2023-10-09 05:18:14

by Anshuman Khandual

Subject: [PATCH] driver: perf: arm_pmuv3: Identify 'event->attr.config1' based attributes

This defines macros to identify long and rdpmc attributes which get carried
inside 'event->attr.config1', thus making the code explicit and clear.

Cc: Will Deacon <[email protected]>
Cc: Mark Rutland <[email protected]>
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Anshuman Khandual <[email protected]>
---
This applies on v6.6-rc5.

drivers/perf/arm_pmuv3.c | 7 +++++--
1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/perf/arm_pmuv3.c b/drivers/perf/arm_pmuv3.c
index 94723d00548e..797ffd5a8fed 100644
--- a/drivers/perf/arm_pmuv3.c
+++ b/drivers/perf/arm_pmuv3.c
@@ -298,16 +298,19 @@ PMU_FORMAT_ATTR(event, "config:0-15");
PMU_FORMAT_ATTR(long, "config1:0");
PMU_FORMAT_ATTR(rdpmc, "config1:1");

+#define ARM_PMUV3_ATTR_LONG 0x01
+#define ARM_PMUV3_ATTR_RDPMC 0x02
+
static int sysctl_perf_user_access __read_mostly;

static inline bool armv8pmu_event_is_64bit(struct perf_event *event)
{
- return event->attr.config1 & 0x1;
+ return event->attr.config1 & ARM_PMUV3_ATTR_LONG;
}

static inline bool armv8pmu_event_want_user_access(struct perf_event *event)
{
- return event->attr.config1 & 0x2;
+ return event->attr.config1 & ARM_PMUV3_ATTR_RDPMC;
}

static struct attribute *armv8_pmuv3_format_attrs[] = {
--
2.25.1

2023-10-13 13:47:56

by Mark Rutland

Subject: Re: [PATCH] driver: perf: arm_pmuv3: Identify 'event->attr.config1' based attributes

On Mon, Oct 09, 2023 at 10:47:53AM +0530, Anshuman Khandual wrote:
> This defines macros to identify long and rdpmc attributes which get carried
> inside 'event->attr.config1', thus making the code explicit and clear.
>
> Cc: Will Deacon <[email protected]>
> Cc: Mark Rutland <[email protected]>
> Cc: [email protected]
> Cc: [email protected]
> Signed-off-by: Anshuman Khandual <[email protected]>
> ---
> This applies on v6.6-rc5.
>
> drivers/perf/arm_pmuv3.c | 7 +++++--
> 1 file changed, 5 insertions(+), 2 deletions(-)

I think this would make sense to do if we also used a common definition of
these bits in the helper functions *and* their format attributes, but as-is
this does nothing to relate the two, and I don't think that it improves
legibility on its own.

Mark.

> diff --git a/drivers/perf/arm_pmuv3.c b/drivers/perf/arm_pmuv3.c
> index 94723d00548e..797ffd5a8fed 100644
> --- a/drivers/perf/arm_pmuv3.c
> +++ b/drivers/perf/arm_pmuv3.c
> @@ -298,16 +298,19 @@ PMU_FORMAT_ATTR(event, "config:0-15");
> PMU_FORMAT_ATTR(long, "config1:0");
> PMU_FORMAT_ATTR(rdpmc, "config1:1");
>
> +#define ARM_PMUV3_ATTR_LONG 0x01
> +#define ARM_PMUV3_ATTR_RDPMC 0x02
> +
> static int sysctl_perf_user_access __read_mostly;
>
> static inline bool armv8pmu_event_is_64bit(struct perf_event *event)
> {
> - return event->attr.config1 & 0x1;
> + return event->attr.config1 & ARM_PMUV3_ATTR_LONG;
> }
>
> static inline bool armv8pmu_event_want_user_access(struct perf_event *event)
> {
> - return event->attr.config1 & 0x2;
> + return event->attr.config1 & ARM_PMUV3_ATTR_RDPMC;
> }
>
> static struct attribute *armv8_pmuv3_format_attrs[] = {
> --
> 2.25.1
>

2023-10-16 03:18:35

by Anshuman Khandual

Subject: Re: [PATCH] driver: perf: arm_pmuv3: Identify 'event->attr.config1' based attributes

On 10/13/23 19:17, Mark Rutland wrote:
> On Mon, Oct 09, 2023 at 10:47:53AM +0530, Anshuman Khandual wrote:
>> This defines macros to identify long and rdpmc attributes which get carried
>> inside 'event->attr.config1', thus making the code explicit and clear.
>>
>> Cc: Will Deacon <[email protected]>
>> Cc: Mark Rutland <[email protected]>
>> Cc: [email protected]
>> Cc: [email protected]
>> Signed-off-by: Anshuman Khandual <[email protected]>
>> ---
>> This applies on v6.6-rc5.
>>
>> drivers/perf/arm_pmuv3.c | 7 +++++--
>> 1 file changed, 5 insertions(+), 2 deletions(-)

I am not sure, if I understand your concerns and/or suggestions properly.

>
> I think this would make sense to do if we also used a common definition of
> these bits in the helper functions *and* their format attributes, but as-is

Common definitions of sysfs formats ? Which helper - linux/perf/arm_pmuv3.h ?

> this does nothing to relate the two, and I don't think that it improves
> legibility on its own.

Currently armv8pmu_event_is_64bit() and armv8pmu_event_want_user_access()
extracts required values from 'event->attr.config1' with hard coded masks
without any explicit co-relation with PMU_FORMAT_ATTR() described formats
above, unless some one really looks into the config1 bits layout.

Could you please provide some more details on how to go about making this
change better.

>
> Mark.
>
>> diff --git a/drivers/perf/arm_pmuv3.c b/drivers/perf/arm_pmuv3.c
>> index 94723d00548e..797ffd5a8fed 100644
>> --- a/drivers/perf/arm_pmuv3.c
>> +++ b/drivers/perf/arm_pmuv3.c
>> @@ -298,16 +298,19 @@ PMU_FORMAT_ATTR(event, "config:0-15");
>> PMU_FORMAT_ATTR(long, "config1:0");
>> PMU_FORMAT_ATTR(rdpmc, "config1:1");
>>
>> +#define ARM_PMUV3_ATTR_LONG 0x01
>> +#define ARM_PMUV3_ATTR_RDPMC 0x02
>> +
>> static int sysctl_perf_user_access __read_mostly;
>>
>> static inline bool armv8pmu_event_is_64bit(struct perf_event *event)
>> {
>> - return event->attr.config1 & 0x1;
>> + return event->attr.config1 & ARM_PMUV3_ATTR_LONG;
>> }
>>
>> static inline bool armv8pmu_event_want_user_access(struct perf_event *event)
>> {
>> - return event->attr.config1 & 0x2;
>> + return event->attr.config1 & ARM_PMUV3_ATTR_RDPMC;
>> }
>>
>> static struct attribute *armv8_pmuv3_format_attrs[] = {
>> --
>> 2.25.1
>>