2022-05-13 00:52:13 +00:00
|
|
|
/*-*- mode:c;indent-tabs-mode:nil;c-basic-offset:2;tab-width:8;coding:utf-8 -*-│
|
|
|
|
│ vi: set et ft=c ts=2 sts=2 sw=2 fenc=utf-8 :vi │
|
|
|
|
╞══════════════════════════════════════════════════════════════════════════════╡
|
|
|
|
│ Copyright 2021 Justine Alexandra Roberts Tunney │
|
|
|
|
│ │
|
|
|
|
│ 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. │
|
|
|
|
│ │
|
|
|
|
│ 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. │
|
|
|
|
╚─────────────────────────────────────────────────────────────────────────────*/
|
2023-06-04 01:32:33 +00:00
|
|
|
#include "libc/atomic.h"
|
Make improvements
- We now serialize the file descriptor table when spawning / executing
processes on Windows. This means you can now inherit more stuff than
just standard i/o. It's needed by bash, which duplicates the console
to file descriptor #255. We also now do a better job serializing the
environment variables, so you're less likely to encounter E2BIG when
using your bash shell. We also no longer coerce environ to uppercase
- execve() on Windows now remotely controls its parent process to make
them spawn a replacement for itself. Then it'll be able to terminate
immediately once the spawn succeeds, without having to linger around
for the lifetime as a shell process for proxying the exit code. When
process worker thread running in the parent sees the child die, it's
given a handle to the new child, to replace it in the process table.
- execve() and posix_spawn() on Windows will now provide CreateProcess
an explicit handle list. This allows us to remove handle locks which
enables better fork/spawn concurrency, with seriously correct thread
safety. Other codebases like Go use the same technique. On the other
hand fork() still favors the conventional WIN32 inheritence approach
which can be a little bit messy, but is *controlled* by guaranteeing
perfectly clean slates at both the spawning and execution boundaries
- sigset_t is now 64 bits. Having it be 128 bits was a mistake because
there's no reason to use that and it's only supported by FreeBSD. By
using the system word size, signal mask manipulation on Windows goes
very fast. Furthermore @asyncsignalsafe funcs have been rewritten on
Windows to take advantage of signal masking, now that it's much more
pleasant to use.
- All the overlapped i/o code on Windows has been rewritten for pretty
good signal and cancelation safety. We're now able to ensure overlap
data structures are cleaned up so long as you don't longjmp() out of
out of a signal handler that interrupted an i/o operation. Latencies
are also improved thanks to the removal of lots of "busy wait" code.
Waits should be optimal for everything except poll(), which shall be
the last and final demon we slay in the win32 i/o horror show.
- getrusage() on Windows is now able to report RUSAGE_CHILDREN as well
as RUSAGE_SELF, thanks to aggregation in the process manager thread.
2023-10-08 12:36:18 +00:00
|
|
|
#include "libc/calls/state.internal.h"
|
2022-05-13 00:52:13 +00:00
|
|
|
#include "libc/calls/struct/ucontext-netbsd.internal.h"
|
2022-05-16 20:20:08 +00:00
|
|
|
#include "libc/dce.h"
|
2025-01-03 02:44:07 +00:00
|
|
|
#include "libc/intrin/asmflag.h"
|
2023-06-04 01:32:33 +00:00
|
|
|
#include "libc/intrin/atomic.h"
|
2023-10-03 21:47:20 +00:00
|
|
|
#include "libc/intrin/ulock.h"
|
2022-05-27 20:25:46 +00:00
|
|
|
#include "libc/limits.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/mem/alloca.h"
|
|
|
|
#include "libc/nt/enum/processcreationflags.h"
|
2022-05-13 00:52:13 +00:00
|
|
|
#include "libc/nt/runtime.h"
|
2022-09-16 21:02:06 +00:00
|
|
|
#include "libc/nt/synchronization.h"
|
2022-05-13 00:52:13 +00:00
|
|
|
#include "libc/nt/thread.h"
|
|
|
|
#include "libc/nt/thunk/msabi.h"
|
|
|
|
#include "libc/runtime/runtime.h"
|
2023-06-04 01:32:33 +00:00
|
|
|
#include "libc/runtime/syslib.internal.h"
|
2022-11-08 18:09:47 +00:00
|
|
|
#include "libc/sock/internal.h"
|
2024-01-06 04:36:57 +00:00
|
|
|
#include "libc/sysv/consts/arch.h"
|
2022-05-13 00:52:13 +00:00
|
|
|
#include "libc/thread/freebsd.internal.h"
|
2022-06-26 01:17:31 +00:00
|
|
|
#include "libc/thread/openbsd.internal.h"
|
2024-12-18 12:59:02 +00:00
|
|
|
#include "libc/thread/posixthread.internal.h"
|
2022-05-13 00:52:13 +00:00
|
|
|
#include "libc/thread/xnu.internal.h"
|
|
|
|
|
2023-06-04 08:57:10 +00:00
|
|
|
