perf report: Show branch type
Subsystems (2)
v2 cover perf report: Show branch type
v2:
---
1. Use 4 bits in perf_branch_entry to record branch type.
2. Pull out some common branch types from FAR_BRANCH. Now the branch types defined in perf_event.h:
PERF_BR_NONE : unknownPERF_BR_JCC_FWD : conditional forward jumpPERF_BR_JCC_BWD : conditional backward jumpPERF_BR_JMP : jumpPERF_BR_IND_JMP : indirect jumpPERF_BR_CALL : callPERF_BR_IND_CALL : indirect callPERF_BR_RET : returnPERF_BR_SYSCALL : syscallPERF_BR_SYSRET : syscall returnPERF_BR_IRQ : hw interrupt/trap/faultPERF_BR_INT : sw interruptPERF_BR_IRET : return from interruptPERF_BR_FAR_BRANCH: others not generic far branch type
3. Use 2 bits in perf_branch_entry for a "cross" metrics checking for branch cross 4K or 2M area. It's an approximate computing for checking if the branch cross 4K page or 2MB page.
For example:
perf record -g --branch-filter any,save_type <command>
perf report --stdio
JCC forward: 27.7%
JCC backward: 9.8%
JMP: 0.0%
IND_JMP: 6.5%
CALL: 26.6%
IND_CALL: 0.0%
RET: 29.3%
IRET: 0.0%
CROSS_4K: 0.0%
CROSS_2M: 14.3%perf report --branch-history --stdio --no-children
-23.60%--main div.c:42 (RET cycles:2)
compute_flag div.c:28 (RET cycles:2) compute_flag div.c:27 (RET CROSS_2M cycles:1) rand rand.c:28 (RET CROSS_2M cycles:1) rand rand.c:28 (RET cycles:1) __random random.c:298 (RET cycles:1) __random random.c:297 (JCC forward cycles:1) __random random.c:295 (JCC forward cycles:1) __random random.c:295 (JCC forward cycles:1) __random random.c:295 (JCC forward cycles:1) __random random.c:295 (RET cycles:9)
Changed:
perf/core: Define the common branch type classification perf/x86/intel: Record branch type
perf report: Show branch type statistics for stdio mode perf report: Show branch type in callchain entry
Not changed:
perf record: Create a new option save_type in --branch-filter
v1:
---
It is often useful to know the branch types while analyzing branch data. For example, a call is very different from a conditional branch.
Currently we have to look it up in binary while the binary may later not be available and even the binary is available but user has to take some time. It is very useful for user to check it directly in perf report.
Perf already has support for disassembling the branch instruction to get the branch type.
The patch series records the branch type and show the branch type with other LBR information in callchain entry via perf report. The patch series also adds the branch type summary at the end of
perf report --stdio.
To keep consistent on kernel and userspace and make the classification more common, the patch adds the common branch type classification in perf_event.h.
The common branch types are:
JCC forward: Conditional forward jump
JCC backward: Conditional backward jump
JMP: Jump imm
IND_JMP: Jump reg/mem
CALL: Call imm
IND_CALL: Call reg/mem
RET: Ret
FAR_BRANCH: SYSCALL/SYSRET, IRQ, IRET, TSX Abort
An example:
1. Record branch type (new option "save_type")
perf record -g --branch-filter any,save_type <command>
2. Show the branch type statistics at the end of perf report --stdio
perf report --stdio
JCC forward: 34.0%
JCC backward: 3.6%
JMP: 0.0%
IND_JMP: 6.5%
CALL: 26.6%
IND_CALL: 0.0%
RET: 29.3%
FAR_BRANCH: 0.0%3. Show branch type in callchain entry
perf report --branch-history --stdio --no-children
--23.91%--main div.c:42 (RET cycles:2)
compute_flag div.c:28 (RET cycles:2) compute_flag div.c:27 (RET cycles:1) rand rand.c:28 (RET cycles:1)
rand rand.c:28 (RET cycles:1) __random random.c:298 (RET cycles:1) __random random.c:297 (JCC forward cycles:1) __random random.c:295 (JCC forward cycles:1) __random random.c:295 (JCC forward cycles:1) __random random.c:295 (JCC forward cycles:1) __random random.c:295 (RET cycles:9)
Jin Yao (5):
perf/core: Define the common branch type classification perf/x86/intel: Record branch type
perf record: Create a new option save_type in --branch-filter perf report: Show branch type statistics for stdio mode perf report: Show branch type in callchain entry
arch/x86/events/intel/lbr.c | 106 ++++++++++++++- include/uapi/linux/perf_event.h | 37 +++++- tools/include/uapi/linux/perf_event.h | 37 +++++- tools/perf/Documentation/perf-record.txt | 1 + tools/perf/builtin-report.c | 212 +++++++++++++++++++++++++++++ tools/perf/util/callchain.c | 221 ++++++++++++++++++++++--------- tools/perf/util/callchain.h | 20 +++ tools/perf/util/event.h | 4 +- tools/perf/util/hist.c | 5 +- tools/perf/util/parse-branch-options.c | 1 + 10 files changed, 577 insertions(+), 67 deletions(-)
Signature · 1 line
2.7.4
PATCH v2 1/5 perf/core: Define the common branch type classification
It is often useful to know the branch types while analyzing branch data. For example, a call is very different from a conditional branch.
Currently we have to look it up in binary while the binary may later not be available and even the binary is available but user has to take some time. It is very useful for user to check it directly in perf report.
Perf already has support for disassembling the branch instruction to get the x86 branch type.
To keep consistent on kernel and userspace and make the classification more common, the patch adds the common branch type classification in perf_event.h.
