mirrors/cosmopolitan
1
0
Fork 0
mirror of https://github.com/jart/cosmopolitan.git synced 2025-02-12 09:17:53 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1
cosmopolitan/libc/intrin/spinlock.h

55 lines
2.4 KiB
C
Raw Normal View History

Improve locks and signals - Introduce fast spinlock API - Double rand64() perf w/ spinlock - Improve raise() on New Technology - Support gettid() across platforms - Implement SA_NODEFER on New Technology - Move the lock intrinsics into LIBC_INTRIN - Make SIGTRAP recoverable on New Technology - Block SIGCHLD in wait4() on New Technology - Add threading prototypes for XNU and FreeBSD - Rewrite abort() fixing its minor bugs on XNU/NT - Shave down a lot of the content in libc/bits/bits.h - Let signal handlers modify CPU registers on New Technology
2022-04-12 12:20:17 +00:00
#ifndef COSMOPOLITAN_LIBC_INTRIN_SPINLOCK_H_
#define COSMOPOLITAN_LIBC_INTRIN_SPINLOCK_H_
Fix many thread and file descriptor issues on Windows The greenbean web server now works nearly perfectly on Windows with over 1000 threads. But some synchronization issues still remain which prevent us from going over nine thousand.
2022-05-21 01:51:41 +00:00
/*───────────────────────────────────────────────────────────────────────────│─╗
│ cosmopolitan § spinlocks ─╬─│┼
╚────────────────────────────────────────────────────────────────────────────│─╝
Make mutex calling code 10x tinier Calls to lock/unlock functions are now NOPs by default. The first time clone() is called, they get turned into CALL instructions. Doing this caused funcctions like fputc() to shrink from 85 bytes to 45+4 bytes. Since the ANSI solution of `(__threaded && lock())` inlines os much superfluous binary content into functions all over the place.
2022-06-13 02:33:42 +00:00
fast tiny inline synchronization routines */
Improve threading support further
2022-05-17 11:14:28 +00:00
Make mutex calling code 10x tinier Calls to lock/unlock functions are now NOPs by default. The first time clone() is called, they get turned into CALL instructions. Doing this caused funcctions like fputc() to shrink from 85 bytes to 45+4 bytes. Since the ANSI solution of `(__threaded && lock())` inlines os much superfluous binary content into functions all over the place.
2022-06-13 02:33:42 +00:00
#ifdef TINY
#define _spinlock(lock) _spinlock_tiny(lock)
Fix bugs and make improvements - Get clone() working on FreeBSD - Increase some Python build quotas - Add more atomic builtins to chibicc - Fix ASAN poisoning of alloca() memory - Make MODE= mandatory link path tinier - Improve the examples folder a little bit - Start working on some more resource limits - Make the linenoise auto-complete UI as good as GNU readline - Update compile.com, avoiding AVX codegen on non-AVX systems - Make sure empty path to syscalls like opendir raises ENOENT - Correctly polyfill ENOENT vs. ENOTDIR on the New Technology - Port bestline's paredit features to //third_party/linenoise - Remove workarounds for RHEL 5.0 bugs that were fixed in 5.1
2022-04-20 16:56:53 +00:00
#else
Make mutex calling code 10x tinier Calls to lock/unlock functions are now NOPs by default. The first time clone() is called, they get turned into CALL instructions. Doing this caused funcctions like fputc() to shrink from 85 bytes to 45+4 bytes. Since the ANSI solution of `(__threaded && lock())` inlines os much superfluous binary content into functions all over the place.
