ship apport ignore hook for usplash BIOS code crashes (SIGTRAP within vesa_setmode->LRMI_int->run_vm86)

StacktraceTop:run_vm86 () at lrmi.c:526
LRMI_int (i=16, r=0xb7f11cc0) at lrmi.c:844
vesa_setmode (mode=3, prv_mode=0) at /build/buildd/usplash-0.4/svgalib/src/vesa.c:213
initialize () at /build/buildd/usplash-0.4/svgalib/src/vga.c:2016
vga_setmode (mode=12) at /build/buildd/usplash-0.4/svgalib/src/vga.c:2245

<mjg59> Crashing in BIOS code? Turns out to be hard to fix.
<sladen> mjg59: can it be hooked, caught and just "failed" rather than leading to a crash?
<mjg59> No
<mjg59> Eh.
<mjg59> You could have a sigsegv handler.
<mjg59> But that is clearly not the right answer
<sladen> vesa_setmode() could catch SIGTRAP before lrmi is called and cleanly exit on failure/trap
<mjg59> I'm not sure how this is preferable
<mjg59> The code either crashes or it doesn't
<sladen> it's the equivalent of try { setmode() } except { printf("failed"); }
<mjg59> sladen: The code tried to write to memory it didn't own. It really should crash.
<sladen> rather than just dying.---in which case apport catches the crash and will want to report it
<mjg59> The right fix is to tell apport to ignore it, not to pretend we're not crashing when we are
<sladen> which is pointless if there's nothing to be done about it
<pitti> mjg59: you can actually do so now in 0.7
<mjg59> pitti: Score

<pitti> sladen: well, usplash can ship a package hook which sets a particular field
<pitti> sladen: that package hook can examine the current crash report and decide wehther or not to ignore it

Target to Feisty release so we don't get tonnes of bug reports from broken machines.

Matthew, Paul, I'm fine with writing the actual apport hook. Can you please describe exactly when it should be supressed? It should both match the nonsymbolic trace below and the attached apport retrace (which does not mention libx86.so.1 any more for some reason).

Perhaps 'First item in StacktraceTop contains libx86.so or run_vm86'?

Yup, anything inside 'run_vm86()' is probably proprietary BIOS code, otherwise we'd have better ways of running it.

If that's the way to match it (see Stacktraces attached) then go for it.

 usplash (0.4-44) feisty; urgency=low
 .
   [ Colin Watson ]
   * Add XS-Vcs-Bzr field to debian/control.
 .
   [ Martin Pitt ]
   * add debian/usplash.py: Apport package hook to check whether a crash
     happened in run_vm86(), which means a crash in the BIOS. We do not want
     crash reports about those, because they are useless. (LP: #94911)
   * debian/usplash.install: Install that hook into
     /usr/share/apport/package-hooks/.