From owner-kaio@oss.sgi.com Sat Jul  7 13:12:07 2001
Received: (from majordomo@localhost)
	by oss.sgi.com (8.11.2/8.11.3) id f67KC7K12317
	for kaio-outgoing; Sat, 7 Jul 2001 13:12:07 -0700
Received: from longsword.omniti.com (IDENT:exim@longsword.omniti.com [216.0.51.134])
	by oss.sgi.com (8.11.2/8.11.3) with SMTP id f67KC5V12314
	for <kaio@oss.sgi.com>; Sat, 7 Jul 2001 13:12:05 -0700
Received: from scout-3.omniti.com ([216.0.51.173] helo=localhost)
	by longsword.omniti.com with esmtp (Exim 3.22 #2)
	id 15IyQt-0001pe-00
	for kaio@oss.sgi.com; Sat, 07 Jul 2001 16:12:03 -0400
Date: Sat, 7 Jul 2001 16:19:14 -0400
Content-Type: text/plain;
	format=flowed;
	charset=us-ascii
Mime-Version: 1.0 (Apple Message framework v388)
From: Theo Schlossnagle <jesus@omniti.com>
To: kaio@oss.sgi.com
X-Mailer: Apple Mail (2.388)
Content-Transfer-Encoding: 7bit
Subject: KAIO, Linux 2.4.5 XFS.
Message-Id: <E15IyQt-0001pe-00@longsword.omniti.com>
Sender: owner-kaio@oss.sgi.com
Precedence: bulk

Hi,

   First, I want to make sure this project isn't dead.  If it is, please 
let me know.

   I am using linux 2.4.5 + xfs (linux-2.4.5-xfs-06112001.patch) and the 
kaio-kern-1.3.1-2.4.2 patch.  There were some conflicts with syscalls, 
but they were very easy to rectify.

   I am employing aio to asynchronously write to files using SIGEV_SIGNAL 
notification.  the baisc flow is to do something like:

register signal handler (SIGUSR1)
open file
aio_write(10 bytes offset 0)
aio_write(100 bytes offset 10)
aio_write(50 bytes offset 110)

The aio_write's uses the sigevent framework to pass in a pointer 
(sigev_value.sival_ptr) and the "last" aio_write on a file descriptor if 
marked as such and the handler calls close().

I include <features.h> and <linux/aio.h> before anything else.  and I 
link against -ldba (not -lpthread as the application is single threaded 
for now).

I am working on an xfs partition and everytime I run my program, the 
file I create is 135036928 bytes in size instead of the just over 100 
bytes it should be.  Also, I have a kaiod-xxxxx processes (thread) that 
will not die and is CPU bound (100% all the time).  Needless to say, it 
make things unusable.

Does KAIO just not work on XFS (I haven't tried it yet on ext2)?  Or am 
I doing something fundamentally wrong?

Any help would be appreciated.

--
Theo Schlossnagle
1024D/82844984/95FD 30F1 489E 4613 F22E  491A 7E88 364C 8284 4984
2047R/33131B65/71 F7 95 64 49 76 5D BA  3D 90 B9 9F BE 27 24 E7