused variables perf defines both __used and __unused variables to use for marking unused variables. The variable __used is defined to __attribute__((__unused__)), which contradicts the kernel definition to __attribute__((__used__)) for new gcc versions. On Android, __used is also defined in system headers and this leads to warnings like: warning: '__used__' attribute ignored __unused is not defined in the kernel and is not a standard definition. If __unused is included everywhere instead of __used, this leads to conflicts with glibc headers, since glibc has a variables with this name in its headers. The best approach is to use __maybe_unused, the definition used in the kernel for __attribute__((unused)). In this way there is only one definition in perf sources (instead of 2 definitions that point to the same thing: __used and __unused) and it works on both Linux and Android. This patch simply replaces all instances of __used and __unused with __maybe_unused. Signed-off-by: Irina Tirdea <irina.tirdea@intel.com> Acked-by: Pekka Enberg <penberg@kernel.org> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1347315303-29906-7-git-send-email-irina.tirdea@intel.com [ committer note: fixed up conflict with a116e05 in builtin-sched.c ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 06:15:03 +08:00
u64 min_hit, struct callchain_param *param __maybe_unused)
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
{
__sort_chain_flat(rb_root, &root->node, min_hit);
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
}
static void __sort_chain_graph_abs(struct callchain_node *node,
u64 min_hit)
{
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
struct rb_node *n;
struct callchain_node *child;
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
node->rb_root = RB_ROOT;
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
n = rb_first(&node->rb_root_in);
while (n) {
child = rb_entry(n, struct callchain_node, rb_node_in);
n = rb_next(n);
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
__sort_chain_graph_abs(child, min_hit);
if (callchain_cumul_hits(child) >= min_hit)
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
rb_insert_callchain(&node->rb_root, child,
CHAIN_GRAPH_ABS);
}
}
static void
sort_chain_graph_abs(struct rb_root *rb_root, struct callchain_root *chain_root,
perf tools: Use __maybe_used for unused variables perf defines both __used and __unused variables to use for marking unused variables. The variable __used is defined to __attribute__((__unused__)), which contradicts the kernel definition to __attribute__((__used__)) for new gcc versions. On Android, __used is also defined in system headers and this leads to warnings like: warning: '__used__' attribute ignored __unused is not defined in the kernel and is not a standard definition. If __unused is included everywhere instead of __used, this leads to conflicts with glibc headers, since glibc has a variables with this name in its headers. The best approach is to use __maybe_unused, the definition used in the kernel for __attribute__((unused)). In this way there is only one definition in perf sources (instead of 2 definitions that point to the same thing: __used and __unused) and it works on both Linux and Android. This patch simply replaces all instances of __used and __unused with __maybe_unused. Signed-off-by: Irina Tirdea <irina.tirdea@intel.com> Acked-by: Pekka Enberg <penberg@kernel.org> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1347315303-29906-7-git-send-email-irina.tirdea@intel.com [ committer note: fixed up conflict with a116e05 in builtin-sched.c ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 06:15:03 +08:00
u64 min_hit, struct callchain_param *param __maybe_unused)
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
{
__sort_chain_graph_abs(&chain_root->node, min_hit);
rb_root->rb_node = chain_root->node.rb_root.rb_node;
perf report: Add support for callchain graph output Currently, the printing of callchains is done in a single vertical level, this is the "flat" mode: 8.25% [k] copy_user_generic_string 4.19% copy_user_generic_string generic_file_aio_read do_sync_read vfs_read sys_pread64 system_call_fastpath pread64 This patch introduces a new "graph" mode which provides a hierarchical output of factorized paths recursively sorted: 8.25% [k] copy_user_generic_string | |--4.31%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--4.19%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --0.12%-- sys_read | system_call_fastpath | __read | |--3.24%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--3.14%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --0.10%-- sys_write [...] The command line has then changed. By providing the -c option, the callchain will output in the flat mode by default. But you can override it: perf report -c graph or perf report -c flat You can also pass the abreviated mode: perf report -c g or perf report -c gra will both make use of the graph mode. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246550301-8954-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-02 23:58:21 +08:00
}
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
static void __sort_chain_graph_rel(struct callchain_node *node,
double min_percent)
