[uml-devel] UML/Debian hwclock bug identified.

[BlaisorBlade <bla...@ya...>]

Finally, with my new broadband connection, I downloaded the Debian root_fs and 
finally identified the patch (it's part of 2.4.24-2) containing the bug.

However, sadly, the patch causing it is the fix for the "I'm tracing myself 
and can't get out" bug. It is attached in two parts.

I've split the patch in two part, but however without part-2 the patch does 
not work *at all*: when starting /sbin/init glibc throws out some messages 
and IIRC hangs; probably there is an unfixed SIGSEGV, because it can't read 
the mmap()'ed memory.

The idea for splitting out the 2nd part is that it allows UML to get an 
infinite number of segfaults, in a loop, which can actually happen. Sadly, 
I've not yet been able to debug it (I've never debugged UML in TT mode).

In fact, with this patch applied, a compiler bug showed up: the gcc-3.3.2 
(hammer branch) included in Mandrake miscompiles UML. In TT mode it still 
works, but only if this patch is excluded.

Also, this patch is also applied on 2.6 kernels, so it's quite important to 
fix it.

I've also seen this comment, which does not make sense to me:

+       /* This is done because to allow SIGSEGV to be delivered inside a SEGV
+        * handler.  This can happen in copy_user, and if SEGV is disabled,
+        * the process will die.
+        */
+       if(sig == SIGSEGV)
+               change_sig(SIGSEGV, 1);
+

If SIGSEGV is blocked, it will just be queued and/or ignored (no more than one 
signal can stay in the queue); in the 2nd case, the process will get it 
again. Could you take a look at this issue, Jeff?
-- 
Paolo Giarrusso, aka Blaisorblade
Linux registered user n. 292729