#define kMaxThreadIds 32768
|
|
|
|
#define kMinThreadId 262144
|
|
|
|
|
2024-01-06 04:36:57 +00:00
|
|
|
#define AMD64_SET_FSBASE 129
|
|
|
|
#define AMD64_SET_GSBASE 131
|
|
|
|
|
2022-05-13 00:52:13 +00:00
|
|
|
#define __NR_thr_new 455
|
|
|
|
#define __NR_clone_linux 56
|
|
|
|
#define __NR__lwp_create 309
|
|
|
|
#define __NR_getcontext_netbsd 307
|
|
|
|
#define __NR_bsdthread_create 0x02000168
|
|
|
|
#define __NR_thread_fast_set_cthread_self 0x03000003
|
|
|
|
#define PTHREAD_START_CUSTOM_XNU 0x01000000
|
|
|
|
#define LWP_DETACHED 0x00000040
|
|
|
|
#define LWP_SUSPENDED 0x00000080
|
|
|
|
|
2022-05-17 14:40:00 +00:00
|
|
|
struct CloneArgs {
|
2025-01-03 02:44:07 +00:00
|
|
|
union {
|
|
|
|
long sp;
|
2022-05-17 14:40:00 +00:00
|
|
|
int64_t tid64;
|
|
|
|
};
|
2024-06-21 03:46:42 +00:00
|
|
|
atomic_int *ptid;
|
|
|
|
atomic_int *ctid;
|
2022-05-17 14:40:00 +00:00
|
|
|
char *tls;
|
2025-01-03 02:44:07 +00:00
|
|
|
int (*func)(void *);
|
2022-05-17 11:14:28 +00:00
|
|
|
void *arg;
|
|
|
|
};
|
2022-05-16 20:20:08 +00:00
|
|
|
|
2024-02-01 11:39:46 +00:00
|
|
|
int sys_set_tls(uintptr_t, void *);
|
2025-01-03 02:44:07 +00:00
|
|
|
int __stack_call(void *, int, long, long, int (*)(void *), long);
|
2023-06-04 08:57:10 +00:00
|
|
|
|
2023-06-04 01:32:33 +00:00
|
|
|
#ifdef __x86_64__
|
|
|
|
|
2022-05-17 14:40:00 +00:00
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
// THE NEW TECHNOLOGY
|
|
|
|
|
2023-06-04 01:32:33 +00:00
|
|
|
__msabi extern typeof(ExitThread) *const __imp_ExitThread;
|
2024-10-08 01:39:25 +00:00
|
|
|
__msabi extern typeof(GetCurrentThreadId) *const __imp_GetCurrentThreadId;
|
2023-06-04 01:32:33 +00:00
|
|
|
__msabi extern typeof(WakeByAddressAll) *const __imp_WakeByAddressAll;
|
|
|
|
|
2025-01-02 06:25:22 +00:00
|
|
|
textwindows dontinstrument wontreturn static void //
|
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
|
|
|
WinThreadEntry(int rdi, // rcx
|
|
|
|
int rsi, // rdx
|
|
|
|
int rdx, // r8
|
|
|
|
struct CloneArgs *wt) { // r9
|
2025-01-03 02:44:07 +00:00
|
|
|
__set_tls_win32(wt->tls);
|
2024-12-31 08:55:15 +00:00
|
|
|
int tid = __imp_GetCurrentThreadId();
|
2025-01-03 02:44:07 +00:00
|
|
|
atomic_int *ctid = wt->ctid;
|
|
|
|
atomic_init(ctid, tid);
|
2024-12-31 08:55:15 +00:00
|
|
|
atomic_init(wt->ptid, tid);
|
2025-01-03 02:44:07 +00:00
|
|
|
int rc = __stack_call(wt->arg, tid, 0, 0, wt->func, wt->sp);
|
2022-05-19 23:57:49 +00:00
|
|
|
// we can now clear ctid directly since we're no longer using our own
|
|
|
|
// stack memory, which can now be safely free'd by the parent thread.
|
2025-01-03 02:44:07 +00:00
|
|
|
atomic_store_explicit(ctid, 0, memory_order_release);
|
|
|
|
__imp_WakeByAddressAll(ctid);
|
2022-05-19 23:57:49 +00:00
|
|
|
// since we didn't indirect this function through NT2SYSV() it's not
|
2022-11-09 11:58:57 +00:00
|
|
|
// safe to simply return, and as such, we need ExitThread().
|
2022-05-19 23:57:49 +00:00
|
|
|
__imp_ExitThread(rc);
|
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
|
|
|
__builtin_unreachable();
|
2022-05-13 00:52:13 +00:00
|
|
|
}
|
|
|
|
|
2025-01-03 02:44:07 +00:00
|
|
|
textwindows static errno_t CloneWindows(int (*func)(void *), char *stk,
|
|
|
|
size_t stksz, void *arg, void *tls,
|
|
|
|
atomic_int *ptid, atomic_int *ctid) {
|
2024-06-21 03:46:42 +00:00
|
|
|
long sp;
|
2022-05-13 00:52:13 +00:00
|
|
|
int64_t h;
|
2025-01-03 02:44:07 +00:00
|
|
|
intptr_t tip;
|
2024-06-21 03:46:42 +00:00
|
|
|
uint32_t utid;
|
2022-05-17 14:40:00 +00:00
|
|
|
struct CloneArgs *wt;
|
2025-01-03 02:44:07 +00:00
|
|
|
sp = tip = (intptr_t)stk + stksz;
|
2024-06-21 03:46:42 +00:00
|
|
|
sp -= sizeof(struct CloneArgs);
|
|
|
|
sp &= -alignof(struct CloneArgs);
|
|
|
|
wt = (struct CloneArgs *)sp;
|
2025-01-03 02:44:07 +00:00
|
|
|
wt->ctid = ctid;
|
|
|
|
wt->ptid = ptid;
|
2022-05-13 00:52:13 +00:00
|
|
|
wt->func = func;
|
|
|
|
wt->arg = arg;
|
2025-01-03 02:44:07 +00:00
|
|
|
wt->tls = tls;
|
|
|
|
wt->sp = tip & -16;
|
2025-01-03 03:09:57 +00:00
|
|
|
if ((h = CreateThread(0, 65536, (void *)WinThreadEntry, wt,
|
2024-06-21 03:46:42 +00:00
|
|
|
kNtStackSizeParamIsAReservation, &utid))) {
|
2025-01-03 02:44:07 +00:00
|
|
|
atomic_init(ptid, utid);
|
|
|
|
struct CosmoTib *tib = tls;
|
|
|
|
atomic_store_explicit(&tib->tib_syshand, h, memory_order_release);
|
2022-11-08 18:09:47 +00:00
|
|
|
return 0;
|
2022-05-13 00:52:13 +00:00
