svn commit: r224683 - head/lib/libthread_db

Marius Strobl marius at alchemy.franken.de
Sun Aug 7 12:33:59 UTC 2011


On Sun, Aug 07, 2011 at 12:22:02PM +0000, Bjoern A. Zeeb wrote:
> 
> On Aug 7, 2011, at 11:52 AM, Marius Strobl wrote:
> 
> > On Sat, Aug 06, 2011 at 07:35:49PM -0700, Marcel Moolenaar wrote:
> >> 
> >> On Aug 6, 2011, at 4:24 PM, Marius Strobl wrote:
> >> 
> >>> On Sun, Aug 07, 2011 at 01:06:48AM +0300, Andriy Gapon wrote:
> >>>> on 07/08/2011 00:41 Garrett Cooper said the following:
> >>>>> It's not just i386. It's other architectures like arm, mips, and pc98
> >>>>> according to the tinderbox reports (this list is potentially
> >>>>> incomplete).
> >>>> 
> >>>> Yeah, confusingly enough thr_pread_long() is declared to take uint64_t* as its
> >>>> third argument, so this commit breaks all platforms where uint64_t is not
> >>>> derived from (unsigned) long.
> >>>> Just in case, thr_pread_int() takes uint32_t* as well.
> >>>> 
> >>> 
> >>> Yes, the type of val is wrong. I'm currently running a fix through a
> >>> universe build
> >> 
> >> Ah, euh, ok, I forgot about this particular quirk:
> >> 
> >> uint64_t is deliberate, because it's the width of long on 64-bit
> >> architectures and thread_db is to be used as a support library
> >> for a cross-tool (i.e. a 32-bit debugger for a 64-bit target).
> >> 
> >> That's why thr_pread_long() takes a pointer to uint64_t and
> >> thr_pread_int() takes a pointer to uint32_t...
> >> 
> >> Sorry for missing this in the review.
> > 
> > Okay, but then I don't know how to properly fix this given that
> > thr_p{read,write}_long() still seem to do the wrong thing as they
> > supply sizeof(long) rather than the size of a long on the target
> > to thr_p{read,write}() as the size of the value in the target
> > address space. If I change the callers of thr_pread_long() to
> > supply a uint64_t this will compile but it still does the wrong
> > thing in the cross-debugging case and I can't even think of how
> > to fix that without additional information about the target, i.e.
> > just using sizeof(uint64_t) obviously also is the wrong thing.
> > Both thr_p{read,write}_ptr() are similarly confusing as they take
> > a psaddr_t which is defined as uint64_t but use sizeof(void *)
> > which again is specific to the host rather tan the target.
> > Do you have a suggestion how to fix these?
> 
> Given this, can you please back it out, find a proper solution together, test it and re-add that then?
> 

I'm currently running a universe build with lwp changed from long
to int64_t. This should compile and work on all architectures for
native-debugging and as such be a step in the right direction. In
a second step thr_p{read,write}_{long,ptr}() then can be fixed to
also do the right thing when cross-debugging.

Marius



More information about the svn-src-all mailing list