Re: [PATCH] PCI: vmd: Flush initiator posted writes before demuxing interrupts on Meteor Lake

[Rickey Bartlett]

On Tue, Sep 08, 2026 at 07:53:25PM -0600, Keith Busch wrote:
> I am just not sure if reading config space is necessary vs memory space
> (maybe it doesn't matter), or if this errata requires reading from every
> potential interrupting device to flush preceeding transactions vs just a
> single posted transaction to anything in the domain.

Thanks for the ack.

I don't think config vs memory matters for ordering either; I only
tested config. It has a register that is safe to read on any child
without knowing what the device is, which is why I used it.

The only negative result I have is the VMD's own BAR, which is above
the domain rather than in it, so that does not answer whether one read
to any child would cover the others. With a single NVMe on this
machine I could not test one sharer flushing another. The erratum text
names the MSI initiator, so I kept it per-initiator rather than assume
the wider case. If someone with two drives behind the same VMD vector
can show one read is enough, dropping to one read per vmd_irq is a
small change on top.