> -----Original Message-----
> From: Junjie Cao <[email protected]>
> Sent: Sunday, September 20, 2026 2:44 PM
> To: Manish Honap <[email protected]>
> Cc: [email protected]; Ankit Agrawal <[email protected]>; [email protected];
> [email protected]; [email protected]; Srirangan
> Madhavan <[email protected]>; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> Krishnakant Jaju <[email protected]>; Vikram Sethi <[email protected]>; Zhi
> Wang <[email protected]>; [email protected]; qemu-
> [email protected]; [email protected]; Shameer Kolothum Thodi
> <[email protected]>
> Subject: Re: [PATCH v2 10/10] hw/pci-host: Emit a _DSM on pxb-cxl to
> preserve firmware PCI config
> 
> External email: Use caution opening links or attachments
> 
> 
> On Thu, 17 Sep 2026 00:14:12 +0530, Manish Honap wrote:
> > under preserve_config, the _DSM. Gate the _DSM on preserve_config: x86
> > q35 passes false as it does not use the accelerated SMMU, so it emits
> > no _DSM and its ACPI tables stay byte for byte what they were.
> 
> Confirmed: on 28e7aad522 plus the series bios-tables-test passes on
> x86_64 (q35/cxl and q35/acpihmat-genericx, the v1 failures, included),
> aarch64, riscv64 and loongarch64.
> 
> accel=on is out of reach under TCG, so I forced preserve_config on virt 
> locally
> and booted an arm64 guest (7.3-rc1 based, defconfig) with a
> cxl-type3 behind a cxl-rp.  With this patch Linux keeps the firmware BARs and
> bridge window under the pxb-cxl; on 09/10 with the same hack it releases and
> reassigns them.  No ACPI errors either way.
> 
> Tested-by: Junjie Cao <[email protected]>
> 
> 08/10 still says the x86 bridge "still emits the ``_DSM`` method, but its
> function 0 returns an empty support mask"; that was v1.

Oops, I missed this line. In current v2, _DSM is gated on preserve_config, so 
x86
q35 emits no _DSM at all rather than an empty support mask.
I will update this.

> 
> Junjie


Reply via email to