Hello,

On Tue, May 06, 2025 at 09:02:20PM +0200, Salvatore Bonaccorso wrote:

Hi,

Pasi Kallinen reported in Debian a regression with perf r5101c4 counter, initially it was found in https://github.com/rr-debugger/rr/issues/3949 but said to be a kernel problem.

What's the '51' part? I don't think it's defined.

$ grep . -r /sys/bus/event_source/devices/cpu/format/ /sys/bus/event_source/devices/cpu/format/event:config:0-7 /sys/bus/event_source/devices/cpu/format/pc:config:19 /sys/bus/event_source/devices/cpu/format/edge:config:18 /sys/bus/event_source/devices/cpu/format/offcore_rsp:config1:0-63 /sys/bus/event_source/devices/cpu/format/ldlat:config1:0-15 /sys/bus/event_source/devices/cpu/format/inv:config:23 /sys/bus/event_source/devices/cpu/format/umask:config:8-15 /sys/bus/event_source/devices/cpu/format/frontend:config1:0-23 /sys/bus/event_source/devices/cpu/format/cmask:config:24-31

Nothing for bit 16, 20 and 22 on the 'config' field.

Is it possible to fix rr to use a correct event encoding instead?

Thanks, Namhyung

On Tue, May 06, 2025 at 07:18:39PM +0300, Pasi Kallinen wrote:

...

Package: src:linux Version: 6.12.25-1 Severity: normal X-Debbugs-Cc: debian-amd64@lists.debian.org, paxed@alt.org User: debian-amd64@lists.debian.org Usertags: amd64

Dear Maintainer,

perf stat -e r5101c4 true

reports "not supported".

The counters worked in kernel 6.11.10.

I first noticed this not working when updating to 6.12.22. Booting back to 6.11.10, the counters work correctly.

Does this ring a bell?

Would you be able to bisect the changes to identify where the behaviour changed?

Regards, Salvatore