PERF_BR_NONE : unknownPERF_BR_JCC_FWD : conditional forward jumpPERF_BR_JCC_BWD : conditional backward jumpPERF_BR_JMP : jumpPERF_BR_IND_JMP : indirect jumpPERF_BR_CALL : callPERF_BR_IND_CALL : indirect callPERF_BR_RET : returnPERF_BR_SYSCALL : syscallPERF_BR_SYSRET : syscall returnPERF_BR_IRQ : hw interrupt/trap/faultPERF_BR_INT : sw interruptPERF_BR_IRET : return from interruptPERF_BR_FAR_BRANCH: others not generic branch type
The patch adds following metrics checking for branches cross 4K or 2MB areas.
PERF_BR_CROSS_NONE: branch not cross an areaPERF_BR_CROSS_4K : branch cross 4K areaPERF_BR_CROSS_2M : branch cross 2MB area
Since the disassembling of branch instruction needs some overhead, a new PERF_SAMPLE_BRANCH_TYPE_SAVE is introduced to indicate if it needs to disassemble the branch instruction and record the branch type.
--- include/uapi/linux/perf_event.h | 37 ++++++++++++++++++++++++++++++++++- tools/include/uapi/linux/perf_event.h | 37 ++++++++++++++++++++++++++++++++++- 2 files changed, 72 insertions(+), 2 deletions(-)
include/uapi/linux/perf_event.h +36 -1
@@ -174,6 +174,8 @@ enum perf_branch_sample_type_shift {PERF_SAMPLE_BRANCH_NO_FLAGS_SHIFT =14 ,/* no flags */ PERF_SAMPLE_BRANCH_NO_CYCLES_SHIFT =15 ,/* no cycles */ +PERF_SAMPLE_BRANCH_TYPE_SAVE_SHIFT =16 ,/* save branch type */ + PERF_SAMPLE_BRANCH_MAX_SHIFT /* non-ABI */ };
@@ -198,9 +200,38 @@ enum perf_branch_sample_type {PERF_SAMPLE_BRANCH_NO_FLAGS =1U <<PERF_SAMPLE_BRANCH_NO_FLAGS_SHIFT ,PERF_SAMPLE_BRANCH_NO_CYCLES =1U <<PERF_SAMPLE_BRANCH_NO_CYCLES_SHIFT , +PERF_SAMPLE_BRANCH_TYPE_SAVE = +1U <<PERF_SAMPLE_BRANCH_TYPE_SAVE_SHIFT , + PERF_SAMPLE_BRANCH_MAX = 1U << PERF_SAMPLE_BRANCH_MAX_SHIFT, }; +/* +* Common flow change classification +*/ +enum { +PERF_BR_NONE =0 ,/* unknown */ +PERF_BR_JCC_FWD =1 ,/* conditional forward jump */ +PERF_BR_JCC_BWD =2 ,/* conditional backward jump */ +PERF_BR_JMP =3 ,/* jump */ +PERF_BR_IND_JMP =4 ,/* indirect jump */ +PERF_BR_CALL =5 ,/* call */ +PERF_BR_IND_CALL =6 ,/* indirect call */ +PERF_BR_RET =7 ,/* return */ +PERF_BR_SYSCALL =8 ,/* syscall */ +PERF_BR_SYSRET =9 ,/* syscall return */ +PERF_BR_IRQ =10 ,/* hw interrupt/trap/fault */ +PERF_BR_INT =11 ,/* sw interrupt */ +PERF_BR_IRET =12 ,/* return from interrupt */ +PERF_BR_FAR_BRANCH =13 ,/* others not generic branch type */ +}; + +enum { +PERF_BR_CROSS_NONE =0 ,/* branch not cross an area */ +PERF_BR_CROSS_4K =1 ,/* branch cross 4K */ +PERF_BR_CROSS_2M =2 ,/* branch cross 2MB */ +}; +#define PERF_SAMPLE_BRANCH_PLM_ALL \ (PERF_SAMPLE_BRANCH_USER|\ PERF_SAMPLE_BRANCH_KERNEL|\
@@ -999,6 +1030,8 @@ union perf_mem_data_src {* in_tx: running in a hardware transaction * abort: aborting a hardware transaction * cycles: cycles from last branch (or 0 if not supported) +* type: branch type +* cross: branch cross 4K or 2MB area */ struct perf_branch_entry {__u64 from;
@@ -1008,7 +1041,9 @@ struct perf_branch_entry { in_tx:1, /* in transaction */ abort:1, /* transaction abort */ cycles:16, /* cycle count to last branch */ - reserved:44; + type:4, /* branch type */ + cross:2, /* branch cross 4K or 2MB area */ + reserved:38; };#endif /* _UAPI_LINUX_PERF_EVENT_H */
tools/include/uapi/linux/perf_event.h +36 -1
@@ -174,6 +174,8 @@ enum perf_branch_sample_type_shift {PERF_SAMPLE_BRANCH_NO_FLAGS_SHIFT =14 ,/* no flags */ PERF_SAMPLE_BRANCH_NO_CYCLES_SHIFT =15 ,/* no cycles */ +PERF_SAMPLE_BRANCH_TYPE_SAVE_SHIFT =16 ,/* save branch type */ + PERF_SAMPLE_BRANCH_MAX_SHIFT /* non-ABI */ };
@@ -198,9 +200,38 @@ enum perf_branch_sample_type {PERF_SAMPLE_BRANCH_NO_FLAGS =1U <<PERF_SAMPLE_BRANCH_NO_FLAGS_SHIFT ,PERF_SAMPLE_BRANCH_NO_CYCLES =1U <<PERF_SAMPLE_BRANCH_NO_CYCLES_SHIFT , +PERF_SAMPLE_BRANCH_TYPE_SAVE = +1U <<PERF_SAMPLE_BRANCH_TYPE_SAVE_SHIFT , + PERF_SAMPLE_BRANCH_MAX = 1U << PERF_SAMPLE_BRANCH_MAX_SHIFT, }; +/* +* Common flow change classification +*/ +enum { +PERF_BR_NONE =0 ,/* unknown */ +PERF_BR_JCC_FWD =1 ,/* conditional forward jump */ +PERF_BR_JCC_BWD =2 ,/* conditional backward jump */ +PERF_BR_JMP =3 ,/* jump */ +PERF_BR_IND_JMP =4 ,/* indirect jump */ +PERF_BR_CALL =5 ,/* call */ +PERF_BR_IND_CALL =6 ,/* indirect call */ +PERF_BR_RET =7 ,/* return */ +PERF_BR_SYSCALL =8 ,/* syscall */ +PERF_BR_SYSRET =9 ,/* syscall return */ +PERF_BR_IRQ =10 ,/* hw interrupt/trap/fault */ +PERF_BR_INT =11 ,/* sw interrupt */ +PERF_BR_IRET =12 ,/* return from interrupt */ +PERF_BR_FAR_BRANCH =13 ,/* others not generic branch type */ +}; + +enum { +PERF_BR_CROSS_NONE =0 ,/* branch not cross an area */ +PERF_BR_CROSS_4K =1 ,/* branch cross 4K */ +PERF_BR_CROSS_2M =2 ,/* branch cross 2MB */ +}; +#define PERF_SAMPLE_BRANCH_PLM_ALL \ (PERF_SAMPLE_BRANCH_USER|\ PERF_SAMPLE_BRANCH_KERNEL|\
