2.6.28 radeon.ko oops, QS22 (Cell) blade.

2.6.28 radeon.ko oops, QS22 (Cell) blade.

[Wartan Hachaturow]

Hi there.

While trying to bring up X.org on this beast, I've stumbled upon this
oops in radeon.ko
(right on the first write attempt):

[drm] Initialized drm 1.1.0 20060810
pci 0005:01:00.0: enabling device (0140 -> 0143)
[drm] Initialized radeon 1.29.0 20080528 on minor 0
[drm] Setting GART location based on new memory map
[drm] GART aligned down from 0x04010000 to 0x04000000
[drm] Loading R500 Microcode
[drm] Num pipes: 1
[drm] writeback test failed
[drm] BEGIN_RING( 8 )
[drm]    OUT_RING( 0x000005c9 ) at 0x6402505
Unable to handle kernel paging request for data at address 0xd00000001a04c414
Faulting instruction address: 0xd000000000fd6d68
Oops: Kernel access of bad area, sig: 11 [#1]
SMP NR_CPUS=128 NUMA Cell
Modules linked in: radeon drm sunrpc pmi tg3 mptsas mptscsih mptbase
scsi_transport_sas
NIP: d000000000fd6d68 LR: d000000000fd6d54 CTR: 0000000000000001
REGS: c0000000fb4f38b0 TRAP: 0300   Not tainted  (2.6.28-lks-wks-alt1)
MSR: 9000000000009032 <EE,ME,IR,DR>  CR: 24004424  XER: 20000000
DAR: d00000001a04c414, DSISR: 0000000042000000
TASK = c0000000fb4e4f50[4088] 'X' THREAD: c0000000fb4f0000 CPU: 2
GPR00: 00000000000005c9 c0000000fb4f3b30 d000000001016c80 d000000000fff578
GPR04: 0000000000000000 ffffffffffffffff c00000000097d6b0 c000000000805554
GPR08: 000000000001ffff 0000000019009414 c00000000097d6b4 0000000000000001
GPR12: d000000000ffab80 c000000000952800 0000000000000005 0000000000000000
GPR16: 00000000100b3e50 0000000010090000 0000000010080000 0000000010090000
GPR20: 00000000100b3e80 0000000000000000 0000000000000006 0000000000000000
GPR24: ffffffffffffffea d000000001043000 c0000000fecba000 000000000003ffff
GPR28: d000000000fff578 0000000006402506 d000000001015b38 c0000000fecbf000
NIP [d000000000fd6d68] .radeon_cp_start+0x358/0x73c [radeon]
LR [d000000000fd6d54] .radeon_cp_start+0x344/0x73c [radeon]
Call Trace:
[c0000000fb4f3b30] [d000000000fd6d54] .radeon_cp_start+0x344/0x73c [radeon]
(unreliable)
[c0000000fb4f3be0] [d000000000d9cbbc] .drm_ioctl+0x230/0x300 [drm]
[c0000000fb4f3ca0] [d000000000ff8e98] .radeon_compat_ioctl+0xb4/0xfc [radeon]
[c0000000fb4f3d40] [c000000000189944] .compat_sys_ioctl+0x140/0x488
[c0000000fb4f3e30] [c000000000008570] syscall_exit+0x0/0x40
Instruction dump:
388005c9 837a002c 3929ffe0 7fa5eb78 7f83e378 913a0030 48023e31 e8410028
7ba91764 380005c9 3bbd0001 7f83e378 <7c19492e> 38800033 7fbdd838 7fbd07b4
---[ end trace b46b031df5997cbf ]---
[drm:drm_release] *ERROR* Device busy: 1 0

Any quick ideas for me to check? The ring address seems to be quite strange.

-- 
Regards, Wartan.

Re: 2.6.28 radeon.ko oops, QS22 (Cell) blade.

