[permalink] [raw]

Subject: Re: [RFC V1 11/11] perf: Capture branch privilege information

On 1/25/22 9:09 PM, James Clark wrote:
>
>
> On 24/01/2022 04:30, Anshuman Khandual wrote:
>> Platforms like arm64 could capture privilege level information for all the
>> branch records. Hence this adds a new element in the struct branch_entry to
>> record the privilege level information, which could be requested through a
>> new event.attr.branch_sample_type flag PERF_SAMPLE_BRANCH_PRIV_SAVE. While
>> here, update the BRBE driver as required.
>>
>> Cc: Will Deacon <[email protected]>
>> Cc: Mark Rutland <[email protected]>
>> Cc: Peter Zijlstra <[email protected]>
>> Cc: Ingo Molnar <[email protected]>
>> Cc: Arnaldo Carvalho de Melo <[email protected]>
>> Cc: Jiri Olsa <[email protected]>
>> Cc: Namhyung Kim <[email protected]>
>> Cc: [email protected]
>> Cc: [email protected]
>> Cc: [email protected]
>> Signed-off-by: Anshuman Khandual <[email protected]>
>> ---
>> drivers/perf/arm_pmu_brbe.c | 28 ++++++++++++++++++++++++
>> include/linux/perf_event.h | 5 +++++
>> include/uapi/linux/perf_event.h | 13 ++++++++++-
>> tools/include/uapi/linux/perf_event.h | 13 ++++++++++-
>> tools/perf/Documentation/perf-record.txt | 1 +
>> tools/perf/util/parse-branch-options.c | 1 +
>> 6 files changed, 59 insertions(+), 2 deletions(-)
>>
> [...]
>> diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
>> index 361fdc6b87a0..4d77710f7a4e 100644
>> --- a/include/uapi/linux/perf_event.h
>> +++ b/include/uapi/linux/perf_event.h
>> @@ -204,6 +204,8 @@ enum perf_branch_sample_type_shift {
>>
>> PERF_SAMPLE_BRANCH_HW_INDEX_SHIFT = 17, /* save low level index of raw branch records */
>>
>> + PERF_SAMPLE_BRANCH_PRIV_SAVE_SHIFT = 18, /* save privillege mode */
>
> privillege -> privilege

Fixed.

>
>> +
>> PERF_SAMPLE_BRANCH_MAX_SHIFT /* non-ABI */
>> };
>>
>> @@ -233,6 +235,8 @@ enum perf_branch_sample_type {
>>
>> PERF_SAMPLE_BRANCH_HW_INDEX = 1U << PERF_SAMPLE_BRANCH_HW_INDEX_SHIFT,
>>
>> + PERF_SAMPLE_BRANCH_PRIV_SAVE = 1U << PERF_SAMPLE_BRANCH_PRIV_SAVE_SHIFT,
>> +
>
> Can you also add the new entry to the perf verbose printer otherwise it looks like
> it's not being set on branch_sample_type
> (as in when using ./perf record -j any_call,u,priv -vvv):

Sure, will do.

>
> static void __p_branch_sample_type(char *buf, size_t size, u64 value)
> {
> #define bit_name(n) { PERF_SAMPLE_BRANCH_##n, #n }
> struct bit_names bits[] = {
> bit_name(USER), bit_name(KERNEL), bit_name(HV), bit_name(ANY),
> bit_name(ANY_CALL), bit_name(ANY_RETURN), bit_name(IND_CALL),
> bit_name(ABORT_TX), bit_name(IN_TX), bit_name(NO_TX),
> bit_name(COND), bit_name(CALL_STACK), bit_name(IND_JUMP),
> bit_name(CALL), bit_name(NO_FLAGS), bit_name(NO_CYCLES),
> bit_name(HW_INDEX),
> { .name = NULL, }
> };
> #undef bit_name
> __p_bits(buf, size, value, bits);
> }
>
> PERF_SAMPLE_BRANCH_TYPE_SAVE_SHIFT is also missing so it probably makes sense to add it
> at the same time as that was expanded for BRBE.

Posted a patch for PERF_SAMPLE_BRANCH_TYPE_SAVE_SHIFT here but will update
for the new 'priv' request via this series.

https://lore.kernel.org/all/[email protected]/

2022-03-14 14:48:09

by Anshuman Khandual

[permalink] [raw]

