linux-mips
[Top] [All Lists]

Re: PREEMPT

To: Ralf Baechle <ralf@linux-mips.org>
Subject: Re: PREEMPT
From: john cooper <john.cooper@timesys.com>
Date: Thu, 06 Oct 2005 15:05:07 -0400
Cc: Greg Weeks <greg.weeks@timesys.com>, linux-mips@linux-mips.org, john cooper <john.cooper@timesys.com>
In-reply-to: <20051006185504.GD15275@linux-mips.org>
Original-recipient: rfc822;linux-mips@linux-mips.org
References: <43456EA9.8020209@timesys.com> <20051006184656.GA12173@linux-mips.org> <43457266.3090208@timesys.com> <20051006185504.GD15275@linux-mips.org>
Sender: linux-mips-bounce@linux-mips.org
User-agent: Mozilla Thunderbird 0.8 (X11/20040913)
Ralf Baechle wrote:
On Thu, Oct 06, 2005 at 02:52:22PM -0400, Greg Weeks wrote:


I'd remembered some problems ages ago, but had thought they'd been fixed. John was just picking my brain about it so I thought I'd ask to be sure.


The problems I recall were all related to being preempted just while
fiddling with the hardware FPU - can't happen on the fpu-less 4Kc.
Another issue fixesd recently even though more cosmetic were a bunch
of global variables.

The code base I'm dealing with is a 2.6.13 derivative
with PREEMPT_RT support.  Looks like the problem was
due to PREEMPT_RT confusing fpu_emulator_cop1Handler()
resulting in a SIGBUS nailing the associated task.

I have it sort of working for soft FPU but expect it
requires some attention to safely access a HW FPU where
emulation assistance is needed.

-john

--
john.cooper@timesys.com

<Prev in Thread] Current Thread [Next in Thread>