From 1dab8ba334a1b53f3c8780bc8a4f8045a6ea0375 Mon Sep 17 00:00:00 2001 From: ringabout <43030857+ringabout@users.noreply.github.com> Date: Sun, 30 Oct 2022 00:11:40 +0800 Subject: move threads out of system (#20674) * move syslocks first * progress * clean up * go on * clean up * clean up * add imports syslocks * remove documentation * public deallocOsPages * fixes genode * fixes more * fixes boehmGC * cover more cases * fixes cyclic deps * fixes genode * cleanup * unpublic fields * cleanup * clean up --- lib/core/locks.nim | 3 +- lib/core/rlocks.nim | 3 +- lib/genode/alloc.nim | 2 +- lib/genode/env.nim | 8 +- lib/pure/concurrency/cpuinfo.nim | 2 +- lib/std/private/syslocks.nim | 234 +++++++++++++++++++++++ lib/std/private/threadtypes.nim | 176 +++++++++++++++++ lib/std/threads.nim | 276 +++++++++++++++++++++++++++ lib/system.nim | 12 +- lib/system/alloc.nim | 1 + lib/system/channels_builtin.nim | 2 + lib/system/gc_regions.nim | 1 + lib/system/syslocks.nim | 234 ----------------------- lib/system/threadimpl.nim | 107 +++++++++++ lib/system/threadlocalstorage.nim | 174 ++--------------- lib/system/threads.nim | 390 -------------------------------------- tools/kochdocs.nim | 1 - 17 files changed, 830 insertions(+), 796 deletions(-) create mode 100644 lib/std/private/syslocks.nim create mode 100644 lib/std/private/threadtypes.nim create mode 100644 lib/std/threads.nim delete mode 100644 lib/system/syslocks.nim create mode 100644 lib/system/threadimpl.nim delete mode 100644 lib/system/threads.nim diff --git a/lib/core/locks.nim b/lib/core/locks.nim index 92967b9db..ad0bff44d 100644 --- a/lib/core/locks.nim +++ b/lib/core/locks.nim @@ -18,8 +18,7 @@ when not compileOption("threads") and not defined(nimdoc): when false: # fix #12330 {.error: "Locks requires --threads:on option.".} -const insideRLocksModule = false -include "system/syslocks" +import std/private/syslocks type Lock* = SysLock ## Nim lock; whether this is re-entrant diff --git a/lib/core/rlocks.nim b/lib/core/rlocks.nim index 0444b9a83..bee5c1655 100644 --- a/lib/core/rlocks.nim +++ b/lib/core/rlocks.nim @@ -16,8 +16,7 @@ when not compileOption("threads") and not defined(nimdoc): # so they can replace each other seamlessly. {.error: "Rlocks requires --threads:on option.".} -const insideRLocksModule = true -include "system/syslocks" +import std/private/syslocks type RLock* = SysLock ## Nim lock, re-entrant diff --git a/lib/genode/alloc.nim b/lib/genode/alloc.nim index 3ddd3074b..efc25ac57 100644 --- a/lib/genode/alloc.nim +++ b/lib/genode/alloc.nim @@ -15,7 +15,7 @@ when not defined(genode): {.error: "Genode only module".} when not declared(GenodeEnv): - include genode/env + import genode/env type RamDataspaceCapability {. importcpp: "Genode::Ram_dataspace_capability", pure.} = object diff --git a/lib/genode/env.nim b/lib/genode/env.nim index ef4a25883..34abf576f 100644 --- a/lib/genode/env.nim +++ b/lib/genode/env.nim @@ -20,10 +20,10 @@ when not defined(genode): {.error: "Genode only include".} type - GenodeEnvObj {.importcpp: "Genode::Env", header: "", pure.} = object - GenodeEnvPtr = ptr GenodeEnvObj + GenodeEnvObj* {.importcpp: "Genode::Env", header: "", pure.} = object + GenodeEnvPtr* = ptr GenodeEnvObj -const runtimeEnvSym = "nim_runtime_env" +const runtimeEnvSym* = "nim_runtime_env" when not defined(nimscript): - var runtimeEnv {.importcpp: runtimeEnvSym.}: GenodeEnvPtr + var runtimeEnv* {.importcpp: runtimeEnvSym.}: GenodeEnvPtr diff --git a/lib/pure/concurrency/cpuinfo.nim b/lib/pure/concurrency/cpuinfo.nim index 16d32002d..1d2ff63e1 100644 --- a/lib/pure/concurrency/cpuinfo.nim +++ b/lib/pure/concurrency/cpuinfo.nim @@ -40,7 +40,7 @@ when defined(macosx) or defined(bsd): importc: "sysctl", nodecl.} when defined(genode): - include genode/env + import genode/env proc affinitySpaceTotal(env: GenodeEnvPtr): cuint {. importcpp: "@->cpu().affinity_space().total()".} diff --git a/lib/std/private/syslocks.nim b/lib/std/private/syslocks.nim new file mode 100644 index 000000000..dff339e9e --- /dev/null +++ b/lib/std/private/syslocks.nim @@ -0,0 +1,234 @@ +# +# +# Nim's Runtime Library +# (c) Copyright 2012 Andreas Rumpf +# +# See the file "copying.txt", included in this +# distribution, for details about the copyright. +# + +# Low level system locks and condition vars. + +{.push stackTrace: off.} + +when defined(windows): + type + Handle = int + + SysLock* {.importc: "CRITICAL_SECTION", + header: "", final, pure.} = object # CRITICAL_SECTION in WinApi + DebugInfo: pointer + LockCount: int32 + RecursionCount: int32 + OwningThread: int + LockSemaphore: int + SpinCount: int + + SysCond* {.importc: "RTL_CONDITION_VARIABLE", header: "".} = object + thePtr {.importc: "ptr".} : Handle + + proc initSysLock*(L: var SysLock) {.importc: "InitializeCriticalSection", + header: "".