linux-kernel - Re: [regression] Bug 217218 - Trying to boot Linux version 6-2.2 kernel with Marvell SATA controller 88SE9235

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: <6a775920-9260-f82e-dace-0b792f5371c4@arm.com>
Date:   Mon, 17 Apr 2023 12:21:58 +0100
From:   Robin Murphy <robin.murphy@....com>
To:     Christoph Hellwig <hch@....de>
Cc:     Linux regressions mailing list <regressions@...ts.linux.dev>,
        LKML <linux-kernel@...r.kernel.org>,
        Linux PCI <linux-pci@...r.kernel.org>, iommu@...ts.linux.dev,
        baolu.lu@...ux.intel.com
Subject: Re: [regression] Bug 217218 - Trying to boot Linux version 6-2.2
 kernel with Marvell SATA controller 88SE9235

On 2023-04-16 07:41, Christoph Hellwig wrote:
> On Thu, Mar 30, 2023 at 01:18:45PM +0100, Robin Murphy wrote:
>> FWIW "Marvell SATA" instantly makes me suspect the phantom function quirk.
>> What *should* happen is the IOMMU driver sees the PCI DMA aliases correctly
>> and sets up context entries for both 07.00.0 and 07.00.1, but it looks like
>> that may be what's gone awry.
> 
> Looking at the bug report it seems this is device 9235, which doesn't
> need the DMA alias quirks.

Indeed that one doesn't appear to be in the quirk list currently. 
However the symptom of DMA traffic from function 1 which the IOMMU 
clearly wasn't expecting firmly suggests that it *does* need the quirk. 
Digging up the original report, the lspci output there suggests that 
07:00.1 isn't a real function, which would further confirm it.

The other thing which catches my interest is the seemingly-conflicting 
"iommu=soft" and "intel_iommu=on" arguments - I could well believe that 
refactoring the x86 IOMMU detection stuff might have subtly changed the 
interaction there, such that previously it ended up not actually using 
the IOMMU for DMA ops, but now it is?

Robin.