|
|
|
} else {
|
2022-11-08 18:09:47 +00:00
|
|
|
return __dos2errno(GetLastError());
|
2022-05-13 00:52:13 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2022-05-17 14:40:00 +00:00
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
// XNU'S NOT UNIX
|
|
|
|
|
2023-06-04 01:32:33 +00:00
|
|
|
void XnuThreadThunk(void *pthread, // rdi x0
|
|
|
|
int machport, // rsi x1
|
|
|
|
void *(*func)(void *), // rdx x2
|
|
|
|
void *arg, // rcx x3
|
|
|
|
intptr_t *stack, // r8 x4
|
|
|
|
unsigned xnuflags); // r9 x5
|
2022-05-27 20:25:46 +00:00
|
|
|
asm("XnuThreadThunk:\n\t"
|
2022-05-13 00:52:13 +00:00
|
|
|
"xor\t%ebp,%ebp\n\t"
|
|
|
|
"mov\t%r8,%rsp\n\t"
|
2022-05-17 14:40:00 +00:00
|
|
|
"push\t%rax\n\t"
|
2022-05-13 00:52:13 +00:00
|
|
|
"jmp\tXnuThreadMain\n\t"
|
|
|
|
".size\tXnuThreadThunk,.-XnuThreadThunk");
|
2023-07-10 11:29:46 +00:00
|
|
|
__attribute__((__used__))
|
2022-05-13 00:52:13 +00:00
|
|
|
|
2025-01-03 02:44:07 +00:00
|
|
|
dontinstrument wontreturn static void
|
|
|
|
XnuThreadMain(void *pthread, // rdi
|
|
|
|
int tid, // rsi
|
|
|
|
int (*func)(void *arg), // rdx
|
|
|
|
void *arg, // rcx
|
|
|
|
struct CloneArgs *wt, // r8
|
|
|
|
unsigned xnuflags) { // r9
|
2024-12-31 08:55:15 +00:00
|
|
|
atomic_init(wt->ctid, tid);
|
|
|
|
atomic_init(wt->ptid, tid);
|
2022-07-10 11:01:17 +00:00
|
|
|
|
2025-01-03 02:44:07 +00:00
|
|
|
// XNU uses the same 0x30 offset as the WIN32 TIB x64. They told the
|
|
|
|
// Go team at Google that they Apply stands by our ability to use it
|
|
|
|
// https://github.com/golang/go/issues/23617#issuecomment-376662373
|
|
|
|
int ax;
|
|
|
|
asm volatile("syscall"
|
|
|
|
: "=a"(ax)
|
|
|
|
: "0"(__NR_thread_fast_set_cthread_self), "D"(wt->tls - 0x30)
|
|
|
|
: "rcx", "rdx", "r8", "r9", "r10", "r11", "memory", "cc");
|
2022-07-10 11:01:17 +00:00
|
|
|
|
2025-01-03 02:44:07 +00:00
|
|
|
func(arg);
|
2022-07-10 11:01:17 +00:00
|
|
|
|
2022-05-19 23:57:49 +00:00
|
|
|
// we no longer use the stack after this point
|
|
|
|
// %rax = int bsdthread_terminate(%rdi = void *stackaddr,
|
|
|
|
// %rsi = size_t freesize,
|
|
|
|
// %rdx = uint32_t port,
|
|
|
|
// %r10 = uint32_t sem);
|
2025-01-03 02:44:07 +00:00
|
|
|
asm volatile("movl\t$0,(%%rsi)\n\t" // *wt->ctid = 0
|
2023-10-03 21:47:20 +00:00
|
|
|
"mov\t$0x101,%%edi\n\t" // wake all
|
|
|
|
"xor\t%%edx,%%edx\n\t" // wake_value
|
|
|
|
"mov\t$0x02000204,%%eax\n\t" // ulock_wake()
|
|
|
|
"syscall\n\t" //
|
|
|
|
"xor\t%%edi,%%edi\n\t" // freeaddr
|
|
|
|
"xor\t%%esi,%%esi\n\t" // freesize
|
|
|
|
"xor\t%%edx,%%edx\n\t" // kport
|
|
|
|
"xor\t%%r10d,%%r10d\n\t" // joinsem
|
|
|
|
"mov\t$0x02000169,%%eax\n\t" // bsdthread_terminate()
|
|
|
|
"syscall"
|
|
|
|
: /* no outputs */
|
2025-01-03 02:44:07 +00:00
|
|
|
: "S"(wt->ctid)
|
2023-10-03 21:47:20 +00:00
|
|
|
: "rax", "rcx", "r10", "r11", "memory");
|
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
|
|
|
__builtin_unreachable();
|
2022-05-13 00:52:13 +00:00
|
|
|
}
|
|
|
|
|
2025-01-03 02:44:07 +00:00
|
|
|
static errno_t CloneXnu(int (*fn)(void *), char *stk, size_t stksz, void *arg,
|
|
|
|
void *tls, atomic_int *ptid, atomic_int *ctid) {
|
2024-06-21 03:46:42 +00:00
|
|
|
|
|
|
|
// perform this weird mandatory system call once
|
2022-05-13 00:52:13 +00:00
|
|
|
static bool once;
|
|
|
|
if (!once) {
|
2025-01-03 02:44:07 +00:00
|
|
|
sys_bsdthread_register(XnuThreadThunk, 0, 0, 0, 0, 0, 0);
|
2022-05-13 00:52:13 +00:00
|
|
|
once = true;
|
|
|
|
}
|
2024-06-21 03:46:42 +00:00
|
|
|
|
|
|
|
// setup stack for thread
|
|
|
|
long sp;
|
|
|
|
struct CloneArgs *wt;
|
|
|
|
sp = (intptr_t)stk + stksz;
|
|
|
|
sp -= sizeof(struct CloneArgs);
|
|
|
|
sp &= -alignof(struct CloneArgs);
|
|
|
|
wt = (struct CloneArgs *)sp;
|
2025-01-03 02:44:07 +00:00
|
|
|
sp &= -16;
|
2024-06-21 03:46:42 +00:00
|
|
|
|
|
|
|
// pass parameters to new thread via xnu
|
2025-01-03 02:44:07 +00:00
|
|
|
wt->ctid = ctid;
|
|
|
|
wt->ptid = ptid;
|
|
|
|
wt->tls = tls;
|
2022-11-08 18:09:47 +00:00
|
|
|
return sys_clone_xnu(fn, arg, wt, 0, PTHREAD_START_CUSTOM_XNU);
|
2022-05-13 00:52:13 +00:00
|
|
|
}
|
|
|
|
|
2022-05-17 14:40:00 +00:00
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
// OPEN BESIYATA DISHMAYA
|
2022-05-13 00:52:13 +00:00
|
|
|
|
2022-07-11 12:55:17 +00:00
|
|
|
// we can't use address sanitizer because:
|
|
|
|
// 1. __asan_handle_no_return wipes stack [todo?]
