Re: [PATCH v10 6/9] copy-on-read: skip non-guest reads if no copy needed

[Andrey Shinkevich]

On 07.10.2020 22:28, Vladimir Sementsov-Ogievskiy wrote:
> 07.10.2020 22:01, Andrey Shinkevich wrote:
> > On 07.10.2020 13:06, Vladimir Sementsov-Ogievskiy wrote:
> > > 29.09.2020 15:38, Andrey Shinkevich wrote:
> > > > If the flag BDRV_REQ_PREFETCH was set, pass it further to the
> > > > COR-driver to skip unneeded reading. It can be taken into account for
> > > > the COR-algorithms optimization. That check is being made during the
> > > > block stream job by the moment.
> > > >
> > > > Signed-off-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
> > > > ---
> > > >   block/copy-on-read.c | 14 ++++++++++----
> > > >   block/io.c           |  2 +-
> > > >   2 files changed, 11 insertions(+), 5 deletions(-)
> > > >
> > > > diff --git a/block/copy-on-read.c b/block/copy-on-read.c
> > > > index f53f7e0..5389dca 100644
> > > > --- a/block/copy-on-read.c
> > > > +++ b/block/copy-on-read.c
> > > > @@ -145,10 +145,16 @@ static int coroutine_fn 
> > > > cor_co_preadv_part(BlockDriverState *bs,
> > > >               }
> > > >           }
> > > > -        ret = bdrv_co_preadv_part(bs->file, offset, n, qiov, 
> > > > qiov_offset,
> > > > -                                  local_flags);
> > > > -        if (ret < 0) {
> > > > -            return ret;
> > > > +        if ((flags & BDRV_REQ_PREFETCH) &
> > >
> > > BDRV_REQ_PREFETCH is documented to be only used with 
> > > BDRV_REQ_COPY_ON_READ. But here
> > > BDRV_REQ_COPY_ON_READ appears intermediately. We should change 
> > > documentation in block.h
> > > in a separate patch (and probably code in bdrv_aligned_preadv())
> >
> > OK, we will come here without the BDRV_REQ_PREFETCH flag set.
>
> flag BDRV_REQ_PREFETCH should be set in stream job. Where should it be 
> handled, I don't follow?

If we leave block/io.c unchanged in this patch, what I'm agreeing with, 
we'll come to the COR-driver with the hardcoded flags = 0 :

#4  0x000055a22bb480cf in cor_co_preadv_part (bs=0x55a22d593710, 
offset=0, bytes=524288, qiov=0x0, qiov_offset=0, flags=0) at 
../block/copy-on-read.c:149
#5  0x000055a22badcb1d in bdrv_driver_preadv (bs=0x55a22d593710, 
offset=0, bytes=524288, qiov=0x0, qiov_offset=0, flags=0) at 
../block/io.c:1129
#6  0x000055a22baddc81 in bdrv_aligned_preadv (child=0x55a22d814780, 
req=0x7f8c1abffce0, offset=0, bytes=524288, align=1, qiov=0x0, 
qiov_offset=0, flags=512) at ../block/io.c:1515
#7  0x000055a22bade59a in bdrv_co_preadv_part (child=0x55a22d814780, 
offset=0, bytes=524288, qiov=0x0, qiov_offset=0, 
flags=BDRV_REQ_PREFETCH) at ../block/io.c:1757
#8  0x000055a22bade3d2 in bdrv_co_preadv (child=0x55a22d814780, 
offset=0, bytes=524288, qiov=0x0, flags=BDRV_REQ_PREFETCH) at 
../block/io.c:1715
#9  0x000055a22baf5d09 in blk_do_preadv (blk=0x55a22d818c00, offset=0, 
bytes=524288, qiov=0x0, flags=BDRV_REQ_PREFETCH) at 
../block/block-backend.c:1211
#10 0x000055a22baf5d61 in blk_co_preadv (blk=0x55a22d818c00, offset=0, 
bytes=524288, qiov=0x0, flags=BDRV_REQ_PREFETCH) at 
../block/block-backend.c:1223
#11 0x000055a22bab4eba in stream_populate (blk=0x55a22d818c00, offset=0, 
bytes=524288) at ../block/stream.c:50
#12 0x000055a22bab52c2 in stream_run (job=0x55a22d810a20, 
errp=0x55a22d810aa0) at ../block/stream.c:162
#13 0x000055a22bab79f0 in job_co_entry (opaque=0x55a22d810a20) at 
../job.c:908