} + ## Initializes the lock `L`. + + proc tryAcquireSysAux(L: var SysLock): int32 {.importc: "TryEnterCriticalSection", + header: "".} + ## Tries to acquire the lock `L`. + + proc tryAcquireSys*(L: var SysLock): bool {.inline.} = + result = tryAcquireSysAux(L) != 0'i32 + + proc acquireSys*(L: var SysLock) {.importc: "EnterCriticalSection", + header: "".} + ## Acquires the lock `L`. + + proc releaseSys*(L: var SysLock) {.importc: "LeaveCriticalSection", + header: "".} + ## Releases the lock `L`. + + proc deinitSys*(L: var SysLock) {.importc: "DeleteCriticalSection", + header: "".} + + proc initializeConditionVariable( + conditionVariable: var SysCond + ) {.stdcall, noSideEffect, dynlib: "kernel32", importc: "InitializeConditionVariable".} + + proc sleepConditionVariableCS( + conditionVariable: var SysCond, + PCRITICAL_SECTION: var SysLock, + dwMilliseconds: int + ): int32 {.stdcall, noSideEffect, dynlib: "kernel32", importc: "SleepConditionVariableCS".} + + + proc signalSysCond*(hEvent: var SysCond) {.stdcall, noSideEffect, + dynlib: "kernel32", importc: "WakeConditionVariable".} + + proc broadcastSysCond*(hEvent: var SysCond) {.stdcall, noSideEffect, + dynlib: "kernel32", importc: "WakeAllConditionVariable".} + + proc initSysCond*(cond: var SysCond) {.inline.} = + initializeConditionVariable(cond) + proc deinitSysCond*(cond: var SysCond) {.inline.} = + discard + proc waitSysCond*(cond: var SysCond, lock: var SysLock) = + discard sleepConditionVariableCS(cond, lock, -1'i32) + +elif defined(genode): + const + Header = "genode_cpp/syslocks.h" + type + SysLock* {.importcpp: "Nim::SysLock", pure, final, + header: Header.} = object + SysCond* {.importcpp: "Nim::SysCond", pure, final, + header: Header.} = object + + proc initSysLock*(L: var SysLock) = discard + proc deinitSys*(L: var SysLock) = discard + proc acquireSys*(L: var SysLock) {.noSideEffect, importcpp.} + proc tryAcquireSys*(L: var SysLock): bool {.noSideEffect, importcpp.} + proc releaseSys*(L: var SysLock) {.noSideEffect, importcpp.} + + proc initSysCond*(L: var SysCond) = discard + proc deinitSysCond*(L: var SysCond) = discard + proc waitSysCond*(cond: var SysCond, lock: var SysLock) {. + noSideEffect, importcpp.} + proc signalSysCond*(cond: var SysCond) {. + noSideEffect, importcpp.} + proc broadcastSysCond*(cond: var SysCond) {. + noSideEffect, importcpp.} + +else: + type + SysLockObj {.importc: "pthread_mutex_t", pure, final, + header: """#include + #include """.} = object + when defined(linux) and defined(amd64): + abi: array[40 div sizeof(clong), clong] + + SysLockAttr* {.importc: "pthread_mutexattr_t", pure, final + header: """#include + #include """.} = object + when defined(linux) and defined(amd64): + abi: array[4 div sizeof(cint), cint] # actually a cint + + SysCondObj {.importc: "pthread_cond_t", pure, final, + header: """#include + #include """.} = object + when defined(linux) and defined(amd64): + abi: array[48 div sizeof(clonglong), clonglong] + + SysCondAttr {.importc: "pthread_condattr_t", pure, final + header: """#include + #include """.} = object + when defined(linux) and defined(amd64): + abi: array[4 div sizeof(cint), cint] # actually a cint + + SysLockType = distinct cint + + proc initSysLockAux(L: var SysLockObj, attr: ptr SysLockAttr) {. + importc: "pthread_mutex_init", header: "", noSideEffect.} + proc deinitSysAux(L: var SysLockObj) {.noSideEffect, + importc: "pthread_mutex_destroy", header: "".} + + proc acquireSysAux(L: var SysLockObj) {.noSideEffect, + importc: "pthread_mutex_lock", header: "".} + proc tryAcquireSysAux(L: var SysLockObj): cint {.noSideEffect, + importc: "pthread_mutex_trylock", header: "".} + + proc releaseSysAux(L: var SysLockObj) {.noSideEffect, + importc: "pthread_mutex_unlock", header: "".} + + when defined(ios): + # iOS will behave badly if sync primitives are moved in memory. In order + # to prevent this once and for all, we're doing an extra malloc when + # initializing the primitive. + type + SysLock* = ptr SysLockObj + SysCond* = ptr SysCondObj + + when not declared(c_malloc): + proc c_malloc(size: csize_t): pointer {. + importc: "malloc", header: "".} + proc c_free(p: pointer) {. + importc: "free", header: "".} + + proc initSysLock*(L: var SysLock, attr: ptr SysLockAttr = nil) = + L = cast[SysLock](c_malloc(csize_t(sizeof(SysLockObj)))) + initSysLockAux(L[], attr) + + proc deinitSys*(L: var SysLock) = + deinitSysAux(L[]) + c_free(L) + + template acquireSys*(L: var SysLock) = + acquireSysAux(L[]) + template tryAcquireSys*(L: var SysLock): bool = + tryAcquireSysAux(L[]) == 0'i32 + template releaseSys*(L: var SysLock) = + releaseSysAux(L[]) + else: + type + SysLock* = SysLockObj + SysCond* = SysCondObj + + template initSysLock*(L: var SysLock, attr: ptr SysLockAttr = nil) = + initSysLockAux(L, attr) + template deinitSys*(L: var SysLock) = + deinitSysAux(L) + template acquireSys*(L: var SysLock) = + acquireSysAux(L) + template tryAcquireSys*(L: var SysLock): bool = + tryAcquireSysAux(L) == 0'i32 + template releaseSys*(L: var SysLock) = + releaseSysAux(L) + + # rlocks + var SysLockType_Reentrant* {.importc: "PTHREAD_MUTEX_RECURSIVE", + header: "".