[rv370/r[cs]4xx] acceleration is ten kinds of weird

[JuanC]

Hi,

I downloaded Kubuntu Breezy Release Candidate from
http://www.kubuntu.org/announcements/breezy-release-candidate.php

I downloaded the AMD64 Install version and install it on my laptop HP Compaq
presario M2000.

After the install , KDE don't start and i try to find a solution and i found a
thread in ubuntuforums.org :
http://ubuntuforums.org/showthread.php?t=69367

I this thread tells to do this :

*****************************************************************
1. Boot into recovery mode. It will get you to a command line.

2. Type: sudo nano /etc/X11/xorg.conf

3. Find the section titled: Section "Device"

4. As the last entry under "Device" add the line: Option "NoAccel"

5. Ctrl-O to save

6. Ctrl-X to exit

7. Reboot in normal mode

8. Rejoice that your mouse will work and Gnome will load and that once again,
everything is good in the world!
******************************************************************

When you do this , KDE starts !!!! and works perfectly.

But i think this is an annoying bug , because if KDE don't start all people
thinks that kubuntu don't works.

I hope this bug would fixed for the final version of Kubuntu.

[Dennis Kaarsemaker (dennis)]

Thank you for your report. Can you please attach /etc/X11/xorg.conf and
/var/log/Xorg.0.log *after resetting the driver to the original, erroneous value*.

[JuanC (juanc1-deactivatedaccount-deactivatedaccount)]

Created an attachment (id=4492)
Xorg Error Log

[JuanC (juanc1-deactivatedaccount-deactivatedaccount)]

Created an attachment (id=4493)
Default xorg.conf file

[JuanC (juanc1-deactivatedaccount-deactivatedaccount)]

I attach the xorg error log and the xorg.conf default file that get this error.

[Daniel Stone (daniels)]

fixed in -76

[JuanC (juanc1-deactivatedaccount-deactivatedaccount)]

What's the meaning of 'fixed in -76'?

It means that this bug is fixed in the final version of Breezy Badger?

[Matt Zimmerman (mdz)]

It means that we believe this bug to be fixed in xorg 6.8.2-76 and newer. The
current (and final, except in case of a disaster) version in Breezy is 6.8.2-77.

[JuanC (juanc1-deactivatedaccount-deactivatedaccount)]

Same error in Kubuntu 5.10 Final version.

[Daniel Stone (daniels)]

*** Bug 23790 has been marked as a duplicate of this bug. ***

[Daniel Stone (daniels)]

*** Bug 24031 has been marked as a duplicate of this bug. ***

[Daniel Stone (daniels)]

*** Bug 23871 has been marked as a duplicate of this bug. ***

[Daniel Stone (daniels)]

*** Bug 24016 has been marked as a duplicate of this bug. ***

[Andy Kwong (andy-diginode)]

Bug 23790 does not appear to be a duplicate of this bug. The aforementioned bug
only exists in 6.8.2-77 and crashes with a signal 4, not with a signal 11 and
there is no "EE" line in the Xorg.conf file at all.

[Andy Kwong (andy-diginode)]

In my previous comment, please substitute xorg.conf with Xorg.0.log

[Daniel Stone (daniels)]

this bug is 'weird crashing out when we attempt to set up accel for whatever
reason', which 17628 is

[Daniel Stone (daniels)]

*** Bug 24249 has been marked as a duplicate of this bug. ***

[Daniel Stone (daniels)]

*** Bug 23375 has been marked as a duplicate of this bug. ***

[Jani Monoses (jani)]

17712, 17628 and 17863 do not seem to be duplicates of this bug even if related.

The changelog for ati-driver 6.8.2-76 has
 *Blacklist all HP X300 RV370s a la -69. Sigh. (closes: Ubuntu#16628).

That change seems to have blacklisted also HP X600 (M24) which is a 380 family
chip not 370.
For some reason this makes X crash with SIGILL on startup.

I recompiled the xorg package with the test for HP ATI pci ids 3150 and 3154
removed from the 025-radeon-patch
and it works fine accelerated just like in hoary and in 6.8.2-75.

Is there a bugzilla entry for those chips behaving erratically?

[Manfred Usselmann (usselmann-m)]

On my HP / Compaq nx8220 notebook the X server still crashes with "Fatal server
error: Caught signal 4. Server aborting" with Breezy release (X 6.8.2-77).

[joe schmid (joe-schmid)]

Also still here the same on HP / Compaq 8220 (X600)

[pepster (jheled)]

I Have a nx 8220 and have not upgraded to breezy yet because of the problem.

Can someone point here to all the valid workarounds? I really really really want
to be able to hibernate that hot beast.

