Re: [PATCH v2 stable 6.12 0/3] s390 KVM PCI backports for 6.12

On 8/11/2026 11:55 AM, Sasha Levin wrote:
>> This patchset backports the failed commits to stable-6.12. I have resolved
>> the merge conflicts and tested them on s390.
> Thanks for the respin - this one does apply, and 3/3 now falls through
> into free_aisb: the way mainline does, so both of my objections to v1
> are addressed.
>
> One thing left in 1/3: the conflict resolution keeps 6.12's existing
> in-loop
>
> 	cur_pages = atomic_long_read(&user->locked_vm);
>
> in account_mem() as well as adding the pre-loop read. It is harmless,
> but it is the only line where 6.12 would then differ from mainline and
> from 6.18/7.1. Could you drop the in-loop read and resend? The same
> artifact is in the 6.6 and 6.1 submissions.
>
> While you are respinning: 0cfe660559e8 ("KVM: s390: pci: Fix aisb
> calculation") is missing from 6.18, 6.12, 6.6 and 6.1 - only 7.1 has it.
> kvm_zpci_set_airq() on these trees still computes the AISB address as
> aift->sbv->vector + (zdev->aisb / 64) * 8, which goes out of bounds past
> 64 forwarded devices. Could you or Matthew include it in the resend?

Yeah, I can send that as part of the re-spin. Will send out updates for 
all the relevant backports.

Thanks

Farhan