]> sourceware.org Git - glibc.git/log
glibc.git
4 years agomanual: Put the istrerrorname_np and strerrordesc_np return type in braces
Adhemerval Zanella [Fri, 7 Aug 2020 20:14:49 +0000 (17:14 -0300)]
manual: Put the istrerrorname_np and strerrordesc_np return type in braces

Otherwise it is not rendered or indexed correctly.

4 years agoLinux: Use faccessat2 to implement faccessat (bug 18683)
Florian Weimer [Fri, 7 Aug 2020 20:06:59 +0000 (22:06 +0200)]
Linux: Use faccessat2 to implement faccessat (bug 18683)

This provides correct AT_EACCESS handling and also takes
Linux security modules into account.

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agomanual: Fix strerrorname_np and strerrordesc_np return type (BZ #26343)
Adhemerval Zanella [Fri, 7 Aug 2020 19:49:59 +0000 (16:49 -0300)]
manual: Fix strerrorname_np and strerrordesc_np return type (BZ #26343)

4 years agomath: Fix inaccuracy of j0f for x >= 2^127 when sin(x)+cos(x) is tiny
Paul Zimmermann [Fri, 7 Aug 2020 19:14:53 +0000 (16:14 -0300)]
math: Fix inaccuracy of j0f for x >= 2^127 when sin(x)+cos(x) is tiny

Checked on x86_64-linux-gnu and i686-linux-gnu.

4 years agoUpdate syscall lists for Linux 5.8.
Joseph Myers [Fri, 7 Aug 2020 14:38:43 +0000 (14:38 +0000)]
Update syscall lists for Linux 5.8.

Linux 5.8 has one new syscall, faccessat2.  Update syscall-names.list
and regenerate the arch-syscall.h headers with build-many-glibcs.py
update-syscalls.

Tested with build-many-glibcs.py.

4 years agoUse Linux 5.8 in build-many-glibcs.py.
Joseph Myers [Fri, 7 Aug 2020 14:38:12 +0000 (14:38 +0000)]
Use Linux 5.8 in build-many-glibcs.py.

This patch makes build-many-glibcs.py use Linux 5.8.

Tested with build-many-glibcs.py (host-libraries, compilers and glibcs
builds).

4 years agohtl: Enable tst-cancelx?[45]
Samuel Thibault [Thu, 6 Aug 2020 23:38:27 +0000 (23:38 +0000)]
htl: Enable tst-cancelx?[45]

* nptl/{tst-cancel4-common.c, tst-cancel4-common.h, tst-cancel4.c,
tst-cancel5.c, tst-cancelx4.c, tst-cancelx5.c}: Move to sysdeps/pthread/
* nptl/Makefile: Move corresponding rules to...
* sysdeps/pthread/Makefile: ... here.

4 years agotst-cancel4: Make blocking on write more portable
Samuel Thibault [Thu, 6 Aug 2020 23:13:17 +0000 (01:13 +0200)]
tst-cancel4: Make blocking on write more portable

* nptl/tst-cancel4.c (tf_send, tf_sendto): Set socket buffer size after
connecting.

4 years agohurd: Add missing hidden def
Samuel Thibault [Thu, 6 Aug 2020 18:10:58 +0000 (20:10 +0200)]
hurd: Add missing hidden def

* sysdeps/mach/hurd/sched_gets.c (__sched_getscheduler): Add hidden def.

4 years agohurd: Rework sbrk
Samuel Thibault [Wed, 5 Aug 2020 21:48:58 +0000 (23:48 +0200)]
hurd: Rework sbrk

Making the brk start exactly at the end of the main application binary was
requiring to get it through the _end symbol, which does not work any more
with recent toolchains, and actually produces in libc.so a confusing
external _end symbol that produces odd results, see
https://sourceware.org/bugzilla/show_bug.cgi?id=23499

Trying to do so is quite outdated anyway with the tendency for address
randomization.

Using _end was also allowing to include the main binary data within
the RLIMIT_DATA, but this also seems outdated with dynamic library
loading, and nowadays' memory consumption via malloc and mmap rather than
statically-allocated data.

This adds a BRK_START macro in <vm_param.h> that just tells where we
want to start the brk, and thus removes the _end symbol.

* sysdeps/mach/hurd/i386/vm_param.h: New file.
* sysdeps/mach/hurd/brk.c: Use BRK_START as brk start instead of _end.
Also ignore __data_start.
* hurd/Versions: Remove _end symbol.
* sysdeps/mach/hurd/i386/libc.abilist: Remove _end symbol.

4 years agohurd: Implement basic sched_get/setscheduler
Samuel Thibault [Wed, 5 Aug 2020 21:46:14 +0000 (23:46 +0200)]
hurd: Implement basic sched_get/setscheduler

* sysdeps/mach/hurd/sched_gets.c: New file.
* sysdeps/mach/hurd/sched_sets.c: New file.

4 years agox86: Rename Intel CPU feature names
H.J. Lu [Wed, 5 Aug 2020 15:20:52 +0000 (08:20 -0700)]
x86: Rename Intel CPU feature names

Intel64 and IA-32 Architectures Software Developer’s Manual has changed
the following CPU feature names:

1. The CPU feature of Enhanced Intel SpeedStep Technology is renamed
from EST to EIST.
2. The CPU feature which supports Platform Quality of Service Monitoring
(PQM) capability is changed to Intel Resource Director Technology
(Intel RDT) Monitoring capability, i.e. PQM is renamed to RDT_M.
3. The CPU feature which supports Platform Quality of Service
Enforcement (PQE) capability is changed to Intel Resource Director
Technology (Intel RDT) Allocation capability, i.e. PQE is renamed to
RDT_A.

4 years agomanual: Fix some @code/@var formatting glitches chapter Date And Time
Florian Weimer [Mon, 9 Sep 2019 09:21:08 +0000 (11:21 +0200)]
manual: Fix some @code/@var formatting glitches chapter Date And Time

4 years agoCopy regex_internal.h from Gnulib
Paul Eggert [Wed, 5 Aug 2020 06:45:27 +0000 (23:45 -0700)]
Copy regex_internal.h from Gnulib

Sync this file from Gnulib, thus incorporating the following
fix for a bug with regexps with 16 or more subexpressions:
* posix/regex_internal.h (struct re_backref_cache_entry):
Use bitset_word_t as the type of eps_reachable_subexps_map,
instead of unsigned short int.  This fixes a bug I introduced
to glibc in 2005-09-28T17:33:18Z!drepper@redhat.com (glibc commit
2c05d33f90861d074dc12808dafbde30f487b1a0, BZ #1302).
Remove unused member 'unused'.

