perf top: Fix freeze on --call-graph flat/folded
Namhyung Kim
maintainerPATCH 1/3 perf top: Fix freeze on --call-graph flat/folded
The callchain rbtree is rebuilt periodically, so it needs to reinitialize the root everytime. Otherwise it can be stuck in the rbtree insertion with stale pointers.
--- tools/perf/util/callchain.c | 1 + 1 file changed, 1 insertion(+)
tools/perf/util/callchain.c +1 -0
@@ -290,6 +290,7 @@ static void sort_chain_flat(struct rb_root *rb_root, struct callchain_root *root, u64 min_hit, struct callchain_param *param __maybe_unused) { + *rb_root =RB_ROOT ;__sort_chain_flat (rb_root, &root->node, min_hit); }
Signature · 1 line
2.6.2
Namhyung Kim
maintainerPATCH 2/3 perf callchain: Stop resolving callchains after invalid address
Unwinding optimized binaries using frame pointer gives garbage. Check callchain address and stop if it's under vm.mmap_min_addr sysctl value.
Before:
$ perf report --stdio --no-children -g callee ...
1.37% perf [kernel.vmlinux] [k] smp_call_function_single
|
---smp_call_function_single
_perf_event_enable
perf_event_for_each_child
perf_ioctl
do_vfs_ioctl
sys_ioctl
entry_SYSCALL_64_fastpath
__GI___ioctl
0
0
0x1c5aa70
0x1c5b910
0x1c5aa70
0x1c5b910
0x1c5aa70
0x1c5b910
0x1c5aa70
0x1c5b910
0x1c5aa70
0x1c5b910
...After:
$ perf report --stdio --no-children -g callee ...
1.37% perf [kernel.vmlinux] [k] smp_call_function_single
|
---smp_call_function_single
_perf_event_enable
perf_event_for_each_child
perf_ioctl
do_vfs_ioctl
sys_ioctl
entry_SYSCALL_64_fastpath
__GI___ioctl$ perf report --stdio --no-children -g caller ...
1.37% perf [kernel.vmlinux] [k] smp_call_function_single
|
---__GI__ioctl
entry_SYSCALL_64_fastpath
sys_ioctl
do_vfs_ioctl
perf_ioctl
perf_event_for_each_child
_perf_event_enable
smp_call_function_single--- tools/perf/perf.c | 6 ++++++ tools/perf/util/machine.c | 13 +++++++++++++ tools/perf/util/util.c | 1 + tools/perf/util/util.h | 1 + 4 files changed, 21 insertions(+)
tools/perf/perf.c +6 -0
@@ -18,6 +18,7 @@#include "util/bpf-loader.h" #include "util/debug.h" #include <api/fs/tracing_path.h> +#include <api/fs/fs.h> #include <pthread.h> const char perf_usage_string[] =
@@ -528,11 +529,16 @@ int main(int argc, const char **argv) {const char *cmd;char sbuf[STRERR_BUFSIZE ]; +int min_addr;/* The page_size is placed in util object. */ page_size =sysconf (_SC_PAGE_SIZE); cacheline_size =sysconf (_SC_LEVEL1_DCACHE_LINESIZE); +if (sysctl__read_int ("vm/mmap_min_addr" , &min_addr) <0 ) + min_addr = page_size; + mmap_min_addr = min_addr; + cmd =perf_extract_argv0_path (argv[0 ]);if (!cmd) cmd ="perf-help" ;
tools/perf/util/machine.c +13 -0
@@ -1860,6 +1860,19 @@ static int thread__resolve_callchain_sample(struct thread *thread,#endif ip = chain->ips[j]; +/* +* Callchain value under mmap_min_addr means it's broken +* or the end of callchain. Stop. +*/ +if (ip < mmap_min_addr) { +if (callchain_param.order ==ORDER_CALLEE ) +break ; + +/* ignore current callchains for CALLER order */ +callchain_cursor_reset (&callchain_cursor); +continue ; + } + err =add_callchain_ip (thread, parent, root_al, &cpumode, ip);if (err)
tools/perf/util/util.c +1 -0
@@ -30,6 +30,7 @@ struct callchain_param callchain_param = {*/ unsigned int page_size;int cacheline_size; +unsigned long mmap_min_addr;bool test_attr__enabled;
tools/perf/util/util.h +1 -0
@@ -281,6 +281,7 @@ void sighandler_dump_stack(int sig);extern unsigned int page_size;extern int cacheline_size; +extern unsigned long mmap_min_addr;void get_term_dimensions (struct winsize *ws);void set_term_quiet_input (struct termios *old);
Signature · 1 line
2.6.2
Re: [PATCH 2/3] perf callchain: Stop resolving callchains after invalid address
Quoted 35 lines · Namhyung Kim
Unwinding optimized binaries using frame pointer gives garbage. Check callchain address and stop if it's under vm.mmap_min_addr sysctl value.