}: SysLockType + proc initSysLockAttr*(a: var SysLockAttr) {. + importc: "pthread_mutexattr_init", header: "", noSideEffect.} + proc setSysLockType*(a: var SysLockAttr, t: SysLockType) {. + importc: "pthread_mutexattr_settype", header: "", noSideEffect.} + + # locks + proc initSysCondAux(cond: var SysCondObj, cond_attr: ptr SysCondAttr = nil) {. + importc: "pthread_cond_init", header: "", noSideEffect.} + proc deinitSysCondAux(cond: var SysCondObj) {.noSideEffect, + importc: "pthread_cond_destroy", header: "".} + + proc waitSysCondAux(cond: var SysCondObj, lock: var SysLockObj): cint {. + importc: "pthread_cond_wait", header: "", noSideEffect.} + proc signalSysCondAux(cond: var SysCondObj) {. + importc: "pthread_cond_signal", header: "", noSideEffect.} + proc broadcastSysCondAux(cond: var SysCondObj) {. + importc: "pthread_cond_broadcast", header: "", noSideEffect.} + + when defined(ios): + proc initSysCond*(cond: var SysCond, cond_attr: ptr SysCondAttr = nil) = + cond = cast[SysCond](c_malloc(csize_t(sizeof(SysCondObj)))) + initSysCondAux(cond[], cond_attr) + + proc deinitSysCond*(cond: var SysCond) = + deinitSysCondAux(cond[]) + c_free(cond) + + template waitSysCond*(cond: var SysCond, lock: var SysLock) = + discard waitSysCondAux(cond[], lock[]) + template signalSysCond*(cond: var SysCond) = + signalSysCondAux(cond[]) + template broadcastSysCond*(cond: var SysCond) = + broadcastSysCondAux(cond[]) + else: + template initSysCond*(cond: var SysCond, cond_attr: ptr SysCondAttr = nil) = + initSysCondAux(cond, cond_attr) + template deinitSysCond*(cond: var SysCond) = + deinitSysCondAux(cond) + + template waitSysCond*(cond: var SysCond, lock: var SysLock) = + discard waitSysCondAux(cond, lock) + template signalSysCond*(cond: var SysCond) = + signalSysCondAux(cond) + template broadcastSysCond*(cond: var SysCond) = + broadcastSysCondAux(cond) + +{.pop.} diff --git a/lib/std/private/threadtypes.nim b/lib/std/private/threadtypes.nim new file mode 100644 index 000000000..a1cdf21dc --- /dev/null +++ b/lib/std/private/threadtypes.nim @@ -0,0 +1,176 @@ +include system/inclrtl + +const hasSharedHeap* = defined(boehmgc) or defined(gogc) # don't share heaps; every thread has its own + +when defined(windows): + type + Handle* = int + SysThread* = Handle + WinThreadProc* = proc (x: pointer): int32 {.stdcall.} + + proc createThread*(lpThreadAttributes: pointer, dwStackSize: int32, + lpStartAddress: WinThreadProc, + lpParameter: pointer, + dwCreationFlags: int32, + lpThreadId: var int32): SysThread {. + stdcall, dynlib: "kernel32", importc: "CreateThread".} + + proc winSuspendThread*(hThread: SysThread): int32 {. + stdcall, dynlib: "kernel32", importc: "SuspendThread".} + + proc winResumeThread*(hThread: SysThread): int32 {. + stdcall, dynlib: "kernel32", importc: "ResumeThread".} + + proc waitForSingleObject*(hHandle: SysThread, dwMilliseconds: int32): int32 {. + stdcall, dynlib: "kernel32", importc: "WaitForSingleObject".} + + proc waitForMultipleObjects*(nCount: int32, + lpHandles: ptr SysThread, + bWaitAll: int32, + dwMilliseconds: int32): int32 {. + stdcall, dynlib: "kernel32", importc: "WaitForMultipleObjects".} + + proc terminateThread*(hThread: SysThread, dwExitCode: int32): int32 {. + stdcall, dynlib: "kernel32", importc: "TerminateThread".} + + proc setThreadAffinityMask*(hThread: SysThread, dwThreadAffinityMask: uint) {. + importc: "SetThreadAffinityMask", stdcall, header: "".} + +elif defined(genode): + const + GenodeHeader* = "genode_cpp/threads.h" + type + SysThread* {.importcpp: "Nim::SysThread", + header: GenodeHeader, final, pure.} = object + GenodeThreadProc* = proc (x: pointer) {.noconv.} + + proc initThread*(s: var SysThread, + env: GenodeEnv, + stackSize: culonglong, + entry: GenodeThreadProc, + arg: pointer, + affinity: cuint) {. + importcpp: "#.initThread(@)".} + + +else: + when not (defined(macosx) or defined(haiku)): + {.passl: "-pthread".} + + when not defined(haiku): + {.passc: "-pthread".} + + const + schedh = "#define _GNU_SOURCE\n#include " + pthreadh* = "#define _GNU_SOURCE\n#include " + + when not declared(Time): + when defined(linux): + type Time = clong + else: + type Time = int + + when (defined(linux) or defined(nintendoswitch)) and defined(amd64): + type + SysThread* {.importc: "pthread_t", + header: "" .} = distinct culong + Pthread_attr* {.importc: "pthread_attr_t", + header: "".} = object + abi: array[56 div sizeof(clong), clong] + elif defined(openbsd) and defined(amd64): + type + SysThread* {.importc: "pthread_t", header: "".} = object + Pthread_attr* {.importc: "pthread_attr_t", + header: "".