tools perf: Add branch speculation info
cover tools perf: Add branch speculation info
AMD Last Branch Record Extension Version 2 (LbrExtV2) provides branch speculation information and the perf UAPI is extended to provide this in a generic manner. Update the tools headers and perf tool to be able to show this additional information.
The LbrExtV2 enablement and UAPI changes can be found at:
https://lore.kernel.org/all/[email protected]/
Sandipan Das (4):
tools headers uapi: Sync perf_event.h with kernel sources tools headers x86: Sync msr-index.h with kernel sources perf script: Show branch speculation info
perf session: Show branch speculation info in raw dump
tools/arch/x86/include/asm/msr-index.h | 5 +++++ tools/include/uapi/linux/perf_event.h | 15 ++++++++++++++- tools/perf/builtin-script.c | 5 +++-- tools/perf/util/branch.c | 15 +++++++++++++++ tools/perf/util/branch.h | 5 ++++- tools/perf/util/evsel.c | 9 ++++++--- tools/perf/util/session.c | 5 +++-- 7 files changed, 50 insertions(+), 9 deletions(-)
Signature · 1 line
2.34.1
PATCH 1/4 tools headers uapi: Sync perf_event.h with kernel sources
Sync perf_event.h with the kernel sources by adding the new "spec" bitfield to branch entries.
--- tools/include/uapi/linux/perf_event.h | 15 ++++++++++++++- 1 file changed, 14 insertions(+), 1 deletion(-)
tools/include/uapi/linux/perf_event.h +14 -1
@@ -256,6 +256,17 @@ enum { PERF_BR_MAX, }; +/* +* Common branch speculation outcome classification +*/ +enum { +PERF_BR_SPEC_NA =0 ,/* Not available */ +PERF_BR_SPEC_WRONG_PATH =1 ,/* Speculative but on wrong path */ +PERF_BR_NON_SPEC_CORRECT_PATH =2 ,/* Non-speculative but on correct path */ +PERF_BR_SPEC_CORRECT_PATH =3 ,/* Speculative and on correct path */ +PERF_BR_SPEC_MAX , +}; +#define PERF_SAMPLE_BRANCH_PLM_ALL \ (PERF_SAMPLE_BRANCH_USER|\ PERF_SAMPLE_BRANCH_KERNEL|\
@@ -1360,6 +1371,7 @@ union perf_mem_data_src {* abort: aborting a hardware transaction * cycles: cycles from last branch (or 0 if not supported) * type: branch type +* spec: branch speculation info (or 0 if not supported) */ struct perf_branch_entry {__u64 from;
@@ -1370,7 +1382,8 @@ struct perf_branch_entry { abort:1, /* transaction abort */ cycles:16, /* cycle count to last branch */ type:4, /* branch type */ - reserved:40; + spec:2, /* branch speculation info */ + reserved:38; };union perf_sample_weight {
Signature · 1 line
2.34.1
PATCH 2/4 tools headers x86: Sync msr-index.h with kernel sources
In reply to Sandipan DasSync msr-index.h with the kernel sources by adding the new AMD Last Branch Record Extension Version 2 (LbrExtV2) MSRs.
--- tools/arch/x86/include/asm/msr-index.h | 5 +++++ 1 file changed, 5 insertions(+)
tools/arch/x86/include/asm/msr-index.h +5 -0
@@ -574,6 +574,9 @@#define MSR_AMD64_PERF_CNTR_GLOBAL_CTL 0xc0000301#define MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR 0xc0000302 +/* AMD Last Branch Record MSRs */ +#define MSR_AMD64_LBR_SELECT 0xc000010e +/* Fam 17h MSRs */ #define MSR_F17H_IRPERF 0xc00000e9
@@ -745,6 +748,8 @@#define MSR_AMD_DBG_EXTN_CFG 0xc000010f#define MSR_AMD_SAMP_BR_FROM 0xc0010300 +#define DBG_EXTN_CFG_LBRV2EN BIT_ULL(6) +#define MSR_IA32_MPERF 0x000000e7#define MSR_IA32_APERF 0x000000e8
Signature · 1 line
2.34.1
PATCH 3/4 perf script: Show branch speculation info
In reply to Sandipan DasShow the branch speculation info if provided by the branch recording hardware feature. This can be useful for optimizing code further.
