> Why is this in drivers/pci/ instead of behind a clean abstraction in
> arch code?

It extends the existing AMD check in cpu_supports_p2pdma(), which
already reads cpuinfo_x86. The table describes a host bridge property,
not a CPU one; it's keyed by CPU model because the host bridge device
isn't visible to VMware and Hyper-V guests.

Thanks,
Pavel

> -----Original Message-----
> From: Christoph Hellwig <[email protected]>
> Sent: Sunday, November 15, 2026 11:22 PM
> To: Popov, Pavel E <[email protected]>
> Cc: Bjorn Helgaas <[email protected]>; Logan Gunthorpe
> <[email protected]>; Arnd Bergmann <[email protected]>; Leon Romanovsky
> <[email protected]>; Lukas Wunner <[email protected]>; Thomas Gleixner
> <[email protected]>; Ingo Molnar <[email protected]>; Borislav Petkov
> <[email protected]>; Dave Hansen <[email protected]>; [email protected];
> Jim Chow <[email protected]>; Radu Rugina
> <[email protected]>; Alexey Makhalov
> <[email protected]>; Wei Liu <[email protected]>; Michael
> Kelley <[email protected]>; Nathan Ciobanu
> <[email protected]>; [email protected];
> [email protected]; [email protected]; linux-
> [email protected]; [email protected]; [email protected]
> Subject: Re: [PATCH v5 3/3] PCI/P2PDMA: Allow P2PDMA on Intel Xeon server
> CPUs
> 
> On Wed, Nov 11, 2026 at 04:44:39PM -0700, Pavel Popov wrote:
> >  #ifdef CONFIG_X86
> > +/* Intel Xeon CPUs whose host bridges support p2pdma */ static const
> > +struct x86_cpu_id p2pdma_intel_cpu_ids[] = {
> 
> Why is this in drivers/pci/ instead of behind a clean abstraction in arch 
> code?
> 
> (and the fact that the CPU vendors still haven't managed to provide a firmware
> interface to expose the P2P capabilities is just insane)

Reply via email to