Added support for TX lock elision of pthread mutexes on s390 and
s390x. This may improve lock scaling of existing programs on TX
capable systems. The lock elision code is only built with
--enable-lock-elision=yes and then requires a GCC version supporting
the TX builtins. With lock elision default mutexes are elided via
__builtin_tbegin, if the cpu supports transactions. By default lock
elision is not enabled and the elision code is not built.
This patch removes the arch specific powerpc implementation and instead
uses the linux default one. Although the current powerpc implementation
already constains the required memory barriers for correct
initialization, the default implementation shows a better performance on
newer chips.
[BZ #15215] This unifies various pthread_once architecture-specific
implementations which were using the same algorithm with slightly different
implementations. It also adds missing memory barriers that are required for
correctness.
This patch moves the __PTHREAD_SPINS definition to arch specific header
since pthread_mutex_t layout is also arch specific. This leads to no
need to defining __PTHREAD_MUTEX_HAVE_ELISION and thus removing of the
undefined compiler warning.
When profiling programs with lock problems with perf record -g dwarf,
libunwind can currently not backtrace through the futex and unlock
functions in pthread. This is because they use out of line sections,
and those are not correctly described in dwarf2 (I believe needs
dwarf3 or 4).
This patch first removes the out of line sections. They only save a
single jump, but cause a lot of pain. Then it converts the now inline
lock code to use the now standard gas .cfi_* commands.
With these changes libunwind/perf can backtrace through the futex
functions now.
Longer term it would be likely better to just use C futex() functions
on x86 like all the other architectures. This would clean the code up
even more.
Support for /proc/self/task/$tid/comm as added in Linux 2.6.33,
therefore since the test tst-setgetname relies on this functionality
to operate we must skip the test in kernels < 2.6.33. We wrap the
checks with __ASSUME_PROC_PID_TASK_COMM such that in the future when
we move arch_minimum_kernel to 2.6.33 we can remove this code.
This commit adds a testcase for pthread_setname_np
and pthread_getname_np. The testcase itself has
four tests to validate that these functions work
as expected. The test is only enabled for Linux
since it requires access to an alternate method
for validating the functions work.
This updates glibc for the changes in the ELFv2 relating to the
stack frame layout. These are described in more detail here:
http://gcc.gnu.org/ml/gcc-patches/2013-11/msg01149.htmlhttp://gcc.gnu.org/ml/gcc-patches/2013-11/msg01146.html
Specifically, the "compiler and linker doublewords" were removed,
which has the effect that the save slot for the TOC register is
now at offset 24 rather than 40 to the stack pointer.
In addition, a function may now no longer necessarily assume that
its caller has set up a 64-byte register save area its use.
To address the first change, the patch goes through all assembler
files and replaces immediate offsets in instructions accessing the
ABI-defined stack slots by symbolic offsets. Those already were
defined in ucontext_i.sym and used in some of the context routines,
but that doesn't really seem like the right place for those defines.
The patch instead defines those symbolic offsets in sysdeps.h,
in two variants for the old and new ABI, and uses them systematically
in all assembler files, not just the context routines.
The second change only affected a few assembler files that used
the save area to temporarily store some registers. In those
cases where this happens within a leaf function, this patch
changes the code to store those registers to the "red zone"
below the stack pointer. Otherwise, the functions already allocate
a stack frame, and the patch changes them to add extra space in
these frames as temporary space for the ELFv2 ABI.
http://sourceware.org/ml/libc-alpha/2013-08/msg00090.html
This patch fixes symbol versioning in setjmp/longjmp. The existing
code uses raw versions, which results in wrong symbol versioning when
you want to build glibc with a base version of 2.19 for LE.
Note that the merging the 64-bit and 32-bit versions in novmx-lonjmp.c
and pt-longjmp.c doesn't result in GLIBC_2.0 versions for 64-bit, due
to the base in shlib_versions.
* sysdeps/powerpc/longjmp.c: Use proper symbol versioning macros.
* sysdeps/powerpc/novmx-longjmp.c: Likewise.
* sysdeps/powerpc/powerpc32/bsd-_setjmp.S: Likewise.
* sysdeps/powerpc/powerpc32/bsd-setjmp.S: Likewise.