The speculation info is appended to the end of the list of fields so any existing tools that use "/" as a delimiter for access fields via an index remain unaffected. Also show "-" instead of "N/A" when speculation info is unavailable because "/" is used as the field separator.
E.g.
$ perf record -j any,u,save_type ./test_branch $ perf script --fields brstacksym
Before:
[...]
check_match+0x60/strcmp+0x0/P/-/-/0/CALL
do_lookup_x+0x3c5/check_match+0x0/P/-/-/0/CALL
[...]
After:
[...] check_match+0x60/strcmp+0x0/P/-/-/0/CALL/NON_SPEC_CORRECT_PATH do_lookup_x+0x3c5/check_match+0x0/P/-/-/0/CALL/NON_SPEC_CORRECT_PATH [...]
--- tools/perf/builtin-script.c | 5 +++-- tools/perf/util/branch.c | 15 +++++++++++++++ tools/perf/util/branch.h | 5 ++++- tools/perf/util/evsel.c | 9 ++++++--- 4 files changed, 28 insertions(+), 6 deletions(-)
tools/perf/builtin-script.c +3 -2
@@ -872,12 +872,13 @@ mispred_str(struct branch_entry *br)static int print_bstack_flags (FILE *fp,struct branch_entry *br) { -return fprintf (fp,"/%c/%c/%c/%d/%s " , +return fprintf (fp,"/%c/%c/%c/%d/%s/%s " ,mispred_str (br), br->flags.in_tx ?'X' :'-' , br->flags.abort ?'A' :'-' , br->flags.cycles, - br->flags.type ?branch_type_name (br->flags.type) :"-" ); + br->flags.type ?branch_type_name (br->flags.type) :"-" , + br->flags.spec ?branch_spec_desc (br->flags.spec) :"-" ); } static int perf_sample__fprintf_brstack(struct perf_sample *sample,
tools/perf/util/branch.c +15 -0
@@ -146,3 +146,18 @@ int branch_type_str(struct branch_type_stat *st, char *bf, int size)return printed; } + +const char *branch_spec_desc (int spec) +{ +const char *branch_spec_outcomes[PERF_BR_SPEC_MAX ] = { +"N/A" , +"SPEC_WRONG_PATH" , +"NON_SPEC_CORRECT_PATH" , +"SPEC_CORRECT_PATH" , + }; + +if (spec >=0 && spec <PERF_BR_SPEC_MAX ) +return branch_spec_outcomes[spec]; + +return NULL ; +}
tools/perf/util/branch.h +4 -1
@@ -24,7 +24,8 @@ struct branch_flags { u64 abort:1; u64 cycles:16; u64 type:4; - u64 reserved:40; + u64 spec:2; + u64 reserved:38; }; }; };
@@ -85,4 +86,6 @@ const char *branch_type_name(int type);void branch_type_stat_display (FILE *fp,struct branch_type_stat *st);int branch_type_str (struct branch_type_stat *st,char *bf,int bfsize); +const char *branch_spec_desc (int spec); +#endif /* _PERF_BRANCH_H */
tools/perf/util/evsel.c +6 -3
@@ -2313,7 +2313,8 @@ u64 evsel__bitfield_swap_branch_flags(u64 value)* abort:1 //transaction abort * cycles:16 //cycle count to last branch * type:4 //branch type -* reserved:40 +* spec:2 //branch speculation info +* reserved:38 * } * } *
@@ -2332,7 +2333,8 @@ u64 evsel__bitfield_swap_branch_flags(u64 value) new_val |=bitfield_swap (value,3 ,1 ); new_val |=bitfield_swap (value,4 ,16 ); new_val |=bitfield_swap (value,20 ,4 ); - new_val |=bitfield_swap (value,24 ,40 ); + new_val |=bitfield_swap (value,24 ,2 ); + new_val |=bitfield_swap (value,26 ,38 ); }else { new_val =bitfield_swap (value,63 ,1 ); new_val |=bitfield_swap (value,62 ,1 );
@@ -2340,7 +2342,8 @@ u64 evsel__bitfield_swap_branch_flags(u64 value) new_val |=bitfield_swap (value,60 ,1 ); new_val |=bitfield_swap (value,44 ,16 ); new_val |=bitfield_swap (value,40 ,4 ); - new_val |=bitfield_swap (value,0 ,40 ); + new_val |=bitfield_swap (value,38 ,2 ); + new_val |=bitfield_swap (value,0 ,38 ); }return new_val;
Signature · 1 line
2.34.1
PATCH 4/4 perf session: Show branch speculation info in raw dump
In reply to Sandipan DasShow the branch speculation info if provided by the branch recording hardware feature. This can be useful for optimizing code further.
