2020-06-15 14:18:57 +00:00
|
|
|
/*-*- mode:c;indent-tabs-mode:nil;c-basic-offset:2;tab-width:8;coding:utf-8 -*-│
|
|
|
|
|
│vi: set net ft=c ts=2 sts=2 sw=2 fenc=utf-8 :vi│
|
|
|
|
|
╞══════════════════════════════════════════════════════════════════════════════╡
|
2022-06-15 23:19:50 +00:00
|
|
|
│ Copyright 2022 Justine Alexandra Roberts Tunney │
|
2020-06-15 14:18:57 +00:00
|
|
|
│ │
|
2020-12-28 01:18:44 +00:00
|
|
|
│ Permission to use, copy, modify, and/or distribute this software for │
|
|
|
|
|
│ any purpose with or without fee is hereby granted, provided that the │
|
|
|
|
|
│ above copyright notice and this permission notice appear in all copies. │
|
2020-06-15 14:18:57 +00:00
|
|
|
│ │
|
2020-12-28 01:18:44 +00:00
|
|
|
│ THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL │
|
|
|
|
|
│ WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED │
|
|
|
|
|
│ WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE │
|
|
|
|
|
│ AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL │
|
|
|
|
|
│ DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR │
|
|
|
|
|
│ PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER │
|
|
|
|
|
│ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR │
|
|
|
|
|
│ PERFORMANCE OF THIS SOFTWARE. │
|
2020-06-15 14:18:57 +00:00
|
|
|
╚─────────────────────────────────────────────────────────────────────────────*/
|
2023-06-03 15:12:13 +00:00
|
|
|
#include "ape/sections.internal.h"
|
2022-06-26 01:17:31 +00:00
|
|
|
#include "libc/calls/syscall-sysv.internal.h"
|
2023-07-04 00:35:11 +00:00
|
|
|
#include "libc/dce.h"
|
2022-06-15 23:19:50 +00:00
|
|
|
#include "libc/errno.h"
|
2022-09-10 18:49:13 +00:00
|
|
|
#include "libc/intrin/asan.internal.h"
|
|
|
|
|
#include "libc/intrin/asancodes.h"
|
2022-10-13 20:44:41 +00:00
|
|
|
#include "libc/intrin/atomic.h"
|
2023-07-06 13:57:28 +00:00
|
|
|
#include "libc/intrin/dll.h"
|
2022-08-20 19:32:51 +00:00
|
|
|
#include "libc/intrin/weaken.h"
|
2022-07-10 11:01:17 +00:00
|
|
|
#include "libc/macros.internal.h"
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
#include "libc/nt/files.h"
|
|
|
|
|
#include "libc/nt/runtime.h"
|
|
|
|
|
#include "libc/nt/synchronization.h"
|
|
|
|
|
#include "libc/nt/thread.h"
|
2022-06-15 23:19:50 +00:00
|
|
|
#include "libc/runtime/internal.h"
|
|
|
|
|
#include "libc/runtime/runtime.h"
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
#include "libc/runtime/syslib.internal.h"
|
2023-08-21 16:00:40 +00:00
|
|
|
#include "libc/str/locale.h"
|
2023-05-10 05:41:57 +00:00
|
|
|
#include "libc/str/str.h"
|
2022-10-11 00:52:41 +00:00
|
|
|
#include "libc/thread/posixthread.internal.h"
|
2022-11-09 11:58:57 +00:00
|
|
|
#include "libc/thread/thread.h"
|
2022-09-10 09:56:25 +00:00
|
|
|
#include "libc/thread/tls.h"
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
#include "third_party/make/gnumake.h"
|
2022-07-10 11:01:17 +00:00
|
|
|
|
2023-05-10 05:41:57 +00:00
|
|
|
#define I(x) ((uintptr_t)x)
|
2020-06-15 14:18:57 +00:00
|
|
|
|
2022-07-10 15:27:50 +00:00
|
|
|
extern unsigned char __tls_mov_nt_rax[];
|
|
|
|
|
extern unsigned char __tls_add_nt_rax[];
|
2022-11-09 11:58:57 +00:00
|
|
|
|
2023-05-10 05:41:57 +00:00
|
|
|
_Alignas(TLS_ALIGNMENT) static char __static_tls[6016];
|
2022-07-10 15:27:50 +00:00
|
|
|
|
2022-07-18 10:33:32 +00:00
|
|
|
/**
|
2022-07-21 16:16:38 +00:00
|
|
|
* Enables thread local storage for main process.
