The previous barrier implementation did not fulfill the POSIX requirements
for when a barrier can be destroyed. Specifically, it was possible that
threads that haven't noticed yet that their round is complete still access
the barrier's memory, and that those accesses can happen after the barrier
has been legally destroyed.
The new algorithm does not have this issue, and it avoids using a lock
internally.
GLIBC benchtest testcases shows SSE2_Unaligned based implementations
are performing faster compare to SSE2 based implementations for
routines: strcmp, strcat, strncat, stpcpy, stpncpy, strcpy, strncpy
and strstr. Flag index_Fast_Unaligned_Load is set for Excavator family
0x15h CPU's. This makes SSE2_Unaligned based implementations as
default for these routines.
[BZ #19467]
* sysdeps/x86/cpu-features.c (init_cpu_features): Set
index_Fast_Unaligned_Load flag for Excavator family CPUs.
Preparation for gcc -fsplit-stack support (gcc bug #68191). The new
field is basically identical to the one on x86. Its TCB offset needs
to be constant, as it'll be hardcoded in gcc.
ChangeLog:
* sysdeps/s390/nptl/tls.h (struct tcbhead_t): Add __private_ss field.
This patch adds some new header definitions from Linux 4.4:
* MCL_ONFAULT is added to bits/mman.h / bits/mman-linux.h (this was
already done for hppa).
* PTRACE_SECCOMP_GET_FILTER is added to sys/ptrace.h. Along with it,
the older PTRACE_GETSIGMASK and PTRACE_SETSIGMASK, added in Linux
3.11 but missed at the time, are also added.
Tested for x86_64 and x86 (testsuite, and that installed stripped
shared libraries are unchanged by the patch).
* bits/mman-linux.h [!MCL_CURRENT] (MCL_ONFAULT): New macro.
* sysdeps/unix/sysv/linux/alpha/bits/mman.h (MCL_ONFAULT):
Likewise.
* sysdeps/unix/sysv/linux/powerpc/bits/mman.h (MCL_ONFAULT):
Likewise.
* sysdeps/unix/sysv/linux/sparc/bits/mman.h (MCL_ONFAULT):
Likewise.
* sysdeps/unix/sysv/linux/sys/ptrace.h (PTRACE_GETSIGMASK): New
enum constant and macro.
(PTRACE_SETSIGMASK): Likewise.
(PTRACE_SECCOMP_GET_FILTER): Likewise.
* sysdeps/unix/sysv/linux/aarch64/sys/ptrace.h
(PTRACE_GETSIGMASK): Likewise.
(PTRACE_SETSIGMASK): Likewise.
(PTRACE_SECCOMP_GET_FILTER): Likewise.
* sysdeps/unix/sysv/linux/ia64/sys/ptrace.h (PTRACE_GETSIGMASK):
Likewise.
(PTRACE_SETSIGMASK): Likewise.
(PTRACE_SECCOMP_GET_FILTER): Likewise.
* sysdeps/unix/sysv/linux/powerpc/sys/ptrace.h
(PTRACE_GETSIGMASK): Likewise.
(PTRACE_SETSIGMASK): Likewise.
(PTRACE_SECCOMP_GET_FILTER): Likewise.
* sysdeps/unix/sysv/linux/s390/sys/ptrace.h (PTRACE_GETSIGMASK):
Likewise.
(PTRACE_SETSIGMASK): Likewise.
(PTRACE_SECCOMP_GET_FILTER): Likewise.
* sysdeps/unix/sysv/linux/sparc/sys/ptrace.h (PTRACE_GETSIGMASK):
Likewise.
(PTRACE_SETSIGMASK): Likewise.
(PTRACE_SECCOMP_GET_FILTER): Likewise.
* sysdeps/unix/sysv/linux/tile/sys/ptrace.h (PTRACE_GETSIGMASK):
Likewise.
(PTRACE_SETSIGMASK): Likewise.
(PTRACE_SECCOMP_GET_FILTER): Likewise.
Work around a GCC behavior with hardware transactional memory built-ins.
GCC doesn't treat the PowerPC transactional built-ins as compiler
barriers, moving instructions past the transaction boundaries and
altering their atomicity.
The attached patch fixes dladdr on hppa.
