It removes calling to pthread_mutex_lock() at the beginning of new
thread, which helps to support thread sanitizer.
Change-Id: Ia3601c476de7976a9177b792bd74bb200cee0e13
It is due to a previous change "Let g_thread_list_lock only protect g_thread_list".
We need to add the newly created thread to thread_list even if
__init_thread fails, so the thread can exit successfully.
Change-Id: I0332df11acfdd181350bcc092b12d90d679057a4
As glibc/netbsd don't protect access to thread struct members by a global
lock, we don't want to do it either. This change reduces the
responsibility of g_thread_list_lock to only protect g_thread_list.
Bug: 19636317
Change-Id: I897890710653dac165d8fa4452c7ecf74abdbf2b
aligned attribute can only control compiler's behavior, but we
are manually allocating pthread_internal_t. So we need to make
sure of alignment manually.
Change-Id: Iea4c46eadf10dfd15dc955c5f41cf6063cfd8536
1. Move the representation of thread join_state from pthread.attr.flag
to pthread.join_state. This clarifies thread state change.
2. Use atomic operations for pthread.join_state. So we don't need to
protect it by g_thread_list_lock. g_thread_list_lock will be reduced
to only protect g_thread_list or even removed in further changes.
Bug: 19636317
Change-Id: I31fb143a7c69508c7287307dd3b0776993ec0f43
Unfortunately, this change provokes random crashes for ART, and
I have seen libc crashes on the device that might be related to it.
Reverting it fixes the ART crashes. there is unfortunately no
stack trace for the crashes, but just a "Segmentation fault" message.
This reverts commit cc5f6543e3.
Change-Id: I68dca8e1e9b9edcce7eb84596e8db619e40e8052
This lock has been here since the original commits, but as far as I can tell
it never served any purpose. We've never had a free list of cached stacks or
anything like that.
Change-Id: I9d665c7eaa9c699ce0659ffb111402a0239fe1f5
gdb won't even try to use this on Android because it knows we don't
support old enough kernels to need it.
Bug: 15470251
Change-Id: Ia6d54585d888bbab8ee0490a148a1586b25437b9
In practice, with this implementation we never need to make a system call.
We get the main thread's tid (which is the same as our pid) back from
the set_tid_address system call we have to make during initialization.
A new pthread will have the same pid as its parent, and a fork child's
main (and only) thread will have a pid equal to its tid, which we get for
free from the kernel before clone returns.
The only time we'd actually have to make a getpid system call now is if
we take a signal during fork and the signal handler calls getpid. (That,
or we call getpid in the dynamic linker while it's still dealing with its
own relocations and hasn't even set up the main thread yet.)
Bug: 15387103
Change-Id: I6d4718ed0a5c912fc75b5f738c49a023dbed5189
The problem with the original patch was that using syscall(3) means that
errno can be set, but pthread_create(3) was abusing the TLS errno slot as
a pthread_mutex_t for the thread startup handshake.
There was also a mistake in the check for syscall failures --- it should
have checked against -1 instead of 0 (not just because that's the default
idiom, but also here because futex(2) can legitimately return values > 0).
This patch stops abusing the TLS errno slot and adds a pthread_mutex_t to
pthread_internal_t instead. (Note that for LP64 sizeof(pthread_mutex_t) >
sizeof(uintptr_t), so we could potentially clobber other TLS slots too.)
I've also rewritten the LP32 compatibility stubs to directly reuse the
code from the .h file.
This reverts commit 75c55ff84e.
Bug: 15195455
Change-Id: I6ffb13e5cf6a35d8f59f692d94192aae9ab4593d
Also let clone(2) set the TLS for x86.
Also ensure we initialize the TLS before we clone(2) for all architectures.
Change-Id: Ie5fa4466e1c9ee116a281dfedef574c5ba60c0b5
Also neuter __isthreaded.
We should come back to try to hide struct FILE's internals for LP64.
Bug: 3453512
Bug: 3453550
Change-Id: I7e115329fb4579246a72fea367b9fc8cb6055d18
Unlike other architectures, on x86 (but not x86-64), CLONE_SETTLS
takes a pointer to a struct user_desc instead of a pointer to the
TLS itself. Rather than have to deal with this here, let's just use
the old __set_tls mechanism we used to use (and still use for the
main thread on all architectures, so it's not going away any time
soon).
Bug: 11826724
Change-Id: I02a27939a73ae6cea1134a3f4c1dd7eafea479da
Let the kernel keep pthread_internal_t::tid updated, including
across forks and for the main thread. This then lets us fix
pthread_join to only return after the thread has really exited.
Also fix the thread attributes of the main thread so we don't
unmap the main thread's stack (which is really owned by the
dynamic linker and contains things like environment variables),
which fixes crashes when joining with an exited main thread
and also fixes problems reported publicly with accessing environment
variables after the main thread exits (for which I've added a new
unit test).
In passing I also fixed a bug where if the clone(2) inside
pthread_create(3) fails, we'd unmap the child's stack and TLS (which
contains the mutex) and then try to unlock the mutex. Boom! It wasn't
until after I'd uploaded the fix for this that I came across a new
public bug reporting this exact failure.
Bug: 8206355
Bug: 11693195
Bug: https://code.google.com/p/android/issues/detail?id=57421
Bug: https://code.google.com/p/android/issues/detail?id=62392
Change-Id: I2af9cf6e8ae510a67256ad93cad891794ed0580b
<pthread.h> was missing nonnull attributes, noreturn on pthread_exit,
and had incorrect cv qualifiers for several standard functions.