* sysdeps/powerpc/powerpc32/fpu/__longjmp.S: Likewise.
* sysdeps/powerpc/powerpc32/fpu/setjmp.S: Likewise.
* sysdeps/powerpc/powerpc32/mcount.c: Likewise.
* sysdeps/powerpc/powerpc32/setjmp.S: Likewise.
* sysdeps/powerpc/powerpc64/setjmp.S: Likewise.
* nptl/sysdeps/unix/sysv/linux/powerpc/pt-longjmp.c: Likewise.
Add elision paths to the basic mutex locks.
The normal path has a check for RTM and upgrades the lock
to RTM when available. Trylocks cannot automatically upgrade,
so they check for elision every time.
We use a 4 byte value in the mutex to store the lock
elision adaptation state. This is separate from the adaptive
spin state and uses a separate field.
Condition variables currently do not support elision.
Recursive mutexes and condition variables may be supported at some point,
but are not in the current implementation. Also "trylock" will
not automatically enable elision unless some other lock call
has been already called on the lock.
This version does not use IFUNC, so it means every lock has one
additional check for elision. Benchmarking showed the overhead
to be negligible.
Lock elision using TSX is a technique to optimize lock scaling
It allows to run locks in parallel using hardware support for
a transactional execution mode in 4th generation Intel Core CPUs.
See http://www.intel.com/software/tsx for more Information.
This patch implements a simple adaptive lock elision algorithm based
on RTM. It enables elision for the pthread mutexes and rwlocks.
The algorithm keeps track whether a mutex successfully elides or not,
and stops eliding for some time when it is not.
When the CPU supports RTM the elision path is automatically tried,
otherwise any elision is disabled.
The adaptation algorithm and its tuning is currently preliminary.
The code adds some checks to the lock fast paths. Micro-benchmarks
show little to no difference without RTM.
This patch implements the low level "lll_" code for lock elision.
Followon patches hook this into the pthread implementation
Changes with the RTM mutexes:
-----------------------------
Lock elision in pthreads is generally compatible with existing programs.
There are some obscure exceptions, which are expected to be uncommon.
See the manual for more details.
- A broken program that unlocks a free lock will crash.
There are ways around this with some tradeoffs (more code in hot paths)
I'm still undecided on what approach to take here; have to wait for testing reports.
- pthread_mutex_destroy of a lock mutex will not return EBUSY but 0.
- There's also a similar situation with trylock outside the mutex,
"knowing" that the mutex must be held due to some other condition.
In this case an assert failure cannot be recovered. This situation is
usually an existing bug in the program.
- Same applies to the rwlocks. Some of the return values changes
(for example there is no EDEADLK for an elided lock, unless it aborts.
However when elided it will also never deadlock of course)
- Timing changes, so broken programs that make assumptions about specific timing
may expose already existing latent problems. Note that these broken programs will
break in other situations too (loaded system, new faster hardware, compiler
optimizations etc.)
- Programs with non recursive mutexes that take them recursively in a thread and
which would always deadlock without elision may not always see a deadlock.
The deadlock will only happen on an early or delayed abort (which typically
happens at some point)
This only happens for mutexes not explicitely set to PTHREAD_MUTEX_NORMAL
or PTHREAD_MUTEX_ADAPTIVE_NP. PTHREAD_MUTEX_NORMAL mutexes do not elide.
The elision default can be set at configure time.
This patch implements the basic infrastructure for elision.
Resolves BZ #15618.
pthread_attr_getaffinity_np may write beyond bounds of the input
cpuset buffer if the size of the input buffer is smaller than the
buffer present in the input pthread attributes. Fix is to copy to the
extent of the minimum of the source and the destination.
It is very very possible that the futex syscall returns an
error and that the caller of lll_futex_wake may want to
look at that error and propagate the failure.
This patch allows a caller to see the syscall error.
There are no users of the syscall error at present, but
future cleanups are now be able to check for the error.
--
nplt/
2013-06-10 Carlos O'Donell <carlos@redhat.com>
* sysdeps/unix/sysv/linux/i386/lowlevellock.h
(lll_futex_wake): Return syscall error.
* sysdeps/unix/sysv/linux/x86_64/lowlevellock.h
(lll_futex_wake): Return syscall error.
