Hi Reinette,
On 5/27/2026 10:24 AM, Reinette Chatre wrote:
Hi Chenyu,
On 5/26/26 6:57 PM, Chen, Yu C wrote:
My thinking was that shareable_bits only describes portions that may
be shared with non-CPU agents, and on an otherwise idle test system
the CAT miss-rate trend can still be observed even if all ways are
marked shareable. But I agree that this is not a good assumption for
kselftest pass/fail.
So I agree that falling back to the full CBM is not the right fix for
this test. I will change the patch to skip L3_CAT when no exclusive
cache portion is available, with a message explaining that the performance
test cannot run reliably because all cache portions are shareable.
Sounds like
https://lore.kernel.org/lkml/[email protected]/
does what you need.
Thanks for pointing this out. I’ve tested the patch, and it works for me.
For platforms where all cache ways are marked shareable, do you think
providing a extra parameter to ask cat_test to "ignore" the share
restriction is acceptable?
While resctrl tests support command line arguments I do not know if they
are actually being used. During a previous rework we found that they were
not behaving as intended and we did not receive any complaints. Even if
somebody uses the hypothetical "ignore" parameter I do not see how the test
result can be trusted unless this is done as part of some hardware validation
that resctrl selftests do not aim to support. resctrl selftests are intended to
test the resctrl kernel subsystem.
I think effort is better spent on creating an actual CAT functional test.
For reference:
https://lore.kernel.org/lkml/[email protected]/
If I understand correctly, what you suggest is that if the machine does
not have
exclusive cache ways, it is not suitable for kselftest performance
tests. The
only reliable tests are functional tests to verify that the CBM is written
correctly, and so on. I'll check if Ilpo's patch set could be leveraged on
this share-cache machine.
thanks,
Chenyu