|
2025-01-02 06:25:22 +00:00
|
|
|
relegated dontinstrument wontreturn static void OpenbsdThreadMain(void *p) {
|
2022-06-26 01:17:31 +00:00
|
|
|
struct CloneArgs *wt = p;
|
2025-01-03 02:44:07 +00:00
|
|
|
int tid = atomic_load_explicit(wt->ctid, memory_order_relaxed);
|
|
|
|
atomic_init(wt->ptid, tid);
|
|
|
|
wt->func(wt->arg);
|
|
|
|
asm volatile("mov\t%1,%%rsp\n\t" // so syscall can validate stack exists
|
|
|
|
"movl\t$0,(%2)\n\t" // *wt->ctid = 0 (old stack now free'd)
|
2022-10-09 06:54:05 +00:00
|
|
|
"syscall\n\t" // futex(int*, op, val) will wake wait0
|
|
|
|
"xor\t%%edi,%%edi\n\t" // so kernel doesn't write to old stack
|
|
|
|
"mov\t$302,%%eax\n\t" // __threxit(int *notdead) doesn't wake
|
2022-07-18 02:59:49 +00:00
|
|
|
"syscall"
|
2025-01-03 02:44:07 +00:00
|
|
|
: /* no outputs */
|
|
|
|
: "a"(83), "m"(__oldstack), "D"(wt->ctid),
|
2022-10-09 06:54:05 +00:00
|
|
|
"S"(2 /* FUTEX_WAKE */), "d"(INT_MAX)
|
2022-05-19 23:57:49 +00:00
|
|
|
: "rcx", "r11", "memory");
|
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
|
|
|
__builtin_unreachable();
|
2022-05-13 00:52:13 +00:00
|
|
|
}
|
|
|
|
|
2025-01-03 02:44:07 +00:00
|
|
|
relegated static errno_t CloneOpenbsd(int (*func)(void *), char *stk,
|
|
|
|
size_t stksz, void *arg, void *tls,
|
|
|
|
atomic_int *ptid, atomic_int *ctid) {
|
2022-11-08 18:09:47 +00:00
|
|
|
int rc;
|
2022-06-26 01:17:31 +00:00
|
|
|
intptr_t sp;
|
|
|
|
struct __tfork *tf;
|
2022-05-17 14:40:00 +00:00
|
|
|
struct CloneArgs *wt;
|
2022-06-26 01:17:31 +00:00
|
|
|
sp = (intptr_t)stk + stksz;
|
|
|
|
sp -= sizeof(struct __tfork);
|
|
|
|
sp &= -alignof(struct __tfork);
|
|
|
|
tf = (struct __tfork *)sp;
|
|
|
|
sp -= sizeof(struct CloneArgs);
|
2023-06-04 15:19:45 +00:00
|
|
|
sp &= -alignof(struct CloneArgs);
|
2022-06-26 01:17:31 +00:00
|
|
|
wt = (struct CloneArgs *)sp;
|
2025-01-03 02:44:07 +00:00
|
|
|
sp &= -16;
|
|
|
|
sp -= 8;
|
|
|
|
*(intptr_t *)sp = (intptr_t)CloneOpenbsd + 1;
|
|
|
|
wt->ctid = ctid;
|
|
|
|
wt->ptid = ptid;
|
2022-05-17 14:40:00 +00:00
|
|
|
wt->arg = arg;
|
2022-06-26 01:17:31 +00:00
|
|
|
wt->func = func;
|
2025-01-03 02:44:07 +00:00
|
|
|
tf->tf_stack = (char *)sp;
|
|
|
|
tf->tf_tcb = tls;
|
|
|
|
tf->tf_tid = ctid;
|
2022-11-08 18:09:47 +00:00
|
|
|
if ((rc = __tfork_thread(tf, sizeof(*tf), OpenbsdThreadMain, wt)) >= 0) {
|
2025-01-03 02:44:07 +00:00
|
|
|
atomic_init(ptid, rc);
|
2022-11-08 18:09:47 +00:00
|
|
|
return 0;
|
|
|
|
} else {
|
|
|
|
return -rc;
|
2022-05-13 00:52:13 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2022-05-17 14:40:00 +00:00
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
// NET BESIYATA DISHMAYA
|
|
|
|
|
2025-01-02 06:25:22 +00:00
|
|
|
wontreturn dontinstrument static void NetbsdThreadMain(
|
2025-01-03 02:44:07 +00:00
|
|
|
void *arg, // rdi
|
|
|
|
int (*func)(void *), // rsi
|
|
|
|
atomic_int *ctid, // rdx
|
|
|
|
atomic_int *ptid) { // rcx
|
|
|
|
int ax;
|
|
|
|
asm("syscall"
|
|
|
|
: "=a"(ax) // man says always succeeds
|
|
|
|
: "0"(311) // _lwp_self()
|
|
|
|
: "rcx", "rdx", "r8", "r9", "r10", "r11", "memory", "cc");
|
|
|
|
atomic_init(ctid, ax);
|
|
|
|
atomic_init(ptid, ax);
|
|
|
|
func(arg);
|
2022-05-19 23:57:49 +00:00
|
|
|
// we no longer use the stack after this point
|
|
|
|
// %eax = int __lwp_exit(void);
|
2025-01-03 02:44:07 +00:00
|
|
|
asm volatile("movl\t$0,(%2)\n\t" // *ztid = 0
|
|
|
|
"syscall" // __lwp_exit()
|
|
|
|
: "=a"(ax)
|
|
|
|
: "0"(310), "r"(ctid)
|
2022-05-19 23:57:49 +00:00
|
|
|
: "rcx", "r11", "memory");
|
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
|
|
|
__builtin_unreachable();
|
2022-05-13 00:52:13 +00:00
|
|
|
}
|
|
|
|
|
2025-01-03 02:44:07 +00:00
|
|
|
static int CloneNetbsd(int (*func)(void *), char *stk, size_t stksz, void *arg,
|
|
|
|
void *tls, atomic_int *ptid, atomic_int *ctid) {
|
2022-05-13 00:52:13 +00:00
|
|
|
// NetBSD has its own clone() and it works, but it's technically a
|
2022-06-19 08:13:03 +00:00
|
|
|
// second-class API, intended to help Linux folks migrate to this.
