perf_clear_branch_entry_bitfields() resets all struct perf_branch_entry bit
fields before capturing branch records. This resets remaining bit fields
except 'new_type', which is valid only when 'type' is PERF_BR_EXTEND_ABI.
Cc: Peter Zijlstra <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Mark Rutland <[email protected]>
Cc: Alexander Shishkin <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: [email protected]>
Cc: [email protected]
Signed-off-by: Anshuman Khandual <[email protected]>
---
This applies on v6.1-rc6
'perf_branch_entry.new_type' can remain uninitialized as explained earlier.
Also there is no PERF_BR_NEW_UNKNOWN to spare, because 'perf_branch_entry.
new_type' enumeration starts at PERF_BR_NEW_FAULT_ALGN, to save a position
for the extended branch types instead.
include/linux/perf_event.h | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 0031f7b4d9ab..c97b5f6f77a4 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -1110,8 +1110,9 @@ static inline void perf_clear_branch_entry_bitfields(struct perf_branch_entry *b
br->in_tx = 0;
br->abort = 0;
br->cycles = 0;
- br->type = 0;
+ br->type = PERF_BR_UNKNOWN;
br->spec = PERF_BR_SPEC_NA;
+ br->priv = PERF_BR_PRIV_UNKNOWN;
br->reserved = 0;
}
--
2.25.1
On 01/12/2022 05:51, Anshuman Khandual wrote:
> perf_clear_branch_entry_bitfields() resets all struct perf_branch_entry bit
> fields before capturing branch records. This resets remaining bit fields
> except 'new_type', which is valid only when 'type' is PERF_BR_EXTEND_ABI.
>
> Cc: Peter Zijlstra <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: Arnaldo Carvalho de Melo <[email protected]>
> Cc: Mark Rutland <[email protected]>
> Cc: Alexander Shishkin <[email protected]>
> Cc: Jiri Olsa <[email protected]>
> Cc: Namhyung Kim <[email protected]>
> Cc: [email protected]>
> Cc: [email protected]
> Signed-off-by: Anshuman Khandual <[email protected]>
> ---
> This applies on v6.1-rc6
>
> 'perf_branch_entry.new_type' can remain uninitialized as explained earlier.
> Also there is no PERF_BR_NEW_UNKNOWN to spare, because 'perf_branch_entry.
> new_type' enumeration starts at PERF_BR_NEW_FAULT_ALGN, to save a position
> for the extended branch types instead.
>
> include/linux/perf_event.h | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> index 0031f7b4d9ab..c97b5f6f77a4 100644
> --- a/include/linux/perf_event.h
> +++ b/include/linux/perf_event.h
> @@ -1110,8 +1110,9 @@ static inline void perf_clear_branch_entry_bitfields(struct perf_branch_entry *b
> br->in_tx = 0;
> br->abort = 0;
> br->cycles = 0;
> - br->type = 0;
> + br->type = PERF_BR_UNKNOWN;
> br->spec = PERF_BR_SPEC_NA;
> + br->priv = PERF_BR_PRIV_UNKNOWN;
> br->reserved = 0;
> }
I would vote for just memsetting the whole struct to 0 at this point and
making it work by ensuring the cleared from and to values are only set
after this function.
Or do the thing where it's wrapped in a union and the 'u64 value' member
is assigned 0. See union perf_mem_data_src. I don't know if this would
be a breaking change, but it doesn't look like it.
Currently this is a bit too fragile and the kind of bugs it will cause
are almost undetectable.
But as my proposal is an extra change on top of this:
Reviewed-by: James Clark <[email protected]>
James
>
On 12/1/22 11:21, Anshuman Khandual wrote:
> perf_clear_branch_entry_bitfields() resets all struct perf_branch_entry bit
> fields before capturing branch records. This resets remaining bit fields
> except 'new_type', which is valid only when 'type' is PERF_BR_EXTEND_ABI.
>
> Cc: Peter Zijlstra <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: Arnaldo Carvalho de Melo <[email protected]>
> Cc: Mark Rutland <[email protected]>
> Cc: Alexander Shishkin <[email protected]>
> Cc: Jiri Olsa <[email protected]>
> Cc: Namhyung Kim <[email protected]>
> Cc: [email protected]>
> Cc: [email protected]
> Signed-off-by: Anshuman Khandual <[email protected]>
> ---
> This applies on v6.1-rc6
>
> 'perf_branch_entry.new_type' can remain uninitialized as explained earlier.
> Also there is no PERF_BR_NEW_UNKNOWN to spare, because 'perf_branch_entry.
> new_type' enumeration starts at PERF_BR_NEW_FAULT_ALGN, to save a position
> for the extended branch types instead.
>
> include/linux/perf_event.h | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> index 0031f7b4d9ab..c97b5f6f77a4 100644
> --- a/include/linux/perf_event.h
> +++ b/include/linux/perf_event.h
> @@ -1110,8 +1110,9 @@ static inline void perf_clear_branch_entry_bitfields(struct perf_branch_entry *b
> br->in_tx = 0;
> br->abort = 0;
> br->cycles = 0;
> - br->type = 0;
> + br->type = PERF_BR_UNKNOWN;
> br->spec = PERF_BR_SPEC_NA;
> + br->priv = PERF_BR_PRIV_UNKNOWN;
> br->reserved = 0;
> }
>
Hello Peter,
Just wondering is there a chance that this patch could make it to v6.2-rc1 ?
- Anshuman