On 02/01/18 22:58, Adam Carter wrote: > AMD coder's patch to disable the new code (to avoid the performance hit) > where he states the issue doesnt exist on AMD processors; > https://lkml.org/lkml/2017/12/27/2
Read LWN, specifically the links to the people who covered the bug. It's a flaw in speculative forward processing, where the security does not travel with the speculative processing. So user code can trigger a page fault that references kernel code, causing that page to be retrieved. OOPPSSSS. AMD keeps security context with the code, causing an attempt to exploit the bug to fail with "invalid security context". And as I understand it the code can be disabled with either a compile time option or command line switch to the kernel. The relevant code is called KAISER, which forces kernel and user address space into different contexts, and causes a nasty context-switching overhead on both Intel and AMD cpus. Cheers, Wol
