Bug 28422

Created attachment 36099 [details]
/proc/config

Created attachment 36100 [details]
dmesg

Created attachment 36101 [details]
lspci

lspci on the day of test. Not my actual one anymore.

Created attachment 36102 [details]
xorg.conf

Created attachment 36103 [details]
Xorg.0.log

I always have/want/need XEN activated and working. For daily use, I have SurroundView off.

Last month, I used to have things working whith vga=791, and FB working in wonsole, and x11-drivers/xf86-video-ati-6.12.6 .

After update to x11-drivers/xf86-video-ati-6.13.0 , I had to disable FB at boot prompt (regression bug ? no time to chat about minor problems that can be fixed by disabling FB when 99% things I attemped ended up in freese or KP).

The only thing that works ATM, is when I use only two monitors on the same card; ideally the MB integrated, or first slot PCI; getting both monitors on second PCI card (and ONLY this second card) is way harder.

Only Windows and Debian Stable could use monitors from two different cards. And, I can not get it in Debian anymore since I activated  hypervision (maybe a coincidence; I have 8 different Linux in the box, and forgot the detail of what I did).

Side note: it's easier to get logs when mouting the / in sync mode.

X -configure usually end up in segfault or freese.

You need to use KMS to use multiple cards.

(In reply to comment #7)
> You need to use KMS to use multiple cards.

Up yo you to look in my config if the required KMS is in or not. KMS have been introduced in 2009. How could my previous computer use 4 cards without KMS since 2002 ?

(In reply to comment #8)
> (In reply to comment #7)
> > You need to use KMS to use multiple cards.
> 
> Up yo you to look in my config if the required KMS is in or not. KMS have been
> introduced in 2009. How could my previous computer use 4 cards without KMS
> since 2002 ?

Support for multi-card was broken in the xserver after 1.4.x due to the pci rework.  If you want to use multi-card on newer systems, you need kms.

And during 2 years, nobody could use multicard system ...

You tell me if KMS was set properly ! because, my config clearly have some KMS Set. Still, i get many problems.

Or you can keep telling me I need KMS.

I would say, that you don't have modesetting enabled:
(II) [KMS] drm report modesetting isn't supported.
(II) [KMS] drm report modesetting isn't supported.

Don't know if it's not enabled in your kernel or in your libdrm. Also don't load any frame buffer driver and remove "vga= " from your boot loader options if you use these. If you want to be able to switch to the console make sure you load "fbcon".

As I thought, reporting such a vaste bug is useless; if maintainers don't even read attachements, it's useless to attach anything. It's obvisouly too difficult for you to read a kernel conf, even when attached. Even when I point all relevant points, you don't care.

People from IRC insisted I should report.

Now, I am sure this will never be fixed. Two years X is broken. Either have to revert to X 1.4, or forget about multicard.

I am *DEEPLY* *VERY* sad Linux systems can have such regression bugs. Bugs unfixed for over a year never get fixed. I give up. And since Xorg is a common project with BSD, I can not switch to BSD. Since no more distro distribute X 1.4, I have to go back to Windows.

No Linux dev will ever have a large enough view to understand the incomings of multicard + xen + MB specific features (I have ATI, but I know Nvidia has a different way to bridge their cards).

Linux used to be a good system. Xorg used to be a good server. I give up both.

(In reply to comment #12)
> As I thought, reporting such a vaste bug is useless; if maintainers don't even
> read attachements, it's useless to attach anything. It's obvisouly too
> difficult for you to read a kernel conf, even when attached. Even when I point
> all relevant points, you don't care.
> 
> People from IRC insisted I should report.
> 
> Now, I am sure this will never be fixed. Two years X is broken. Either have to
> revert to X 1.4, or forget about multicard.

Myself and others have used multi-card with kms just fine.

> 
> I am *DEEPLY* *VERY* sad Linux systems can have such regression bugs. Bugs
> unfixed for over a year never get fixed. I give up. And since Xorg is a common
> project with BSD, I can not switch to BSD. Since no more distro distribute X
> 1.4, I have to go back to Windows.
> 
> No Linux dev will ever have a large enough view to understand the incomings of
> multicard + xen + MB specific features (I have ATI, but I know Nvidia has a
> different way to bridge their cards).
> 
> Linux used to be a good system. Xorg used to be a good server. I give up both.

Your config is busted and we are trying to help you.  You need to participate if you want help.  From your dmesg, you have several issues:

[drm] Initialized drm 1.1.0 20060810
[drm] VGACON disable radeon kernel modesetting.
pci 0000:01:00.0: PCI INT A -> GSI 18 (level, low) -> IRQ 18
pci 0000:01:00.0: setting latency timer to 64
[drm] Initialized radeon 1.31.0 20080528 for 0000:01:00.0 on minor 0
pci 0000:02:00.0: enabling device (0000 -> 0003)
pci 0000:02:00.0: PCI INT A -> GSI 16 (level, low) -> IRQ 16
pci 0000:02:00.0: setting latency timer to 64
[drm] Initialized radeon 1.31.0 20080528 for 0000:02:00.0 on minor 1
fglrx: disagrees about version of symbol module_layout
fglrx: disagrees about version of symbol module_layout

First, you seem to have a fglrx installed.  You need to remove that to use the open drivers, secondly, as Martin pointed out, vgacon is preventing radeon from loading, so remove any vga= lines from your kernel command line if you have them.

1. I'm a normal user like you and I'm not experienced in tracking down bugs
2. I understand that you're pissed but try to calm down, it just doesn't help anybody

Your boot option seems to be: vga=794 video__=1280x1024 vga_=normal nomodeset_
Try removing those entries.

From the kernel name it reads like you're using gentoo: http://en.gentoo-wiki.com/wiki/Radeon

I would also suggest that you try the 2.6.33 or 2.6.34. A lot of work happened in the Ati driver section

1: fglrx being installed can cause a problem ? what leads me to this simple question: since kernel could not load it, how can this driver be a problem anyway ?

2: even with an underscore, nomodeset_ disables any FB querry set by VGA= or VIDEO=. As long as this word is in the line, even with one or more underscors, nomodeset disables FB. So, I think there is a bug in the kernel, or the driver responsable to interpret this switch. If i remove nomdeset vga= and video= , kernel will automatically set FB anyway; and if not kernel, Gentoo console scripts will (and I did not find how to disable them). So, anyway, you need to tell me what to pass to kernel to prevent kernel and Gentoo scripts doing stupid things.