@@ -999,6 +1030,8 @@ union perf_mem_data_src {* in_tx: running in a hardware transaction * abort: aborting a hardware transaction * cycles: cycles from last branch (or 0 if not supported) +* type: branch type +* cross: branch cross 4K or 2MB area */ struct perf_branch_entry {__u64 from;
@@ -1008,7 +1041,9 @@ struct perf_branch_entry { in_tx:1, /* in transaction */ abort:1, /* transaction abort */ cycles:16, /* cycle count to last branch */ - reserved:44; + type:4, /* branch type */ + cross:2, /* branch cross 4K or 2MB area */ + reserved:38; };#endif /* _UAPI_LINUX_PERF_EVENT_H */
Signature · 1 line
2.7.4
PATCH v2 2/5 perf/x86/intel: Record branch type
In reply to Jin YaoPerf already has support for disassembling the branch instruction and using the branch type for filtering. The patch just records the branch type in perf_branch_entry.
Before recording, the patch converts the x86 branch classification to common branch classification and compute for checking if the branches cross 4K or 2MB areas. It's an approximate computing for crossing 4K page or 2MB page.
--- arch/x86/events/intel/lbr.c | 106 +++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 105 insertions(+), 1 deletion(-)
arch/x86/events/intel/lbr.c +105 -1
@@ -109,6 +109,9 @@ enum {X86_BR_ZERO_CALL =1 <<15 ,/* zero length call */ X86_BR_CALL_STACK =1 <<16 ,/* call stack */ X86_BR_IND_JMP =1 <<17 ,/* indirect jump */ + + X86_BR_TYPE_SAVE = 1 << 18,/* indicate to save branch type */ + };#define X86_BR_PLM (X86_BR_USER | X86_BR_KERNEL)
@@ -139,6 +142,9 @@ enum { X86_BR_IRQ |\ X86_BR_INT) +#define AREA_4K 4096 +#define AREA_2M (2 * 1024 * 1024) +static void intel_pmu_lbr_filter (struct cpu_hw_events *cpuc);/*
@@ -670,6 +676,10 @@ static int intel_pmu_setup_sw_lbr_filter(struct perf_event *event)if (br_type &PERF_SAMPLE_BRANCH_CALL ) mask |=X86_BR_CALL |X86_BR_ZERO_CALL ; + +if (br_type &PERF_SAMPLE_BRANCH_TYPE_SAVE ) + mask |=X86_BR_TYPE_SAVE ; +/* * stash actual user request into reg, it may * be used by fixup code for some CPU
@@ -923,6 +933,84 @@ static int branch_type(unsigned long from, unsigned long to, int abort)return ret; } +static int +common_branch_type (int type,u64 from,u64 to) +{ +int ret; + + type = type & (~(X86_BR_KERNEL |X86_BR_USER )); + +switch (type) { +case X86_BR_CALL : +case X86_BR_ZERO_CALL : + ret =PERF_BR_CALL ; +break ; + +case X86_BR_RET : + ret =PERF_BR_RET ; +break ; + +case X86_BR_SYSCALL : + ret =PERF_BR_SYSCALL ; +break ; + +case X86_BR_SYSRET : + ret =PERF_BR_SYSRET ; +break ; + +case X86_BR_INT : + ret =PERF_BR_INT ; +break ; + +case X86_BR_IRET : + ret =PERF_BR_IRET ; +break ; + +case X86_BR_IRQ : + ret =PERF_BR_IRQ ; +break ; + +case X86_BR_ABORT : + ret =PERF_BR_FAR_BRANCH ; +break ; + +case X86_BR_JCC : +if (to > from) + ret =PERF_BR_JCC_FWD ; +else + ret =PERF_BR_JCC_BWD ; +break ; + +case X86_BR_JMP : + ret =PERF_BR_JMP ; +break ; + +case X86_BR_IND_CALL : + ret =PERF_BR_IND_CALL ; +break ; + +case X86_BR_IND_JMP : + ret =PERF_BR_IND_JMP ; +break ; + +default : + ret =PERF_BR_NONE ; + } + +return ret; +} + +static bool +cross_area (u64 addr1,u64 addr2,int size) +{ +u64 align1, align2; + + align1 = addr1 & ~(size -1 ); + align2 = addr2 & ~(size -1 ); + +return (align1 != align2) ? true : false; +} +/* * implement actual branch filter based on user demand. * Hardware may not exactly satisfy that request, thus
@@ -939,7 +1027,8 @@ intel_pmu_lbr_filter(struct cpu_hw_events *cpuc)bool compress = false;/* if sampling all branches, then nothing to filter */ -if ((br_sel &X86_BR_ALL ) ==X86_BR_ALL ) +if (((br_sel &X86_BR_ALL ) ==X86_BR_ALL ) && + ((br_sel &X86_BR_TYPE_SAVE ) !=X86_BR_TYPE_SAVE ))return ;for (i =0 ; i < cpuc->lbr_stack.nr; i++) {
@@ -960,6 +1049,21 @@ intel_pmu_lbr_filter(struct cpu_hw_events *cpuc) cpuc->lbr_entries[i].from =0 ; compress = true; } + +if ((br_sel &X86_BR_TYPE_SAVE ) ==X86_BR_TYPE_SAVE ) { + cpuc->lbr_entries[i].type =common_branch_type (type, + from, + to); +if (cross_area (from, to,AREA_2M )) + cpuc->lbr_entries[i].cross =PERF_BR_CROSS_2M ; +else if (cross_area (from, to,AREA_4K )) + cpuc->lbr_entries[i].cross =PERF_BR_CROSS_4K ; +else + cpuc->lbr_entries[i].cross =PERF_BR_CROSS_NONE ; + }else { + cpuc->lbr_entries[i].type =PERF_BR_NONE ; + cpuc->lbr_entries[i].cross =PERF_BR_CROSS_NONE ; + } }if (!compress)
Signature · 1 line
2.7.4
Peter Zijlstra
maintainerRe: [PATCH v2 2/5] perf/x86/intel: Record branch type
Quoted 8 lines · Jin Yao
Perf already has support for disassembling the branch instruction and using the branch type for filtering. The patch just records the branch type in
perf_branch_entry.Before recording, the patch converts the x86 branch classification to common branch classification and compute for checking if the branches cross 4K or 2MB areas. It's an approximate computing for crossing 4K page or 2MB page.