Thanks, Joseph

[Jani Monoses (jani)]

Created an attachment (id=4823)
modified ati driver .deb

Install with sudo dpkg -i
it should fix X600 (pci id 3150) and possibly X700 (if it is 3154)

[Daniel Stone (daniels)]

jani: does that just add to the special casing in patch #025?

[Jani Monoses (jani)]

Just modified the patch to have FALSE instead of TRUE for these two HP ids
The same effect as not having that if() there but I did not want to change too
many characters in the patch.
This way it applied cleanly.

     case PCI_CHIP_RV380_3150:
     case PCI_CHIP_RV380_3154:
         info->IsMobility = TRUE;
+ if (info->PciInfo->subsysVendor == PCI_VENDOR_HP) {
+ info->NoAccel = FALSE;
+ }
     case PCI_CHIP_RV380_3E50:
     case PCI_CHIP_RV380_3E54:
         info->ChipFamily = CHIP_FAMILY_RV380;

[JuanC (juanc1-deactivatedaccount-deactivatedaccount)]

Any .deb package for Amd64 that patch or fixed this problem ?

[pepster (jheled)]

(In reply to comment #23)
> jani: does that just add to the special casing in patch #025?

(In reply to comment #22)
> Created an attachment (id=4823) [edit]
> modified ati driver .deb
>
> Install with sudo dpkg -i
> it should fix X600 (pci id 3150) and possibly X700 (if it is 3154)

I tested the .deb on my machine and it works. Thanks, Jani!!

[Jani Monoses (jani)]

JuanMa: no amd64 deb, unless you recompile the sources for yourself :(
either that or find an older version of the ati driver deb for amd64 - for
instance on the preview iso
or whichever daily iso you have until around the 5th of october. Install that
manually and it should be ok if it's version
6.8.2-75 or older (<75)

[Jani Monoses (jani)]

Think I found the real problem:

whether to accel or not is checked in 3 places in radeon, but only one check
uses the NoAccel field in info
the other two use the xorg.conf setting if there is one.
When this per-chip field corresponds to what is the xorg.conf setting (or lack
thereof of Option "noaccel")
all is well - either both say noaccel and things go slow but don't crash, or
both are accel and all is fine and fast.
When they disagree some codepaths take accel some don't and this results in
stack corruption and then SIGILL.

This is why either setting NoAccel in xorg or setting NoAccel=FALSE in radeon
cured the problem for X600, it made
one setting say what the other says too.

The patch below should make it not crash on any card.

And the other issue is that with this change X600/X700 will not crash but be
slow, so it would be nice either to
apply the patch in the previous comment since they seem to work fine
accelerated.Another way is to say option "accel"
in xorg.conf.

diff -rN -u old-ati/radeon_driver.c new-ati/radeon_driver.c
--- old-ati/radeon_driver.c 2005-11-10 10:05:37.000000000 +0200
+++ new-ati/radeon_driver.c 2005-11-10 10:05:37.000000000 +0200
@@ -4778,7 +4778,7 @@
                            info->CurrentLayout.pixel_bytes);
        int maxy = info->FbMapSize / width_bytes;

- if (xf86ReturnOptValBool(info->Options, OPTION_NOACCEL, FALSE)) {
+ if (xf86ReturnOptValBool(info->Options, OPTION_NOACCEL, info->NoAccel)) {
            xf86DrvMsg(scrnIndex, X_WARNING,
                       "Acceleration disabled, not initializing the DRI\n");
            info->directRenderingEnabled = FALSE;
@@ -5127,7 +5127,7 @@
     }

                                /* Acceleration setup */
- if (!xf86ReturnOptValBool(info->Options, OPTION_NOACCEL, FALSE)) {
+ if (!xf86ReturnOptValBool(info->Options, OPTION_NOACCEL, info->NoAccel)) {
        if (RADEONAccelInit(pScreen)) {
            xf86DrvMsg(scrnIndex, X_INFO, "Acceleration enabled\n");
            info->accelOn = TRUE;

[Daniel Stone (daniels)]

argh! good catch, thanks. i'll be looking to push this one into breezy-updates.

[Daniel Stone (daniels)]

*** Bug 24315 has been marked as a duplicate of this bug. ***

[Adolfo González Blázquez (infinito)]

Will this hit breezy-updates recently?

[Daniel Stone (daniels)]

always been fixed in breezy

[Daniel Stone (daniels)]

i mean 'in dapper', christ

[Daniel Stone (daniels)]

*** Bug 27545 has been marked as a duplicate of this bug. ***

[Adolfo González Blázquez (infinito)]

Does anyone know why the fixed package hasn't hitted breezy-updates?

It's very strange to have updates like ubuntu-docs, and not xserver-xorg-driver-ati, which i think is _very_ important for users (even more than ubuntu-docs)

Hope it will get uploaded any day before dapper is released...