On 01/09/2018 02:11 AM, Tim W wrote:
On Tuesday, January 9, 2018 at 1:16:10 AM UTC-5, Sven Semmler wrote:
Great time to be using a AMD chipset as they are not effected.
Just got back from a small seminar on the topic. All modern processors with speculative execution units are likely effected by this.
Wonder if something like this would have been caught years ago if the microcode was open?
It would not make any difference, as a microcode patch is not able to fix the underlying problems in the architecture. The problem lies in the kernel memory cache system vs the speculative branch prediction portions of the CPU, and microcode does not generally coordinate these separate hardware units.
When you have multiple branches of code independently executing in a given CPU core the kernel can be tricked into loading kernel memory into cache, which is then able to be accessed/hammered to copy that data back out into userspace. I heard one quote that the kernel data can be read at up to 5 kbits/sec by a carefully constructed application.
Since it takes a locally running application to do this trick the flaw is disastrous for cloud services. Thus allowing anyone to execute arbitrary code in your virtualization system could be giving away all the other VM's secrets. Probably not a problem if you trust the code you are running on a single user system like Qubes, but even signed code from your repo should be considered suspect for data exfiltration purposes with this issue unpatched.
This is a big one in terms of the effects it has when mitigated at the software level. I wonder what the performance hit will be from application of whatever patch route Qubes takes? Projections of 5-30% hit. As I said Great day for AMD stock LOL
Not a good day for any CPU vendor as far as I can see, because anything advanced enough to give good performance via speculative execution now needs to pull back on the reigns until there is a architectural solution. Likely the next-gen processors will actually fix it, but that could take years given the modern development cycle time frames.
There are all kinds of patches being worked on to get around this, but they all show poor performance. We may see patches with better performance as time goes on in specific instances, but for right now "slow", by actually defeating speculative execution, seems to be the solution.
-- You received this message because you are subscribed to the Google Groups "qubes-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to qubes-users+unsubscr...@googlegroups.com. To post to this group, send email to qubes-users@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/qubes-users/5ba19c63-c3a1-afb1-da9a-155507263978%40jhuapl.edu. For more options, visit https://groups.google.com/d/optout.
