On 1/16/17 5:46 AM, Stefan Hajnoczi wrote:
> On Fri, Jan 13, 2017 at 09:15:49AM -0600, Doug Goldstein wrote:
>> On 1/13/17 6:02 AM, Stefan Hajnoczi wrote:
>>> On Thu, Jan 12, 2017 at 10:57:53AM -0600, Doug Goldstein wrote:
>>>> On 1/12/17 5:46 AM, Stefan Hajnoczi wrote:
>>>>> The virtio_queue_set_notification() nesting introduced for AioContext polling
>>>>> raised an assertion with virtio-net (even in non-polling mode).  Converting
>>>>> virtio-net and virtio-crypto to use virtio_queue_set_notification() in a
>>>>> nesting fashion would be invasive and isn't worth it.
>>>>>
>>>>> Patch 1 contains the revert to resolve the bug that Doug noticed.
>>>>>
>>>>> Patch 2 is a less efficient but safe alternative.
>>>>>
>>>>> Stefan Hajnoczi (2):
>>>>>   Revert "virtio: turn vq->notification into a nested counter"
>>>>>   virtio: disable notifications again after poll succeeded
>>>>>
>>>>>  hw/virtio/virtio.c | 21 +++++++++------------
>>>>>  1 file changed, 9 insertions(+), 12 deletions(-)
>>>>>
>>>>
>>>> So I just gave this series a whirl and it fixes the assert but causes
>>>> another issue for me. While iPXE is getting a DHCP address the screen
>>>> immediately flashes over to the UEFI shell. Its like a timeout is
>>>> getting hit and just dropping me to the shell.
>>>
>>> Sounds like an separate problem.
>>>
>>> Stefan
>>>
>>
>> Is there any debug output that I can provide to help troubleshoot it?
>> I've built 23425cc and UEFI via OVMF is able to get an IP address via
>> DHCP inside of iPXE. I've also taken master and only applied the first
>> patch in this series (the revert) and it too works. Its only when I add
>> the 2nd patch into the mix or don't revert out the "virtio: turn
>> vq->notification into a nested counter" patch that it fails.
> 
> The code in Patch 2 should not be executed in your QEMU configuration,
> so I wonder how Patch 2 can cause the DHCP failure.
> 
> Please verify as follows:
> 
>   $ gdb --args path/to/qemu-system-x86_64 ...
>   (gdb) handle SIGUSR1 noprint nostop pass
>   (gdb) handle SIGPIPE noprint nostop pass
>   (gdb) b virtio_queue_host_notifier_aio_poll_begin
>   (gdb) r
> 
> I predict the breakpoint will not be hit.
> 
> Stefan
> 


Correct. That did not hit. I reapplied both patches to my current master
and it did not hit but it still dropped me to the EFI shell while iPXE
was negotiating DHCP.

I then rebased on top of the latest master
(2ccede18bd24fce5db83fef3674563a1f256717b) and applied both patches. The
iPXE issue went away and the assert issue went away.

So I will concur with you, the issue was not in either of these patches
and is now resolved.

-- 
Doug Goldstein