} = object + else: + type + SysThread* {.importc: "pthread_t", header: "".} = int + Pthread_attr* {.importc: "pthread_attr_t", + header: "".} = object + type + Timespec* {.importc: "struct timespec", header: "".} = object + tv_sec*: Time + tv_nsec*: clong + + proc pthread_attr_init*(a1: var Pthread_attr): cint {. + importc, header: pthreadh.} + proc pthread_attr_setstack*(a1: ptr Pthread_attr, a2: pointer, a3: int): cint {. + importc, header: pthreadh.} + proc pthread_attr_setstacksize*(a1: var Pthread_attr, a2: int): cint {. + importc, header: pthreadh.} + proc pthread_attr_destroy*(a1: var Pthread_attr): cint {. + importc, header: pthreadh.} + + proc pthread_create*(a1: var SysThread, a2: var Pthread_attr, + a3: proc (x: pointer): pointer {.noconv.}, + a4: pointer): cint {.importc: "pthread_create", + header: pthreadh.} + proc pthread_join*(a1: SysThread, a2: ptr pointer): cint {. + importc, header: pthreadh.} + + proc pthread_cancel*(a1: SysThread): cint {. + importc: "pthread_cancel", header: pthreadh.} + + type CpuSet* {.importc: "cpu_set_t", header: schedh.} = object + when defined(linux) and defined(amd64): + abi: array[1024 div (8 * sizeof(culong)), culong] + + proc cpusetZero*(s: var CpuSet) {.importc: "CPU_ZERO", header: schedh.} + proc cpusetIncl*(cpu: cint; s: var CpuSet) {. + importc: "CPU_SET", header: schedh.} + + when defined(android): + # libc of android doesn't implement pthread_setaffinity_np, + # it exposes pthread_gettid_np though, so we can use that in combination + # with sched_setaffinity to set the thread affinity. + type Pid* {.importc: "pid_t", header: "".} = int32 # From posix_other.nim + + proc setAffinityTID*(tid: Pid; setsize: csize_t; s: var CpuSet) {. + importc: "sched_setaffinity", header: schedh.} + + proc pthread_gettid_np*(thread: SysThread): Pid {. + importc: "pthread_gettid_np", header: pthreadh.} + + proc setAffinity*(thread: SysThread; setsize: csize_t; s: var CpuSet) = + setAffinityTID(pthread_gettid_np(thread), setsize, s) + else: + proc setAffinity*(thread: SysThread; setsize: csize_t; s: var CpuSet) {. + importc: "pthread_setaffinity_np", header: pthreadh.} + + +const + emulatedThreadVars* = compileOption("tlsEmulation") +# we preallocate a fixed size for thread local storage, so that no heap +# allocations are needed. Currently less than 16K are used on a 64bit machine. +# We use `float` for proper alignment: +const nimTlsSize {.intdefine.} = 16000 +type + ThreadLocalStorage* = array[0..(nimTlsSize div sizeof(float)), float] + PGcThread* = ptr GcThread + GcThread* {.pure, inheritable.} = object + when emulatedThreadVars: + tls*: ThreadLocalStorage + else: + nil + when hasSharedHeap: + next*, prev*: PGcThread + stackBottom*, stackTop*: pointer + stackSize*: int + else: + nil + +const hasAllocStack* = defined(zephyr) # maybe freertos too? + +type + Thread*[TArg] = object + core*: PGcThread + sys*: SysThread + when TArg is void: + dataFn*: proc () {.nimcall, gcsafe.} + else: + dataFn*: proc (m: TArg) {.nimcall, gcsafe.} + data*: TArg + when hasAllocStack: + rawStack*: pointer + +proc `=copy`*[TArg](x: var Thread[TArg], y: Thread[TArg]) {.error.} diff --git a/lib/std/threads.nim b/lib/std/threads.nim new file mode 100644 index 000000000..5726a5cdc --- /dev/null +++ b/lib/std/threads.nim @@ -0,0 +1,276 @@ +# +# +# Nim's Runtime Library +# (c) Copyright 2012 Andreas Rumpf +# +# See the file "copying.txt", included in this +# distribution, for details about the copyright. +# + +## Thread support for Nim. +## +## Nim's memory model for threads is quite different from other common +## programming languages (C, Pascal): Each thread has its own +## (garbage collected) heap and sharing of memory is restricted. This helps +## to prevent race conditions and improves efficiency. See `the manual for +## details of this memory model `_. +## +## Examples +## ======== +## +## .. code-block:: Nim +## +## import std/locks +## +## var +## thr: array[0..4, Thread[tuple[a,b: int]]] +## L: Lock +## +## proc threadFunc(interval: tuple[a,b: int]) {.thread.} = +## for i in interval.a..interval.b: +## acquire(L) # lock stdout +## echo i +## release(L) +## +## initLock(L) +## +## for i in 0..high(thr): +## createThread(thr[i], threadFunc, (i*10, i*10+5)) +## joinThreads(thr) +## +## deinitLock(L) + +import std/private/[threadtypes] +export Thread + +import system/ansi_c + +when defined(nimPreviewSlimSystem): + import std/assertions + +when defined(genode): + import genode/env + + +when hasAllocStack or defined(zephyr) or defined(freertos): + const + nimThreadStackSize {.intdefine.} = 8192 + nimThreadStackGuard {.intdefine.