E.g.
$ perf record -j any,u ./test_branch $ perf report --dump-raw-trace
Before:
[...]
8380958377610 0x40b178 [0x1b0]: PERF_RECORD_SAMPLE(IP, 0x2): 7952/7952: 0x4f851a period: 48973 addr: 0 ... branch stack: nr:16
..... 0: 00000000004b52fd -> 00000000004f82c0 0 cycles P 0 ..... 1: ffffffff8220137c -> 00000000004b52f0 0 cycles M 0 ..... 2: 000000000041d1c4 -> 00000000004b52f0 0 cycles P 0 ..... 3: 00000000004e7ead -> 000000000041d1b0 0 cycles M 0 ..... 4: 00000000004e7f91 -> 00000000004e7ead 0 cycles P 0 ..... 5: 00000000004e7ea8 -> 00000000004e7f70 0 cycles P 0 ..... 6: 00000000004e7e52 -> 00000000004e7e98 0 cycles M 0 ..... 7: 00000000004e7e1f -> 00000000004e7e40 0 cycles M 0 ..... 8: 00000000004e7f60 -> 00000000004e7df0 0 cycles P 0 ..... 9: 00000000004e7f58 -> 00000000004e7f60 0 cycles M 0 ..... 10: 000000000041d85d -> 00000000004e7f50 0 cycles P 0 ..... 11: 000000000043306a -> 000000000041d840 0 cycles P 0 ..... 12: ffffffff8220137c -> 0000000000433040 0 cycles M 0 ..... 13: 000000000041e4a1 -> 0000000000433040 0 cycles P 0 ..... 14: ffffffff8220137c -> 000000000041e490 0 cycles M 0 ..... 15: 000000000041d89b -> 000000000041e487 0 cycles P 0
... thread: test_branch:7952
...... dso: /data/sandipan/test_branch
[...]
After:
[...]
8380958377610 0x40b178 [0x1b0]: PERF_RECORD_SAMPLE(IP, 0x2): 7952/7952: 0x4f851a period: 48973 addr: 0 ... branch stack: nr:16
..... 0: 00000000004b52fd -> 00000000004f82c0 0 cycles P 0 NON_SPEC_CORRECT_PATH ..... 1: ffffffff8220137c -> 00000000004b52f0 0 cycles M 0 NON_SPEC_CORRECT_PATH ..... 2: 000000000041d1c4 -> 00000000004b52f0 0 cycles P 0 NON_SPEC_CORRECT_PATH ..... 3: 00000000004e7ead -> 000000000041d1b0 0 cycles M 0 NON_SPEC_CORRECT_PATH ..... 4: 00000000004e7f91 -> 00000000004e7ead 0 cycles P 0 NON_SPEC_CORRECT_PATH ..... 5: 00000000004e7ea8 -> 00000000004e7f70 0 cycles P 0 NON_SPEC_CORRECT_PATH ..... 6: 00000000004e7e52 -> 00000000004e7e98 0 cycles M 0 SPEC_CORRECT_PATH ..... 7: 00000000004e7e1f -> 00000000004e7e40 0 cycles M 0 NON_SPEC_CORRECT_PATH ..... 8: 00000000004e7f60 -> 00000000004e7df0 0 cycles P 0 NON_SPEC_CORRECT_PATH ..... 9: 00000000004e7f58 -> 00000000004e7f60 0 cycles M 0 NON_SPEC_CORRECT_PATH ..... 10: 000000000041d85d -> 00000000004e7f50 0 cycles P 0 NON_SPEC_CORRECT_PATH ..... 11: 000000000043306a -> 000000000041d840 0 cycles P 0 NON_SPEC_CORRECT_PATH ..... 12: ffffffff8220137c -> 0000000000433040 0 cycles M 0 NON_SPEC_CORRECT_PATH ..... 13: 000000000041e4a1 -> 0000000000433040 0 cycles P 0 NON_SPEC_CORRECT_PATH ..... 14: ffffffff8220137c -> 000000000041e490 0 cycles M 0 NON_SPEC_CORRECT_PATH ..... 15: 000000000041d89b -> 000000000041e487 0 cycles P 0 NON_SPEC_CORRECT_PATH
... thread: test_branch:7952
...... dso: /data/sandipan/test_branch
[...]