4 years agoCopy regex BITSET_WORD_BITS porting from Gnulib
Paul Eggert [Wed, 5 Aug 2020 06:39:20 +0000 (23:39 -0700)]
Copy regex BITSET_WORD_BITS porting from Gnulib

* posix/regex.c (__STDC_WANT_IEC_60559_BFP_EXT__):
Define, for ULONG_WIDTH.  This syncs regex.c from Gnujlib.
* posix/regex_internal.h (ULONG_WIDTH):
Use a more-portable fallback, from Gnulib.
(BITSET_WORD_BITS): Now defined in terms of ULONG_WIDTH.

4 years agoSync regex.h from Gnulib
Paul Eggert [Wed, 5 Aug 2020 06:24:03 +0000 (23:24 -0700)]
Sync regex.h from Gnulib

* posix/regex.h: Remove an ‘#ifndef _CRAY’ that hasn’t been needed
for years in Gnulib (and was needed only because of Gnulib).

4 years agoSync mktime.c from Gnulib
Paul Eggert [Wed, 5 Aug 2020 06:15:31 +0000 (23:15 -0700)]
Sync mktime.c from Gnulib

* time/mktime.c: Sync from Gnulib.
This micro-optimizes three division-related computations.

4 years agoSync intprops.h from Gnulib
Paul Eggert [Wed, 5 Aug 2020 05:58:58 +0000 (22:58 -0700)]
Sync intprops.h from Gnulib

* include/intprops.h: Sync from Gnulib.  This improves
performance of INT_MULTIPLY_WRAPV on recent GCC, which affects
glibc only in the support library.

4 years agoOpen master branch for glibc 2.33 development. glibc-2.32.9000
Carlos O'Donell [Wed, 5 Aug 2020 03:41:43 +0000 (23:41 -0400)]
Open master branch for glibc 2.33 development.

Happy hacking!

4 years agoPrepare for glibc 2.32 release. glibc-2.32
Carlos O'Donell [Wed, 5 Aug 2020 02:17:00 +0000 (22:17 -0400)]
Prepare for glibc 2.32 release.

Update version.h, features.h, and ChangeLog.old/ChangeLog.21.

4 years agoRegenerate configure scripts.
Carlos O'Donell [Wed, 5 Aug 2020 01:36:19 +0000 (21:36 -0400)]
Regenerate configure scripts.

4 years agoUpdate NEWS with bugs.
Carlos O'Donell [Wed, 5 Aug 2020 00:39:09 +0000 (20:39 -0400)]
Update NEWS with bugs.

4 years agoUpdate translations.
Carlos O'Donell [Wed, 5 Aug 2020 00:31:51 +0000 (20:31 -0400)]
Update translations.

Incorporate updates from translationproject.org.

4 years agoDon't mix linker error messages into edited scripts
Alan Modra [Wed, 29 Jul 2020 06:58:57 +0000 (16:28 +0930)]
Don't mix linker error messages into edited scripts

* Makerules (shlib.lds): Discard linker warning output.
(format.lds): Likewise.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agobenchtests/README update.
Paul Zimmermann [Tue, 4 Aug 2020 11:27:39 +0000 (13:27 +0200)]
benchtests/README update.

Improve documentation of the 'name' directive and the 'workload' mechanism.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoRISC-V: Update lp64d libm-test-ulps according to HiFive Unleashed
Maciej W. Rozycki [Tue, 4 Aug 2020 12:00:17 +0000 (13:00 +0100)]
RISC-V: Update lp64d libm-test-ulps according to HiFive Unleashed

Produced with HiFive Unleashed hardware using Linux 5.8-rc5 exactly and
GCC 10.0.1 20200426.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoaarch64: update NEWS about branch protection
Szabolcs Nagy [Wed, 29 Jul 2020 07:27:27 +0000 (08:27 +0100)]
aarch64: update NEWS about branch protection

After some discussions it seems the original news was not clear
and that it is valid to manually pass the branch protection flags
iff GCC target libs are built with them too. The main difference
between manually passing the flags and using the configure
option is that the latter also makes branch protection the
default in GCC which may not be desirable in some cases.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoAdd NEWS entry for CVE-2016-10228 (bug 19519)
Aurelien Jarno [Thu, 30 Jul 2020 08:07:33 +0000 (10:07 +0200)]
Add NEWS entry for CVE-2016-10228 (bug 19519)

4 years agopowerpc: Fix incorrect cache line size load in memset (bug 26332)
Florian Weimer [Mon, 3 Aug 2020 16:07:19 +0000 (18:07 +0200)]
powerpc: Fix incorrect cache line size load in memset (bug 26332)

__GLRO loaded the word after the requested variable on big-endian
PowerPC, where LOWORD is 4.  This can cause the memset implement
go wrong because the masking with the cache line size produces
wrong results, particularly if the loaded value happens to be 1.

The __GLRO macro is not used in any place where loading the lower
32-bit word of a 64-bit value is desired, so the +4 offset is always
wrong.

Fixes commit 18363b4f010da9ba459b13310b113ac0647c2fcc
("powerpc: Move cache line size to rtld_global_ro") and bug 26332.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoUpdate Nios II libm-test-ulps file.
Chung-Lin Tang [Mon, 3 Aug 2020 08:42:48 +0000 (01:42 -0700)]
Update Nios II libm-test-ulps file.

