| Hard crashes with device,/retain=2 ?? [message #34478] |
Thu, 20 March 2003 10:22  |
condor
Messages: 35
Registered: January 2002
|
Member |
|
|
For some reason the default retain=1 started working for me a while
ago -- I never had to care about backing store but somewhere during
the upgrade from 5.3 to 5.5 or there around I lost the ability to
obsure windows. This was when I got a new computer (P4-2.8 if it makes
a difference, running RedHat8.0 with whatever version of XF86 is the
default, so the problem might well have to do with X and not IDL).
No problem I thought and added the line "device,retain=2" to my
startup file. Over the next week or two I experience hard crashes on
my bot -- about once a day. I've been noodling around unixoid systems
for a decade now and I've NEVER had to hit a reset button - until now.
At some point I started suspecting it might have to do with the
omnious warnings in the IDL help that retain=2 should only be used for
really important windows and I took the line out of my startup file.
Lo, I haven't had any of those crashes since.
Coincidence? Possibly -- so I figured I'd ask around if others have
experienced similar strangenesses around retain=2 or if this is maybe
even a well-known problem with 5.5 or something. (Yes, I have
installed the patch for those color-bands ...) Or if there's just
something funny with my box.
|
|
|
|
| Re: Hard crashes with device,/retain=2 ?? [message #34546 is a reply to message #34478] |
Wed, 26 March 2003 08:43  |
Karl Schultz
Messages: 341
Registered: October 1999
|
Senior Member |
|
|
"Big Bird" <condor@biosys.net> wrote in message
news:df160b8f.0303252001.2c26459a@posting.google.com...
> Nigel Wade <nmw@ion.le.ac.uk> wrote in message
news:<b5en3i$j8l$1@south.jnrs.ja.net>...
> {quoting myself in full for coherence -- and correcting some horrible
> typos in the process:}
>
>> Big Bird wrote:
>>
>>> For some reason the default retain=1 stopped working for me a while
>>> ago -- I never had to care about backing store but somewhere during
>>> the upgrade from 5.3 to 5.5 or there around I lost the ability to
>>> obscure windows. This was when I got a new computer (P4-2.8 if it
makes
>>> a difference, running RedHat8.0 with whatever version of XF86 is the
>>> default, so the problem might well have to do with X and not IDL).
>>>
>>> No problem I thought and added the line "device,retain=2" to my
>>> startup file. Over the next week or two I experienced hard crashes on
>>> my box -- about once a day. I've been noodling around unixoid systems
>>> for a decade now and I've NEVER had to hit a reset button - until now.
>>> At some point I started suspecting it might have to do with the
>>> omnious warnings in the IDL help that retain=2 should only be used for
>>> really important windows and I took the line out of my startup file.
>>> Lo, I haven't had any of those crashes since.
>>>
>>> Coincidence? Possibly -- so I figured I'd ask around if others have
>>> experienced similar strangenesses around retain=2 or if this is maybe
>>> even a well-known problem with 5.5 or something. (Yes, I have
>>> installed the patch for those color-bands ...) Or if there's just
>>> something funny with my box.
>>
>> That sounds like a video card driver issue to me.
>>
>> What's the video card? (It's not an NVidia, using their drivers, by any
>> chance is it?).
>
> By any chance it is indeed some generic Nvidia GF4 (whatever Dell
> ships with their 8250) and it does indeed use the latest detonator
> (4191? Something like that) drivers.
>
> Your response makes me hopeful that this might be a well-known
> problem?
This almost certainly is a problem with the X server and/or its drivers.
The X server runs in its own process and theoretically it should be
impossible for any client program, IDL or something else, to crash the X
server. But X servers have their bugs and so do the drivers. You are
working with some relatively new hardware and software, so there can easily
be some problems, even though the entire Linux/X picture has improved quite
a bit over the past couple of years.
Also, changing IDL's RETAIN setting affects how IDL uses the X server a
great deal, which in turn determines what code paths get executed in the X
server as IDL runs. This may explain why you have problems with one RETAIN
setting and not another. For example, the server may be having trouble
copying the "backing store" pixmap to your window.
What to do? Here are a few ideas:
- Make sure you have all the possible patches/upgrades for the X server
itself.
- (More important) Make sure you have the latest drivers for your video
card. It looks like you have already done this. I have had some rather
astounding results by just upgrading drivers. nVidia is known for their
driver stability and I've often been pleasantly surprised by the results
people get after they update their drivers. This may sound like an ISV
cop-out, but it is true.