} = 128 + + StackGuardSize = nimThreadStackGuard + ThreadStackSize = nimThreadStackSize - nimThreadStackGuard +else: + const + StackGuardSize = 4096 + ThreadStackMask = + when defined(genode): + 1024*64*sizeof(int)-1 + else: + 1024*256*sizeof(int)-1 + + ThreadStackSize = ThreadStackMask+1 - StackGuardSize + + +when defined(gcDestructors): + proc allocThreadStorage(size: int): pointer = + result = c_malloc(csize_t size) + zeroMem(result, size) +else: + template allocThreadStorage(size: untyped): untyped = allocShared0(size) + +#const globalsSlot = ThreadVarSlot(0) +#sysAssert checkSlot.int == globalsSlot.int + +# Zephyr doesn't include this properly without some help +when defined(zephyr): + {.emit: """/*INCLUDESECTION*/ + #include + """.} + + +# We jump through some hops here to ensure that Nim thread procs can have +# the Nim calling convention. This is needed because thread procs are +# ``stdcall`` on Windows and ``noconv`` on UNIX. Alternative would be to just +# use ``stdcall`` since it is mapped to ``noconv`` on UNIX anyway. + + + +proc onThreadDestruction*(handler: proc () {.closure, gcsafe, raises: [].}) = + ## Registers a *thread local* handler that is called at the thread's + ## destruction. + ## + ## A thread is destructed when the `.thread` proc returns + ## normally or when it raises an exception. Note that unhandled exceptions + ## in a thread nevertheless cause the whole process to die. + nimThreadDestructionHandlers.add handler + + +{.push stack_trace:off.} +when defined(windows): + proc threadProcWrapper[TArg](closure: pointer): int32 {.stdcall.} = + nimThreadProcWrapperBody(closure) + # implicitly return 0 +elif defined(genode): + proc threadProcWrapper[TArg](closure: pointer) {.noconv.} = + nimThreadProcWrapperBody(closure) +else: + proc threadProcWrapper[TArg](closure: pointer): pointer {.noconv.} = + nimThreadProcWrapperBody(closure) +{.pop.} + +proc running*[TArg](t: Thread[TArg]): bool {.inline.} = + ## Returns true if `t` is running. + result = t.dataFn != nil + +proc handle*[TArg](t: Thread[TArg]): SysThread {.inline.} = + ## Returns the thread handle of `t`. + result = t.sys + +when hostOS == "windows": + const MAXIMUM_WAIT_OBJECTS = 64 + + proc joinThread*[TArg](t: Thread[TArg]) {.inline.} = + ## Waits for the thread `t` to finish. + discard waitForSingleObject(t.sys, -1'i32) + + proc joinThreads*[TArg](t: varargs[Thread[TArg]]) = + ## Waits for every thread in `t` to finish. + var a: array[MAXIMUM_WAIT_OBJECTS, SysThread] + var k = 0 + while k < len(t): + var count = min(len(t) - k, MAXIMUM_WAIT_OBJECTS) + for i in 0..(count - 1): a[i] = t[i + k].sys + discard waitForMultipleObjects(int32(count), + cast[ptr SysThread](addr(a)), 1, -1) + inc(k, MAXIMUM_WAIT_OBJECTS) + +elif defined(genode): + proc joinThread*[TArg](t: Thread[TArg]) {.importcpp.} + ## Waits for the thread `t` to finish. + + proc joinThreads*[TArg](t: varargs[Thread[TArg]]) = + ## Waits for every thread in `t` to finish. + for i in 0..t.high: joinThread(t[i]) + +else: + proc joinThread*[TArg](t: Thread[TArg]) {.inline.} = + ## Waits for the thread `t` to finish. + discard pthread_join(t.sys, nil) + + proc joinThreads*[TArg](t: varargs[Thread[TArg]]) = + ## Waits for every thread in `t` to finish. + for i in 0..t.high: joinThread(t[i]) + +when false: + # XXX a thread should really release its heap here somehow: + proc destroyThread*[TArg](t: var Thread[TArg]) = + ## Forces the thread `t` to terminate. This is potentially dangerous if + ## you don't have full control over `t` and its acquired resources. + when hostOS == "windows": + discard TerminateThread(t.sys, 1'i32) + else: + discard pthread_cancel(t.sys) + when declared(registerThread): unregisterThread(addr(t)) + t.dataFn = nil + ## if thread `t` already exited, `t.core` will be `null`. + if not isNil(t.core): + deallocThreadStorage(t.core) + t.core = nil + +when hostOS == "windows": + proc createThread*[TArg](t: var Thread[TArg], + tp: proc (arg: TArg) {.thread, nimcall.}, + param: TArg) = + ## Creates a new thread `t` and starts its execution. + ## + ## Entry point is the proc `tp`. + ## `param` is passed to `tp`. `TArg` can be `void` if you + ## don't need to pass any data to the thread. + t.core = cast[PGcThread](allocThreadStorage(sizeof(GcThread))) + + when TArg isnot void: t.data = param + t.dataFn = tp + when hasSharedHeap: t.core.stackSize = ThreadStackSize + var dummyThreadId: int32 + t.sys = createThread(nil, ThreadStackSize, threadProcWrapper[TArg], + addr(t), 0'i32, dummyThreadId) + if t.sys <= 0: + raise newException(ResourceExhaustedError, "cannot create thread") + + proc pinToCpu*[Arg](t: var Thread[Arg]; cpu: Natural) = + ## Pins a thread to a `CPU`:idx:. + ## + ## In other words sets a thread's `affinity`:idx:. + ## If you don't know what this means, you shouldn't use this proc. + setThreadAffinityMask(t.sys, uint(1 shl cpu)) + +elif defined(genode): + var affinityOffset: cuint = 1 + ## CPU affinity offset for next thread, safe to roll-over. + + proc createThread*[TArg](t: var Thread[TArg], + tp: proc (arg: TArg) {.thread, nimcall.