| Age | Commit message (Collapse) | Author |
|---|
|
Other arches may also benefit from this iff it can do
unaligned stores.
|
|
|
|
_dl_strchr,_dl_strrchr,_dl_strstr,_dl_memcmp:
Optimize for archs which can do pre increment/decrement and load/store
in one instruction.
|
|
handle "" strings and optimze it.
_dl_simple_ltoa,_dl_simple_ltoahex:
Optimize for archs which can do pre increment/decrement and load/store
in one instruction.
|
|
|
|
i was sticking some libraries in non-standard (i.e. not-hardcoded) paths and
then updating ld.so.conf accordingly ... oddly though, i couldnt get binaries
that linked against those libraries to run ... `ldconfig -v` showed the
required libraries being detected/cached, so something seemed wrong ...
turns out the libraries were being tagged as LIB_ELF_LIBC0 in the cache and
the loader currently only accepts LIB_ELF_LIBC{5,6} ... since readsoname.c in
ldconfig defines needed_tab with the mapping 'libc.so.0' -> LIB_ELF_LIBC0 it
seems like the loader should accept this type too
find attached a small patch which adds LIB_ELF_LIBC0 along side
LIB_ELF_LIBC{5,6} as an accepted lib type in the loader when dealing with the
ld.so.cache file
|
|
in libdl, the XXFLAGS is set incorrectly if DODEBUG=y ... it has 'I.' added to
it instead of '-I.' ... this of course pisses of gcc :)
find attached a patch that cleans up the XXFLAGS setting so as to minimize
this kind of thing ... when asked, mjn3 said it was a kosher idea ;)
and later writes:
hmm it seems ldso also utilizes USE_CACHE in the code (but LIBRARY_CACHE never
propogates down to the Makefiles)
so the fix is to add $(LIBRARY_CACHE) while setting $(XXFLAGS)
Applied with minor fixes and also fixed up ldso/Makefile to match
libdl/Makefile.
|
|
However RTLD_LOCAL still doesn't work. Everything is RTLD_GLOBAL.
|
|
PPC32, SPARC32/64 and S390 includes the PLT in its RELA size. This caused ldso
to always do unlazy relocation of the JMPRELs. This patch fixes it.
|
|
This patch makes -fpic work for PCC and optimzes the relcation by moving the cache
flushing stuff to JMP relocs only. Actually PPCs ldso can only handle small
GOT tables(<=8192 entries)anyhow, so it makes little sense to compile PPC with -fPIC.
libuClibc shrunk from 340724 to 330780 bytes with -fpic.
|
|
Some utilities, such as valgrind, have a legitimate reason to know the address
of the current brk. Since we know such utils will peek under our skirt, we
might as well give them what they expect and not use a gratuitously different
symbol name.
-Erik
|
|
rather than _dl_malloc, fprintf rather than _dl_printf, etc.
-Erik
|
|
|
|
|
|
|
|
tree. Bad boy, No doughnut.
-Erik
|
|
aligned pointer that may be aligned up to page_size. Also add _dl_free,
|
|
Hi yet again :)
in dl-startup.c when performing boot strap relocation the following test
exists to make sure that only "_dl_" symbols are relocated:
/* We only do a partial dynamic linking right now. The user
is not supposed to define any symbols that start with a
'_dl', so we can do this with confidence. */
if (!symname || !_dl_symbol(symname)) {
continue;
}
However on PPC(and the other archs as well I suspect) all symbols are
"_dl_" symbols so the test is never true. The test can be removed and the
whole loop simplified(smaller). This also makes it possible to
simplify elfinterp.c
This remove the scanning of ldso.so relocs, making relocation faster.
I have tested this on PPC and it works well.
Do you think this optimization will work for the other arches as well?
I can't see why not.
Jocke
* Tested on x86, arm, mipsel, and powerpc by Erik and works nicely
-Erik
|
|
only occur when shared libs are not compiled with -fPIC
|
|
|
|
shared libraries with -fPIC
|
|
Make old weak symbol handling work, if someone wants to
enable it.
|
|
Since MIPS don't have COPY relocs it seems apropiate to define
DL_NO_COPY_RELOCS. This will optimize dl_find_hash somewhat.
|
|
Glibc does not use COPY class for SHN_COMMOM in MIPS. Make uClibc ldso
the same.
|
|
dl-string.h references do_rem, but do_rem is a #define in <arch>/dl-sysdep.h
which is not included by dl-sysdep.h. This causes a problem in libdl:
In file included from ../../ldso/include/ldso.h:27, from libdl.c:33:
../../ldso/include/dl-string.h: In function `_dl_simple_ltoa':
../../ldso/include/dl-string.h:216: warning: implicit declaration of
function `do_rem'
Include dl-sysdep.h in dl-string.h before using do_rem.
|
|
Hello,
I managed to get ldso (and thus shared linking to uClibc) to work on
sparc (actually sparc64 kernel with 32-bit userspace), at least on
simple "hello world" program (more complex ones not tested).