2022-06-13 02:33:42 +00:00
#define _spinlock(lock) _spinlock_cooperative(lock)
Fix bugs and make improvements - Get clone() working on FreeBSD - Increase some Python build quotas - Add more atomic builtins to chibicc - Fix ASAN poisoning of alloca() memory - Make MODE= mandatory link path tinier - Improve the examples folder a little bit - Start working on some more resource limits - Make the linenoise auto-complete UI as good as GNU readline - Update compile.com, avoiding AVX codegen on non-AVX systems - Make sure empty path to syscalls like opendir raises ENOENT - Correctly polyfill ENOENT vs. ENOTDIR on the New Technology - Port bestline's paredit features to //third_party/linenoise - Remove workarounds for RHEL 5.0 bugs that were fixed in 5.1
2022-04-20 16:56:53 +00:00
#endif
Improve locks and signals - Introduce fast spinlock API - Double rand64() perf w/ spinlock - Improve raise() on New Technology - Support gettid() across platforms - Implement SA_NODEFER on New Technology - Move the lock intrinsics into LIBC_INTRIN - Make SIGTRAP recoverable on New Technology - Block SIGCHLD in wait4() on New Technology - Add threading prototypes for XNU and FreeBSD - Rewrite abort() fixing its minor bugs on XNU/NT - Shave down a lot of the content in libc/bits/bits.h - Let signal handlers modify CPU registers on New Technology
2022-04-12 12:20:17 +00:00
Make mutex calling code 10x tinier Calls to lock/unlock functions are now NOPs by default. The first time clone() is called, they get turned into CALL instructions. Doing this caused funcctions like fputc() to shrink from 85 bytes to 45+4 bytes. Since the ANSI solution of `(__threaded && lock())` inlines os much superfluous binary content into functions all over the place.
2022-06-13 02:33:42 +00:00
#define _spunlock(lock) __atomic_store_n(lock, 0, __ATOMIC_RELAXED)
Add raw memory visualization tool to redbean This change introduces a `-W /dev/pts/1` flag to redbean. What it does is use the mincore() system call to create a dual-screen terminal display that lets you troubleshoot the virtual address space. This is useful since page faults are an important thing to consider when using a forking web server. Now we have a colorful visualization of which pages are going to fault and which ones are resident in memory. The memory monitor, if enabled, spawns as a thread that just outputs ANSI codes to the second terminal in a loop. In order to make this happen using the new clone() polyfill, stdio is now thread safe. This change also introduces some new demo pages to redbean. It also polishes the demos we already have, to look a bit nicer and more presentable for the upcoming release, with better explanations too.
2022-05-14 11:33:58 +00:00
Make mutex calling code 10x tinier Calls to lock/unlock functions are now NOPs by default. The first time clone() is called, they get turned into CALL instructions. Doing this caused funcctions like fputc() to shrink from 85 bytes to 45+4 bytes. Since the ANSI solution of `(__threaded && lock())` inlines os much superfluous binary content into functions all over the place.
2022-06-13 02:33:42 +00:00
#define _seizelock(lock, value) \
({ \
Make fixes and improvements - Document more compiler flags - Expose new __print_maps() api - Better overflow checking in mmap() - Improve the shell example somewhat - Fix minor runtime bugs regarding stacks - Make kill() on fork()+execve()'d children work - Support CLONE_CHILD_CLEARTID for proper joining - Fix recent possible deadlock regression with --ftrace
2022-05-19 23:57:49 +00:00
autotype(lock) __lock = (lock); \
Fix many thread and file descriptor issues on Windows The greenbean web server now works nearly perfectly on Windows with over 1000 threads. But some synchronization issues still remain which prevent us from going over nine thousand.
2022-05-21 01:51:41 +00:00
typeof(*__lock) __x = (value); \
Make fixes and improvements - Document more compiler flags - Expose new __print_maps() api - Better overflow checking in mmap() - Improve the shell example somewhat - Fix minor runtime bugs regarding stacks - Make kill() on fork()+execve()'d children work - Support CLONE_CHILD_CLEARTID for proper joining - Fix recent possible deadlock regression with --ftrace
2022-05-19 23:57:49 +00:00
__atomic_store(__lock, &__x, __ATOMIC_RELEASE); \
Make mutex calling code 10x tinier Calls to lock/unlock functions are now NOPs by default. The first time clone() is called, they get turned into CALL instructions. Doing this caused funcctions like fputc() to shrink from 85 bytes to 45+4 bytes. Since the ANSI solution of `(__threaded && lock())` inlines os much superfluous binary content into functions all over the place.