The changelog is completely empty of rationale. Why do we care?
Not having the binary is a very bad reason; you can't do much of anything if that's missing.
Quoted 41 lines
arch/x86/events/intel/lbr.c +37
-0@@ -923,6 +933,84 @@ static int branch_type(unsigned long from, unsigned long to, int abort)return ret; } +static int +common_branch_type (int type,u64 from,u64 to) +{ +int ret; + + type = type & (~(X86_BR_KERNEL |X86_BR_USER )); + +switch (type) { +case X86_BR_CALL : +case X86_BR_ZERO_CALL : + ret =PERF_BR_CALL ; +break ; + +case X86_BR_RET : + ret =PERF_BR_RET ; +break ; + +case X86_BR_SYSCALL : + ret =PERF_BR_SYSCALL ; +break ; + +case X86_BR_SYSRET : + ret =PERF_BR_SYSRET ; +break ; + +case X86_BR_INT : + ret =PERF_BR_INT ; +break ; + +case X86_BR_IRET : + ret =PERF_BR_IRET ; +break ; + +case X86_BR_IRQ : + ret =PERF_BR_IRQ ; +break ; +
arch/x86/events/intel/lbr.c static int branch_type(unsigned long from, unsigned long to, int abort) · line 982
Quoted 10 lines
arch/x86/events/intel/lbr.c +10
-0+case X86_BR_ABORT : + ret =PERF_BR_FAR_BRANCH ; +break ; + +case X86_BR_JCC : +if (to > from) + ret =PERF_BR_JCC_FWD ; +else + ret =PERF_BR_JCC_BWD ; +break ;
This seems like superfluous information; we already get to and from, so this comparison is pointless.
The rest looks like something you can simpler implement using a lookup table.
Quoted 49 lines
arch/x86/events/intel/lbr.c +34
-1+ +case X86_BR_JMP : + ret =PERF_BR_JMP ; +break ; + +case X86_BR_IND_CALL : + ret =PERF_BR_IND_CALL ; +break ; + +case X86_BR_IND_JMP : + ret =PERF_BR_IND_JMP ; +break ; + +default : + ret =PERF_BR_NONE ; + } + +return ret; +} + +static bool +cross_area (u64 addr1,u64 addr2,int size) +{ +u64 align1, align2; + + align1 = addr1 & ~(size -1 ); + align2 = addr2 & ~(size -1 ); + +return (align1 != align2) ? true : false; +} +/* * implement actual branch filter based on user demand. * Hardware may not exactly satisfy that request, thus @@ -939,7 +1027,8 @@ intel_pmu_lbr_filter(struct cpu_hw_events *cpuc)bool compress = false;/* if sampling all branches, then nothing to filter */ -if ((br_sel &X86_BR_ALL ) ==X86_BR_ALL ) +if (((br_sel &X86_BR_ALL ) ==X86_BR_ALL ) && + ((br_sel &X86_BR_TYPE_SAVE ) !=X86_BR_TYPE_SAVE ))return ;for (i =0 ; i < cpuc->lbr_stack.nr; i++) {@@ -960,6 +1049,21 @@ intel_pmu_lbr_filter(struct cpu_hw_events *cpuc) cpuc->lbr_entries[i].from =0 ; compress = true; } +
arch/x86/events/intel/lbr.c intel_pmu_lbr_filter(struct cpu_hw_events *cpuc) · line 1062
Quoted 10 lines
arch/x86/events/intel/lbr.c +10
-0+if ((br_sel &X86_BR_TYPE_SAVE ) ==X86_BR_TYPE_SAVE ) { + cpuc->lbr_entries[i].type =common_branch_type (type, + from, + to); +if (cross_area (from, to,AREA_2M )) + cpuc->lbr_entries[i].cross =PERF_BR_CROSS_2M ; +else if (cross_area (from, to,AREA_4K )) + cpuc->lbr_entries[i].cross =PERF_BR_CROSS_4K ; +else + cpuc->lbr_entries[i].cross =PERF_BR_CROSS_NONE ;
This again is superfluous information; it is already fully contained in to and from, which we have.