4 years agoMove NEWS entry for CVE-2020-1751 to the 2.31 section
Florian Weimer [Fri, 31 Jul 2020 10:07:06 +0000 (12:07 +0200)]
Move NEWS entry for CVE-2020-1751 to the 2.31 section

It was fixed in commit d93769405996dfc11d216ddbe415946617b5a494
("Fix array overflow in backtrace on PowerPC (bug 25423)"), which
went into glibc 2.31.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoNEWS: Deprecate weak libpthread symbols for single-threaded checks
Florian Weimer [Fri, 24 Jul 2020 14:50:38 +0000 (16:50 +0200)]
NEWS: Deprecate weak libpthread symbols for single-threaded checks

Recommend the new __libc_single_thread variable instead.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoNEWS: Deprecate nss_hesiod
Florian Weimer [Fri, 24 Jul 2020 14:46:23 +0000 (16:46 +0200)]
NEWS: Deprecate nss_hesiod

Storing user databases in DNS, without client-side DNSSEC validation,
is problematic from a security point of view.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agonptl: Zero-extend arguments to SETXID syscalls [BZ #26248]
H.J. Lu [Thu, 16 Jul 2020 10:37:10 +0000 (03:37 -0700)]
nptl: Zero-extend arguments to SETXID syscalls [BZ #26248]

nptl has

/* Opcodes and data types for communication with the signal handler to
   change user/group IDs.  */
struct xid_command
{
  int syscall_no;
  long int id[3];
  volatile int cntr;
  volatile int error;
};

 /* This must be last, otherwise the current thread might not have
     permissions to send SIGSETXID syscall to the other threads.  */
  result = INTERNAL_SYSCALL_NCS (cmdp->syscall_no, 3,
                                 cmdp->id[0], cmdp->id[1], cmdp->id[2]);

But the second argument of setgroups syscal is a pointer:

       int setgroups (size_t size, const gid_t *list);

But on x32, pointers passed to syscall must have pointer type so that
they will be zero-extended.  The kernel XID arguments are unsigned and
do not require sign extension.  Change xid_command to

struct xid_command
{
  int syscall_no;
  unsigned long int id[3];
  volatile int cntr;
  volatile int error;
};

so that all arguments are zero-extended.  A testcase is added for x32 and
setgroups returned with EFAULT when running as root without the fix.

4 years agoUse binutils 2.35 branch in build-many-glibcs.py.
Joseph Myers [Mon, 27 Jul 2020 14:55:10 +0000 (14:55 +0000)]
Use binutils 2.35 branch in build-many-glibcs.py.

This patch makes build-many-glibcs.py use binutils 2.35 branch.

Tested with build-many-glibcs.py (compilers and glibcs builds).

4 years agoaarch64: Use future HWCAP2_MTE in ifunc resolver
Szabolcs Nagy [Fri, 24 Jul 2020 11:37:23 +0000 (12:37 +0100)]
aarch64: Use future HWCAP2_MTE in ifunc resolver

Make glibc MTE-safe on systems where MTE is available. This allows
using heap tagging with an LD_PRELOADed malloc implementation that
enables MTE. We don't document this as guaranteed contract yet, so
glibc may not be MTE safe when HWCAP2_MTE is set (older glibcs
certainly aren't). This is mainly for testing and debugging.

The HWCAP flag is not exposed in public headers until Linux adds it
to its uapi. The HWCAP value reservation will be in Linux 5.9.

4 years agoUpdate x86-64 libm-test-ulps
Andreas K. Hüttel [Sat, 25 Jul 2020 09:36:00 +0000 (12:36 +0300)]
Update x86-64 libm-test-ulps

x86_64 Intel(R) Core(TM) i5-8265U
gcc (Gentoo 10.1.0-r2 p3) 10.1.0
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoaarch64: Respect p_flags when protecting code with PROT_BTI
Szabolcs Nagy [Mon, 13 Jul 2020 10:28:18 +0000 (11:28 +0100)]
aarch64: Respect p_flags when protecting code with PROT_BTI

Use PROT_READ and PROT_WRITE according to the load segment p_flags
when adding PROT_BTI.

This is before processing relocations which may drop PROT_BTI in
case of textrels.  Executable stacks are not protected via PROT_BTI
either.  PROT_BTI is hardening in case memory corruption happened,
it's value is reduced if there is writable and executable memory
available so missing it on such memory is fine, but we should
respect the p_flags and should not drop PROT_WRITE.

4 years agoDisable warnings due to deprecated libselinux symbols used by nss and nscd
Arjun Shankar [Thu, 23 Jul 2020 10:20:38 +0000 (12:20 +0200)]
Disable warnings due to deprecated libselinux symbols used by nss and nscd

The SELinux API deprecated several symbols in its 3.1 release, including
security_context_t, matchpathcon, avc_init, and sidput, which are used in
makedb and nscd.  While the usage of these should eventually be replaced by
newer interfaces, this commit disables GCC warnings due to the use of the
above symbols.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
Tested-by: Carlos O'Donell <carlos@redhat.com>
4 years agoRegenerate INSTALL for ARC port updates.
Carlos O'Donell [Thu, 23 Jul 2020 03:32:07 +0000 (23:32 -0400)]
Regenerate INSTALL for ARC port updates.

4 years agoUpdate libc.pot for 2.32 release.
Carlos O'Donell [Thu, 23 Jul 2020 03:28:05 +0000 (23:28 -0400)]
Update libc.pot for 2.32 release.

4 years agopowerpc: Fix POWER10 selection
Tulio Magno Quites Machado Filho [Tue, 21 Jul 2020 21:01:39 +0000 (18:01 -0300)]
powerpc: Fix POWER10 selection

Add a line that was missing from a previous commit.
Without increasing str, the null-byte is not validated, and
_dl_string_platform returns -1.

Fixes: d2ba3677da7a ("powerpc: Add support for POWER10")
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agopowerpc64le: guarantee a .gnu.attributes section [BZ #26220]
Paul E. Murphy [Wed, 8 Jul 2020 21:49:34 +0000 (16:49 -0500)]
powerpc64le: guarantee a .gnu.attributes section [BZ #26220]

Upstream GCC 11 development is now building the ibm128 runtime
support (in libgcc) without a .gnu.attributes section on ppc64le.
Ensure we have one to replace by building one ibm128 file in
libc and libm with attributes.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Tulio Magno Quites Machado Filho <tuliom@linux.ibm.com>
4 years agolibio: Remove __libc_readline_unlocked
Florian Weimer [Wed, 15 Jul 2020 10:37:01 +0000 (12:37 +0200)]
libio: Remove __libc_readline_unlocked

__nss_readline supersedes it.  This reverts part of commit
3f5e3f5d066dcffb80af48ae2cf35a01a85a8f10 ("libio: Implement
internal function __libc_readline_unlocked").  The internal
aliases __fseeko64 and __ftello64 are preserved because
they are needed by __nss_readline as well.

