linux-3.3-rc2 and radeon kms failure on ppc32 with Radeon 9250 pci

[Michel Dänzer]

On Son, 2012-02-05 at 00:41 +0100, acrux wrote:
> unable to have a working radeon kms framebuffer with linux-3.3-rc2 on ppc
> video card: Radeon 9250 PCI

Is this a regression? If yes, can you bisect?


> Machine check in kernel mode.
> Data Write PLB Error
> Machine Check exception is imprecise
> Oops: Machine check, sig: 7 [#1]
> Canyonlands
> Modules linked in:
> NIP: c000a580 LR: c0399084 CTR: 000bfffb
> REGS: efff7f10 TRAP: 0214   Not tainted  (3.3.0-rc2)
> MSR: 00029000 <CE,EE,ME>  CR: 24714222  XER: 00000000
> TASK = ef830000[1] 'swapper' THREAD: ef834000
> GPR00: 00000000 ef835c30 ef830000 f5500000 00000000 00300000 ef835bd8 00000000
> GPR08: ef835b38 f5500014 00000000 000c0001 24714284 8500682f ef8f7800 ef17c1c0
> GPR16: 00000020 c05f0000 c06105de 00000000 ef835d08 c0610351 c056a6f0 c0610600
> GPR24: fffffff4 ef8ca47c ef9ffe00 ef9fff38 ef9e7c00 ef835cb8 ef8ea000 ef8ca400
> NIP [c000a580] _memset_io+0x54/0x90
> LR [c0399084] radeon_fb_find_or_create_single+0x234/0x42c
> Call Trace:
> [ef835c30] [c0399068] radeon_fb_find_or_create_single+0x218/0x42c (unreliable)

Again looks like the problem occurs when first accessing VRAM, in this
case for clearing the visible framebuffer contents. 

I wonder if we're missing something to handle device memory access
properly on your machine(s)... Is ioremap_wc() working on them with
other drivers? 


-- 
Earthling Michel Dänzer           |                   http://www.amd.com
Libre software enthusiast         |          Debian, X and DRI developer