Am 2021-12-30 12:33, schrieb Paul Menzel:
Hi Paul,
Thank you for the wording improvements to the commit message,
incorporated for the next
patch version.
For x86_64, this removes 8 out of 18 benchmark loops which each take
16 jiffies, so up to 160 jiffies saved on module load (640ms on a
250HZ
kernel)
On what system?
on a x86_64 system with avx512 capabilities. before this patch it was
doing 3x avx512, 3x avx2 and 3x sse2 xor() benchmark runs (so 9 total,
plus 9 gen() runs as well, leading to the 18 above). with this patch
applied the 9 xor() runs become just 1, saving 8. exact timing depends
on the CONFIG_HZ setting in use, as the benchmark timescale is in
jiffies (which is a problem on its own, but that is for another patch).
The new message below is logged?
raid6: skipped pq benchmark and selected …
its the same message like before, just worded slightly differently. I
can undo the wording change if requested.
I am booting my non-RAID systems with `cryptomgr.notests` to avoid
this boot time penalty.
the benchmark option is recommended to be turned on, and I'm trying to
reduce the cost of that. turning it off avoids the cost altogether, but
I'm not able to judge (yet?) whether that's a better thing to do.
Thanks
Dirk
