trapped reads in 7.3-rc1 since nv50 instmem handling change

trapped reads in 7.3-rc1 since nv50 instmem handling change

[Jamie Heilman]

On my workstation with a Quadro NVS 290 running 7.3-rc1 I'm seeing
this sort of thing show up spuriously:

kernel: nouveau 0000:01:00.0: fb: trapped read at 0100ef56e0 on channel -1 [0fee0000 unknown] engine 06 [BAR] client 08 [PFIFO_READ] subclient 01 [IN] reason 00000002 [PAGE_NOT_PRESENT]
kernel: nouveau 0000:01:00.0: fb: trapped read at 0100f85788 on channel -1 [0fee0000 unknown] engine 06 [BAR] client 08 [PFIFO_READ] subclient 01 [IN] reason 00000002 [PAGE_NOT_PRESENT]
(repeated 16 times)

I bisected it back to 34e27b90552a (nouveau/instmem: use iomapping
interface for instmem handling).  The rest of the salient dmesg during
initialization is:

kernel: nouveau 0000:01:00.0: NVIDIA G86 (086f00a2)
kernel: nouveau 0000:01:00.0: bios: version 60.86.6c.00.21
kernel: nouveau 0000:01:00.0: vgaarb: deactivate vga console
kernel: Console: switching to colour dummy device 80x25
kernel: nouveau 0000:01:00.0: bios: M0203T not found
kernel: nouveau 0000:01:00.0: bios: M0203E not matched!
kernel: nouveau 0000:01:00.0: fb: 256 MiB DDR2
kernel: nouveau 0000:01:00.0: drm: VRAM: 256 MiB
kernel: nouveau 0000:01:00.0: drm: GART: 1048576 MiB
kernel: nouveau 0000:01:00.0: drm: TMDS table version 2.0
kernel: nouveau 0000:01:00.0: drm: MM: using CRYPT for buffer copies
kernel: [drm] Initialized nouveau 1.4.3 for 0000:01:00.0 on minor 0
kernel: Console: switching to colour frame buffer device 240x75
kernel: nouveau 0000:01:00.0: [drm] fb0: nouveaudrmfb frame buffer device

It doesn't always happen at boot, sometimes it takes a while to show
up as it seems to be related to how much I have going on at any point
in time.  That commit reverts cleanly for the moment, but I'm happy to
test any ideas or enable further debugging as needed, assuming this
wasn't intentional.

-- 
Jamie Heilman                     http://audible.transient.net/~jamie/

Re: trapped reads in 7.3-rc1 since nv50 instmem handling change

[Jamie Heilman]

Jamie Heilman wrote:
> On my workstation with a Quadro NVS 290 running 7.3-rc1 I'm seeing
> this sort of thing show up spuriously:
> 
> kernel: nouveau 0000:01:00.0: fb: trapped read at 0100ef56e0 on channel -1 [0fee0000 unknown] engine 06 [BAR] client 08 [PFIFO_READ] subclient 01 [IN] reason 00000002 [PAGE_NOT_PRESENT]
> kernel: nouveau 0000:01:00.0: fb: trapped read at 0100f85788 on channel -1 [0fee0000 unknown] engine 06 [BAR] client 08 [PFIFO_READ] subclient 01 [IN] reason 00000002 [PAGE_NOT_PRESENT]
> (repeated 16 times)
> 
> I bisected it back to 34e27b90552a (nouveau/instmem: use iomapping
> interface for instmem handling).  The rest of the salient dmesg during
> initialization is:
> 
> kernel: nouveau 0000:01:00.0: NVIDIA G86 (086f00a2)
> kernel: nouveau 0000:01:00.0: bios: version 60.86.6c.00.21
> kernel: nouveau 0000:01:00.0: vgaarb: deactivate vga console
> kernel: Console: switching to colour dummy device 80x25
> kernel: nouveau 0000:01:00.0: bios: M0203T not found
> kernel: nouveau 0000:01:00.0: bios: M0203E not matched!
> kernel: nouveau 0000:01:00.0: fb: 256 MiB DDR2
> kernel: nouveau 0000:01:00.0: drm: VRAM: 256 MiB
> kernel: nouveau 0000:01:00.0: drm: GART: 1048576 MiB
> kernel: nouveau 0000:01:00.0: drm: TMDS table version 2.0
> kernel: nouveau 0000:01:00.0: drm: MM: using CRYPT for buffer copies
> kernel: [drm] Initialized nouveau 1.4.3 for 0000:01:00.0 on minor 0
> kernel: Console: switching to colour frame buffer device 240x75
> kernel: nouveau 0000:01:00.0: [drm] fb0: nouveaudrmfb frame buffer device
> 
> It doesn't always happen at boot, sometimes it takes a while to show
> up as it seems to be related to how much I have going on at any point
> in time.  That commit reverts cleanly for the moment, but I'm happy to
> test any ideas or enable further debugging as needed, assuming this
> wasn't intentional.

This is still present after deced5fa01c5 ("nouveau/instmem: handle
iomapping already existing") and in 7.3-rc4.  Apart from the log spam
(which has made journald angry once by flooding the buffer) it doesn't
seem to have any other serious consequences yet though.

-- 
Jamie Heilman                     http://audible.transient.net/~jamie/