Previously we blindly tried to use libunwind whenever header is
detected. Even if actually working libunwind library is missing. This
is now fixed, so we attempt to use libunwind when it actually works.
Somehow recent freebsd ships libunwind.h (which seems to belong to
llvm's implementation), but apparently without matching .so. So then building
and linking failed.
We got bitten again by compiler optimizing out free(malloc(sz))
combination. We replace calls to malloc/free with calls to global
function operator new/delete. Those are apparently forbidden by
standard to be optimized out.
If for some reason benchmark function is too fast (like when it got
optimized out to nothing), we'd previously hang in infinite loop. Now
we'll catch this condition due to iterations count overflowing.
On modern machines xoring volatile int is fast enough to occasionally
make specific test case's profiling results not stable enough. So we
now force atomics into the picture which is slightly slower (could be
lots on some platforms and/or optimizer flags).
As part of that we also upgrade required autoconf version to 2.69
which is what I see in rhel/centos 7 and ubuntu 14.04. Both are old
enough. And, of course, .tar.gz releases still ship "packaged" configure,
so will work on older distros.
This fixes issue #1335.
In Monterey, it seems that free_definite_size is required (despite
the documentation saying that is optional). The implementation
just forward the call to free.
Signed-off-by: Natale Patriciello <natale.patriciello@gmail.com>
This commit replaces the "golang" moniker in the README file with the
appropriate name of "Go" per https://golang.org/doc/faq#go_or_golang.
In this context, no disambiguation is required.
We previously tested wrong assumption that larger than page size size
classes have addresses aligned on page size. New code is making proper
check of size class.
Also added is unit test coverage for this previously failing
condition. And we now also run "assert-ful" unittests for big tcmalloc
too, not only tcmalloc_minimal configuration.
This fixes github issue #1254
tcmalloc contains some thread-safety annotations; however those
annotations have not been exercised for some time, as they used
macros/attributes only supported by a legacy branch of gcc.
Pull request #1251 converted those macros to support modern
versions of clang; this CR fixes the annotations that were
enabled. For the most part, this just requires re-enabling
annotations on member functions that take/release locks. For the
tcmalloc fork (pre-fork and post-fork) handlers, we mark the
functions as exempt from this analysis, as it takes a dynamic
number of locks.
tcmalloc contains its own copy of thread_annotations.h, wrapper
macros for static thread-safety analysis expressions. These thread-
safety expressions allow asserting (at compile time) that certain
locks are held or excluded or certain data is protected by specific
locks; they are checked at compile-time by recent versions of clang
or a gcc branch (https://gcc.gnu.org/wiki/ThreadSafetyAnnotation).
Convert the #if-guard and macro names from the no-longer-supported
gcc branch's defines & macros to the versions supported by recent
versions of clang.
Older glibc seemingly doesn't have right unwind info for signal frame,
so fails recently upgraded stacktrace_unittest. But since this version
is not supported anymore, lets just test newer glibc.
It is actually needed for libgcc backtracer from time to time. And
we've seen libunwind to need it too. Plus we've not heard of any
problems with it. So lets just always enable it.
This should fix github issue #1248.
Libunwind is mostly safe on x86-64 and most of OS (e.g. libc) is
compiled without frame pointers anyways. Even when libc is built with
frame pointers, various libc asm routines (e.g. memcpy) are not. And
even for code built with frame pointers we actually sometimes get
SIGPROF inside function prologues, before frame pointer is
established. Which then causes imperfect backtraces.
Libunwind should handle this all (when it doesn't crash or deadlock).
This supports frame pointer backtracing on x86-64, aarch64 and
riscv-s (should work for both 32 and 64 bits).
Also added is detection of borked libunwind on aarch64-s. In this case
frame pointer unwinder is preferred.
Previously it only was respected on x86_64, but this days lots of modern
ABIs are without frame pointers by default (e.g. arm64 and riscv, and
even older mips).
