Re: aiodone daemon kernel used before initialisation

[Chuck Silvers <chuq%chuq.com@localhost>]

On Mon, Dec 28, 2020 at 05:14:52AM +0100, Pierre Pronchery wrote:
> 			Hi tech-pkg@,
> 
> While trying to debug an issue with my Radeon HD4870 during initialisation (the UVD firmware does not load) I noticed an error in the kernel trap. The important part of the backtrace is as follows: (copied manually)
> 
> > kernel: page fault trap, code=0
> > Stopped in pid 0.100 (system) at netbsd:workqueue_enqueue+0x16: testb $0x2,10(%rdi)
> > workqueue_enqueue() at netbsd:workqueue_enqueue+0x16
> > biodone2() at netbsd:biodone2+0xd8
> > [â?¦]
> > trap() at netbsd:trap+0x70c
> > --- trap (number 6) ---
> > copy() at netbsd:kcopy+0x15
> > uiomove() at netbsd:uiomove+0x93
> > [â?¦]
> > firmware_read() at netbsd:firmware_read+0x5e
> 
> 
> I figured that the call to workqueue_enqueue() is performed on an uninitialised queue, in main() (â??aiodonedâ??).
> When moving its initialisation before mounting the root file system, the error while loading the firmware seems to be caught:
> 
> > radeon0: unable to attach drm: 22
> 
> 
> Let me know if the patch attached is therefore correct. (I am working on the netbsd-9 branch)

That change is fine for netbsd-9.
I fixed it differently in HEAD, but I made some mistakes and it took a few commits
to get it straightened out, and I'm not sure I could find all the related commits now.


> Of course, any help for the actual problem preventing the radeon DRM code from loading RV770_uvd.bin is welcome.
> It is the 5th file out of 5 to be loaded, and the 4 previous firmware files load correctly, with seemingly identical code.
> The correct file is present in the filesystem (booting with a ramdisk) and request_firmware() sees the correct size
> (90668 bytes compared to 3392, 5440, 4096, and 16656 earlier).

I have no idea about this part, my only suggestion is to keep adding printf's down the call stack.

-Chuck