Quoted 10 lines
arch/x86/events/intel/lbr.c +4
-0+ }else { + cpuc->lbr_entries[i].type =PERF_BR_NONE ; + cpuc->lbr_entries[i].cross =PERF_BR_CROSS_NONE ; + } }if (!compress)Signature · 1 line
2.7.4
Re: [PATCH v2 2/5] perf/x86/intel: Record branch type
Quoted 10 lines · Peter Zijlstra
Quoted 8 lines · Jin Yao
Perf already has support for disassembling the branch instruction and using the branch type for filtering. The patch just records the branch type in
perf_branch_entry.Before recording, the patch converts the x86 branch classification to common branch classification and compute for checking if the branches cross 4K or 2MB areas. It's an approximate computing for crossing 4K page or 2MB page.
Quoted 4 lines
The changelog is completely empty of rationale. Why do we care?
Not having the binary is a very bad reason; you can't do much of anything if that's missing.
It's a somewhat common situation with partially JITed code, if you don't have an agent. You can still do a lot of useful things.
We found it useful to have this extra information during workload analysis. Forward conditionals and page crossing jumps are indications of frontend problems.
-Andi
Peter Zijlstra
maintainerRe: [PATCH v2 2/5] perf/x86/intel: Record branch type
Quoted 11 lines · Andi Kleen
Quoted 10 lines · Peter Zijlstra
Quoted 8 lines · Jin Yao
Perf already has support for disassembling the branch instruction and using the branch type for filtering. The patch just records the branch type in
perf_branch_entry.Before recording, the patch converts the x86 branch classification to common branch classification and compute for checking if the branches cross 4K or 2MB areas. It's an approximate computing for crossing 4K page or 2MB page.
Quoted 7 lines
Quoted 4 lines
The changelog is completely empty of rationale. Why do we care?
Not having the binary is a very bad reason; you can't do much of anything if that's missing.
It's a somewhat common situation with partially JITed code, if you don't have an agent. You can still do a lot of useful things.
Like what? How can you say anything about code you don't have?
Quoted 3 lines
We found it useful to have this extra information during workload analysis. Forward conditionals and page crossing jumps are indications of frontend problems.
But you already have the exact same information in {to,from}, why would you need to repackage information already contained?
Re: [PATCH v2 2/5] perf/x86/intel: Record branch type
Quoted 4 lines
Quoted 2 lines
It's a somewhat common situation with partially JITed code, if you don't have an agent. You can still do a lot of useful things.
Like what? How can you say anything about code you don't have?
For example if you combine the PMU topdown measurement, and see if it's frontend bound, and then you see it has lots of forward conditionals, then dynamic basic block reordering will help. If you have lots of cross page jumps then function reordering will help. etc.
Quoted 6 lines
Quoted 3 lines
We found it useful to have this extra information during workload analysis. Forward conditionals and page crossing jumps are indications of frontend problems.
But you already have the exact same information in {to,from}, why would you need to repackage information already contained?
Without this patch, we don't know if it's conditional or something else. And the kernel already knows this for its filtering, so it can as well report it.
Right the CROSS_* and forward backward information could be computed later.
-Andi
Re: [PATCH v2 2/5] perf/x86/intel: Record branch type
Quoted 10 lines
Without this patch, we don't know if it's conditional or something else. And the kernel already knows this for its filtering, so it can as well report it.
Right the CROSS_* and forward backward information could be computed later.
-Andi
OK, I will move CROSS_* and JCC forward/backward computing to user-space though it makes user-space code to be complicated.
Thanks
Jin Yao
PATCH v2 3/5 perf record: Create a new option save_type in --branch-filter
In reply to Jin YaoThe option indicates the kernel to save branch type during sampling.
One example:
perf record -g --branch-filter any,save_type <command>
--- tools/perf/Documentation/perf-record.txt | 1 + tools/perf/util/parse-branch-options.c | 1 + 2 files changed, 2 insertions(+)
tools/perf/Documentation/perf-record.txt +1 -0
@@ -332,6 +332,7 @@ following filters are defined: - no_tx: only when the target is not in a hardware transaction - abort_tx: only when the target is a hardware transaction abort - cond: conditional branches + - save_type: save branch type during sampling in case binary is not available later + The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
tools/perf/util/parse-branch-options.c +1 -0
@@ -28,6 +28,7 @@ static const struct branch_mode branch_modes[] = {BRANCH_OPT ("cond" ,PERF_SAMPLE_BRANCH_COND ),BRANCH_OPT ("ind_jmp" ,PERF_SAMPLE_BRANCH_IND_JUMP ),BRANCH_OPT ("call" ,PERF_SAMPLE_BRANCH_CALL ), +BRANCH_OPT ("save_type" ,PERF_SAMPLE_BRANCH_TYPE_SAVE ), BRANCH_END };
Signature · 1 line
2.7.4
PATCH v2 4/5 perf report: Show branch type statistics for stdio mode
In reply to Jin YaoShow the branch type statistics at the end of perf report --stdio.
For example:
perf report --stdio
JCC forward: 27.7%
JCC backward: 9.8%
JMP: 0.0%
IND_JMP: 6.5%
CALL: 26.6%
IND_CALL: 0.0%
RET: 29.3%
IRET: 0.0%
CROSS_4K: 0.0%
CROSS_2M: 14.3%The branch types are:
---------------------
JCC forward: Conditional forward jump
JCC backward: Conditional backward jump
JMP: Jump imm
IND_JMP: Jump reg/mem
CALL: Call imm
IND_CALL: Call reg/mem
RET: Ret
SYSCALL: Syscall
SYSRET: Syscall return
IRQ: HW interrupt/trap/fault
INT: SW interrupt
IRET: Return from interrupt
FAR_BRANCH: Others not generic branch type
CROSS_4K and CROSS_2M:
----------------------
They are the metrics checking for branches cross 4K or 2MB pages. It's an approximate computing. We don't know if the area is 4K or 2MB, so always compute both.
To make the output simple, if a branch crosses 2M area, CROSS_4K will not be incremented.