Before:
$ perf report --stdio --no-children -g callee ...1.37% perf [kernel.vmlinux] [k] smp_call_function_single | ---smp_call_function_single _perf_event_enable perf_event_for_each_child perf_ioctl do_vfs_ioctl sys_ioctl entry_SYSCALL_64_fastpath __GI___ioctl 0 0 0x1c5aa70 0x1c5b910 0x1c5aa70 0x1c5b910 0x1c5aa70 0x1c5b910 0x1c5aa70 0x1c5b910 0x1c5aa70 0x1c5b910 ...After:
$ perf report --stdio --no-children -g callee ...
Quoted 10 lines
1.37% perf [kernel.vmlinux] [k] smp_call_function_single
|
---smp_call_function_single
_perf_event_enable
perf_event_for_each_child
perf_ioctl
do_vfs_ioctl
sys_ioctl
entry_SYSCALL_64_fastpath
__GI___ioctlIn addition to that, would it make sense to terminate the callchain with an indicator that we found something anomalous? Such an extra line:
...
would not be intrusive, but would tell the informed reader that it's not a normal ending of the call chain.
This assumes that we can tell apart 'normal end of call chain' from 'seems to end with garbage poiner' cases - can do we that?
Thanks,
Ingo
Namhyung Kim
maintainerRe: [PATCH 2/3] perf callchain: Stop resolving callchains after invalid address
Hi Ingo,
Quoted 52 lines · Ingo Molnar
Quoted 35 lines · Namhyung Kim
Unwinding optimized binaries using frame pointer gives garbage. Check callchain address and stop if it's under vm.mmap_min_addr sysctl value.
Before:
$ perf report --stdio --no-children -g callee ...1.37% perf [kernel.vmlinux] [k] smp_call_function_single | ---smp_call_function_single _perf_event_enable perf_event_for_each_child perf_ioctl do_vfs_ioctl sys_ioctl entry_SYSCALL_64_fastpath __GI___ioctl 0 0 0x1c5aa70 0x1c5b910 0x1c5aa70 0x1c5b910 0x1c5aa70 0x1c5b910 0x1c5aa70 0x1c5b910 0x1c5aa70 0x1c5b910 ...After:
$ perf report --stdio --no-children -g callee ...Quoted 10 lines
1.37% perf [kernel.vmlinux] [k] smp_call_function_single | ---smp_call_function_single _perf_event_enable perf_event_for_each_child perf_ioctl do_vfs_ioctl sys_ioctl entry_SYSCALL_64_fastpath __GI___ioctlIn addition to that, would it make sense to terminate the callchain with an indicator that we found something anomalous? Such an extra line:
Quoted 7 lines
...would not be intrusive, but would tell the informed reader that it's not a normal ending of the call chain.
This assumes that we can tell apart 'normal end of call chain' from 'seems to end with garbage poiner' cases - can do we that?
In case of fp unwind, I'm not sure we can determine whether it's normal end or not especially for optimized binaries. It seems kernel also can stop callchain anytime if it sees a broken frame.
For dwarf unwind, I think it's also hard to tell since it can be stopped for various reasons like insufficient dump size or broken CFI,
...
Thanks,
Namhyung
Re: [PATCH 2/3] perf callchain: Stop resolving callchains after invalid address
Quoted 63 lines · Namhyung Kim
Hi Ingo,
Quoted 52 lines · Ingo Molnar
Quoted 35 lines · Namhyung Kim
Unwinding optimized binaries using frame pointer gives garbage. Check callchain address and stop if it's under vm.mmap_min_addr sysctl value.
