diff options
| author | Andy Lutomirski <luto@kernel.org> | 2016-04-07 17:31:46 -0700 |
|---|---|---|
| committer | Ingo Molnar <mingo@kernel.org> | 2016-04-13 10:20:41 +0200 |
| commit | 7a5d67048745e3eab62779c6d043a2e3d95dc848 (patch) | |
| tree | c3052c7d6e5b675ced4c3a7edb2f7a6c2c7bb65f /arch/x86/kernel | |
| parent | d47b50e7a111bb7a56fb1c974728b56209d7f515 (diff) | |
x86/cpu: Probe the behavior of nulling out a segment at boot time
AMD and Intel do different things when writing zero to a segment
selector. Since neither vendor documents the behavior well and it's
easy to test the behavior, try nulling fs to see what happens.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Reviewed-by: Borislav Petkov <bp@suse.de>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rudolf Marek <r.marek@assembler.cz>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/61588ba0e0df35beafd363dc8b68a4c5878ef095.1460075211.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Diffstat (limited to 'arch/x86/kernel')
| -rw-r--r-- | arch/x86/kernel/cpu/common.c | 31 |
1 files changed, 31 insertions, 0 deletions
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c index 7fea4079d102..8e40eee5843a 100644 --- a/arch/x86/kernel/cpu/common.c +++ b/arch/x86/kernel/cpu/common.c @@ -889,6 +889,35 @@ static void detect_nopl(struct cpuinfo_x86 *c) #endif } +static void detect_null_seg_behavior(struct cpuinfo_x86 *c) +{ +#ifdef CONFIG_X86_64 + /* + * Empirically, writing zero to a segment selector on AMD does + * not clear the base, whereas writing zero to a segment + * selector on Intel does clear the base. Intel's behavior + * allows slightly faster context switches in the common case + * where GS is unused by the prev and next threads. + * + * Since neither vendor documents this anywhere that I can see, + * detect it directly instead of hardcoding the choice by + * vendor. + * + * I've designated AMD's behavior as the "bug" because it's + * counterintuitive and less friendly. + */ + + unsigned long old_base, tmp; + rdmsrl(MSR_FS_BASE, old_base); + wrmsrl(MSR_FS_BASE, 1); + loadsegment(fs, 0); + rdmsrl(MSR_FS_BASE, tmp); + if (tmp != 0) + set_cpu_bug(c, X86_BUG_NULL_SEG); + wrmsrl(MSR_FS_BASE, old_base); +#endif +} + static void generic_identify(struct cpuinfo_x86 *c) { c->extended_cpuid_level = 0; @@ -921,6 +950,8 @@ static void generic_identify(struct cpuinfo_x86 *c) get_model_name(c); /* Default name */ detect_nopl(c); + + detect_null_seg_behavior(c); } static void x86_init_cache_qos(struct cpuinfo_x86 *c) |