[RFC PATCH] perf/x86/intel: downgrade alloc_bts_buffer() WARN() to pr_err()

[RFC PATCH] perf/x86/intel: downgrade alloc_bts_buffer() WARN() to pr_err()

[Sergey Senozhatsky]

We see alloc_bts_buffer() allocation failures under memory pressure,
while those allocation failures are non-fatal the WARN_ONCE() triggers
our fleet monitoring.  Given that alloc_bts_buffer() can be called from
various paths (e.g. syscall() or kvm_emulate_wrmsr()) and fail multiple
times, there seem to be little value in that one single backtrace (for
first allocation failure only).

Downgrade WARN_ONCE() to pr_err_once().

Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
---
 arch/x86/events/intel/ds.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/x86/events/intel/ds.c b/arch/x86/events/intel/ds.c
index 8940f0292229..29c24b2925a7 100644
--- a/arch/x86/events/intel/ds.c
+++ b/arch/x86/events/intel/ds.c
@@ -918,7 +918,7 @@ static int alloc_bts_buffer(int cpu)
 
 	buffer = dsalloc_pages(BTS_BUFFER_SIZE, GFP_KERNEL | __GFP_NOWARN, cpu);
 	if (unlikely(!buffer)) {
-		WARN_ONCE(1, "%s: BTS buffer allocation failure\n", __func__);
+		pr_err_once(1, "%s: BTS buffer allocation failure\n", __func__);
 		return -ENOMEM;
 	}
 	hwev->ds_bts_vaddr = buffer;
-- 
2.55.0.897.gb25b4bd76c-goog

Re: [RFC PATCH] perf/x86/intel: downgrade alloc_bts_buffer() WARN() to pr_err()

[Sergey Senozhatsky]

On (26/08/27 15:55), Sergey Senozhatsky wrote:
> diff --git a/arch/x86/events/intel/ds.c b/arch/x86/events/intel/ds.c
> index 8940f0292229..29c24b2925a7 100644
> --- a/arch/x86/events/intel/ds.c
> +++ b/arch/x86/events/intel/ds.c
> @@ -918,7 +918,7 @@ static int alloc_bts_buffer(int cpu)
>  
>  	buffer = dsalloc_pages(BTS_BUFFER_SIZE, GFP_KERNEL | __GFP_NOWARN, cpu);
>  	if (unlikely(!buffer)) {
> -		WARN_ONCE(1, "%s: BTS buffer allocation failure\n", __func__);
> +		pr_err_once(1, "%s: BTS buffer allocation failure\n", __func__);

                           ^^^  compile-testing patches?  Haven't heard of that...

Re: [RFC PATCH] perf/x86/intel: downgrade alloc_bts_buffer() WARN() to pr_err()

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] Replacing `WARN_ONCE(1, ...)` with `pr_err_once(1, ...)` without removing the condition argument causes a build failure.
--

commit e234f9d8f1cd00344467d9d8095e6f5f291f3900
Author: Sergey Senozhatsky <senozhatsky@chromium.org>

perf/x86/intel: downgrade alloc_bts_buffer() WARN() to pr_err()

This commit downgrades the BTS buffer allocation failure warning from
WARN_ONCE to pr_err_once to prevent triggering fleet monitoring alarms
under memory pressure.

> diff --git a/arch/x86/events/intel/ds.c b/arch/x86/events/intel/ds.c
> index 91a093d8cf2e7..02ec21ebb7af2 100644
> --- a/arch/x86/events/intel/ds.c
> +++ b/arch/x86/events/intel/ds.c
> @@ -918,7 +918,7 @@ static int alloc_bts_buffer(int cpu)
>  
>  	buffer = dsalloc_pages(BTS_BUFFER_SIZE, GFP_KERNEL | __GFP_NOWARN, cpu);
>  	if (unlikely(!buffer)) {
> -		WARN_ONCE(1, "%s: BTS buffer allocation failure\n", __func__);
> +		pr_err_once(1, "%s: BTS buffer allocation failure\n", __func__);

[Severity: Low]
Does this cause a build failure? Unlike WARN_ONCE which takes a condition
as its first argument, pr_err_once expects the format string as its first
argument. Leaving the integer 1 here will attempt to concatenate the KERN_ERR
macro string literal with an integer during expansion, which usually results
in a compile-time syntax error.

>  		return -ENOMEM;
>  	}
>  	hwev->ds_bts_vaddr = buffer;

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260827065529.3835720-1-senozhatsky@chromium.org?part=1

Re: [RFC PATCH] perf/x86/intel: downgrade alloc_bts_buffer() WARN() to pr_err()

[Sergey Senozhatsky]

On (26/08/27 15:59), Sergey Senozhatsky wrote:
> On (26/08/27 15:55), Sergey Senozhatsky wrote:
> > diff --git a/arch/x86/events/intel/ds.c b/arch/x86/events/intel/ds.c
> > index 8940f0292229..29c24b2925a7 100644
> > --- a/arch/x86/events/intel/ds.c
> > +++ b/arch/x86/events/intel/ds.c
> > @@ -918,7 +918,7 @@ static int alloc_bts_buffer(int cpu)
> >  
> >  	buffer = dsalloc_pages(BTS_BUFFER_SIZE, GFP_KERNEL | __GFP_NOWARN, cpu);
> >  	if (unlikely(!buffer)) {
> > -		WARN_ONCE(1, "%s: BTS buffer allocation failure\n", __func__);
> > +		pr_err_once(1, "%s: BTS buffer allocation failure\n", __func__);

I guess also can just remove it altogether.