Before:
$ perf report --stdio --no-children -g callee ...1.37% perf [kernel.vmlinux] [k] smp_call_function_single | ---smp_call_function_single _perf_event_enable perf_event_for_each_child perf_ioctl do_vfs_ioctl sys_ioctl entry_SYSCALL_64_fastpath __GI___ioctl 0 0 0x1c5aa70 0x1c5b910 0x1c5aa70 0x1c5b910 0x1c5aa70 0x1c5b910 0x1c5aa70 0x1c5b910 0x1c5aa70 0x1c5b910 ...After:
$ perf report --stdio --no-children -g callee ...Quoted 10 lines
1.37% perf [kernel.vmlinux] [k] smp_call_function_single | ---smp_call_function_single _perf_event_enable perf_event_for_each_child perf_ioctl do_vfs_ioctl sys_ioctl entry_SYSCALL_64_fastpath __GI___ioctlIn addition to that, would it make sense to terminate the callchain with an indicator that we found something anomalous? Such an extra line:
Quoted 7 lines
...would not be intrusive, but would tell the informed reader that it's not a normal ending of the call chain.
This assumes that we can tell apart 'normal end of call chain' from 'seems to end with garbage poiner' cases - can do we that?
Quoted 6 lines
In case of fp unwind, I'm not sure we can determine whether it's normal end or not especially for optimized binaries. It seems kernel also can stop callchain anytime if it sees a broken frame.
For dwarf unwind, I think it's also hard to tell since it can be stopped for various reasons like insufficient dump size or broken CFI,
But but. Doesn't your patch 'detect' an anomaly to begin with?
+ /*
+ * Callchain value under mmap_min_addr means it's broken + * or the end of callchain. Stop. + */
+ if (ip < mmap_min_addr) {
+ if (callchain_param.order == ORDER_CALLEE)
+ break;
all I'm asking for is to indicate it in some low-key visual fashion when we encounter such a 'broken' call-chain.
I presume the 'old' way of ending the call-chain was that 'ip' was zero, right? We should not print the indicator in that case.
Also, in the dwarf case I'd also see value in indicating if any of these events occured:
Quoted 2 lines
For dwarf unwind, I think it's also hard to tell since it can be stopped for various reasons like insufficient dump size or broken CFI,
even if we cannot catch all anomalies. Performance analysis must stand firm on a hard rock of reliability and dependability, and we should always propagate information about possible profiling data corruption/unreliability. That's why we print the 'IO overload' messages during perf record for example.
Even if the problem is not caused by perf, but by external factors such as the compiler/linker.
Thanks,
Ingo
Re: [PATCH 2/3] perf callchain: Stop resolving callchains after invalid address
In reply to Namhyung Kimtools/perf/perf.c int main(int argc, const char **argv) · line 538
Quoted 11 lines · Namhyung Kim
tools/perf/perf.c +2
-0@@ -528,11 +529,16 @@ int main(int argc, const char **argv) {const char *cmd;char sbuf[STRERR_BUFSIZE ]; +int min_addr;/* The page_size is placed in util object. */ page_size =sysconf (_SC_PAGE_SIZE); cacheline_size =sysconf (_SC_LEVEL1_DCACHE_LINESIZE); +if (sysctl__read_int ("vm/mmap_min_addr" , &min_addr) <0 )
This assumes the record and analysis are done on the same system.
David
Namhyung Kim
maintainerRe: [PATCH 2/3] perf callchain: Stop resolving callchains after invalid address
Hi David,
Quoted 7 lines · David Ahern
Quoted 6 lines · Namhyung Kim
tools/perf/perf.c +1
-0@@ -528,11 +529,16 @@ int main(int argc, const char **argv) {const char *cmd;char sbuf[STRERR_BUFSIZE ]; +int min_addr;
Quoted 7 lines
tools/perf/perf.c int main(int argc, const char **argv) · line 538
Quoted 5 lines
tools/perf/perf.c +1
-0/* The page_size is placed in util object. */ page_size =sysconf (_SC_PAGE_SIZE); cacheline_size =sysconf (_SC_LEVEL1_DCACHE_LINESIZE); +if (sysctl__read_int ("vm/mmap_min_addr" , &min_addr) <0 )This assumes the record and analysis are done on the same system.
Right. Maybe we can just use minimal size (or page size?) or save and pass it through somewhere in the feature bit?
Thanks,
Namhyung
Re: [PATCH 2/3] perf callchain: Stop resolving callchains after invalid address
Quoted 14 lines · Namhyung Kim
Hi David,
Quoted 11 lines · David Ahern
Quoted 10 lines · Namhyung Kim
tools/perf/perf.c +1
-0@@ -528,11 +529,16 @@ int main(int argc, const char **argv) {const char *cmd;char sbuf[STRERR_BUFSIZE ]; +int min_addr;/* The page_size is placed in util object. */ page_size =sysconf (_SC_PAGE_SIZE); cacheline_size =sysconf (_SC_LEVEL1_DCACHE_LINESIZE);
Quoted 6 lines
Quoted 3 lines
tools/perf/perf.c int main(int argc, const char **argv) · line 538
Quoted 1 line
tools/perf/perf.c +1
-0+if (sysctl__read_int ("vm/mmap_min_addr" , &min_addr) <0 )This assumes the record and analysis are done on the same system.
