linuxconsole-dev

ruby 2.6.3 died with "divide error" when switching to fbcon

[Helge H. <hel...@ai...>]

Configuration: matrox G550 dualhead, everything compiled-in (fbcon & 
framebuffer) no modules, smp, no preempt.

(Some stuff omitted, but I have a digital photo of it
if it matters.)

The framebuffers came up fine, then I got this:
Console: switching to colour MATROX 160x64 vc:1-16
divide error: 0000 [#1]
CPU:1
EIP: 060:[<c02afccb>] Not tainted
EFLAGS: 00010246
EIP is at cfb_imageblit+0x27b/0x610
eax:00000020 ebx:00000000 ecx: 00000020 edx: 00000000
esi:00000000 edi:00000000 ebp: 00000000 esp: c13dfd84
ds:007b es:007b ss:0068
Process swapper (pid:1 ...
stack: ...
Call trace:
move_buf_aligned
accel_putc
fbcon_putc
hide_softcursor
update_screen
vt_map_display
fbcon_add
cache_alloc_refill
fbcon_event_notify
notifier_call_chain
register_framebuffer
matroxfb_db_regit
matroxfb_db_register
fbcon_add
matroxfb_crtc2_probe
matroxfb_register_driver
matroxfb_crtc2_init
do_initcalls
init_workqueues
init
init
kernel_thread_helper
Code: f7 f7 85 d2 0f 85 ab 01 00 00 85 ed 0f 85 a3 01 00 00 8b 54
<0>Kernel panic: Attempted to kill init!
<0>Rebooting in 300 seconds

I checked out ruby from cvs yesterday, and copied it into
a 2.6.3 source tree.

Helge Hafting

Re: ruby 2.6.3 died with "divide error" when switching to fbcon

[James S. <jsi...@in...>]

I just looked at the code. The only way the could happen is if 
info->var.bits_per_pixel equals zero. Need to check the initialization to 
see if it sets this.
 

> Configuration: matrox G550 dualhead, everything compiled-in (fbcon & 
> framebuffer) no modules, smp, no preempt.
> 
> (Some stuff omitted, but I have a digital photo of it
> if it matters.)
> 
> The framebuffers came up fine, then I got this:
> Console: switching to colour MATROX 160x64 vc:1-16
> divide error: 0000 [#1]
> CPU:1
> EIP: 060:[<c02afccb>] Not tainted
> EFLAGS: 00010246
> EIP is at cfb_imageblit+0x27b/0x610
> eax:00000020 ebx:00000000 ecx: 00000020 edx: 00000000
> esi:00000000 edi:00000000 ebp: 00000000 esp: c13dfd84
> ds:007b es:007b ss:0068
> Process swapper (pid:1 ...
> stack: ...
> Call trace:
> move_buf_aligned
> accel_putc
> fbcon_putc
> hide_softcursor
> update_screen
> vt_map_display
> fbcon_add
> cache_alloc_refill
> fbcon_event_notify
> notifier_call_chain
> register_framebuffer
> matroxfb_db_regit
> matroxfb_db_register
> fbcon_add
> matroxfb_crtc2_probe
> matroxfb_register_driver
> matroxfb_crtc2_init
> do_initcalls
> init_workqueues
> init
> init
> kernel_thread_helper
> Code: f7 f7 85 d2 0f 85 ab 01 00 00 85 ed 0f 85 a3 01 00 00 8b 54
> <0>Kernel panic: Attempted to kill init!
> <0>Rebooting in 300 seconds
> 
> I checked out ruby from cvs yesterday, and copied it into
> a 2.6.3 source tree.
> 
> Helge Hafting
> 
> 
> -------------------------------------------------------
> SF.Net is sponsored by: Speed Start Your Linux Apps Now.
> Build and deploy apps & Web services for Linux with
> a free DVD software kit from IBM. Click Now!
> http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click
> _______________________________________________
> Linuxconsole-dev mailing list
> Lin...@li...
> https://lists.sourceforge.net/lists/listinfo/linuxconsole-dev
> 

Re: ruby 2.6.3 died with "divide error", ruby 2.6.5 has the same problem

[Helge H. <hel...@ai...>]

