oss-sec mailing list archives

Security sensitive bug in the i915 kernel driver (CVE-2022-4139)


From: Andrzej Hajda <andrzej.hajda () intel com>
Date: Wed, 30 Nov 2022 11:22:16 +0100

Hi all,

[This is a public disclosure of an issue reported 7 days ago to linux-distros () vs openwall org. CVE-2022-4139 has been assigned to the issue since.]

Incorrect GPU TLB flush code has been discovered in i915 kernel driver.
In some cases (Gen12 hardware with specific types of engine) the engine's TLB is not flushed at all. Depending on whether the GPU is running behindĀ an active IOMMU there are two possible scenarios which can happen, due to stale TLB mapping: 1. Without IOMMU - GPU can still access physical memory which could be already assigned by OS to different process. 2. With IOMMU - GPU can access any memory, if the malicious process is able to create/reuse necessary IOMMU mappings.

It is currently not known if specific memory could be targeted, but random memory corruption or data leaks are a known possibility.

All Intel integrated and discrete GPUs Gen12 are affected, including Tiger Lake, Rocket Lake, Alder Lake, DG1, Raptor Lake, DG2, Arctic Sound, Meteor Lake. Fix has already been developed and consists of fixing the method of writing to specific registers. I am attaching a set of back-ported patches which implement the fix for all affected stable branches (all since 5.4).

This vulnerability has similar impact as CVE-2022-0330[1].

I will try to follow Linux Security Process[2]. So I hope to send the fix for public mailing list after 7 days.

[1]:https://nvd.nist.gov/vuln/detail/cve-2022-0330
[2]:https://www.kernel.org/doc/html/latest/admin-guide/security-bugs.html

Regards
Andrzej


Attachment: media-tlb.tar
Description:


Current thread: