perf thread-stack: Fix thread stack return from kernel for kernel-only case
Adrian Hunter
reviewercover perf thread-stack: Fix thread stack return from kernel for kernel-only case
Hi
Here is one non-urgent fix and a subsequent tidy-up.
Adrian Hunter (2):
perf thread-stack: Fix thread stack return from kernel for kernel-only case perf thread-stack: Eliminate code duplicating thread_stack__pop_ks()
tools/perf/util/thread-stack.c | 48 ++++++++++++++++++++++++++++++------------ 1 file changed, 35 insertions(+), 13 deletions(-)
Regards
Adrian
Adrian Hunter
reviewerPATCH 1/2 perf thread-stack: Fix thread stack return from kernel for kernel-only case
Commit f08046cb3082 ("perf thread-stack: Represent jmps to the start of a different symbol") had the side-effect of introducing more stack entries before return from kernel space. When user space is also traced, those entries are popped before entry to user space, but when user space is not traced, they get stuck at the bottom of the stack, making the stack grow progressively larger. Fix by detecting a return-from-kernel branch type, and popping kernel addresses from the stack then.
Note, the problem and fix affect the exported Call Graph / Tree but not the callindent option used by "perf script --call-trace".
Example:
perf-with-kcore record example -e intel_pt//k -- ls perf-with-kcore script --itrace=bep -s ~/libexec/perf-core/scripts/python/export-to-sqlite.py example.db branches calls ~/libexec/perf-core/scripts/python/exported-sql-viewer.py example.db
Menu option: Reports -> Context-Sensitive Call Graph
Before: (showing Call Path column only)
Call Path
▶ perf
▼ ls
▼ 12111:12111
▶ setup_new_exec
▶ __task_pid_nr_ns
▶ perf_event_pid_type
▶ perf_event_comm_output
▶ perf_iterate_ctx
▶ perf_iterate_sb
▶ perf_event_comm
▶ __set_task_comm
▶ load_elf_binary
▶ search_binary_handler
▶ __do_execve_file.isra.41
▶ __x64_sys_execve
▶ do_syscall_64
▼ entry_SYSCALL_64_after_hwframe
▼ swapgs_restore_regs_and_return_to_usermode
▼ native_iret
▶ error_entry
▶ do_page_fault
▼ error_exit
▼ retint_user
▶ prepare_exit_to_usermode
▼ native_iret
▶ error_entry
▶ do_page_fault
▼ error_exit
▼ retint_user
▶ prepare_exit_to_usermode
▼ native_iret
▶ error_entry
▶ do_page_fault
▼ error_exit
▼ retint_user
▶ prepare_exit_to_usermode
▶ native_iretAfter: (showing Call Path column only)
Call Path
▶ perf
▼ ls
▼ 12111:12111
▶ setup_new_exec
▶ __task_pid_nr_ns
▶ perf_event_pid_type
▶ perf_event_comm_output
▶ perf_iterate_ctx
▶ perf_iterate_sb
▶ perf_event_comm
▶ __set_task_comm
▶ load_elf_binary
▶ search_binary_handler
▶ __do_execve_file.isra.41
▶ __x64_sys_execve
▶ do_syscall_64
▶ entry_SYSCALL_64_after_hwframe
▶ page_fault
▼ entry_SYSCALL_64
▼ do_syscall_64
▶ __x64_sys_brk
▶ __x64_sys_access
▶ __x64_sys_openat
▶ __x64_sys_newfstat
▶ __x64_sys_mmap
▶ __x64_sys_close
▶ __x64_sys_read
▶ __x64_sys_mprotect
▶ __x64_sys_arch_prctl
▶ __x64_sys_munmap
▶ exit_to_usermode_loop
▶ __x64_sys_set_tid_address
▶ __x64_sys_set_robust_list
▶ __x64_sys_rt_sigaction
▶ __x64_sys_rt_sigprocmask
▶ __x64_sys_prlimit64
▶ __x64_sys_statfs
▶ __x64_sys_ioctl
▶ __x64_sys_getdents64
▶ __x64_sys_write
▶ __x64_sys_exit_group--- tools/perf/util/thread-stack.c | 30 +++++++++++++++++++++++++++++- 1 file changed, 29 insertions(+), 1 deletion(-)
tools/perf/util/thread-stack.c +29 -1
