lkml.org 
[lkml]   [1997]   [Aug]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [yet another attempt] Re: 2.0.31-pre5: Couldn't get a free page..... (fwd)

OK...I've been using the patch from Gadi against a clean 2.0.31-pre6 that
only has the modified aic7xxx code that I'm working on. I'm able to do
things like 3 100 MB bonnies simultaneously to three disks without any bad
messages or any problems. However, that's not to say there aren't side
effects. Specifically, people need to look into the following side effect
of his patch to see if there should be a fix, or if it will have any
unknown side effects.

So, here's the problem with the patch (or it may be that &tq_scheduler is
a bad task queue to use, but I would have thought it would get run fairly
regularly). The basic idea behind the patch is to let the kernel kflushd
have the time it needs to get the number of dirty buffers down. In so
doing, it blocks other tasks. Specifically, I've been working on (and had
it working until I applied Gadi's patch :) a bottom half routine for the
aic7xxx driver. The routine calls the scsi_done function outside of the
interrupt context in order to keep interrupt latencies low. The problem
is, adding the bottom half to &tq_immediate didn't work, so I used
&tq_scheduler in order to ensure the bottom half got ran as soon as
possible after the hardware interrupts. This was great, until Gadi's
patch. With Gadi's patch, it is possible that kflushd can write so much
information, that bottom half handlers or other items on &tq_schedule
won't get run for extended periods of time. Switching this to &tq_timer
solved that problem for me, but I wanted other people to be aware that
this condition exists in case it might impact their code. Anyone who is
dependant on reasonable length scheduler intervals should check out how
their code works with this patch under heavy dirty buffer loads such as
badblocks -w or bonnie. Other than this one reservation, which should be
checked to see if it might cause any other code problems, I think the
patch does a wonderful job :)


*****************************************************************************
* Doug Ledford * Unix, Novell, Dos, Windows 3.x, *
* dledford@dialnet.net 873-DIAL * WfW, Windows 95 & NT Technician *
* PPP access $14.95/month *****************************************
* Springfield, MO and surrounding * Usenet news, e-mail and shell account.*
* communities. Sign-up online at * Web page creation and hosting, other *
* 873-9000 V.34 * services available, call for info. *
*****************************************************************************


\
 
 \ /
  Last update: 2005-03-22 13:40    [W:0.027 / U:1.464 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site