RE: Geode GX1, video acceleration -> crash

RE: Geode GX1, video acceleration -> crash

[Keyser Soze]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset=us-ascii, Size: 551 bytes --]

> But the real problem is, that I wanted to benchmark
> the system while the scrolling continues, and issued
> a dd if=/dev/mem of=/dev/null bs=1024 count=32768
> command. For the second go, the system freezed like
> a good refrigerator. No kernel panic, nothing, just
> freezed. 

Try turning off ide dma and see if that helps.  You
will lose very little by turning off udma on this
system and I'll bet you end up being more stable.


__________________________________
Do you Yahoo!?
SBC Yahoo! DSL - Now only $29.95 per month!
http://sbc.yahoo.com

RE: Geode GX1, video acceleration -> crash

[Alan Cox]

On Gwe, 2003-07-11 at 19:10, Keyser Soze wrote:
> Try turning off ide dma and see if that helps.  You
> will lose very little by turning off udma on this
> system and I'll bet you end up being more stable.

You'll lose a lot of disk performance turning off UDMA on a Geode GX1,
down from full UDMA33 to about 1Mbyte/second. It should be rock solid
at least my Geode's are. Slow... but solid

Re: Geode GX1, video acceleration -> crash

[Ferenc Engard]

Keyser Soze wrote:
> 
> > But the real problem is, that I wanted to benchmark
> > the system while the scrolling continues, and issued
> > a dd if=/dev/mem of=/dev/null bs=1024 count=32768
> > command. For the second go, the system freezed like
> > a good refrigerator. No kernel panic, nothing, just
> > freezed.
> 
> Try turning off ide dma and see if that helps.  You
> will lose very little by turning off udma on this
> system and I'll bet you end up being more stable.

I will try it on Monday, as the eval board is in my workplace. What is
the connection between ide dma, memory read and the hw video accel? The
ide dma setting alters the way /dev/mem is read? :-O

I had a feeling that maybe it is a dma-related problem, but I do not
know what does 'dd if=/dev/mem......' _really_ do. It uses a dma
channel? The GX1 manual do not mention that the bitblk engine use up
dma; in fact, I do think that it is a serious design error that using
the bitlbk engine alters the way the memory can be used (except memory
bandwith usage, naturally). 

Anyway, if it solves the problem, I must understand why happened this
error. The product will be a medical bedside monitor; so it is not
enough that it is now _more_ stable, I have to know the origin of the
problem.

I can tell you more on Monday,
Circum

PS: The system will run from flash (on IDE controller), so I suspect
that it will be a _real big_ slowdown if in the last product I have to
turn of the dma, but hopefully there will be no gigabytes of data...

Re: Geode GX1, video acceleration -> crash

[Jan Rychter]

[-- Attachment #1: Type: text/plain, Size: 1061 bytes --]

>>>>> "Ferenc" == Ferenc Engard <ferenc@engard.hu> writes:
 Ferenc> Keyser Soze wrote:
 >>
 > But the real problem is, that I wanted to benchmark the system while
 > the scrolling continues, and issued a dd if=/dev/mem of=/dev/null
 > bs=1024 count=32768 command. For the second go, the system freezed
 > like a good refrigerator. No kernel panic, nothing, just freezed.
 >>
 >> Try turning off ide dma and see if that helps.  You will lose very
 >> little by turning off udma on this system and I'll bet you end up
 >> being more stable.

 Ferenc> I will try it on Monday, as the eval board is in my
 Ferenc> workplace. What is the connection between ide dma, memory read
 Ferenc> and the hw video accel? The ide dma setting alters the way
 Ferenc> /dev/mem is read? :-O
[...]

It can have interesting things in common, but NS won't tell you about
them without NDAs, I'm afraid -- at least that's how it used to be.
I've had IDE DMA interfere with video input, for instance. So asking NS
is indeed the best answer in this thread.

--J.

[-- Attachment #2: Type: application/pgp-signature, Size: 188 bytes --]

Re: Geode GX1, video acceleration -> crash

[NunO fELICIO]

Yes, i have some troubles with that, do not use the fb driver from national,
look at http://www.directfb.org/ , they have very good drivers =).


