From: "Yao, Yuan" <yuan.yao@intel.com>
To: Dave Hansen <dave.hansen@linux.intel.com>,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Cc: "Bae, Chang Seok" <chang.seok.bae@intel.com>,
"x86@kernel.org" <x86@kernel.org>,
"Hansen, Dave" <dave.hansen@intel.com>,
Thomas Gleixner <tglx@linutronix.de>
Subject: RE: [PATCH] x86/fpu: Remove dynamic features from xcomp_bv for init_fpstate
Date: Thu, 13 Oct 2022 01:33:11 +0000 [thread overview]
Message-ID: <BYAPR11MB3717FC0492B36C16AAB95F2095259@BYAPR11MB3717.namprd11.prod.outlook.com> (raw)
In-Reply-To: <20221011222425.866137-1-dave.hansen@linux.intel.com>
>-----Original Message-----
>From: Dave Hansen <dave.hansen@linux.intel.com>
>Sent: Wednesday, October 12, 2022 06:24
>To: linux-kernel@vger.kernel.org
>Cc: Bae, Chang Seok <chang.seok.bae@intel.com>; x86@kernel.org; Yao, Yuan <yuan.yao@intel.com>; Hansen, Dave
><dave.hansen@intel.com>; Thomas Gleixner <tglx@linutronix.de>
>Subject: [PATCH] x86/fpu: Remove dynamic features from xcomp_bv for init_fpstate
>
>From: Yuan Yao <yuan.yao@intel.com>
>
>This was found a couple of months ago in a big old AMX
>backport. But, it looks to be a problem in mainline too.
>Please let me know if this looks OK. I'd also especially
>appreciate some testing from folks that have AMX hardware
>handy.
>
>Builds and survives a quick boot test on non-AMX hardware.
>
>--
>
>== Background ==
>
>'init_fpstate' is a sort of template for all of the fpstates
>that come after it. It is copied when new processes are
>execve()'d or XRSTOR'd to get fpregs into a known state.
>
>That means that it represents the *starting* state for a
>process's fpstate which includes only the 'default' features.
>Dynamic features can, of course, be acquired later, but
>processes start with only default_features.
>
>During boot the kernel decides whether all fpstates will be
>kept in the compacted or uncompacted format. This choice is
>communicated to the hardware via the XCOMP_BV field in all
>XSAVE buffers, including 'init_fpstate'.
>
>== Problem ==
>
>But, the existing XCOMP_BV calculation is incorrect. It uses
>the set of 'max_features', not the default features.
>
>As a result, when XRSTOR operates on buffers derived from
>'init_fpstate', it may attempt to "tickle" dynamic features which
>are at offsets for which there is no space allocated in the
>fpstate.
>
>== Scope ==
>
>This normally results in a relatively harmless out-of-bounds
>memory read. It's harmless because it never gets consumed. But,
>if the fpstate is next to some unmapped memory, this "tickle" can
>cause a page fault and an oops.
>
>This only causes issues on systems when dynamic features are
>available and when an XSAVE buffer is next to uninitialized
>memory. In other words, it only affects recent Intel server
>CPUs, and in relatively few memory locations.
>
>Those two things are why it took relatively long to catch this.
>
>== Solution ==
>
>Use 'default_features' to establish the init_fpstate
>xcomp_bv value. Reset individual fpstate xcomp_bv values
>when the rest of the fpstate is reset.
>
>[ dhansen: add reset code from tglx, rewrites
> commit message and comments ]
Thanks to Dave's help on this! The new comment message is more
clear and easier to understand then before.
>
>Fixes: 1c253ff2287f ("x86/fpu: Move xstate feature masks to fpu_*_cfg")
>Suggested-by: Dave Hansen <dave.hansen@intel.com>
>Suggested-by: Thomas Gleixner <tglx@linutronix.de>
>Signed-off-by: Yuan Yao <yuan.yao@intel.com>
>Cc: stable@vger.kernel.org
>---
> arch/x86/kernel/fpu/core.c | 3 +++
> arch/x86/kernel/fpu/xstate.c | 7 ++++++-
> 2 files changed, 9 insertions(+), 1 deletion(-)
>
>diff --git a/arch/x86/kernel/fpu/core.c b/arch/x86/kernel/fpu/core.c
>index 3b28c5b25e12..4d64de34da12 100644
>--- a/arch/x86/kernel/fpu/core.c
>+++ b/arch/x86/kernel/fpu/core.c
>@@ -526,6 +526,9 @@ static void __fpstate_reset(struct fpstate *fpstate, u64 xfd)
> fpstate->xfeatures = fpu_kernel_cfg.default_features;
> fpstate->user_xfeatures = fpu_user_cfg.default_features;
> fpstate->xfd = xfd;
>+
>+ /* Ensure that xcomp_bv matches ->xfeatures */
>+ xstate_init_xcomp_bv(&fpstate->regs.xsave, fpstate->xfeatures);
> }
>
> void fpstate_reset(struct fpu *fpu)
>diff --git a/arch/x86/kernel/fpu/xstate.c b/arch/x86/kernel/fpu/xstate.c
>index c8340156bfd2..f9f45610c72f 100644
>--- a/arch/x86/kernel/fpu/xstate.c
>+++ b/arch/x86/kernel/fpu/xstate.c
>@@ -360,7 +360,12 @@ static void __init setup_init_fpu_buf(void)
>
> print_xstate_features();
>
>- xstate_init_xcomp_bv(&init_fpstate.regs.xsave, fpu_kernel_cfg.max_features);
>+ /*
>+ * 'init_fpstate' is sized for the default feature
>+ * set without any of the dynamic features.
>+ */
>+ xstate_init_xcomp_bv(&init_fpstate.regs.xsave,
>+ fpu_kernel_cfg.default_features);
>
> /*
> * Init all the features state with header.xfeatures being 0x0
>--
>2.34.1
next prev parent reply other threads:[~2022-10-13 1:33 UTC|newest]
Thread overview: 15+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-10-11 22:24 [PATCH] x86/fpu: Remove dynamic features from xcomp_bv for init_fpstate Dave Hansen
2022-10-11 22:47 ` Chang S. Bae
2022-10-13 1:33 ` Yao, Yuan [this message]
2022-10-13 1:47 ` Chang S. Bae
2022-10-13 3:35 ` Yao, Yuan
2022-10-13 16:23 ` Chang S. Bae
2022-10-13 17:21 ` Dave Hansen
2022-10-13 17:33 ` Chang S. Bae
2022-10-13 17:44 ` Dave Hansen
2022-10-14 3:53 ` Chang S. Bae
2022-10-14 4:10 ` Yao, Yuan
2022-10-14 4:26 ` Chang S. Bae
2022-10-14 4:03 ` Yao, Yuan
2022-10-13 18:04 ` Dave Hansen
2022-10-17 22:39 ` Chang S. Bae
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=BYAPR11MB3717FC0492B36C16AAB95F2095259@BYAPR11MB3717.namprd11.prod.outlook.com \
--to=yuan.yao@intel.com \
--cc=chang.seok.bae@intel.com \
--cc=dave.hansen@intel.com \
--cc=dave.hansen@linux.intel.com \
--cc=linux-kernel@vger.kernel.org \
--cc=tglx@linutronix.de \
--cc=x86@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).