perf report: Add support for callchain graph output Currently, the printing of callchains is done in a single vertical level, this is the "flat" mode: 8.25% [k] copy_user_generic_string 4.19% copy_user_generic_string generic_file_aio_read do_sync_read vfs_read sys_pread64 system_call_fastpath pread64 This patch introduces a new "graph" mode which provides a hierarchical output of factorized paths recursively sorted: 8.25% [k] copy_user_generic_string | |--4.31%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--4.19%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --0.12%-- sys_read | system_call_fastpath | __read | |--3.24%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--3.14%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --0.10%-- sys_write [...] The command line has then changed. By providing the -c option, the callchain will output in the flat mode by default. But you can override it: perf report -c graph or perf report -c flat You can also pass the abreviated mode: perf report -c g or perf report -c gra will both make use of the graph mode. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246550301-8954-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-02 23:58:21 +08:00
{
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
struct rb_node *n;
perf report: Add support for callchain graph output Currently, the printing of callchains is done in a single vertical level, this is the "flat" mode: 8.25% [k] copy_user_generic_string 4.19% copy_user_generic_string generic_file_aio_read do_sync_read vfs_read sys_pread64 system_call_fastpath pread64 This patch introduces a new "graph" mode which provides a hierarchical output of factorized paths recursively sorted: 8.25% [k] copy_user_generic_string | |--4.31%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--4.19%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --0.12%-- sys_read | system_call_fastpath | __read | |--3.24%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--3.14%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --0.10%-- sys_write [...] The command line has then changed. By providing the -c option, the callchain will output in the flat mode by default. But you can override it: perf report -c graph or perf report -c flat You can also pass the abreviated mode: perf report -c g or perf report -c gra will both make use of the graph mode. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246550301-8954-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-02 23:58:21 +08:00
struct callchain_node *child;
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
u64 min_hit;
perf report: Add support for callchain graph output Currently, the printing of callchains is done in a single vertical level, this is the "flat" mode: 8.25% [k] copy_user_generic_string 4.19% copy_user_generic_string generic_file_aio_read do_sync_read vfs_read sys_pread64 system_call_fastpath pread64 This patch introduces a new "graph" mode which provides a hierarchical output of factorized paths recursively sorted: 8.25% [k] copy_user_generic_string | |--4.31%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--4.19%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --0.12%-- sys_read | system_call_fastpath | __read | |--3.24%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--3.14%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --0.10%-- sys_write [...] The command line has then changed. By providing the -c option, the callchain will output in the flat mode by default. But you can override it: perf report -c graph or perf report -c flat You can also pass the abreviated mode: perf report -c g or perf report -c gra will both make use of the graph mode. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246550301-8954-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-02 23:58:21 +08:00
node->rb_root = RB_ROOT;
min_hit = ceil(node->children_hit * min_percent);
perf report: Add support for callchain graph output Currently, the printing of callchains is done in a single vertical level, this is the "flat" mode: 8.25% [k] copy_user_generic_string 4.19% copy_user_generic_string generic_file_aio_read do_sync_read vfs_read sys_pread64 system_call_fastpath pread64 This patch introduces a new "graph" mode which provides a hierarchical output of factorized paths recursively sorted: 8.25% [k] copy_user_generic_string | |--4.31%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--4.19%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --0.12%-- sys_read | system_call_fastpath | __read | |--3.24%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--3.14%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --0.10%-- sys_write [...] The command line has then changed. By providing the -c option, the callchain will output in the flat mode by default. But you can override it: perf report -c graph or perf report -c flat You can also pass the abreviated mode: perf report -c g or perf report -c gra will both make use of the graph mode. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246550301-8954-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-02 23:58:21 +08:00
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
n = rb_first(&node->rb_root_in);
while (n) {
child = rb_entry(n, struct callchain_node, rb_node_in);
n = rb_next(n);
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
__sort_chain_graph_rel(child, min_percent);
if (callchain_cumul_hits(child) >= min_hit)
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
rb_insert_callchain(&node->rb_root, child,
CHAIN_GRAPH_REL);