|
|
|
|
|
*
|
2023-05-10 05:41:57 +00:00
|
|
|
* Here's the TLS memory layout on x86_64:
|
|
|
|
|
*
|
|
|
|
|
* __get_tls()
|
2022-07-21 16:16:38 +00:00
|
|
|
* │
|
2023-05-10 05:41:57 +00:00
|
|
|
* %fs Linux/BSDs
|
|
|
|
|
* _Thread_local │
|
2022-07-21 16:16:38 +00:00
|
|
|
* ┌───┬──────────┬──────────┼───┐
|
|
|
|
|
* │pad│ .tdata │ .tbss │tib│
|
|
|
|
|
* └───┴──────────┴──────────┼───┘
|
|
|
|
|
* │
|
|
|
|
|
* Windows/Mac %gs
|
2022-07-19 05:26:11 +00:00
|
|
|
*
|
2023-05-10 05:41:57 +00:00
|
|
|
* Here's the TLS memory layout on aarch64:
|
|
|
|
|
*
|
2023-05-19 02:05:08 +00:00
|
|
|
* x28
|
2023-05-10 05:41:57 +00:00
|
|
|
* %tpidr_el0
|
|
|
|
|
* │
|
|
|
|
|
* │ _Thread_local
|
|
|
|
|
* ┌───┼───┬──────────┬──────────┐
|
|
|
|
|
* │tib│dtv│ .tdata │ .tbss │
|
|
|
|
|
* ├───┴───┴──────────┴──────────┘
|
|
|
|
|
* │
|
|
|
|
|
* __get_tls()
|
|
|
|
|
*
|
2022-07-19 05:26:11 +00:00
|
|
|
* This function is always called by the core runtime to guarantee TLS
|
|
|
|
|
* is always available to your program. You must build your code using
|
|
|
|
|
* -mno-tls-direct-seg-refs if you want to use _Thread_local.
|
|
|
|
|
*
|
|
|
|
|
* You can use __get_tls() to get the linear address of your tib. When
|
|
|
|
|
* accessing TLS via privileged code you must use __get_tls_privileged
|
|
|
|
|
* because we need code morphing to support The New Technology and XNU
|
|
|
|
|
*
|
|
|
|
|
* On XNU and The New Technology, this function imposes 1ms of latency
|
|
|
|
|
* during startup for larger binaries like Python.
|
|
|
|
|
*
|
|
|
|
|
* If you don't want TLS and you're sure you're not using it, then you
|
|
|
|
|
* can disable it as follows:
|
|
|
|
|
*
|
|
|
|
|
* int main() {
|
2023-05-19 02:05:08 +00:00
|
|
|
* __tls_enabled_set(false);
|
2022-07-19 05:26:11 +00:00
|
|
|
* // do stuff
|
|
|
|
|
* }
|
|
|
|
|
*
|
|
|
|
|
* This is useful if you want to wrestle back control of %fs using the
|
|
|
|
|
* arch_prctl() function. However, such programs might not be portable
|
|
|
|
|
* and your `errno` variable also won't be thread safe anymore.
|
2022-07-18 10:33:32 +00:00
|
|
|
*/
|
2023-05-10 08:10:28 +00:00
|
|
|
textstartup void __enable_tls(void) {
|
2022-10-13 20:44:41 +00:00
|
|
|
int tid;
|
2023-09-02 03:49:13 +00:00
|
|
|
size_t siz;
|
2022-07-10 11:01:17 +00:00
|
|
|
char *mem, *tls;
|
2023-05-10 05:41:57 +00:00
|
|
|
struct CosmoTib *tib;
|
2022-11-09 11:58:57 +00:00
|
|
|
|
2023-05-10 05:41:57 +00:00
|
|
|
// Here's the layout we're currently using:
|
|
|
|
|
//
|
2023-07-30 01:44:15 +00:00
|
|
|
// .balign 4096
|
2023-05-10 05:41:57 +00:00
|
|
|
// _tdata_start:
|
|
|
|
|
// .tdata
|
|
|
|
|
// _tdata_size = . - _tdata_start
|
2023-07-30 01:44:15 +00:00
|
|
|
// .balign 4096
|
2023-05-10 05:41:57 +00:00
|
|
|
// _tbss_start:
|
|
|
|
|
// _tdata_start + _tbss_offset:
|
|
|
|
|
// .tbss
|
2023-05-19 02:05:08 +00:00
|
|
|
// .balign TLS_ALIGNMENT
|
2023-05-10 05:41:57 +00:00
|
|
|
// _tbss_size = . - _tbss_start
|
|
|
|
|
// _tbss_end:
|
|
|
|
|
// _tbss_start + _tbss_size:
|
|
|
|
|
// _tdata_start + _tls_size:
|
|
|
|
|
//
|
2023-09-21 14:30:39 +00:00
|
|
|
// unassert(_tbss_start == _tdata_start + I(_tbss_offset));
|
|
|
|
|
// unassert(_tbss_start + I(_tbss_size) == _tdata_start + I(_tls_size));
|
2023-05-10 05:41:57 +00:00
|
|
|
|
|
|
|
|
#ifdef __x86_64__
|
|
|
|
|
|
2023-07-10 11:29:46 +00:00
|
|
|
siz = ROUNDUP(I(_tls_size) + sizeof(*tib), TLS_ALIGNMENT);
|
2022-07-19 05:26:11 +00:00
|
|
|
if (siz <= sizeof(__static_tls)) {
|
|
|
|
|
// if tls requirement is small then use the static tls block
|
|
|
|
|
// which helps avoid a system call for appes with little tls
|
|
|
|
|
// this is crucial to keeping life.com 16 kilobytes in size!