Right. Maybe we can just use minimal size (or page size?) or save and pass it through somewhere in the feature bit?
no preference, but it should work with cross arch analysis as well (e.g., record on arm/ppc and analysis on x86)
Re: [PATCH 2/3] perf callchain: Stop resolving callchains after invalid address
Quoted 19 lines · David Ahern
Quoted 18 lines · Namhyung Kim
Hi David,
Quoted 14 lines · David Ahern
tools/perf/perf.c int main(int argc, const char **argv) · line 538
Quoted 11 lines · Namhyung Kim
tools/perf/perf.c +2
-0@@ -528,11 +529,16 @@ int main(int argc, const char **argv) {const char *cmd;char sbuf[STRERR_BUFSIZE ]; +int min_addr;/* The page_size is placed in util object. */ page_size =sysconf (_SC_PAGE_SIZE); cacheline_size =sysconf (_SC_LEVEL1_DCACHE_LINESIZE); +if (sysctl__read_int ("vm/mmap_min_addr" , &min_addr) <0 )This assumes the record and analysis are done on the same system.
Quoted 5 lines
Quoted 2 lines
Right. Maybe we can just use minimal size (or page size?) or save and pass it through somewhere in the feature bit?
no preference, but it should work with cross arch analysis as well (e.g., record on arm/ppc and analysis on x86)
I think we should store it in perf.data in features, but seems like a waste to spend one bit just for this number.
I remember commenting on new CPU related FEATURE data, that would contain cpu specific data in extensible form like TAG,VALUE,TAG,VALUE..
but I think the design changed or something, because I cannot find it in now ;-)
maybe we could add something like that
jirka
Arnaldo Carvalho de Melo
maintainerRe: [PATCH 2/3] perf callchain: Stop resolving callchains after invalid address
Em Thu, Nov 26, 2015 at 04:58:57PM +0100, Jiri Olsa escreveu:
Quoted 12 lines · David Ahern
Quoted 11 lines · Namhyung Kim
Quoted 10 lines
Hi David,
Quoted 7 lines · David Ahern
Quoted 6 lines · Namhyung Kim
tools/perf/perf.c +1
-0@@ -528,11 +529,16 @@ int main(int argc, const char **argv) {const char *cmd;char sbuf[STRERR_BUFSIZE ]; +int min_addr;
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
tools/perf/perf.c int main(int argc, const char **argv) · line 538
Quoted 5 lines
tools/perf/perf.c +1
-0/* The page_size is placed in util object. */ page_size =sysconf (_SC_PAGE_SIZE); cacheline_size =sysconf (_SC_LEVEL1_DCACHE_LINESIZE); +if (sysctl__read_int ("vm/mmap_min_addr" , &min_addr) <0 )
Please put this in that symbol_conf kitchen sink :-)
I'm unsure though if there would be a reason for having both the local min_addr and the one at perf record time, i.e. from perf.data :-\
Quoted 11 lines
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
This assumes the record and analysis are done on the same system.
Right. Maybe we can just use minimal size (or page size?) or save and pass it through somewhere in the feature bit?
no preference, but it should work with cross arch analysis as well (e.g., record on arm/ppc and analysis on x86)
I think we should store it in perf.data in features, but seems like a waste to spend one bit just for this number.
Quoted 6 lines
I remember commenting on new CPU related FEATURE data, that would contain cpu specific data in extensible form like TAG,VALUE,TAG,VALUE..
but I think the design changed or something, because I cannot find it in now ;-)
maybe we could add something like that
Right, isn't that one part of the perf/stat patchkit?
