Yu, Regarding your patch to disable AVX with eager fpu disabled in the 4.4 stable queue: https://git.kernel.org/pub/scm/linux/kernel/git/stable/stable-queue.git/tree/queue-4.4/x86-fpu-disable-mpx-when-eagerfpu-is-off.patch Why would this be necessary? In the Intel SDM Vol 2, Chapter 2.4, FPU/SIMD instructions are grouped into different classes and for all of them CR0.TS triggers a #NM exception in all modes but real mode and virtual 8086 mode. It seems to me lazy context switching of AVX is perfectly functional (disregarding the recently disclosed security issue). Am I missing something? Julian
Attachment:
signature.asc
Description: OpenPGP digital signature