| Message ID | 20240903144729.37218-1-thenzl@redhat.com (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | 24d7071d964574cca41fa72a10c211221af37aec |
| Headers | show |
| Series | [v2] mpi3mr: a performance fix | expand |
Tomas, > Commit 0c52310f2600 ("hrtimer: Ignore slack time for RT tasks in > schedule_hrtimeout_range()") effectivelly shortens a sleep in a > polling function in the driver. That is causing a performance > regression as the new value of just 2us is too low, in certain tests > the perf drop is ~30%. Fix this by adjusting the sleep to 20us (close > to the previous value). Applied to 6.12/scsi-staging, thanks!
On Tue, 03 Sep 2024 16:47:29 +0200, Tomas Henzl wrote: > Commit 0c52310f2600 ("hrtimer: Ignore slack time for RT tasks in schedule_hrtimeout_range()") > effectivelly shortens a sleep in a polling function in the driver. > That is causing a performance regression as the new value > of just 2us is too low, in certain tests the perf drop is ~30%. > Fix this by adjusting the sleep to 20us (close to the previous value). > > The '+1' added to the max parameter in usleep_range is there just > to silence static code analyzers like checkpatch. > > [...] Applied to 6.12/scsi-queue, thanks! [1/1] mpi3mr: a performance fix https://git.kernel.org/mkp/scsi/c/24d7071d9645
diff --git a/drivers/scsi/mpi3mr/mpi3mr.h b/drivers/scsi/mpi3mr/mpi3mr.h index dc2cdd5f0311..249c1a7285d6 100644 --- a/drivers/scsi/mpi3mr/mpi3mr.h +++ b/drivers/scsi/mpi3mr/mpi3mr.h @@ -178,7 +178,7 @@ extern atomic64_t event_counter; #define MPI3MR_DEFAULT_SDEV_QD 32 /* Definitions for Threaded IRQ poll*/ -#define MPI3MR_IRQ_POLL_SLEEP 2 +#define MPI3MR_IRQ_POLL_SLEEP 20 #define MPI3MR_IRQ_POLL_TRIGGER_IOCOUNT 8 /* Definitions for the controller security status*/ diff --git a/drivers/scsi/mpi3mr/mpi3mr_fw.c b/drivers/scsi/mpi3mr/mpi3mr_fw.c index c196dc14ad20..5695c95fca15 100644 --- a/drivers/scsi/mpi3mr/mpi3mr_fw.c +++ b/drivers/scsi/mpi3mr/mpi3mr_fw.c @@ -710,7 +710,7 @@ static irqreturn_t mpi3mr_isr_poll(int irq, void *privdata) mpi3mr_process_op_reply_q(mrioc, intr_info->op_reply_q); - usleep_range(MPI3MR_IRQ_POLL_SLEEP, 10 * MPI3MR_IRQ_POLL_SLEEP); + usleep_range(MPI3MR_IRQ_POLL_SLEEP, MPI3MR_IRQ_POLL_SLEEP + 1); } while (atomic_read(&intr_info->op_reply_q->pend_ios) && (num_op_reply < mrioc->max_host_ios));
Commit 0c52310f2600 ("hrtimer: Ignore slack time for RT tasks in schedule_hrtimeout_range()") effectivelly shortens a sleep in a polling function in the driver. That is causing a performance regression as the new value of just 2us is too low, in certain tests the perf drop is ~30%. Fix this by adjusting the sleep to 20us (close to the previous value). The '+1' added to the max parameter in usleep_range is there just to silence static code analyzers like checkpatch. Reported-by: Jan Jurca <jjurca@redhat.com> Signed-off-by: Tomas Henzl <thenzl@redhat.com> --- V2: added a '+ 1' to silence checkpatch. --- drivers/scsi/mpi3mr/mpi3mr.h | 2 +- drivers/scsi/mpi3mr/mpi3mr_fw.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)