On Thu, Dec 30, 2004 at 06:48:25PM +1100, Dave Airlie wrote:
> http://www.skynet.ie/~airlied/patches/dri/drm_core_split-26bk.diff
Applying this patch to 2.6.10 shows that the oops-and-crash in
i810_mmap_buffers I reported in the bk-drm (and -mm) tree is
here. :( Vanilla 2.6.10 is fine.
I'll repeat what I worked out about this for anyone interested; I
was able to get the i810 DRM working again by reverting this change:
http://drm.bkbits.net:8080/drm-2.6/cset@418a0608l2KOkX2bWPW4DYyiQfa69A?nav=index.html|ChangeSet@-4w
I made this patch with some ignorant copy-and-paste coding, which
is wrong but works for me:
diff -aurN a/drivers/char/drm/i810_drv.c b/drivers/char/drm/i810_drv.c
--- a/drivers/char/drm/i810_drv.c 2004-11-26 23:18:35.000000000 -0500
+++ b/drivers/char/drm/i810_drv.c 2004-11-26 23:38:36.000000000 -0500
@@ -112,7 +112,7 @@
.open = drm_open,
.release = drm_release,
.ioctl = drm_ioctl,
- .mmap = i810_mmap_buffers,
+ .mmap = drm_mmap,
.poll = drm_poll,
.fasync = drm_fasync,
},
A copy of the Oops output (this one from 2.6.10-rc2-mm1):
Unable to handle kernel NULL pointer dereference at virtual address 00000038
printing eip:
c0235d04
*pde = 08aed067
*pte = 00000000
Oops: 0000 [#1]
PREEMPT
Modules linked in: lp af_packet 3c59x mii yenta_socket pcmcia_core snd_intel8x0
snd_ac97_codec i2c_i801 i2c_core uhci_hcd usbcore i8xx_tco hw_random floppy parport_pc
parport tsdev joydev nls_iso8859_1 ntfs dm_mod snd_seq_oss snd_seq_midi_event snd_seq
snd_seq_device snd_pcm_oss snd_pcm snd_timer snd_page_alloc snd_mixer_oss snd soundcore
ide_cd cdrom
CPU: 0
EIP: 0060:[<c0235d04>] Not tainted VLI
EFLAGS: 00013246 (2.6.10-rc2-mm1)
EIP is at i810_mmap_buffers+0x24/0x80
eax: 00000000 ebx: c90a0e34 ecx: 00000000 edx: cb1d5570
esi: c9b7dee0 edi: cb515560 ebp: c943ff40 esp: c943ff2c
ds: 007b es: 007b ss: 0068
Process XFree86 (pid: 3632, threadinfo=c943e000 task=cb1d5570)
Stack: 000cca1c c943ff80 c90a0e34 cb515560 ffffffea c943ff98 c01494fe c943ff84
c943ff88 c943ff4c bffff890 c015559d c943ff60 00000000 00000001 00000000
001000fb cb11c9d8 cb68adc0 00002000 b7f9c000 c90f0ac4 c90f0ae4 c90f0adc
Call Trace:
[<c0103e1a>] show_stack+0x7a/0x90
[<c0103f99>] show_registers+0x149/0x1b0
[<c010418d>] die+0xdd/0x170
[<c0113b6a>] do_page_fault+0x33a/0x68a
[<c0103a9b>] error_code+0x2b/0x30
[<c01494fe>] do_mmap_pgoff+0x33e/0x6f0
[<c0108b6c>] sys_mmap2+0x6c/0xa0
[<c0103011>] sysenter_past_esp+0x52/0x71
Code: ff ff eb d1 8d 76 00 55 89 e5 83 ec 14 89 5d f4 89 d3 89 7d fc 89 c7 89 75 f8 8b 70
7c e8 85 fd 0c 00 8b 46 20 8b 80 04 07 00 00 <8b> 40 38 8b 40 44 81 4b 14 00 40 02 00 89 7b
4c c7 40 08 01 00
--
Joseph Fannin
[email protected] || [email protected]
>
> Applying this patch to 2.6.10 shows that the oops-and-crash in
> i810_mmap_buffers I reported in the bk-drm (and -mm) tree is
> here. :( Vanilla 2.6.10 is fine.
>
> I'll repeat what I worked out about this for anyone interested; I
> was able to get the i810 DRM working again by reverting this change:
> http://drm.bkbits.net:8080/drm-2.6/cset@418a0608l2KOkX2bWPW4DYyiQfa69A?nav=index.html|ChangeSet@-4w
>
I've figured this out, your change is actually partly correct, there is a
bit more that needs to be done though, I'll post a patch tomorrow
hopefully...
Dave.
--
David Airlie, Software Engineer
http://www.skynet.ie/~airlied / airlied at skynet.ie
pam_smb / Linux DECstation / Linux VAX / ILUG person
>
> I'll repeat what I worked out about this for anyone interested; I
> was able to get the i810 DRM working again by reverting this change:
> http://drm.bkbits.net:8080/drm-2.6/cset@418a0608l2KOkX2bWPW4DYyiQfa69A?nav=index.html|ChangeSet@-4w
>
actually the patch is correct, the DRM CVS had some extra code but I never
moved that to the kernel as I was unsure of its correctness,
I'll follow this up with a tree pull request for it..
Dave.
--
David Airlie, Software Engineer
http://www.skynet.ie/~airlied / airlied at skynet.ie
pam_smb / Linux DECstation / Linux VAX / ILUG person