Re: [PATCH] hw/virtio: Update vring after modifying its queue size

[Philippe Mathieu-Daudé]

On 8/26/21 10:40 AM, Philippe Mathieu-Daudé wrote:
> On 8/26/21 5:28 AM, Jason Wang wrote:
>> On Thu, Aug 26, 2021 at 6:43 AM Philippe Mathieu-Daudé
>> <philmd@redhat.com> wrote:
>>>
>>> When a ring queue size is modified, we need to call
>>> virtio_queue_update_rings() to re-init the memory region
>>> caches. Otherwise the region might have outdated memory
>>> size, and later load/store access might trigger an
>>> assertion, such:
>>>
>>>   qemu-system-i386: include/exec/memory_ldst_cached.h.inc:30: uint16_t 
>>> address_space_lduw_le_cached(MemoryRegionCache *, hwaddr, MemTxAttrs, 
>>> MemTxResult *):
>>>   Assertion `addr < cache->len && 2 <= cache->len - addr' failed.
>>>   Thread 1 "qemu-system-i38" received signal SIGABRT, Aborted.
>>>   0x00007ffff4d312a2 in raise () from /lib64/libc.so.6
>>>   (gdb) bt
>>>   #1  0x00007ffff4d1a8a4 in abort () at /lib64/libc.so.6
>>>   #4  0x0000555558f2b8ec in address_space_lduw_le_cached 
>>> (cache=0x61300010a7c0, addr=134, attrs=..., result=0x0) at 
>>> include/exec/memory_ldst_cached.h.inc:30
>>>   #5  0x0000555558f2ac6b in lduw_le_phys_cached (cache=0x61300010a7c0, 
>>> addr=134) at include/exec/memory_ldst_phys.h.inc:67
>>>   #6  0x0000555558f2a3cd in virtio_lduw_phys_cached (vdev=0x62d00003a680, 
>>> cache=0x61300010a7c0, pa=134) at include/hw/virtio/virtio-access.h:166
>>>   #7  0x0000555558f300ea in vring_avail_ring (vq=0x7fffdd55d8a0, i=65) at 
>>> hw/virtio/virtio.c:326
>>>   #8  0x0000555558f33b10 in vring_get_used_event (vq=0x7fffdd55d8a0) at 
>>> hw/virtio/virtio.c:332
>>>   #9  0x0000555558f33677 in virtio_split_should_notify 
>>> (vdev=0x62d00003a680, vq=0x7fffdd55d8a0) at hw/virtio/virtio.c:2471
>>>   #10 0x0000555558f1859f in virtio_should_notify (vdev=0x62d00003a680, 
>>> vq=0x7fffdd55d8a0) at hw/virtio/virtio.c:2523
>>>   #11 0x0000555558f188cc in virtio_notify (vdev=0x62d00003a680, 
>>> vq=0x7fffdd55d8a0) at hw/virtio/virtio.c:2565
>>>   #12 0x0000555557c2bd52 in virtio_input_handle_sts (vdev=0x62d00003a680, 
>>> vq=0x7fffdd55d8a0) at hw/input/virtio-input.c:100
>>>   #13 0x0000555558f16df7 in virtio_queue_notify (vdev=0x62d00003a680, n=1) 
>>> at hw/virtio/virtio.c:2363
>>>   #14 0x00005555583f45c0 in virtio_pci_notify_write (opaque=0x62d000032400, 
>>> addr=7, val=0, size=1) at hw/virtio/virtio-pci.c:1369
>>>   #15 0x0000555558b80b04 in memory_region_write_accessor 
>>> (mr=0x62d000033190, addr=7, value=0x7fffffff8eb0, size=1, shift=0, 
>>> mask=255, attrs=...) at softmmu/memory.c:492
>>>
>>> Reported-by: Qiuhao Li <Qiuhao.Li@outlook.com>
>>> Fixes: ab223c9518e ("virtio: allow virtio-1 queue layout")
>>> Resolves: https://gitlab.com/qemu-project/qemu/-/issues/302
>>> BugLink: https://bugs.launchpad.net/qemu/+bug/1913510
>>> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
>>> ---
>>>  hw/virtio/virtio.c | 1 +
>>>  1 file changed, 1 insertion(+)
>>>
>>> diff --git a/hw/virtio/virtio.c b/hw/virtio/virtio.c
>>> index 874377f37a7..04ffe5f420e 100644
>>> --- a/hw/virtio/virtio.c
>>> +++ b/hw/virtio/virtio.c
>>> @@ -2255,6 +2255,7 @@ void virtio_queue_set_num(VirtIODevice *vdev, int n, 
>>> int num)
>>>          return;
>>>      }
>>>      vdev->vq[n].vring.num = num;
>>> +    virtio_queue_update_rings(vdev, n);
>>>  }
>>>
>>
>> Spec said:
>>
>> "
>> The driver MUST configure the other virtqueue fields before enabling
>> the virtqueue with queue_enable.
>> "
>>
>> So I think we should forbid the num to be changed if the virtqueue is ready?

What about virtio_queue_set_addr() and virtio_queue_set_align()?