2022-06-13 02:33:42 +00:00
})
Improve locks and signals - Introduce fast spinlock API - Double rand64() perf w/ spinlock - Improve raise() on New Technology - Support gettid() across platforms - Implement SA_NODEFER on New Technology - Move the lock intrinsics into LIBC_INTRIN - Make SIGTRAP recoverable on New Technology - Block SIGCHLD in wait4() on New Technology - Add threading prototypes for XNU and FreeBSD - Rewrite abort() fixing its minor bugs on XNU/NT - Shave down a lot of the content in libc/bits/bits.h - Let signal handlers modify CPU registers on New Technology
2022-04-12 12:20:17 +00:00
Make mutex calling code 10x tinier Calls to lock/unlock functions are now NOPs by default. The first time clone() is called, they get turned into CALL instructions. Doing this caused funcctions like fputc() to shrink from 85 bytes to 45+4 bytes. Since the ANSI solution of `(__threaded && lock())` inlines os much superfluous binary content into functions all over the place.
2022-06-13 02:33:42 +00:00
#define _spinlock_tiny(lock) \
({ \
autotype(lock) __lock = (lock); \
while (_trylock(__lock)) { \
__builtin_ia32_pause(); \
} \
0; \
})
Add raw memory visualization tool to redbean This change introduces a `-W /dev/pts/1` flag to redbean. What it does is use the mincore() system call to create a dual-screen terminal display that lets you troubleshoot the virtual address space. This is useful since page faults are an important thing to consider when using a forking web server. Now we have a colorful visualization of which pages are going to fault and which ones are resident in memory. The memory monitor, if enabled, spawns as a thread that just outputs ANSI codes to the second terminal in a loop. In order to make this happen using the new clone() polyfill, stdio is now thread safe. This change also introduces some new demo pages to redbean. It also polishes the demos we already have, to look a bit nicer and more presentable for the upcoming release, with better explanations too.
2022-05-14 11:33:58 +00:00
Make fixes and improvements - Document more compiler flags - Expose new __print_maps() api - Better overflow checking in mmap() - Improve the shell example somewhat - Fix minor runtime bugs regarding stacks - Make kill() on fork()+execve()'d children work - Support CLONE_CHILD_CLEARTID for proper joining - Fix recent possible deadlock regression with --ftrace
2022-05-19 23:57:49 +00:00
#define _spinlock_cooperative(lock) \
Make mutex calling code 10x tinier Calls to lock/unlock functions are now NOPs by default. The first time clone() is called, they get turned into CALL instructions. Doing this caused funcctions like fputc() to shrink from 85 bytes to 45+4 bytes. Since the ANSI solution of `(__threaded && lock())` inlines os much superfluous binary content into functions all over the place.
2022-06-13 02:33:42 +00:00
({ \
Make fixes and improvements - Document more compiler flags - Expose new __print_maps() api - Better overflow checking in mmap() - Improve the shell example somewhat - Fix minor runtime bugs regarding stacks - Make kill() on fork()+execve()'d children work - Support CLONE_CHILD_CLEARTID for proper joining - Fix recent possible deadlock regression with --ftrace
2022-05-19 23:57:49 +00:00
autotype(lock) __lock = (lock); \
typeof(*__lock) __x; \
Use re-entrant locks on stdio
2022-05-22 15:13:13 +00:00
unsigned __tries = 0; \
Make fixes and improvements - Document more compiler flags - Expose new __print_maps() api - Better overflow checking in mmap() - Improve the shell example somewhat - Fix minor runtime bugs regarding stacks - Make kill() on fork()+execve()'d children work - Support CLONE_CHILD_CLEARTID for proper joining - Fix recent possible deadlock regression with --ftrace
2022-05-19 23:57:49 +00:00
for (;;) { \
__atomic_load(__lock, &__x, __ATOMIC_RELAXED); \
Make mutex calling code 10x tinier Calls to lock/unlock functions are now NOPs by default. The first time clone() is called, they get turned into CALL instructions. Doing this caused funcctions like fputc() to shrink from 85 bytes to 45+4 bytes. Since the ANSI solution of `(__threaded && lock())` inlines os much superfluous binary content into functions all over the place.