perf report: Add support for callchain graph output Currently, the printing of callchains is done in a single vertical level, this is the "flat" mode: 8.25% [k] copy_user_generic_string 4.19% copy_user_generic_string generic_file_aio_read do_sync_read vfs_read sys_pread64 system_call_fastpath pread64 This patch introduces a new "graph" mode which provides a hierarchical output of factorized paths recursively sorted: 8.25% [k] copy_user_generic_string | |--4.31%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--4.19%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --0.12%-- sys_read | system_call_fastpath | __read | |--3.24%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--3.14%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --0.10%-- sys_write [...] The command line has then changed. By providing the -c option, the callchain will output in the flat mode by default. But you can override it: perf report -c graph or perf report -c flat You can also pass the abreviated mode: perf report -c g or perf report -c gra will both make use of the graph mode. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246550301-8954-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-02 23:58:21 +08:00
}
}
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
static void
sort_chain_graph_rel(struct rb_root *rb_root, struct callchain_root *chain_root,
perf tools: Use __maybe_used for unused variables perf defines both __used and __unused variables to use for marking unused variables. The variable __used is defined to __attribute__((__unused__)), which contradicts the kernel definition to __attribute__((__used__)) for new gcc versions. On Android, __used is also defined in system headers and this leads to warnings like: warning: '__used__' attribute ignored __unused is not defined in the kernel and is not a standard definition. If __unused is included everywhere instead of __used, this leads to conflicts with glibc headers, since glibc has a variables with this name in its headers. The best approach is to use __maybe_unused, the definition used in the kernel for __attribute__((unused)). In this way there is only one definition in perf sources (instead of 2 definitions that point to the same thing: __used and __unused) and it works on both Linux and Android. This patch simply replaces all instances of __used and __unused with __maybe_unused. Signed-off-by: Irina Tirdea <irina.tirdea@intel.com> Acked-by: Pekka Enberg <penberg@kernel.org> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1347315303-29906-7-git-send-email-irina.tirdea@intel.com [ committer note: fixed up conflict with a116e05 in builtin-sched.c ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 06:15:03 +08:00
u64 min_hit __maybe_unused, struct callchain_param *param)
perf report: Add support for callchain graph output Currently, the printing of callchains is done in a single vertical level, this is the "flat" mode: 8.25% [k] copy_user_generic_string 4.19% copy_user_generic_string generic_file_aio_read do_sync_read vfs_read sys_pread64 system_call_fastpath pread64 This patch introduces a new "graph" mode which provides a hierarchical output of factorized paths recursively sorted: 8.25% [k] copy_user_generic_string | |--4.31%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--4.19%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --0.12%-- sys_read | system_call_fastpath | __read | |--3.24%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--3.14%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --0.10%-- sys_write [...] The command line has then changed. By providing the -c option, the callchain will output in the flat mode by default. But you can override it: perf report -c graph or perf report -c flat You can also pass the abreviated mode: perf report -c g or perf report -c gra will both make use of the graph mode. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246550301-8954-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-02 23:58:21 +08:00
{
__sort_chain_graph_rel(&chain_root->node, param->min_percent / 100.0);
rb_root->rb_node = chain_root->node.rb_root.rb_node;
}
int callchain_register_param(struct callchain_param *param)
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
{
switch (param->mode) {
case CHAIN_GRAPH_ABS:
param->sort = sort_chain_graph_abs;
break;
case CHAIN_GRAPH_REL:
param->sort = sort_chain_graph_rel;
break;
case CHAIN_FLAT:
param->sort = sort_chain_flat;
break;
perf: Enable more compiler warnings Related to a shadowed variable bug fix Valdis Kletnieks noticed that perf does not get built with -Wshadow, which could have helped us avoid the bug. So enable -Wshadow and also enable the following warnings on perf builds, in addition to the already enabled -Wall -Wextra -std=gnu99 warnings: -Wcast-align -Wformat=2 -Wshadow -Winit-self -Wpacked -Wredundant-decls -Wstack-protector -Wstrict-aliasing=3 -Wswitch-default -Wswitch-enum -Wno-system-headers -Wundef -Wvolatile-register-var -Wwrite-strings -Wbad-function-cast -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wold-style-definition -Wstrict-prototypes -Wdeclaration-after-statement And change/fix the perf code to build cleanly under GCC 4.3.2. The list of warnings enablement is rather arbitrary: it's based on my (quick) reading of the GCC manpages and trying them on perf. I categorized the warnings based on individually enabling them and looking whether they trigger something in the perf build. If i liked those warnings (i.e. if they trigger for something that arguably could be improved) i enabled the warning. If the warnings seemed to come from language laywers spamming the build with tons of nuisance warnings i generally kept them off. Most of the sign conversion related warnings were in this category. (A second patch enabling some of the sign warnings might be welcome - sign bugs can be nasty.) I also kept warnings that seem to make sense from their manpage description and which produced no actual warnings on our code base. These warnings might still be turned off if they end up being a nuisance. I also left out a few warnings that are not supported in older compilers. [ Note that these changes might break the build on older compilers i did not test, or on non-x86 architectures that produce different warnings, so more testing would be welcome. ] Reported-by: Valdis.Kletnieks@vt.edu Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-08-15 18:26:57 +08:00
