Re: [PATCH v2] vpci: introduce per-domain lock to protect vpci structure

[Jan Beulich <jbeulich@xxxxxxxx>]

On 15.02.2022 17:28, Oleksandr Andrushchenko wrote:
> On 15.02.22 18:18, Jan Beulich wrote:
>> On 15.02.2022 16:46, Oleksandr Andrushchenko wrote:
>>> Question: can anyone please explain why pcidevs is a recursive lock?
>> Well, assuming you did look at the change making it so, can you be a
>> little more specific with your question? Are you perhaps suggesting
>> the original reason has disappeared, and no new one has appeared? I'm
>> afraid I have to repeat what I did say before: If you want to remove
>> the recursive nature of the lock, then it is all on you to prove that
>> there's no code path where the lock is taken recursively. IOW even if
>> no-one knew of a reason, you'd still need to provide this proof.
>> Unless of course we'd all agree we're okay to take the risk; I don't
>> see us doing so, though.
> The question was exactly as asked: I don't understand why it is
> recursive and for what reason. I am not suggesting we blindly
> change it to a normal spinlock.

But the reason for changing it to be recursive is stated in the
description of the respective commit bfa493f52e89:

IOMMU: make the pcidevs_lock a recursive one

The pcidevs_lock is going to be recursively taken for hiding ATS
device, when VT-d Device-TLB flush timed out.

Before asking such a question, I would have assumed that you looked
up that very commit. Hence my asking to be more specific.

Jan