Go Back   nV News Forums > Linux Support Forums > NVIDIA Linux

Newegg Daily Deals

Reply
 
Thread Tools
Old 08-14-06, 07:31 PM   #25
StevenChamberla
Registered User
 
Join Date: Jul 2006
Posts: 14
Default Re: Crashing when SMP enabled

Quote:
Originally Posted by JaXXoN
Just to be sure: did you run cyclictest as root user?
Yes I did, it won't start as a non-privileged user.

But what I do notice is for a second or two the output is steady, and reads something like this:
Code:
T: 0 ( 1559) P: 0 I:    1000 C:    5812 Min:     839 Act:    1264 Avg:    1545 Max:    1871
I recompiled this kernel without debugging support, the HRT timer is still enabled, and now the numbers from cyclictest seem to increase at a slower rate after 500000. After several minutes I see:
Code:
T: 0 ( 1559) P: 0 I:    1000 C:  258356 Min:     839 Act: 1242887 Avg:  622159 Max: 1242994
...and after several more minutes...
T: 0 ( 1559) P: 0 I:    1000 C:  622329 Min:     839 Act: 2938731 Avg: 1461496 Max: 2938779
The value for 'Max' now only increases in random bursts.

Perhaps this is really an accurate figure? If it is measured in nanoseconds this really could be the latency I'm encountering. It's quite a disappointment if that's true. 'jackd' only rarely suffers xruns now, but I imagine there's still something very wrong. I've done away with 'reiserfs' and moved everything to 'ext3' but that hasn't helped.
StevenChamberla is offline   Reply With Quote
Old 08-14-06, 08:17 PM   #26
JaXXoN
Registered User
 
Join Date: Jul 2005
Location: Munich
Posts: 910
Default Re: Crashing when SMP enabled

You are getting truely strange numbers! The measurement is in
microseconds, btw. Max should stay below 100 for a nanosleep call:

Code:
cyclictest -c 1 -n -p 80 -t 1 -i 1000
options:
c = clock select: 0=CLOCK_MONOTONIC, 1=CLOCK_REALTIME
n = use clock_nanosleep
p = task priority
r = use relative timer instead of absolute timer
i = base intervall in microseconds
t = number of threads.

You may try different options and check if it makes any difference.
This might help isolating the problem.

Did you also applied "idle=poll" as kernel boot option?

regards

Bernhard
JaXXoN is offline   Reply With Quote
Old 08-14-06, 09:02 PM   #27
StevenChamberla
Registered User
 
Join Date: Jul 2006
Posts: 14
Default Re: Crashing when SMP enabled

Quote:
Originally Posted by JaXXoN
Max should stay below 100 for a nanosleep call:
Code:
cyclictest -c 1 -n -p 80 -t 1 -i 1000
Ah, much better, but well above 100. The '-n' option alone stops the number from increasing beyond 1200 or so.
Code:
T: 0 (26423) P:80 I:    1000 C:  240520 Min:       5 Act:     420 Avg:     512 Max:    1189
I left cyclictest running in X and started a compile running. 'jackd' suffered an xrun but the figure for 'Max' did not increase.
Code:
From 'jackd':
**** alsa_pcm: xrun of at least 1.108 msecs
I even tried without X or 'nvidia' module loaded, and managed a max figure of '1014' when running a compile in the background. Switching virtual terminals brings the figure to over 4000 but I believe there are unavoidable latencies invovled with VT switching.

Quote:
Originally Posted by JaXXoN
Did you also applied "idle=poll" as kernel boot option?
Yes, I was using this boot option, as well as 'pci=conf1' and 'noirqbalancer' as suggested earlier. I also used 'noapic' because without APIC the IRQs for my devices are such that the sound card has a higher priority than the disks, video, and network interface.

I tried booting without any of these options and still got a figure of '1184' for max. Again, I ran a compile which caused 'jackd' to xrun but the value in cyclictest did not increase. The compile was running under the 'TS' scheduler; the main 'jackd' thread, as well as the threads 'feeding' JACK, were all running with 'FF' scheduler.

So, am I right in thinking this indicates a problem with the realtime patch? Or rather, there is a latency somewhere which has not been fixed by the realtime patch. If so, I should probably learn how to debug the realtime patch and hopefully find out where the latencies are coming from.
StevenChamberla is offline   Reply With Quote
Old 08-14-06, 10:03 PM   #28
JaXXoN
Registered User
 
Join Date: Jul 2005
Location: Munich
Posts: 910
Default Re: Crashing when SMP enabled

Do you have PAT support re-enabled (as described in the other
thread)? But even then, for me, worst case latency has never
been more then something about 600 microseconds (because
of cache flushing when a 3D app start or stops).

When no 3D app runs, latencies should actually never be
higher that 30 microseconds. If you have high latencies,
especially under high disk load, i then suspect there might
be a similar somewhat hardware related problem as with
the nforce4 SATA controller. For details, please check the
following thread:
http://www.uwsg.iu.edu/hypermail/lin...03.1/2539.html

BTW.: I never recognized high latencies while switching VTs, but
i have to admit that i didn't checked that extensivly, because
the nvidia driver tends to freeze the machine when doing so :-)

Can you please re-run your tests with fbdev and/or the OSS
nvidia driver? If you still get high latencies, then you should
contact Ingo Molnar and Thomas Gleixner (the realtime preemption
patch maintainers).

regards

Bernhard
JaXXoN is offline   Reply With Quote
Reply


Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Forum Jump


Similar Threads
Thread Thread Starter Forum Replies Last Post
Acer Completes its Windows 8 Lineup with Touch Enabled All-in-Ones News Archived News Items 0 06-04-12 07:10 AM
OpenGL apps crashing in NVIDIA driver when switching desktop dinosaur NVIDIA Linux 0 06-01-12 08:09 PM
AGP not being enabled on Geforce2 mx frankjr NVIDIA Linux 1 09-22-02 03:01 PM
SMP system hangs with OpenGL chazmati NVIDIA Linux 9 08-26-02 10:28 AM
X freezes when agp 4x is enabled jinksed NVIDIA Linux 1 08-07-02 05:56 PM

All times are GMT -5. The time now is 07:47 AM.


Powered by vBulletin® Version 3.7.1
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Copyright 1998 - 2014, nV News.