Tested-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoshadow: Implement fgetspent_r using __nss_fgetent_r
Florian Weimer [Thu, 16 Jul 2020 15:34:43 +0000 (17:34 +0200)]
shadow: Implement fgetspent_r using __nss_fgetent_r

Tested-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agopwd: Implement fgetpwent_r using __nss_fgetent_r
Florian Weimer [Thu, 16 Jul 2020 15:34:19 +0000 (17:34 +0200)]
pwd: Implement fgetpwent_r using __nss_fgetent_r

Tested-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agogshadow: Implement fgetsgent_r using __nss_fgetent_r (bug 20338)
Florian Weimer [Thu, 16 Jul 2020 15:31:20 +0000 (17:31 +0200)]
gshadow: Implement fgetsgent_r using __nss_fgetent_r (bug 20338)

Tested-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agogrp: Implement fgetgrent_r using __nss_fgetent_r
Florian Weimer [Thu, 16 Jul 2020 15:28:28 +0000 (17:28 +0200)]
grp: Implement fgetgrent_r using __nss_fgetent_r

Tested-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agonss: Add __nss_fgetent_r
Florian Weimer [Wed, 15 Jul 2020 11:41:31 +0000 (13:41 +0200)]
nss: Add __nss_fgetent_r

And helper functions __nss_readline, __nss_readline_seek,
 __nss_parse_line_result.

This consolidates common code for handling overlong lines and
parse files.  Use the new functionality in internal_getent
in nss/nss_files/files-XXX.c.

Tested-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agolibio: Add fseterr_unlocked for internal use
Florian Weimer [Thu, 16 Jul 2020 15:00:46 +0000 (17:00 +0200)]
libio: Add fseterr_unlocked for internal use

Tested-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agonss_files: Use generic result pointer in parse_line
Florian Weimer [Thu, 16 Jul 2020 14:55:45 +0000 (16:55 +0200)]
nss_files: Use generic result pointer in parse_line

As a result, all parse_line functions have the same prototype, except
for that producing struct hostent.  This change is ABI-compatible, so
it does not alter the internal GLIBC_PRIVATE ABI (otherwise we should
probably have renamed the exported functions).

A future change will use this to implement a generict fget*ent_r
function.

Tested-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agonss_files: Consolidate line parse declarations in <nss_files.h>
Florian Weimer [Thu, 16 Jul 2020 14:40:44 +0000 (16:40 +0200)]
nss_files: Consolidate line parse declarations in <nss_files.h>

These functions should eventually have the same type, so it makes
sense to declare them together.

Tested-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agonss_compat: Do not use mmap to read database files (bug 26258)
Florian Weimer [Thu, 16 Jul 2020 14:21:28 +0000 (16:21 +0200)]
nss_compat: Do not use mmap to read database files (bug 26258)

This avoids crashes in case the files are truncated for some reason.
For typically file sizes, it is also going to be slightly faster.
Using __nss_files_fopen instead mirrors what nss_files does.

Tested-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agonss_files: Consolidate file opening in __nss_files_fopen
Florian Weimer [Thu, 16 Jul 2020 14:12:46 +0000 (16:12 +0200)]
nss_files: Consolidate file opening in __nss_files_fopen

Tested-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoUpdate powerpc-nofpu libm-test-ulps.
Joseph Myers [Mon, 20 Jul 2020 20:16:25 +0000 (20:16 +0000)]
Update powerpc-nofpu libm-test-ulps.

4 years agoUse MPFR 4.1.0 in build-many-glibcs.py.
Joseph Myers [Mon, 20 Jul 2020 18:16:50 +0000 (18:16 +0000)]
Use MPFR 4.1.0 in build-many-glibcs.py.

This patch makes build-many-glibcs.py use the new MPFR 4.1.0 release.

Tested with build-many-glibcs.py (host-libraries, compilers and glibcs
builds).

4 years agoelf: Change TLS static surplus default back to 1664
Florian Weimer [Mon, 20 Jul 2020 11:30:45 +0000 (13:30 +0200)]
elf: Change TLS static surplus default back to 1664

Make the computation in elf/dl-tls.c more transparent, and add
an explicit test for the historic value.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agohurd: Fix longjmp check for sigstate
Samuel Thibault [Sat, 18 Jul 2020 13:12:56 +0000 (15:12 +0200)]
hurd: Fix longjmp check for sigstate

* sysdeps/mach/hurd/i386/____longjmp_chk.S,__longjmp.S: Properly check for
sigstate being NULL.

4 years agohurd: Fix longjmp early in initialization
Samuel Thibault [Sat, 18 Jul 2020 13:08:03 +0000 (15:08 +0200)]
hurd: Fix longjmp early in initialization

When e.g. an LD_PRELOAD fails, _dl_signal_exception/error longjmps, but TLS
is not initialized yet, let along signal state.  We thus mustn't look at
them within __longjmp.

* sysdeps/mach/hurd/i386/____longjmp_chk.S,__longjmp.S: Check for
initialized value of %gs, and that sigstate is non-NULL.

4 years agomanual: New signal and errno string functions are AS-safe
Florian Weimer [Fri, 17 Jul 2020 15:13:59 +0000 (17:13 +0200)]
manual: New signal and errno string functions are AS-safe

The annotations for sigabbrev_np, sigdescr_np, strerrordesc_np,
strerrorname_np are not preliminary.  These functions were
added precisely because they are AS-safe.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoAArch64: Improve strlen_asimd performance (bug 25824)
Wilco Dijkstra [Fri, 17 Jul 2020 13:09:36 +0000 (14:09 +0100)]
AArch64: Improve strlen_asimd performance (bug 25824)

Optimize strlen using a mix of scalar and SIMD code.  On modern micro
architectures large strings are 2.6 times faster than existing
strlen_asimd and 35% faster than the new MTE version of strlen.

On a random strlen benchmark using small sizes the speedup is 7% vs
strlen_asimd and 40% vs the MTE strlen.  This fixes the main strlen
regressions on Cortex-A53 and other cores with a simple Neon unit.

Rename __strlen_generic to __strlen_mte, and select strlen_asimd when
MTE is not enabled (this is waiting on support for a HWCAP_MTE bit).