@@ -637,6 +637,23 @@ static int thread_stack__bottom(struct thread_stack *ts, true, false); } +static int thread_stack__pop_ks (struct thread *thread,struct thread_stack *ts, +struct perf_sample *sample,u64 ref) +{ +u64 tm = sample->time; +int err; + +/* Return to userspace, so pop all kernel addresses */ +while (thread_stack__in_kernel (ts)) { + err =thread_stack__call_return (thread, ts, --ts->cnt, + tm, ref, true); +if (err) +return err; + } + +return 0 ; +} + static int thread_stack__no_call_return(struct thread *thread, struct thread_stack *ts, struct perf_sample *sample,
@@ -919,7 +936,18 @@ int thread_stack__process(struct thread *thread, struct comm *comm, ts->rstate =X86_RETPOLINE_DETECTED ; }else if (sample->flags &PERF_IP_FLAG_RETURN ) { -if (!sample->ip || !sample->addr) +if (!sample->addr) { +u32 return_from_kernel =PERF_IP_FLAG_SYSCALLRET | +PERF_IP_FLAG_INTERRUPT ; + +if (!(sample->flags & return_from_kernel)) +return 0 ; + +/* Pop kernel stack */ +return thread_stack__pop_ks (thread, ts, sample, ref); + } + +if (!sample->ip)return 0 ;/* x86 retpoline 'return' doesn't match the stack */
Signature · 1 line
2.17.1
Em Wed, Jun 19, 2019 at 09:44:28AM +0300, Adrian Hunter escreveu:
Quoted 8 lines
Commit
f08046cb3082("perf thread-stack: Represent jmps to the start of a different symbol") had the side-effect of introducing more stack entries before return from kernel space. When user space is also traced, those entries are popped before entry to user space, but when user space is not traced, they get stuck at the bottom of the stack, making the stack grow progressively larger. Fix by detecting a return-from-kernel branch type, and popping kernel addresses from the stack then.
Quoted 8 lines
Note, the problem and fix affect the exported Call Graph / Tree but not the callindent option used by "perf script --call-trace".
Example:
perf-with-kcore record example -e intel_pt//k -- ls perf-with-kcore script --itrace=bep -s ~/libexec/perf-core/scripts/python/export-to-sqlite.py example.db branches calls ~/libexec/perf-core/scripts/python/exported-sql-viewer.py example.db
Trying with you above instructions I hit some snags, 'perf-with-kcore' expects the first arg to be the directory where it'll find the perf.data file, etc, right? I.e. to work I had to do:
[root@quaco tmp]# rm -rf bep
[root@quaco tmp]# rm -rf example.db
[root@quaco tmp]# perf-with-kcore record bep -e intel_pt//k -- ls
Recording
Using /root/bin/perf
perf version 5.2.rc4.gd1d5628fa057/root/bin/perf record -o bep/perf.data -e intel_pt//k -- ls
<SNIP>
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.123 MB bep/perf.data ]
Copying kcore
Done
[root@quaco tmp]# perf-with-kcore script bep --itrace=bep -s ~acme/libexec/perf-core/scripts/python/export-to-sqlite.py example.db branches calls Using /root/bin/perf
perf version 5.2.rc4.gd1d5628fa057/root/bin/perf script -i bep/perf.data --kallsyms=bep/kcore_dir/kallsyms --itrace=bep -s /home/acme/libexec/perf-core/scripts/python/export-to-sqlite.py example.db branches calls
2019-06-24 15:29:54.910209 Creating database ... 2019-06-24 15:29:54.913351 Writing records... 2019-06-24 15:29:58.009226 Adding indexes 2019-06-24 15:29:58.024644 Done
[root@quaco tmp]#
I.e. add 'bep' after 'script' and before '--itrace-bep'.
I'm fixing this up in the comment, please check.
With that in place I can reproduce your results below.