Instead of using the generic version of _dl_lookup_address, we use an
implementation more or less modeled after __canonicalize_funcptr_for_compare()
in gcc. The function pointer is analyzed and if it points to the
trampoline used to call _dl_runtime_resolve just before the global
offset table, then we call _dl_fixup to resolve the function pointer.
Then, we return the instruction pointer from the first word of the
descriptor.
The change fixes the testcase provided in [BZ #19415] and the Debian
nss package now builds successfully.
We define __ASSUME_ST_INO_64_BIT by default for Linux targets, and then
undef it for alpha/sh targets. But the code that uses it looks at its
value (as 0/1) rather than whether it's defined (like all other assume
knobs). Change the code to see if it's defined to fix build Wundef build
errors for alpha/sh.
Since internal unistd functions are only used internally in ld.so and
libc.so, they can be made hidden. __close, __getcwd, __getpid,
__libc_read and __libc_write can't be hidden in ld.so on Hurd since they
will be preempted by the ones in libc.so after bootstrap.
[BZ #19122]
* include/unistd.h [IS_IN (rtld)]: Include <dl-unistd.h>.
* sysdeps/generic/dl-unistd.h: New file.
* sysdeps/mach/hurd/dl-unistd.h: Likewise.
Since ld.so internal mmap functions are only used internally in ld.so,
they can be made hidden. Don't hide __mmap on Hurd, since __mmap in
ld.so will be preempted by the one in libc.so after bootstrap.
[BZ #19122]
* include/sys/mman.h [IS_IN (rtld)]: Include <dl-mman.h>.
* sysdeps/generic/dl-mman.h: New file.
* sysdeps/mach/hurd/dl-mman.h: Likewise.
When looking at the code generated for pow() on ppc64 I noticed quite
a few sign extensions. Making the array indices unsigned reduces the
number of sign extensions from 24 to 7.
Tested for powerpc64le and x86_64.
The attached patch adds some upstream defines like MAP_HUGETLB and MAP_STACK
in mman.h for the hppa architecture.
The existing MADV_xxK_PAGES defines were dropped upstream, because they were
originally added many years ago based on a proposed patch for the Linux kernel
which was never applied. So, this patch drops those unneeded defines.
The rework in commit d709042a6e broke
buiding on ia64 due to compat_symbol expanding into ... in some cases.
The common files were wrapped in a BUILD_LGAMMA check, but the ia64
ones were not. Add that logic to the ia64 files too.
The personality system call, starting with linux kernel commit
v2.6.29-6609-g11d06b2a1e5658f448a308aa3beb97bacd64a940, always
successfully changes the personality if requested. The syscall
wrapper, however, still can return an error in the following cases:
- the value returned by the system call looks like an error
due to architecture limitations of 32-bit kernels;
- a personality greater than 0xffffffff is passed to the system call,
and the 64-bit kernel does not have commit
v2.6.35-rc1-372-g485d527686850d68a0e9006dd9904f19f122485e
that would truncate this value to unsigned int;
- on sparc64, the value returned by the system call looks like an error
due to sparc64 kernel sign extension bug.
The solution is three-fold:
- move generic syscalls.list personality entry to generic 64-bit
syscalls.list file;
- for each 32-bit architecture that use negated errno semantics,
add a NOERRNO personality entry to their syscalls.list file;
- for sparc64 and 32-bit architectures that use dedicated registers
to flag syscall errors, add a wrapper around personality syscall;
if the system call return value is flagged as an error, this wrapper
returns the negated "would be errno" value, otherwise it returns
the system call return value; on sparc64, it also truncates the
personality argument to unsigned int before passing it to the kernel.
[BZ #19408]
* sysdeps/unix/sysv/linux/personality.c: New file.
* sysdeps/unix/sysv/linux/sparc/sparc64/personality.c: Likewise.
* sysdeps/unix/sysv/linux/tst-personality.c: Likewise.
* sysdeps/unix/sysv/linux/Makefile [$(subdir) == misc]
(sysdep_routines): Add personality.
(tests): Add tst-personality.
* sysdeps/unix/sysv/linux/syscalls.list (personality): Move ...
* sysdeps/unix/sysv/linux/wordsize-64/syscalls.list: ... here.