|
2024-06-21 03:46:42 +00:00
|
|
|
int ax;
|
2022-05-13 00:52:13 +00:00
|
|
|
bool failed;
|
|
|
|
intptr_t dx, sp;
|
|
|
|
static bool once;
|
2022-06-19 08:13:03 +00:00
|
|
|
struct ucontext_netbsd *ctx;
|
2022-05-13 00:52:13 +00:00
|
|
|
static struct ucontext_netbsd netbsd_clone_template;
|
2022-05-19 23:57:49 +00:00
|
|
|
|
|
|
|
// memoize arbitrary valid processor state structure
|
2022-05-13 00:52:13 +00:00
|
|
|
if (!once) {
|
|
|
|
asm volatile(CFLAG_ASM("syscall")
|
|
|
|
: CFLAG_CONSTRAINT(failed), "=a"(ax)
|
|
|
|
: "1"(__NR_getcontext_netbsd), "D"(&netbsd_clone_template)
|
2022-09-09 11:07:08 +00:00
|
|
|
: "rcx", "rdx", "r8", "r9", "r10", "r11", "memory");
|
2022-05-13 00:52:13 +00:00
|
|
|
once = true;
|
|
|
|
}
|
2024-06-21 03:46:42 +00:00
|
|
|
sp = (intptr_t)stk + stksz;
|
2022-06-19 08:13:03 +00:00
|
|
|
|
|
|
|
// align the stack
|
2025-01-03 02:44:07 +00:00
|
|
|
sp &= -16;
|
2022-06-19 08:13:03 +00:00
|
|
|
|
|
|
|
// simulate call to misalign stack and ensure backtrace looks good
|
2022-05-19 23:57:49 +00:00
|
|
|
sp -= 8;
|
2022-06-19 08:13:03 +00:00
|
|
|
*(intptr_t *)sp = (intptr_t)CloneNetbsd + 1;
|
|
|
|
|
|
|
|
// place the giant 784 byte ucontext structure in the red zone!
|
|
|
|
// it only has to live long enough for the thread to come alive
|
2025-01-03 02:44:07 +00:00
|
|
|
ctx = (struct ucontext_netbsd *)((sp - sizeof(struct ucontext_netbsd)) & -64);
|
2022-06-19 08:13:03 +00:00
|
|
|
|
2022-05-19 23:57:49 +00:00
|
|
|
// pass parameters in process state
|
2022-06-19 08:13:03 +00:00
|
|
|
memcpy(ctx, &netbsd_clone_template, sizeof(*ctx));
|
|
|
|
ctx->uc_link = 0;
|
|
|
|
ctx->uc_mcontext.rbp = 0;
|
|
|
|
ctx->uc_mcontext.rsp = sp;
|
|
|
|
ctx->uc_mcontext.rip = (intptr_t)NetbsdThreadMain;
|
|
|
|
ctx->uc_mcontext.rdi = (intptr_t)arg;
|
|
|
|
ctx->uc_mcontext.rsi = (intptr_t)func;
|
2025-01-03 02:44:07 +00:00
|
|
|
ctx->uc_mcontext.rdx = (intptr_t)ctid;
|
|
|
|
ctx->uc_mcontext.rcx = (intptr_t)ptid;
|
2022-06-19 08:13:03 +00:00
|
|
|
ctx->uc_flags |= _UC_STACK;
|
|
|
|
ctx->uc_stack.ss_sp = stk;
|
|
|
|
ctx->uc_stack.ss_size = stksz;
|
|
|
|
ctx->uc_stack.ss_flags = 0;
|
2025-01-03 02:44:07 +00:00
|
|
|
ctx->uc_flags |= _UC_TLSBASE;
|
|
|
|
ctx->uc_mcontext._mc_tlsbase = (intptr_t)tls;
|
2022-05-19 23:57:49 +00:00
|
|
|
|
|
|
|
// perform the system call
|
2024-12-18 12:59:02 +00:00
|
|
|
int tid = 0;
|
2022-05-13 00:52:13 +00:00
|
|
|
asm volatile(CFLAG_ASM("syscall")
|
|
|
|
: CFLAG_CONSTRAINT(failed), "=a"(ax), "=d"(dx)
|
2024-12-18 12:59:02 +00:00
|
|
|
: "1"(__NR__lwp_create), "D"(ctx), "S"(LWP_DETACHED), "2"(&tid)
|
2022-09-09 11:07:08 +00:00
|
|
|
: "rcx", "r8", "r9", "r10", "r11", "memory");
|
2022-05-13 00:52:13 +00:00
|
|
|
if (!failed) {
|
2025-01-03 02:44:07 +00:00
|
|
|
atomic_init(ptid, tid);
|
2022-11-08 18:09:47 +00:00
|
|
|
return 0;
|
2022-05-13 00:52:13 +00:00
|
|
|
} else {
|
2022-11-08 18:09:47 +00:00
|
|
|
return ax;
|
2022-05-13 00:52:13 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2023-05-09 08:56:56 +00:00
|
|
|
#endif /* __x86_64__ */
|
|
|
|
|
2023-12-30 04:11:23 +00:00
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
// FREE BESIYATA DISHMAYA
|
|
|
|
|
2025-01-02 06:25:22 +00:00
|
|
|
wontreturn dontinstrument static void FreebsdThreadMain(void *p) {
|
2023-12-30 04:11:23 +00:00
|
|
|
struct CloneArgs *wt = p;
|
|
|
|
#ifdef __aarch64__
|
|
|
|
asm volatile("mov\tx28,%0" : /* no outputs */ : "r"(wt->tls));
|
2024-01-06 04:36:57 +00:00
|
|
|
#elif defined(__x86_64__)
|
Release Cosmopolitan v3.3
This change upgrades to GCC 12.3 and GNU binutils 2.42. The GNU linker
appears to have changed things so that only a single de-duplicated str
table is present in the binary, and it gets placed wherever the linker
wants, regardless of what the linker script says. To cope with that we
need to stop using .ident to embed licenses. As such, this change does
significant work to revamp how third party licenses are defined in the
codebase, using `.section .notice,"aR",@progbits`.
This new GCC 12.3 toolchain has support for GNU indirect functions. It
lets us support __target_clones__ for the first time. This is used for
optimizing the performance of libc string functions such as strlen and
friends so far on x86, by ensuring AVX systems favor a second codepath
that uses VEX encoding. It shaves some latency off certain operations.
It's a useful feature to have for scientific computing for the reasons
explained by the test/libcxx/openmp_test.cc example which compiles for
fifteen different microarchitectures. Thanks to the upgrades, it's now
also possible to use newer instruction sets, such as AVX512FP16, VNNI.
Cosmo now uses the %gs register on x86 by default for TLS. Doing it is
helpful for any program that links `cosmo_dlopen()`. Such programs had
to recompile their binaries at startup to change the TLS instructions.