Thanks,
- Arnaldo
Quoted 34 lines
Menu option: Reports -> Context-Sensitive Call Graph
Before: (showing Call Path column only)
Call Path ▶ perf ▼ ls ▼ 12111:12111 ▶ setup_new_exec ▶ __task_pid_nr_ns ▶ perf_event_pid_type ▶ perf_event_comm_output ▶ perf_iterate_ctx ▶ perf_iterate_sb ▶ perf_event_comm ▶ __set_task_comm ▶ load_elf_binary ▶ search_binary_handler ▶ __do_execve_file.isra.41 ▶ __x64_sys_execve ▶ do_syscall_64 ▼ entry_SYSCALL_64_after_hwframe ▼ swapgs_restore_regs_and_return_to_usermode ▼ native_iret ▶ error_entry ▶ do_page_fault ▼ error_exit ▼ retint_user ▶ prepare_exit_to_usermode ▼ native_iret ▶ error_entry ▶ do_page_fault ▼ error_exit ▼ retint_user
Quoted 8 lines
▶ prepare_exit_to_usermode
▼ native_iret
▶ error_entry
▶ do_page_fault
▼ error_exit
▼ retint_user
▶ prepare_exit_to_usermode
▶ native_iretT
Quoted 102 lines
After: (showing Call Path column only)
Call Path ▶ perf ▼ ls ▼ 12111:12111 ▶ setup_new_exec ▶ __task_pid_nr_ns ▶ perf_event_pid_type ▶ perf_event_comm_output ▶ perf_iterate_ctx ▶ perf_iterate_sb ▶ perf_event_comm ▶ __set_task_comm ▶ load_elf_binary ▶ search_binary_handler ▶ __do_execve_file.isra.41 ▶ __x64_sys_execve ▶ do_syscall_64 ▶ entry_SYSCALL_64_after_hwframe ▶ page_fault ▼ entry_SYSCALL_64 ▼ do_syscall_64 ▶ __x64_sys_brk ▶ __x64_sys_access ▶ __x64_sys_openat ▶ __x64_sys_newfstat ▶ __x64_sys_mmap ▶ __x64_sys_close ▶ __x64_sys_read ▶ __x64_sys_mprotect ▶ __x64_sys_arch_prctl ▶ __x64_sys_munmap ▶ exit_to_usermode_loop ▶ __x64_sys_set_tid_address ▶ __x64_sys_set_robust_list ▶ __x64_sys_rt_sigaction ▶ __x64_sys_rt_sigprocmask ▶ __x64_sys_prlimit64 ▶ __x64_sys_statfs ▶ __x64_sys_ioctl ▶ __x64_sys_getdents64 ▶ __x64_sys_write ▶ __x64_sys_exit_groupSigned-off-by Adrian Hunter <[email protected]>Fixes f08046cb3082 ("perf thread-stack: Represent jmps to the start of a different symbol")--- tools/perf/util/thread-stack.c | 30 +++++++++++++++++++++++++++++- 1 file changed, 29 insertions(+), 1 deletion(-)tools/perf/util/thread-stack.c +29
-1@@ -637,6 +637,23 @@ static int thread_stack__bottom(struct thread_stack *ts, true, false); } +static int thread_stack__pop_ks (struct thread *thread,struct thread_stack *ts, +struct perf_sample *sample,u64 ref) +{ +u64 tm = sample->time; +int err; + +/* Return to userspace, so pop all kernel addresses */ +while (thread_stack__in_kernel (ts)) { + err =thread_stack__call_return (thread, ts, --ts->cnt, + tm, ref, true); +if (err) +return err; + } + +return 0 ; +} + static int thread_stack__no_call_return(struct thread *thread, struct thread_stack *ts, struct perf_sample *sample,@@ -919,7 +936,18 @@ int thread_stack__process(struct thread *thread, struct comm *comm, ts->rstate =X86_RETPOLINE_DETECTED ; }else if (sample->flags &PERF_IP_FLAG_RETURN ) { -if (!sample->ip || !sample->addr) +if (!sample->addr) { +u32 return_from_kernel =PERF_IP_FLAG_SYSCALLRET | +PERF_IP_FLAG_INTERRUPT ; + +if (!(sample->flags & return_from_kernel)) +return 0 ; + +/* Pop kernel stack */ +return thread_stack__pop_ks (thread, ts, sample, ref); + } + +if (!sample->ip)return 0 ;/* x86 retpoline 'return' doesn't match the stack */ Signature · 1 line
2.17.1
Signature · 1 line
- Arnaldo
tip-bot for Adrian Hunter
merge botperf thread-stack: Fix thread stack return from kernel for kernel-only case
Commit-ID: 97860b483c5597663a174ff7405be957b4838391 Gitweb: https://git.kernel.org/tip/97860b483c5597663a174ff7405be957b4838391 Author: Adrian Hunter <[email protected]> AuthorDate: Wed, 19 Jun 2019 09:44:28 +0300 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Tue, 25 Jun 2019 08:47:10 -0300
perf thread-stack: Fix thread stack return from kernel for kernel-only case
Commit f08046cb3082 ("perf thread-stack: Represent jmps to the start of a different symbol") had the side-effect of introducing more stack entries before return from kernel space.
When user space is also traced, those entries are popped before entry to user space, but when user space is not traced, they get stuck at the bottom of the stack, making the stack grow progressively larger.
