linux-kernel - Re: [PATCH v1] platform: Fix race condition during DMA configure at IOMMU probe time

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

Message-ID: <20250423165711.GG1213339@ziepe.ca>
Date: Wed, 23 Apr 2025 13:57:11 -0300
From: Jason Gunthorpe <jgg@...pe.ca>
To: Robin Murphy <robin.murphy@....com>
Cc: Bjorn Helgaas <helgaas@...nel.org>,
	Will McVicker <willmcvicker@...gle.com>,
	Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
	"Rafael J. Wysocki" <rafael@...nel.org>,
	Danilo Krummrich <dakr@...nel.org>,
	"Rob Herring (Arm)" <robh@...nel.org>,
	Lorenzo Pieralisi <lpieralisi@...nel.org>,
	Joerg Roedel <jroedel@...e.de>, Bjorn Helgaas <bhelgaas@...gle.com>,
	iommu@...ts.linux.dev, Saravana Kannan <saravanak@...gle.com>,
	kernel-team@...roid.com, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v1] platform: Fix race condition during DMA configure at
 IOMMU probe time

On Wed, Apr 23, 2025 at 04:50:29PM +0100, Robin Murphy wrote:

> Indeed in my half-written version of this patch I was leaning towards
> removing the drv variable altogether (just doing
> to_x_driver(dev->driver)->driver_managed_dma inline), or at least doing the
> same as Will's previous diff. I figure the one-liner replacing
> "!dev->driver" with "!&drv->driver" would be too disgustingly non-obvious
> for anyone else's tastes...

The READ_ONCE alerts the reader that something weird is going on here :\

> For consistency we should really fix all the buses the same way - sorry for
> the bother (I can write up the other patches if you'd like). FWIW this part
> really was the most temporary stopgap, as my planned next step is to propose
> moving driver_managed_dma and the use_default_domain() call up into the
> driver core and so removing all this bus-level code anyway, hence trying to
> minimise the effort spent churning it. Oh well.

It started out this way but Greg really didn't want it, so this was
the compromise. Perhaps you can convince Greg of the merit.

Jason