Nuno Felicio
---------------------
Systems and Coca Cola

----- Original Message -----
From: "Ferenc Engard" <ferenc@engard.hu>
To: <linux-kernel@vger.kernel.org>
Sent: Thursday, July 03, 2003 5:24 PM
Subject: Geode GX1, video acceleration -> crash


> Hello there,
>
> First of all, I am new to the framebuffer business, and looks like
> that the linux-fbdev.org site is down. Where can I find a mailing list
> or other information resource about framebuffers in general, and about
> Geode hardware specifically? It is a nightmare to find any information
> for Geode on linux... :((
>
> I am developing a graphical app for an Advantech PCM-5820 (I think :),
> i.e. a Geode GX1 / CS5530 board, 64MB RAM, which I want to run a linux
> / framebuffer application on.
>
> Now, about my problem:
>
> I have installed the 2.4.17 kernel, patched with NSC's original Geode
> fb driver (nsc-kfb-driver-2.7.7.tar.gz), and compiled it successfully.
> Also, as I couldn't find in other places, I have downloaded the
> nsc_xfree_2.7.6.tgz package just to compile the GAL library
> (nsc_galfns.c) in it, as this was what I needed.
>
> First surprise: I cannot switch into 32bpp modes! Did I miss
> something? fbset refuses it, setting at boot time do not work either.
>
> Next, I have written a small application to test the processor's
> bitblt capability. The program calls Gal_screen_to_screen_blt() to
> scroll a rectangle on the screen, then usleeps a bit. It does work,
> although the scheduler do not give back the run to the task until
> approx. 20ms elapses (no problem at now), and in 16bpp, it do not
> scroll the region but inverses it(?!).  I suppose that it is not good
> that the console writes out things while the bitblt engine works.
> Question: should I disable writing to the console while my app runs?
> How?
>
> But the real problem is, that I wanted to benchmark the system while
> the scrolling continues, and issued a
> dd if=/dev/mem of=/dev/null bs=1024 count=32768
> command. For the second go, the system freezed like a good
> refrigerator. No kernel panic, nothing, just freezed. It can be
> repeated, if I copy just the 1st MB of RAM, then it freezes for the
> 5-6th go. :((
>
> What can I do? How to debug?
>
> Please cc your answers to ferenc@engard.hu, too!
>
> Thank you:
> Circum
>
> -
> 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/
>

Re: Geode GX1, video acceleration -> crash

[Alan Cox]

On Iau, 2003-07-03 at 17:24, Ferenc Engard wrote:
> I have installed the 2.4.17 kernel, patched with NSC's original Geode
> fb driver (nsc-kfb-driver-2.7.7.tar.gz), and compiled it successfully.
> Also, as I couldn't find in other places, I have downloaded the
> nsc_xfree_2.7.6.tgz package just to compile the GAL library
> (nsc_galfns.c) in it, as this was what I needed

Really you should take to the geode folks at NSC (or whatever NSC spun
them off as), and whatever user lists they have

Geode GX1, video acceleration -> crash

[Ferenc Engard]

Hello there,

First of all, I am new to the framebuffer business, and looks like
that the linux-fbdev.org site is down. Where can I find a mailing list
or other information resource about framebuffers in general, and about
Geode hardware specifically? It is a nightmare to find any information
for Geode on linux... :((

I am developing a graphical app for an Advantech PCM-5820 (I think :),
i.e. a Geode GX1 / CS5530 board, 64MB RAM, which I want to run a linux
/ framebuffer application on.

Now, about my problem:

I have installed the 2.4.17 kernel, patched with NSC's original Geode
fb driver (nsc-kfb-driver-2.7.7.tar.gz), and compiled it successfully.
Also, as I couldn't find in other places, I have downloaded the
nsc_xfree_2.7.6.tgz package just to compile the GAL library
(nsc_galfns.c) in it, as this was what I needed.

First surprise: I cannot switch into 32bpp modes! Did I miss
something? fbset refuses it, setting at boot time do not work either.

Next, I have written a small application to test the processor's
bitblt capability. The program calls Gal_screen_to_screen_blt() to
scroll a rectangle on the screen, then usleeps a bit. It does work,
although the scheduler do not give back the run to the task until
approx. 20ms elapses (no problem at now), and in 16bpp, it do not
scroll the region but inverses it(?!).  I suppose that it is not good
that the console writes out things while the bitblt engine works.
Question: should I disable writing to the console while my app runs?
How?

But the real problem is, that I wanted to benchmark the system while
the scrolling continues, and issued a
dd if=/dev/mem of=/dev/null bs=1024 count=32768
command. For the second go, the system freezed like a good
refrigerator. No kernel panic, nothing, just freezed. It can be
repeated, if I copy just the 1st MB of RAM, then it freezes for the
5-6th go. :((

What can I do? How to debug?

Please cc your answers to ferenc@engard.hu, too!

Thank you:
Circum