Re: [PATCH] virtio_ring: fix KMSAN error for premapped mode

["Michael S. Tsirkin" <mst@redhat.com>]

On Mon, Jul 15, 2024 at 05:30:49PM +0200, Ilya Leoshkevich wrote:
> On Mon, 2024-07-15 at 10:20 -0400, Michael S. Tsirkin wrote:
> > On Mon, Jul 15, 2024 at 03:24:17PM +0200, Alexander Potapenko wrote:
> > > On Mon, Jun 24, 2024 at 3:57 PM Michael S. Tsirkin <mst@redhat.com>
> > > wrote:
> > > > 
> > > > On Mon, Jun 24, 2024 at 03:51:50PM +0200, Alexander Potapenko
> > > > wrote:
> > > > > On Fri, Jun 7, 2024 at 8:34 AM Jason Wang <jasowang@redhat.com>
> > > > > wrote:
> > > > > > 
> > > > > > On Thu, Jun 6, 2024 at 7:13 PM Xuan Zhuo
> > > > > > <xuanzhuo@linux.alibaba.com> wrote:
> > > > > > > 
> > > > > > > Add kmsan for virtqueue_dma_map_single_attrs to fix:
> > > > > > > 
> > > > > > > BUG: KMSAN: uninit-value in receive_buf+0x45ca/0x6990
> > > > > > >  receive_buf+0x45ca/0x6990
> > > > > > >  virtnet_poll+0x17e0/0x3130
> > > > > > >  net_rx_action+0x832/0x26e0
> > > > > > >  handle_softirqs+0x330/0x10f0
> > > > > > >  [...]
> > > > > > > 
> > > > > > > Uninit was created at:
> > > > > > >  __alloc_pages_noprof+0x62a/0xe60
> > > > > > >  alloc_pages_noprof+0x392/0x830
> > > > > > >  skb_page_frag_refill+0x21a/0x5c0
> > > > > > >  virtnet_rq_alloc+0x50/0x1500
> > > > > > >  try_fill_recv+0x372/0x54c0
> > > > > > >  virtnet_open+0x210/0xbe0
> > > > > > >  __dev_open+0x56e/0x920
> > > > > > >  __dev_change_flags+0x39c/0x2000
> > > > > > >  dev_change_flags+0xaa/0x200
> > > > > > >  do_setlink+0x197a/0x7420
> > > > > > >  rtnl_setlink+0x77c/0x860
> > > > > > >  [...]
> > > > > > > 
> > > > > > > Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
> > > > > > > Tested-by: Alexander Potapenko <glider@google.com>
> > > > > > 
> > > > > > Acked-by: Jason Wang <jasowang@redhat.com>
> > > > > 
> > > > > Jason, Michael, any chance this patch can make it to 6.10?
> > > > > The bug in question has essentially been blocking KMSAN on
> > > > > syzbot for
> > > > > the past month:
> > > > > https://syzkaller.appspot.com/bug?id=2819883f3d376b703e686e700fde00a01eec62fa
> > > > > 
> > > > > The patch should probably have
> > > > >   Fixes: b6253b4e21939 ("virtio_ring: introduce dma map api for
> > > > > virtqueue")
> > > > > 
> > > > > , but I am not exactly sure why syzkaller only started
> > > > > triggering the
> > > > > bug this year, given that the patch dates back to 2023.
> > > > 
> > > > Yes that's probably a good idea. I will queue it with this tag.
> > > 
> > > Hi Michael,
> > > 
> > > Looks like the patch is still stuck in the limbo :(
> > 
> > It should be in linux next. Don't you see it there?
> 
> FWIW I ran linux-next with KMSAN today and this fix was already there.
> 
> But in the meantime another similarly looking issue crept in:
> 
> https://lore.kernel.org/netdev/20240712195759.GA2972562@myrica/T/

I remember picking up that fix for next, too. Not there?
Not at the testing box right now.

-- 
MST