linux-kernel - Re: math_state_restore and kernel_fpu

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <alpine.DEB.2.10.1401191150520.2663@minerva.lan>
Date:	Sun, 19 Jan 2014 12:02:22 -0700 (MST)
From:	Nate Eldredge <nate@...tsmathematics.com>
To:	George Spelvin <linux@...izon.com>
cc:	adilger@...ger.ca, jack@...e.cz, linux-kernel@...r.kernel.org,
	viro@...iv.linux.org.uk, Suresh Siddha <suresh.b.siddha@...el.com>,
	Arjan van de Ven <arjan@...ux.intel.com>,
	Ingo Molnar <mingo@...e.hu>,
	Thomas Gleixner <tglx@...utronix.de>,
	Maarten Baert <maarten-baert@...mail.com>
Subject: Re: math_state_restore and kernel_fpu_end disable interrupts?

[Here's my original message, since George's reply didn't quote or 
reference it: https://lkml.org/lkml/2014/1/18/3.  Summary: 
math_state_restore() always leaves interrupts disabled, and I think this 
is a bug.]

On Sun, 19 Jan 2014, George Spelvin wrote:

> THANK YOU!
>
> I've been having a problem with ext4 metadata checksums, which use SSE
> for large blocks, and traced it to kernel_fpu_end() disabling interrupts,
> but had paused to debug this (I assumed well-tested) piece of kernel
> code before pushing it harder.

Interesting.  I guess it's not surprising this has other effects.

Here's the commit that added the code in question (about 6 years ago): 
https://github.com/torvalds/linux/commit/aa283f49276e7d840a40fb01eee6de97eaa7e012

It's credited to Suresh Siddha, whom I've cc'ed (along with others who 
signed off).  Suresh, if you're still around, could you comment on why 
math_state_restore always leaves interrupts disabled, regardless of their 
state on entry?  Is there a deep reason or is it a bug?

Assuming it's a bug, here's the obvious patch:

--- linux-source-3.11.0/arch/x86/kernel/traps.c	2013-09-02 14:46:10.000000000 -0600
+++ linux-source-3.11.0-nate/arch/x86/kernel/traps.c	2014-01-19 11:25:32.977221476 -0700
@@ -624,6 +624,9 @@
  	struct task_struct *tsk = current;

  	if (!tsk_used_math(tsk)) {
+		unsigned long flags;
+
+		local_save_flags(flags);
  		local_irq_enable();
  		/*
  		 * does a slab alloc which can sleep
@@ -635,7 +638,7 @@
  			do_group_exit(SIGKILL);
  			return;
  		}
-		local_irq_disable();
+		local_irq_restore(flags);
  	}

  	__thread_fpu_begin(tsk);

I tested it briefly: the kernel still boots fine, and it fixes the problem 
I was seeing (BUG() when core dumping on ecryptfs).  George, does it help 
your problem?

Thanks everyone!

> (Search October-December LKML archives for "3.11.4: kernel BUG at
> fs/buffer.c:1268".")

-- 
Nate Eldredge
nate@...tsmathematics.com

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/