On 2024-09-17 at 08:32:11 -0700, Reinette Chatre wrote:
Hi Maciej,
On 8/27/24 1:15 AM, Maciej Wieczor-Retman wrote:
On 2024-08-12 at 16:40:10 -0700, Reinette Chatre wrote:
On 7/12/24 2:04 AM, Maciej Wieczor-Retman wrote:
- if ((get_vendor() == ARCH_INTEL) && snc_unreliable) {
ksft_print_msg("Sub-NUMA Clustering could not be detected properly (see earlier messages for details).\n");ksft_print_msg("Intel CAT may be inaccurate.\n");- }
This is still relevant but unclear why previous message checked "ret" but above does not.
The above check tries to explain why a failure happened.
This check is a reminder about a false positive - the test passes but "snc_unreliable" was set. I guess we could make this check to test "!ret"?
Thinking about this more ... if the test results cannot be trusted at all (whether tests pass or fail) when snc_reliable is true then it seems more appropriate to just skip these tests when SNC detection is unreliable.
Okay, I'll just skip the test if the snc_unreliable is true.