linux-kernel - Re: [PATCH for-linus 1/1] PCI: Revert early bridge resource set up

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: <20251014203906.GA905971@bhelgaas>
Date: Tue, 14 Oct 2025 15:39:06 -0500
From: Bjorn Helgaas <helgaas@...nel.org>
To: Ilpo Järvinen <ilpo.jarvinen@...ux.intel.com>
Cc: Bjorn Helgaas <bhelgaas@...gle.com>, linux-pci@...r.kernel.org,
	linux-kernel@...r.kernel.org, Val Packett <val@...kett.cool>,
	Guenter Roeck <linux@...ck-us.net>
Subject: Re: [PATCH for-linus 1/1] PCI: Revert early bridge resource set up

On Tue, Oct 14, 2025 at 07:36:02PM +0300, Ilpo Järvinen wrote:
> The commit a43ac325c7cb ("PCI: Set up bridge resources earlier") moved
> bridge window resources set up earlier than before. The change was
> necessary to support another change that got pulled on the last minute
> due to breaking s390 and other systems.
> 
> The presence of valid bridge window resources earlier than before
> allows pci_assign_unassigned_root_bus_resources() call from
> pci_host_probe() assign the bridge windows. Some host bridges, however,
> have to wait first for the link up event before they can enumerate
> successfully (see e.g. qcom_pcie_global_irq_thread()) and thus the bus
> has not been enumerated yet while calling pci_host_probe().
> 
> Calling pci_assign_unassigned_root_bus_resources() without results from
> enumeration can result in sizing bridge windows with too small sizes
> which cannot be later corrected after the enumeration has completed
> because bridge windows have become pinned in place by the other
> resources.
> 
> Interestingly, it seems pci_read_bridge_bases() is not called at all in
> the problematic case and the bridge window resource type setup is done
> by pci_bridge_check_ranges() and sizing by the usual resource fitting
> logic.
> 
> The root problem behind all this looks pretty generic. If resource
> fitting is called too early, the hotplug reservation and old size lower
> bounding cause the bridge windows to be assigned without children but
> with non-zero size, which leads to these pinning problems. As such,
> this can likely be solved on the general level but the solution does
> not look trivial.
> 
> As the commit a43ac325c7cb ("PCI: Set up bridge resources earlier") was
> prequisite for other change that did not end up into kernel yet, revert
> it to resolve the resource assignment failures and give time to code
> and test a generic solution.
> 
> Reported-by: Val Packett <val@...kett.cool>
> Reported-by: Guenter Roeck <linux@...ck-us.net>
> Fixes: a43ac325c7cb ("PCI: Set up bridge resources earlier")
> Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@...ux.intel.com>

Applied to pci/for-linus for v6.18, thanks!

> ---
> 
> This revert should go to for-linus.
> 
> 
> I'm not sure whether Guenter's case is exactly the same problem as
> described in the commit message, I only know for sure his bisection
> landed on the same commit.
> 
> My plan is to retry these changes with more supporting changes. It
> looks PCI core could delay assigning the bridge window resources if
> there are no child resource to put into the bridge windows. Or
> alternatively the resource fitting algorithm could release empty bridge
> windows as the first step. But that is too complicated change to make
> now and would benefit from time spent in -next.
> 
> ---
>  drivers/pci/probe.c | 13 +++----------
>  1 file changed, 3 insertions(+), 10 deletions(-)
> 
> diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> index c83e75a0ec12..0ce98e18b5a8 100644
> --- a/drivers/pci/probe.c
> +++ b/drivers/pci/probe.c
> @@ -538,14 +538,10 @@ static void pci_read_bridge_windows(struct pci_dev *bridge)
>  	}
>  	if (io) {
>  		bridge->io_window = 1;
> -		pci_read_bridge_io(bridge,
> -				   pci_resource_n(bridge, PCI_BRIDGE_IO_WINDOW),
> -				   true);
> +		pci_read_bridge_io(bridge, &res, true);
>  	}
>  
> -	pci_read_bridge_mmio(bridge,
> -			     pci_resource_n(bridge, PCI_BRIDGE_MEM_WINDOW),
> -			     true);
> +	pci_read_bridge_mmio(bridge, &res, true);
>  
>  	/*
>  	 * DECchip 21050 pass 2 errata: the bridge may miss an address
> @@ -583,10 +579,7 @@ static void pci_read_bridge_windows(struct pci_dev *bridge)
>  			bridge->pref_64_window = 1;
>  	}
>  
> -	pci_read_bridge_mmio_pref(bridge,
> -				  pci_resource_n(bridge,
> -						 PCI_BRIDGE_PREF_MEM_WINDOW),
> -				  true);
> +	pci_read_bridge_mmio_pref(bridge, &res, true);
>  }
>  
>  void pci_read_bridge_bases(struct pci_bus *child)
> 
> base-commit: 2f2c7254931f41b5736e3ba12aaa9ac1bbeeeb92
> -- 
> 2.39.5
>