--- tools/perf/util/session.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
tools/perf/util/session.c +3 -2
@@ -1172,7 +1172,7 @@ static void branch_stack__printf(struct perf_sample *sample, bool callstack)struct branch_entry *e = &entries[i];if (!callstack) { - printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x %s\n", + printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x %s %s\n", i, e->from, e->to, (unsigned short)e->flags.cycles, e->flags.mispred ? "M" : " ",
@@ -1180,7 +1180,8 @@ static void branch_stack__printf(struct perf_sample *sample, bool callstack) e->flags.abort ?"A" :" " , e->flags.in_tx ?"T" :" " , (unsigned )e->flags.reserved, - e->flags.type ? branch_type_name(e->flags.type) : ""); + e->flags.type ?branch_type_name (e->flags.type) :"" , + e->flags.spec ? branch_spec_desc(e->flags.spec) : ""); }else {if (i ==0 ) { printf("..... %2"PRIu64": %016" PRIx64 "\n"
Signature · 1 line
2.34.1
Re: [PATCH 2/4] tools headers x86: Sync msr-index.h with kernel sources
In reply to Sandipan DasQuoted 13 lines · Sandipan Das
Sync
msr-index.hwith the kernel sources by adding the new AMD Last Branch Record Extension Version 2 (LbrExtV2) MSRs.Signed-off-by Sandipan Das <[email protected]>--- tools/arch/x86/include/asm/msr-index.h | 5 +++++ 1 file changed, 5 insertions(+)tools/arch/x86/include/asm/msr-index.h +0
-0
tools/arch/x86/include/asm/msr-index.h line 578
Quoted 5 lines
tools/arch/x86/include/asm/msr-index.h +2
-0@@ -574,6 +574,9 @@#define MSR_AMD64_PERF_CNTR_GLOBAL_CTL 0xc0000301#define MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR 0xc0000302 +/* AMD Last Branch Record MSRs */ +#define MSR_AMD64_LBR_SELECT 0xc000010e
curious do we actualy use this in tools somewhere?
jirka
Quoted 16 lines
tools/arch/x86/include/asm/msr-index.h +3
-0+/* Fam 17h MSRs */ #define MSR_F17H_IRPERF 0xc00000e9@@ -745,6 +748,8 @@#define MSR_AMD_DBG_EXTN_CFG 0xc000010f#define MSR_AMD_SAMP_BR_FROM 0xc0010300 +#define DBG_EXTN_CFG_LBRV2EN BIT_ULL(6) +#define MSR_IA32_MPERF 0x000000e7#define MSR_IA32_APERF 0x000000e8Signature · 1 line
2.34.1
Re: [PATCH 2/4] tools headers x86: Sync msr-index.h with kernel sources
Hi Jiri,
Quoted 17 lines · Jiri Olsa
Quoted 16 lines · Sandipan Das
Sync
msr-index.hwith the kernel sources by adding the new AMD Last Branch Record Extension Version 2 (LbrExtV2) MSRs.Signed-off-by Sandipan Das <[email protected]>--- tools/arch/x86/include/asm/msr-index.h | 5 +++++ 1 file changed, 5 insertions(+)tools/arch/x86/include/asm/msr-index.h +0
-0@@ -574,6 +574,9 @@#define MSR_AMD64_PERF_CNTR_GLOBAL_CTL 0xc0000301#define MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR 0xc0000302
Quoted 7 lines
tools/arch/x86/include/asm/msr-index.h line 578
Quoted 2 lines
tools/arch/x86/include/asm/msr-index.h +2
-0+/* AMD Last Branch Record MSRs */ +#define MSR_AMD64_LBR_SELECT 0xc000010ecurious do we actualy use this in tools somewhere?
jirka
Commit 9dde6cadb92b ("tools arch x86: Sync the msr-index.h copy with the kernel sources") from Arnaldo says that adding these new MSR definitions in the tools headers allows the beautification scripts to pick up new entries which can be used for filtering MSR access traces.