This fixes big-endian bug 25824. Passes GLIBC regression tests.

Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
4 years agoMove <rpc/netdb.h> from sunrpc to inet
Florian Weimer [Fri, 17 Jul 2020 13:19:35 +0000 (15:19 +0200)]
Move <rpc/netdb.h> from sunrpc to inet

Restore <rpc/netdb.h> as an installed header. Delete the dummy header
resolv/rpc/netdb.h because inet is not an optional glibc component
(so its <rpc/netdb.h> is always available).

Fixes commit acb527929d0c2b3bb0798472c42ddb3203729708 ("Move
non-deprecated RPC-related functions from sunrpc to inet") in
combination with commit 5500cdba4018ddbda7909bc7f4f9718610b43cf0
("Remove --enable-obsolete-rpc configure flag").

4 years agoen_US: Minimize changes to date_fmt (Bug 25923)
Carlos O'Donell [Wed, 8 Jul 2020 19:58:06 +0000 (15:58 -0400)]
en_US: Minimize changes to date_fmt (Bug 25923)

In 2000 when date_fmt was originally added as an extension the
en_US locale did not have a date_fmt specifier and so used the
default which resulted in the abbreviated month name coming
before the day of the month (as expected in the US and other
locales).  In commit 7395f3a0efad9fc51bb54fa383ef6524702e0c49 the
date_fmt was added to en_US with a 12H time to better align with
US user expectations.  Unfortunately the abbreviated month name
and day were inverted during that transition, and that was seen
as a regression and reported against Fedora 32:
https://bugzilla.redhat.com/show_bug.cgi?id=1830623

The progression of date_fmt looks like this:
"%a %b %e %H:%M:%S %Z %Y"    <- Originally (2000)
"%a %d %b %Y %I:%M:%S %p %Z" <- glibc 2.29 (2019)
"%a %b %e %r %Z %Y"          <- glibc 2.32 (2020) [this commit]

Note: "%r" is "%I:%M:%S %p" in en_US and so shorter to write.

Likewise the year is in the wrong place in commit
7395f3a0efad9fc51bb54fa383ef6524702e0c49 and this is corrected in
this patch.

For reference d_t_fmt:
"%a %d %b %Y %r %Z"          <- d_t_fmt    (1997)

Yes, d_t_fmt and date_fmt are *not* the same, this is just the
history of this locale. This commit does not change d_t_fmt to
better align with date_fmt. No users have requested we change
d_t_fmt or given any justification for such a change.

The only goals of this change are to place the abbreviated month
name before the day of the month as it has been printed since
2000, and place the year at the end. This minimizes the change
from commit 7395f3a0efad9fc51bb54fa383ef6524702e0c49 and makes
good on changing only from 24H clock to 12H clock.

Reviewed-by: Florian Weimer <fweimer@redhat.com>
4 years agoLinux: Remove rseq support
Florian Weimer [Thu, 16 Jul 2020 15:55:35 +0000 (17:55 +0200)]
Linux: Remove rseq support

The kernel ABI is not finalized, and there are now various proposals
to change the size of struct rseq, which would make the glibc ABI
dependent on the version of the kernels used for building glibc.
This is of course not acceptable.

This reverts commit 48699da1c468543ade14777819bd1b4d652709de ("elf:
Support at least 32-byte alignment in static dlopen"), commit
8f4632deb3545b2949cec5454afc3cb21a0024ea ("Linux: rseq registration
tests"), commit 6e29cb3f61ff5432c78a1c84b0d9b123a350ab36 ("Linux: Use
rseq in sched_getcpu if available"), and commit
0c76fc3c2b346dc5401dc055d97d4279632b0fb3 ("Linux: Perform rseq
registration at C startup and thread creation"), resolving the conflicts
introduced by the ARC port and the TLS static surplus changes.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agomanual: Use Unicode instead HTML entities for characters (bug 19737)
Florian Weimer [Thu, 16 Jul 2020 08:17:31 +0000 (10:17 +0200)]
manual: Use Unicode instead HTML entities for characters (bug 19737)

Texinfo no longer treats arguments to @set in @ifhtml blocks as
literal HTML, so the & in the entity references was encoded as
@amp; in HTML.  Using the equivalent Unicode characters avoids
this issue.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
Tested-by: Carlos O'Donell <carlos@redhat.com>
4 years agoAdd NEWS entry for CVE-2020-6096 (bug 25620)
Aurelien Jarno [Sun, 12 Jul 2020 19:58:43 +0000 (21:58 +0200)]
Add NEWS entry for CVE-2020-6096 (bug 25620)

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoarm: remove string/tst-memmove-overflow XFAIL
Aurelien Jarno [Mon, 13 Jul 2020 20:37:41 +0000 (22:37 +0200)]
arm: remove string/tst-memmove-overflow XFAIL

The arm string/tst-memmove-overflow XFAIL has been added in commit
eca1b233322 ("arm: XFAIL string/tst-memmove-overflow due to bug 25620")
as a way to reproduce the reported bug.

Now that this bug has been fixed in commits 79a4fa341b8 ("arm:
CVE-2020-6096: fix memcpy and memmove for negative length [BZ #25620]")
and beea3610507 ("arm: CVE-2020-6096: Fix multiarch memcpy for negative
length [BZ #25620]"), let's remove the XFAIL.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoAArch64: Rename IS_ARES to IS_NEOVERSE_N1
Wilco Dijkstra [Wed, 15 Jul 2020 15:58:07 +0000 (16:58 +0100)]
AArch64: Rename IS_ARES to IS_NEOVERSE_N1

Rename IS_ARES to IS_NEOVERSE_N1 since that is a bit clearer.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoAArch64: Add optimized Q-register memcpy
Wilco Dijkstra [Wed, 15 Jul 2020 15:55:07 +0000 (16:55 +0100)]
AArch64: Add optimized Q-register memcpy

Add a new memcpy using 128-bit Q registers - this is faster on modern
cores and reduces codesize.  Similar to the generic memcpy, small cases
include copies up to 32 bytes.  64-128 byte copies are split into two
cases to improve performance of 64-96 byte copies.  Large copies align
the source rather than the destination.

bench-memcpy-random is ~9% faster than memcpy_falkor on Neoverse N1,
so make this memcpy the default on N1 (on Centriq it is 15% faster than
memcpy_falkor).

