lkml.org 
[lkml]   [2018]   [May]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC v3 3/5] virtio_ring: add packed ring support
On Thu, May 10, 2018 at 05:49:20PM +0800, Jason Wang wrote:
> On 2018年05月10日 16:56, Tiwei Bie wrote:
> > On Thu, May 10, 2018 at 03:34:50PM +0800, Jason Wang wrote:
> > > On 2018年05月10日 15:32, Jason Wang wrote:
> > > > On 2018年04月25日 13:15, Tiwei Bie wrote:
> > > > > +    /* We're using some buffers from the free list. */
> > > > > +    vq->vq.num_free -= descs_used;
> > > > > +
> > > > > +    /* Update free pointer */
> > > > > +    if (indirect) {
> > > > > +        n = head + 1;
> > > > > +        if (n >= vq->vring_packed.num) {
> > > > > +            n = 0;
> > > > > +            vq->wrap_counter ^= 1;
> > > > > +        }
> > > > > +        vq->next_avail_idx = n;
> > > > > +    } else
> > > > > +        vq->next_avail_idx = i;
> > > > During testing zerocopy (out of order completion), I found driver may
> > > > submit two identical buffer id to vhost. So the above code may not work
> > > > well.
> > > >
> > > > Consider the case that driver adds 3 buffer and virtqueue size is 8.
> > > >
> > > > a) id = 0,count = 2,next_avail = 2
> > > >
> > > > b) id = 2,count = 4,next_avail = 2
> > > next_avail should be 6 here.
> > >
> > > > c) id = 4,count = 2,next_avail = 0
> > > >
> > > id should be 6 here.
> > >
> > > Thanks
> > >
> > > > if packet b is done before packet a, driver may think buffer id 0 is
> > > > available and try to use it if even if the real buffer 0 was not done.
> > > >
> > > > Thanks
> > Nice catch! Thanks a lot!
> > I'll implement an ID allocator.
> >
> > Best regards,
> > Tiwei Bie
>
> Sounds good.
>
> Another similar issue is detac_buf_packed(). It did:
>
>         for (j = 0; j < vq->desc_state[head].num; j++) {
>                 desc = &vq->vring_packed.desc[i];
>                 vring_unmap_one_packed(vq, desc);
>                 i++;
>                 if (i >= vq->vring_packed.num)
>                         i = 0;
>         }
>
> This probably won't work for out of order too and according to the spec:
>
> """
> Driver needs to keep track of the size of the list corresponding to each
> buffer ID, to be able to skip to where the next used descriptor is written
> by the device.
> """
>
> Looks like we should not depend on the descriptor ring.

Yeah, the previous ID allocation is too simple..
Let me fix it in the next version.

Thanks!

>
> Thanks

\
 
 \ /
  Last update: 2018-05-10 12:50    [W:0.077 / U:0.172 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site