--- tools/perf/builtin-report.c | 212 ++++++++++++++++++++++++++++++++++++++++++++ tools/perf/util/event.h | 4 +- tools/perf/util/hist.c | 5 +- 3 files changed, 216 insertions(+), 5 deletions(-)
tools/perf/builtin-report.c +212 -0
@@ -43,6 +43,24 @@#include <linux/bitmap.h> #include <linux/stringify.h> +struct branch_type_stat { +u64 jcc_fwd; +u64 jcc_bwd; +u64 jmp; +u64 ind_jmp; +u64 call; +u64 ind_call; +u64 ret; +u64 syscall; +u64 sysret; +u64 irq; +u64 intr; +u64 iret; +u64 far_branch; +u64 cross_4k; +u64 cross_2m; +}; +struct report {struct perf_tool tool;struct perf_session *session;
@@ -66,6 +84,7 @@ struct report {u64 queue_size;int socket_filter;DECLARE_BITMAP (cpu_bitmap,MAX_NR_CPUS ); +struct branch_type_stat brtype_stat; };static int report__config (const char *var,const char *value,void *cb)
@@ -144,6 +163,91 @@ static int hist_iter__report_callback(struct hist_entry_iter *iter,return err; } +static void branch_type_count (struct report *rep,struct branch_info *bi) +{ +struct branch_type_stat *stat = &rep->brtype_stat; +struct branch_flags *flags = &bi->flags; + +switch (flags->type) { +case PERF_BR_JCC_FWD : + stat->jcc_fwd++; +break ; + +case PERF_BR_JCC_BWD : + stat->jcc_bwd++; +break ; + +case PERF_BR_JMP : + stat->jmp++; +break ; + +case PERF_BR_IND_JMP : + stat->ind_jmp++; +break ; + +case PERF_BR_CALL : + stat->call++; +break ; + +case PERF_BR_IND_CALL : + stat->ind_call++; +break ; + +case PERF_BR_RET : + stat->ret++; +break ; + +case PERF_BR_SYSCALL : + stat->syscall++; +break ; + +case PERF_BR_SYSRET : + stat->sysret++; +break ; + +case PERF_BR_IRQ : + stat->irq++; +break ; + +case PERF_BR_INT : + stat->intr++; +break ; + +case PERF_BR_IRET : + stat->iret++; +break ; + +case PERF_BR_FAR_BRANCH : + stat->far_branch++; +break ; + +default : +break ; + } + +if (flags->cross ==PERF_BR_CROSS_2M ) + stat->cross_2m++; +else if (flags->cross ==PERF_BR_CROSS_4K ) + stat->cross_4k++; +} + +static int hist_iter__branch_callback (struct hist_entry_iter *iter, +struct addr_location *al__maybe_unused , +bool single__maybe_unused , +void *arg) +{ +struct hist_entry *he = iter->he; +struct report *rep = arg; +struct branch_info *bi; + +if (sort__mode ==SORT_MODE__BRANCH ) { + bi = he->branch_info; +branch_type_count (rep, bi); + } + +return 0 ; +} + static int process_sample_event(struct perf_tool *tool, union perf_event *event, struct perf_sample *sample,
@@ -182,6 +286,8 @@ static int process_sample_event(struct perf_tool *tool,*/ if (!sample->branch_stack) goto out_put; + + iter.add_entry_cb = hist_iter__branch_callback; iter.ops = &hist_iter_branch; }else if (rep->mem_mode) { iter.ops = &hist_iter_mem;
@@ -369,6 +475,107 @@ static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct reportreturn ret +fprintf (fp,"\n#\n" ); } +static void branch_type_stat_display (FILE *fp,struct branch_type_stat *stat) +{ +u64 total =0 ; + + total += stat->jcc_fwd; + total += stat->jcc_bwd; + total += stat->jmp; + total += stat->ind_jmp; + total += stat->call; + total += stat->ind_call; + total += stat->ret; + total += stat->syscall; + total += stat->sysret; + total += stat->irq; + total += stat->intr; + total += stat->iret; + total += stat->far_branch; + +if (total ==0 ) +return ; + +fprintf (fp,"\n#" ); +fprintf (fp,"\n# Branch Statistics:" ); +fprintf (fp,"\n#" ); + +if (stat->jcc_fwd >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"JCC forward" , +100.0 * (double )stat->jcc_fwd / (double )total); + +if (stat->jcc_bwd >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"JCC backward" , +100.0 * (double )stat->jcc_bwd / (double )total); + +if (stat->jmp >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"JMP" , +100.0 * (double )stat->jmp / (double )total); + +if (stat->ind_jmp >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"IND_JMP" , +100.0 * (double )stat->ind_jmp / (double )total); + +if (stat->call >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"CALL" , +100.0 * (double )stat->call / (double )total); + +if (stat->ind_call >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"IND_CALL" , +100.0 * (double )stat->ind_call / (double )total); + +if (stat->ret >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"RET" , +100.0 * (double )stat->ret / (double )total); + +if (stat->syscall >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"SYSCALL" , +100.0 * (double )stat->syscall / (double )total); + +if (stat->sysret >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"SYSRET" , +100.0 * (double )stat->sysret / (double )total); + +if (stat->irq >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"IRQ" , +100.0 * (double )stat->irq / (double )total); + +if (stat->intr >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"INT" , +100.0 * (double )stat->intr / (double )total); + +if (stat->iret >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"IRET" , +100.0 * (double )stat->iret / (double )total); + +if (stat->far_branch >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"FAR_BRANCH" , +100.0 * (double )stat->far_branch / (double )total); + +if (stat->cross_4k >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"CROSS_4K" , +100.0 * (double )stat->cross_4k / (double )total); + +if (stat->cross_2m >0 ) +fprintf (fp,"\n%12s: %5.1f%%" , +"CROSS_2M" , +100.0 * (double )stat->cross_2m / (double )total); +} +static int perf_evlist__tty_browse_hists (struct perf_evlist *evlist,struct report *rep,const char *help)
@@ -404,6 +611,9 @@ static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,perf_read_values_destroy (&rep->show_threads_values); } +if (sort__mode ==SORT_MODE__BRANCH ) +branch_type_stat_display (stdout, &rep->brtype_stat); +return 0 ; }
@@ -936,6 +1146,8 @@ int cmd_report(int argc, const char **argv)if (has_br_stack && branch_call_mode) symbol_conf.show_branchflag_count = true; +memset (&report.brtype_stat,0 ,sizeof (struct branch_type_stat )); +/* * Branch mode is a tristate: * -1 means default, so decide based on the file having branch data.
