Message-ID: <49507534.5080401@ct.jp.nec.com>
Date: Mon, 22 Dec 2008 21:20:52 -0800
From: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
User-Agent: Thunderbird 2.0.0.18 (Windows/20081105)
MIME-Version: 1.0
To: Ingo Molnar <mingo@elte.hu>, Thomas Gleixner <tglx@linutronix.de>,
       "H. Peter Anvin" <hpa@zytor.com>
Cc: linux-kernel@vger.kernel.org
Subject: [RFC -tip 0/4] x86: improve uaccess in signal
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1900
Lines: 57

This patch series is experimental.

This is another proposal to improve uaccess in signal.
This patch series will improve signal handling performance. The normal path of
signal handling should be same as this;
http://lkml.org/lkml/2008/9/25/332

I think there is a lot of fixup code in signal generated by __{get|put}_user().
In most case kernel only needs to know whether there is error, and error value
is not important. And when an exception has occurred, kernel doesn't need to
continue following __{get|put}_user() in that function.

I think in fixup code to use direct jump to the tail of function and returns
error, will save fixup code size. This fixup code can be used __{get|put}_user()
calls in same function.

The concept code is;
int func()
{
	int err = 0;
.section .fixup
error_out:
	err = -EFAULT;
	goto out;
.previous
	__{get|put}_user(); /* fixup: jump to error_out */
	__{get|put}_user();
	__{get|put}_user();
	:
out:
	return err;
}

and the results are;
$ size *signal*.o.*
   text	   data	    bss	    dec	    hex	filename
   4646	      0	      0	   4646	   1226	ia32_signal.o.new
   6006	      0	      0	   6006	   1776	ia32_signal.o.old
   3557	      0	      0	   3557	    de5	signal.o.new
   4540	      0	      0	   4540	   11bc	signal.o.old
   3840	      0	      0	   3840	    f00	signal32.o.new
   4876	      0	      0	   4876	   130c	signal32.o.old

However, this code is tricky, I can't say compiler guarantees to keep the
order of this framework code and this code relies on the order, fixup section
is before the all __{get|put}_user() codes and the out label is just before
the return.

Comments are welcome.

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