* sysdeps/unix/sysv/linux/arm/syscalls.list (personality): New entry.
* sysdeps/unix/sysv/linux/hppa/syscalls.list (personality): Likewise.
* sysdeps/unix/sysv/linux/i386/syscalls.list (personality): Likewise.
* sysdeps/unix/sysv/linux/m68k/syscalls.list (personality): Likewise.
* sysdeps/unix/sysv/linux/microblaze/syscalls.list (personality):
Likewise.
* sysdeps/unix/sysv/linux/mips/mips64/n32/syscalls.list (personality):
Likewise.
* sysdeps/unix/sysv/linux/s390/s390-32/syscalls.list (personality):
Likewise.
* sysdeps/unix/sysv/linux/sh/syscalls.list (personality): Likewise.
* sysdeps/unix/sysv/linux/x86_64/x32/syscalls.list (personality):
Likewise.
Since GLIBC requires a minimum 2.6.32 kernel, the sysctl (CTL_BUS,
CTL_BUS_ISA, ISA_*) is always available. We can therefore remove the
fallback code reading /etc/arm_systype or parsing /proc/cpuinfo.
Remove fscanf from localplt.data as it is no longer called from within
GLIBC.
* sysdeps/unix/sysv/linux/arm/ioperm.c: Do not include <string.h>.
(PATH_ARM_SYSTYPE): Remove.
(PATH_CPUINFO): Likewise.
(IO_BASE_FOOTBRIDGE): Likewise.
(IO_SHIFT_FOOTBRIDGE): Likewise.
(struct platform): Likewise.
(init_iosys): Remove compatibility code for 2.4 kernels.
* sysdeps/unix/sysv/linux/arm/localplt.data: Remove fscanf.
The attached patch fixes BZ #19270 and the Debian gmt package now builds
successfully. Aside from the comment, the define of __NO_LONG_DOUBLE_MATH
is similar to that in the generic version of glibc.
Build tested on hppa-unknown-linux-gnu with no observed regressions.
POSIX and C++11 require that a thread can destroy a mutex if no other
thread owns the mutex, is blocked on the mutex, or will try to acquire
it in the future. After destroying the mutex, it can reuse or unmap the
underlying memory. Thus, we must not access a mutex' memory after
releasing it. Currently, we can load the private flag after releasing
the mutex, which is fixed by this patch.
See https://sourceware.org/bugzilla/show_bug.cgi?id=13690 for more
background.
We need to call futex_wake on the lock after releasing it, however. This
is by design, and can lead to spurious wake-ups on unrelated futex words
(e.g., when the mutex memory is reused for another mutex). This behavior
is documented in the glibc-internal futex API and in recent drafts of the
Linux kernel's futex documentation (see the draft_futex branch of
git://git.kernel.org/pub/scm/docs/man-pages/man-pages.git).
Commit 67385a01d2 added a new feature for
powerpc, where we store HWCAP/Platform bits in the TCB. In the dynamic
linking case, we use the versioned symbol
'__parse_hwcap_and_convert_at_platform' to verify if this feature is
available. However, the same symbol was not exported to libc.a, making
it not possible for GCC to check for it prior to link time.
This fixes build when _IO_funlockfile is a macro, fixes build where
_IO_acquire_lock_clear_flags2 is used, and fixes unlocking on unexpected
stack unwind.
* sysdeps/generic/stdio-lock.h [__EXCEPTIONS] (_IO_acquire_lock,
_IO_release_lock ): Use cleanup attribute on new
_IO_acquire_lock_file variable instead of assuming that
_IO_release_lock will be called.
[!__EXCEPTIONS] (_IO_acquire_lock): Define to non-existing
_IO_acquire_lock_needs_exceptions_enabled.
(_IO_acquire_lock_clear_flags2): New macro.
Like the previous change, exploit the fact that computation for sin
and cos is identical except that it is apart by a quadrant. Also
remove csloww, csloww1 and csloww2 since they can easily be expressed
in terms of sloww, sloww1 and sloww2.
The sin and cos computation for this range of input is identical
except for a difference in quadrants by 1. Exploit that fact and the
common argument reduction to reduce computations for sincos.
Range reduction needs to be done only once for sin and cos, so copy
over all of the relevant functions (__sin, __cos, reduce_and_compute)
and consolidate common code.