- Arnaldo
Namhyung Kim
maintainerRe: [PATCH 2/3] perf callchain: Stop resolving callchains after invalid address
Quoted 17 lines · Arnaldo Carvalho de Melo
Em Thu, Nov 26, 2015 at 04:58:57PM +0100, Jiri Olsa escreveu:
Quoted 16 lines · David Ahern
Quoted 15 lines · Namhyung Kim
Quoted 14 lines
Hi David,
Quoted 11 lines · David Ahern
Quoted 10 lines · Namhyung Kim
tools/perf/perf.c +1
-0@@ -528,11 +529,16 @@ int main(int argc, const char **argv) {const char *cmd;char sbuf[STRERR_BUFSIZE ]; +int min_addr;/* The page_size is placed in util object. */ page_size =sysconf (_SC_PAGE_SIZE); cacheline_size =sysconf (_SC_LEVEL1_DCACHE_LINESIZE);
Quoted 6 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
tools/perf/perf.c int main(int argc, const char **argv) · line 538
Quoted 1 line
tools/perf/perf.c +1
-0+if (sysctl__read_int ("vm/mmap_min_addr" , &min_addr) <0 )Please put this in that
symbol_confkitchen sink :-)I'm unsure though if there would be a reason for having both the local
min_addrand the one at perf record time, i.e. from perf.data :-\
Otherwise, it can use different value and can treat valid addresses as invalid especially if the difference is huge.
Btw, I think it should be kernel to cut off these invalid callchains in the first place, no?
Quoted 12 lines
Quoted 12 lines
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
This assumes the record and analysis are done on the same system.
Right. Maybe we can just use minimal size (or page size?) or save and pass it through somewhere in the feature bit?
no preference, but it should work with cross arch analysis as well (e.g., record on arm/ppc and analysis on x86)
I think we should store it in perf.data in features, but seems like a waste to spend one bit just for this number.
Quoted 8 lines
Quoted 6 lines
I remember commenting on new CPU related FEATURE data, that would contain cpu specific data in extensible form like TAG,VALUE,TAG,VALUE..
but I think the design changed or something, because I cannot find it in now ;-)
maybe we could add something like that
Right, isn't that one part of the perf/stat patchkit?
I'll take a look. Jiri, could you give the link if you have it?
Thanks,
Namhyung
Namhyung Kim
maintainerPATCH 3/3 perf callchain: Honor hide_unresolved
In reply to Namhyung KimIf user requested to hide unresolved entries, skip unresolved callchains as well as hist entries.
--- tools/perf/builtin-report.c | 7 +++---- tools/perf/util/machine.c | 5 +++++ tools/perf/util/symbol.h | 3 ++- 3 files changed, 10 insertions(+), 5 deletions(-)
tools/perf/builtin-report.c +3 -4
@@ -45,7 +45,6 @@ struct report {struct perf_tool tool;struct perf_session *session;bool use_tui, use_gtk, use_stdio; -bool hide_unresolved;bool dont_use_callchains;bool show_full_info;bool show_threads;
@@ -146,7 +145,7 @@ static int process_sample_event(struct perf_tool *tool,struct hist_entry_iter iter = { .evsel = evsel, .sample = sample, - .hide_unresolved = rep->hide_unresolved, + .hide_unresolved = symbol_conf.hide_unresolved, .add_entry_cb = hist_iter__report_callback, };int ret =0 ;
@@ -157,7 +156,7 @@ static int process_sample_event(struct perf_tool *tool,return -1 ; } -if (rep->hide_unresolved && al.sym ==NULL ) +if (symbol_conf.hide_unresolved && al.sym ==NULL ) goto out_put;if (rep->cpu_list && !test_bit (sample->cpu, rep->cpu_bitmap))
@@ -740,7 +739,7 @@ int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)OPT_STRING_NOEMPTY ('t' ,"field-separator" , &symbol_conf.field_sep,"separator" ,"separator for columns, no spaces will be added between " "columns '.' is reserved." ), -OPT_BOOLEAN ('U' ,"hide-unresolved" , &report.hide_unresolved, +OPT_BOOLEAN ('U' ,"hide-unresolved" , &symbol_conf.hide_unresolved,"Only display entries resolved to a symbol" ),OPT_STRING (0 ,"symfs" , &symbol_conf.symfs,"directory" , "Look for files with symbols relative to this directory"),
tools/perf/util/machine.c +5 -0
@@ -1618,6 +1618,8 @@ static int add_callchain_ip(struct thread *thread, } } +if (symbol_conf.hide_unresolved && al.sym ==NULL ) +return 0 ;return callchain_cursor_append (&callchain_cursor, al.addr, al.map, al.sym); }
@@ -1885,6 +1887,9 @@ static int thread__resolve_callchain_sample(struct thread *thread,static int unwind_entry (struct unwind_entry *entry,void *arg) {struct callchain_cursor *cursor = arg; + +if (symbol_conf.hide_unresolved && entry->sym ==NULL ) +return 0 ;return callchain_cursor_append (cursor, entry->ip, entry->map, entry->sym); }
tools/perf/util/symbol.h +2 -1
@@ -108,7 +108,8 @@ struct symbol_conf { show_hist_headers, branch_callstack, has_filter, - show_ref_callgraph; + show_ref_callgraph, + hide_unresolved;const char *vmlinux_name, *kallsyms_name, *source_prefix,
Signature · 1 line
2.6.2
Re: [PATCH 3/3] perf callchain: Honor hide_unresolved
Quoted 4 lines · Namhyung Kim
If user requested to hide unresolved entries, skip unresolved callchains as well as hist entries.