[Arnd Bergmann]

On Tuesday 10 February 2009, Wartan Hachaturow wrote:

> [drm] Initialized drm 1.1.0 20060810
> pci 0005:01:00.0: enabling device (0140 -> 0143)
> [drm] Initialized radeon 1.29.0 20080528 on minor 0
> [drm] Setting GART location based on new memory map
> [drm] GART aligned down from 0x04010000 to 0x04000000
> [drm] Loading R500 Microcode
> [drm] Num pipes: 1
> [drm] writeback test failed
> [drm] BEGIN_RING( 8 )
> [drm]    OUT_RING( 0x000005c9 ) at 0x6402505
> Unable to handle kernel paging request for data at address 0xd00000001a04c414
> Faulting instruction address: 0xd000000000fd6d68
> Oops: Kernel access of bad area, sig: 11 [#1]
> SMP NR_CPUS=128 NUMA Cell
> Modules linked in: radeon drm sunrpc pmi tg3 mptsas mptscsih mptbase
> scsi_transport_sas
> NIP: d000000000fd6d68 LR: d000000000fd6d54 CTR: 0000000000000001
> REGS: c0000000fb4f38b0 TRAP: 0300   Not tainted  (2.6.28-lks-wks-alt1)
> MSR: 9000000000009032 <EE,ME,IR,DR>  CR: 24004424  XER: 20000000
> DAR: d00000001a04c414, DSISR: 0000000042000000
> TASK = c0000000fb4e4f50[4088] 'X' THREAD: c0000000fb4f0000 CPU: 2
> GPR00: 00000000000005c9 c0000000fb4f3b30 d000000001016c80 d000000000fff578
> GPR04: 0000000000000000 ffffffffffffffff c00000000097d6b0 c000000000805554
> GPR08: 000000000001ffff 0000000019009414 c00000000097d6b4 0000000000000001
> GPR12: d000000000ffab80 c000000000952800 0000000000000005 0000000000000000
> GPR16: 00000000100b3e50 0000000010090000 0000000010080000 0000000010090000
> GPR20: 00000000100b3e80 0000000000000000 0000000000000006 0000000000000000
> GPR24: ffffffffffffffea d000000001043000 c0000000fecba000 000000000003ffff
> GPR28: d000000000fff578 0000000006402506 d000000001015b38 c0000000fecbf000
> NIP [d000000000fd6d68] .radeon_cp_start+0x358/0x73c [radeon]
> LR [d000000000fd6d54] .radeon_cp_start+0x344/0x73c [radeon]
> Call Trace:
> [c0000000fb4f3b30] [d000000000fd6d54] .radeon_cp_start+0x344/0x73c [radeon]
> (unreliable)
> [c0000000fb4f3be0] [d000000000d9cbbc] .drm_ioctl+0x230/0x300 [drm]
> [c0000000fb4f3ca0] [d000000000ff8e98] .radeon_compat_ioctl+0xb4/0xfc [radeon]
> [c0000000fb4f3d40] [c000000000189944] .compat_sys_ioctl+0x140/0x488
> [c0000000fb4f3e30] [c000000000008570] syscall_exit+0x0/0x40
> Instruction dump:
> 388005c9 837a002c 3929ffe0 7fa5eb78 7f83e378 913a0030 48023e31 e8410028
> 7ba91764 380005c9 3bbd0001 7f83e378 <7c19492e> 38800033 7fbdd838 7fbd07b4
> ---[ end trace b46b031df5997cbf ]---
> [drm:drm_release] *ERROR* Device busy: 1 0
> 
> Any quick ideas for me to check? The ring address seems to be quite strange.
> 

Right, the ring address is definitely broken. 0xd0000000xxxxxxxxx is
probably the right location for the ring, but the offset 0x6402505
into it seems ways off. This comes from dev_priv->ring.tail, which is
set up in radeon_do_cp_reset() and radeon_cp_init_ring_buffer().
You can instrument these two functions with printk() to see whether
one of them is called at all, and where the value comes from.

	Arnd <><

Re: 2.6.28 radeon.ko oops, QS22 (Cell) blade.

[Benjamin Herrenschmidt]

On Tue, 2009-02-10 at 14:18 +0300, Wartan Hachaturow wrote:
> Hi there.
> 
> While trying to bring up X.org on this beast, I've stumbled upon this
> oops in radeon.ko
> (right on the first write attempt):

You may want to get whatever latest patches are in Dave Airlie's
dri-next tree. I've fixed a number of bugs in the DRI relative to having
PCI physical addresses above 4G, which probably will help.

