Re: [PATCH Bugfix v2 2/4] x86/xsaves: Define and use user_xstate_size for xstate size in signal context

From: Dave Hansen <dave.hansen@intel.com>
To: Fenghua Yu <fenghua.yu@intel.com>,
	"H. Peter Anvin" <hpa@linux.intel.com>,
	Ingo Molnar <mingo@elte.hu>, Thomas Gleixner <tglx@linutronix.de>,
	Asit K Mallick <asit.k.mallick@intel.com>,
	Glenn Williamson <glenn.p.williamson@intel.com>
Cc: linux-kernel <linux-kernel@vger.kernel.org>, x86 <x86@kernel.org>
Subject: Re: [PATCH Bugfix v2 2/4] x86/xsaves: Define and use user_xstate_size for xstate size in signal context
Date: Tue, 28 Apr 2015 15:09:17 -0700	[thread overview]
Message-ID: <5540050D.2000008@intel.com> (raw)
In-Reply-To: <1429678319-61356-3-git-send-email-fenghua.yu@intel.com>

On 04/21/2015 09:51 PM, Fenghua Yu wrote:
> +	/*
> +	 * Clear xcomp_bv[63] in user's xsave area header to indicate
> +	 * buf_fx is in standard format.
> +	 */
> +	xcomp_bv = xsave->xsave_hdr.xcomp_bv;
> +	user_xsave = buf_fx;
> +	xcomp_bv &= ~((u64)1 << 63);

I ran in to another bug.  xrestor_user() is hitting a #GP with these values:

[    6.258743] xrestore_user() tmp: ffff88003f813000
[    6.261122] 	     XCR0: 000000000000001f
[    6.261868] 	xstate_bv: 0000000000000003
[    6.262613] 	 xcomp_bv: 000000000000001f

I think it is because bit 63 is clear in xcomp_bv, but there are other
bits set in there.

I think the above needs to just do:

	/*
	 * We are uncompacting the state for the user buffer.  We need
	 * to clear out the xcomp_bv field entirely.  The uncompacted
	 * form of xsave/xrstor treats this field as reserved.
	 */
	if (__put_user(0, &user_xsave->xsave_hdr.xcomp_bv))
		return -1;