On Fri, Feb 27, 2004 at 05:30:36PM +0000, James Simmons wrote:
> 
> I just looked at the code. The only way the could happen is if 
> info->var.bits_per_pixel equals zero. Need to check the initialization to 
> see if it sets this.
>  
Did you find anything? I tried ruby-2.6.5, and got almost
exactly the same oops again.  I get the penguins and a message
about switchin to framebuffer, then this.  Only the first
lines are different from the 2.6.3 case:

divide error: 9999 [#1]
EIP at cfb_imageblit+0x262/0x5c0
trace:
buffered_runqueue
fbcon_putc
gotoxay
clear_region
hide_softcursor
update_screen
vt_map_display
fbcon_add
cache_alloc_refill
fbcon_event_notify
notifier_call_chain
register_framebuffer
matroxfb_db_regit
matroxfb_db_register
matroxfb_xrtx2_probe
matroxfb_register_driver
matroxfb_crtc2_init
do_initcalls
init
init
kernel_thread_helper
Kernel panic: attempted to kill init!

At this stage I only see the first framebuffer.  I don't know how far
the second get, because userland has to run "matroxset" before it 
gets routed to an actual screen.

The first framebuffer seems to work, it displays the penguins and
the oops flawlessly.  Could it be something wrong in initializing the
second fb?  Might a well-placed printk or something help?

Helge Hafting

Rest of the 2.6.3 ruby report, with a very similiar trace.
> > Configuration: matrox G550 dualhead, everything compiled-in (fbcon & 
> > framebuffer) no modules, smp, no preempt.
> > 
> > (Some stuff omitted, but I have a digital photo of it
> > if it matters.)
> > 
> > The framebuffers came up fine, then I got this:
> > Console: switching to colour MATROX 160x64 vc:1-16
> > divide error: 0000 [#1]
> > CPU:1
> > EIP: 060:[<c02afccb>] Not tainted
> > EFLAGS: 00010246
> > EIP is at cfb_imageblit+0x27b/0x610
> > eax:00000020 ebx:00000000 ecx: 00000020 edx: 00000000
> > esi:00000000 edi:00000000 ebp: 00000000 esp: c13dfd84
> > ds:007b es:007b ss:0068
> > Process swapper (pid:1 ...
> > stack: ...
> > Call trace:
> > move_buf_aligned
> > accel_putc
> > fbcon_putc
> > hide_softcursor
> > update_screen
> > vt_map_display
> > fbcon_add
> > cache_alloc_refill
> > fbcon_event_notify
> > notifier_call_chain
> > register_framebuffer
> > matroxfb_db_regit
> > matroxfb_db_register
> > fbcon_add
> > matroxfb_crtc2_probe
> > matroxfb_register_driver
> > matroxfb_crtc2_init
> > do_initcalls
> > init_workqueues
> > init
> > init
> > kernel_thread_helper
> > Code: f7 f7 85 d2 0f 85 ab 01 00 00 85 ed 0f 85 a3 01 00 00 8b 54
> > <0>Kernel panic: Attempted to kill init!
> > <0>Rebooting in 300 seconds
> > 
> > I checked out ruby from cvs yesterday, and copied it into
> > a 2.6.3 source tree.
> > 
> > Helge Hafting
> > 
> > 
> > -------------------------------------------------------
> > SF.Net is sponsored by: Speed Start Your Linux Apps Now.
> > Build and deploy apps & Web services for Linux with
> > a free DVD software kit from IBM. Click Now!
> > http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click
> > _______________________________________________
> > Linuxconsole-dev mailing list
> > Lin...@li...
> > https://lists.sourceforge.net/lists/listinfo/linuxconsole-dev
> > 

Re: ruby 2.6.3 died with "divide error", ruby 2.6.5 has the same problem

[Aivils <ai...@un...>]

On Friday 07 May 2004 00:24, Helge Hafting wrote:
> On Fri, Feb 27, 2004 at 05:30:36PM +0000, James Simmons wrote:
> > 
> > I just looked at the code. The only way the could happen is if 
> > info->var.bits_per_pixel equals zero. Need to check the initialization to 
> > see if it sets this.
> >  
> Did you find anything? I tried ruby-2.6.5, and got almost
> exactly the same oops again.  I get the penguins and a message
> about switchin to framebuffer, then this.  Only the first
> lines are different from the 2.6.3 case:
> 
> divide error: 9999 [#1]
> EIP at cfb_imageblit+0x262/0x5c0
> trace:
> buffered_runqueue
> fbcon_putc
> gotoxay
> clear_region
> hide_softcursor
> update_screen
> vt_map_display
> fbcon_add
> cache_alloc_refill
> fbcon_event_notify
> notifier_call_chain
> register_framebuffer
> matroxfb_db_regit
> matroxfb_db_register
> matroxfb_xrtx2_probe
> matroxfb_register_driver
> matroxfb_crtc2_init
> do_initcalls
> init
> init
> kernel_thread_helper
> Kernel panic: attempted to kill init!
> 
> At this stage I only see the first framebuffer.  I don't know how far
> the second get, because userland has to run "matroxset" before it 
> gets routed to an actual screen.
> 
> The first framebuffer seems to work, it displays the penguins and
> the oops flawlessly.  Could it be something wrong in initializing the
> second fb?  Might a well-placed printk or something help?