The i386 ULPs are actually the i686/multiarch ones. The i686/multiarch
float ULPs are more precise as the SSE2 version (when available) uses
double for the cosf and sinf functions.
On the other hand the higher precision of the x86 FPU improves the
precision for a few other math functions.
* sysdeps/i386/fpu/libm-test-ulps: Move to ....
* sysdeps/i386/i686/multiarch/fpu/libm-test-ulps: ...here.
* sysdeps/i386/fpu/libm-test-ulps: Regenerate.
It shows improvement up to 28% over AVX2 memset (performance results
attached at <https://sourceware.org/ml/libc-alpha/2015-12/msg00052.html>).
* sysdeps/x86_64/multiarch/memset-avx512-no-vzeroupper.S: New file.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Added new file.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c: Added new tests.
* sysdeps/x86_64/multiarch/memset.S: Added new IFUNC branch.
* sysdeps/x86_64/multiarch/memset_chk.S: Likewise.
* sysdeps/x86/cpu-features.h (bit_Prefer_No_VZEROUPPER,
index_Prefer_No_VZEROUPPER): New.
* sysdeps/x86/cpu-features.c (init_cpu_features): Set the
Prefer_No_VZEROUPPER for Knights Landing.
This patch fixes the SYSCALL_CANCEL macro for usage with zero argument
number (for instance SYSCALL_CANCEL (pause)) using a similar approach
used for SOCKETCALL_CANCEL.
GLIBC build still does not hit this issue still since SYSCALL_CANCEL
is not currently being used for zero arguments calls.
Tested on i386, x86_64, powerpc64le, aarch64.
* sysdeps/unix/sysdep.h (SYSCALL_CANCEL): Fix macro for zero argument
syscalls.
(__SYSCALL0): New macro.
(__SYSCALL1): Likewise.
(__SYSCALL2): Likewise.
(__SYSCALL3): Likewise.
(__SYSCALL4): Likewise.
(__SYSCALL5): Likewise.
(__SYSCALL6): Likewise.
(__SYSCALL7): Likewise.
(__SYSCALL_CONCAT_X): Likewise.
(__SYSCALL_CONCAT): Likewise.
(__SYSCALL_DIST): Likewise.
(__SYSCALL_CALL): Likewise.
Since times returns 64-bit clock_t on x32, we need to provide x32 times
by redefining INTERNAL_SYSCALL_NCS and INTERNAL_SYSCALL_ERROR_P with
64-bit return type for syscall. All system calls returning 64-bit
integer, which are lseek, time and times, must be handled specially for
x32. lseek is handled by x32 lseek.S and time doesn't check syscall
return. times is the only missed one. Before this patch, there are
0000000 <__times>:
0: b8 64 00 00 40 mov $0x40000064,%eax
5: 0f 05 syscall
7: 48 63 d0 movslq %eax,%rdx
^^^^^^^^^^ Incorrect signed extension
a: 48 83 fa f2 cmp $0xfffffffffffffff2,%rdx
e: 75 07 jne 17 <__times+0x17>
10: 3d 00 f0 ff ff cmp $0xfffff000,%eax
^^^^^^^^^^^^^^^^^^^^^ 32-bit compare
15: 77 11 ja 28 <__times+0x28>
17: 48 83 fa ff cmp $0xffffffffffffffff,%rdx
1b: b8 00 00 00 00 mov $0x0,%eax
20: 48 0f 45 c2 cmovne %rdx,%rax
24: c3 retq
After this patch, there are
00000000 <__times>:
0: b8 64 00 00 40 mov $0x40000064,%eax
5: 0f 05 syscall
7: 48 83 f8 f2 cmp $0xfffffffffffffff2,%rax
b: 75 08 jne 15 <__times+0x15>
d: 48 3d 00 f0 ff ff cmp $0xfffffffffffff000,%rax
13: 77 13 ja 28 <__times+0x28>
15: 48 83 f8 ff cmp $0xffffffffffffffff,%rax
19: ba 00 00 00 00 mov $0x0,%edx
1e: 48 0f 44 c2 cmove %rdx,%rax
22: c3 retq
The incorrect signed extension and 32-bit compare are gone.
