How to use perf to log page faults

Taras Glek

unread,

May 12, 2010, 3:40:02 PM5/12/10

to

Hi,
From the docs it seems that perf may be useful for logging page faults.
I tried
perf record -g -d -f -e page-faults <mycommand>
but all I got were PERF_RECORD_MMAP events.

I'm using 2.6.32. Ideally I'd like to collect fault addresses and
userspace stacks that caused them. Isn't this what the page-fault event
is for?

Taras
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majo...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Frederic Weisbecker

unread,

May 12, 2010, 4:10:02 PM5/12/10

to

On Wed, May 12, 2010 at 12:38:30PM -0700, Taras Glek wrote:
> Hi,
> From the docs it seems that perf may be useful for logging page faults.
> I tried
> perf record -g -d -f -e page-faults <mycommand>
> but all I got were PERF_RECORD_MMAP events.
>
> I'm using 2.6.32. Ideally I'd like to collect fault addresses and
> userspace stacks that caused them. Isn't this what the page-fault event
> is for?
>
> Taras

Could you try with -c 1 ?

Otherwise, does that work on the latest kernel?

Thanks.

Andi Kleen

unread,

May 12, 2010, 4:10:01 PM5/12/10

to

Taras Glek <tg...@mozilla.com> writes:

> Hi,
> From the docs it seems that perf may be useful for logging page
> faults. I tried
> perf record -g -d -f -e page-faults <mycommand>
> but all I got were PERF_RECORD_MMAP events.
>
> I'm using 2.6.32. Ideally I'd like to collect fault addresses and
> userspace stacks that caused them. Isn't this what the page-fault
> event is for?

Some time ago I used systemtap to do this for my pbitmaps
works. This was the old systemtap script.

-Andi

probe vm.pagefault {
if (task_execname(task_current()) == "executable" && address < 0x100000000) {
printf("%u\n", address);
}

}

--
a...@linux.intel.com -- Speaking for myself only.

Taras Glek

unread,

May 12, 2010, 5:20:01 PM5/12/10

to

On 05/12/2010 12:59 PM, Frederic Weisbecker wrote:
> On Wed, May 12, 2010 at 12:38:30PM -0700, Taras Glek wrote:
>
>> Hi,
>> From the docs it seems that perf may be useful for logging page faults.
>> I tried
>> perf record -g -d -f -e page-faults<mycommand>
>> but all I got were PERF_RECORD_MMAP events.
>>
>> I'm using 2.6.32. Ideally I'd like to collect fault addresses and
>> userspace stacks that caused them. Isn't this what the page-fault event
>> is for?
>>
>> Taras
>>
>
> Could you try with -c 1 ?
>
> Otherwise, does that work on the latest kernel?
>

this just gets me PERF_RECORD_SAMPLE events which don't seem to be
related to page-faults(ie they show up with/without -e page-faults).

I don't have the latest kernel installed here. Can you get perf
page-fault events on your system?

Taras

Frederic Weisbecker

unread,

May 12, 2010, 5:30:03 PM5/12/10

to

On Wed, May 12, 2010 at 02:18:08PM -0700, Taras Glek wrote:
> On 05/12/2010 12:59 PM, Frederic Weisbecker wrote:
>> On Wed, May 12, 2010 at 12:38:30PM -0700, Taras Glek wrote:
>>
>>> Hi,
>>> From the docs it seems that perf may be useful for logging page faults.
>>> I tried
>>> perf record -g -d -f -e page-faults<mycommand>
>>> but all I got were PERF_RECORD_MMAP events.
>>>
>>> I'm using 2.6.32. Ideally I'd like to collect fault addresses and
>>> userspace stacks that caused them. Isn't this what the page-fault event
>>> is for?
>>>
>>> Taras
>>>
>>
>> Could you try with -c 1 ?
>>
>> Otherwise, does that work on the latest kernel?
>>
> this just gets me PERF_RECORD_SAMPLE events which don't seem to be
> related to page-faults(ie they show up with/without -e page-faults).
>
> I don't have the latest kernel installed here. Can you get perf
> page-fault events on your system?

Yeah, no problem on 2.6.34-rc7.