On 5/9/18 3:08 AM, Omar Sandoval wrote: > From: Omar Sandoval <osandov@xxxxxx> > > Here's the timestamp field rework rebased on for-4.18/block. The only > conflict was the blk stat bits, which this series removes in favor of > rq->wbt_flags. No other changes since v2. > > Cover letter from v1: > > Currently, struct request has four timestamp fields: > > - start_time (jiffies), marked at get_request() time and used for > iostats > - issue_stat (ktime nanoseconds, with some bits shared with wbt and > io.low), marked at start_request() time and used for accounting the > time spent in the device for blk-stats (i.e., wbt, kyber, hybrid > polling) > - start_time_ns (sched_clock nanoseconds), also marked at get_request() > time, used for CFQ and BFQ cgroups support > - io_start_time_ns (sched_clock nanoseconds), also marked at > start_request() time and also used for CFQ and BFQ > > Clearly, there's some redundancy here. This patch series consolidates > these four timestamps into two, both of which are in ktime nanoseconds. > This shaves off up to 16 bytes from struct request, gets rid of the > usage of sched_clock(), and actually removes lines of code. > > Patches 1 and 2 do some prep in wbt. Patch 3 does some prep in io.low. > Patch 4 splits up issue_stat into just a timestamp and separate fields > to fill a hole in struct request. Patch 5 does some prep in CFQ and BFQ. > Patch 6 allows us to save a call to ktime_get(). Patch 7 does the final > consolidation. > > Please take a look. I added a blktest (block/018) to sanity check > iostats, but other stuff is harder to test so it could use some eyes. > Thanks! Thanks Omar, applied. -- Jens Axboe