}, + param: TArg) = + t.core = cast[PGcThread](allocThreadStorage(sizeof(GcThread))) + + when TArg isnot void: t.data = param + t.dataFn = tp + when hasSharedHeap: t.stackSize = ThreadStackSize + t.sys.initThread( + runtimeEnv, + ThreadStackSize.culonglong, + threadProcWrapper[TArg], addr(t), affinityOffset) + inc affinityOffset + + proc pinToCpu*[Arg](t: var Thread[Arg]; cpu: Natural) = + {.hint: "cannot change Genode thread CPU affinity after initialization".} + discard + +else: + proc createThread*[TArg](t: var Thread[TArg], + tp: proc (arg: TArg) {.thread, nimcall.}, + param: TArg) = + ## Creates a new thread `t` and starts its execution. + ## + ## Entry point is the proc `tp`. `param` is passed to `tp`. + ## `TArg` can be `void` if you + ## don't need to pass any data to the thread. + t.core = cast[PGcThread](allocThreadStorage(sizeof(GcThread))) + + when TArg isnot void: t.data = param + t.dataFn = tp + when hasSharedHeap: t.core.stackSize = ThreadStackSize + var a {.noinit.}: Pthread_attr + doAssert pthread_attr_init(a) == 0 + when hasAllocStack: + var + rawstk = allocThreadStorage(ThreadStackSize + StackGuardSize) + stk = cast[pointer](cast[uint](rawstk) + StackGuardSize) + let setstacksizeResult = pthread_attr_setstack(addr a, stk, ThreadStackSize) + t.rawStack = rawstk + else: + let setstacksizeResult = pthread_attr_setstacksize(a, ThreadStackSize) + + when not defined(ios): + # This fails on iOS + doAssert(setstacksizeResult == 0) + if pthread_create(t.sys, a, threadProcWrapper[TArg], addr(t)) != 0: + raise newException(ResourceExhaustedError, "cannot create thread") + doAssert pthread_attr_destroy(a) == 0 + + proc pinToCpu*[Arg](t: var Thread[Arg]; cpu: Natural) = + ## Pins a thread to a `CPU`:idx:. + ## + ## In other words sets a thread's `affinity`:idx:. + ## If you don't know what this means, you shouldn't use this proc. + when not defined(macosx): + var s {.noinit.}: CpuSet + cpusetZero(s) + cpusetIncl(cpu.cint, s) + setAffinity(t.sys, csize_t(sizeof(s)), s) + +proc createThread*(t: var Thread[void], tp: proc () {.thread, nimcall.}) = + createThread[void](t, tp) + +when not defined(gcOrc): + include system/threadids diff --git a/lib/system.nim b/lib/system.nim index c28479a44..ff2bff45f 100644 --- a/lib/system.nim +++ b/lib/system.nim @@ -1115,7 +1115,7 @@ elif defined(nimdoc): proc quit*(errorcode: int = QuitSuccess) {.magic: "Exit", noreturn.} elif defined(genode): - include genode/env + import genode/env var systemEnv {.exportc: runtimeEnvSym.}: GenodeEnvPtr @@ -1609,8 +1609,7 @@ when notJSnotNims: {.push stackTrace: off.} when not defined(js) and hasThreadSupport and hostOS != "standalone": - const insideRLocksModule = false - include "system/syslocks" + import std/private/syslocks include "system/threadlocalstorage" when not defined(js) and defined(nimV2): @@ -2092,7 +2091,12 @@ when not defined(js): when declared(initAllocator): initAllocator() when hasThreadSupport: - when hostOS != "standalone": include "system/threads" + when hostOS != "standalone": + include system/threadimpl + + import std/threads + export threads + elif not defined(nogc) and not defined(nimscript): when not defined(useNimRtl) and not defined(createNimRtl): initStackBottom() when declared(initGC): initGC() diff --git a/lib/system/alloc.nim b/lib/system/alloc.nim index 88f680500..5350c1db8 100644 --- a/lib/system/alloc.nim +++ b/lib/system/alloc.nim @@ -11,6 +11,7 @@ {.push profiler:off.} include osalloc +import std/private/syslocks template track(op, address, size) = when defined(memTracker): diff --git a/lib/system/channels_builtin.nim b/lib/system/channels_builtin.nim index 2ad589383..50d740b23 100644 --- a/lib/system/channels_builtin.nim +++ b/lib/system/channels_builtin.nim @@ -139,6 +139,8 @@ when not declared(ThisIsSystem): {.error: "You must not import this module explicitly".} +import std/private/syslocks + type pbytes = ptr UncheckedArray[byte] RawChannel {.pure, final.} = object ## msg queue for a thread diff --git a/lib/system/gc_regions.nim b/lib/system/gc_regions.nim index 6ced04c99..d96de7eac 100644 --- a/lib/system/gc_regions.nim +++ b/lib/system/gc_regions.nim @@ -7,6 +7,7 @@ # # "Stack GC" for embedded devices or ultra performance requirements. +import std/private/syslocks when defined(memProfiler): proc nimProfile(requestedSize: int) {.benign.} diff --git a/lib/system/syslocks.nim b/lib/system/syslocks.nim deleted file mode 100644 index 2f0c8b0ba..000000000 --- a/lib/system/syslocks.nim +++ /dev/null @@ -1,234 +0,0 @@ -# -# -# Nim's Runtime Library -# (c) Copyright 2012 Andreas Rumpf -# -# See the file "copying.txt", included in this -# distribution, for details about the copyright. -# - -# Low level system locks and condition vars. - -{.push stackTrace: off.} - -when defined(windows): - type - Handle = int - - SysLock {.importc: "CRITICAL_SECTION", - header: "", final, pure.