That's not great, since it means every page in the executable needs to
be faulted. The work of rewriting TLS-related x86 opcodes, is moved to
fixupobj.com instead. This is great news for MacOS x86 users, since we
previously needed to morph the binary every time for that platform but
now that's no longer necessary. The only platforms where we need fixup
of TLS x86 opcodes at runtime are now Windows, OpenBSD, and NetBSD. On
Windows we morph TLS to point deeper into the TIB, based on a TlsAlloc
assignment, and on OpenBSD/NetBSD we morph %gs back into %fs since the
kernels do not allow us to specify a value for the %gs register.
OpenBSD users are now required to use APE Loader to run Cosmo binaries
and assimilation is no longer possible. OpenBSD kernel needs to change
to allow programs to specify a value for the %gs register, or it needs
to stop marking executable pages loaded by the kernel as mimmutable().
This release fixes __constructor__, .ctor, .init_array, and lastly the
.preinit_array so they behave the exact same way as glibc.
We no longer use hex constants to define math.h symbols like M_PI.
2024-02-20 19:12:09 +00:00
|
|
|
sys_set_tls(AMD64_SET_GSBASE, wt->tls);
|
2023-12-30 04:11:23 +00:00
|
|
|
#endif
|
2025-01-03 02:44:07 +00:00
|
|
|
atomic_init(wt->ctid, wt->tid64);
|
|
|
|
atomic_init(wt->ptid, wt->tid64);
|
|
|
|
wt->func(wt->arg);
|
2023-12-30 04:11:23 +00:00
|
|
|
// we no longer use the stack after this point
|
|
|
|
// void thr_exit(%rdi = long *state);
|
|
|
|
#ifdef __x86_64__
|
2025-01-03 02:44:07 +00:00
|
|
|
asm volatile("movl\t$0,%0\n\t" // *wt->ctid = 0
|
|
|
|
"syscall\n\t" // _umtx_op(wt->ctid, WAKE, INT_MAX)
|
2023-12-30 04:11:23 +00:00
|
|
|
"movl\t$431,%%eax\n\t" // thr_exit(long *nonzeroes_and_wake)
|
|
|
|
"xor\t%%edi,%%edi\n\t" // sad we can't use this free futex op
|
|
|
|
"syscall\n\t" // thr_exit() fails if thread is orphaned
|
|
|
|
"movl\t$1,%%eax\n\t" // _exit()
|
|
|
|
"syscall" //
|
2025-01-03 02:44:07 +00:00
|
|
|
: "=m"(*wt->ctid)
|
|
|
|
: "a"(454), "D"(wt->ctid), "S"(UMTX_OP_WAKE), "d"(INT_MAX)
|
2023-12-30 04:11:23 +00:00
|
|
|
: "rcx", "r8", "r9", "r10", "r11", "memory");
|
|
|
|
#elif defined(__aarch64__)
|
2025-01-03 02:44:07 +00:00
|
|
|
register long x0 asm("x0") = (long)wt->ctid;
|
2023-12-30 04:11:23 +00:00
|
|
|
register long x1 asm("x1") = UMTX_OP_WAKE;
|
|
|
|
register long x2 asm("x2") = INT_MAX;
|
|
|
|
register long x8 asm("x8") = 454; // _umtx_op
|
2025-01-03 02:44:07 +00:00
|
|
|
asm volatile("str\twzr,%0\n\t" // *wt->ctid = 0
|
|
|
|
"svc\t0\n\t" // _umtx_op(wt->ctid, WAKE, INT_MAX)
|
2023-12-30 04:11:23 +00:00
|
|
|
"mov\tx0,#0\n\t" // arg0 = 0
|
|
|
|
"mov\tx8,#431\n\t" // thr_exit
|
|
|
|
"svc\t0\n\t" // thr_exit(long *nonzeroes_and_wake = 0)
|
|
|
|
"mov\tx8,#1\n\t" // _exit
|
|
|
|
"svc\t0" // _exit(long *nonzeroes_and_wake = 0)
|
2025-01-03 02:44:07 +00:00
|
|
|
: "=m"(*wt->ctid)
|
2023-12-30 04:11:23 +00:00
|
|
|
: "r"(x0), "r"(x1), "r"(x2), "r"(x8));
|
|
|
|
#else
|
|
|
|
#error "unsupported architecture"
|
|
|
|
#endif
|
|
|
|
__builtin_unreachable();
|
|
|
|
}
|
|
|
|
|
2025-01-03 02:44:07 +00:00
|
|
|
static errno_t CloneFreebsd(int (*func)(void *), char *stk, size_t stksz,
|
|
|
|
void *arg, void *tls, atomic_int *ptid,
|
2024-06-21 03:46:42 +00:00
|
|
|
atomic_int *ctid) {
|
|
|
|
long sp;
|
2025-01-03 02:44:07 +00:00
|
|
|
int64_t tid64;
|
2023-12-30 04:11:23 +00:00
|
|
|
struct CloneArgs *wt;
|
2024-06-21 03:46:42 +00:00
|
|
|
sp = (intptr_t)stk + stksz;
|
|
|
|
sp -= sizeof(struct CloneArgs);
|
|
|
|
sp &= -alignof(struct CloneArgs);
|
|
|
|
wt = (struct CloneArgs *)sp;
|
2025-01-03 02:44:07 +00:00
|
|
|
sp &= -16;
|
|
|
|
wt->ctid = ctid;
|
|
|
|
wt->ptid = ptid;
|
2023-12-30 04:11:23 +00:00
|
|
|
wt->tls = tls;
|
|
|
|
wt->func = func;
|
|
|
|
wt->arg = arg;
|
|
|
|
struct thr_param params = {
|
|
|
|
.start_func = FreebsdThreadMain,
|
|
|
|
.arg = wt,
|
|
|
|
.stack_base = stk,
|
2024-06-21 03:46:42 +00:00
|
|
|
.stack_size = sp - (long)stk,
|
2025-01-03 02:44:07 +00:00
|
|
|
.tls_base = tls,
|
2023-12-30 04:11:23 +00:00
|
|
|
.tls_size = 64,
|
|
|
|
.child_tid = &wt->tid64,
|
2025-01-03 02:44:07 +00:00
|
|
|
.parent_tid = &tid64,
|
2023-12-30 04:11:23 +00:00
|
|
|
};
|
|
|
|
#ifdef __x86_64__
|