> Try removing those entries.

if kernel don't switch to FB, Gentoo scrpts will.

IRC people could override this manually with nomodeset or radeon.modeset=1 . When using radeon.modeset, things were better.

> From the kernel name it reads like you're using gentoo

I said I am using Gentoo in the initial report ... 

> I would also suggest that you try the 2.6.33 or 2.6.34. A lot of work happened
in the Ati driver section

it's 3 months I am waiting for maintainer to push them in portage. http://bugs.gentoo.org/show_bug.cgi?id=307123 if you like URLs.

> http://en.gentoo-wiki.com/wiki/Radeon

page is useless because xen kernels don't apply the same set of patches.

And I forgot to mention that I can have freese, segfault, or panic even when trying to use only one monitor (no xrandr) on only one card (simple layout).

Now coming today's test: as you asked me ... 

Result is: garbage on both monitors (head #0 of cards #2 and #3). Numlock was non responsive. no KP, i started a "sleep 30 && reboot" in the background.

Created attachment 36109 [details]
z01 demsg

Created attachment 36110 [details]
z01 lspci

Created attachment 36111 [details]
z01 Xorg.0.log

Created attachment 36112 [details]
z01 xorg.conf

Hm,
can you try to start the xserver without the xorg.conf?

My goal is that you have a simple running xserver with KMS without bothering with any multi monitor setup. 2.6.32 should be working too, but the driver got a lot better with later kernel releases. Too bad that Xen seems to support only 2.6.31 and 2.6.32. I guess you depend on Xen?

XAA isn't supported anymore: Option         "AccelMethod" "XAA"
I guess this option is ignored by the driver, but I would delete/comment out that line (or just start the xserver without xorg.conf).

This one caught my attention: RADEON(0): Setting screen physical size to 338 x 270
I think this is wrong.

> can you try to start the xserver without the xorg.conf?

yes, and it gives a standard segfault. I'll try to get a log.

> 2.6.32 should be working too, but the driver got a lot better with later kernel releases

in 2009, 2.6.32 was not even a draft on Linus desktop.

Oh, yes, I could also use a verbatim kernel, disactivate hypervisor, disactivate AMD64 flag, boot my old good P3 800 EB, and extract an archive of known to work X 1.4 ... Unfortunately, we are in 2010.

> I think this is wrong.

It may happen when DDC does not report resolutions properly; that's why I push virtual screen to a bug value, and do fine tuning with xrandr. DDC may fail for thouthands reasons; and really, I don't mind the detail.

Which output do you want ?
- X
- Xorg
- startx
- X -configure ?

Each X attempt costs me 2 reboot. Each X failure costs me an extra reset plus an extra reboot.

console output or just X log ? i am not sure details of segfault are reported in X logs :)

I can reproduce easily on the z01 configuration. It produces an error different from the previous one, but I can reproduce it on demand for now ... until I am forced to do major changes (or just update the distro).

I don't use neither Xen nor KVM, but here's what I think: If I've understood it correctly they're both designed to do the same thing (they're both hypervisors for virtual machines, correct me if I'm wrong). So if you're able to use KVM it would have the following advantages:
  -the kernel component is included in the kernel, so no need to patch the kernel (Xen will most likely never be integrated into the kernel)
  -you can use a newer kernel ;-)
I'm using Archlinux and I found the following package: http://aur.archlinux.org/packages.php?ID=29023
The interesting part is, that the Xen patches are taken from gentoo. If you want to give it a try and build the kernel remember to install the microcode for your cards: http://gentoo-portage.com/x11-drivers/radeon-ucode

There are some other things you could try out:
You say that you got a segfault. I think this could happen when the Xorg stack wasn't compiled in the correct order:
 1. kernel
 (2. protos)
 3. xorg-server
 4. libdrm
 5. xf86-video-ati
 6. mesa
You can try to compile 3, 4, 5 and then 6.

If that doesn't help you can try to remove all cards but one and see if you still get the error. If you still get the segfault with only one card (and one monitor) installed then I would guess that there is a problem that the installed Xorg packages don't play nicely together.

Trouble-shooting is unfortunately often a time hog.

Good luck!

Yes and no. Xen is like vmware: it can do virtualisation using at least 4 different ways. KVM is one of them. Somehow, KVM is a subset of XEN. To make it VERY short. Other point of view: XEN symbols are for the host dom0, while KVM symbols are for hosted domUs (XEN people would byte me for this). And KVM has limitations: it can only load Linux domUs; I need HVM for non Linux domUs.

> remember to install the microcode for your cards: http://gentoo-portage.com/x11-drivers/radeon-ucode

if you had read this bug report, and the page you advise me, you would have notice that the uCodes are only used for R6xxx and R7xxx, while I am using R4350; so, either you did not read the web page, or my initial report, or know nothing about Radeon chipset (this difference is explained in all forums dealing with hardware acceleration on ATI: firmware update are only available for R6* and R7* cards).

Please, stop advising me things that are "obviously" useless.

And for the part I have not explicitely stated yet: i am not sure what is exactly the chipset integrated in the MB (card at adress 1:5:0), but it's in the R4xxx serie, a model slightly weaker than r4350. Probably an 42xx from memory. R4xxx for sure. I took similar models so that SurroundView can work with three chipsets together, and, as much as possible, the three chipsets should use the same driver (but it was not the cas under Windows: the driver CD for MB could not manage the daughter cards).

By the way, I forgot to give you the z01 kernel config.

Before I do more tests, if you want, tell me *EXACTLY* what you want. Every single detail is important, because any single detail can make the difference between a segfault, a freese, and a KP.

> You say that you got a segfault. I think this could happen when the Xorg stack
wasn't compiled in the correct order

Things were compiled by gentoo; do you want me to rebuild everything ? in a specific order ? so, converted in Gentoo langage:

eix proto | grep x11-pro | grep -v xcalibrateproto | cut -d " " -f2 | xargs emerge -v1 && emerge -v1 xorg-server x11-base/xorg-x11 && emerge -v1 x11-libs/libdrm && emerge -v1 xf86-video-ati && emerge -v1 media-libs/mesa x11-apps/mesa-progs

> If that doesn't help you can try to remove all cards but one and see if you
still get the error.

