Bug 5926

Summary:	radeonfb somtimes hangs on x300
Product:	Drivers	Reporter:	Carlo Contavalli (ccontavalli)
Component:	Console/Framebuffers	Assignee:	James Simmons (jsimmons)
Status:	REJECTED INSUFFICIENT_DATA
Severity:	normal	CC:	adaplas, benh, bunk, ccontavalli
Priority:	P2
Hardware:	i386
OS:	Linux
Kernel Version:	any 2.6.x, up to 2.6.14	Subsystem:
Regression:	---	Bisected commit-id:

Description Carlo Contavalli 2006-01-19 16:36:01 UTC

Most recent kernel where this bug did not occur: 
  I have only used 2.6 kernels, and it has always showed up,
  with all the versions I tested, up to 2.6.14.

Distribution: 
  I'm using Debian GNU Linux from the testing archive. Tested both
  with vanilla kernel from kernel.org and the distribution one.

Hardware Environment:
  Dell Inspiron 6000 with x300 radeon Video card.
01:00.0 VGA compatible controller ATI Technologies Inc 5460 00 00 Dell 2003

Problem Description:
  Ok. I use the console as much as I can. Most often without running
  X. I use the radeonfb driver. With some ncurses applications, most notably
  mutt, when certain operations are performed my computer completely hangs.
  If mp3 were running in background, they stop. sysreq+s, sysreq+... others,
  do not work. I need to reboot using sysreq + b, which seems to be
  the only one working (or pushing the power off button for 5 seconds). 
  Once it shows up, I can reproduce the bug on a regular basis (100%). It 
  usually shows up when I use the 'v' key to see the attachments in
  a mail or when browsing a deep mail thread. It usually hangs in the 
  middle of redrawing the screen for a scroll down/scroll up event,
  with half the screen redrawn and the lower half still with the previous
  page shown. It is quite annoying. It is not funny. Don't laugh at it :), 
  even if it saves me from many flame wars. 
  If I try to connect to the computer while it is hanged with ssh,
  it does not work. If I leave the computer on, just for the curiosity 
  to see what happens, the console does not blank, but the redrawing goes
  on at a really slow speed (like 10/15 minutes to see a couple more lines,
  which come out all together). I don't think it's a bug in mutt or in
  the ncurses library :-) 
  Obviously, if I don't modprobe radeonfb, the bug doesn't show up...
  I haven't seen any difference with the various kernel versions.

Steps to reproduce:
  hem.. subscribe lkml, buy a dell inspiron 6000, use radeonfb, wait
  for the first flame war and try to take part in the discussion. It 
  may as well be some kind of flame hardware protection or some sort
  of divine punishment... If you really need to, I can provide my own 
  muttrc. I can do whatever patching, dumping, testing, debugging, 
  coding may become necessary to fix the problem. I have a couple 
  mails/threads I'd really like to read from mutt  which always hang 
  my computer. I know it sounds crazy, but I have yet to figure out
  where the problem lies.

Comment 1 Antonino Daplas 2006-01-23 05:24:12 UTC

Try booting with video=radeonfb:noaccel.

If the above doesn't help, try 

fbset -vyres n

where n = vertical resolution of your display.

Let us know what happens.

Comment 2 Carlo Contavalli 2006-01-23 13:57:22 UTC

On Mon, Jan 23, 2006 at 05:24:15AM -0800, bugme-daemon@bugzilla.kernel.org wrote:
> Try booting with video=radeonfb:noaccel.
uhm .. passed the parameter with modprobe. With noaccel, couldn't
hang the pc in anyway. I could read the whole dm versus md thread
without troubles :)

reboot + modprobe _without noaccel_ hanged almost immediately.

Is there anything I can do to help fixing the accelerated code?

Thanks,
Cheers,
Carlo

Comment 3 Benjamin Herrenschmidt 2006-01-23 15:11:56 UTC

What if you do radeonfb.default_dynclk=0 instead of noaccel ?

Comment 4 Carlo Contavalli 2006-01-24 01:56:04 UTC

On Mon, Jan 23, 2006 at 03:12:00PM -0800, bugme-daemon@bugzilla.kernel.org happily wrote:
> What if you do radeonfb.default_dynclk=0 instead of noaccel ? 
ok, with default_dynclk set to 0, I get the same problem described
in the first mail. A crash after a few mails...
So:
  modprobe radeonfb -> crash
  modprobe radeonfb default_dynclk=0 -> crash
  modprobe radeonfb noaccel -> works fine

(for every command, I verified the dmesg output to see if the
option was actually accepted).

Cheers,
Carlo

Comment 5 Antonino Daplas 2006-01-24 04:01:15 UTC

Can you open drivers/video/aty/radeon_base.c and look for

static struct fb_ops radeonfb_ops?

Then replace radeonfb_imageblit|fillrect|copyarea with
cfb_imageblit|fillrect|copyarea.  Then put back the accelerated versions one by
one so we know which of the accel functions is causing the hang.

Comment 6 Adrian Bunk 2006-04-22 12:39:04 UTC

Please reopen this bug if:
- it is still present in kernel 2.6.16 and
- you can provide the requested information.