AGP PCI-E Issues (Rialto chip) No Interupt (IRQ) assigned, Poor performance.

[oliver]

Hi, I'm starting this new thread as the old one was getting to off-topic (http://www.phoronix.com/forums/showthread.php?t=6112)

The problem, AGP Cards, with PCI-E chips on them, and thus using the realto bridge chip, aren't getting Interupts assigned resulting in utterly poor performance.

The DMESG snipped from loading/activating the driver:

fglrx: module license 'Proprietary. (C) 2002 - ATI Technologies, Starnberg, GERMANY' taints kernel.
[fglrx] Maximum main memory to use for locked dma buffers: 1898 MBytes.
[fglrx] ASYNCIO init succeed!
[fglrx] PAT is enabled successfully!
[fglrx] module loaded - fglrx 8.43.2 [Nov  9 2007] on minor 0
ACPI: PCI Interrupt 0000:01:00.0[A] -> GSI 16 (level, low) -> IRQ 19
[fglrx] IRQ_MGR is disabled untill GART_CACHABLE memory will be implemented
[fglrx] Internal AGP support requested, but kernel AGP support active.
[fglrx] Have to use kernel AGP support to avoid conflicts.
[fglrx] AGP detected, AgpState   = 0x1f000a1b (hardware caps of chipset)
agpgart: Found an AGP 3.0 compliant device at 0000:00:00.0.
agpgart: Putting AGP V3 device at 0000:00:00.0 into 8x mode
agpgart: Putting AGP V3 device at 0000:01:00.0 into 8x mode
[fglrx] AGP enabled,  AgpCommand = 0x1f000312 (selected caps)

The relevant part from the Xorg.log:

(II) fglrx(0): [drm] DRM interface version 1.0
(II) fglrx(0): [drm] created "fglrx" driver at busid "PCI:1:0:0"
(II) fglrx(0): [drm] added 8192 byte SAREA at 0x2000
(II) fglrx(0): [drm] mapped SAREA 0x2000 to 0xb7f9e000
(II) fglrx(0): [drm] framebuffer handle = 0x3000
(II) fglrx(0): [drm] added 1 reserved context for kernel
(II) fglrx(0): DRIScreenInit done
(II) fglrx(0): Kernel Module Version Information:
(II) fglrx(0):     Name: fglrx
(II) fglrx(0):     Version: 8.43.2
(II) fglrx(0):     Date: Nov  9 2007
(II) fglrx(0):     Desc: ATI FireGL DRM kernel module
(II) fglrx(0): Kernel Module version matches driver.
(II) fglrx(0): Kernel Module Build Time Information:
(II) fglrx(0):     Build-Kernel UTS_RELEASE:        2.6.23-gentoo-r1
(II) fglrx(0):     Build-Kernel MODVERSIONS:        no
(II) fglrx(0):     Build-Kernel __SMP__:            yes
(II) fglrx(0):     Build-Kernel PAGE_SIZE:          0x1000
(II) fglrx(0): [drm] register handle = 0x00004000
(EE) fglrx(0): Failed to enable interrupts.
(II) fglrx(0): [pci] find AGP GART
(II) fglrx(0): [agp] Mode=0x1f000a1b bridge: 0x1106/0x0282
(II) fglrx(0): [agp] AGP v1/2 disable mask 0x00000000
(II) fglrx(0): [agp] AGP v3 disable mask   0x00000000
(II) fglrx(0): [agp] enabling AGP with mode=0x1f000b1a
(II) fglrx(0): [agp] Remapping MC AGP space (new MCAGPBase = 0xc0000000)
(II) fglrx(0): [agp] AGP protocol is enabled for graphics board. (cmd=0x1f000312)
(II) fglrx(0): [agp] graphics chipset has AGP v3.0 (native mode)
(II) fglrx(0): DRI initialization successfull!
(II) fglrx(0): FBADPhys: 0xe0100000 FBMappedSize: 0x01000400
(II) fglrx(0): FBMM initialized for area (0,0)-(1280,3277)
(II) fglrx(0): FBMM auto alloc for area (0,0)-(1280,1024) (front color buffer - assumption)
(II) fglrx(0): Largest offscreen area available: 1280 x 2253
(==) fglrx(0): Backing store disabled

This has been going on since .. all the 'new' 8.X driver series, the only other driver that was usable before that was 8.34.5.

