On Thu, 29 Aug 2024 03:35:58 -0400, "Michael S. Tsirkin" <mst@xxxxxxxxxx> wrote:
> On Thu, Aug 29, 2024 at 03:26:00PM +0800, Xuan Zhuo wrote:
> > On Thu, 29 Aug 2024 12:51:31 +0800, Jason Wang <jasowang@xxxxxxxxxx> wrote:
> > > On Wed, Aug 28, 2024 at 7:21 PM Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx> wrote:
> > > >
> > > > On Tue, 27 Aug 2024 11:38:45 +0800, Jason Wang <jasowang@xxxxxxxxxx> wrote:
> > > > > On Tue, Aug 20, 2024 at 3:19 PM Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx> wrote:
> > > > > >
> > > > > > leads to regression on VM with the sysctl value of:
> > > > > >
> > > > > > - net.core.high_order_alloc_disable=1
> > > > > >
> > > > > > which could see reliable crashes or scp failure (scp a file 100M in size
> > > > > > to VM):
> > > > > >
> > > > > > The issue is that the virtnet_rq_dma takes up 16 bytes at the beginning
> > > > > > of a new frag. When the frag size is larger than PAGE_SIZE,
> > > > > > everything is fine. However, if the frag is only one page and the
> > > > > > total size of the buffer and virtnet_rq_dma is larger than one page, an
> > > > > > overflow may occur. In this case, if an overflow is possible, I adjust
> > > > > > the buffer size. If net.core.high_order_alloc_disable=1, the maximum
> > > > > > buffer size is 4096 - 16. If net.core.high_order_alloc_disable=0, only
> > > > > > the first buffer of the frag is affected.
>
> I don't exactly get it, when you say "only the first buffer of the frag
> is affected" what do you mean? Affected how?
I should say that if the frag is 32k, that is safe.
Only when that frag is 4k, that is not safe.
Thanks.
>
> > > > >
> > > > > I wonder instead of trying to make use of headroom, would it be
> > > > > simpler if we allocate dedicated arrays of virtnet_rq_dma？
> > > >
> > > > Sorry for the late reply. My mailbox was full, so I missed the reply to this
> > > > thread. Thanks to Si-Wei for reminding me.
> > > >
> > > > If the virtnet_rq_dma is at the headroom, we can get the virtnet_rq_dma by buf.
> > > >
> > > > struct page *page = virt_to_head_page(buf);
> > > >
> > > > head = page_address(page);
> > > >
> > > > If we use a dedicated array, then we need pass the virtnet_rq_dma pointer to
> > > > virtio core, the array has the same size with the rx ring.
> > > >
> > > > The virtnet_rq_dma will be:
> > > >
> > > > struct virtnet_rq_dma {
> > > > dma_addr_t addr;
> > > > u32 ref;
> > > > u16 len;
> > > > u16 need_sync;
> > > > + void *buf;
> > > > };
> > > >
> > > > That will be simpler.
> > >
> > > I'm not sure I understand here, did you mean using a dedicated array is simpler?
> >
> > I found the old version(that used a dedicated array):
> >
> > http://lore.kernel.org/all/20230710034237.12391-11-xuanzhuo@xxxxxxxxxxxxxxxxx
> >
> > If you think that is ok, I can port a new version based that.
> >
> > Thanks.
> >
>
> That one got a bunch of comments that likely still apply.
> And this looks like a much bigger change than what this
> patch proposes.
>
> > >
> > > >
> > > > >
> > > > > Btw, I see it has a need_sync, I wonder if it can help for performance
> > > > > or not? If not, any reason to keep that?
> > > >
> > > > I think yes, we can skip the cpu sync when we do not need it.
> > >
> > > I meant it looks to me the needs_sync is not necessary in the
> > > structure as we can call need_sync() any time if we had dma addr.
> > >
> > > Thanks
> > >
> > > >
> > > > Thanks.
> > > >
> > > >
> > > > >
> > > > > >
> > > > > > Fixes: f9dac92ba908 ("virtio_ring: enable premapped mode whatever use_dma_api")
> > > > > > Reported-by: "Si-Wei Liu" <si-wei.liu@xxxxxxxxxx>
> > > > > > Closes: http://lore.kernel.org/all/8b20cc28-45a9-4643-8e87-ba164a540c0a@xxxxxxxxxx
> > > > > > Signed-off-by: Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx>
> > > > > > ---
> > > > > > drivers/net/virtio_net.c | 12 +++++++++---
> > > > > > 1 file changed, 9 insertions(+), 3 deletions(-)
> > > > > >
> > > > > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > > > > index c6af18948092..e5286a6da863 100644
> > > > > > --- a/drivers/net/virtio_net.c
> > > > > > +++ b/drivers/net/virtio_net.c
> > > > > > @@ -918,9 +918,6 @@ static void *virtnet_rq_alloc(struct receive_queue *rq, u32 size, gfp_t gfp)
> > > > > > void *buf, *head;
> > > > > > dma_addr_t addr;
> > > > > >
> > > > > > - if (unlikely(!skb_page_frag_refill(size, alloc_frag, gfp)))
> > > > > > - return NULL;
> > > > > > -
> > > > > > head = page_address(alloc_frag->page);
> > > > > >
> > > > > > dma = head;
> > > > > > @@ -2421,6 +2418,9 @@ static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
> > > > > > len = SKB_DATA_ALIGN(len) +
> > > > > > SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
> > > > > >
> > > > > > + if (unlikely(!skb_page_frag_refill(len, &rq->alloc_frag, gfp)))
> > > > > > + return -ENOMEM;
> > > > > > +
> > > > > > buf = virtnet_rq_alloc(rq, len, gfp);
> > > > > > if (unlikely(!buf))
> > > > > > return -ENOMEM;
> > > > > > @@ -2521,6 +2521,12 @@ static int add_recvbuf_mergeable(struct virtnet_info *vi,
> > > > > > */
> > > > > > len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len, room);
> > > > > >
> > > > > > + if (unlikely(!skb_page_frag_refill(len + room, alloc_frag, gfp)))
> > > > > > + return -ENOMEM;
> > > > > > +
> > > > > > + if (!alloc_frag->offset && len + room + sizeof(struct virtnet_rq_dma) > alloc_frag->size)
> > > > > > + len -= sizeof(struct virtnet_rq_dma);
> > > > > > +
> > > > > > buf = virtnet_rq_alloc(rq, len + room, gfp);
> > > > > > if (unlikely(!buf))
> > > > > > return -ENOMEM;
> > > > > > --
> > > > > > 2.32.0.3.g01195cf9f
> > > > >
> > > > > Thanks
> > > > >
> > > > > >
> > > > >
> > > > >
> > > > >
> > > >
> > >
>