|
|
|
|
|
mem = __static_tls;
|
|
|
|
|
} else {
|
|
|
|
|
// if this binary needs a hefty tls block then we'll bank on
|
|
|
|
|
// malloc() being linked, which links _mapanon(). otherwise
|
2023-07-26 20:54:49 +00:00
|
|
|
// if you exceed this, you need to __static_yoink("_mapanon").
|
2022-07-19 05:26:11 +00:00
|
|
|
// please note that it's probably too early to call calloc()
|
2022-09-13 06:10:38 +00:00
|
|
|
mem = _weaken(_mapanon)(siz);
|
2022-07-19 05:26:11 +00:00
|
|
|
}
|
2022-11-09 11:58:57 +00:00
|
|
|
|
2022-09-10 18:49:13 +00:00
|
|
|
if (IsAsan()) {
|
|
|
|
|
// poison the space between .tdata and .tbss
|
2023-05-10 05:41:57 +00:00
|
|
|
__asan_poison(mem + I(_tdata_size), I(_tbss_offset) - I(_tdata_size),
|
2022-09-10 18:49:13 +00:00
|
|
|
kAsanProtected);
|
|
|
|
|
}
|
2022-11-09 11:58:57 +00:00
|
|
|
|
2023-05-10 05:41:57 +00:00
|
|
|
tib = (struct CosmoTib *)(mem + siz - sizeof(*tib));
|
|
|
|
|
tls = mem + siz - sizeof(*tib) - I(_tls_size);
|
|
|
|
|
|
|
|
|
|
#elif defined(__aarch64__)
|
|
|
|
|
|
2023-09-02 03:49:13 +00:00
|
|
|
size_t hiz = ROUNDUP(sizeof(*tib) + 2 * sizeof(void *), I(_tls_align));
|
2023-06-03 15:12:13 +00:00
|
|
|
siz = hiz + I(_tls_size);
|
2023-05-10 05:41:57 +00:00
|
|
|
if (siz <= sizeof(__static_tls)) {
|
|
|
|
|
mem = __static_tls;
|
|
|
|
|
} else {
|
|
|
|
|
mem = _weaken(_mapanon)(siz);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (IsAsan()) {
|
|
|
|
|
// there's a roundup(pagesize) gap between .tdata and .tbss
|
|
|
|
|
// poison that empty space
|
2023-06-03 15:12:13 +00:00
|
|
|
__asan_poison(mem + hiz + I(_tdata_size), I(_tbss_offset) - I(_tdata_size),
|
|
|
|
|
kAsanProtected);
|
2023-05-10 05:41:57 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
tib = (struct CosmoTib *)mem;
|
2023-06-03 15:12:13 +00:00
|
|
|
tls = mem + hiz;
|
2023-05-10 05:41:57 +00:00
|
|
|
|
|
|
|
|
// Set the DTV.
|
|
|
|
|
//
|
|
|
|
|
// We don't support dynamic shared objects at the moment. The APE
|
|
|
|
|
// linker script will only produce a single PT_TLS program header
|
|
|
|
|
// therefore our job is relatively simple.