Passes GLIBC regression tests.

Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
4 years agoAArch64: Align ENTRY to a cacheline
Wilco Dijkstra [Wed, 15 Jul 2020 15:50:02 +0000 (16:50 +0100)]
AArch64: Align ENTRY to a cacheline

Given almost all uses of ENTRY are for string/memory functions,
align ENTRY to a cacheline to simplify things.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoCorrect timespec implementation [BZ #26232]
H.J. Lu [Mon, 13 Jul 2020 23:15:56 +0000 (16:15 -0700)]
Correct timespec implementation [BZ #26232]

commit 04deeaa9ea74b0679dfc9d9155a37b6425f19a9f
Author: Lucas A. M. Magalhaes <lamm@linux.ibm.com>
Date:   Fri Jul 10 19:41:06 2020 -0300

    Fix time/tst-cpuclock1 intermitent failures

has 2 issues:

1. It assumes time_t == long which is false on x32.
2. tst-timespec.c is compiled without -fexcess-precision=standard which
generates incorrect results on i686 in support_timespec_check_in_range:

  double ratio = (double)observed_norm / expected_norm;
  return (lower_bound <= ratio && ratio <= upper_bound);

This patch does

1. Compile tst-timespec.c with -fexcess-precision=standard.
2. Replace long with time_t.
3. Replace LONG_MIN and LONG_MAX with TYPE_MINIMUM (time_t) and
TYPE_MAXIMUM (time_t).

4 years agoRemove --enable-obsolete-rpc configure flag
Petr Vorel [Mon, 13 Jul 2020 17:36:35 +0000 (19:36 +0200)]
Remove --enable-obsolete-rpc configure flag

Sun RPC was removed from glibc. This includes rpcgen program, librpcsvc,
and Sun RPC headers. Also test for bug #20790 was removed
(test for rpcgen).

Backward compatibility for old programs is kept only for architectures
and ABIs that have been added in or before version 2.28.

libtirpc is mature enough, librpcsvc and rpcgen are provided in
rpcsvc-proto project.

NOTE: libnsl code depends on Sun RPC (installed libnsl headers use
installed Sun RPC headers), thus --enable-obsolete-rpc was a dependency
for --enable-obsolete-nsl (removed in a previous commit).

The arc ABI list file has to be updated because the port was added
with the sunrpc symbols

Tested-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agohurd: Fix build-many-glibcs.py
Adhemerval Zanella [Fri, 10 Jul 2020 17:48:17 +0000 (14:48 -0300)]
hurd: Fix build-many-glibcs.py

It fixes the issue report by Joseph [1].

Checked with a build-many-glibcs.py build for i686-gnu.

[1] https://sourceware.org/pipermail/libc-alpha/2020-July/116134.html

4 years agox86: Support usable check for all CPU features
H.J. Lu [Mon, 29 Jun 2020 23:36:08 +0000 (16:36 -0700)]
x86: Support usable check for all CPU features

Support usable check for all CPU features with the following changes:

1. Change struct cpu_features to

struct cpuid_features
{
  struct cpuid_registers cpuid;
  struct cpuid_registers usable;
};

struct cpu_features
{
  struct cpu_features_basic basic;
  struct cpuid_features features[COMMON_CPUID_INDEX_MAX];
  unsigned int preferred[PREFERRED_FEATURE_INDEX_MAX];
...
};

so that there is a usable bit for each cpuid bit.
2. After the cpuid bits have been initialized, copy the known bits to the
usable bits.  EAX/EBX from INDEX_1 and EAX from INDEX_7 aren't used for
CPU feature detection.
3. Clear the usable bits which require OS support.
4. If the feature is supported by OS, copy its cpuid bit to its usable
bit.
5. Replace HAS_CPU_FEATURE and CPU_FEATURES_CPU_P with CPU_FEATURE_USABLE
and CPU_FEATURE_USABLE_P to check if a feature is usable.
6. Add DEPR_FPU_CS_DS for INDEX_7_EBX_13.
7. Unset MPX feature since it has been deprecated.

The results are

1. If the feature is known and doesn't requre OS support, its usable bit
is copied from the cpuid bit.
2. Otherwise, its usable bit is copied from the cpuid bit only if the
feature is known to supported by OS.
3. CPU_FEATURE_USABLE/CPU_FEATURE_USABLE_P are used to check if the
feature can be used.
4. HAS_CPU_FEATURE/CPU_FEATURE_CPU_P are used to check if CPU supports
the feature.

4 years agostring: Make tst-strerror/tst-strsignal unsupported if msgfmt is not installed
Adhemerval Zanella [Thu, 9 Jul 2020 16:24:04 +0000 (13:24 -0300)]
string: Make tst-strerror/tst-strsignal unsupported if msgfmt is not installed

Without msgfmt libc.mo files are not generated and its loading failure
is silent ignored with xsetlocale.

Also unset LANGUAGE environment variable to avoid it taking precedence
when loading the message catalog.  Although not strictly required
(since the test is issued with test-container and it sets a strict
environment variable) it follows other tests that deal with
translation.

Checked on x86_64-linux-gnu.

4 years agomalloc: Deprecate more hook-related functionality
Florian Weimer [Mon, 13 Jul 2020 07:58:37 +0000 (09:58 +0200)]
malloc: Deprecate more hook-related functionality

__morecore, __after_morecore_hook, and __default_morecore had not
been deprecated in commit 7d17596c198f11fa85cbcf9587443f262e63b616
("Mark malloc hook variables as deprecated"), probably by accident.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agoelf: Support at least 32-byte alignment in static dlopen
Florian Weimer [Sun, 12 Jul 2020 15:53:58 +0000 (17:53 +0200)]
elf: Support at least 32-byte alignment in static dlopen

Otherwise loading a dynamically linked libc with rseq support fails,
as result of the __rseq_abi TLS variable, which has an alignment
of 32 bytes.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
Tested-by: Vineet Gupta <vgupta@synopsys.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Tested-by: Alistair Francis <alistair.francis@wdc.com>
4 years agox86: Remove __ASSEMBLER__ check in init-arch.h
H.J. Lu [Sat, 11 Jul 2020 17:03:05 +0000 (10:03 -0700)]
x86: Remove __ASSEMBLER__ check in init-arch.h

Since

commit 430388d5dc0e1861b869096f4f5d946d7d74232a
Author: H.J. Lu <hjl.tools@gmail.com>
Date:   Fri Aug 3 08:04:49 2018 -0700

    x86: Don't include <init-arch.h> in assembly codes

removed all usages of <init-arch.h> from assembly codes, we can remove
__ASSEMBLER__ check in init-arch.h.