Signed-off-by Namhyung Kim <[email protected]>
thanks,
jirka
Re: [PATCH 3/3] perf callchain: Honor hide_unresolved
Em Thu, Nov 26, 2015 at 09:46:20AM +0100, Jiri Olsa escreveu:
Quoted 7 lines · Namhyung Kim
Quoted 4 lines
If user requested to hide unresolved entries, skip unresolved callchains as well as hist entries.
Signed-off-by Namhyung Kim <[email protected]>Acked-by Jiri Olsa <[email protected]>
Applied.
- Arnaldo
tip-bot for Namhyung Kim
merge bot[tip:perf/core] perf callchain: Honor hide_unresolved
In reply to Namhyung Kimperf callchain: Honor hide_unresolved
Commit-ID: b49a8fe52626814968b9a9d27d7ad1cadc5532ed Gitweb: http://git.kernel.org/tip/b49a8fe52626814968b9a9d27d7ad1cadc5532ed Author: Namhyung Kim <[email protected]> AuthorDate: Thu, 26 Nov 2015 16:08:20 +0900 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Thu, 26 Nov 2015 13:19:39 -0300
perf callchain: Honor hide_unresolved
If user requested to hide unresolved entries, skip unresolved callchains as well as hist entries.
--- tools/perf/builtin-report.c | 7 +++---- tools/perf/util/machine.c | 5 +++++ tools/perf/util/symbol.h | 3 ++- 3 files changed, 10 insertions(+), 5 deletions(-)
tools/perf/builtin-report.c +3 -4
@@ -45,7 +45,6 @@ struct report {struct perf_tool tool;struct perf_session *session;bool use_tui, use_gtk, use_stdio; -bool hide_unresolved;bool dont_use_callchains;bool show_full_info;bool show_threads;
@@ -146,7 +145,7 @@ static int process_sample_event(struct perf_tool *tool,struct hist_entry_iter iter = { .evsel = evsel, .sample = sample, - .hide_unresolved = rep->hide_unresolved, + .hide_unresolved = symbol_conf.hide_unresolved, .add_entry_cb = hist_iter__report_callback, };int ret =0 ;
@@ -157,7 +156,7 @@ static int process_sample_event(struct perf_tool *tool,return -1 ; } -if (rep->hide_unresolved && al.sym ==NULL ) +if (symbol_conf.hide_unresolved && al.sym ==NULL ) goto out_put;if (rep->cpu_list && !test_bit (sample->cpu, rep->cpu_bitmap))
@@ -740,7 +739,7 @@ int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)OPT_STRING_NOEMPTY ('t' ,"field-separator" , &symbol_conf.field_sep,"separator" ,"separator for columns, no spaces will be added between " "columns '.' is reserved." ), -OPT_BOOLEAN ('U' ,"hide-unresolved" , &report.hide_unresolved, +OPT_BOOLEAN ('U' ,"hide-unresolved" , &symbol_conf.hide_unresolved,"Only display entries resolved to a symbol" ),OPT_STRING (0 ,"symfs" , &symbol_conf.symfs,"directory" , "Look for files with symbols relative to this directory"),
tools/perf/util/machine.c +5 -0
@@ -1618,6 +1618,8 @@ static int add_callchain_ip(struct thread *thread, } } +if (symbol_conf.hide_unresolved && al.sym ==NULL ) +return 0 ;return callchain_cursor_append (&callchain_cursor, al.addr, al.map, al.sym); }
@@ -1872,6 +1874,9 @@ check_calls:static int unwind_entry (struct unwind_entry *entry,void *arg) {struct callchain_cursor *cursor = arg; + +if (symbol_conf.hide_unresolved && entry->sym ==NULL ) +return 0 ;return callchain_cursor_append (cursor, entry->ip, entry->map, entry->sym); }
tools/perf/util/symbol.h +2 -1
@@ -108,7 +108,8 @@ struct symbol_conf { show_hist_headers, branch_callstack, has_filter, - show_ref_callgraph; + show_ref_callgraph, + hide_unresolved;const char *vmlinux_name, *kallsyms_name, *source_prefix,
Quoted 14 lines · Namhyung Kim
The callchain rbtree is rebuilt periodically, so it needs to reinitialize the root everytime. Otherwise it can be stuck in the rbtree insertion with stale pointers.