case CHAIN_NONE:
perf report: Add "Fractal" mode output - support callchains with relative overhead rate The current callchain displays the overhead rates as absolute: relative to the total overhead. This patch provides relative overhead percentage, in which each branch of the callchain tree is a independant instrumentated object. This provides a 'fractal' view of the call-chain profile: each sub-graph looks like a profile in itself - relative to its parent. You can produce such output by using the "fractal" mode that you can abbreviate via f, fr, fra, frac, etc... ./perf report -s sym -c fractal Example: 8.46% [k] copy_user_generic_string | |--52.01%-- generic_file_aio_read | do_sync_read | vfs_read | | | |--97.20%-- sys_pread64 | | system_call_fastpath | | pread64 | | | --2.81%-- sys_read | system_call_fastpath | __read | |--39.85%-- generic_file_buffered_write | __generic_file_aio_write_nolock | generic_file_aio_write | do_sync_write | reiserfs_file_write | vfs_write | | | |--97.05%-- sys_pwrite64 | | system_call_fastpath | | __pwrite64 | | | --2.95%-- sys_write | system_call_fastpath | __write_nocancel [...] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Anton Blanchard <anton@samba.org> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1246772361-9960-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-07-05 13:39:21 +08:00
default:
return -1;
}
return 0;
}
/*
* Create a child for a parent. If inherit_children, then the new child
* will become the new parent of it's parent children
*/
static struct callchain_node *
create_child(struct callchain_node *parent, bool inherit_children)
{
struct callchain_node *new;
new = zalloc(sizeof(*new));
if (!new) {
perror("not enough memory to create child for code path tree");
return NULL;
}
new->parent = parent;
INIT_LIST_HEAD(&new->val);
if (inherit_children) {
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
struct rb_node *n;
struct callchain_node *child;
new->rb_root_in = parent->rb_root_in;
parent->rb_root_in = RB_ROOT;
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
n = rb_first(&new->rb_root_in);
while (n) {
child = rb_entry(n, struct callchain_node, rb_node_in);
child->parent = new;
n = rb_next(n);
}
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
/* make it the first child */
rb_link_node(&new->rb_node_in, NULL, &parent->rb_root_in.rb_node);
rb_insert_color(&new->rb_node_in, &parent->rb_root_in);
}
return new;
}
perf: Fix orphan callchain branches Callchains have markers inside their capture to tell we enter a context (kernel, user, ...). Those are not displayed in the callchains but they are incidentally an active part of the radix tree where callchains are stored, just like any other address. If we have the two following callchains: addr1 -> addr2 -> user context -> addr3 addr1 -> addr2 -> user context -> addr4 addr1 -> addr2 -> addr 5 This is pretty common if addr1 and addr2 are part of an interrupt path, addr3 and addr4 are user addresses and addr5 is a kernel non interrupt path. This will be stored as follows in the tree: addr1 addr2 / \ / addr5 user context / \ addr3 addr4 But we ignore the context markers in the report, hence the addr3 and addr4 will appear as orphan branches: |--28.30%-- hrtimer_interrupt | smp_apic_timer_interrupt | apic_timer_interrupt | | <------------- here, no parent! | | | | | |--11.11%-- 0x7fae7bccb875 | | | | | |--11.11%-- 0xffffffffff60013b | | | | | |--11.11%-- __pthread_mutex_lock_internal | | | | | |--11.11%-- __errno_location Fix this by removing the context markers when we process the callchains to the tree. Reported-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1269274173-20328-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-23 00:09:33 +08:00
/*
* Fill the node with callchain values
*/
static void
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
fill_node(struct callchain_node *node, struct callchain_cursor *cursor)
{
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
struct callchain_cursor_node *cursor_node;
node->val_nr = cursor->nr - cursor->pos;
if (!node->val_nr)
pr_warning("Warning: empty node in callchain tree\n");
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
cursor_node = callchain_cursor_current(cursor);
while (cursor_node) {
struct callchain_list *call;
call = zalloc(sizeof(*call));
if (!call) {
perror("not enough memory for the code path tree");
return;
}
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
call->ip = cursor_node->ip;
call->ms.sym = cursor_node->sym;
call->ms.map = cursor_node->map;
list_add_tail(&call->list, &node->val);
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
callchain_cursor_advance(cursor);
cursor_node = callchain_cursor_current(cursor);
}
}
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
static struct callchain_node *
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
add_child(struct callchain_node *parent,
struct callchain_cursor *cursor,
u64 period)
{
struct callchain_node *new;
new = create_child(parent, false);
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
fill_node(new, cursor);
new->children_hit = 0;
new->hit = period;
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
return new;
}
static s64 match_chain(struct callchain_cursor_node *node,
struct callchain_list *cnode)
{
struct symbol *sym = node->sym;
if (cnode->ms.sym && sym &&
callchain_param.key == CCKEY_FUNCTION)
return cnode->ms.sym->start - sym->start;
else
return cnode->ip - node->ip;
}
/*
* Split the parent in two parts (a new child is created) and
* give a part of its callchain to the created child.
