Floating point exception in Xorg

Floating point exception in Xorg

[Neil MacMunn]

I've started getting a floating point exception when gpe-dm starts Xorg 
or I start it manually. It only happens when I'm using a small display 
i.e. 4.3"LCD, works fine with DVI monitor. Here is the end of my Xorg.0.log

> (II) Loading /usr/lib/xorg/modules/drivers/omapfb_drv.so
> (II) Module omapfb: vendor="X.Org Foundation"
> 	compiled for 1.7.4, module version = 0.1.1
> 	ABI class: X.Org Video Driver, version 6.0
> (II) omapfb: Driver for OMAP framebuffer (omapfb) and external LCD controllers:
> 	omap1/2/3, S1D13745, HWA742
> (WW) Falling back to old probe method for OMAPFB
> (WW) VGA arbiter: cannot open kernel arbiter, no multi-card support
> (II) Running in FRAMEBUFFER Mode
> (WW) Error opening /sys/devices/platform/omapfb/ctrl/name: No such file or directory
> (WW) Can't autodetect LCD controller, assuming internal
> (II) LCD controller: internal
> (II) omapfb(0): Base plane capabilities:
> (II) omapfb(0): Base plane supports the following image formats:
> (II) omapfb(0): VideoRAM: 512KiB (SDRAM)
> (II) omapfb(0): Creating default Display subsection in Screen section
> 	"Screen0" for depth/fbbpp 24/32
> (--) omapfb(0): Depth 24, (==) framebuffer bpp 32
> (==) omapfb(0): RGB weight 888
> (==) omapfb(0): Default visual is TrueColor
> (II) omapfb(0): Output LCD using monitor section Monitor0
>
> Backtrace:
> 0: Xorg (xorg_backtrace+0x2c) [0x57ca0]
> Floating point exception at address 0x734
>
> Fatal server error:
> Caught signal 8 (Floating point exception). Server aborting

I ran a backtrace with GDB and got the following.

> #0  0x401820ac in raise () from /lib/libpthread.so.0
> #1  0x40661fbc in __div0 ()
>     at /home/neil/overo-oe/tmp-unstable/work/armv7a-angstrom-linux-gnueabi/gcc-cross-4.3.3-r23.2/gcc-4.3.3/libgcc/../gcc/config/arm/lib1funcs.asm:1079
> #2  0x40661f90 in __udivsi3 ()
>     at /home/neil/overo-oe/tmp-unstable/work/armv7a-angstrom-linux-gnueabi/gcc-cross-4.3.3-r23.2/gcc-4.3.3/libgcc/../gcc/config/arm/lib1funcs.asm:834
> #3  0x4065edc4 in OMAPFBOutputGetModes (output=<value optimized out>)
>     at omapfb-output.c:128
> #4  0x000f8c60 in xf86ProbeOutputModes (scrn=0x90a990, maxX=2048, maxY=2048)
>     at xf86Crtc.c:1575
> #5  0x000f95dc in xf86InitialConfiguration (scrn=0x2, canGrow=1)
>     at xf86Crtc.c:2345
> #6  0x4065e47c in OMAPFBPreInit (pScrn=0x90a990, flags=<value optimized out>)
>     at omapfb-driver.c:422
> #7  0x0006f6c0 in InitOutput (pScreenInfo=0x1, argc=7, argv=0x0)
>     at xf86Init.c:817
> #8  0x00021fbc in main (argc=7, argv=0x0, envp=<value optimized out>)
>     at main.c:204

I suspected that there was a change in Xorg or omapfb_drv.so so I 
replaced them with their counterparts from a working system to no 
effect. xorg.conf hasn't changed. I'm using a Gumstix Overo with desktop 
and palmtop images. Has anyone seen a floating point exception from Xorg 
or have any ideas where it might originate?

Thanks.

--
Neil

Re: Floating point exception in Xorg

[Phil Blundell]

On Fri, 2011-03-04 at 11:57 -0800, Neil MacMunn wrote:
> > #1  0x40661fbc in __div0 ()
> >     at /home/neil/overo-oe/tmp-unstable/work/armv7a-angstrom-linux-gnueabi/gcc-cross-4.3.3-r23.2/gcc-4.3.3/libgcc/../gcc/config/arm/lib1funcs.asm:1079
> > #2  0x40661f90 in __udivsi3 ()
> >     at /home/neil/overo-oe/tmp-unstable/work/armv7a-angstrom-linux-gnueabi/gcc-cross-4.3.3-r23.2/gcc-4.3.3/libgcc/../gcc/config/arm/lib1funcs.asm:834
> > #3  0x4065edc4 in OMAPFBOutputGetModes (output=<value optimized out>)
> >     at omapfb-output.c:128
> > #4  0x000f8c60 in xf86ProbeOutputModes (scrn=0x90a990, maxX=2048, maxY=2048)
> >     at xf86Crtc.c:1575

That's a division by zero in what looks like the mode selection code.
At a guess, some parameter which the server is relying on to calculate
the video timings is not being set correctly by the kernel.  You'd need
to inspect the source for OMAPFBOutputGetModes() to figure out what's
up, though.

p.

Re: Floating point exception in Xorg

[Neil MacMunn]

On 11-03-04 12:26 PM, Phil Blundell wrote:
> On Fri, 2011-03-04 at 11:57 -0800, Neil MacMunn wrote:
>>> #1  0x40661fbc in __div0 ()
>>>      at /home/neil/overo-oe/tmp-unstable/work/armv7a-angstrom-linux-gnueabi/gcc-cross-4.3.3-r23.2/gcc-4.3.3/libgcc/../gcc/config/arm/lib1funcs.asm:1079
>>> #2  0x40661f90 in __udivsi3 ()
>>>      at /home/neil/overo-oe/tmp-unstable/work/armv7a-angstrom-linux-gnueabi/gcc-cross-4.3.3-r23.2/gcc-4.3.3/libgcc/../gcc/config/arm/lib1funcs.asm:834
>>> #3  0x4065edc4 in OMAPFBOutputGetModes (output=<value optimized out>)
>>>      at omapfb-output.c:128
>>> #4  0x000f8c60 in xf86ProbeOutputModes (scrn=0x90a990, maxX=2048, maxY=2048)
>>>      at xf86Crtc.c:1575
>
> That's a division by zero in what looks like the mode selection code.
> At a guess, some parameter which the server is relying on to calculate
> the video timings is not being set correctly by the kernel.  You'd need
> to inspect the source for OMAPFBOutputGetModes() to figure out what's
> up, though.
>
> p.
>


Right. This line is the problem.

>  mode->Clock = PICOS2KHZ(ofb->state_info.pixclock);

because

>  #define PICOS2KHZ(a) (1000000000UL/(a))

I didn't have any luck figuring out where pixclock was supposed to be 
set so I just reverted xf86-video-omapfb_git.bb and my LCDs work again.

> git checkout 3b039e5c54265c5e1e254736687ad1cefed8f534 -- xf86-video-omapfb_git.bb

Thanks for the help.

--
Neil