E.g. one can trace the hardware LBR branch filter bits getting written to the LBR_SELECT MSR while recording branches.
$ perf record -j any,u true $ perf record -j any_call,u true
$ sudo perf trace -e msr:write_msr/max-stack=32/ --filter="msr == AMD64_LBR_SELECT"
[...]
224568.130 perf/9093 msr:write_msr(msr: AMD64_LBR_SELECT, val: 1)
do_trace_write_msr ([kernel.kallsyms])
do_trace_write_msr ([kernel.kallsyms])
native_write_msr ([kernel.kallsyms])
amd_pmu_lbr_enable_all ([kernel.kallsyms])
amd_pmu_v2_enable_all ([kernel.kallsyms])
x86_pmu_enable ([kernel.kallsyms])
ctx_resched ([kernel.kallsyms])
perf_event_exec ([kernel.kallsyms])
begin_new_exec ([kernel.kallsyms])
load_elf_binary ([kernel.kallsyms])
bprm_execve ([kernel.kallsyms])
do_execveat_common.isra.0 ([kernel.kallsyms])
__x64_sys_execve ([kernel.kallsyms])
do_syscall_64 ([kernel.kallsyms])
entry_SYSCALL_64 ([kernel.kallsyms])
execve (/usr/lib/x86_64-linux-gnu/libc.so.6)
[...]
302748.439 perf/9126 msr:write_msr(msr: AMD64_LBR_SELECT, val: 229)
do_trace_write_msr ([kernel.kallsyms])
do_trace_write_msr ([kernel.kallsyms])
native_write_msr ([kernel.kallsyms])
amd_pmu_lbr_enable_all ([kernel.kallsyms])
amd_pmu_v2_enable_all ([kernel.kallsyms])
x86_pmu_enable ([kernel.kallsyms])
ctx_resched ([kernel.kallsyms])
perf_event_exec ([kernel.kallsyms])
begin_new_exec ([kernel.kallsyms])
load_elf_binary ([kernel.kallsyms])
bprm_execve ([kernel.kallsyms])
do_execveat_common.isra.0 ([kernel.kallsyms])
__x64_sys_execve ([kernel.kallsyms])
do_syscall_64 ([kernel.kallsyms])
entry_SYSCALL_64 ([kernel.kallsyms])
execve (/usr/lib/x86_64-linux-gnu/libc.so.6)[...]
I can add this example to the commit message in the next revision.
- Sandipan
Arnaldo Carvalho de Melo
maintainerRe: [PATCH 2/4] tools headers x86: Sync msr-index.h with kernel sources
Em Fri, Aug 12, 2022 at 02:33:46PM +0530, Sandipan Das escreveu:
Quoted 25 lines
Hi Jiri,
Quoted 22 lines · Jiri Olsa
Quoted 13 lines · Sandipan Das
Sync
msr-index.hwith the kernel sources by adding the new AMD Last Branch Record Extension Version 2 (LbrExtV2) MSRs.Signed-off-by Sandipan Das <[email protected]>--- tools/arch/x86/include/asm/msr-index.h | 5 +++++ 1 file changed, 5 insertions(+)tools/arch/x86/include/asm/msr-index.h +0
-0tools/arch/x86/include/asm/msr-index.h line 578
Quoted 5 lines
tools/arch/x86/include/asm/msr-index.h +2
-0@@ -574,6 +574,9 @@#define MSR_AMD64_PERF_CNTR_GLOBAL_CTL 0xc0000301#define MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR 0xc0000302 +/* AMD Last Branch Record MSRs */ +#define MSR_AMD64_LBR_SELECT 0xc000010ecurious do we actualy use this in tools somewhere?
Quoted 7 lines
Quoted 2 lines
jirka
Commit
9dde6cadb92b("tools arch x86: Sync themsr-index.hcopy with the kernel sources") from Arnaldo says that adding these new MSR definitions in the tools headers allows the beautification scripts to pick up new entries which can be used for filtering MSR access traces.