Signed-off-by Namhyung Kim <[email protected]>--- tools/perf/util/callchain.c | 1 + 1 file changed, 1 insertion(+)tools/perf/util/callchain.c +0
-0
tools/perf/util/callchain.c static void · line 293
Quoted 4 lines
tools/perf/util/callchain.c +1
-0@@ -290,6 +290,7 @@ static void sort_chain_flat(struct rb_root *rb_root, struct callchain_root *root, u64 min_hit, struct callchain_param *param __maybe_unused) { + *rb_root =RB_ROOT ;
it seems ok, but I did not find how this could be called twice?
the only sort I can see is done within:
__hists__insert_output_entry
do we allow resorting of the callchains?
thanks,
jirka
Quoted 6 lines
tools/perf/util/callchain.c +0
-0__sort_chain_flat (rb_root, &root->node, min_hit); }Signature · 1 line
2.6.2
Namhyung Kim
maintainerHi Jiri,
Quoted 22 lines · Jiri Olsa
Quoted 14 lines · Namhyung Kim
The callchain rbtree is rebuilt periodically, so it needs to reinitialize the root everytime. Otherwise it can be stuck in the rbtree insertion with stale pointers.
Signed-off-by Namhyung Kim <[email protected]>--- tools/perf/util/callchain.c | 1 + 1 file changed, 1 insertion(+)tools/perf/util/callchain.c +0
-0tools/perf/util/callchain.c static void · line 293
Quoted 4 lines
tools/perf/util/callchain.c +1
-0@@ -290,6 +290,7 @@ static void sort_chain_flat(struct rb_root *rb_root, struct callchain_root *root, u64 min_hit, struct callchain_param *param __maybe_unused) { + *rb_root =RB_ROOT ;it seems ok, but I did not find how this could be called twice?
Quoted 4 lines
the only sort I can see is done within:
__hists__insert_output_entrydo we allow resorting of the callchains?
No, but I think it's possible though.
It's called from perf top's display thread.
display_thread()
-> while (!done)
-> perf_top__print_sym_table()
-> hists__output_resort()
-> __hists__insert_output_entry()Thanks,
Namhyung
Quoted 25 lines · Namhyung Kim
Hi Jiri,
Quoted 22 lines · Jiri Olsa
Quoted 14 lines · Namhyung Kim
The callchain rbtree is rebuilt periodically, so it needs to reinitialize the root everytime. Otherwise it can be stuck in the rbtree insertion with stale pointers.
Signed-off-by Namhyung Kim <[email protected]>--- tools/perf/util/callchain.c | 1 + 1 file changed, 1 insertion(+)tools/perf/util/callchain.c +0
-0tools/perf/util/callchain.c static void · line 293
Quoted 4 lines
tools/perf/util/callchain.c +1
-0@@ -290,6 +290,7 @@ static void sort_chain_flat(struct rb_root *rb_root, struct callchain_root *root, u64 min_hit, struct callchain_param *param __maybe_unused) { + *rb_root =RB_ROOT ;it seems ok, but I did not find how this could be called twice?
Quoted 8 lines
Quoted 4 lines
the only sort I can see is done within:
__hists__insert_output_entrydo we allow resorting of the callchains?
No, but I think it's possible though.
It's called from perf top's display thread.
ah right.. top ;-) ok
isn't there analogical issue with the other sorts? graph_abs, graph_rel... I dont see that code doing this
thanks,
jirka
Quoted 9 lines
display_thread()
-> while (!done)-> perf_top__print_sym_table() -> hists__output_resort() -> __hists__insert_output_entry()Thanks,
Namhyung
Namhyung Kim
maintainerQuoted 31 lines · Jiri Olsa
Quoted 30 lines · Namhyung Kim
Hi Jiri,
Quoted 26 lines · Jiri Olsa
Quoted 14 lines · Namhyung Kim
The callchain rbtree is rebuilt periodically, so it needs to reinitialize the root everytime. Otherwise it can be stuck in the rbtree insertion with stale pointers.