} = object # CRITICAL_SECTION in WinApi - DebugInfo: pointer - LockCount: int32 - RecursionCount: int32 - OwningThread: int - LockSemaphore: int - SpinCount: int - - SysCond {.importc: "RTL_CONDITION_VARIABLE", header: "".} = object - thePtr {.importc: "ptr".} : Handle - - proc initSysLock(L: var SysLock) {.importc: "InitializeCriticalSection", - header: "".} - ## Initializes the lock `L`. - - proc tryAcquireSysAux(L: var SysLock): int32 {.importc: "TryEnterCriticalSection", - header: "".} - ## Tries to acquire the lock `L`. - - proc tryAcquireSys(L: var SysLock): bool {.inline.} = - result = tryAcquireSysAux(L) != 0'i32 - - proc acquireSys(L: var SysLock) {.importc: "EnterCriticalSection", - header: "".} - ## Acquires the lock `L`. - - proc releaseSys(L: var SysLock) {.importc: "LeaveCriticalSection", - header: "".} - ## Releases the lock `L`. - - proc deinitSys(L: var SysLock) {.importc: "DeleteCriticalSection", - header: "".} - - proc initializeConditionVariable( - conditionVariable: var SysCond - ) {.stdcall, noSideEffect, dynlib: "kernel32", importc: "InitializeConditionVariable".} - - proc sleepConditionVariableCS( - conditionVariable: var SysCond, - PCRITICAL_SECTION: var SysLock, - dwMilliseconds: int - ): int32 {.stdcall, noSideEffect, dynlib: "kernel32", importc: "SleepConditionVariableCS".} - - - proc signalSysCond(hEvent: var SysCond) {.stdcall, noSideEffect, - dynlib: "kernel32", importc: "WakeConditionVariable".} - - proc broadcastSysCond(hEvent: var SysCond) {.stdcall, noSideEffect, - dynlib: "kernel32", importc: "WakeAllConditionVariable".} - - proc initSysCond(cond: var SysCond) {.inline.} = - initializeConditionVariable(cond) - proc deinitSysCond(cond: var SysCond) {.inline.} = - discard - proc waitSysCond(cond: var SysCond, lock: var SysLock) = - discard sleepConditionVariableCS(cond, lock, -1'i32) - -elif defined(genode): - const - Header = "genode_cpp/syslocks.h" - type - SysLock {.importcpp: "Nim::SysLock", pure, final, - header: Header.} = object - SysCond {.importcpp: "Nim::SysCond", pure, final, - header: Header.} = object - - proc initSysLock(L: var SysLock) = discard - proc deinitSys(L: var SysLock) = discard - proc acquireSys(L: var SysLock) {.noSideEffect, importcpp.} - proc tryAcquireSys(L: var SysLock): bool {.noSideEffect, importcpp.} - proc releaseSys(L: var SysLock) {.noSideEffect, importcpp.} - - proc initSysCond(L: var SysCond) = discard - proc deinitSysCond(L: var SysCond) = discard - proc waitSysCond(cond: var SysCond, lock: var SysLock) {. - noSideEffect, importcpp.} - proc signalSysCond(cond: var SysCond) {. - noSideEffect, importcpp.} - proc broadcastSysCond(cond: var SysCond) {. - noSideEffect, importcpp.} - -else: - type - SysLockObj {.importc: "pthread_mutex_t", pure, final, - header: """#include - #include """.} = object - when defined(linux) and defined(amd64): - abi: array[40 div sizeof(clong), clong] - - SysLockAttr {.importc: "pthread_mutexattr_t", pure, final - header: """#include - #include """.} = object - when defined(linux) and defined(amd64): - abi: array[4 div sizeof(cint), cint] # actually a cint - - SysCondObj {.importc: "pthread_cond_t", pure, final, - header: """#include - #include """.} = object - when defined(linux) and defined(amd64): - abi: array[48 div sizeof(clonglong), clonglong] - - SysCondAttr {.importc: "pthread_condattr_t", pure, final - header: """#include - #include """.} = object - when defined(linux) and defined(amd64): - abi: array[4 div sizeof(cint), cint] # actually a cint - - SysLockType = distinct cint - - proc initSysLockAux(L: var SysLockObj, attr: ptr SysLockAttr) {. - importc: "pthread_mutex_init", header: "", noSideEffect.} - proc deinitSysAux(L: var SysLockObj) {.noSideEffect, - importc: "pthread_mutex_destroy", header: "".} - - proc acquireSysAux(L: var SysLockObj) {.noSideEffect, - importc: "pthread_mutex_lock", header: "".} - proc tryAcquireSysAux(L: var SysLockObj): cint {.noSideEffect, - importc: "pthread_mutex_trylock", header: "".} - - proc releaseSysAux(L: var SysLockObj) {.noSideEffect, - importc: "pthread_mutex_unlock", header: "".} - - when defined(ios): - # iOS will behave badly if sync primitives are moved in memory. In order - # to prevent this once and for all, we're doing an extra malloc when - # initializing the primitive. - type - SysLock = ptr SysLockObj - SysCond = ptr SysCondObj - - when not declared(c_malloc): - proc c_malloc(size: csize_t): pointer {. - importc: "malloc", header: "".} - proc c_free(p: pointer) {. - importc: "free", header: "".