Since i also have an 9500Pro AGP, i can luckly use that one, which works great. So it's definably driver related. The card also works fine in Windos. (Yes i actually installed ONLY to see if the card would work properly there. I'm that desperate).

So please, Micheal, ATi-devs .. somebody, acknowledge this bug. At least then we don't have to test these drivers constantly, but can simply see in the known bugs section there's still an issue here. Also If none of the beta testers have AGP PCI-e cards, i'm still happy to sign up as an BETA tester.

bug #914 on the bugzilla (http://ati.cchtml.com/show_bug.cgi?id=914) also is about this issue.

[Crunchy]

I get exactly the same error on my x1950 Pro AGP under Gentoo with any Xorg version.

[liqor_mortis]

Me too, with gentoo-sources 2.6.23r2 and a 1950pro AGP. I wonder if it's a problem with gentoo specifically, or whether other distros are seeing the same problem.

Would a different kernel make a difference?

[figvam]

Why do you think it's related to the kernel? Looks more like the unfortunate combination of hardware with inability of fglrx driver to handle it to me.

Have any of you tried disabling ACPI (acpi=off in the kernel command line)? Maybe manually assigning interrupts via BIOS would work?

Another thing that comes to mind: would setting AGP aperture as non-cacheable via MTRR work?

[oliver]

Originally Posted by figvam

Why do you think it's related to the kernel? Looks more like the unfortunate combination of hardware with inability of fglrx driver to handle it to me.

Have any of you tried disabling ACPI (acpi=off in the kernel command line)? Maybe manually assigning interrupts via BIOS would work?

Another thing that comes to mind: would setting AGP aperture as non-cacheable via MTRR work?

All fair points, however:
Why does the exact same config (bios +sofwatware) and an 9500Pro work, but the X1950Pro (with the bridge chip!) not.

Why does the exact same hardware all work fine with 8.34 but not the 8.4* series?

I'm willing to try endless configs, that's not it, but it's far easier to point to the driver

[Crunchy]

I've been fiddling with the MTRR and I'm convinced that the driver simply doesn't support something yet.

Here's the MTRR, as you would expect:

[fglrx] IRQ_MGR is disabled untill GART_CACHABLE memory will be implemented

I'd interpret this as saying that the IRQ manager is disabled till full MTRR support is written.

I disabled the MTRR register that should've been used for the AGP apeture but it wasn't recreated when I launched X as I would expect. Here's what I did:

> cat /proc/mtrr
reg00: base=0x00000000 (   0MB), size=1024MB: write-back, count=1
reg01: base=0xe0000000 (3584MB), size= 256MB: write-combining, count=1

reg01 is the area that should be used for the AGP apeture. So I disable it:

> echo "disable=1" >| /proc/mtrr
> cat /proc/mtrr
reg00: base=0x00000000 (   0MB), size=1024MB: write-back, count=1

After restarting X the area should be recreated by the driver, it isn't. I tried setting the "mtrr" option, in the device section of xorg.conf, to both on and off and there was no difference. I would expect the driver to recreate the area, complain or crash.

[uzi-]

I've tried some things which all won't help.
PCI-E support in kernel on/off
ACPI on/off
with and without kernel agp

I can also confirm that this only happens to rialto-based cards, tried a X1650 AGP, 9500 AGP and mobile xpress 1100(complete ati chipset).
My desktop chipset is an ancient nForce2

The performance is so poor, but it looks pretty funny to turn a transparent compiz-cube in slow-motion

[Crunchy]

I've tried just about every kernel option I can think of to do with interrupts, pci, mtrr arg

[mabco]

Same issue here with the X1950 PRO AGP(512) still waiting.....for a miracle to save me from Vi$ta. Can at least get proper 2D support now, next should be 3D fixes along with the AGP/PCI-E Bridge issues with IRQs and wrong detection.

[figvam]

Originally Posted by oliver

Why does the exact same config (bios +sofwatware) and an 9500Pro work, but the X1950Pro (with the bridge chip!) not.

Why does the exact same hardware all work fine with 8.34 but not the 8.4* series?

I'm willing to try endless configs, that's not it, but it's far easier to point to the driver

I'm not arguing it's fglrx inability to handle the cetrtain hardware, I was just replying to some previous poster's suggestion it's the kernel fault.