Messages in this thread | | | Date | Thu, 18 Apr 2024 17:48:45 -0700 | From | Josh Poimboeuf <> | Subject | Re: [PATCH v3] x86/bugs: Only harden syscalls when needed |
| |
On Wed, Apr 17, 2024 at 07:01:54PM +0100, Andrew Cooper wrote: > On 17/04/2024 6:57 pm, Josh Poimboeuf wrote: > > On Wed, Apr 17, 2024 at 09:45:14AM -0700, Pawan Gupta wrote: > >> On Wed, Apr 17, 2024 at 04:14:26PM +0100, Andrew Cooper wrote: > >>> On 17/04/2024 12:02 am, Josh Poimboeuf wrote: > >>>> diff --git a/arch/x86/kernel/cpu/bugs.c b/arch/x86/kernel/cpu/bugs.c > >>>> index ca295b0c1eee..dcb97cc2758f 100644 > >>>> --- a/arch/x86/kernel/cpu/bugs.c > >>>> +++ b/arch/x86/kernel/cpu/bugs.c > >>>> @@ -1678,6 +1687,21 @@ static void __init spectre_v2_select_mitigation(void) > >>>> enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline(); > >>>> enum spectre_v2_mitigation mode = SPECTRE_V2_NONE; > >>>> > >>>> + /* > >>>> + * X86_FEATURE_INDIRECT_SAFE indicates whether indirect calls can be > >>>> + * considered safe. That means either: > >>>> + * > >>>> + * - the CPU isn't vulnerable to Spectre v2 or its variants; > >>>> + * > >>>> + * - a hardware mitigation is in place (e.g., IBRS, BHI_DIS_S); or > >>>> + * > >>>> + * - the user turned off mitigations altogether. > >>>> + * > >>>> + * Assume innocence until proven guilty: set the cap bit now, then > >>>> + * clear it later if/when needed. > >>>> + */ > >>>> + setup_force_cpu_cap(X86_FEATURE_INDIRECT_SAFE); > >>> Following on from the (re)discovery that X86_FEATURE_RETPOLINE is a poor > >>> name given what it *actually* does, can I recommend s/SAFE/OK/ here? > >> Or simply X86_FEATURE_USE_INDIRECT_BRANCH. > >> > >>> This flag really is "do I want indirect branches or not", which - as > >>> noted here - is more than just a judgement of whether indirect branches > >>> are "safe". > > X86_FEATURE_USE_INDIRECT_BRANCH sounds good. It's a bit long but does > > describe it better. > > Works for me. Definitely an improvement over SAFE.
USE_INDIRECT_BRANCH is now irking me: "use indirect branch for what? when? why?"
At the moment I'm leaning towards Andrew's suggestion of X86_FEATURE_INDIRECT_OK as it at least says it's "ok" (safe or don't care) to use indirect branches when desired (typically performance raisins). I'll probably go with that (or maybe X86_FEATURE_INDIRECT_BRANCH_OK) unless anybody yells.
-- Josh
| |