linux-kernel - Re: [PATCH 1/1] pci: Add subordinate check before pci_add_new

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 for Android: free password hash cracker in your pocket

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

Message-ID: <20250814203612.GA346111@bhelgaas>
Date: Thu, 14 Aug 2025 15:36:12 -0500
From: Bjorn Helgaas <helgaas@...nel.org>
To: Rui He <rui.he@...driver.com>
Cc: Bjorn Helgaas <bhelgaas@...gle.com>, linux-pci@...r.kernel.org,
	linux-kernel@...r.kernel.org, Prashant.Chikhalkar@...driver.com,
	Jiguang.Xiao@...driver.com
Subject: Re: [PATCH 1/1] pci: Add subordinate check before pci_add_new_bus()

On Thu, Aug 14, 2025 at 05:39:37PM +0800, Rui He wrote:
> For preconfigured PCI bridge, child bus created on the first scan.
> While for some reasons(e.g register mutation), the secondary, and subordiante
> register reset to 0 on the second scan, which caused to create
> PCI bus twice for the same PCI device.

I don't quite follow this.  Do you mean something is changing the
bridge configuration between the first and second scans?

> Following is the related log:
> [Wed May 28 20:38:36 CST 2025] pci 0000:0b:01.0: PCI bridge to [bus 0d]
> [Wed May 28 20:38:36 CST 2025] pci 0000:0b:05.0: bridge configuration invalid ([bus 00-00]), reconfiguring
> [Wed May 28 20:38:36 CST 2025] pci 0000:0b:01.0: PCI bridge to [bus 0e-10]
> [Wed May 28 20:38:36 CST 2025] pci 0000:0b:05.0: PCI bridge to [bus 0f-10]

Drop the timestamps (since they don't contribute to understanding the
problem) and indent the logs a couple spaces.

> Here PCI device 000:0b:01.0 assigend to bus 0d and 0e.

It looks like the [bus 0f-10] range is assigned to both bridges
(0b:01.0 and 0b:05.0), which would definitely be a problem.

I'm surprised that we haven't tripped over this before, and I'm
curious about how we got here.  Can you set CONFIG_DYNAMIC_DEBUG=y,
boot with the dyndbg="file drivers/pci/* +p" kernel parameter, and
collect the complete dmesg log?

> This patch checks if child PCI bus has been created on the second scan
> of bridge. If yes, return directly instead of create a new one.
> 
> Signed-off-by: Rui He <rui.he@...driver.com>
> ---
>  drivers/pci/probe.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> index f41128f91ca76..ec67adbf31738 100644
> --- a/drivers/pci/probe.c
> +++ b/drivers/pci/probe.c
> @@ -1444,6 +1444,9 @@ static int pci_scan_bridge_extend(struct pci_bus *bus, struct pci_dev *dev,
>  			goto out;
>  		}
>  
> +		if(pci_has_subordinate(dev))
> +			goto out;

Follow the coding style, i.e., add a space in "if (pci_..."

>  		/* Clear errors */
>  		pci_write_config_word(dev, PCI_STATUS, 0xffff);
>  
> -- 
> 2.43.0
>