[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <9b764f1c-3c69-b825-5d5e-640d26be79e4@huawei.com>
Date: Wed, 2 Jun 2021 18:49:57 +0800
From: "liuqi (BA)" <liuqi115@...wei.com>
To: Will Deacon <will@...nel.org>
CC: Linuxarm <linuxarm@...wei.com>,
<linux-arm-kernel@...ts.infradead.org>,
<linux-kernel@...r.kernel.org>,
Peter Zijlstra <peterz@...radead.org>,
"Ingo Molnar" <mingo@...hat.com>,
Arnaldo Carvalho de Melo <acme@...nel.org>,
"Mark Rutland" <mark.rutland@....com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>
Subject: Re: [PATCH v2 1/9] perf: Add EVENT_ATTR_ID to simplify event
attributes
Hi Will,
On 2021/6/2 17:49, Will Deacon wrote:
> On Wed, Jun 02, 2021 at 04:45:23PM +0800, liuqi (BA) wrote:
>>
>> Hi Will,
>>
>> Thanks for reviewing this patch.
>>
>> On 2021/6/1 21:10, Will Deacon wrote:
>>> On Wed, May 19, 2021 at 05:51:51PM +0800, Qi Liu wrote:
>>>> Similar EVENT_ATTR macros are defined in many PMU drivers,
>>>> like HiSilicon PMU driver, Arm PMU driver, Arm SMMU PMU
>>>> driver. So Add a generic macro to simplify code.
>>>>
>>>> Cc: Peter Zijlstra <peterz@...radead.org>
>>>> Cc: Ingo Molnar <mingo@...hat.com>
>>>> Cc: Arnaldo Carvalho de Melo <acme@...nel.org>
>>>> Cc: Mark Rutland <mark.rutland@....com>
>>>> Cc: Alexander Shishkin <alexander.shishkin@...ux.intel.com>
>>>> Signed-off-by: Qi Liu <liuqi115@...wei.com>
>>>> ---
>>>> include/linux/perf_event.h | 6 ++++++
>>>> kernel/events/core.c | 2 ++
>>>> 2 files changed, 8 insertions(+)
>>>>
>>>> diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
>>>> index f5a6a2f..d0aa74e 100644
>>>> --- a/include/linux/perf_event.h
>>>> +++ b/include/linux/perf_event.h
>>>> @@ -1576,6 +1576,12 @@ static struct perf_pmu_events_attr _var = { \
>>>> .event_str = _str, \
>>>> };
>>>> +#define PMU_EVENT_ATTR_ID(_name, _id) \
>>>> + (&((struct perf_pmu_events_attr[]) { \
>>>> + { .attr = __ATTR(_name, 0444, perf_event_sysfs_show, NULL), \
>>>> + .id = _id, } \
>>>> + })[0].attr.attr)
>>>> +
>>>> #define PMU_FORMAT_ATTR(_name, _format) \
>>>> static ssize_t \
>>>> _name##_show(struct device *dev, \
>>>> diff --git a/kernel/events/core.c b/kernel/events/core.c
>>>> index 0ac818b..330d9cc 100644
>>>> --- a/kernel/events/core.c
>>>> +++ b/kernel/events/core.c
>>>> @@ -13295,6 +13295,8 @@ ssize_t perf_event_sysfs_show(struct device *dev, struct device_attribute *attr,
>>>> if (pmu_attr->event_str)
>>>> return sprintf(page, "%s\n", pmu_attr->event_str);
>>>> + else
>>>> + return sprintf(page, "config=%#llx\n", pmu_attr->id);
>>>
>>> I think it's a really bad idea to hardcode this here. For example, I think
>>> this patch series breaks user ABI for the SMMU PMU which used to print:
>>>
>>> "event=0x%02llx\n"
>>>
>>> and by the looks of it many of the other conversions are unsound too.
>>>
>> Got it, so I'll use pmu_attr->event_str here, for example,
>> SMMU_EVENT_ATTR(cycles, "event=0x00")
>
> You could, but honestly I don't really see this being any better than the
> current code. The advantage of using "event=0x%02llx\n" is that things are
> consistent by construction and therefore userspace can parse the information
> easily. We lose that if we just hardcode the string and it's error-prone to
> extend.
>
> Will
Got it, thanks.
So how about adding a common macro like this:
#define PMU_EVENT_ATTR_ID(_name,_func, _id) \
(&((struct perf_pmu_events_attr[]) { \
{ .attr = __ATTR(_name, 0444, _func, NULL), \
.id = _id, } \
})[0].attr.attr)
Drivers could define thir private macro, like SMMU_PMU_EVENT_ATTR, and
use their private event_show() functions if they need, or use the common
function perf_event_sysfs_show().
Thanks,
Qi
> .
>
Powered by blists - more mailing lists