Messages in this thread | | | From | Andy Lutomirski <> | Date | Thu, 23 Apr 2015 08:52:16 -0700 | Subject | Re: [PATCH] x86/asm/entry/64: better check for canonical address |
| |
On Thu, Apr 23, 2015 at 8:49 AM, Borislav Petkov <bp@alien8.de> wrote: > On Thu, Apr 23, 2015 at 08:41:15AM -0700, Andy Lutomirski wrote: >> I was rather vague there. Let me try again: >> >> If anyone in the AMD camp really cared, we could add a new bug flag >> X86_BUG_SYSRET_NEEDS_CANONICAL_RCX and set it on Intel chips only, so >> we could use alternatives to patch out the check when running on >> sensible AMD hardware. This would speed the slow path up by a couple >> of cycles on AMD chips. >> >> Does that make more sense? We could call it >> X86_BUG_SYSRET_NEEDS_CANONICAL_RIP if that makes more sense. > > Actually "...NEEDS_CANONICAL_RCX" makes more sense as this is what we're > going to patch out eventually, if it makes sense - the RIP canonicalness > test is being done as part of SYSRET, just RCX is not being tested. > > Tell you what - how about I perf stat this first by commenting out that > couple of instructions on AMD to see whether it brings anything. > > Got an idea for a workload other than a kernel build? :-) > > Although a kernel build should do a lot of syscalls too...
Kernel build should be fine. Or "timing_test_64 10 sys_enosys 1" or "perf_self_monitor" (warning: WIP). Make sure you either have context tracking forced on or something else (audit?) that forces the slow path, though, or you won't see it at all.
https://git.kernel.org/cgit/linux/kernel/git/luto/misc-tests.git/
--Andy
> > Thanks. > > -- > Regards/Gruss, > Boris. > > ECO tip #101: Trim your mails when you reply. > --
-- Andy Lutomirski AMA Capital Management, LLC
| |