* Then create another child to host the given callchain of new branch
*/
static void
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
split_add_child(struct callchain_node *parent,
struct callchain_cursor *cursor,
struct callchain_list *to_split,
u64 idx_parents, u64 idx_local, u64 period)
{
struct callchain_node *new;
struct list_head *old_tail;
unsigned int idx_total = idx_parents + idx_local;
/* split */
new = create_child(parent, true);
/* split the callchain and move a part to the new child */
old_tail = parent->val.prev;
list_del_range(&to_split->list, old_tail);
new->val.next = &to_split->list;
new->val.prev = old_tail;
to_split->list.prev = &new->val;
old_tail->next = &new->val;
/* split the hits */
new->hit = parent->hit;
new->children_hit = parent->children_hit;
parent->children_hit = callchain_cumul_hits(new);
new->val_nr = parent->val_nr - idx_local;
parent->val_nr = idx_local;
/* create a new child for the new branch if any */
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
if (idx_total < cursor->nr) {
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
struct callchain_node *first;
struct callchain_list *cnode;
struct callchain_cursor_node *node;
struct rb_node *p, **pp;
parent->hit = 0;
parent->children_hit += period;
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
node = callchain_cursor_current(cursor);
new = add_child(parent, cursor, period);
/*
* This is second child since we moved parent's children
* to new (first) child above.
*/
p = parent->rb_root_in.rb_node;
first = rb_entry(p, struct callchain_node, rb_node_in);
cnode = list_first_entry(&first->val, struct callchain_list,
list);
if (match_chain(node, cnode) < 0)
pp = &p->rb_left;
else
pp = &p->rb_right;
rb_link_node(&new->rb_node_in, p, pp);
rb_insert_color(&new->rb_node_in, &parent->rb_root_in);
} else {
parent->hit = period;
}
}
static int
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
append_chain(struct callchain_node *root,
struct callchain_cursor *cursor,
u64 period);
static void
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
append_chain_children(struct callchain_node *root,
struct callchain_cursor *cursor,
u64 period)
{
struct callchain_node *rnode;
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
struct callchain_cursor_node *node;
struct rb_node **p = &root->rb_root_in.rb_node;
struct rb_node *parent = NULL;
node = callchain_cursor_current(cursor);
if (!node)
return;
/* lookup in childrens */
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
while (*p) {
s64 ret;
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
parent = *p;
rnode = rb_entry(parent, struct callchain_node, rb_node_in);
/* If at least first entry matches, rely to children */
ret = append_chain(rnode, cursor, period);
if (ret == 0)
goto inc_children_hit;
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
if (ret < 0)
p = &parent->rb_left;
else
p = &parent->rb_right;
}
/* nothing in children, add to the current node */
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
rnode = add_child(root, cursor, period);
rb_link_node(&rnode->rb_node_in, parent, p);
rb_insert_color(&rnode->rb_node_in, &root->rb_root_in);
inc_children_hit:
root->children_hit += period;
}
static int
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
append_chain(struct callchain_node *root,
struct callchain_cursor *cursor,
u64 period)
{
struct callchain_list *cnode;
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
u64 start = cursor->pos;
bool found = false;
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
u64 matches;
int cmp = 0;
/*
* Lookup in the current node
* If we have a symbol, then compare the start to match
* anywhere inside a function, unless function
* mode is disabled.