Right, in this specific case that header should be moved to tools/perf/trace/beauty/include/, as it is only used to harvest new MSRs by:
⬢[acme@toolbox perf]$ tools/perf/trace/beauty/tracepoints/x86_msr.sh | wc -l 328
⬢[acme@toolbox perf]$ tools/perf/trace/beauty/tracepoints/x86_msr.sh | head
static const char *x86_MSRs[] = {
[0x00000000] = "IA32_P5_MC_ADDR",
[0x00000001] = "IA32_P5_MC_TYPE",
[0x00000010] = "IA32_TSC",
[0x00000017] = "IA32_PLATFORM_ID",
[0x0000001b] = "IA32_APICBASE",
[0x00000020] = "KNC_PERFCTR0",
[0x00000021] = "KNC_PERFCTR1",
[0x00000028] = "KNC_EVNTSEL0",
[0x00000029] = "KNC_EVNTSEL1",⬢[acme@toolbox perf]$
Other headers we have copies from the kernel may be used both for harvesting non-enum enumerations like those to pretty print and use in tracepoint filter expressions in 'perf trace', and in building perf on older systems.
The preferred way is for kernel developers not to update the copies when they add new stuff to the original file, so that perf developers have the opportunity of checking if the new original file in the kernel don't break the scripts in tools/perf/, or if the new additions are note matched by regexps in the perf/tools/trace/beauty/ extraction scripts.
But kernel developers are welcome to update it iif they do these checks themselves, at the very least do a 'make -C tools/perf' to check if it builds before/after the update.
- Arnaldo
Quoted 50 lines
E.g. one can trace the hardware LBR branch filter bits getting written to the
LBR_SELECTMSR while recording branches.$ perf record -j any,u true $ perf record -j any_call,u true$ sudo perf trace -e msr:write_msr/max-stack=32/ --filter="msr == AMD64_LBR_SELECT"[...]
224568.130 perf/9093 msr:write_msr(msr:AMD64_LBR_SELECT, val: 1)do_trace_write_msr ([kernel.kallsyms]) do_trace_write_msr ([kernel.kallsyms]) native_write_msr ([kernel.kallsyms]) amd_pmu_lbr_enable_all ([kernel.kallsyms]) amd_pmu_v2_enable_all ([kernel.kallsyms]) x86_pmu_enable ([kernel.kallsyms]) ctx_resched ([kernel.kallsyms]) perf_event_exec ([kernel.kallsyms]) begin_new_exec ([kernel.kallsyms]) load_elf_binary ([kernel.kallsyms]) bprm_execve ([kernel.kallsyms]) do_execveat_common.isra.0 ([kernel.kallsyms]) __x64_sys_execve ([kernel.kallsyms]) do_syscall_64 ([kernel.kallsyms]) entry_SYSCALL_64 ([kernel.kallsyms]) execve (/usr/lib/x86_64-linux-gnu/libc.so.6) [...] 302748.439 perf/9126 msr:write_msr(msr: AMD64_LBR_SELECT, val: 229) do_trace_write_msr ([kernel.kallsyms]) do_trace_write_msr ([kernel.kallsyms]) native_write_msr ([kernel.kallsyms]) amd_pmu_lbr_enable_all ([kernel.kallsyms]) amd_pmu_v2_enable_all ([kernel.kallsyms]) x86_pmu_enable ([kernel.kallsyms]) ctx_resched ([kernel.kallsyms]) perf_event_exec ([kernel.kallsyms]) begin_new_exec ([kernel.kallsyms]) load_elf_binary ([kernel.kallsyms]) bprm_execve ([kernel.kallsyms]) do_execveat_common.isra.0 ([kernel.kallsyms]) __x64_sys_execve ([kernel.kallsyms]) do_syscall_64 ([kernel.kallsyms]) entry_SYSCALL_64 ([kernel.kallsyms]) execve (/usr/lib/x86_64-linux-gnu/libc.so.6)[...]
I can add this example to the commit message in the next revision.