Useless. Even when all are inserted, I easily can make any of them work, and most of the time, get the two outputs work. That's the case of actual z01 config: right now, I am typing with two monitors on card #2, except that I just commented screen_3 ... and as I uncomment, i get garbage and freese.

(In reply to comment #24)
> Yes and no. Xen is like vmware: it can do virtualisation using at least 4
> different ways. KVM is one of them. Somehow, KVM is a subset of XEN. To make it
> VERY short. Other point of view: XEN symbols are for the host dom0, while KVM
> symbols are for hosted domUs (XEN people would byte me for this). And KVM has
> limitations: it can only load Linux domUs; I need HVM for non Linux domUs.
> 
> > remember to install the microcode for your cards: http://gentoo-portage.com/x11-drivers/radeon-ucode
> 
> if you had read this bug report, and the page you advise me, you would have
> notice that the uCodes are only used for R6xxx and R7xxx, while I am using
> R4350; so, either you did not read the web page, or my initial report, or know
> nothing about Radeon chipset (this difference is explained in all forums
> dealing with hardware acceleration on ATI: firmware update are only available
> for R6* and R7* cards).
> 
> Please, stop advising me things that are "obviously" useless.

Your cards ARE r6xx and r7xx based:
01:00.0 VGA compatible controller: ATI Technologies Inc RV710 [Radeon HD 4350]
02:00.0 VGA compatible controller: ATI Technologies Inc RV710 [Radeon HD 4350]
01:05.0 VGA compatible controller: ATI Technologies Inc Device 9710
The RV710 cards are R7xx cards and and the integrated card is R6xx based.  So the firmware is not useless and is in fact required for your cards.  However, it appears that your kernel is too old to use the new firmware files which are required for the interrupt controller.

If you notice, the integrated chip does does not show up in your original lspci.  You need to enable surround view in order to use the integrated card.

I would suggest trying a newer kernel without external xen patches.  That should help narrow down the problem.

Created attachment 36119 [details]
z01 kernel config

Created attachment 36120 [details]
startx >log 2>&1 - when no xorg.conf

Created attachment 36121 [details]
Xorg.0.log for previous startx when no xorg.conf

When starting X in z01 configuration (removed all vga= video= and nomodeset) without xorg.conf, I get garbage on screen, i can play with numlock button, and mouse device moves a square of garbage in front of funny screen.

To get a usable X, I have to use ... identical kernel and bios settings, just add vga=794 video=1280x1024 vgaa_=normal nomodeset ... with z01 xorg.conf (and commenting the second screen, called screen_3 IIRC).

Note that without nomodeset, console enters FB mode, of course.

At the moment, I booted with
vga=794 video=1280x1024 vgaa_=normal nomodeset___
with no xorg.conf (because I actually forgot to restaure it), and startx worked sucessfully. I do not have FB. xrandr failed to switch on the second head (I may play with xrandr when I finished this report). In short, i got one monitor working properly.

On Mon, Jun 07, 2010 at 03:37:01PM -0700, bugzilla-daemon@freedesktop.org wrote:
> if you had read this bug report, and the page you advise me, you would have
> notice that the uCodes are only used for R6xxx and R7xxx, while I am using
> R4350; so, either you did not read the web page, or my initial report, or know
> nothing about Radeon chipset (this difference is explained in all forums
> dealing with hardware acceleration on ATI: firmware update are only available
> for R6* and R7* cards).
> 
> Please, stop advising me things that are "obviously" useless.

A Radeon 4350 is an r7xx - the former refers to card names, where as the
latter refers to chip generations.

Please, stop being so fucking obnoxious for no reason.  These people are
helping you for free.  And, if you're going to be incredibly rude at
great length, at least have the decency to not be completely wrong.

Hope that helps.

(In reply to comment #25)
> If you notice, the integrated chip does does not show up in your original
> lspci.

Because my BIOS settings were different. For the original report, I did, to the letter, what an ATI dev/maintainer told me to do, to the letter. And he asked me to disable the integrated chipset for my initial report.

> You need to enable surround view in order to use the integrated card.

Wrong. Windows can use integrated card with *AND/OR* without SurroundView. I spend 2h especially for this test, removing all y disks, and installing Windows on the mail system (instead of HVM), just to make this test. In less than 2h, I could install Windows, get 6 monitors work, and check that they work with AND without SourroundView.

I think that ATM it is disabled, because, from memory, it's easier to get X work without SRV for me; i ll check on next reboot.

> I would suggest trying a newer kernel without external xen patches.  That
> should help narrow down the problem.

It would work. I don't know how, but stable Debian without Xen support at all could let X use at least two cards. It's not possible to do this test anymore, because I have not been able to make it work a second time (debian stable is available with too many tastes and different kernels, I could not find agaun the working combination), and because many other factors impact.

I can not spend hours testing an exponential number of combinations of factors. Over 40h just on X since january is already very much. Plus 3 tonight. Plus 2 for Windows (to check that the problem was NOT hardware).

Without XEN, it can work with three cards. With XEN, it can work with one card.

***

Things have been recompiled as said.

ucode is now installed. What's next ?

(In reply to comment #31)
> Without XEN, it can work with three cards. With XEN, it can work with one card.

So without the XEN patches, it all works?  That would point to a problem with XEN then.

It works with a non xen 2.6.28 and X 1.4, yes. And preferably without SourroundView.

ATM, surroundview is ENABLED, and it was also on for z01 tests.

As I installed ucode and rebooted, I attach logs of the working configuration.

Created attachment 36122 [details]
z02 dmesg

Created attachment 36123 [details]
z02 xorg.conf

Created attachment 36124 [details]
z02 Xorg.0.log

The xorg.conf is almost the same as for z01, except I commented second screen.

kernel config is the same as for z01.

With this, after launching xrandr, I get two monitors on one card.

In the console, I don't have FB working.

I have no idea what you're doing. I guess you're going to tell me that I didn't read you log files completely (and you would be right), or haven't read all the postings of this bug report (and you would be right again).

But here's what I read so far:
"Linux version 2.6.32-xen-r1-Gentoo-uranus-1-15"
It still seems to be kernel 2.6.32 with xen patches. I have no idea how the kernel build process works in Gentoo and if you just left the name as it was before but compiled the kernel without the xen patches.