|
|
|
int ax;
|
|
|
|
bool failed;
|
|
|
|
asm volatile(CFLAG_ASM("syscall")
|
|
|
|
: CFLAG_CONSTRAINT(failed), "=a"(ax)
|
|
|
|
: "1"(__NR_thr_new), "D"(¶ms), "S"(sizeof(params))
|
|
|
|
: "rcx", "rdx", "r8", "r9", "r10", "r11", "memory");
|
|
|
|
if (failed)
|
|
|
|
return ax;
|
|
|
|
#elif defined(__aarch64__)
|
|
|
|
register long x0 asm("x0") = (long)¶ms;
|
|
|
|
register long x1 asm("x1") = sizeof(params);
|
|
|
|
register int x8 asm("x8") = 0x1c7; // thr_new
|
|
|
|
asm volatile("svc\t0" : "+r"(x0) : "r"(x1), "r"(x8) : "memory");
|
|
|
|
if (x0)
|
|
|
|
return x0;
|
|
|
|
#else
|
|
|
|
#error "unsupported architecture"
|
|
|
|
#endif
|
2025-01-03 02:44:07 +00:00
|
|
|
atomic_init(ptid, tid64);
|
2023-12-30 04:11:23 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2023-06-04 01:32:33 +00:00
|
|
|
#ifdef __aarch64__
|
|
|
|
|
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
// APPLE SILICON
|
|
|
|
|
2025-01-02 06:25:22 +00:00
|
|
|
dontinstrument static void *SiliconThreadMain(void *arg) {
|
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
|
|
|
struct CloneArgs *wt = arg;
|
2025-01-03 02:44:07 +00:00
|
|
|
atomic_int *ctid = wt->ctid;
|
|
|
|
int tid = atomic_load_explicit(ctid, memory_order_relaxed);
|
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
|
|
|
asm volatile("mov\tx28,%0" : /* no outputs */ : "r"(wt->tls));
|
2025-01-03 02:44:07 +00:00
|
|
|
__stack_call(wt->arg, tid, 0, 0, wt->func, wt->sp);
|
|
|
|
atomic_store_explicit(ctid, 0, memory_order_release);
|
|
|
|
ulock_wake(UL_COMPARE_AND_WAIT | ULF_WAKE_ALL, ctid, 0);
|
2023-06-04 01:32:33 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2025-01-03 02:44:07 +00:00
|
|
|
static errno_t CloneSilicon(int (*fn)(void *), char *stk, size_t stksz,
|
|
|
|
void *arg, void *tls, atomic_int *ptid,
|
2024-06-21 03:46:42 +00:00
|
|
|
atomic_int *ctid) {
|
2025-01-03 02:44:07 +00:00
|
|
|
|
|
|
|
// assign tid to new thread
|
2023-06-04 08:57:10 +00:00
|
|
|
static atomic_uint tids;
|
2025-01-03 02:44:07 +00:00
|
|
|
unsigned tid = atomic_fetch_add_explicit(&tids, 1, memory_order_relaxed);
|
|
|
|
tid %= kMaxThreadIds;
|
|
|
|
tid += kMinThreadId;
|
|
|
|
atomic_init(ctid, tid);
|
|
|
|
atomic_init(ptid, tid);
|
|
|
|
|
|
|
|
// pass temp data on stack
|
|
|
|
intptr_t sp, tip;
|
|
|
|
struct CloneArgs *wt;
|
|
|
|
sp = tip = (intptr_t)stk + stksz;
|
2024-06-21 03:46:42 +00:00
|
|
|
sp -= sizeof(struct CloneArgs);
|
|
|
|
sp &= -alignof(struct CloneArgs);
|
|
|
|
wt = (struct CloneArgs *)sp;
|
2023-06-04 01:32:33 +00:00
|
|
|
wt->func = fn;
|
|
|
|
wt->arg = arg;
|
2025-01-03 02:44:07 +00:00
|
|
|
wt->tls = tls;
|
|
|
|
wt->ctid = ctid;
|
|
|
|
wt->sp = tip & -16;
|
|
|
|
|
|
|
|
// ask apple libc to spawn thread
|
|
|
|
errno_t res;
|
|
|
|
pthread_t th;
|
|
|
|
size_t babystack = __syslib->__pthread_stack_min;
|
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
|
|
|
#pragma GCC push_options
|
|
|
|
#pragma GCC diagnostic ignored "-Walloca-larger-than="
|
2025-01-03 02:44:07 +00:00
|
|
|
void *attr = alloca(__syslib->__sizeof_pthread_attr_t);
|
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
|
|
|
#pragma GCC pop_options
|
2025-01-03 02:44:07 +00:00
|
|
|
__syslib->__pthread_attr_init(attr);
|
|
|
|
__syslib->__pthread_attr_setguardsize(attr, 0);
|
|
|
|
__syslib->__pthread_attr_setstacksize(attr, babystack);
|
2024-12-24 18:30:59 +00:00
|
|
|
if (!(res = __syslib->__pthread_create(&th, attr, SiliconThreadMain, wt))) {
|
2025-01-03 02:44:07 +00:00
|
|
|
atomic_init(ptid, tid);
|
|
|
|
struct CosmoTib *tib = tls;
|
|
|
|
atomic_store_explicit(&tib[-1].tib_syshand, th, memory_order_release);
|
2023-06-04 08:57:10 +00:00
|
|
|
}
|
2025-01-03 02:44:07 +00:00
|
|
|
__syslib->__pthread_attr_destroy(attr);
|
2023-06-04 08:57:10 +00:00
|
|
|
return res;
|
2023-06-04 01:32:33 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
#endif /* __aarch64__ */
|
|
|
|
|
2022-05-17 14:40:00 +00:00
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
// GNU/SYSTEMD
|
|
|
|
|
2024-01-06 04:36:57 +00:00
|
|
|
struct LinuxCloneArgs {
|
2025-01-03 02:44:07 +00:00
|
|
|
int (*func)(void *);
|
2024-01-06 04:36:57 +00:00
|
|
|
void *arg;
|
|
|
|
char *tls;
|
|
|
|
};
|
|
|
|
|
2024-06-21 03:46:42 +00:00
|
|
|
