Re: libpthread breaking various pkg builds

[Piotr Meyer <aniou%smutek.pl@localhost>]

On Sun, Dec 26, 2010 at 11:53:24AM -0500, Christos Zoulas wrote:
 
> I know. Any help debugging it is appreciated, but I think that
> in the end TLS will come to the rescue.

I think, in this case we hit protected pages due to change from
rev. 1.119 of src/lib/libpthread/pthread.c (mprotect base only
if we move it). Program faults in pthread__stackid_setup on write:

t->pt_stack.ss_size = size - 2 * pagesize;

Data from two faults (: marks my own debug from pthread__initmain
and pthread__stackid_setup):

stack 0xbfc00000 d=0x5ff000 0xbf601000
applying to 0xbbbf8000
: pthread__sp: bf5ff494
:: redaddr: 0xbf401000 base: 0xbf400000
[segfault]

stack 0xbfc00000 d=0xdfe000 0xbee02000
applying to 0xbbbf8000 orig_addr=0x0 f=1002
result 0xb78b1000
applying to 0xbbbf
: pthread__sp: bedffc80
:: redaddr: 0xbec01000 base: 0xbec00000 
[segfault]


...but with unconditionally mprotect we hit already mentioned
problem, when pthread_self() returns different bases for same
thread when stack pointer crosses boundaries, for example:

* sp: 0xbf800060 va: 0xbf800000 
* sp: 0xbf8001b0 va: 0xbf800000
* sp: 0xbf7ff3b0 va: 0xbf600000
* sp: 0xbf7ff394 va: 0xbf600000
[hang]

Regards,
-- 
Piotr 'aniou' Meyer