According to the Gentoo wiki you only need the firmware for kernel >=2.6.33 (that was in the link I posted earlier)

I thought we were trying to get KMS working: "[drm] VGACON disable radeon kernel modesetting."

You really make sure that modesetting isn't used?:
"vga=794 video=1280x1024 vgaa_=normal nomodeset"
Why don't you just copy that line, then comment out the first line and do the changes in the second line? Have you searched the Gentoo wiki on how to enable modesetting? If Gentoo wants to load framebuffer driver, try to search the wiki on how to disable/blacklist drivers.

I have compiled these two driver into my initial ramdisk: radeon, fbcon
This is distribution specific how you can do this.

I'm using Archlinux, so I'm making a little advertisement here:
 -it supports KMS out of the box
 -there is a package in AUR (Arch user repository) with a 2.6.34 kernel with xen
No promises that it will work though. This decision is up to you if you want to spent more time on trying to get Gentoo with KMS to work, or if you want to spent time trying out another distro without you knowing if it will be worth the trouble...

> kernel without the xen patches

The initial report was done with a non XEN kernel: compile uranus-1-16.

The z01/z02 configurations are using uranus-1-15 which have XEN enabled with HVM support for dom0.

I really don't understand what you talk about: compile 15 has all possible XEN features activated. If you have questions about how I compile my kernel, what's wrong with kernel config files I provided ?

compile 16 have been done for IRC people who wanted me to try to repro on XEN disabled configuration. The initial report complains about radeon driver not being able to get to work, even with a non xen kernel (there, you are right, I was using the standard gentoo xen-source package with XEN disabled; but this is specific to build 16).

In short, i can get two monitors on same card work with and without xen, and I get bugs even without xen when I try to activate a second card ...

But, really, that day, I did 80 reboots in 12h, so, the exact detail of what I did each time faded away.

> According to the Gentoo wiki you only need the firmware for kernel >=2.6.33
(that was in the link I posted earlier)

When xen-source-2.6.33 is available. http://bugs.gentoo.org/show_bug.cgi?id=307123 . 3 months I am waiting.

> I thought we were trying to get KMS working

tell me what to do.

> You really make sure that modesetting isn't used?

I gave dmesg for all 3 problems. 

> Have you searched the Gentoo wiki on how to enable modesetting?

#gentoo told me to use nomodeset
#radeon told me to use radeon.modeset=1

YOU tell me what you want, and I create a new grub entry for you.

> Have you searched the Gentoo wiki on how to enable modesetting?

you know, on the same official wiki, I can find sometimes 2 pages saying contradictory statements; sometimes even in the same page. I am having lots of way to produce many different bugs, and they vary with very fine /etc adjustements. I have tried every things wikis and forums say. Now, I report a bug. I am having a probme with X since over 16 months; http://bugs.gentoo.org/show_bug.cgi?id=267080
 clearly told me three times that this bug can not be fixed at Gentoo level. Now, it's Xorg team turn to try t fix it.

> try to search the wiki on how to disable/blacklist drivers.

It's too late. and, nomodeset/radeon.modeset=1 can do it at kernel level. And, if some symbol bores you in my conf, just ask me to rebuild after removing some driver from kernel ... and I would create a kernel just for you with a second grub entry with your name (and the kernel name will be uranus-1-17 ).

> This is distribution specific how you can do this.

z02 logs were generated after installing the firmare and rebooting; if the firmwre could make any diference, you should see it in logs.

I have also 3 debian testing (as dom0), but I fear they may not help at all.

> it supports KMS out of the box

both kconfig I provided show some KMS symbol; what's wrong with them ?

> there is a package in AUR (Arch user repository) with a 2.6.34 kernel with
xen

so you want me to install Arch just to test X ? if I find a free partition ... (i have already installed Windows just to proove Xorg bugs; why not a 7th Linux) Tell me by advance:
- kernel exact version to use
- boot options to alter
- xorg.conf to use

If I can remember correctly the vga= option in grub will load a framebuffer driver which conflicts with KMS. That's the reason you have to remove this option.

The nomodeset option disables kms and radeon.modeset=1 enables it. For modesetting to work you have to remove the nomodeset option. I think you have to add the radeon.modeset=1 option with the kernel you're using.

You can try to add "radeon" and "fbcon" to /etc/modules.autoload.d/kernel-2.6 it they're not included there.

If this still doesn't enable KMS you can try the following (I think this doesn't really belong to a bug report about Xorg but anyways...):
-Unmask the 2.6.34 vanilla kernel sources and emerge them as described here: http://www.gentoo.org/doc/en/handbook/handbook-x86.xml?part=1&chap=7.
-get "microcode_core-xen.patch" and "http://gentoo-xen-kernel.googlecode.com/files/xen-patches-2.6.34-1.tar.bz2" and "config.xen" or "config.xen.x86_64" from here: http://aur.archlinux.org/packages.php?ID=29023
-apply the patches to your source (which should be in /usr/src/linux)
-copy the corresponding config file (64 bit: config.xen.x86_64) to /usr/src/linux/.config
-configure your kernel with "make xconfig" as described in the Gentoo documentation. Make sure you have "Enable modesetting on radeon by default" enabled in "Direct Rendering Manager...". In "general" change "Local version - append to kernel release" to a string you like (e.g. "-xen-kms")
-install the ucode firmware files

Install the kernel as described in the documentation and create a new grub entry.

Cross your fingers and boot into the new kernel. If there are problems using the new kernel (eg. Firmware are missing) I have to refer you to the Gentoo people.

For later reference: You can blacklist drivers in  /etc/modprobe.d/blacklist 

If you really want to try Archlinux you can use the packages in the repositories. You shouldn't have to configure anything (There is a nice install guide: http://wiki.archlinux.org/index.php/Official_Arch_Linux_Install_Guide and this one http://wiki.archlinux.org/index.php/Beginners'_Guide). When you want to have a kernel with Xen support you have to use the package in AUR which I mentioned earlier. Have a look at http://wiki.archlinux.org/index.php/AUR#Installing_Packages_from_the_AUR.

> If this still doesn't enable KMS you can try the following

I have absolutely no clue why is KMS required, and how to check if it's loaded. I installed sys-kernel/xen-sources-2.6.34 (from the URL given in quoted bug): it installs 72 patches ... the first of which is a fbcon patch ! 

