Fix crazy CPU accounting usage for SQPOLL
Get series with b4
b4 am [email protected]Versions newest first · linked by author and title
- v2 Latest2 patches 0 repliesNew
- v12 patches 2 repliesSuperseded
Changes in v2 from the cover letter
- Store time in usec rather than convert between nsec and timespec64.
Patches in v2 No applies recorded yet
| # | Subject and files | Size | Replies | Applied |
|---|---|---|---|---|
| 1/2 | io_uring/sqpoll: switch away from getrusage() for CPU accountingio_uring/ io_uring/ io_uring/ | +23 | 0 replies | |
| 2/2 | io_uring/sqpoll: be smarter on when to update the stime usageio_uring/ | +32 | 0 replies |
Reviews trailers per patch
1 of 2 patches has a Reviewed-by or Acked-by.
| Patch | Gabriel Krisman Bertazi |
|---|---|
| 1/2 | R-b |
| 2/2 |
Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes.
Cover letter [PATCHSET v2 0/2] Fix crazy CPU accounting usage for SQPOLL
Hi,
Fengnan Chang reports [1] excessive CPU usage from the SQPOLL work vs total time accounting, and that indeed does look like a problem. The issue seems to be just eager time querying, and using essentially the wrong interface. Patch 1 gets rid of getrusage() for this, and patch 2 attempts to be a bit smarter in how often the time needs to get queried. Profile results in patch 2 as well.
Since v1:
- Store time in usec rather than convert between nsec and timespec64.
io_uring/fdinfo.c | 8 +++---- io_uring/sqpoll.c | 65 ++++++++++++++++++++++++++++++++++++++----------------- io_uring/sqpoll.h | 1 + 3 files changed, 50 insertions(+), 24 deletions(-)
Signature · 1 line
Jens Axboe