linux-kernel - Re: [PATCH] PCI/portdrv: Allow probing even without child services

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: <aWio-NB1csIhZJen@wunner.de>
Date: Thu, 15 Jan 2026 09:44:40 +0100
From: Lukas Wunner <lukas@...ner.de>
To: Brian Norris <briannorris@...omium.org>
Cc: Bjorn Helgaas <bhelgaas@...gle.com>, linux-pci@...r.kernel.org,
	linux-kernel@...r.kernel.org,
	Manivannan Sadhasivam <mani@...nel.org>
Subject: Re: [PATCH] PCI/portdrv: Allow probing even without child services

On Wed, Jan 14, 2026 at 11:30:19AM -0800, Brian Norris wrote:
> On Sat, Jan 10, 2026 at 07:10:52AM +0100, Lukas Wunner wrote:
> > On Fri, Jan 09, 2026 at 03:20:13PM -0800, Brian Norris wrote:
> > > @@ -355,29 +355,18 @@ static int pcie_port_device_register(struct pci_dev *dev)
> > >  	if (status) {
> > >  		capabilities &= PCIE_PORT_SERVICE_HP;
> > >  		if (!capabilities)
> > > -			goto error_disable;
> > > +			return 0;
> > >  	}
> > 
> > This will keep the Bus Master Enable bit set (see call to
> > pci_set_master() further up in the function), even though
> > no MSIs are expected from the device.  (I *think* these
> > would be the only memory writes that a port would perform.)
> > 
> > That doesn't seem right.  If there are no services, it seems
> > prudent to clear Bus Master Enable again (as is done by
> > pci_disable_device() right now).
> 
> Seems like a reasonable suggestion. I'll try pci_clear_master() in some
> of these no-op non-failure cases.
> 
> Do you have the same concerns if pcie_init_service_irqs() falls back to
> INTx but does not fail? It seems like a potentially fraught exercise to
> guess what child services might need bus mastering though, so maybe it's
> better to limit this only to nr_service==0 cases?

Sounds reasonable to me to constrain to nr_service==0.
Basically just retain the existing behavior.

I note that pcie_portdrv_remove() calls pci_disable_device()
unconditionally.  You may need an extra struct with an extra flag
to remember whether pci_disable_device() needs to be called on remove.

Thanks,

Lukas