[BZ #19363]
* sysdeps/unix/sysv/linux/x86_64/x32/times.c: New file.
The optimized POWER7 logb implementation does not use the absolute
value of the word extracted from the input to apply the leading 0-bits
builtin (to ignore the float sign). This patch fixes it by
clearing the signal bit in the resulting word.
It fixes the subnormal tests failures when running on POWER7 ou
newer chip.
Tested on powerpc64le (POWER8).
[BZ# 19375]
* sysdeps/powerpc/power7/fpu/s_logb.c (__logb): Fix return for
negative subnormals.
X86-64 system calls use a different calling convention, which clobbers
CC, %r11 an %rcx registers. Define REGISTERS_CLOBBERED_BY_SYSCALL for
x86-64 inline asm statements.
* sysdeps/unix/sysv/linux/x86_64/sysdep.h
(REGISTERS_CLOBBERED_BY_SYSCALL): New.
(INTERNAL_SYSCALL_NCS): Use it.
(INTERNAL_SYSCALL_NCS_TYPES): Likewise.
According to Silvermont software optimization guide, for 64-bit
applications, branch prediction performance can be negatively impacted
when the target of a branch is more than 4GB away from the branch. Add
the Prefer_MAP_32BIT_EXEC bit so that mmap will try to map executable
pages with MAP_32BIT first. NB: MAP_32BIT will map to lower 2GB, not
lower 4GB, address. Prefer_MAP_32BIT_EXEC reduces bits available for
address space layout randomization (ASLR), which is always disabled for
SUID programs and can only be enabled by setting environment variable,
LD_PREFER_MAP_32BIT_EXEC.
On Fedora 23, this patch speeds up GCC 5 testsuite by 3% on Silvermont.
[BZ #19367]
* sysdeps/unix/sysv/linux/wordsize-64/mmap.c: New file.
* sysdeps/unix/sysv/linux/x86_64/64/dl-librecon.h: Likewise.
* sysdeps/unix/sysv/linux/x86_64/64/mmap.c: Likewise.
* sysdeps/x86/cpu-features.h (bit_Prefer_MAP_32BIT_EXEC): New.
(index_Prefer_MAP_32BIT_EXEC): Likewise.
Knights Landing processor is based on Silvermont. This patch enables
Silvermont optimizations for Knights Landing.
* sysdeps/x86/cpu-features.c (init_cpu_features): Enable
Silvermont optimizations for Knights Landing.
Various Linux kernel syscalls have become obsolete over time.
Specifically, the following are obsolete in all kernel versions
supported by glibc, are not present for architectures more recently
added to the kernel, and as such, the wrapper functions for them
should be compat symbols, not in static libc and not available for new
links with shared libc.
* bdflush: in Linux 2.6, does nothing if present.
* create_module get_kernel_syms query_module: Linux 2.4 module
interface, syscalls not present in Linux 2.6.
* uselib: part of the mechanism for loading a.out shared libraries,
irrelevant with ELF.
This patch adds support for syscalls.list to list syscall aliases of
the form NAME@VERSION:OBSOLETED, with SHLIB_COMPAT conditionals being
generated for such aliases. Those five syscalls are then made into
compat symbols (obsoleted in glibc 2.23, so future ports won't have
these symbols at all), with the header <sys/kdaemon.h> declaring
bdflush being removed. When we move to 3.2 as minimum kernel version,
the same can be done for nfsservctl (removed in Linux 3.1) as well.
Tested for x86_64 and x86 (testsuite, as well as checking that the
symbols in question indeed become compat symbols, that they are indeed
omitted from static libc, and that the generated SHLIB_COMPAT
conditionals look right).
[BZ #18472]
* sysdeps/unix/Makefile ($(objpfx)stub-syscalls.c): Handle entries
for the form NAME@VERSION:OBSOLETED and generate SHLIB_COMPAT
conditionals for them.
* sysdeps/unix/make-syscalls.sh (emit_weak_aliases): Likewise.
* sysdeps/unix/sysv/linux/sys/kdaemon.h: Remove file.
* sysdeps/unix/sysv/linux/Makefile (sysdep_headers): Remove
sys/kdaemon.h.
