linux-kernel - Issue seen since commit f5ff79fddf0e ("dma-mapping: remove CONFIG_DMA

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

Date:   Mon, 10 Oct 2022 11:57:39 -0700
From:   Jerry Snitselaar <jsnitsel@...hat.com>
To:     Christoph Hellwig <hch@....de>, Robin Murphy <robin.murphy@....com>
Cc:     Joerg Roedel <joro@...tes.org>, iommu@...ts.linux.dev,
        linux-kernel@...r.kernel.org
Subject: Issue seen since commit f5ff79fddf0e ("dma-mapping: remove
 CONFIG_DMA_REMAP")

I've been looking at an odd issue that shows up with commit
f5ff79fddf0e ("dma-mapping: remove CONFIG_DMA_REMAP").  What is being
seen is the bnx2fc driver calling dma_free_coherent(), and eventually
hits the BUG_ON() in vunmap().  bnx2fc_free_session_resc() does a
spin_lock_bh() around the dma_free_coherent() calls, and looking at
preempt.h that will trigger in_interrupt() to return positive, so that
makes sense. The really odd part is this only happens with the
shutdown of the kernel after a system install. Reboots after that do not
hit the BUG_ON() in vunmap().

I still need to grab a system and try to see what it is doing on the
subsequent shutdowns, because it seems to me that any time
bnx2fc_free_session_resc() is called it will end up there, unless the
allocs are not coming from vmalloc() in the later boots. Between the
comments in dma_free_attrs(), and preempt.h, dma_free_coherent()
shouldn't be called under a spin_lock_bh(), yes? I think the comments
in dma_free_attrs() might be out of date with commit f5ff79fddf0e
("dma-mapping: remove CONFIG_DMA_REMAP") in place since now it is more
general that you can land in vunmap(). Also, should that WARN_ON() in
dma_free_attrs() trigger as well for the BH disabled case?

It was also reproduced with a 6.0-rc5 kernel build[1].

Regards,
Jerry

[1] https://koji.fedoraproject.org/koji/buildinfo?buildID=2061881