summaryrefslogtreecommitdiff
path: root/libc/misc/internals/__errno_location.c
AgeCommit message (Collapse)Author
2014-04-23weak symbols need to be "defined" weak but "declared" strongVineet Gupta
Patch "LT.old: Make __errno_location/__h_errno_location thread safe" uncovered yet another bug with static linking and errno (hopefully this is last of them all). Currently, __errno_location is declared weak but is defined strong. While this provides with the desired weak semantics in dso, it is subtly broken in static links. Quoting Joern Rennecke (ARC gcc expert): | I think the issue is that you declare the function as weak in the | header file. That is a rare instance where you want the reference | use declaration that differs a bit from the definition. | If the reference uses a weakly declared function, that creates a | weakref, i.e. the linker won't bother to look for this symbol at | all - if it gets linked in for some other reason, fine, | otherwise, it stays zero. So the solution to declare strong, define weak. Supporting data ----------------- orig code: ARM mmap wrapper (LT.old build + my prev patch for errno) _mmap: @ args = 8, pretend = 0, frame = 0 @ frame_needed = 0, uses_anonymous_args = 0 stmfd sp!, {r4, r5, r7, lr} ldr r5, [sp, #20] movs ip, r5, asl #20 beq .L2 bl __errno_location(PLT) mov r3, #22 str r3, [r0, #0] mvn r0, #0 ... ... .weak __errno_location A statically linked hello world program which uses mmap too. As we can see__errno_location is completely gone - which is semantically wrong - we need functional errno. 00008274 <__GI_mmap>: 8274: e92d40b0 push {r4, r5, r7, lr} 8278: e59d5014 ldr r5, [sp, #20] 827c: e1b0ca05 lsls ip, r5, #20 8280: 0a000004 beq 8298 8284: e320f000 nop {0} ^^^^^^^^^^ 8288: e3a03016 mov r3, #22 828c: e5803000 str r3, [r0] 8290: e3e00000 mvn r0, #0 This in turn is due to a fixup in ARM ld which transforms branch-to-null into a nop. It is better than crashing but still wrong since errno handling is removed. With the patch, errno_location is restored back in test program. 00008274 <__GI_mmap>: 8274: e92d40b0 push {r4, r5, r7, lr} 8278: e59d5014 ldr r5, [sp, #20] 827c: e1b0ca05 lsls ip, r5, #20 8280: 0a000004 beq 8298 <__GI_mmap+0x24> 8284: eb000010 bl 82cc <__errno_location> 8288: e3a03016 mov r3, #22 828c: e5803000 str r3, [r0] Cc: Christian Ruppert <christian.ruppert@abilis.com> CC: Francois Bedard <Francois.Bedard@synopsys.com> Cc: Anton Kolesov <Anton.Kolesov@synopsys.com> Cc: Joern Rennecke <joern.rennecke@embecosm.com> Cc: Jeremy Bennett <jeremy.bennett@embecosm.com> Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com> Cc: Peter Korsgaard <peter@korsgaard.com> Cc: Khem Raj <raj.khem@gmail.com> Signed-off-by: Vineet Gupta <vgupta@synopsys.com> Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
2014-04-23LT.old: Make __errno_location/__h_errno_location thread safeVineet Gupta
WHY: errno in uClibc is not thread safe HOW: __errno_location and it's sibling __h_errno_location need to be called via PLT - even from within libc. That way when linked with pthread, intra-uClibc callers will also use the thread safe version. This is achieved by removing the GI alias for these functions, ensuring that they get called normally (via PLT) Verified with ARC LT.old and ARM cubieboard2 buildroot (LT.old) NPTL is unaffected by this bug. ------------ History behind this patch --------------------- This is a long standing bug (https://bugs.busybox.net/show_bug.cgi?id=2089) Others have tried to fix it in past (alteast Peter Korsgaard's patch in 2010), but somehow failed to be merged (or were backed out afterwards). http://lists.uclibc.org/pipermail/uclibc/2010-July/044176.html One of the causes could be side effect of atleast one more bug related to pthreads and static link which has now been fixed. http://lists.uclibc.org/pipermail/uclibc/2013-October/047958.html I have solved this w/o looking at other pacthes but would like to give credit to Peter and others for confirming that it makes sense. Cc: Christian Ruppert <christian.ruppert@abilis.com> CC: Francois Bedard <Francois.Bedard@synopsys.com> Cc: Joern Rennecke <joern.rennecke@embecosm.com> Cc: Jeremy Bennett <jeremy.bennett@embecosm.com> Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com> Cc: Peter Korsgaard <peter@korsgaard.com> Cc: Khem Raj <raj.khem@gmail.com> Cc: buildroot@busybox.net Signed-off-by: Vineet Gupta <vgupta@synopsys.com> Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
2013-11-06Fix weak/strong attribute of __errno_location and it's __GI aliasVineet Gupta
