linux-kernel - Re: [PATCH] PCI/pwrctrl: Only destroy alongside host bridge

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: <CAMRc=Meepp_5WS2Tdu2gevUbv-_D_Xb-NfAneP5UBYJNck22Vw@mail.gmail.com>
Date: Thu, 17 Jul 2025 21:20:14 +0200
From: Bartosz Golaszewski <brgl@...ev.pl>
To: Brian Norris <briannorris@...omium.org>
Cc: Manivannan Sadhasivam <mani@...nel.org>, Bjorn Helgaas <bhelgaas@...gle.com>, linux-pci@...r.kernel.org, 
	Rob Herring <robh@...nel.org>, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] PCI/pwrctrl: Only destroy alongside host bridge

On Thu, Jul 17, 2025 at 12:25 AM Brian Norris <briannorris@...omium.org> wrote:
>
> On Wed, Jul 16, 2025 at 09:47:41AM -0700, Brian Norris wrote:
> > (2) Even after resolving 1, I'm seeing pci_free_host_bridge() exit with
> >     a bridge->dev.kboj.kref refcount of 1 in some cases. I don't yet
> >     have an explanation of that one.
>
> Ah, well now I have an explanation:
> One should always be skeptical of out-of-tree drivers.
>
> In this case, one of my endpoint drivers was mismanaging a pci_dev_put()
> reference count, and that cascades to all its children and links,
> including the host bridge.
>
> Once I fix that (and the aforementioned problem (1)), it seems my
> problems go away.
>
> I'll let a v2 soak in my local environment, and unless I hear some news
> from Bartosz about OF_POPULATED to change my mind, I'll send it out
> eventually.
>
> Brian

Hi! Sorry for the late reply, I would really like to be able to assist
with these changes (although Mani is doing a great job!) I'm currently
really busy with other stuff. :( FWIW I just spent 30 minutes looking
at the tree as of commit f1536585588b~1 and I am no longer sure what
exactly did I refer to when I said that the PCI core clears the
OF_POPULATED flag but I'm 100% sure I was facing this issue and seeing
OF nodes associated with a device that's registered without this flag.

Looking at it again now, it's no longer obvious, I wish I had been
more verbose in the commit message. Feel free to try and revert this
change, maybe over a year later it's no longer needed (or never was).
If it is, we should quickly see some issues triggered by it.

Bartosz