linux-kernel - Re: [PATCH v7 4/5] PCI/AER: Clear both AER fatal and non-fatal status

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <aYGsmgyVwvbj83YI@wunner.de>
Date: Tue, 3 Feb 2026 09:06:50 +0100
From: Lukas Wunner <lukas@...ner.de>
To: Shuai Xue <xueshuai@...ux.alibaba.com>
Cc: linux-pci@...r.kernel.org, linux-kernel@...r.kernel.org,
	linuxppc-dev@...ts.ozlabs.org, bhelgaas@...gle.com,
	kbusch@...nel.org, sathyanarayanan.kuppuswamy@...ux.intel.com,
	mahesh@...ux.ibm.com, oohall@...il.com, Jonathan.Cameron@...wei.com,
	terry.bowman@....com, tianruidong@...ux.alibaba.com
Subject: Re: [PATCH v7 4/5] PCI/AER: Clear both AER fatal and non-fatal status

On Sat, Jan 24, 2026 at 03:45:56PM +0800, Shuai Xue wrote:
> The DPC driver clears AER fatal status for the port that reported the
> error, but not for the downstream device that deteced the error.  The
> current recovery code only clears non-fatal AER status, leaving fatal
> status bits set in the error device.

That's not quite accurate:

The error device has undergone a Hot Reset as a result of the Link Down
event.  To be able to use it again, pci_restore_state() is invoked by
the driver's ->slot_reset() callback.  And pci_restore_state() does
clear fatal status bits.

pci_restore_state()
  pci_aer_clear_status()
    pci_aer_raw_clear_status()

> Use pci_aer_raw_clear_status() to clear both fatal and non-fatal error
> status in the error device, ensuring all AER status bits are properly
> cleared after recovery.

Well, pci_restore_state() already clears all AER status bits so why
is this patch necessary?

> +++ b/drivers/pci/pcie/err.c
> @@ -285,7 +285,7 @@ pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
>  	 */
>  	if (host->native_aer || pcie_ports_native) {
>  		pcie_clear_device_status(dev);
> -		pci_aer_clear_nonfatal_status(dev);
> +		pci_aer_raw_clear_status(dev);
>  	}

This code path is for the case when pcie_do_recovery() is called
with state=pci_channel_io_normal, i.e. in the nonfatal case.
That's why only the nonfatal bits need to be cleared here.

In the fatal case clearing of the error bits is done by
pci_restore_state().

I understand that this is subtle and should probably be changed
to improve clarity, but this patch doesn't look like a step
in that direction.

Thanks,

Lukas