- Sandipan
Signature · 1 line
- Arnaldo
Arnaldo Carvalho de Melo
maintainerRe: [PATCH 2/4] tools headers x86: Sync msr-index.h with kernel sources
In reply to Sandipan DasEm Fri, Aug 12, 2022 at 02:33:46PM +0530, Sandipan Das escreveu:
Quoted 77 lines
Hi Jiri,
Quoted 17 lines · Jiri Olsa
Quoted 16 lines · Sandipan Das
Sync
msr-index.hwith the kernel sources by adding the new AMD Last Branch Record Extension Version 2 (LbrExtV2) MSRs.Signed-off-by Sandipan Das <[email protected]>--- tools/arch/x86/include/asm/msr-index.h | 5 +++++ 1 file changed, 5 insertions(+)tools/arch/x86/include/asm/msr-index.h +0
-0@@ -574,6 +574,9 @@#define MSR_AMD64_PERF_CNTR_GLOBAL_CTL 0xc0000301#define MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR 0xc0000302Quoted 7 lines
tools/arch/x86/include/asm/msr-index.h line 578
Quoted 2 lines
tools/arch/x86/include/asm/msr-index.h +2
-0+/* AMD Last Branch Record MSRs */ +#define MSR_AMD64_LBR_SELECT 0xc000010ecurious do we actualy use this in tools somewhere?
jirka
Commit
9dde6cadb92b("tools arch x86: Sync themsr-index.hcopy with the kernel sources") from Arnaldo says that adding these new MSR definitions in the tools headers allows the beautification scripts to pick up new entries which can be used for filtering MSR access traces.E.g. one can trace the hardware LBR branch filter bits getting written to the
LBR_SELECTMSR while recording branches.$ perf record -j any,u true $ perf record -j any_call,u true$ sudo perf trace -e msr:write_msr/max-stack=32/ --filter="msr == AMD64_LBR_SELECT"[...]
224568.130 perf/9093 msr:write_msr(msr:AMD64_LBR_SELECT, val: 1)do_trace_write_msr ([kernel.kallsyms]) do_trace_write_msr ([kernel.kallsyms]) native_write_msr ([kernel.kallsyms]) amd_pmu_lbr_enable_all ([kernel.kallsyms]) amd_pmu_v2_enable_all ([kernel.kallsyms]) x86_pmu_enable ([kernel.kallsyms]) ctx_resched ([kernel.kallsyms]) perf_event_exec ([kernel.kallsyms]) begin_new_exec ([kernel.kallsyms]) load_elf_binary ([kernel.kallsyms]) bprm_execve ([kernel.kallsyms]) do_execveat_common.isra.0 ([kernel.kallsyms]) __x64_sys_execve ([kernel.kallsyms]) do_syscall_64 ([kernel.kallsyms]) entry_SYSCALL_64 ([kernel.kallsyms]) execve (/usr/lib/x86_64-linux-gnu/libc.so.6) [...] 302748.439 perf/9126 msr:write_msr(msr: AMD64_LBR_SELECT, val: 229) do_trace_write_msr ([kernel.kallsyms]) do_trace_write_msr ([kernel.kallsyms]) native_write_msr ([kernel.kallsyms]) amd_pmu_lbr_enable_all ([kernel.kallsyms]) amd_pmu_v2_enable_all ([kernel.kallsyms]) x86_pmu_enable ([kernel.kallsyms]) ctx_resched ([kernel.kallsyms]) perf_event_exec ([kernel.kallsyms]) begin_new_exec ([kernel.kallsyms]) load_elf_binary ([kernel.kallsyms]) bprm_execve ([kernel.kallsyms]) do_execveat_common.isra.0 ([kernel.kallsyms]) __x64_sys_execve ([kernel.kallsyms]) do_syscall_64 ([kernel.kallsyms]) entry_SYSCALL_64 ([kernel.kallsyms]) execve (/usr/lib/x86_64-linux-gnu/libc.so.6)
Quoted 3 lines
[...]
I can add this example to the commit message in the next revision.
Right, you can reuse something from my messages updating this file or, better, show an example like you did above, for the thing that you are working at that moment :-)
One of the things we need to do is:
perf trace -e msr:write_msr/max-stack=32/ --filter-help
And then it should look what are the strings that can be used for that specific msr:write_msr tracepoint.
