Re: [PATCH] x86/hweight: Don't clobber %rdi

[Linus Torvalds]

On Mon, Aug 8, 2016 at 11:04 AM, Borislav Petkov <bp@xxxxxxxxx> wrote:
>
> Hmm, I've been staring at asm for the last hours and my head is spinning
> now, so can you please point me at the exact where this happens. I can't
> find it in x86_perf_event_update().

The corruption could easily have happened long long before.

With a random register clobbered (by a function that is *not* a C
function - it's a replacement for the "popcnt %rdi, %rax"
instruction), you might end up with memory corruption somewhere, and
then an oops much later.

That said, in this case it's likely something like the the
intel_pmu_init() doing

    c->weight = hweight64(c->idxmsk64);

in arch/x86/events/intel/core.c, which then corrupts something related
to the event constraints, and then you get the oops in
x86_perf_event_update() later.

> In any case, it is not a good idea to do the compiler's work. :-\

The compiler has absolutely nothing to do with this. It's all assembly
language and an inline asm.

We *used* to try to have the compiler generate the code. That's what
caused problems.

               Linus