On 8/27/22 9:32 AM, Jakub Kicinski wrote:
On Fri, 26 Aug 2022 17:51:27 +0800 D. Wythe wrote:
This patch set attempts to optimize the parallelism of SMC-R connections,
mainly to reduce unnecessary blocking on locks, and to fix exceptions that
occur after thoses optimization.
According to Off-CPU graph, SMC worker's off-CPU as that:
smc_close_passive_work (1.09%)
smcr_buf_unuse (1.08%)
smc_llc_flow_initiate (1.02%)
smc_listen_work (48.17%)
__mutex_lock.isra.11 (47.96%)
The patches should be ordered so that the prerequisite changes are
first, then the removal of locks. Looks like there are 3 patches here
which carry a Fixes tag, for an old commit but in fact IIUC there is no
bug in those old commits, the problem only appears after the locking is
removed?
Thank you for your suggestion, this is indeed my ill-consideration.
The first PATCH with the Fix tag is indeed a prerequisite for removing the lock,
and it do should be placed before. The other two with PATCH fixes theoretically
can also appear before, but after the lock is removed the probability of it will
be greatly increased. I see it can also be placed before.
That said please wait for IBM folks to review first before reshuffling
the patches, I presume the code itself won't change.
Thanks your suggestion again, I will reshuffling the order of it after you
have reviewed it all.
Also I still haven't see anyone reply to Al Viro, IIRC he was
complaining about changes someone from your team has made.
I consider this a blocker for applying new patches from your team :(
Sorry to bother you and your team, my colleague will explain to you soon.
Thanks.
D. Wythe