*/
list_for_each_entry(cnode, &root->val, list) {
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
struct callchain_cursor_node *node;
perf: Fix orphan callchain branches Callchains have markers inside their capture to tell we enter a context (kernel, user, ...). Those are not displayed in the callchains but they are incidentally an active part of the radix tree where callchains are stored, just like any other address. If we have the two following callchains: addr1 -> addr2 -> user context -> addr3 addr1 -> addr2 -> user context -> addr4 addr1 -> addr2 -> addr 5 This is pretty common if addr1 and addr2 are part of an interrupt path, addr3 and addr4 are user addresses and addr5 is a kernel non interrupt path. This will be stored as follows in the tree: addr1 addr2 / \ / addr5 user context / \ addr3 addr4 But we ignore the context markers in the report, hence the addr3 and addr4 will appear as orphan branches: |--28.30%-- hrtimer_interrupt | smp_apic_timer_interrupt | apic_timer_interrupt | | <------------- here, no parent! | | | | | |--11.11%-- 0x7fae7bccb875 | | | | | |--11.11%-- 0xffffffffff60013b | | | | | |--11.11%-- __pthread_mutex_lock_internal | | | | | |--11.11%-- __errno_location Fix this by removing the context markers when we process the callchains to the tree. Reported-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1269274173-20328-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-23 00:09:33 +08:00
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
node = callchain_cursor_current(cursor);
if (!node)
break;
perf: Fix orphan callchain branches Callchains have markers inside their capture to tell we enter a context (kernel, user, ...). Those are not displayed in the callchains but they are incidentally an active part of the radix tree where callchains are stored, just like any other address. If we have the two following callchains: addr1 -> addr2 -> user context -> addr3 addr1 -> addr2 -> user context -> addr4 addr1 -> addr2 -> addr 5 This is pretty common if addr1 and addr2 are part of an interrupt path, addr3 and addr4 are user addresses and addr5 is a kernel non interrupt path. This will be stored as follows in the tree: addr1 addr2 / \ / addr5 user context / \ addr3 addr4 But we ignore the context markers in the report, hence the addr3 and addr4 will appear as orphan branches: |--28.30%-- hrtimer_interrupt | smp_apic_timer_interrupt | apic_timer_interrupt | | <------------- here, no parent! | | | | | |--11.11%-- 0x7fae7bccb875 | | | | | |--11.11%-- 0xffffffffff60013b | | | | | |--11.11%-- __pthread_mutex_lock_internal | | | | | |--11.11%-- __errno_location Fix this by removing the context markers when we process the callchains to the tree. Reported-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1269274173-20328-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-23 00:09:33 +08:00
cmp = match_chain(node, cnode);
if (cmp)
break;
perf: Fix orphan callchain branches Callchains have markers inside their capture to tell we enter a context (kernel, user, ...). Those are not displayed in the callchains but they are incidentally an active part of the radix tree where callchains are stored, just like any other address. If we have the two following callchains: addr1 -> addr2 -> user context -> addr3 addr1 -> addr2 -> user context -> addr4 addr1 -> addr2 -> addr 5 This is pretty common if addr1 and addr2 are part of an interrupt path, addr3 and addr4 are user addresses and addr5 is a kernel non interrupt path. This will be stored as follows in the tree: addr1 addr2 / \ / addr5 user context / \ addr3 addr4 But we ignore the context markers in the report, hence the addr3 and addr4 will appear as orphan branches: |--28.30%-- hrtimer_interrupt | smp_apic_timer_interrupt | apic_timer_interrupt | | <------------- here, no parent! | | | | | |--11.11%-- 0x7fae7bccb875 | | | | | |--11.11%-- 0xffffffffff60013b | | | | | |--11.11%-- __pthread_mutex_lock_internal | | | | | |--11.11%-- __errno_location Fix this by removing the context markers when we process the callchains to the tree. Reported-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1269274173-20328-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-23 00:09:33 +08:00
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
found = true;
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
callchain_cursor_advance(cursor);
}
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
/* matches not, relay no the parent */
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
if (!found) {
WARN_ONCE(!cmp, "Chain comparison error\n");
return cmp;
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
}
matches = cursor->pos - start;
/* we match only a part of the node. Split it and add the new chain */
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
if (matches < root->val_nr) {
split_add_child(root, cursor, cnode, start, matches, period);
return 0;
}
/* we match 100% of the path, increment the hit */
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
if (matches == root->val_nr && cursor->pos == cursor->nr) {
root->hit += period;
return 0;
}
