On Fri, 2007-05-25 at 12:13 +0200, Geert Uytterhoeven wrote:

On Fri, 4 May 2007, Antonino A. Daplas wrote:

quoted

On Fri, 2007-05-04 at 10:50 +0200, Geert Uytterhoeven wrote:

quoted

It looks like unregister_framebuffer() doesn't cause fbcon_del_cursor_timer()
to be called.

Hence if cursor_timer_handler() runs, it will try to schedule work on
fb_info.queue, which is no longer valid, and queue_work() will BUG_ON().

Or am I missing something?

The console cursor is not part of fbdev.  fbcon_del_cursor_timer() is
done in fbcon_exit() in fbcon.c. You cannot (or, rather, should not)
call unregister_framebuffer() unless fbcon unbinds the underlying
driver.

Indeed, when unbinding the console first using

    echo 0 > /sys/class/vtconsole/vtcon1/bind
    
the problem no longer happens on halt (Geoff: and ps3-hack-fbcon-shutdown.diff
is no longer needed).

However there are still a few remaining issues:
  - unbinding requires CONFIG_VT_HW_CONSOLE_BINDING=y,
  - unbinding can be done using sysfs only (for now),
  - worse, unbinding may fail, e.g. if the console is in graphics mode due to
    running X.

Our problem is that we have to shutdown ps3fb on reboot/kexec, to release the
hypervisor resources used by ps3fb. If unbinding fails, we cannot do that.

Both you and drmfb has the same problem, and Jesse submitted a patch
that will unbind fbcon prior to unregister_framebuffer(). The patch
needs a few tweaks but is fine by me in principle.

Search lkml for [PATCH 1/3] allow console unregistration.

As for the first, maybe a 

'select VT_HW_CONSOLE_BINDING if FRAMEBUFFER_CONSOLE' under ps3fb?

As for the second and third concerns, Jesse's patch would take care of
that.

Tony



-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/