Subject: Re: [RFC V1 11/11] perf: Capture branch privilege information

On 1/26/22 22:57, James Clark wrote:
>
> On 24/01/2022 04:30, Anshuman Khandual wrote:
>> Platforms like arm64 could capture privilege level information for all the
>> branch records. Hence this adds a new element in the struct branch_entry to
>> record the privilege level information, which could be requested through a
>> new event.attr.branch_sample_type flag PERF_SAMPLE_BRANCH_PRIV_SAVE. While
>> here, update the BRBE driver as required.
>>
>> Cc: Will Deacon <[email protected]>
>> Cc: Mark Rutland <[email protected]>
>> Cc: Peter Zijlstra <[email protected]>
>> Cc: Ingo Molnar <[email protected]>
>> Cc: Arnaldo Carvalho de Melo <[email protected]>
>> Cc: Jiri Olsa <[email protected]>
>> Cc: Namhyung Kim <[email protected]>
>> Cc: [email protected]
>> Cc: [email protected]
>> Cc: [email protected]
>> Signed-off-by: Anshuman Khandual <[email protected]>
>> ---
>> drivers/perf/arm_pmu_brbe.c | 28 ++++++++++++++++++++++++
>> include/linux/perf_event.h | 5 +++++
>> include/uapi/linux/perf_event.h | 13 ++++++++++-
>> tools/include/uapi/linux/perf_event.h | 13 ++++++++++-
>> tools/perf/Documentation/perf-record.txt | 1 +
>> tools/perf/util/parse-branch-options.c | 1 +
>> 6 files changed, 59 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/perf/arm_pmu_brbe.c b/drivers/perf/arm_pmu_brbe.c
>> index 7cd1208c6c58..d4cbea74c148 100644
>> --- a/drivers/perf/arm_pmu_brbe.c
>> +++ b/drivers/perf/arm_pmu_brbe.c
>> @@ -270,6 +270,25 @@ static int brbe_fetch_perf_type(u64 brbinf)
>> }
>> }
>>
>> +static int brbe_fetch_perf_priv(u64 brbinf)
>> +{
>> + int brbe_el = brbe_fetch_el(brbinf);
>> +
>> + switch (brbe_el) {
>> + case BRBINF_EL_EL0:
>> + return PERF_BR_USER;
>> + case BRBINF_EL_EL1:
>> + return PERF_BR_KERNEL;
>> + case BRBINF_EL_EL2:
>> + if (is_kernel_in_hyp_mode())
>> + return PERF_BR_KERNEL;
>> + return PERF_BR_HV;
>> + default:
>> + pr_warn("unknown branch privilege captured\n");
>> + return -1;
>> + }
>> +}
>> +
>> static void capture_brbe_flags(struct pmu_hw_events *cpuc, struct perf_event *event,
>> u64 brbinf, int idx)
>> {
>> @@ -302,6 +321,15 @@ static void capture_brbe_flags(struct pmu_hw_events *cpuc, struct perf_event *ev
>> cpuc->brbe_entries[idx].in_tx = brbinf & BRBINF_TX;
>> }
>> }
>> +
>> + if (branch_sample_priv(event)) {
>> + /*
>> + * All these information (i.e branch privilege level) are not
>> + * available for source only branch records.
>> + */
>> + if (type != BRBINF_VALID_SOURCE)
>> + cpuc->brbe_entries[idx].priv = brbe_fetch_perf_priv(brbinf);
>> + }
>> }
>>
>> /*
>> diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
>> index 916ce5102b33..8021b6a30d86 100644
>> --- a/include/linux/perf_event.h
>> +++ b/include/linux/perf_event.h
>> @@ -1688,4 +1688,9 @@ static inline bool branch_sample_hw_index(const struct perf_event *event)
>> {
>> return event->attr.branch_sample_type & PERF_SAMPLE_BRANCH_HW_INDEX;
>> }
>> +
>> +static inline bool branch_sample_priv(const struct perf_event *event)
>> +{
>> + return event->attr.branch_sample_type & PERF_SAMPLE_BRANCH_PRIV_SAVE;
>> +}
>> #endif /* _LINUX_PERF_EVENT_H */
>> diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
>> index 361fdc6b87a0..4d77710f7a4e 100644
>> --- a/include/uapi/linux/perf_event.h
>> +++ b/include/uapi/linux/perf_event.h
>> @@ -204,6 +204,8 @@ enum perf_branch_sample_type_shift {
>>
>> PERF_SAMPLE_BRANCH_HW_INDEX_SHIFT = 17, /* save low level index of raw branch records */
>>
>> + PERF_SAMPLE_BRANCH_PRIV_SAVE_SHIFT = 18, /* save privillege mode */
>> +
>> PERF_SAMPLE_BRANCH_MAX_SHIFT /* non-ABI */
>> };
>>
>> @@ -233,6 +235,8 @@ enum perf_branch_sample_type {
>>
>> PERF_SAMPLE_BRANCH_HW_INDEX = 1U << PERF_SAMPLE_BRANCH_HW_INDEX_SHIFT,
>>
>> + PERF_SAMPLE_BRANCH_PRIV_SAVE = 1U << PERF_SAMPLE_BRANCH_PRIV_SAVE_SHIFT,
>> +
>> PERF_SAMPLE_BRANCH_MAX = 1U << PERF_SAMPLE_BRANCH_MAX_SHIFT,
>> };
>>
>> @@ -265,6 +269,12 @@ enum {
>> PERF_BR_MAX,
>> };
>>
>> +enum {
>> + PERF_BR_USER = 0,
>> + PERF_BR_KERNEL = 1,
>> + PERF_BR_HV = 2,
>> +};
>> +
> Can we have 0 as "UNKNOWN". It's going to be difficult to parse files when privilege information
> isn't saved and get accurate results without that. For example if it's not set then presumably
> the field would be 0 (PERF_BR_USER), but that doesn't mean the samples are user in that case.
>
> I know you might be able to go backwards and look at what arguments were passed to the kernel but
> it's not guaranteed that the kernel honored the request anyway. There are also other platforms
> to think about etc.
>
> If you look at the branch type definitions above they start at 0 (PERF_BR_UNKNOWN) which I think
> works out quite nicely in the userspace code.

This is being taken care in the new BRBE related perf ABI changes series (V3).

https://lore.kernel.org/all/[email protected]/