4 years agox86: Remove the unused __x86_prefetchw
H.J. Lu [Sat, 11 Jul 2020 16:04:34 +0000 (09:04 -0700)]
x86: Remove the unused __x86_prefetchw

Since

commit c867597bff2562180a18da4b8dba89d24e8b65c4
Author: H.J. Lu <hjl.tools@gmail.com>
Date:   Wed Jun 8 13:57:50 2016 -0700

    X86-64: Remove previous default/SSE2/AVX2 memcpy/memmove

removed the only usage of __x86_prefetchw, we can remove the unused
__x86_prefetchw.

4 years agoDocumentation for ARC port
Vineet Gupta [Mon, 20 Nov 2017 20:46:17 +0000 (12:46 -0800)]
Documentation for ARC port

(a) ABI doc:
    https://github.com/foss-for-synopsys-dwc-arc-processors/toolchain/wiki/files/ARCv2_ABI.pdf

(b) Programmer's Reference Manual (PRM) : needs a download request to be filled
    https://www.synopsys.com/dw/ipdir.php?ds=arc-hs44-hs46-hs48
    https://www.synopsys.com/dw/doc.php/ds/cc/programmers-reference-manual-ARC-HS.pdf

As of port merge (and Florian's patch to fix static TLS), glibc testsuite
(cross-test setup) has following failures:

FAIL: elf/tst-audit14
FAIL: elf/tst-audit15
FAIL: elf/tst-audit16
FAIL: elf/tst-ldconfig-ld_so_conf-update
FAIL: elf/tst-libc_dlvsym
FAIL: elf/tst-libc_dlvsym-static
FAIL: iconv/test-iconvconfig # Needs gconv installed
FAIL: io/ftwtest # Requires execution by non-root
FAIL: io/tst-lockf
FAIL: libio/tst-wfile-sync
FAIL: locale/tst-localedef-path-norm
FAIL: nptl/test-cond-printers # needs Python3 and target GDB on target
FAIL: nptl/test-condattr-printers #    ditto
FAIL: nptl/test-mutex-printers #    ditto
FAIL: nptl/test-mutexattr-printers #    ditto
FAIL: nptl/test-rwlock-printers #    ditto
FAIL: nptl/test-rwlockattr-printers #    ditto
FAIL: nptl/tst-umask1 # passes if run natively on target (NFS ACLv3 support needed)
FAIL: nss/bug-erange
FAIL: nss/tst-nss-files-hosts-getent
FAIL: nss/tst-nss-files-hosts-multi
FAIL: posix/bug-ga2
FAIL: posix/globtest # require same user on target and host
FAIL: posix/tst-getaddrinfo5
FAIL: stdio-common/tst-vfprintf-width-prec
FAIL: stdio-common/tst-vfprintf-width-prec-alloc
FAIL: stdio-common/tst-vfprintf-width-prec-mem
FAIL: string/tst-strerror
FAIL: string/tst-strsignal
FAIL: sunrpc/bug20790 # missing cpp on target
FAIL: timezone/tst-tzset

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agobuild-many-glibcs.py: Enable ARC builds
Vineet Gupta [Fri, 13 Mar 2020 01:29:59 +0000 (18:29 -0700)]
build-many-glibcs.py: Enable ARC builds

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agoARC: Build Infrastructure
Vineet Gupta [Tue, 23 Oct 2018 22:39:50 +0000 (15:39 -0700)]
ARC: Build Infrastructure

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agoARC: ABI lists
Vineet Gupta [Tue, 23 Oct 2018 22:37:08 +0000 (15:37 -0700)]
ARC: ABI lists

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agoARC: Linux Startup and Dynamic Loading
Vineet Gupta [Tue, 23 Oct 2018 21:58:39 +0000 (14:58 -0700)]
ARC: Linux Startup and Dynamic Loading

A big shoutout to Cupertino Miranda <cmiranda@synopsys.com> for his
valuable contribution in initial bringup and debugging on Linux and
later in solving pesky unwinding/cancelation failures in testsuite.

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agoARC: Linux ABI
Vineet Gupta [Tue, 23 Oct 2018 21:56:39 +0000 (14:56 -0700)]
ARC: Linux ABI

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agoARC: Linux Syscall Interface
Vineet Gupta [Tue, 23 Oct 2018 21:44:30 +0000 (14:44 -0700)]
ARC: Linux Syscall Interface

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agoARC: hardware floating point support
Vineet Gupta [Tue, 5 Nov 2019 00:20:37 +0000 (16:20 -0800)]
ARC: hardware floating point support

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agoARC: math soft float support
Vineet Gupta [Wed, 24 Oct 2018 21:22:54 +0000 (14:22 -0700)]
ARC: math soft float support

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agoARC: Atomics and Locking primitives
Vineet Gupta [Tue, 23 Oct 2018 21:36:39 +0000 (14:36 -0700)]
ARC: Atomics and Locking primitives

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agoARC: Thread Local Storage support
Vineet Gupta [Tue, 23 Oct 2018 21:18:04 +0000 (14:18 -0700)]
ARC: Thread Local Storage support

This includes all 4 TLS addressing models

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agoARC: startup and dynamic linking code
Vineet Gupta [Tue, 23 Oct 2018 19:36:32 +0000 (12:36 -0700)]
ARC: startup and dynamic linking code

Code for C runtime startup and dynamic loading including PLT layout.

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agoARC: ABI Implementation
Vineet Gupta [Tue, 23 Oct 2018 18:38:32 +0000 (11:38 -0700)]
ARC: ABI Implementation

This code deals with the ARC ABI.

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
4 years agoFix time/tst-cpuclock1 intermitent failures
Lucas A. M. Magalhaes [Fri, 10 Jul 2020 22:41:06 +0000 (19:41 -0300)]
Fix time/tst-cpuclock1 intermitent failures

This test fails intermittently in systems with heavy load as
CLOCK_PROCESS_CPUTIME_ID is subject to scheduler pressure.  Thus the
test boundaries were relaxed to keep it from failing on such systems.

