linux-kernel - Re: [GIT PULL] x86 fixes for v3.14-rc6

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>] [day] [month] [year] [list]

Message-ID: <20140308052410.14411.qmail@science.horizon.com>
Date:	8 Mar 2014 00:24:10 -0500
From:	"George Spelvin" <linux@...izon.com>
To:	hpa@...or.com
Cc:	linux@...izon.com, linux-kernel@...r.kernel.org, sbsiddha@...il.com
Subject: Re: [GIT PULL] x86 fixes for v3.14-rc6

> The FPU stuff is still pending, I fear.  I haven't heard anything from
> Suresh so I suspect I'm going to have to dig into the init specifics
> myself and fix up the patchset.

As I mentioned a couple of weeks ago in the "[PATCH] Make
math_state_restore() save and restore the interrupt flag" thread,
I've been running with just Suresh's second patch for the last month,
and it's cured all my symptoms.

Although, just as a minor style point, I might reduce the
number of levels if indentation by rewriting it as:

 void __kernel_fpu_end(void)
 {
-	if (use_eager_fpu())
-		math_state_restore();
-	else
+	if (!use_eager_fpu()) {
 		stts();
+	} else if (likely(tsk_used_math(current)))
+		math_state_restore();
+	}
+	/*
+	 * When using eager fpu, tsk_used_math() is almost always true,
+	 * and we restore the user math state eagerly. In some special
+	 * cases during thread exit, signal handling etc, tsk_used_math()
+	 * is false. Those few places will take proper actions, so we
+	 * don't need to restore the math here.
+	 */
 }
 EXPORT_SYMBOL(__kernel_fpu_end);

Truthfully, the idea I like best is Linus's proposal to postpone all
such adjustments to the processor state until the return to user space,
so multiple kernel_fpu_begin()/_end() pairs become much cheaper:
http://marc.info/?l=linux-kernel&m=139128398717816

Hoerver, that deep into Here Be Dragons code is somewhere I fear to tread,
so I'm hoping someone braver than I is inspired.
--
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/