[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <1159876414.17553.29.camel@localhost.localdomain>
Date: Tue, 03 Oct 2006 12:53:34 +0100
From: Alan Cox <alan@...rguk.ukuu.org.uk>
To: Andi Kleen <ak@...e.de>
Cc: Randy Dunlap <rdunlap@...otime.net>,
Linus Torvalds <torvalds@...l.org>, akpm <akpm@...l.org>,
Jesper Juhl <jesper.juhl@...il.com>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH/RFC] Math-emu kills the kernel on Athlon64 X2 II --
it's terminally broken
Ar Maw, 2006-10-03 am 12:11 +0200, ysgrifennodd Andi Kleen:
> On Tuesday 03 October 2006 05:24, Randy Dunlap wrote:
>
> Actually I looked at the code more closely. It looks like kernel math
> emulation is much more broken. e.g. kernel_fpu_begin() is missing
> code and lots of other paths in i387 that need to check HAVE_HWFP don't.
That check would be wrong anyway. A kernel built with FPU emulation
boots, runs and uses the hardware FPU code correctly. The problem area
is the X86_FEATURE_foo stuff. I think that comes down to a single thing
- if we have FPU disabled clear X86_FEATURE_FXSR|X86_FEATURE_MMX|
X86_FEATURE_SSE* in the boot cpu features during early option
parsing/setup. Basically the emulated FPU is forgetting to tell the
truth about the fact its a very basic FPU.
Alan
-
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/
Powered by blists - more mailing lists