[v5.0-rc6 regression] "perf/core: Don't WARN() for impossible ring-buffer sizes"

[v5.0-rc6 regression] "perf/core: Don't WARN() for impossible ring-buffer sizes"

[Mathieu Poirier]

Hi Mark,

CoreSight trace collection is broken on v5.0-rc6 due to this commit:

9dff0aa95a32 perf/core: Don't WARN() for impossible ring-buffer sizes

Before:
root@juno:/home/linaro# perf record -e cs_etm/@20070000.etr/u
--per-thread  uname
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.036 MB perf.data ]
root@juno:/home/linaro#

After:
root@juno:/home/linaro# perf record -e cs_etm/@20070000.etr/u
--per-thread  uname
failed to mmap with 12 (Cannot allocate memory)
root@juno:/home/linaro#

The problem is related to the order_base_2() [1] test with a size of
1264, stemming from nr_pages equal to 128.  The combination yields an
order of 11, something leading directly to the error path.

The results are the same with linux-next 20190213.  This was tested on
a Juno R0 and R1 with a 4K page configuration.  I haven't tried but
I'm pretty sure it breaks IntelPT as well.

Please have a look when you have a minute.  Leo Yan and I will be
happy to test patches.

Thanks,
Mathieu

[1].  https://elixir.bootlin.com/linux/v5.0-rc6/source/kernel/events/ring_buffer.c#L737

Re: [v5.0-rc6 regression] "perf/core: Don't WARN() for impossible ring-buffer sizes"

[Leo Yan]

Hi Mathieu, Mark,

On Wed, Feb 13, 2019 at 02:26:25PM -0700, Mathieu Poirier wrote:
> Hi Mark,
> 
> CoreSight trace collection is broken on v5.0-rc6 due to this commit:
> 
> 9dff0aa95a32 perf/core: Don't WARN() for impossible ring-buffer sizes
> 
> Before:
> root@juno:/home/linaro# perf record -e cs_etm/@20070000.etr/u
> --per-thread  uname
> [ perf record: Woken up 1 times to write data ]
> [ perf record: Captured and wrote 0.036 MB perf.data ]
> root@juno:/home/linaro#
> 
> After:
> root@juno:/home/linaro# perf record -e cs_etm/@20070000.etr/u
> --per-thread  uname
> failed to mmap with 12 (Cannot allocate memory)
> root@juno:/home/linaro#
> 
> The problem is related to the order_base_2() [1] test with a size of
> 1264, stemming from nr_pages equal to 128.  The combination yields an
> order of 11, something leading directly to the error path.
> 
> The results are the same with linux-next 20190213.  This was tested on
> a Juno R0 and R1 with a 4K page configuration.  I haven't tried but
> I'm pretty sure it breaks IntelPT as well.

I noticed in another thread [1] Peter gave out the fixing for this
issue; applied this fixing on my local code and it fixes CoreSight
tracing issue.

Thanks,
Leo Yan

[1] https://lkml.org/lkml/2019/2/12/452