I've also marked the non-standard stuff (where I count glibc rather
than POSIX as "standard") so we can revisit this cruft for LP64 and
try to ensure we're compatible with glibc.
I've also broken out the pthread_cond* functions into a new file.
I've made the remaining pthread files (plus ptrace) part of the bionic code
and fixed all the warnings.
I've added a few more smoke tests for chunks of untested pthread functionality.
We no longer need the libc_static_common_src_files hack for any of the
pthread implementation because we long since stripped out the rest of
the armv5 support, and this hack was just to ensure that __get_tls in libc.a
went via the kernel if necessary.
This patch also finishes the job of breaking up the pthread.c monolith, and
adds a handful of new tests.
Change-Id: Idc0ae7f5d8aa65989598acd4c01a874fe21582c7
Also remove the SIGSEGV special case, which was probably because
hand-written __exit_with_stack_teardown stubs used to try to cause
SIGSEGV if the exit system call returned (which it never does, so
that dead code disappeared).
Also move the sigprocmask into the only case where it's necessary ---
the one where we unmap the stack that would be used by a signal
handler.
Change-Id: Ie40d20c1ae2f5e7125131b6b492cba7a2c6d08e9
We couldn't fix this for 32-bit because there's too much broken
code out there. (Pretty much everyone asks for real-time
scheduling for all their threads, and the kernel says "don't be
stupid".)
Change-Id: I43c5271e6b6bb91278b9a19eec08cbf05391e3c4
The x86_64 build was failing because clone.S had a call to __thread_entry which
was being added to a different intermediate .a on the way to making libc.so,
and the linker couldn't guarantee statically that such a relocation would be
possible.
ld: error: out/target/product/generic_x86_64/obj/STATIC_LIBRARIES/libc_common_intermediates/libc_common.a(clone.o): requires dynamic R_X86_64_PC32 reloc against '__thread_entry' which may overflow at runtime; recompile with -fPIC
This patch addresses that by ensuring that the caller and callee end up in the
same intermediate .a. While I'm here, I've tried to clean up some of the mess
that led to this situation too. In particular, this removes libc/private/ from
the default include path (except for the DNS code), and splits out the DNS
code into its own library (since it's a weird special case of upstream NetBSD
code that's diverged so heavily it's unlikely ever to get back in sync).
There's more cleanup of the DNS situation possible, but this is definitely a
step in the right direction, and it's more than enough to get x86_64 building
cleanly.
Change-Id: I00425a7245b7a2573df16cc38798187d0729e7c4
If __get_tls has the right type, a lot of confusing casting can disappear.
It was probably a mistake that __get_tls was exposed as a function for mips
and x86 (but not arm), so let's (a) ensure that the __get_tls function
always matches the macro, (b) that we have the function for arm too, and
(c) that we don't have the function for any 64-bit architecture.
Change-Id: Ie9cb989b66e2006524ad7733eb6e1a65055463be
This reverts commits eb1b07469f and
d14dc3b87f, and fixes the bug where
we were calling mmap (which might cause errno to be set) before
__set_tls (which is required to implement errno).
Bug: 8557703
Change-Id: I2c36d00240c56e156e1bb430d8c22a73a068b70c
We notify debuggerd of problems by installing signal handlers. That's
fine except for when the signal is caused by us running off the end of
a thread's stack and into the guard page.
Bug: 8557703
Change-Id: I1ef65b4bb3bbca7e9a9743056177094921e60ed3
pthread_getattr_np was reporting the values supplied to us, not the values we
actually used, which is kinda the whole point of pthread_getattr_np.
pthread_attr_setguardsize and pthread_attr_setstacksize were reporting EINVAL
for any size that wasn't a multiple of the system page size. This is
unnecessary. We can just round like POSIX suggests and glibc already does.
Also improve the error reporting for pthread_create failures.
Change-Id: I7ebc518628a8a1161ec72e111def911d500bba71
Removed 'join_count' from pthread_internal_t and switched to using the flag
PTHREAD_ATTR_FLAG_JOINED to indicate if a thread is being joined. Combined with
a switch to a while loop in pthread_join, this fixes spurious wake-ups but
prevents a thread from being joined multiple times. This is fine for
two reasons:
1) The pthread_join specification allows for undefined behavior when multiple
threads try to join a single thread.
2) There is no thread safe way to allow multiple threads to join a single
thread with the pthread interface. The second thread calling pthread_join
could be pre-empted until the thread is destroyed and its handle reused for
a different thread. Therefore multi-join is always an error.
Bug: https://code.google.com/p/android/issues/detail?id=52255
Change-Id: I8b6784d47620ffdcdbfb14524e7402e21d46c5f7
pthread_create returns EAGAIN when it can't allocate a pthread_internal_t,
when it can't allocate a stack for the new thread, or when clone(2) fails
because there are too many threads. It's useful to be able to know why your
pthread_create just failed, so add some logging.
Bug: 8470684
Change-Id: I1bb4497d4f7528eacce0db35c2014771cba64569
We only need one logging API, and I prefer the one that does no
allocation and is thus safe to use in any context.
Also use O_CLOEXEC when opening the /dev/log files.
Move everything logging-related into one header file.
Change-Id: Ic1e3ea8e9b910dc29df351bff6c0aa4db26fbb58