A simple statically linked hello world program was segfaulting for ARC in linuxthreads.old configuration (although the root casue applies cross-arch for NPTL as well as linuxthreads.old as described) The crash was due to branch to NULL in _stdio_init 0001026c <_stdio_init>: 1026c: push_s blink 1026e: st.a r13,[sp,-8] 10272: bl.d 0 --> supposed call to __errno_location The call was NOT getting patched to libc internal only alias __GI___errno_location, because it was weak while it's exported cousin, __errno_location was strong/normal. arc-linux-uclibc-nm libc/misc/internals/__errno_location.os 00000000 W __GI___errno_location 00000000 T __errno_location This is exactly opposite to what is expected. Quoting Peter S. Mazinger, commit 87936cd013041 "errno and *_init cleanup" | The rule adopted: | for enabled threads we make in libc the __GI_x() variants strong, x() weak | and (should) provide another strong x() in libpthread. | If threads are disabled, even the __GI_x() variants are weak. With the fix, we see the right settings as below 00000000 T __GI___errno_location 00000000 W __errno_location Note that problem won't show up in a static busybox build as it references errno and that seems to elide the issue. I can confirm the same/more issues with latest ARM buildroot builds w/o my fix. (1). linuxthreads.old (broken just like ARC) arm-linux-nm uclibc-snapshot/libc/misc/internals/__errno_location.os 00000000 W __GI___errno_location 00000000 T __errno_location But presumably the issue there is NOT catestrophic because ARM linker is likely smarter and patches a NOP instead of NULL branch. 00008388 <_stdio_init>: 8388: e92d4038 push {r3, r4, r5, lr} 838c: e320f000 nop {0} (2) NPTL build (exported version is not weak) 00000000 T __GI___errno_location 00000000 T __errno_location This causes a static link with libpthread and test program referencing errno to fail to link. #include <errno.h> int main(void) { printf("%d\n", errno); } arm-linux-gcc -static -pthread -o tst tst.o arm-buildroot-linux-uclibcgnueabi/sysroot/usr/lib/libc.a(__errno_location.os): In function `__errno_location': __errno_location.c:(.text+0x0): multiple definition of `__errno_location' arm-buildroot-linux-uclibcgnueabi/sysroot/usr/lib/libpthread.a (errno_location.os):errno_location.c:(.text+0x0): first defined here Signed-off-by: Vineet Gupta <vgupta@synopsys.com> Cc: Baruch Siach <baruch@tkos.co.il> Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
2012-06-15errno, h_errno: correct them for non-TLSPeter S. Mazinger
Move h_errno related stuff to separate file. Do not hide errno and h_errno for non-TLS, else it keeps being 0. Make __[h_]errno_location weak for non-TLS. Signed-off-by: Peter S. Mazinger <ps.m@gmx.net> Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
2012-06-15errno and *_init cleanupPeter S. Mazinger
Define a common view of __errno_location, __h_errno_location in common header and use that everywhere, __uClibc_main.c is no special. The rule adopted: for enabled threads we make in libc the __GI_x() variants strong, x() weak and (should) provide another strong x() in libpthread. If threads are disabled, even the __GI_x() variants are weak. _stdio_init,_stdio_term,_locale_init: make all hidden weak in common header Signed-off-by: Peter S. Mazinger <ps.m@gmx.net> Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
2010-07-06Revert "don't make __errno_location / __h_errno_location hidden"Austin Foxley
This reverts commit 0d6ee549bc86fd330672a79d9a87d2c3825eea67. We need to find a solution that will work in shared and static libraries Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
2010-07-05don't make __errno_location / __h_errno_location hiddenPeter Korsgaard
Closes #2089 (https://bugs.busybox.net/show_bug.cgi?id=2089) __errno_location / __h_errno_location access has to go through the PLT like malloc/free, so the linuxthread variants gets used instead when compiling with -pthread. Based on http://github.com/mat-c/uClibc/commit/328d392c54aa5dc2b8e7f398a419087de497de2b Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk> Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
2010-06-30nptl/errno: Use a separate __errno_location for libpthread.Khem Raj
Signed-off-by: Khem Raj <raj.khem@gmail.com>
2009-09-18trim Experimentally off and uncommented hiddenBernhard Reutner-Fischer
sed -i -e '/Experimentally off - /d' $(grep -rl "Experimentally off - " *) sed -i -e '/^\/\*[[:space:]]*libc_hidden_proto(/d' $(grep -rl "libc_hidden_proto" *) should be a nop Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
2008-11-20Last portion of libc_hidden_proto removal.Denis Vlasenko
Appears to build fine (several .configs tried)
2007-02-21redo errno handling so that it should work both in thread and non-threaded ↵Mike Frysinger
situations
2006-08-23sync with psm: update errno handling to be the same on all archesMike Frysinger
2006-02-02even the hidden version of the weaks in libc (as strong in libpthread) have ↵Peter S. Mazinger
to be weaks
2006-01-16Correct a typoPeter S. Mazinger
2006-01-16Last relocs jump and global data, (even locales) that I could remove are ↵Peter S. Mazinger
gone from libc. The remaining are left as exercise for others ;-)
2002-02-20Merge in the pthread library. This is the linuxthreads library taken fromEric Andersen
glibc 2.1.3 and ported to work with uClibc by Stefan Soucek and Erik Andersen (me). Stefan has hacked things up such that linuxthreads runs on MMU-less systems (tested only on arm-nommu). Erik cleaned things up and made it work properly as a shared library. -Erik