Re: [PATCH v7 1/1] vfio/nvgrace-gpu: Add Blackwell-Next GPU readiness check via CXL DVSEC

[Ankit Agrawal]

> The legacy path already has this same issue, but the timeout is 30s
> rather than 256s.  Both locations could pretty trivially change to:
>     msleep_interruptible(POLL_QUANTUM_MS);
>     if (fatal_signal_pending(current))
>             return -EINTR;

Good point, thanks. I'll fold this into both the CXL and legacy waits
for v8.

> We could make this a bit more robust to such a race with a retry, but
> it would be contingent on the interruptible msleep above

Sounds good, I'll add the retry in nvgrace_gpu_vfio_pci_huge_fault() as
you have it.

> Note that the read/write paths also have this gap [...] userspace will
> already automatically handle the -EAGAIN

Agreed.

Thanks for the review!  I'll send v8 with both changes.