* sysdeps/unix/sysv/linux/syscalls.list (bdflush): Make into
compat-only syscall, obsoleted in glibc 2.23.
(create_module): Likewise.
(get_kernel_syms): Likewise.
(query_module): Likewise.
(uselib): Likewise.
* manual/sysinfo.texi (System Parameters): Do not mention bdflush.
According to POSIX the grantpt() function does the following:
The grantpt() function shall change the mode and ownership of the
slave pseudo-terminal device associated with its master
pseudo-terminal counterpart. The fildes argument is a file descriptor
that refers to a master pseudo-terminal device. The user ID of the
slave shall be set to the real UID of the calling process and the
group ID shall be set to an unspecified group ID. The permission
mode of the slave pseudo-terminal shall be set to readable and
writable by the owner, and writable by the group.
Historically the GNU libc has been responsible to setup the permission
mode to 0620 and the group to 'tty' usually number 5, using the pt_chown
helper, badly known for its security issues. With the creation of the
devpts filesytem in the Linux kernel, this responsibility has been moved
to the Linux kernel. The system is responsible to mount the devpts
filesystem in /dev/pts with the options gid=5 and mode=0620. In that
case the GNU libc has nothing to do and pt_chown is not need anymore. So
far so good.
The problem is that by default the devpts filesystem is shared between
all mounts, and that contrary to other filesystem, the mount options are
honored at the second mount, including for the default mount options.
Given it corresponds to mode=0600 without gid parameter (that is the
filesystem GID of the creating process), it's common to see systems
where the devpts filesystem is mounted using these options. It is enough
to run a "mount -t devpts devpts /mychroot/dev/pts" to come into this
situation, and it's unfortunately wrongly used in a lot of scripts
dealing with chroots, or for creating virtual machines images.
When this happens the GNU libc tries to fix the group and permission
mode of the pty nodes, and given it fails to do so for non-root users,
grantpt() almost always fail. It means users are not able to open new
terminals.
This patch changes grantpt() to not enforce this anymore, while still
enforcing minimum security measures to the permission mode. Therefore
the responsibility to follow POSIX is now shared at the system level,
i.e. kernel + system scripts + GNU libc. It stops trying to change the
group, and makes the pty node readable and writable by the owner, and
writable by the group only when originally writable and when the group
is the tty one.
As a result, on a system wrongly mounted with gid=0 and mode=0600, the
pty nodes won't be accessible by the tty group, but the grantpt()
function will succeed and users will have a working system. The system
is not fully POSIX compliant (which might be an admin choice to default
to "mesg n" mode), but the GNU libc is not to blame here, as without the
pt_chown helper it can't do anything.
With this patch there should not be any reason left to build the GNU
libc with the --enable-pt_chown configure option on a GNU/Linux system.
The ldbl-128ibm implementation of logl is inaccurate for arguments
near 1, because when deciding whether to bypass a series expansion for
log(1+z), where z = x-1, it compares the square of z rather than z
itself with an epsilon value. This patch fixes that comparison, so
eliminating the test failures for inaccuracy of logl in such cases.
Tested for powerpc.
[BZ #19351]
* sysdeps/ieee754/ldbl-128ibm/e_logl.c (__ieee754_logl): When
expanding log(1+z), compare z rather than its square with epsilon
to determine when to avoid evaluating the expansion.
The ldbl-128ibm implementation of sinhl uses a slightly too small
overflow threshold (similar to bug 16407 for coshl). This patch fixes
it to use a safe threshold (so that values whose high part is above
the value compared with definitely result in an overflow in all
rounding modes).
Tested for powerpc.
[BZ #19350]
* sysdeps/ieee754/ldbl-128ibm/e_sinhl.c (__ieee754_sinhl):
Increase overflow threshold.
The ldbl-128ibm implementation of tanhl is inaccurate for small
arguments, because it returns x*(1+x) (maybe in an attempt to raise
"inexact") when x itself would be the accurate return value but
multiplying by 1+x introduces large errors. This patch fixes it to
return x in that case (when the mathematical result is x plus a
negligible remainder on the order of x^3) to avoid those errors.
Tested for powerpc.
[BZ #19349]
* sysdeps/ieee754/ldbl-128ibm/s_tanhl.c (__tanhl): Return argument
when small.