I can help only with randomly-placed printk :)
First try this patch
--- ruby-CVS-20040430/drivers/video/console/fbcon.c     2004-04-30 09:45:06.000000000 +0300
+++ ruby-test/drivers/video/console/fbcon.c     2004-05-07 09:41:49.000000000 +0300
@@ -499,7 +499,7 @@ static const char *fbcon_startup(struct
        vc->vc_can_do_color = info->var.bits_per_pixel != 1;

        /* We trust the mode the driver supplies. */
-       if (!done && info->fbops->fb_set_par)
+       if (info->fbops->fb_set_par)
                info->fbops->fb_set_par(info);

        DPRINTK("mode:   %s\n", info->fix.id);


I add this !done because my ati-rage3d dies as secondary fb. May be matrox
does all vice-versa, because more sane is calling fb_set_par().

fbcon_strartup() already have printk
        DPRINTK("res:    %dx%d-%d\n", info->var.xres,
                info->var.yres,
                info->var.bits_per_pixel);
Seems You cannot see that on the screen. Oops push out important text.

Aivils

Re: ruby 2.6.3 died with "divide error", ruby 2.6.5 has the same problem

[Helge H. <hel...@ai...>]

On Fri, May 07, 2004 at 10:09:57AM +0300, Aivils wrote:
> On Friday 07 May 2004 00:24, Helge Hafting wrote:
> > On Fri, Feb 27, 2004 at 05:30:36PM +0000, James Simmons wrote:
> > > 
> > > I just looked at the code. The only way the could happen is if 
> > > info->var.bits_per_pixel equals zero. Need to check the initialization to 
> > > see if it sets this.
> > >  
> > Did you find anything? I tried ruby-2.6.5, and got almost
> > exactly the same oops again.  I get the penguins and a message
> > about switchin to framebuffer, then this.  Only the first
> > lines are different from the 2.6.3 case:
> > 
> > divide error: 9999 [#1]
> > EIP at cfb_imageblit+0x262/0x5c0
> > trace:
> > buffered_runqueue
> > fbcon_putc
> > gotoxay
> > clear_region
> > hide_softcursor
> > update_screen
> > vt_map_display
> > fbcon_add
> > cache_alloc_refill
> > fbcon_event_notify
> > notifier_call_chain
> > register_framebuffer
> > matroxfb_db_regit
> > matroxfb_db_register
> > matroxfb_xrtx2_probe
> > matroxfb_register_driver
> > matroxfb_crtc2_init
> > do_initcalls
> > init
> > init
> > kernel_thread_helper
> > Kernel panic: attempted to kill init!
> > 
> > At this stage I only see the first framebuffer.  I don't know how far
> > the second get, because userland has to run "matroxset" before it 
> > gets routed to an actual screen.
> > 
> > The first framebuffer seems to work, it displays the penguins and
> > the oops flawlessly.  Could it be something wrong in initializing the
> > second fb?  Might a well-placed printk or something help?
> 
> I can help only with randomly-placed printk :)
> First try this patch
> --- ruby-CVS-20040430/drivers/video/console/fbcon.c     2004-04-30 09:45:06.000000000 +0300
> +++ ruby-test/drivers/video/console/fbcon.c     2004-05-07 09:41:49.000000000 +0300
> @@ -499,7 +499,7 @@ static const char *fbcon_startup(struct
>         vc->vc_can_do_color = info->var.bits_per_pixel != 1;
> 
>         /* We trust the mode the driver supplies. */
> -       if (!done && info->fbops->fb_set_par)
> +       if (info->fbops->fb_set_par)
>                 info->fbops->fb_set_par(info);

Thanks for the tip.  I tried it, but it looks like the problem
is something else because I got exactly the same oops this time.
Something with the matrox second fb driver perhaps?  I guess few
people actually use that.

Helge Hafting