Re: [PATCH 1/2] x86/PCI: Disable enhanced atomics on some AMD PCIe ports

[Mario Limonciello <mario.limonciello@amd.com>]

On 9/8/26 12:14, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> - [Critical] Hardware state loss across suspend/resume causing data corruption
> --
> 
> commit 6c1c185449d3f44e1db8d42c0f5b1f2a64c2eeb8
> Author: Mario Limonciello <mario.limonciello@amd.com>
> 
> x86/PCI: Disable enhanced atomics on some AMD PCIe ports
> 
> This commit disables enhanced atomic operations on PCIe ports for AMD
> NBIO 7.7 and 7.11 based models to prevent data corruption with 64-bit DMA.
> The issue occurs when the 32-bit IOVA space is exhausted on certain BIOSes.
> 
> Closes: https://lore.kernel.org/all/178789300872.392066.15963676631650361573@gmail.com/
> Closes: https://lore.kernel.org/linux-ide/20260406222335.379935-1-artmoty@gmail.com/
> Closes: https://lore.kernel.org/linux-ide/20260621100844.1224301-1-alvinwylim@gmail.com/
> 
>> diff --git a/arch/x86/pci/fixup.c b/arch/x86/pci/fixup.c
>> index b301c6c8df753..0857fd9365ec2 100644
>> --- a/arch/x86/pci/fixup.c
>> +++ b/arch/x86/pci/fixup.c
> 
> [ ... ]
> 
>> +/* Phoenix, Hawk Point (NBIO 7.7) */
>> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_AMD, 0x14E8,
>> +			quirk_amd_nbio_7_7_disable_enhanced_atomic);
>> +
>> +/* Strix, Krackan, Strix Halo (NBIO 7.11) */
>> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_AMD, 0x1507,
>> +			quirk_amd_nbio_7_11_disable_enhanced_atomic);
>> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_AMD, 0x1122,
>> +			quirk_amd_nbio_7_11_disable_enhanced_atomic);
> 
> [Severity: Critical]
> Is this fixup lost across suspend and resume cycles?
> 
> Because the quirk is only registered for the final boot phase using
> DECLARE_PCI_FIXUP_FINAL, the PCI core will not re-apply it during system
> wake (which relies on DECLARE_PCI_FIXUP_RESUME).
> 
> If the system suspends to S3 or s2idle, the NBIO power domain is gated and
> SMN registers lose their state. When the system resumes, the firmware
> re-initializes these SMN registers to their default state (which has
> enhanced atomics enabled).
> 
> Because the kernel does not re-apply the quirk on resume, any subsequent
> 64-bit DMA traffic after waking from sleep could trigger the data
> corruption this code is intended to prevent.
> 
> Should these quirks also be registered with DECLARE_PCI_FIXUP_RESUME?
> 

This looks like valid feedback.  I checked and these registers don't 
surive suspend/resume.  I'll modify it for a v2.