lkml.org 
[lkml]   [2016]   [Feb]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: WARNING: CPU: 0 PID: 3031 at ./arch/x86/include/asm/fpu/internal.h:530 fpu__restore+0x90/0x130()
    On Thu, Feb 11, 2016 at 05:16:00PM -0800, Andy Lutomirski wrote:
    > Are you running 32-bit userspace by any chance?

    Sure, that's a 32-bit kernel testing partition. :)

    > I'm guessing you're hitting this in __fpu_restore_sig:

    Yeah, I was looking at that too.

    > fpu__drop(fpu);
    > if (__copy_from_user(&fpu->state.xsave, buf_fx, state_size) ||
    > __copy_from_user(&env, buf, sizeof(env))) {
    > fpstate_init(&fpu->state);
    > err = -1;
    > } else {
    > sanitize_restored_xstate(tsk, &env, xfeatures, fx_only);
    > }
    >
    > fpu->fpstate_active = 1;
    >
    > <-- preempted right here

    Yeah, that could explain why I'm seeing it.

    > if (use_eager_fpu()) {
    > preempt_disable();
    > fpu__restore(fpu);
    > preempt_enable();
    > }
    >
    > I don't see why this code deserves to work. If I'm right, it can be
    > fixed by pulling the preempt_disable out of the if (use_eager_fpu())
    > to right above the fpstate_active = 1 line. Don't bother trying to
    > optimize the !use_eager_fpu() case.

    Right.

    > Once someone gets around to eagerly *allocating* the FPU context and
    > dropping CR0.TS usage entirely, then even that won't be enough unless
    > we do my suggesting of deferring FPU restore to
    > prepare_exit_to_usermode. (Doing that will make all of this much,
    > much more sane.)

    Sounds good to me.

    So the thing with this issue is, is that I don't have a reproducer yet.
    It happened randomly.

    So let me ask it this way: can anything go wrong if we pull up the
    preemption disabled region? I mean, do we even care about the !eager FPU
    case? I'd much prefer that vs FPU state corruption...

    --
    Regards/Gruss,
    Boris.

    ECO tip #101: Trim your mails when you reply.

    \
     
     \ /
      Last update: 2016-02-15 20:21    [W:7.208 / U:0.184 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site