Also have tab expansion in the --filter= part :-)
- Arnaldo
Re: [PATCH 2/4] tools headers x86: Sync msr-index.h with kernel sources
Hi Arnaldo,
Quoted 86 lines · Arnaldo Carvalho de Melo
Em Fri, Aug 12, 2022 at 02:33:46PM +0530, Sandipan Das escreveu:
Quoted 77 lines
Hi Jiri,
Quoted 17 lines · Jiri Olsa
Quoted 16 lines · Sandipan Das
Sync
msr-index.hwith the kernel sources by adding the new AMD Last Branch Record Extension Version 2 (LbrExtV2) MSRs.Signed-off-by Sandipan Das <[email protected]>--- tools/arch/x86/include/asm/msr-index.h | 5 +++++ 1 file changed, 5 insertions(+)tools/arch/x86/include/asm/msr-index.h +0
-0@@ -574,6 +574,9 @@#define MSR_AMD64_PERF_CNTR_GLOBAL_CTL 0xc0000301#define MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR 0xc0000302Quoted 7 lines
tools/arch/x86/include/asm/msr-index.h line 578
Quoted 2 lines
tools/arch/x86/include/asm/msr-index.h +2
-0+/* AMD Last Branch Record MSRs */ +#define MSR_AMD64_LBR_SELECT 0xc000010ecurious do we actualy use this in tools somewhere?
jirka
Commit
9dde6cadb92b("tools arch x86: Sync themsr-index.hcopy with the kernel sources") from Arnaldo says that adding these new MSR definitions in the tools headers allows the beautification scripts to pick up new entries which can be used for filtering MSR access traces.E.g. one can trace the hardware LBR branch filter bits getting written to the
LBR_SELECTMSR while recording branches.$ perf record -j any,u true $ perf record -j any_call,u true$ sudo perf trace -e msr:write_msr/max-stack=32/ --filter="msr == AMD64_LBR_SELECT"[...]
224568.130 perf/9093 msr:write_msr(msr:AMD64_LBR_SELECT, val: 1)do_trace_write_msr ([kernel.kallsyms]) do_trace_write_msr ([kernel.kallsyms]) native_write_msr ([kernel.kallsyms]) amd_pmu_lbr_enable_all ([kernel.kallsyms]) amd_pmu_v2_enable_all ([kernel.kallsyms]) x86_pmu_enable ([kernel.kallsyms]) ctx_resched ([kernel.kallsyms]) perf_event_exec ([kernel.kallsyms]) begin_new_exec ([kernel.kallsyms]) load_elf_binary ([kernel.kallsyms]) bprm_execve ([kernel.kallsyms]) do_execveat_common.isra.0 ([kernel.kallsyms]) __x64_sys_execve ([kernel.kallsyms]) do_syscall_64 ([kernel.kallsyms]) entry_SYSCALL_64 ([kernel.kallsyms]) execve (/usr/lib/x86_64-linux-gnu/libc.so.6) [...] 302748.439 perf/9126 msr:write_msr(msr: AMD64_LBR_SELECT, val: 229) do_trace_write_msr ([kernel.kallsyms]) do_trace_write_msr ([kernel.kallsyms]) native_write_msr ([kernel.kallsyms]) amd_pmu_lbr_enable_all ([kernel.kallsyms]) amd_pmu_v2_enable_all ([kernel.kallsyms]) x86_pmu_enable ([kernel.kallsyms]) ctx_resched ([kernel.kallsyms]) perf_event_exec ([kernel.kallsyms]) begin_new_exec ([kernel.kallsyms]) load_elf_binary ([kernel.kallsyms]) bprm_execve ([kernel.kallsyms]) do_execveat_common.isra.0 ([kernel.kallsyms]) __x64_sys_execve ([kernel.kallsyms]) do_syscall_64 ([kernel.kallsyms]) entry_SYSCALL_64 ([kernel.kallsyms]) execve (/usr/lib/x86_64-linux-gnu/libc.so.6)Quoted 3 lines
[...]
I can add this example to the commit message in the next revision.
Right, you can reuse something from my messages updating this file or, better, show an example like you did above, for the thing that you are working at that moment :-)
Quoted 9 lines
One of the things we need to do is:
perf trace -e msr:write_msr/max-stack=32/ --filter-help
And then it should look what are the strings that can be used for that specific msr:write_msr tracepoint.
Also have tab expansion in the --filter= part :-)
Thanks for the suggestions. Will incorporate them in the next revision.
- Sandipan