Re: [patch 0/8] x86/fpu: Mop up XSAVES and related damage

[Thomas Gleixner]

On Fri, Jun 04 2021 at 15:04, Dave Hansen wrote:
> The nice Intel 0day folks threw some tests at this series.  It managed
> to trigger an oops.  I can't right this moment publish the source for
> the test, but it looks pretty trivial.  It basically creates a 0'd XSAVE
> buffer, sets XCOMP_BV to:
>
> #define XSAVES_FEATURES ( \
>         XFEATURE_MASK_PT | \
>         XFEATURE_MASK_SHSTK_USER | \
>         XFEATURE_MASK_SHSTK_KERNEL | \
>         0x8000000000000000 \
>         )
>
> Then passes that buffer in to ptrace(PTRACE_SETREGSET, ...).
>
> The oops is below.  It doesn't *look* to be XSAVES-related.  The oops
> looks like it's happening in xstateregs_set() itself as opposed to down
> in the code actually concerned with supervisor state.
>
> No bug is jumping out of the code as I took a brief look at it.  The
> xbuf versus kbuf code looks a bit wonky, but I can't find a hole in it.

I can....

--- a/arch/x86/kernel/fpu/regset.c
+++ b/arch/x86/kernel/fpu/regset.c
@@ -128,7 +128,7 @@ int xstateregs_set(struct task_struct *t
 		xbuf = vmalloc(count);
 		if (!xbuf)
 			return -ENOMEM;
-		ret = user_regset_copyin(&pos, &count, NULL, &ubuf, xbuf, 0, -1);
+		ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, xbuf, 0, -1);
 		if (ret)
 			goto out;
 	}