Some notes on attached patch (against 0.9.26, would require some work
to apply on current CVS - but I tested 0.9.26, not CVS):
- ELF magic cannot be examined by _dl_strncmp so early, probably because of
string constant, like on ppc/mips/sh
(note that early SEND_STDERR still crashes when trying to do _dl_strlen
- I suppose that string constants require relocation; but adding
load_addr didn't help, just ELF header was displayed instead of crash)
- mmap() is syscall6 like on ppc/mips/sh, not old i386 mmap()
- for generic sparc (i.e. not sparcv8/sparcv9) gcc produces .udiv/.urem
calls for unsigned integer / and % operators - so these operations
must be avoided. I copied do_rem definition from arm header.
But / and % are used also in _dl_simple_ltoa() and
_dl_simple_ltoahex(); in ltoahex gcc optimizes it to shifts (but
I think it's safer to use shifts explicitly, not rely on
optimization...).
I changed % in ltoa to do_rem, but as there was no do_div definition,
I changed all "%d" specifiers to "%x" to avoid crashes (this changes
wouldn't be needed if _dl_simple_ltoa() were fixed to not use
division on sparc).
- "#define SOLARIS_COMPATIBLE" in ld_sysdep.h broke ldso on Linux
because of redefining _dl_linux_resolve only in some places (it was
still referenced in INIT_GOT before redefinition). So
_dl_linux_resolve redefinition should be moved before INIT_GOT
definition or removed.
- sparc64 kernel requires mmap() addresses to be aligned to 8192, not
4096, otherwise mmap() call failed
- reloc_entry must be shifted by 10, not 12 (I found similar operation
in glibc sources)
Aside of sparc-specific fixes:
- I moved some _dl_dprintf()s inside if(_dl_debug_*) conditions (to avoid
debugging messages when LD_DEBUG is not defined)
- it seems that there was possible off-by-one in ltoa and ltoahex?
they are called with char[22] as 1st argument, and then '\0' is stored
in local[22] (_before_ p decrementation)... or am I missing something?
If not, fix is included in patch.
|
|
Here are a number of minor changes to FR-V-specific bits of the uClibc
port:
- I've adjusted the definition of _dl_mmap to cope with the fact that
there will be a definition for an mmap2-only system.
- We don't have COPY relocs, so optimize the copy reloc-related code
away.
- Change the page size to 16KiB, to match the ABI spec, and not a
stale value I'd copied from a linker config file. Oops.
- Fix error handling in clone and vfork; parts of the changes by David
Howells <dhowells@redhat.com>
- Rearrange includes in crtreloc.c.
- Change the API of __syscall_error, to reduce code size.
- Improve __syscall_return in terms of code size, so as to enable
tail-calling of __syscall_error, at least within libc. Ideally,
__syscall_error should be hidden within libc.so, but this didn't
work because of libdl. I haven't looked into why, and figured I'd
leave it visible for now.
- Rename enumerators and macros in sys/ucontext.h to reduce namespace
pollution.
|
|
|
|
Oops, this adds the page alignment offset to the mprotected size.
Regards,
Brad
|
|
ld.so mprotect rwx .dyanmic segment for mips DT_DEBUG
Allow writing debug_addr into the .dynamic segment.
Even though the program header is marked RWE, the kernel
gives it to us rx.
|
|
useful syscall failure diagnostics.
|
|
hard coded 4096 instead of PAGE_SIZE. Because I'm an idiot.
|
|
by extracting the value from the ELF header.
|
|
If I change that ELF_RTYPE_CLASS_COPY to ELF_RTYPE_CLASS_PLT to tell
_dl_find_hash to ignore stubs when resolving undefined functions without
stubs, the dlopen tests all pass. dlopen gets a pointer to the libc.so
malloc instead of a pointer to the libpthread malloc stub. Yay! :-)
|
|
|
|
|
|
with __NR_mmap is not available (i.e. only __NR_mmap2)
|
|
|
|
Hi
I just noticed that gcc has an "b"(Address base register) operand that
will match all "r" registers but r0. It is a better fix then adding
r0 to the clobber list.
What do you think?
|
|
|
|
in libdl pointing to the local 'foobar' function is garbage. This cleans all
that up and makes the code much less horrible. Now it is only really really
ugly (which is a marked improvement),
|
|
white space cleanup and optimize the inner loop in dl_find_hash.
|
|
lookup on 'atexit' got me excited and I started fixing the same
problem elsewhere. Except these were correct as-was, and were
not a problem....
|
|
|
|
mips mostly working. From there, I tracked the bash failure to a
bad _dl_atexit address and Erik took it from there.
|
|
|
|
|
|
This patch adds code to uClibc to support a new ABI designed for the
FR-V architecture, that enables text segments of executables and
shared libraries to be shared by multiple processes on an OS such as
uClinux, that can run on FR-V processors without an MMU.
Patches for binutils and GCC have just been posted in the
corresponding mailing lists. The binutils patch was approved,
but there's one additional patch pending review, that I posted
this week. An updated GCC patch will be posted to
gcc-patches@gcc.gnu.org as soon as I complete testing (I used a
known-good compiler to test the uClibc patch below).
Since the existing dynamic loader code didn't support independent
relocation of segments, it required changes that were somewhat
extensive. I've added a number of new machine-specific macros to try
to keep the platform and ABI-specific details outside the generic
code. I hope this is not a problem.
|
|
|
|
Begin converting some big ugly macros to inline functions
instead
|
