[CFR] Adding a function to rtld-elf.so, how to handle Symbol.map?

[Ian Lepore]

On Fri, 2014-07-18 at 16:36 +0300, Konstantin Belousov wrote:
> On Fri, Jul 18, 2014 at 07:23:59AM -0600, Ian Lepore wrote:
> > On Fri, 2014-07-18 at 11:14 +0300, Konstantin Belousov wrote:
> > > On Thu, Jul 17, 2014 at 06:17:41PM -0600, Ian Lepore wrote:
> > > > On Thu, 2014-07-17 at 20:29 +0300, Konstantin Belousov wrote:
> > > > > On Thu, Jul 17, 2014 at 11:09:50AM -0600, Ian Lepore wrote:
> > > > > > On Thu, 2014-07-17 at 03:45 +0300, Konstantin Belousov wrote:
> > > > > > > On Wed, Jul 16, 2014 at 03:23:53PM -0600, Ian Lepore wrote:
> > > > > > [snip]
> > > > 
> > > > > > I did some looking around and Netbsd, Android, and Risc Os all
> > > > > > implemented this in their dynamic loaders, so that seemed like the way
> > > > > > to go.  Android actually puts a function with this __gnu name in its
> > > > > > libc, but all that function does is calls dl_unwind_find_exidx() which
> > > > > > is implemented in their loader.
> > > > > > 
> > > > > > I've just discovered that the arm unwind support code that will arrive
> > > > > > as part of clang 3.5 appears to assume the Android way of things unless
> > > > > > __LINUX__ is defined, so maybe it would be good to follow that model
> > > > > > ourselves and add a dl_unwind_find_exidx() stub to libc/gen/dlfcn.c and
> > > > > > name the new implementation in ld-elf to match.
> > > > > I think that Android/__LINUX__ combination does the right thing, by
> > > > > providing the symbol in libc. A libc implementation does not need any
> > > > > additional service from rtld, except already existing _rtld_addr_phdr().
> > > > > 
> > > > 
> > > > Android provides a stub of dl_unwind_find_exidx() in libdl and the
> > > > shared-object implementation in the dynamic linker.  What it puts in
> > > > libc is the __gnu_Unwind_Find_exidx() symbol, which just calls through
> > > > to the dl_unwind_find_exidx() implementation in the dynamic linker.
> > > > 
> > > > That aside, I've reworked my code so it all lives in libc instead of
> > > > rtld, as you suggested.  It seems to work fine, and I guess I'm agnostic
> > > > about whether we're exporting a new function from libc versus rtld.  It
> > > > seems a bit strange to me to have just one dl_something() function with
> > > > its shared/dynamic implementation in libc, while all the other functions
> > > > with dl-prefix names are implemented in rtld.  But not so weird that
> > > > it's a big deal.
> > > The new patch is fine with me.  
> > > 
> > > Could you, please, comment why did you decided to export the
> > > dl_unwind_find_exidx alias ? It was absent in the original patch,
> > > and from your description, it seems to be an implementation detail
> > > on Linux.
> > 
> > I think you might have misunderstood what I said earlier.  According to
> > comments in some clang 3.5 sources I saw, the clang project considers
> > dl_unwind_find_exidx() to be "the BSD interface" for finding the exidx
> > data.  They fall back to the gnu name only when clang is compiled with
> > __LINUX__ defined.  By providing the functionality with this name, clang
> > 3.5 will just work right on freebsd without needing to be modified to
> > also use the gnu name when __FreeBSD__ is defined.
> > 
> > Android and Netbsd provide dl_unwind_find_exidx(); I haven't checked
> > other BSDs.  It certainly is a better name for an interface shared by
> > different toolchains than __gnu_Unwind_Find_exidx(), although we do need
> > to also provide that symbol for anyone using gcc.
> 
> Yes, I indeed misunderstood your description, thank you for the clarification.
> I.e. clang on Linux and gcc use __gnu_Unwind_Find_exidx(), while (future)
> clang on non-Linuxes uses dl_unwind_find_exidx() ?  And there is no
> ABI statement on the symbol, right ?

Right, except I don't understand what you are asking in your last
sentence.

-- Ian