So, the only way for the COR-filter driver to differ between guests 
reads and the stream job is to check the qiov pointer for NULL and reset 
the flags as appropriate. This is what I am going to do in the next version.

Andrey

> > To differ between guest reads and the stream job ones, we would set it 
> > here by checking for the qiov NULL pointer:
> >
> >
> > diff --git a/block/copy-on-read.c b/block/copy-on-read.c
> > index 4e3b1c5..df2c2ab 100644
> > --- a/block/copy-on-read.c
> > +++ b/block/copy-on-read.c
> > @@ -144,6 +144,9 @@ static int coroutine_fn 
> > cor_co_preadv_part(BlockDriverState *bs,
> >                                             n, &n);
> >               if (ret) {
> >                   local_flags |= BDRV_REQ_COPY_ON_READ;
> > +                if (!qiov) {
> > +                    local_flags |= BDRV_REQ_PREFETCH;
>
> if qiov is NULL, this means that flags must include BDRV_REQ_PREFETCH. 
> local_flags should inherit flags I think.
>
> > +                }
> >               }
> >           }
> >
> > Andrey
> >
> > > > +            !(local_flags & BDRV_REQ_COPY_ON_READ)) {
> > > > +            /* Skip non-guest reads if no copy needed */
> > > > +        } else {
> > > > +
> > >
> > > extra new-line ?
> > >
> > > > +            ret = bdrv_co_preadv_part(bs->file, offset, n, qiov, 
> > > > qiov_offset,
> > > > +                                      local_flags);
> > > > +            if (ret < 0) {
> > > > +                return ret;
> > > > +            }
> > > >           }
> > > >           offset += n;
> > > > diff --git a/block/io.c b/block/io.c
> > > > index 11df188..62b75a5 100644
> > > > --- a/block/io.c
> > > > +++ b/block/io.c
> > > > @@ -1388,7 +1388,7 @@ static int coroutine_fn 
> > > > bdrv_co_do_copy_on_readv(BdrvChild *child,
> > > >               qemu_iovec_init_buf(&local_qiov, bounce_buffer, pnum);
> > > >               ret = bdrv_driver_preadv(bs, cluster_offset, pnum,
> > > > -                                     &local_qiov, 0, 0);
> > > > +                                     &local_qiov, 0, flags & 
> > > > BDRV_REQ_PREFETCH);
> > >
> > > Why? In this place we want to read. We'll write back the data a few 
> > > lines below. What will we write,
> > > if underlying driver decide to do nothing because of BDRV_REQ_PREFETCH?
> >
> > See my comment above please.
>
> Anyway, BDRV_REQ_PREFETCH here is wrong. You should not pass any qiov, 
> if you set BDRV_REQ_PREFETCH flag.
>
> If we come to bdrv_co_do_copy_on_readv, it means that we have 
> COPY_ON_READ flag. And therefore, we will handle
> PREFETCH and COPY_ON_READ here in generic layer. And therefore, we 
> shouldn't pass them to driver.
>
> On the contrary, if we have PREFETCH flag in bdrv_co_aligned_preadv, but 
> don't have COPY_ON_READ in the same time,
> this means that we must pass PREFETCH flag to the driver if it supports 
> it. And do nothing if driver
> doesn't support PREFETCH. That's how I see it.
>
> > > >               if (ret < 0) {
> > > >                   goto err;
> > > >               }