/* We match the node and still have a part remaining */
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
append_chain_children(root, cursor, period);
return 0;
}
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
int callchain_append(struct callchain_root *root,
struct callchain_cursor *cursor,
u64 period)
{
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
if (!cursor->nr)
perf: Fix orphan callchain branches Callchains have markers inside their capture to tell we enter a context (kernel, user, ...). Those are not displayed in the callchains but they are incidentally an active part of the radix tree where callchains are stored, just like any other address. If we have the two following callchains: addr1 -> addr2 -> user context -> addr3 addr1 -> addr2 -> user context -> addr4 addr1 -> addr2 -> addr 5 This is pretty common if addr1 and addr2 are part of an interrupt path, addr3 and addr4 are user addresses and addr5 is a kernel non interrupt path. This will be stored as follows in the tree: addr1 addr2 / \ / addr5 user context / \ addr3 addr4 But we ignore the context markers in the report, hence the addr3 and addr4 will appear as orphan branches: |--28.30%-- hrtimer_interrupt | smp_apic_timer_interrupt | apic_timer_interrupt | | <------------- here, no parent! | | | | | |--11.11%-- 0x7fae7bccb875 | | | | | |--11.11%-- 0xffffffffff60013b | | | | | |--11.11%-- __pthread_mutex_lock_internal | | | | | |--11.11%-- __errno_location Fix this by removing the context markers when we process the callchains to the tree. Reported-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1269274173-20328-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-23 00:09:33 +08:00
return 0;
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
callchain_cursor_commit(cursor);
perf: Fix orphan callchain branches Callchains have markers inside their capture to tell we enter a context (kernel, user, ...). Those are not displayed in the callchains but they are incidentally an active part of the radix tree where callchains are stored, just like any other address. If we have the two following callchains: addr1 -> addr2 -> user context -> addr3 addr1 -> addr2 -> user context -> addr4 addr1 -> addr2 -> addr 5 This is pretty common if addr1 and addr2 are part of an interrupt path, addr3 and addr4 are user addresses and addr5 is a kernel non interrupt path. This will be stored as follows in the tree: addr1 addr2 / \ / addr5 user context / \ addr3 addr4 But we ignore the context markers in the report, hence the addr3 and addr4 will appear as orphan branches: |--28.30%-- hrtimer_interrupt | smp_apic_timer_interrupt | apic_timer_interrupt | | <------------- here, no parent! | | | | | |--11.11%-- 0x7fae7bccb875 | | | | | |--11.11%-- 0xffffffffff60013b | | | | | |--11.11%-- __pthread_mutex_lock_internal | | | | | |--11.11%-- __errno_location Fix this by removing the context markers when we process the callchains to the tree. Reported-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1269274173-20328-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-23 00:09:33 +08:00
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
append_chain_children(&root->node, cursor, period);
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
if (cursor->nr > root->max_depth)
root->max_depth = cursor->nr;
perf: Fix orphan callchain branches Callchains have markers inside their capture to tell we enter a context (kernel, user, ...). Those are not displayed in the callchains but they are incidentally an active part of the radix tree where callchains are stored, just like any other address. If we have the two following callchains: addr1 -> addr2 -> user context -> addr3 addr1 -> addr2 -> user context -> addr4 addr1 -> addr2 -> addr 5 This is pretty common if addr1 and addr2 are part of an interrupt path, addr3 and addr4 are user addresses and addr5 is a kernel non interrupt path. This will be stored as follows in the tree: addr1 addr2 / \ / addr5 user context / \ addr3 addr4 But we ignore the context markers in the report, hence the addr3 and addr4 will appear as orphan branches: |--28.30%-- hrtimer_interrupt | smp_apic_timer_interrupt | apic_timer_interrupt | | <------------- here, no parent! | | | | | |--11.11%-- 0x7fae7bccb875 | | | | | |--11.11%-- 0xffffffffff60013b | | | | | |--11.11%-- __pthread_mutex_lock_internal | | | | | |--11.11%-- __errno_location Fix this by removing the context markers when we process the callchains to the tree. Reported-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> LKML-Reference: <1269274173-20328-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-23 00:09:33 +08:00
return 0;
}
static int
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
merge_chain_branch(struct callchain_cursor *cursor,
struct callchain_node *dst, struct callchain_node *src)
{
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
struct callchain_cursor_node **old_last = cursor->last;
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
struct callchain_node *child;
struct callchain_list *list, *next_list;
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
struct rb_node *n;
