vacancies advertise contact news tip The Vault
Countdown to 2019 and win big with the HEXUS Epic Giveaway [x]
facebook rss twitter

AMD provides update on Spectre and Meltdown patches

by Mark Tyson on 12 January 2018, 11:06

Tags: AMD (NYSE:AMD), NVIDIA (NASDAQ:NVDA), Microsoft (NASDAQ:MSFT), Intel (NASDAQ:INTC)

Quick Link: HEXUS.net/qadptd

Add to My Vault: x

AMD updates

AMD has published an update on its processor security in the wake of the Spectre and Meltdown saga. In summary it echoes Intel in that "security is our top priority and we are continually working to ensure the safety of our users as new risks arise". Discussing more timely matters, AMD says it has been working closely with OS providers to patch GPZ Variant 1 (Spectre) and OS and motherboard makers to patch GPZ Variant 2 (Spectre). For GPZ Variant 3 (Meltdown), AMD still asserts that "no mitigation is required," as its processors are not susceptible.

For GPZ Variant 1, AMD mentions that it has been working closely with Microsoft to patch for AMD processors old and new. Unfortunately there have been reports on some AMD Windows PCs 'bricking', especially with patches for older AMD processors (AMD Opteron, Athlon and AMD Turion X2 Ultra families). However, safe working updates from Microsoft should resume by next week.

For GPZ Variant 2, AMD says that while its architectures make it "difficult to exploit," it is still working with OS software developers and on issuing a microcode update. Ryzen and EPYC processor microcode updates will be available this week - from both system and OS vensors.

Moving on to GPUs, AMD says that the Radeon architectures don't use speculative execution "and thus are not susceptible to these threats".

Nvidia updates

Neatly linking to where we left off with AMD, Nvidia has issued a security bulletin regarding its driver updates for CPU speculative side channel vulnerabilities. Like AMD it asserts that its GPUs are "immune" for architectural reasons. However it is supplying driver updates to "mitigate the CPU security issue".

Nvidia's software developers will be, like AMD, providing collaborative help to OS vendors to strengthen mitigations for affected CPUs that could be affected by For GPZ Variant 1 and 2 (Spectre).



HEXUS Forums :: 24 Comments

Login with Forum Account

Don't have an account? Register today!
So Nvidia GPUs aren't vulnerable but they're releasing an update anyway? Eh?
spacein_vader
So Nvidia GPUs aren't vulnerable but they're releasing an update anyway? Eh?

It would appear that certain calls in the nvidia GPU driver could be an avenue of attack against vulnerable CPUs. Presumably some software contains more instructions and pathways that could be used as a channel for the attacks, and since GPU drivers will cross the user and kernel memory spaces to some degree it's plausible to me that you could tweak a GPU driver to make it less vulnerable to use as an attack vector.

CAT-THE-FIFTH has pointed out elsewhere that nvidia moved a lot of the scheduling for their GPUs to software a few generations ago, which may be a partial explanation for why nvidia have been particularly pro-active in optimising their drivers.
scaryjim
CAT-THE-FIFTH has pointed out elsewhere that nvidia moved a lot of the scheduling for their GPUs to software a few generations ago, which may be a partial explanation for why nvidia have been particularly pro-active in optimising their drivers.

I would hope that scheduling would be in a user mode driver, so no impact.
DanceswithUnix
I would hope that scheduling would be in a user mode driver, so no impact.

True but doesn't it, at some point, have to do a context switch so the Kernel can dispatch the data to the GPU?
Corky34
True but doesn't it, at some point, have to do a context switch so the Kernel can dispatch the data to the GPU?

No, the GPU is memory mapped and so during set up can be mapped such that a user mode process can write to it.

Kernel needs to do support stuff like route interrupts and initial hardware setup, but the main fast path you would hope will be user only.

Other hardware I have worked on functions that way, I presume Nvidia would do the same.