You tell me to build a kernel, but not what I should do with it. I know a dozen of different ways to start Xorg ... In this bug, I do what you ask, not what you think. In comment 40, you do not ask me anything.

(In reply to comment #41)
> > If this still doesn't enable KMS you can try the following
> 
> I have absolutely no clue why is KMS required, and how to check if it's loaded.
> I installed sys-kernel/xen-sources-2.6.34 (from the URL given in quoted bug):
> it installs 72 patches ... the first of which is a fbcon patch ! 
> 
To quote Alex: "Support for multi-card was broken in the xserver after 1.4.x due to the pci rework.  If you want to use multi-card on newer systems, you need kms." So I don't understand why you wrote that you have no clue why KMS is required.

> You tell me to build a kernel, but not what I should do with it. I know a dozen
> of different ways to start Xorg ... In this bug, I do what you ask, not what
> you think. In comment 40, you do not ask me anything.

So if I understood you correctly you now have a 2.6.34 kernel with Xen patches installed?
Check the following: Go to the kernel source directory and start "make xconfig". Make sure that you have set "Enable modesetting on radeon by default" in "Direct Rendering Manager". "ATI Radeon (DRM Radeon)" should be compiled as a module.

Did you add "radeon" and "fbcon" to your autoload config?
What is your grub entry for the new kernel?
Have you installed the ucode firmware?
Does the new kernel boot correctly?
Start X without a xorg.conf.
Output of lsmod could probably also be useful.

If KMS is enabled you should have the following line in your Xorg.0.log: [KMS] Kernel modesetting enabled.

After installing Archlinux, I could start X at once:
- X -configure crashed, but generated a usable conf
- using radeon driver
- the conf generated was fine and working: each card using two monitors in miror, all 3 cards wotking together ( :0.0 :0.1 :0.2 ). Xrandr could let me isolate each monitor (two per display). Only mouse could move from display to display. HW acceleration working for 2D and 3D. No crash, no freese
- after a bit of playing, i could switch to xinerama: all 6 monitors in the same DISPLAY. I still had partial 3D accell with bugs
- after forcing software 3D acceleration, i lost HW accell, and bugs disapeared.

But, this was on a non Xen kernel. So, tomorrow, I will try to install a XEN kernel in Archlinux.

Martin: this distro is very pleasant to use; but, it seems very heavy to maintain.

Attaching all usefull logs.

Created attachment 36441 [details]
arch 01 xorg.conf auto generated works fine with basic non xen kernel without xen

Created attachment 36442 [details]
Arch 01 Xorg.0.log

Created attachment 36443 [details]
Arch 01 lspci

Created attachment 36444 [details]
Arch 01 dmesg

Created attachment 36445 [details]
Arch 01 kernel config

Created attachment 36446 [details]
Arch 02 xorg.conf - just adding xinerama and LIBGL_ALWAYS_SOFTWARE=1 keeping previous kernel dmesg and pci

Created attachment 36447 [details]
Arch 02 Xorg.log

That's all for now. Tomorow, I try a Xen kernel.

I start Xinerama conf with:
LIBGL_ALWAYS_SOFTWARE=1 startx

Just one minor bug, likely to be off topic: on the rotated screen, the image if fine at the begining, but after the first time I move the mouse on this monitor, the image slides on the side, like when the physical screen is larger than the virtual one (maybe I can fix this in the conf, did not dig yet). It slides only to the left (left, when the physical LCD is physically vertical, portrait orientation, with the buttons that are usually down are put on the left side), and upside, but, can come down to the right place, and, never left again.

z02 xorg.conf works as is on Gentoo with and without SurroundView . If I want to use Screen1 instead of Screen2 (dual head), I get a freese with SRV, and it works without SRV.

I have compiled a Gentoo kernel from package gentoo-source (without xen), using as base conf, my xen conf z01 kernel config . Then, z02 xorg.conf works like on Xen kernel. When using Arch worg.conf, in both case, I get a segfault when doing just startx; if i set LIBGL_ALWAYS_SOFTWARE=1 then i don't get segfault: all 6 monitors startup (and 3 indicate signal out of sync), and numlock refuses to blink.

I have tried to compile gentoo-source from Arch 01 kernel config but it does not boot for now.

So, i just have more and more cases where X crashes without explicit message, send out of sync signal to monitors (what X >1.5 claim to be impossible since >1.5 tend to use DDC; in practice, it does not *really* do it), freese, or segfault.

And, you still did not tell me which kind of tests you want.

In fact, none of my previous bugs have been fixed. I just thing this Bugzilla is for prestige, and that you never try to fix issues users report. As usual, I am on my own, and, many people tell me to do many stupid things, but in fact, nobody really care about me, and, nobody really know what they are asking me to do.

If you did, you would have tell me: put these symboles in your kernel, and remove those ones. Make sure you have this set up, and this NOT set up. That's things get sorted on other BTS.

For example, you are asking me to activate SRV, while SRV is the feature that breaks my box ATM. Like my other bugs, in 5y, it will still be open.

But, this remains a regression bug compared to X 1.4 .

On Wed, Jun 30, 2010 at 05:25:47AM -0700, bugzilla-daemon@freedesktop.org wrote:
> In fact, none of my previous bugs have been fixed. I just thing this Bugzilla
> is for prestige, and that you never try to fix issues users report. As usual, I
> am on my own, and, many people tell me to do many stupid things, but in fact,
> nobody really care about me, and, nobody really know what they are asking me to
> do.

Are you this much of an asshole to everyone you encounter? Your mother?

No; 80% of bugs I reported in Debian and Gentoo have been fixed withint a few months. I honor them.

X 1.5 dropped the multicard support; we had to wait over a year before having it back. <<break first, and see what happen later when the users complains>> is a mentality i will *NEVER* make mine.

Especially, most maintainers I have met since that time all said me very stupid things, such as "oh, Matrox stuff is old and crap"; but the bug was the same for Nvidia; and, the same again with ATI cards.

- <<you must have SRV>>, but it breaks my box
- << you *must* use nomodeset>>, but Arch and Xinerama can work even when FB is on
- <<update to X 1.7 and will work like a charm>>; 1.8 still bugs
- <<you must use this FB Driver>>, changes nothing
- <<all your issues are because you have fglrx driver installed, remove it and it will work>>; removed it, still not working ...
- <<try this distro>> (3 times) ... Archlinux was fun to install, but, building their Xen kernel is not a peace of joke ...
- << build any non Xen kernel, it will work straight>> ... build 3, no result.

And, when i ask: which symbole do you want me to add or remove from my kernel ? no answer.

I have been given dozens of peaces of advice; no-one really knew what they were speaking about.

I started Archlinux with the gentoo XEN kernel ( z01 kernel config ) (because I was too lame to try to understand how to compile a kernel under this distro), and I could run the Xinerama configuration (Arch 02 xorg.conf ... and I even forgot to pass LIBGL_ALWAYS_SOFTWARE=1). It was working like a charm. 6 monitors exactly as I want. Straight away.

I also tried the accelerated configuration (Arch 01 xorg.conf), and, i was not surprised to see it did not work (missing or uncompatible low level DRI/DRM / i don't want to know what ... drivers).

So, the problem is NOT the kernel, kernel version, or kernel configuration. 2.6.34-xen-Gentoo-uranus-1-19 (or maybe it was 1-15 last time) is perfectly suitable for multiple card use.

Arch is using X.Org X Server 1.8.1.902 (1.8.2 RC 2)
My Gentoo is actually using 1.8.1.902 (1.8.2 RC 2)

The problem is not:
- kernel
- xorg.conf (last night, I was using the working conf from Arch)

Very few things left could differ.

I also tried to use Arch 02 xorg.conf on kernel 2.6.34-gentoo-r1-Gentoo-uranus-1-20 (which is not XEN capable; thus, xend failed to start), and it also freesed. So, the issue is not xend being started.

Few things left:
- libc
- X drivers
- (x)randr and mesa versions ...

and X still does not say why it's crashing.

I started Gentoo-xen kernel on Archlinux partition, and chrooted Gentoo to startx. The xinerama conf (Arch 02 xorg.conf) worked. => the X binary build by Gentoo is fine, and can run on Gentoo kernel.

So, the problem is between the kernel and X.

I starter gentoo in rescue mode: init=/bin/bash

I am using this init script:
loadkeys fr
mount -o remount,rw,sync /dev/md5 /
mount -t proc proc /proc/
rm /lib64/rc/init.d/started/* # required for Gentoo reasons
/etc/init.d/udev start

Udev start starts the following depencies:
fsck  hwclock  localmount  lvm  modules  mtab  net.eth0  root  sysfs  udev  udev-mount

Here is the console output:
 * /dev is already mounted
 * Starting udevd ... [ !! ]
 * Populating /dev with existing devices through uevents ... [ ok ]
 * Waiting for uevents to be processed ... [ ok ]
 * Device initiated services: fsck hwclock localmount lvm modules mtab net.eth0 root

I need to start udev, to fill /sys, or X complains it does not find any hardware (even when lspci says cards are in).

Aboce this, I tried 3 xorg.conf:
- z02 works fine with everything (minimal WM working inside; twm i think)
- Arch 01 (xrandr accelerated) will not show WM, but, started with X -retro, I can get the grey grid; mouse will not move - no error in the logs; just saying it's disabling DRI, but this does not look fatal.
- Arch 02 (xinerama, tried with and without LIBGL definition),
also with X -retro, will not show the grey grid.

So, being in rescue mode does not prevent X working, but, loads enough services to make the multicard conf freese. I don't know how to reduce the number of services. I really reached bleeding edge tests.

Created attachment 36829 [details]
z03 xorg.conf

My setup now works better.

You will see boot options in the X.log later.

It works with AND without Suroundview. For bot xinerama and xrandr.

This xinerama conf provides what I need (6 monitors unified in the same "server"); it works perfectly when i comment the rotate line.

The bug still stand: when X crash, freese, or make KP, it does not say anything, and, very often the log file is not even created (I have to mount / with sync to get anything). Recent X ( since 1.5) do not report enough, do not log enough. And when it crashes, it does not say always why.

Sometimes, it even does not take care about xorg.conf, and, does not either even way it, or ignores sections (see the many Gentoo bugs I created last year).

The minor bug in this is: when i use the rotate feature, and move the mouse down (bottom of VISIBLE space) the rotated monitor, the visible spaces moves right (what means, up when you think the absolute hardware coordinates). After it have moved right, it will move up when pushing the cursor left; and down again when moving cursor right. It will not move up/down when moving cursor right/left of this monitor BEFORE going down. After moving down, the left black space is lost forever. I think it's a virtual screen problem, but ... and this is AGAIN the point of this bug ... nothing is mentioned about this in the logs: log don't mention the monitor is rotated (except in the header where it lists the conf commands), and, it does not mention anything about screens or monitor. So, again, no way to track the issue. Because of lack of logs. it will be the z05 log.

Created attachment 36830 [details]
z03 dmesg

Created attachment 36831 [details]
z03 /var/log/Xorg.0.log

Created attachment 36832 [details]
z03 kernel config 2.6.34-xen-Gentoo-uranus-1-31

This config is very close from 1-15 . There should not be significant changes related to X. But in case i am wrong ... 

This 1-31 conf is common with the soon coming z04 and z05 logs.

Now, you have:
- deeply broken conf, the guy from IRC asked me to report
- several intermediate steps with different breakage level
- fully working set-up

Everything you need to compare logs, and track the bugs :)

Side note: I don't use LIBGL_ALWAYS_SOFTWARE=1 . Without it, some accelerated applications are not always shown properly, but when I use it, most applications are slower when scrolling (long web pages in FF, long emails in TB, or when moving large windows around). I know it's a limitation of Xinerama, and have to accept it.

Created attachment 36833 [details]
z05 xorg.conf

rotated monitor shows black strips, and moves as if virtual screen was larger than physical.

Created attachment 36834 [details]
z05 dmesg

Created attachment 36835 [details]
z05 Xorg.0.log

Created attachment 36836 [details]
z04 xorg.conf

Created attachment 36837 [details]
z04 dmesg

Created attachment 36838 [details]
z04 xorg.log

(In reply to comment #42)
> > You tell me to build a kernel, but not what I should do with it. I know a dozen
> > of different ways to start Xorg ... In this bug, I do what you ask, not what
> > you think. In comment 40, you do not ask me anything.
> 
> So if I understood you correctly you now have a 2.6.34 kernel with Xen patches
> installed?
> Check the following: Go to the kernel source directory and start "make
> xconfig". Make sure that you have set "Enable modesetting on radeon by default"
> in "Direct Rendering Manager". "ATI Radeon (DRM Radeon)" should be compiled as
> a module.