- Carefully read the driver release notes and documentation. See if there
are any settings in the XFree86 config file that you can play with to try
some experiments. For example, there may be a setting to turn off the usage
of hardware accelerators on the card. This would let you test the server in
"software" mode, which may bypass some bug or problem in the vendor driver
code. (For example, set the "driver" field to "vga" in the Device section.)
You may not want to run in this mode all the time, but it would at least let
you isolate some causes.
- Along the same lines, look for an option that would turn on server backing
store. The reason why you switched from RETAIN=1 to RETAIN=2 is because
your new X server no longer did its own backing store. You might be able to
turn on an option to restore this behavior, letting you use RETAIN=1 again.
Another way to try doing this is to provide the -wm option when you start
the X server, which forces backing store on when the window is mapped. (You
may have to set your runlevel to 3 and start the server with startx or xinit
to try this). You should be able to turn on WhenMapped in the ServerOptions
section of the XF86Config-4 config file, but I can't tell from the docs if
you can do that.
Hope this helps,
Karl
|
|
|
|
| Re: Hard crashes with device,/retain=2 ?? [message #34549 is a reply to message #34478] |
Wed, 26 March 2003 03:48 |
Nigel Wade
Messages: 286
Registered: March 1998
|
Senior Member |
|
|
Big Bird wrote:
>> That sounds like a video card driver issue to me.
>>
>> What's the video card? (It's not an NVidia, using their drivers, by any
>> chance is it?).
>
> By any chance it is indeed some generic Nvidia GF4 (whatever Dell
> ships with their 8250) and it does indeed use the latest detonator
> (4191? Something like that) drivers.
>
> Your response makes me hopeful that this might be a well-known
> problem?
The NVidia Linux drivers are notoriously poor. I've had all sorts of
problems with them with TwinView, hard lockups, not setting the mode
correctly for switching between X and a consoles, failing to set the right
mode when exiting X and more. But, being closed source no-one but NVidia
can fix them to work properly.
You do need to be sure that you have the correct NVidia drivers for the
kernel you are using, it's very easy to download the wrong ones, or to
upgrade your kernel and not re-create the kernel interface module for the
drivers.
What happens if you use the generic xv driver supplied with XFree86? I know
you won't get any hardware acceleration via OpenGL, but if you're not using
it the driver might be more stable.
--
Nigel Wade, System Administrator, Space Plasma Physics Group,
University of Leicester, Leicester, LE1 7RH, UK
E-mail : nmw@ion.le.ac.uk
Phone : +44 (0)116 2523548, Fax : +44 (0)116 2523555
|
|
|
|
|
|
|
|
| Re: Hard crashes with device,/retain=2 ?? [message #34911 is a reply to message #34670] |
Fri, 25 April 2003 10:39 |
condor
Messages: 35
Registered: January 2002
|
Member |
|
|
This is posted mostly for reference for those poor souls who run into
the same problem in the future and google for it and wonder what
actually helped.
dmarino@digitalglobe.com (dmarino) wrote in message news:<1c23e5f8.0304080600.51f5f77f@posting.google.com>...
> It turned out the Nvidia cards had some kind of 30 bpp mode
> running that wasn't a valid color class for IDL. Maybe some
> games and scanners can use 30 bpp? I never saw a 10/10/10 bit
> display before.
I think this part is covered by appendix D in the nvidia driver
README:
Option "Dac8Bit" "boolean"
Most Quadro parts by default use a 10 bit color look
up table (LUT) by default; setting this option to TRUE
forces
these graphics chips to use an 8 bit (LUT). Default:
a 10 bit LUT is used, when available.
> One other thing we tried that seemed to help, there's some settings in
> your XF86Config file that deal with off-screen memory pixmaps. Turn
> them on and off and see what happens. I think having that enabled and
> using retain=2 may cause a conflict.
As it turned out, none of this made a difference. What helped finally
was the following two lines in the "Screen" section of the XF86Config:
Section "Screen"
Option "backing_store"
Option "save_unders"
Yes, the "Screen"-section, not the "ServerFlags" section. No, this
does not seem to be documented anywhere: I got this from a usenet post
and the poster admitted that he got that from a usenet post as well.
It sure isn't anywhere in 'man Xserver', 'man XF86Config', the nvidia
driver documentation or the www.xfree86.org web site.
I am now running with the default "retain" and after a couple weeks of
almost daily crashes, I haven't had one in almost a week, so I'm
assuming this is what needed fixing.
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