int sys_clone_linux(int flags, // rdi
|
|
|
|
long sp, // rsi
|
|
|
|
atomic_int *ptid, // rdx
|
|
|
|
atomic_int *ctid, // rcx
|
|
|
|
void *tls, // r8
|
|
|
|
void *func, // r9
|
|
|
|
void *arg); // 8(rsp)
|
2022-07-10 11:01:17 +00:00
|
|
|
|
2025-01-03 02:44:07 +00:00
|
|
|
dontinstrument static int AmdLinuxThreadEntry(void *arg) {
|
2024-01-06 04:36:57 +00:00
|
|
|
struct LinuxCloneArgs *wt = arg;
|
2025-01-03 06:19:49 +00:00
|
|
|
#if defined(__x86_64__)
|
2024-01-06 04:36:57 +00:00
|
|
|
sys_set_tls(ARCH_SET_GS, wt->tls);
|
2025-01-03 06:19:49 +00:00
|
|
|
#endif
|
2025-01-03 02:44:07 +00:00
|
|
|
return wt->func(wt->arg);
|
2024-01-06 04:36:57 +00:00
|
|
|
}
|
|
|
|
|
2025-01-03 02:44:07 +00:00
|
|
|
static int CloneLinux(int (*func)(void *), char *stk, size_t stksz, int flags,
|
|
|
|
void *arg, void *tls, atomic_int *ptid,
|
2024-06-21 03:46:42 +00:00
|
|
|
atomic_int *ctid) {
|
2025-01-03 02:44:07 +00:00
|
|
|
long sp = (intptr_t)stk + stksz;
|
|
|
|
|
|
|
|
#if defined(__x86_64__)
|
2024-01-06 04:36:57 +00:00
|
|
|
sp -= sizeof(struct LinuxCloneArgs);
|
2024-06-21 03:46:42 +00:00
|
|
|
sp &= -alignof(struct LinuxCloneArgs);
|
2025-01-03 02:44:07 +00:00
|
|
|
struct LinuxCloneArgs *wt = (struct LinuxCloneArgs *)sp;
|
|
|
|
sp &= -16; // align the stack
|
|
|
|
wt->arg = arg;
|
|
|
|
wt->tls = tls;
|
|
|
|
wt->func = func;
|
|
|
|
func = AmdLinuxThreadEntry;
|
|
|
|
arg = wt;
|
|
|
|
#elif defined(__aarch64__)
|
|
|
|
sp &= -128; // for kernels <=4.6
|
2024-01-06 04:36:57 +00:00
|
|
|
#endif
|
2025-01-03 02:44:07 +00:00
|
|
|
|
|
|
|
int rc;
|
2022-11-08 18:09:47 +00:00
|
|
|
if ((rc = sys_clone_linux(flags, sp, ptid, ctid, tls, func, arg)) >= 0) {
|
|
|
|
// clone() is documented as setting ptid before return
|
|
|
|
return 0;
|
|
|
|
} else {
|
|
|
|
return -rc;
|
|
|
|
}
|
2022-07-10 11:01:17 +00:00
|
|
|
}
|
2022-05-13 00:52:13 +00:00
|
|
|
|
2022-05-17 14:40:00 +00:00
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
// COSMOPOLITAN
|
|
|
|
|
2022-05-13 00:52:13 +00:00
|
|
|
/**
|
2025-01-03 02:44:07 +00:00
|
|
|
* Creates thread without malloc() being linked.
|
2022-05-13 00:52:13 +00:00
|
|
|
*
|
2025-01-03 02:44:07 +00:00
|
|
|
* If you use clone() you're on your own.
|
2022-05-13 00:52:13 +00:00
|
|
|
*/
|
2022-11-08 18:09:47 +00:00
|
|
|
errno_t clone(void *func, void *stk, size_t stksz, int flags, void *arg,
|
|
|
|
void *ptid, void *tls, void *ctid) {
|
2024-12-18 12:59:02 +00:00
|
|
|
errno_t err;
|
|
|
|
|
|
|
|
atomic_fetch_add(&_pthread_count, 1);
|
2022-05-17 14:40:00 +00:00
|
|
|
|
2025-01-03 02:44:07 +00:00
|
|
|
if (IsLinux()) {
|
2024-12-18 12:59:02 +00:00
|
|
|
err = CloneLinux(func, stk, stksz, flags, arg, tls, ptid, ctid);
|
2022-05-13 00:52:13 +00:00
|
|
|
} else if (IsXnu()) {
|
2025-01-03 02:44:07 +00:00
|
|
|
#if defined(__x86_64__)
|
|
|
|
err = CloneXnu(func, stk, stksz, arg, tls, ptid, ctid);
|
2023-06-04 01:32:33 +00:00
|
|
|
#elif defined(__aarch64__)
|
2025-01-03 02:44:07 +00:00
|
|
|
err = CloneSilicon(func, stk, stksz, arg, tls, ptid, ctid);
|
2023-06-04 01:32:33 +00:00
|
|
|
#else
|
|
|
|
#error "unsupported architecture"
|
|
|
|
#endif
|
2022-05-13 00:52:13 +00:00
|
|
|
} else if (IsFreebsd()) {
|
2025-01-03 02:44:07 +00:00
|
|
|
err = CloneFreebsd(func, stk, stksz, arg, tls, ptid, ctid);
|
|
|
|
#if defined(__x86_64__)
|
|
|
|
} else if (IsWindows()) {
|
|
|
|
err = CloneWindows(func, stk, stksz, arg, tls, ptid, ctid);
|
2022-05-13 00:52:13 +00:00
|
|
|
} else if (IsNetbsd()) {
|
2025-01-03 02:44:07 +00:00
|
|
|
err = CloneNetbsd(func, stk, stksz, arg, tls, ptid, ctid);
|
2022-05-13 00:52:13 +00:00
|
|
|
} else if (IsOpenbsd()) {
|
2025-01-03 02:44:07 +00:00
|
|
|
err = CloneOpenbsd(func, stk, stksz, arg, tls, ptid, ctid);
|
2023-05-09 08:56:56 +00:00
|
|
|
#endif /* __x86_64__ */
|
2022-05-13 00:52:13 +00:00
|
|
|
} else {
|
2024-12-18 12:59:02 +00:00
|
|
|
err = ENOSYS;
|
2022-05-19 23:57:49 +00:00
|
|
|
}
|
|
|
|
|
2024-12-18 12:59:02 +00:00
|
|
|
if (SupportsBsd() && err == EPROCLIM)
|
|
|
|
err = EAGAIN;
|
|
|
|
|
|
|
|
if (err)
|
2025-01-03 02:44:07 +00:00
|
|
|
atomic_fetch_sub(&_pthread_count, 1);
|
2023-09-07 10:24:46 +00:00
|
|
|
|
2024-12-18 12:59:02 +00:00
|
|
|
return err;
|
2022-05-13 00:52:13 +00:00
|
|
|
}
|