Module names, and module place change every 6 months. Inside the 2.6 tree, I have seen several drivers change place 4 times. Do not "describe" with words. Be technical, and tell me the symbole name.

An other reason to use symbole names: if you knew it, you could grep directly in my kernel conf ... if I already have it or not.

An other reason to use symbole names: if I don't find it at the place you describe, I can search it. Then, I would either see the new place, or see the deps that make it not appear (last week I had to activate 8 deps before I could make my new wifi card driver appear in the menus).

> should be compiled as a module.

Why not hardlinked ?

> Did you add "radeon" and "fbcon" to your autoload config?

This is irrelevant. Because, even when modules are not declared in /etc, my distribution discovers them anyway. It may happen I list them in autoload, and it refuses to load them (and of course, does not say why, or life would be too easy). The "hotplug" stack in Linux distributions became too big for me, too many changes, too many services, I can do not master it whole anymore: hal halevt devfs (when i started, it just came out, and I still have to use it on very old machines, so, devfs is still in use on some of my box) udev (which changed shape 3 times in 10y), coldplug (now removed from Debian), hotplug (partly replaced by udev in some distros), sysfs (which has some bits in kernel, and also contributes loading modules and firmwares), X (which sometimes asks udev to handle some things like keyboard layout).

And sometimes, X developers fight on IRC because they do not agree between themselves when I ask in which section of xorg.conf I should declare some features (or wonder why some keywords work when I put them at the wrong place).

> What is your grub entry for the new kernel?

This kind of answer already is in Xlogs (search pattern "Kernel command line"). If you are an X dev, I wonder why you ask me such questions.

> Have you installed the ucode firmware?

Now, yes.

> Does the new kernel boot correctly?

Yes.

> Start X without a xorg.conf.

In Gentoo, it always leaded me to freese (or use only one card, and bugs even with only one). Archlinux could start X nicely, and generate a working conf; this conf now works in Gentoo.

No clue why. 

> Output of lsmod could probably also be useful.

I am not using a Debian with "all existing modules compiled in". I have a Gentoo, and build my kernel myself. Look at my kernel conf.

> If KMS is enabled you should have the following line in your Xorg.0.log: [KMS]
> Kernel modesetting enabled.

Look at the dmesg. I have always provided them.

I provide all the logs you could need, and until now, most things I am asked have answers in logs I had previously provided.

Now, let's talk about today. Since 5 days after have the 6 monitors conf working, I have random freeses. See the z03 configuration.

Some logs are given here: http://forums.gentoo.org/viewtopic-p-6361675.html

In short, at some time, I feel the mouse moving slowly, and the system getting loaded. Them I see this appear in system messages:

Jul 12 04:13:47 uranus kernel: [TTM] Error restricting pfn 32df0: -12
Jul 12 04:13:47 uranus kernel: [TTM] Error restricting pfn 33085: -12
Jul 12 04:13:47 uranus kernel: [TTM] Error restricting pfn 33084: -12
Jul 12 04:13:47 uranus kernel: [TTM] Error restricting pfn 33083: -12
Jul 12 04:13:47 uranus kernel: [TTM] Error restricting pfn 32def: -12
Jul 12 04:13:47 uranus kernel: PCI-DMA: Out of SW-IOMMU space for 4096 bytes at device 0000:01:05.0
Jul 12 04:13:47 uranus kernel: [drm:radeon_ttm_backend_bind] *ERROR* failed to bind 4 pages at 0x00A0E000
Jul 12 04:13:47 uranus kernel: [TTM] Couldn't bind backend.
Jul 12 04:13:47 uranus kernel: [TTM] Buffer eviction failed
Jul 12 04:13:47 uranus kernel: radeon 0000:01:05.0: object_init failed for (4001792, 0x00000004)
Jul 12 04:13:47 uranus kernel: [drm:radeon_gem_object_create] *ERROR* Failed to allocate GEM object (4001792, 4, 4096, -12)
Jul 12 04:13:47 uranus kernel: [TTM] Error restricting pfn 32def: -12 

A few minutes later, Thunderbird (ALWAYS displayed on card PCI:1:5-VGA) get's killed; then firefox (always on PCI:1:5-DVI) a few minutes later. Then, the machine freeses: cursor don't move, gkrellm stops moving, music stops.

All windows on card pci:1:5 get killed. Windows on other cards stay alive (until the freese).

Sometimes, I just get the freese, without any application crash or slowdown.

Just after starting X, I see this in syslog:
Jul 24 13:49:46 uranus kernel: vgaarb: device changed decodes: PCI:0000:03:00.0,olddecodes=io+mem,decodes=none:owns=none
Jul 24 13:49:46 uranus kernel: vgaarb: device changed decodes: PCI:0000:02:00.0,olddecodes=io+mem,decodes=none:owns=none
Jul 24 13:49:46 uranus kernel: vgaarb: device changed decodes: PCI:0000:01:05.0,olddecodes=io+mem,decodes=none:owns=io+mem
Jul 24 13:49:47 uranus kernel: [drm:drm_mode_getfb] *ERROR* invalid framebuffer id
Jul 24 13:49:47 uranus kernel: [drm:drm_mode_getfb] *ERROR* invalid framebuffer id
Jul 24 13:49:47 uranus kernel: [drm:drm_mode_getfb] *ERROR* invalid framebuffer id
Jul 24 13:49:47 uranus kernel: [drm:drm_mode_getfb] *ERROR* invalid framebuffer id

I wonder why DRM is loaded ... at the same time as I have Xinerama in my conf. LIBGL_ALWAYS_SOFTWARE=1 does not help.

I wonder where the issue could be: kernel, X, drm, dri, FB, ucode, X radeon driver ... 

The only thing I can tell is that ... PCI:1:5 is a video card integrated to motherboard, and it shares memory with the system (video ram is stolen in the CPU DDR).

Playing with Surround does not impact on bug. Happen in both cases.

The freese usually happens after 2 to 6 hours of desktop use. After a freese, and pressing the reset button (aka, not letting the machine switch compleetely down), the probability of a second/thid freese gets very high (usually 5 to 30mn).

