Re: [PATCH v3] pipe: use __pipe_{lock,unlock} instead of spinlock

[Sedat Dilek]

On Mon, Jan 16, 2023 at 3:16 AM Hongchen Zhang
<zhanghongchen@xxxxxxxxxxx> wrote:
>
> Hi sedat,
>
>
> On 2023/1/16 am9:52, Sedat Dilek wrote:
> > On Fri, Jan 13, 2023 at 10:32 AM Sedat Dilek <sedat.dilek@xxxxxxxxx> wrote:
> >>
> >> On Fri, Jan 13, 2023 at 4:19 AM Hongchen Zhang
> >> <zhanghongchen@xxxxxxxxxxx> wrote:
> >>>
> >>> Hi All,
> >>> any question about this patch, can it be merged?
> >>>
> >>> Thanks
> >>> On 2023/1/7 am 9:23, Hongchen Zhang wrote:
> >>>> Use spinlock in pipe_read/write cost too much time,IMO
> >>>> pipe->{head,tail} can be protected by __pipe_{lock,unlock}.
> >>>> On the other hand, we can use __pipe_{lock,unlock} to protect
> >>>> the pipe->{head,tail} in pipe_resize_ring and
> >>>> post_one_notification.
> >>>>
> >>>> Reminded by Matthew, I tested this patch using UnixBench's pipe
> >>>> test case on a x86_64 machine,and get the following data:
> >>>> 1) before this patch
> >>>> System Benchmarks Partial Index  BASELINE       RESULT    INDEX
> >>>> Pipe Throughput                   12440.0     493023.3    396.3
> >>>>                                                           ========
> >>>> System Benchmarks Index Score (Partial Only)              396.3
> >>>>
> >>>> 2) after this patch
> >>>> System Benchmarks Partial Index  BASELINE       RESULT    INDEX
> >>>> Pipe Throughput                   12440.0     507551.4    408.0
> >>>>                                                           ========
> >>>> System Benchmarks Index Score (Partial Only)              408.0
> >>>>
> >>>> so we get ~3% speedup.
> >>>>
> >>>> Reminded by Andrew, I tested this patch with the test code in
> >>>> Linus's 0ddad21d3e99 add get following result:
> >>
> >> Happy new 2023 Hongchen Zhang,
> >>
> >> Thanks for the update and sorry for the late response.
> >>
> >> Should be "...s/add/and get following result:"
> >>
> >> I cannot say much about the patch itself or tested it in my build-environment.
> >>
> >> Best regards,
> >> -Sedat-
> >>
> >
> > I have applied v3 on top of Linux v6.2-rc4.
> >
> > Used pipebench for a quick testing.
> >
> > # fdisk -l /dev/sdb
> > Disk /dev/sdb: 14,91 GiB, 16013942784 bytes, 31277232 sectors
> > Disk model: SanDisk iSSD P4
> > Units: sectors of 1 * 512 = 512 bytes
> > Sector size (logical/physical): 512 bytes / 512 bytes
> > I/O size (minimum/optimal): 512 bytes / 512 bytes
> > Disklabel type: dos
> > Disk identifier: 0x74f02dea
> >
> > Device     Boot Start      End  Sectors  Size Id Type
> > /dev/sdb1        2048 31277231 31275184 14,9G 83 Linux
> >
> > # cat /dev/sdb | pipebench > /dev/null
> > Summary:
> > Piped   14.91 GB in 00h01m34.20s:  162.12 MB/second
> >
> > Not tested/benchmarked with the kernel w/o your patch.
> >
> > -Sedat-
> >
> OK, If there is any problem, let's continue to discuss it
> and hope it can be merged into the main line.
>

Can you give me a hand on the perf stat line?

I tried with:

$ /usr/bin/perf stat --repeat=1 ./0ddad21d3e99

But that gives in both cases no context-switches and cpu-migrations values.

-Sedat-