Fix by detecting a return-from-kernel branch type, and popping kernel addresses from the stack then.
Note, the problem and fix affect the exported Call Graph / Tree but not the callindent option used by "perf script --call-trace".
Example:
perf-with-kcore record example -e intel_pt//k -- ls perf-with-kcore script example --itrace=bep -s ~/libexec/perf-core/scripts/python/export-to-sqlite.py example.db branches calls ~/libexec/perf-core/scripts/python/exported-sql-viewer.py example.db
Menu option: Reports -> Context-Sensitive Call Graph
Before: (showing Call Path column only)
Call Path
▶ perf
▼ ls
▼ 12111:12111
▶ setup_new_exec
▶ __task_pid_nr_ns
▶ perf_event_pid_type
▶ perf_event_comm_output
▶ perf_iterate_ctx
▶ perf_iterate_sb
▶ perf_event_comm
▶ __set_task_comm
▶ load_elf_binary
▶ search_binary_handler
▶ __do_execve_file.isra.41
▶ __x64_sys_execve
▶ do_syscall_64
▼ entry_SYSCALL_64_after_hwframe
▼ swapgs_restore_regs_and_return_to_usermode
▼ native_iret
▶ error_entry
▶ do_page_fault
▼ error_exit
▼ retint_user
▶ prepare_exit_to_usermode
▼ native_iret
▶ error_entry
▶ do_page_fault
▼ error_exit
▼ retint_user
▶ prepare_exit_to_usermode
▼ native_iret
▶ error_entry
▶ do_page_fault
▼ error_exit
▼ retint_user
▶ prepare_exit_to_usermode
▶ native_iretAfter: (showing Call Path column only)
Call Path
▶ perf
▼ ls
▼ 12111:12111
▶ setup_new_exec
▶ __task_pid_nr_ns
▶ perf_event_pid_type
▶ perf_event_comm_output
▶ perf_iterate_ctx
▶ perf_iterate_sb
▶ perf_event_comm
▶ __set_task_comm
▶ load_elf_binary
▶ search_binary_handler
▶ __do_execve_file.isra.41
▶ __x64_sys_execve
▶ do_syscall_64
▶ entry_SYSCALL_64_after_hwframe
▶ page_fault
▼ entry_SYSCALL_64
▼ do_syscall_64
▶ __x64_sys_brk
▶ __x64_sys_access
▶ __x64_sys_openat
▶ __x64_sys_newfstat
▶ __x64_sys_mmap
▶ __x64_sys_close
▶ __x64_sys_read
▶ __x64_sys_mprotect
▶ __x64_sys_arch_prctl
▶ __x64_sys_munmap
▶ exit_to_usermode_loop
▶ __x64_sys_set_tid_address
▶ __x64_sys_set_robust_list
▶ __x64_sys_rt_sigaction
▶ __x64_sys_rt_sigprocmask
▶ __x64_sys_prlimit64
▶ __x64_sys_statfs
▶ __x64_sys_ioctl
▶ __x64_sys_getdents64
▶ __x64_sys_write
▶ __x64_sys_exit_groupCommitter notes:
The first arg to the perf-with-kcore needs to be the same for the 'record' and 'script' lines, otherwise we'll record the perf.data file and kcore_dir/ files in one directory ('example') to then try to use it from the 'bep' directory, fix the instructions above it so that both use 'example'.
--- tools/perf/util/thread-stack.c | 30 +++++++++++++++++++++++++++++- 1 file changed, 29 insertions(+), 1 deletion(-)
tools/perf/util/thread-stack.c +29 -1
@@ -628,6 +628,23 @@ static int thread_stack__bottom(struct thread_stack *ts, true, false); } +static int thread_stack__pop_ks (struct thread *thread,struct thread_stack *ts, +struct perf_sample *sample,u64 ref) +{ +u64 tm = sample->time; +int err; + +/* Return to userspace, so pop all kernel addresses */ +while (thread_stack__in_kernel (ts)) { + err =thread_stack__call_return (thread, ts, --ts->cnt, + tm, ref, true); +if (err) +return err; + } + +return 0 ; +} + static int thread_stack__no_call_return(struct thread *thread, struct thread_stack *ts, struct perf_sample *sample,
@@ -910,7 +927,18 @@ int thread_stack__process(struct thread *thread, struct comm *comm, ts->rstate =X86_RETPOLINE_DETECTED ; }else if (sample->flags &PERF_IP_FLAG_RETURN ) { -if (!sample->ip || !sample->addr) +if (!sample->addr) { +u32 return_from_kernel =PERF_IP_FLAG_SYSCALLRET | +PERF_IP_FLAG_INTERRUPT ; + +if (!(sample->flags & return_from_kernel)) +return 0 ; + +/* Pop kernel stack */ +return thread_stack__pop_ks (thread, ts, sample, ref); + } + +if (!sample->ip)return 0 ;/* x86 retpoline 'return' doesn't match the stack */
Adrian Hunter
reviewerPATCH 2/2 perf thread-stack: Eliminate code duplicating thread_stack__pop_ks()
In reply to Adrian HunterUse new function thread_stack__pop_ks() in place of equivalent code.
