Re: [Qemu-block] [Qemu-devel] [PATCH 0/3] ahci: fix completion race condition

[Bruce Rogers]

>>> On 5/31/2018 at 7:06 AM, Philippe Mathieu-Daudé <address@hidden> wrote:
> Hi Bruce,
> 
> On 05/31/2018 09:21 AM, Bruce Rogers wrote:
>>>>> On 5/30/2018 at 6:43 PM, John Snow <address@hidden> wrote:
>>> Commit d759c951f changed the main thread lock release/reacquisition,
>>> and in so doing apparently jostled loose a race condition in the AHCI
>>> code.
>>>
>>> Patch 2 should be sufficient to fix this, and patches 1 and 3 are just
>>> little trivial fixes.
>>>
>>> This might be sufficient to fix the bug as reported at
>>> https://bugs.launchpad.net/qemu/+bug/1769189 
>>> but the nature of the timing changes make it difficult to confirm,
>>> so I am posting this patchset for the reporters to help test.
>>>
>>> John Snow (3):
>>>   ahci: trim signatures on raise/lower
>>>   ahci: fix PxCI register race
>>>   ahci: don't schedule unnecessary BH
>>>
>>>  hw/ide/ahci.c | 24 +++++++++++‑‑‑‑‑‑‑‑‑‑‑‑‑
>>>  1 file changed, 11 insertions(+), 13 deletions(‑)
>>>
>>> ‑‑ 
>>> 2.14.3
>> 
>> In my case, I applied these 3 patches on top of v2.12 qemu under which
>> I can fairly reliably reproduce Windows10 disk delays. With these patches,
>> after quite a number of attempts, I no longer can reproduce the failure
>> case, so from my perspective it solves the issue.
> 
> Does that implicitly mean John can use your "Tested-by: Bruce Rogers
> <address@hidden>" tag?

Yes. I guess I should have added that.
Bruce