Also, try at first with a kernel compiled for 4K pages, I know that
works with an R5xx on the 440 setup I've been hacking that on, and I
also know that I've been having problems when trying to use 64K pages,
which I haven't had time to investigate and fix yet.

You may also want to look at the batch of patch that Dave Miller sent to
the DRM list yesterday or so, fixing a whole bunch of additional bugs in
the DRM, among other, the wrong direction being set in the iommu which
would have caused problems on Cell.

With all those patches together, the latest xf86-video-ati from X.org
git, and a libpciaccess-enabled X server, there are some chances that it
will work.

Cheers,
Ben.

> [drm] Initialized drm 1.1.0 20060810
> pci 0005:01:00.0: enabling device (0140 -> 0143)
> [drm] Initialized radeon 1.29.0 20080528 on minor 0
> [drm] Setting GART location based on new memory map
> [drm] GART aligned down from 0x04010000 to 0x04000000
> [drm] Loading R500 Microcode
> [drm] Num pipes: 1
> [drm] writeback test failed
> [drm] BEGIN_RING( 8 )
> [drm]    OUT_RING( 0x000005c9 ) at 0x6402505
> Unable to handle kernel paging request for data at address 0xd00000001a04c414
> Faulting instruction address: 0xd000000000fd6d68
> Oops: Kernel access of bad area, sig: 11 [#1]
> SMP NR_CPUS=128 NUMA Cell
> Modules linked in: radeon drm sunrpc pmi tg3 mptsas mptscsih mptbase
> scsi_transport_sas
> NIP: d000000000fd6d68 LR: d000000000fd6d54 CTR: 0000000000000001
> REGS: c0000000fb4f38b0 TRAP: 0300   Not tainted  (2.6.28-lks-wks-alt1)
> MSR: 9000000000009032 <EE,ME,IR,DR>  CR: 24004424  XER: 20000000
> DAR: d00000001a04c414, DSISR: 0000000042000000
> TASK = c0000000fb4e4f50[4088] 'X' THREAD: c0000000fb4f0000 CPU: 2
> GPR00: 00000000000005c9 c0000000fb4f3b30 d000000001016c80 d000000000fff578
> GPR04: 0000000000000000 ffffffffffffffff c00000000097d6b0 c000000000805554
> GPR08: 000000000001ffff 0000000019009414 c00000000097d6b4 0000000000000001
> GPR12: d000000000ffab80 c000000000952800 0000000000000005 0000000000000000
> GPR16: 00000000100b3e50 0000000010090000 0000000010080000 0000000010090000
> GPR20: 00000000100b3e80 0000000000000000 0000000000000006 0000000000000000
> GPR24: ffffffffffffffea d000000001043000 c0000000fecba000 000000000003ffff
> GPR28: d000000000fff578 0000000006402506 d000000001015b38 c0000000fecbf000
> NIP [d000000000fd6d68] .radeon_cp_start+0x358/0x73c [radeon]
> LR [d000000000fd6d54] .radeon_cp_start+0x344/0x73c [radeon]
> Call Trace:
> [c0000000fb4f3b30] [d000000000fd6d54] .radeon_cp_start+0x344/0x73c [radeon]
> (unreliable)
> [c0000000fb4f3be0] [d000000000d9cbbc] .drm_ioctl+0x230/0x300 [drm]
> [c0000000fb4f3ca0] [d000000000ff8e98] .radeon_compat_ioctl+0xb4/0xfc [radeon]
> [c0000000fb4f3d40] [c000000000189944] .compat_sys_ioctl+0x140/0x488
> [c0000000fb4f3e30] [c000000000008570] syscall_exit+0x0/0x40
> Instruction dump:
> 388005c9 837a002c 3929ffe0 7fa5eb78 7f83e378 913a0030 48023e31 e8410028
> 7ba91764 380005c9 3bbd0001 7f83e378 <7c19492e> 38800033 7fbdd838 7fbd07b4
> ---[ end trace b46b031df5997cbf ]---
> [drm:drm_release] *ERROR* Device busy: 1 0
> 
> Any quick ideas for me to check? The ring address seems to be quite strange.
>