--- tools/perf/util/thread-stack.c | 18 ++++++------------ 1 file changed, 6 insertions(+), 12 deletions(-)
tools/perf/util/thread-stack.c +6 -12
@@ -673,12 +673,9 @@ static int thread_stack__no_call_return(struct thread *thread,if (ip >= ks && addr < ks) {/* Return to userspace, so pop all kernel addresses */ -while (thread_stack__in_kernel (ts)) { - err =thread_stack__call_return (thread, ts, --ts->cnt, - tm, ref, true); -if (err) -return err; - } + err =thread_stack__pop_ks (thread, ts, sample, ref); +if (err) +return err;/* If the stack is empty, push the userspace address */ if (!ts->cnt) {
@@ -688,12 +685,9 @@ static int thread_stack__no_call_return(struct thread *thread, } }else if (thread_stack__in_kernel (ts) && ip < ks) {/* Return to userspace, so pop all kernel addresses */ -while (thread_stack__in_kernel (ts)) { - err =thread_stack__call_return (thread, ts, --ts->cnt, - tm, ref, true); -if (err) -return err; - } + err =thread_stack__pop_ks (thread, ts, sample, ref); +if (err) +return err; }if (ts->cnt)
Signature · 1 line
2.17.1
tip-bot for Adrian Hunter
merge bot[tip:perf/core] perf thread-stack: Eliminate code duplicating thread_stack__pop_ks()
perf thread-stack: Eliminate code duplicating thread_stack__pop_ks()
Commit-ID: eb5d854456f5a4ccec6f9681b7196cf056df8cfa Gitweb: https://git.kernel.org/tip/eb5d854456f5a4ccec6f9681b7196cf056df8cfa Author: Adrian Hunter <[email protected]> AuthorDate: Wed, 19 Jun 2019 09:44:29 +0300 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Tue, 25 Jun 2019 08:47:10 -0300
perf thread-stack: Eliminate code duplicating thread_stack__pop_ks()
Use new function thread_stack__pop_ks() in place of equivalent code.
--- tools/perf/util/thread-stack.c | 18 ++++++------------ 1 file changed, 6 insertions(+), 12 deletions(-)
tools/perf/util/thread-stack.c +6 -12
@@ -664,12 +664,9 @@ static int thread_stack__no_call_return(struct thread *thread,if (ip >= ks && addr < ks) {/* Return to userspace, so pop all kernel addresses */ -while (thread_stack__in_kernel (ts)) { - err =thread_stack__call_return (thread, ts, --ts->cnt, - tm, ref, true); -if (err) -return err; - } + err =thread_stack__pop_ks (thread, ts, sample, ref); +if (err) +return err;/* If the stack is empty, push the userspace address */ if (!ts->cnt) {
@@ -679,12 +676,9 @@ static int thread_stack__no_call_return(struct thread *thread, } }else if (thread_stack__in_kernel (ts) && ip < ks) {/* Return to userspace, so pop all kernel addresses */ -while (thread_stack__in_kernel (ts)) { - err =thread_stack__call_return (thread, ts, --ts->cnt, - tm, ref, true); -if (err) -return err; - } + err =thread_stack__pop_ks (thread, ts, sample, ref); +if (err) +return err; }if (ts->cnt)
Em Wed, Jun 19, 2019 at 09:44:27AM +0300, Adrian Hunter escreveu:
Quoted 3 lines
Hi
Here is one non-urgent fix and a subsequent tidy-up.
Thanks, both applied.
- Arnaldo
Quoted 11 lines
Adrian Hunter (2):
perf thread-stack: Fix thread stack return from kernel for kernel-only case perf thread-stack: Eliminate code duplicatingthread_stack__pop_ks()tools/perf/util/thread-stack.c | 48 ++++++++++++++++++++++++++++++------------ 1 file changed, 35 insertions(+), 13 deletions(-)Regards
Adrian
Signature · 1 line
- Arnaldo