Messages in this thread | | | Date | Mon, 30 Jan 2023 14:57:04 -0700 | Subject | Re: [GIT PULL] iov_iter: Improve page extraction (pin or just list) | From | Jens Axboe <> |
| |
On 1/30/23 2:55 PM, Jens Axboe wrote: > On 1/30/23 2:33 PM, Jens Axboe wrote: >> On 1/30/23 4:14 AM, David Howells wrote: >>> Hi Jens, >>> >>> Could you consider pulling this patchset into the block tree? I think that >>> Al's fears wrt to pinned pages being removed from page tables causing deadlock >>> have been answered. Granted, there is still the issue of how to handle >>> vmsplice and a bunch of other places to fix, not least skbuff handling. >>> >>> I also have patches to fix cifs in a separate branch that I would also like to >>> push in this merge window - and that requires the first two patches from this >>> series also, so would it be possible for you to merge at least those two >>> rather than manually applying them? >> >> I've pulled this into a separate branch, but based on the block branch, >> for-6.3/iov-extract. It's added to for-next as well. > > This does cause about a 2.7% regression for me, using O_DIRECT on a raw > block device. Looking at a perf diff, here's the top: > > +2.71% [kernel.vmlinux] [k] mod_node_page_state > +2.22% [kernel.vmlinux] [k] iov_iter_extract_pages > > and these two are gone: > > 2.14% [kernel.vmlinux] [k] __iov_iter_get_pages_alloc > 1.53% [kernel.vmlinux] [k] iov_iter_get_pages > > rest is mostly in the noise, but mod_node_page_state() sticks out like > a sore thumb. They seem to be caused by the node stat accounting done > in gup.c for FOLL_PIN.
Confirmed just disabling the node_stat bits in mm/gup.c and now the performance is back to the same levels as before.
An almost 3% regression is a bit hard to swallow...
-- Jens Axboe
| |