tools/perf/util/event.h +3 -1
@@ -142,7 +142,9 @@ struct branch_flags { u64 in_tx:1; u64 abort:1; u64 cycles:16; - u64 reserved:44; + u64 type:4; + u64 cross:2; + u64 reserved:38; };struct branch_entry {
tools/perf/util/hist.c +1 -4
@@ -745,12 +745,9 @@ iter_prepare_branch_entry(struct hist_entry_iter *iter, struct addr_location *al }static int -iter_add_single_branch_entry (struct hist_entry_iter *iter, +iter_add_single_branch_entry (struct hist_entry_iter *iter__maybe_unused ,struct addr_location *al__maybe_unused ) { -/* to avoid calling callback function */ - iter->he =NULL ; -return 0 ; }
Signature · 1 line
2.7.4
PATCH v2 5/5 perf report: Show branch type in callchain entry
In reply to Jin YaoShow branch type in callchain entry. The branch type is printed with other LBR information (such as cycles/abort/...).
One example:
perf report --branch-history --stdio --no-children
-23.60%--main div.c:42 (RET cycles:2)
compute_flag div.c:28 (RET cycles:2) compute_flag div.c:27 (RET CROSS_2M cycles:1) rand rand.c:28 (RET CROSS_2M cycles:1) rand rand.c:28 (RET cycles:1) __random random.c:298 (RET cycles:1) __random random.c:297 (JCC forward cycles:1) __random random.c:295 (JCC forward cycles:1) __random random.c:295 (JCC forward cycles:1) __random random.c:295 (JCC forward cycles:1) __random random.c:295 (RET cycles:9)
--- tools/perf/util/callchain.c | 221 ++++++++++++++++++++++++++++++++------------ tools/perf/util/callchain.h | 20 ++++ 2 files changed, 182 insertions(+), 59 deletions(-)
tools/perf/util/callchain.c +162 -59
@@ -428,6 +428,89 @@ create_child(struct callchain_node *parent, bool inherit_children)return new; } +static const char *br_tag[BR_IDX_MAX ] = { +"JCC forward" , +"JCC backward" , +"JMP" , +"IND_JMP" , +"CALL" , +"IND_CALL" , +"RET" , +"SYSCALL" , +"SYSRET" , +"IRQ" , +"INT" , +"IRET" , +"FAR_BRANCH" , +"CROSS_4K" , +"CROSS_2M" , +}; + +static void +branch_type_count (int *counts,struct branch_flags *flags) +{ +switch (flags->type) { +case PERF_BR_JCC_FWD : + counts[BR_IDX_JCC_FWD ]++; +break ; + +case PERF_BR_JCC_BWD : + counts[BR_IDX_JCC_BWD ]++; +break ; + +case PERF_BR_JMP : + counts[BR_IDX_JMP ]++; +break ; + +case PERF_BR_IND_JMP : + counts[BR_IDX_IND_JMP ]++; +break ; + +case PERF_BR_CALL : + counts[BR_IDX_CALL ]++; +break ; + +case PERF_BR_IND_CALL : + counts[BR_IDX_IND_CALL ]++; +break ; + +case PERF_BR_RET : + counts[BR_IDX_RET ]++; +break ; + +case PERF_BR_SYSCALL : + counts[BR_IDX_SYSCALL ]++; +break ; + +case PERF_BR_SYSRET : + counts[BR_IDX_SYSRET ]++; +break ; + +case PERF_BR_IRQ : + counts[BR_IDX_IRQ ]++; +break ; + +case PERF_BR_INT : + counts[BR_IDX_INT ]++; +break ; + +case PERF_BR_IRET : + counts[BR_IDX_IRET ]++; +break ; + +case PERF_BR_FAR_BRANCH : + counts[BR_IDX_FAR_BRANCH ]++; +break ; + +default : +break ; + } + +if (flags->cross ==PERF_BR_CROSS_2M ) + counts[BR_IDX_CROSS_2M ]++; +else if (flags->cross ==PERF_BR_CROSS_4K ) + counts[BR_IDX_CROSS_4K ]++; +}/* * Fill the node with callchain values
@@ -467,6 +550,9 @@ fill_node(struct callchain_node *node, struct callchain_cursor *cursor) call->cycles_count = cursor_node->branch_flags.cycles; call->iter_count = cursor_node->nr_loop_iter; call->samples_count = cursor_node->samples; + +branch_type_count (call->brtype_count, + &cursor_node->branch_flags); }list_add_tail (&call->list, &node->val);
@@ -579,6 +665,9 @@ static enum match_result match_chain(struct callchain_cursor_node *node, cnode->cycles_count += node->branch_flags.cycles; cnode->iter_count += node->nr_loop_iter; cnode->samples_count += node->samples; + +branch_type_count (cnode->brtype_count, + &node->branch_flags); }return MATCH_EQ ;
@@ -1105,95 +1194,108 @@ int callchain_branch_counts(struct callchain_root *root, cycles_count); } +static int branch_type_str (int *counts,char *bf,int bfsize) +{ +int i, printed =0 ; +bool brace = false; + +for (i =0 ; i <BR_IDX_MAX ; i++) { +if (printed == bfsize -1 ) +return printed; + +if (counts[i] >0 ) { +if (!brace) { + brace = true; + printed +=scnprintf (bf + printed, + bfsize - printed, +" (%s" , br_tag[i]); + }else + printed +=scnprintf (bf + printed, + bfsize - printed, +" %s" , br_tag[i]); + } + } + +return printed; +} +static int counts_str_build (char *bf,int bfsize,u64 branch_count,u64 predicted_count,u64 abort_count,u64 cycles_count, -u64 iter_count,u64 samples_count) +u64 iter_count,u64 samples_count, +int *brtype_count) { -double predicted_percent =0.0 ; -const char *null_str ="" ; -char iter_str[32 ]; -char cycle_str[32 ]; -char *istr, *cstr;u64 cycles; +int printed, i =0 ;if (branch_count ==0 )return scnprintf (bf, bfsize," (calltrace)" ); + printed =branch_type_str (brtype_count, bf, bfsize); +if (printed) + i++; + cycles = cycles_count / branch_count; +if (cycles) { +if (i++) + printed +=scnprintf (bf + printed, bfsize - printed, +" cycles:%" PRId64"" , cycles); +else + printed +=scnprintf (bf + printed, bfsize - printed, +" (cycles:%" PRId64"" , cycles); + }if (iter_count && samples_count) { -if (cycles >0 ) -scnprintf (iter_str,sizeof (iter_str), -" iterations:%" PRId64"" , - iter_count / samples_count); +if (i++) + printed +=scnprintf (bf + printed, bfsize - printed, +" iterations:%" PRId64"" , + iter_count / samples_count);else -scnprintf (iter_str,sizeof (iter_str), -"iterations:%" PRId64"" , - iter_count / samples_count); - istr = iter_str; - }else - istr = (char *)null_str; - -if (cycles >0 ) { -scnprintf (cycle_str,sizeof (cycle_str), -"cycles:%" PRId64"" , cycles); - cstr = cycle_str; - }else - cstr = (char *)null_str; - - predicted_percent = predicted_count *100.0 / branch_count; + printed +=scnprintf (bf + printed, bfsize - printed, +" (iterations:%" PRId64"" , + iter_count / samples_count); + } -if ((predicted_count == branch_count) && (abort_count ==0 )) { -if ((cycles >0 ) || (istr != (char *)null_str)) -return scnprintf (bf, bfsize," (%s%s)" , cstr, istr); +if (predicted_count < branch_count) { +if (i++) + printed +=scnprintf (bf + printed, bfsize - printed, +" predicted:%.1f%%" , + predicted_count *100.0 / branch_count);else -return scnprintf (bf, bfsize,"%s" , (char *)null_str); - } - -if ((predicted_count < branch_count) && (abort_count ==0 )) { -if ((cycles >0 ) || (istr != (char *)null_str)) -return scnprintf (bf, bfsize, -" (predicted:%.1f%% %s%s)" , - predicted_percent, cstr, istr); -else { -return scnprintf (bf, bfsize, -" (predicted:%.1f%%)" , - predicted_percent); - } + printed +=scnprintf (bf + printed, bfsize - printed, +" (predicted:%.1f%%" , + predicted_count *100.0 / branch_count); } -if ((predicted_count == branch_count) && (abort_count >0 )) { -if ((cycles >0 ) || (istr != (char *)null_str)) -return scnprintf (bf, bfsize, -" (abort:%" PRId64" %s%s)" , - abort_count, cstr, istr); +if (abort_count) { +if (i++) + printed +=scnprintf (bf + printed, bfsize - printed, +" abort:%.1f%%" , + abort_count *100.0 / branch_count);else -return scnprintf (bf, bfsize, -" (abort:%" PRId64")" , - abort_count); + printed +=scnprintf (bf + printed, bfsize - printed, +" (abort:%.1f%%" , + abort_count *100.0 / branch_count); } -if ((cycles >0 ) || (istr != (char *)null_str)) -return scnprintf (bf, bfsize, -" (predicted:%.1f%% abort:%" PRId64" %s%s)" , - predicted_percent, abort_count, cstr, istr); +if (i) +return scnprintf (bf + printed, bfsize - printed,")" ); -return scnprintf (bf, bfsize, -" (predicted:%.1f%% abort:%" PRId64")" , - predicted_percent, abort_count); + bf[0 ] =0 ; +return 0 ; }static int callchain_counts_printf (FILE *fp,char *bf,int bfsize,u64 branch_count,u64 predicted_count,u64 abort_count,u64 cycles_count, -u64 iter_count,u64 samples_count) +u64 iter_count,u64 samples_count, +int *brtype_count) {char str[128 ];counts_str_build (str,sizeof (str), branch_count, predicted_count, abort_count, cycles_count, - iter_count, samples_count); + iter_count, samples_count, brtype_count);if (fp)return fprintf (fp,"%s" , str);
@@ -1225,7 +1327,8 @@ int callchain_list_counts__printf_value(struct callchain_node *node,return callchain_counts_printf (fp, bf, bfsize, branch_count, predicted_count, abort_count, - cycles_count, iter_count, samples_count); + cycles_count, iter_count, samples_count, + clist->brtype_count); }static void free_callchain_node (struct callchain_node *node)
tools/perf/util/callchain.h +20 -0
@@ -106,6 +106,25 @@ struct callchain_param {extern struct callchain_param callchain_param;extern struct callchain_param callchain_param_default; +enum { +BR_IDX_JCC_FWD =0 , +BR_IDX_JCC_BWD =1 , +BR_IDX_JMP =2 , +BR_IDX_IND_JMP =3 , +BR_IDX_CALL =4 , +BR_IDX_IND_CALL =5 , +BR_IDX_RET =6 , +BR_IDX_SYSCALL =7 , +BR_IDX_SYSRET =8 , +BR_IDX_IRQ =9 , +BR_IDX_INT =10 , +BR_IDX_IRET =11 , +BR_IDX_FAR_BRANCH =12 , +BR_IDX_CROSS_4K =13 , +BR_IDX_CROSS_2M =14 , +BR_IDX_MAX , +}; +struct callchain_list {u64 ip;struct map_symbol ms;
@@ -119,6 +138,7 @@ struct callchain_list {u64 cycles_count;u64 iter_count;u64 samples_count; +int brtype_count[BR_IDX_MAX ];char *srcline;struct list_head list; };
Signature · 1 line
2.7.4