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
int old_pos = cursor->nr;
int err = 0;
list_for_each_entry_safe(list, next_list, &src->val, list) {
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
callchain_cursor_append(cursor, list->ip,
list->ms.map, list->ms.sym);
list_del(&list->list);
free(list);
}
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
if (src->hit) {
callchain_cursor_commit(cursor);
append_chain_children(dst, cursor, src->hit);
}
perf callchain: Convert children list to rbtree Current collapse stage has a scalability problem which can be reproduced easily with a parallel kernel build. This is because it needs to traverse every children of callchains linearly during the collapse/merge stage. Converting it to a rbtree reduced the overhead significantly. On my 400MB perf.data file which recorded with make -j32 kernel build: $ time perf --no-pager report --stdio > /dev/null before: real 6m22.073s user 6m18.683s sys 0m0.706s after: real 0m20.780s user 0m19.962s sys 0m0.689s During the perf report the overhead on append_chain_children went down from 96.69% to 18.16%: - 18.16% perf perf [.] append_chain_children - append_chain_children - 77.48% append_chain_children + 69.79% merge_chain_branch - 22.96% append_chain_children + 67.44% merge_chain_branch + 30.15% append_chain_children + 2.41% callchain_append + 7.25% callchain_append + 12.26% callchain_append + 10.22% merge_chain_branch + 11.58% perf perf [.] dso__find_symbol + 8.02% perf perf [.] sort__comm_cmp + 5.48% perf libc-2.17.so [.] malloc_consolidate Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1381468543-25334-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-10-11 13:15:36 +08:00
n = rb_first(&src->rb_root_in);
while (n) {
child = container_of(n, struct callchain_node, rb_node_in);
n = rb_next(n);
rb_erase(&child->rb_node_in, &src->rb_root_in);
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
err = merge_chain_branch(cursor, dst, child);
if (err)
break;
free(child);
}
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
cursor->nr = old_pos;
cursor->last = old_last;
return err;
}
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
int callchain_merge(struct callchain_cursor *cursor,
struct callchain_root *dst, struct callchain_root *src)
{
return merge_chain_branch(cursor, &dst->node, &src->node);
}
int callchain_cursor_append(struct callchain_cursor *cursor,
u64 ip, struct map *map, struct symbol *sym)
{
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
struct callchain_cursor_node *node = *cursor->last;
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
if (!node) {
node = calloc(1, sizeof(*node));
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
if (!node)
return -ENOMEM;
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
*cursor->last = node;
}
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
node->ip = ip;
node->map = map;
node->sym = sym;
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
cursor->nr++;
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
cursor->last = &node->next;
return 0;
}
int sample__resolve_callchain(struct perf_sample *sample, struct symbol **parent,
struct perf_evsel *evsel, struct addr_location *al,
int max_stack)
{
if (sample->callchain == NULL)
return 0;
if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain ||
sort__has_parent) {
return thread__resolve_callchain(al->thread, evsel, sample,
parent, al, max_stack);
}
return 0;
}
int hist_entry__append_callchain(struct hist_entry *he, struct perf_sample *sample)
{
if (!symbol_conf.use_callchain || sample->callchain == NULL)
return 0;
return callchain_append(he->callchain, &callchain_cursor, sample->period);
}
int fill_callchain_info(struct addr_location *al, struct callchain_cursor_node *node,
bool hide_unresolved)
{
al->map = node->map;
al->sym = node->sym;
if (node->map)
al->addr = node->map->map_ip(node->map, node->ip);
else
al->addr = node->ip;
if (al->sym == NULL) {
if (hide_unresolved)
return 0;
if (al->map == NULL)
goto out;
}
if (al->map->groups == &al->machine->kmaps) {
if (machine__is_host(al->machine)) {
al->cpumode = PERF_RECORD_MISC_KERNEL;
al->level = 'k';
} else {
al->cpumode = PERF_RECORD_MISC_GUEST_KERNEL;
al->level = 'g';
}
} else {
if (machine__is_host(al->machine)) {
al->cpumode = PERF_RECORD_MISC_USER;
al->level = '.';
} else if (perf_guest) {
al->cpumode = PERF_RECORD_MISC_GUEST_USER;
al->level = 'u';
} else {
al->cpumode = PERF_RECORD_MISC_HYPERVISOR;
al->level = 'H';
}
}
out:
return 1;
}
char *callchain_list__sym_name(struct callchain_list *cl,
char *bf, size_t bfsize, bool show_dso)
{
int printed;
if (cl->ms.sym) {
if (callchain_param.key == CCKEY_ADDRESS &&
cl->ms.map && !cl->srcline)
cl->srcline = get_srcline(cl->ms.map->dso,
map__rip_2objdump(cl->ms.map,
cl->ip),
cl->ms.sym, false);
if (cl->srcline)
printed = scnprintf(bf, bfsize, "%s %s",
cl->ms.sym->name, cl->srcline);
else
printed = scnprintf(bf, bfsize, "%s", cl->ms.sym->name);
} else
printed = scnprintf(bf, bfsize, "%#" PRIx64, cl->ip);
if (show_dso)
scnprintf(bf + printed, bfsize - printed, " %s",
cl->ms.map ?
cl->ms.map->dso->short_name :
"unknown");
return bf;
}