I *never* had any freese before july, before updating to "latest everything" (kernel, drivers, X), before being able to use all 3 cards together.

I wonder if it could be a hardware issue (the motherboard not happy to use internal card with two external ones ? capacitor getting aged just *NOW* ? missplugged RAM ?). I have fans everywhere to cool everything.

Just before a freese occurs, once, I have seen garbage in firefox: some website using flash/JS/Java to display some menu, at some point, some objects of the page were replaced by "funny patterns". The external shape was apropriate, but the visible content of object was garbage (colorfull pixels). Only for 1s (after i click on a button, and until the page was reloaded); only some elements, not all of them. The machine frose 2mn later.

I have tried to rebuild a kernel without DRM (because TTM is a dep of DRM, so, removing DRM is required to remove TTM): X frose at start.

=> I can not use a kernel without TTM.
=> the issue does not have WA. It must be fixed.

Since i do not understand the lingo on most of this, i just want you to know, my LibreOffice is getting to the point i can not use it over 3 minutes before it crashes (wipes out and closes) and i lose any work i have done.

(In reply to comment #70)
> Since i do not understand the lingo on most of this, i just want you to know,
> my LibreOffice is getting to the point i can not use it over 3 minutes before
> it crashes (wipes out and closes) and i lose any work i have done.

Please, file a new bug so it doesn't interfere here. This link can help: https://www.libreoffice.org/get-help/bug/

Benoit-Pierre: very old issue here.
Do you still reproduce the bug with new LibreOffice version? (3.5.6 or 3.6.1)
(I think a brand new LO profile is important too)

Benoit-Pierre: sorry someone changed product and component to LibreOffice.
I reverted this change.

Nancy: normal you didn't understand this lingo, it's in fact Xorg/driver Radeon bug.

It was such a PITA to get X just start without getting a kernel panick that I never tried to update the box. My box is now almost stable. I really said *almost*. Still have random freeses.

The bug is not due or related to Radeon; I used to have the issue with X 1.5, when I had not any Radeon hardware installed, or driver compiled.

And I have no reasons to even try to update my software by now. I am waiting for Xinerama to support Xrandr since 2003; I at last have my Xinerama box the way I need it; the next step is to gain Xrandr. As long as there is no official announcement for Xinerama+Xrandr, and my box is almost stable, any update attempt would higly probably make the box less stable and less usable.

I reported the bug just when 1.9 came out.

The issue is that you don't fix the bug: you make a new release without fixing existing bugs; new releases may, or may not have bugs, depending on if the dev alter or not the code. Making a release when there are known bugs is just pointless; or renders bug-reporting completely useless.

I will update X when you will tell me "I have read your report, and found the bogous line, and I fixed it". Otherwise, just randomly ask me to check if a new version still have the bug is pointless: if no dev have read my bug, they can't have fixed the bug; so, if no dev came to tell the bug was fixed ... then, the bug is still in.

A part of the bug I complain in here, that I have encontoured with 1.9 was already in 1.5; I had reported it previously; it was not fixed, and as I just said ... as no dev tried to fix it in 1.5, the bug propagated to 1.9. 

But just waiting a few years won't change anything. The bug will be fixed only when a dev will try to fix it. No mistery.

And one last time: my bug is not Radeon specific. It can happen in a radeon conf, but can also happen in a 100% radeon free (hard+soft) conf. I already had the bug in my previous desktop which was 100% different hard, and 99% different soft. If you search for my bugs in Gentoo, you will find that I asked to keep Xorg 1.6 to not be stabilised because it was completely bugged. When X 1.6 stabilisation was a hot topic, I was completely opposed; when it became stable, I asked for immediate removal from stable tree (then I blocked it, and stuck to 1.5). When I bought a new computer, I could not block to 1.5, because 1.5 was not in tree anymore; so I had to do with 1.8; 1.8 NEVER worked for me; I had a lucky day with 1.9, and never touched anything since that day (mid july 2010). Will never update any thing untill I buy a new computer; this won't happen before 2015; my actual desktop works fine; no reason to mess it.

... unless an official news tells about Xinerama+Xrandr; in this case, I would make a complete backup of my system, and try to update.

X is soo unstable, I won't take any risk now. It's soo fragile (switch XEN on in BIOS or Linux; swap two video cards on bus; activate the wrong option in the BIOS, update or downgrade any video driver, install the wrong udev version, start application FOO with BAR env var declared, start a second server, try to play with FrameBuffer boot options, unplug a VGA plug ... any single tiny change at any level can make X crash, freese, or refuse to start, or induce a KP). And of course, it will silently ignore the conf file provided, and will never give any kind of error message (I even tried to get dmesg on a physical serial port: pointless). Now that I got something that works, I won't touch it before 2015.

To put it clearly, I can't help you with this.
I try to contribute to LibreOffice and saw this tracker which didn't concern LibreOffice at all. I've got a Debian but know nothing about Xorg or video drivers internals.

ritch bitch why shuould we help someone with that many monitors ?

i don't see evidence it's not the window manager

and are you saying there's no problem in XFree86 X.4.8.0 but there is in X11R7.x ?  meaning even with Xnest or vesa or framebuffer servers ?

if you know of anyone that has it workign did you ask them if they ran into the same difficulty ?

also did you compile this yourself ?  gentoo may have auto magically got and compiles X11R7.x but did you make 2>&1 World.log 2>&1 and check everything built and installed ?

------------------
the truth.  there are people who'd distract and delay your goal because it profits them if only the microsoft way works.  infact one can even run into fake complaints that way.

you did a nice job posting your info.  that's only the begginning of a bug report

we now need any info after the crash that got logged (to terminal) or logfile or in Xorg.log

============================
deeply sad linux has such bugs someone says?

REALLY.  well they always are lobbed from the same sources is what i'd say

the rest of it always worked and still works damn well

(In reply to DEMAINE Benoit-Pierre from comment #74)

> Now that I got something that works, I won't touch it before 2015.

Hopefully you've had a better experience with X since then. If you have further issues please do file them.

No; just did not update anything since 2010. Took me too long to get it working; no more time to loose on something that works. And since no other bug have been really fixed, I think that reinstalling or upgrading would break it all.

I have updated the disks, without touching the software. Just migrated data.