Signed-off-by Namhyung Kim <[email protected]>--- tools/perf/util/callchain.c | 1 + 1 file changed, 1 insertion(+)tools/perf/util/callchain.c +0
-0tools/perf/util/callchain.c static void · line 293
Quoted 4 lines
tools/perf/util/callchain.c +1
-0@@ -290,6 +290,7 @@ static void sort_chain_flat(struct rb_root *rb_root, struct callchain_root *root, u64 min_hit, struct callchain_param *param __maybe_unused) { + *rb_root =RB_ROOT ;it seems ok, but I did not find how this could be called twice?
the only sort I can see is done within:
__hists__insert_output_entrydo we allow resorting of the callchains?
Quoted 8 lines
Quoted 3 lines
No, but I think it's possible though.
It's called from perf top's display thread.
ah right.. top ;-) ok
isn't there analogical issue with the other sorts?
graph_abs,graph_rel... I dont see that code doing this
The sort_chain_graph_abs/rel() already do this. Unlike flat/folded callchains, they put child into each callchain_node's rbtree. So they reinitialize node->rb_root in __sort_chain_graph_abs/rel() and reset rb_root after finishing sort.
Thanks,
Namhyung
Jiri Olsa
Acked-byQuoted 38 lines · Namhyung Kim
Quoted 37 lines · Jiri Olsa
Quoted 25 lines · Namhyung Kim
Hi Jiri,
Quoted 22 lines · Jiri Olsa
Quoted 14 lines · Namhyung Kim
The callchain rbtree is rebuilt periodically, so it needs to reinitialize the root everytime. Otherwise it can be stuck in the rbtree insertion with stale pointers.
Signed-off-by Namhyung Kim <[email protected]>--- tools/perf/util/callchain.c | 1 + 1 file changed, 1 insertion(+)tools/perf/util/callchain.c +0
-0tools/perf/util/callchain.c static void · line 293
Quoted 4 lines
tools/perf/util/callchain.c +1
-0@@ -290,6 +290,7 @@ static void sort_chain_flat(struct rb_root *rb_root, struct callchain_root *root, u64 min_hit, struct callchain_param *param __maybe_unused) { + *rb_root =RB_ROOT ;it seems ok, but I did not find how this could be called twice?
Quoted 8 lines
Quoted 4 lines
the only sort I can see is done within:
__hists__insert_output_entrydo we allow resorting of the callchains?
No, but I think it's possible though.
It's called from perf top's display thread.
ah right.. top ;-) ok
Quoted 7 lines
Quoted 2 lines
isn't there analogical issue with the other sorts?
graph_abs,graph_rel... I dont see that code doing thisThe sort_chain_graph_abs/rel() already do this. Unlike flat/folded callchains, they put child into each callchain_node's rbtree. So they reinitialize
node->rb_rootin __sort_chain_graph_abs/rel() and resetrb_rootafter finishing sort.
ook, thanks for explanation
jirka
Em Thu, Nov 26, 2015 at 04:08:18PM +0900, Namhyung Kim escreveu:
Quoted 5 lines
The callchain rbtree is rebuilt periodically, so it needs to reinitialize the root everytime. Otherwise it can be stuck in the rbtree insertion with stale pointers.
Signed-off-by Namhyung Kim <[email protected]>
Applied.
- Arnaldo
tip-bot for Namhyung Kim
merge botperf top: Fix freeze on --call-graph flat/folded
Commit-ID: 0356218a68551f051998f4fb5074a1eed7a346fe Gitweb: http://git.kernel.org/tip/0356218a68551f051998f4fb5074a1eed7a346fe Author: Namhyung Kim <[email protected]> AuthorDate: Thu, 26 Nov 2015 16:08:18 +0900 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Thu, 26 Nov 2015 13:32:08 -0300
perf top: Fix freeze on --call-graph flat/folded
The callchain rbtree is rebuilt periodically, so it needs to reinitialize the root everytime. Otherwise it can be stuck in the rbtree insertion with stale pointers.
--- tools/perf/util/callchain.c | 1 + 1 file changed, 1 insertion(+)
tools/perf/util/callchain.c +1 -0
@@ -290,6 +290,7 @@ static void sort_chain_flat(struct rb_root *rb_root, struct callchain_root *root, u64 min_hit, struct callchain_param *param __maybe_unused) { + *rb_root =RB_ROOT ;__sort_chain_flat (rb_root, &root->node, min_hit); }