A refactor of the spent time checking was made with some support
functions.  With the advantage to representing time jitter in percent
of the target.

The values used by the test boundaries are all empirical.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
4 years agopowerpc64: Fix calls when r2 is not used [BZ #26173]
Tulio Magno Quites Machado Filho [Fri, 10 Jul 2020 22:41:06 +0000 (19:41 -0300)]
powerpc64: Fix calls when r2 is not used [BZ #26173]

Teach the linker that __mcount_internal, __sigjmp_save_symbol,
__syscall_error and __GI_exit do not use r2, so that it does not need to
recover r2 after the call.

Test at configure time if the assembler supports @notoc and define
USE_PPC64_NOTOC.

4 years agoAdd NEWS entry for Update to Unicode 13.0.0 [BZ #25819]
Mike FABIAN [Wed, 8 Jul 2020 12:41:23 +0000 (14:41 +0200)]
Add NEWS entry for Update to Unicode 13.0.0 [BZ #25819]

Unicode 13.0.0 Support: Character encoding, character type info, and
transliteration tables are all updated to Unicode 13.0.0, using
the generator scripts contributed by Mike FABIAN (Red Hat).

Total added characters in newly generated CHARMAP: 5930
Total added characters in newly generated WIDTH: 5536

4 years agoUpdate i686 libm-test-ulps
Patsy Franklin [Thu, 9 Jul 2020 20:06:17 +0000 (16:06 -0400)]
Update i686 libm-test-ulps

Without my ULP patch these 18 tests fail on i686:
 https://koji.fedoraproject.org/koji/taskinfo?taskID=46467301

+ cat /proc/cpuinfo
processor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 85
model name : Intel Xeon Processor (Cascadelake)

FAIL: math/test-double-j0
FAIL: math/test-double-y0
FAIL: math/test-float-erfc
FAIL: math/test-float-j0
FAIL: math/test-float-j1
FAIL: math/test-float-lgamma
FAIL: math/test-float-tgamma
FAIL: math/test-float-y0
FAIL: math/test-float32-erfc
FAIL: math/test-float32-j0
FAIL: math/test-float32-j1
FAIL: math/test-float32-lgamma
FAIL: math/test-float32-tgamma
FAIL: math/test-float32-y0
FAIL: math/test-float32x-j0
FAIL: math/test-float32x-y0
FAIL: math/test-float64-j0
FAIL: math/test-float64-y0

With my ULP patch applied these tests now pass:
 https://koji.fedoraproject.org/koji/taskinfo?taskID=46436310

4 years agoFix memory leak in __printf_fp_l (bug 26215).
Joseph Myers [Thu, 9 Jul 2020 21:52:24 +0000 (21:52 +0000)]
Fix memory leak in __printf_fp_l (bug 26215).

__printf_fp_l has a memory leak in the case of some I/O errors, where
both buffer and wbuffer have been malloced but the handling of I/O
errors only frees wbuffer.  This patch fixes this by moving the
declaration of buffer to an outer scope and ensuring that it is freed
when wbuffer is freed.

Tested for x86_64 and x86.

4 years agoFix double free in __printf_fp_l (bug 26214).
Joseph Myers [Thu, 9 Jul 2020 21:51:49 +0000 (21:51 +0000)]
Fix double free in __printf_fp_l (bug 26214).

__printf_fp_l has a double free bug in the case where it allocates
memory with malloc internally, then has an I/O error while outputting
trailing padding and tries to free that already-freed memory when the
error occurs.  This patch fixes this by setting the relevant pointer
to NULL after the first free (the only free of this pointer that isn't
immediately followed by returning from the function).

Tested for x86_64 and x86.

4 years agolinux: Fix syscall list generation instructions
Maciej W. Rozycki [Thu, 9 Jul 2020 16:43:57 +0000 (17:43 +0100)]
linux: Fix syscall list generation instructions

Make the instructions for syscall list generation match Makefile and
refer to `update-syscall-lists'; there has been no `update-arch-syscall'
target.  Also use single quotes around the command to stick to the ASCII
character set.

Fixes 4cf0d223052d ("Linux: Add tables with system call numbers").

Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
4 years agosysv: linux: Add 64-bit time_t variant for shmctl
Adhemerval Zanella [Tue, 30 Jun 2020 17:08:22 +0000 (14:08 -0300)]
sysv: linux: Add 64-bit time_t variant for shmctl

To provide a y2038 safe interface a new symbol __shmctl64 is added
and __shmctl is change to call it instead (it adds some extra buffer
copying for the 32 bit time_t implementation).

Two new structures are added:

  1. kernel_shmid64_ds: used internally only on 32-bit architectures
     to issue the syscall.  A handful of architectures (hppa, i386,
     mips, powerpc32, and sparc32) require specific implementations
     due to their kernel ABI.

  2. shmid_ds64: this is only for __TIMESIZE != 64 to use along with
     the 64-bit shmctl.  It is different than the kernel struct because
     the exported 64-bit time_t might require different alignment
     depending on the architecture ABI.

So the resulting implementation does:

  1. For 64-bit architectures it assumes shmid_ds already contains
     64-bit time_t fields and will result in just the __shmctl symbol
     using the __shmctl64 code.  The shmid_ds argument is passed as-is
     to the syscall.

  2. For 32-bit architectures with default 64-bit time_t (newer ABIs
     such riscv32 or arc), it will also result in only one exported
     symbol but with the required high/low time handling.

  3. Finally for 32-bit architecture with both 32-bit and 64-bit time_t
     support we follow the already set way to provide one symbol with
     64-bit time_t support and implement the 32-bit time_t support
     using of the 64-bit one.

     The default 32-bit symbol will allocate and copy the shmid_ds
     over multiple buffers, but this should be deprecated in favor
     of the __shmctl64 anyway.

Checked on i686-linux-gnu and x86_64-linux-gnu.  I also did some sniff
tests on powerpc, powerpc64, mips, mips64, armhf, sparcv9, and
sparc64.

Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Tested-by: Alistair Francis <alistair.francis@wdc.com>
Tested-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
This page took 0.078769 seconds and 5 git commands to generate.