On Fri, Jun 17, 2005 at 09:29:48AM -0500, Brian Ford wrote: >On Wed, 8 Jun 2005, Christopher Faylor wrote: >>On Wed, Jun 08, 2005 at 10:36:56AM -0400, Christopher Faylor wrote: >>>I understand the problem. It's a binutils bug that I'm investigating. >>>Either gas or ld is marking NOLOAD sections as "read only" and that is >>>causing windows to actually load them into memory. >> >>There's a new "test" version of binutils available which seems to fix >>this problem. I'm awaiting approval for a patch in the binutils >>mailing list before making this version official. >> >>This turns out to have been a long-time bug in binutils, AFAICT. The >>debug sections that the GNU linker produces were not properly marked as >>"noload" so debug data was actually loaded into memory. > >I wish I would have known this was possible while working on DWARF 2 >debugging. Everything I read stated: "If it has a section header, >Windows' brain dead loader will load it regardless of section >attributes". I hope it really works on all supported platforms.
It's a documented field in PE's section header, so I don't see why it wouldn't work. The only brain-dead thing that the loader does is to go ahead and load a section into memory if it is marked as readable or writable as well as noload. >Do the RVA's for these sections still have to be consecutive? As far as I can tell, yes. Anything else seems to result in a non-loadable binary. But, I didn't investigate this too thoroughly so it is possible that there's another binutils bug lurking there. cgf -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Problem reports: http://cygwin.com/problems.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