2022-06-13 02:33:42 +00:00
if (!__x && !_trylock(__lock)) { \
Make fixes and improvements - Document more compiler flags - Expose new __print_maps() api - Better overflow checking in mmap() - Improve the shell example somewhat - Fix minor runtime bugs regarding stacks - Make kill() on fork()+execve()'d children work - Support CLONE_CHILD_CLEARTID for proper joining - Fix recent possible deadlock regression with --ftrace
2022-05-19 23:57:49 +00:00
break; \
} else if (++__tries & 7) { \
__builtin_ia32_pause(); \
} else { \
Reduce makefile dependencies by 10% The includes in libc/calls/calls.h have now been refactored so that functions with struct parameters are declared in libc/calls/struct/
2022-06-09 03:01:28 +00:00
_spinlock_yield(); \
Make fixes and improvements - Document more compiler flags - Expose new __print_maps() api - Better overflow checking in mmap() - Improve the shell example somewhat - Fix minor runtime bugs regarding stacks - Make kill() on fork()+execve()'d children work - Support CLONE_CHILD_CLEARTID for proper joining - Fix recent possible deadlock regression with --ftrace
2022-05-19 23:57:49 +00:00
} \
} \
Make mutex calling code 10x tinier Calls to lock/unlock functions are now NOPs by default. The first time clone() is called, they get turned into CALL instructions. Doing this caused funcctions like fputc() to shrink from 85 bytes to 45+4 bytes. Since the ANSI solution of `(__threaded && lock())` inlines os much superfluous binary content into functions all over the place.
2022-06-13 02:33:42 +00:00
0; \
})
#define _trylock(lock) __atomic_test_and_set(lock, __ATOMIC_SEQ_CST)
Add raw memory visualization tool to redbean This change introduces a `-W /dev/pts/1` flag to redbean. What it does is use the mincore() system call to create a dual-screen terminal display that lets you troubleshoot the virtual address space. This is useful since page faults are an important thing to consider when using a forking web server. Now we have a colorful visualization of which pages are going to fault and which ones are resident in memory. The memory monitor, if enabled, spawns as a thread that just outputs ANSI codes to the second terminal in a loop. In order to make this happen using the new clone() polyfill, stdio is now thread safe. This change also introduces some new demo pages to redbean. It also polishes the demos we already have, to look a bit nicer and more presentable for the upcoming release, with better explanations too.
2022-05-14 11:33:58 +00:00
Reduce makefile dependencies by 10% The includes in libc/calls/calls.h have now been refactored so that functions with struct parameters are declared in libc/calls/struct/
2022-06-09 03:01:28 +00:00
void _spinlock_yield(void);
Add raw memory visualization tool to redbean This change introduces a `-W /dev/pts/1` flag to redbean. What it does is use the mincore() system call to create a dual-screen terminal display that lets you troubleshoot the virtual address space. This is useful since page faults are an important thing to consider when using a forking web server. Now we have a colorful visualization of which pages are going to fault and which ones are resident in memory. The memory monitor, if enabled, spawns as a thread that just outputs ANSI codes to the second terminal in a loop. In order to make this happen using the new clone() polyfill, stdio is now thread safe. This change also introduces some new demo pages to redbean. It also polishes the demos we already have, to look a bit nicer and more presentable for the upcoming release, with better explanations too.
2022-05-14 11:33:58 +00:00
Improve locks and signals - Introduce fast spinlock API - Double rand64() perf w/ spinlock - Improve raise() on New Technology - Support gettid() across platforms - Implement SA_NODEFER on New Technology - Move the lock intrinsics into LIBC_INTRIN - Make SIGTRAP recoverable on New Technology - Block SIGCHLD in wait4() on New Technology - Add threading prototypes for XNU and FreeBSD - Rewrite abort() fixing its minor bugs on XNU/NT - Shave down a lot of the content in libc/bits/bits.h - Let signal handlers modify CPU registers on New Technology
2022-04-12 12:20:17 +00:00
#endif /* COSMOPOLITAN_LIBC_INTRIN_SPINLOCK_H_ */
Reference in a new issue Copy permalink