Revert 737a3bb9416ce2a7c7a4170852473a4fcc9c67e8 ?

[Gabriel Paubert]

On Wed, Apr 06, 2011 at 06:46:55PM +1000, Dave Airlie wrote:
> 2011/4/6 Uwe Kleine-König <u.kleine-koenig at pengutronix.de>:
> > Hi Gabriel,
> >
> > On Tue, Apr 05, 2011 at 01:52:59AM +0200, Gabriel Paubert wrote:
> >> I've had the following funny crashes on PPC machines, with
> >> cataleptic X server as a consequence:
> >>
> >> kernel: [drm] Setting GART location based on new memory map
> >> kernel: Oops: Exception in kernel mode, sig: 4 [#1]
> >> kernel: CHRP
> >> kernel: last sysfs file: /sys/devices/pci0001:01/0001:01:08.0/resource
> >> kernel: NIP: c05648fc LR: c0226f58 CTR: 00000008
> >> kernel: REGS: ddb53d20 TRAP: 0700   Not tainted  (2.6.38)
> >> kernel: MSR: 00089032 <EE,ME,IR,DR>  CR: 48044482  XER: 00000000
> >> kernel: TASK = ddab12b0[3040] 'Xorg' THREAD: ddb52000
> >> kernel: GPR00: c0226f34 ddb53dd0 ddab12b0 00000000 c0509e6c 00000000 00000000 00000000
> >> kernel: GPR08: 00000000 00000000 00000000 00000000 28044488 101f3d8c bf8166b4 00002c00
> >> kernel: GPR16: 101b9458 1006f1a0 101ebe0c 00000001 101ebe08 00000000 df9efc20 df9efc00
> >> kernel: GPR24: c0591e54 80546440 ddacf660 df9efc00 c0506048 c0480210 00a00000 df9ef800
> >> kernel: NIP [c05648fc] platform_device_register_resndata+0x4/0xa4
> >> kernel: LR [c0226f58] radeon_cp_init+0xd08/0x10c4
> >> kernel: Call Trace:
> >> kernel: [ddb53dd0] [c0226f34] radeon_cp_init+0xce4/0x10c4 (unreliable)
> >> kernel: [ddb53df0] [c020801c] drm_ioctl+0x2c0/0x3e4
> >> kernel: [ddb53eb0] [c0091264] do_vfs_ioctl+0x674/0x710
> >> kernel: [ddb53f10] [c0091340] sys_ioctl+0x40/0x70
> >> kernel: [ddb53f40] [c00111a8] ret_from_syscall+0x0/0x38
> >> kernel: --- Exception: c01 at 0xfc54a78
> >> kernel:     LR = 0xfc549dc
> >> kernel: Instruction dump:
> >> kernel: 736f2e31 32002f75 73722f6c 69622f6c 6962786b 6c617669 65722e73 6f2e3132
> >> kernel: 006c6962 786b6266 696c652e 736f2e31 <002f7573> 722f6c69 622f6c69 62786b62
> >> kernel: ---[ end trace ed79daba161e31d9 ]---
> >>
> >> As you can see, the processor is trying to execute ASCII strings like
> >> "/usr/lib/libxkb" and has trouble digesting them :-)
> >>
> >> The backtrace is actually missing radeon_cp_init_microcode and radeon_do_init_cp
> >> which are inlined inside radeon_cp_init.
> >>
> >> The trouble is that radeon_cp_init_microcode calls platform_device_register_simple
> >> which is a simple inline wrapper around platform_device_register_resndata, which
> >> happens to be already freed and overwritten with something looking like a list
> >> of filenames, since I have a non modular kernel.
> >>
> >> For now I have locally reverted 737a3bb9416ce2a7c7a4170852473a4fcc9c67e8
> >> which simply added an _init_or_module section attribute to
> >> platform_device_register_resndata, and X is up again...
> >>
> >> Now it may be that it is the ioctl that does not have the right to do
> >> this. Actually I thought that the name radeon_cp that is registered there
> >> would appear somwhere under /sys (or /proc) but failed to find it...
> > I don't know for sure, but it looks strange to me that an ioctl can
> > register a device. But the fear for such code in the kernel made me
> > choose not to squash 737a3bb941 into 44f28bdea094. So my POV is that if
> > the maintainer of the radeon driver thinks registering the device is OK,
> > reverting 737a3bb9416 is fine for me.
> 
> This is the old DRM driver for radeon, which relies on userspace to
> start X then calls the kernel
> to initialise the hardware. Due to this model, there is no device we
> can hang off (the PCI device
> might already be bound to fbdev), so we are forced to create a
> platform device to load the firmware.
> 
> So its ugly, unless someone can suggest a better device to hang things
> off I don't know of another way.
> 

The probem is that, at least on one of my machines, the new driver
does not work: the system hangs (apparently solid, but it's before
networking starts up and I've not yet hooked up a serial console), 
after the "radeon: ib pool ready" message.

With the old driver, I've found some combinations of configuration
options that works. They all fail when DRM_RADEON_KMS is enabled.

	Gabriel
> Dave.