|
|
|
|
|
//
|
|
|
|
|
// @see musl/src/env/__init_tls.c
|
|
|
|
|
// @see https://chao-tic.github.io/blog/2018/12/25/tls
|
|
|
|
|
((uintptr_t *)tls)[-2] = 1;
|
|
|
|
|
((void **)tls)[-1] = tls;
|
|
|
|
|
|
|
|
|
|
#else
|
|
|
|
|
#error "unsupported architecture"
|
|
|
|
|
#endif /* __x86_64__ */
|
|
|
|
|
|
2022-11-09 11:58:57 +00:00
|
|
|
// initialize main thread tls memory
|
2022-09-10 09:56:25 +00:00
|
|
|
tib->tib_self = tib;
|
|
|
|
|
tib->tib_self2 = tib;
|
|
|
|
|
tib->tib_errno = __errno;
|
2022-11-02 05:36:03 +00:00
|
|
|
tib->tib_strace = __strace;
|
|
|
|
|
tib->tib_ftrace = __ftrace;
|
2023-08-21 16:00:40 +00:00
|
|
|
tib->tib_locale = (intptr_t)&__c_dot_utf8_locale;
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
tib->tib_pthread = (pthread_t)&_pthread_static;
|
|
|
|
|
if (IsWindows()) {
|
|
|
|
|
intptr_t threadhand, pseudo = GetCurrentThread();
|
|
|
|
|
DuplicateHandle(GetCurrentProcess(), pseudo, GetCurrentProcess(),
|
|
|
|
|
&threadhand, 0, false, kNtDuplicateSameAccess);
|
|
|
|
|
atomic_store_explicit(&tib->tib_syshand, threadhand, memory_order_relaxed);
|
|
|
|
|
} else if (IsXnuSilicon()) {
|
|
|
|
|
tib->tib_syshand = __syslib->__pthread_self();
|
|
|
|
|
}
|
2023-06-04 08:57:10 +00:00
|
|
|
if (IsLinux() || IsXnuSilicon()) {
|
2022-07-19 05:26:11 +00:00
|
|
|
// gnu/systemd guarantees pid==tid for the main thread so we can
|
|
|
|
|
// avoid issuing a superfluous system call at startup in program
|
2022-10-13 20:44:41 +00:00
|
|
|
tid = __pid;
|
2022-07-19 05:26:11 +00:00
|
|
|
} else {
|
2022-10-13 20:44:41 +00:00
|
|
|
tid = sys_gettid();
|
2022-07-19 05:26:11 +00:00
|
|
|
}
|
2022-10-13 20:44:41 +00:00
|
|
|
atomic_store_explicit(&tib->tib_tid, tid, memory_order_relaxed);
|
2022-11-09 11:58:57 +00:00
|
|
|
|
|
|
|
|
// initialize posix threads
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
_pthread_static.tib = tib;
|
|
|
|
|
_pthread_static.pt_flags = PT_STATIC;
|
|
|
|
|
dll_init(&_pthread_static.list);
|
|
|
|
|
_pthread_list = &_pthread_static.list;
|
|
|
|
|
atomic_store_explicit(&_pthread_static.ptid, tid, memory_order_relaxed);
|
|
|
|
|
if (IsWindows()) {
|
2023-09-21 14:30:39 +00:00
|
|
|
if (!(_pthread_static.semaphore = CreateSemaphore(0, 0, 1, 0))) {
|
|
|
|
|
notpossible;
|
|
|
|
|
}
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
}
|
2022-11-09 11:58:57 +00:00
|
|
|
|
|
|
|
|
// copy in initialized data section
|
2023-05-10 05:41:57 +00:00
|
|
|
if (I(_tdata_size)) {
|
2023-07-04 00:35:11 +00:00
|
|
|
if (IsAsan()) {
|
|
|
|
|
__asan_memcpy(tls, _tdata_start, I(_tdata_size));
|
|
|
|
|
} else {
|
|
|
|
|
memcpy(tls, _tdata_start, I(_tdata_size));
|
|
|
|
|
}
|
2023-05-10 05:41:57 +00:00
|
|
|
}
|
2022-07-10 11:01:17 +00:00
|
|
|
|
|
|
|
|
// ask the operating system to change the x86 segment register
|
2022-11-02 05:36:03 +00:00
|
|
|
__set_tls(tib);
|
2022-07-10 11:01:17 +00:00
|
|
|
|
2023-05-02 02:43:59 +00:00
|
|
|
#ifdef __x86_64__
|
2022-11-02 05:36:03 +00:00
|
|
|
// rewrite the executable tls opcodes in memory
|
2023-07-10 02:47:46 +00:00
|
|
|
if (IsWindows() || IsXnu()) {
|
|
|
|
|
__morph_tls();
|
|
|
|
|
}
|
2023-05-02 02:43:59 +00:00
|
|
|
#endif
|
2022-07-10 11:01:17 +00:00
|
|
|
|
|
|
|
|
// we are now allowed to use tls
|
2023-05-19 02:05:08 +00:00
|
|
|
__tls_enabled_set(true);
|
2022-06-15 23:19:50 +00:00
|
|
|
}
|