} - - proc initSysLock(L: var SysLock, attr: ptr SysLockAttr = nil) = - L = cast[SysLock](c_malloc(csize_t(sizeof(SysLockObj)))) - initSysLockAux(L[], attr) - - proc deinitSys(L: var SysLock) = - deinitSysAux(L[]) - c_free(L) - - template acquireSys(L: var SysLock) = - acquireSysAux(L[]) - template tryAcquireSys(L: var SysLock): bool = - tryAcquireSysAux(L[]) == 0'i32 - template releaseSys(L: var SysLock) = - releaseSysAux(L[]) - else: - type - SysLock = SysLockObj - SysCond = SysCondObj - - template initSysLock(L: var SysLock, attr: ptr SysLockAttr = nil) = - initSysLockAux(L, attr) - template deinitSys(L: var SysLock) = - deinitSysAux(L) - template acquireSys(L: var SysLock) = - acquireSysAux(L) - template tryAcquireSys(L: var SysLock): bool = - tryAcquireSysAux(L) == 0'i32 - template releaseSys(L: var SysLock) = - releaseSysAux(L) - - when insideRLocksModule: - let SysLockType_Reentrant {.importc: "PTHREAD_MUTEX_RECURSIVE", - header: "".}: SysLockType - proc initSysLockAttr(a: var SysLockAttr) {. - importc: "pthread_mutexattr_init", header: "", noSideEffect.} - proc setSysLockType(a: var SysLockAttr, t: SysLockType) {. - importc: "pthread_mutexattr_settype", header: "", noSideEffect.} - - else: - proc initSysCondAux(cond: var SysCondObj, cond_attr: ptr SysCondAttr = nil) {. - importc: "pthread_cond_init", header: "", noSideEffect.} - proc deinitSysCondAux(cond: var SysCondObj) {.noSideEffect, - importc: "pthread_cond_destroy", header: "".} - - proc waitSysCondAux(cond: var SysCondObj, lock: var SysLockObj): cint {. - importc: "pthread_cond_wait", header: "", noSideEffect.} - proc signalSysCondAux(cond: var SysCondObj) {. - importc: "pthread_cond_signal", header: "", noSideEffect.} - proc broadcastSysCondAux(cond: var SysCondObj) {. - importc: "pthread_cond_broadcast", header: "", noSideEffect.} - - when defined(ios): - proc initSysCond(cond: var SysCond, cond_attr: ptr SysCondAttr = nil) = - cond = cast[SysCond](c_malloc(csize_t(sizeof(SysCondObj)))) - initSysCondAux(cond[], cond_attr) - - proc deinitSysCond(cond: var SysCond) = - deinitSysCondAux(cond[]) - c_free(cond) - - template waitSysCond(cond: var SysCond, lock: var SysLock) = - discard waitSysCondAux(cond[], lock[]) - template signalSysCond(cond: var SysCond) = - signalSysCondAux(cond[]) - template broadcastSysCond(cond: var SysCond) = - broadcastSysCondAux(cond[]) - else: - template initSysCond(cond: var SysCond, cond_attr: ptr SysCondAttr = nil) = - initSysCondAux(cond, cond_attr) - template deinitSysCond(cond: var SysCond) = - deinitSysCondAux(cond) - - template waitSysCond(cond: var SysCond, lock: var SysLock) = - discard waitSysCondAux(cond, lock) - template signalSysCond(cond: var SysCond) = - signalSysCondAux(cond) - template broadcastSysCond(cond: var SysCond) = - broadcastSysCondAux(cond) - -{.pop.} diff --git a/lib/system/threadimpl.nim b/lib/system/threadimpl.nim new file mode 100644 index 000000000..73f2807d2 --- /dev/null +++ b/lib/system/threadimpl.nim @@ -0,0 +1,107 @@ +var + nimThreadDestructionHandlers* {.rtlThreadVar.}: seq[proc () {.closure, gcsafe, raises: [].}] +when not defined(boehmgc) and not hasSharedHeap and not defined(gogc) and not defined(gcRegions): + proc deallocOsPages() {.rtl, raises: [].} +proc threadTrouble() {.raises: [], gcsafe.} +# create for the main thread. Note: do not insert this data into the list +# of all threads; it's not to be stopped etc. +when not defined(useNimRtl): + #when not defined(createNimRtl): initStackBottom() + when declared(initGC): + initGC() + when not emulatedThreadVars: + type ThreadType {.pure.} = enum + None = 0, + NimThread = 1, + ForeignThread = 2 + var + threadType {.rtlThreadVar.}: ThreadType + + threadType = ThreadType.NimThread + +when defined(gcDestructors): + proc allocThreadStorage(size: int): pointer = + result = c_malloc(csize_t size) + zeroMem(result, size) + + proc deallocThreadStorage(p: pointer) = c_free(p) +else: + template allocThreadStorage(size: untyped): untyped = allocShared0(size) + template deallocThreadStorage(p: pointer) = deallocShared(p) + +template afterThreadRuns() = + for i in countdown(nimThreadDestructionHandlers.len-1, 0): + nimThreadDestructionHandlers[i]() + +when defined(boehmgc): + type GCStackBaseProc = proc(sb: pointer, t: pointer) {.noconv.} + proc boehmGC_call_with_stack_base(sbp: GCStackBaseProc, p: pointer) + {.importc: "GC_call_with_stack_base", boehmGC.} + proc boehmGC_register_my_thread(sb: pointer) + {.importc: "GC_register_my_thread", boehmGC.} + proc boehmGC_unregister_my_thread() + {.importc: "GC_unregister_my_thread", boehmGC.} + + proc threadProcWrapDispatch[TArg](sb: pointer, thrd: pointer) {.noconv, raises: [].} = + boehmGC_register_my_thread(sb) + try: + let thrd = cast[ptr Thread[TArg]](thrd) + when TArg is void: + thrd.dataFn() + else: + thrd.dataFn(thrd.data) + except: + threadTrouble() + finally: + afterThreadRuns() + boehmGC_unregister_my_thread() +else: + proc threadProcWrapDispatch[TArg](thrd: ptr Thread[TArg]) {.raises: [].} = + try: + when TArg is void: + thrd.dataFn() + else: + when defined(nimV2): + thrd.dataFn(thrd.data) + else: + var x: TArg + deepCopy(x, thrd.data) + thrd.dataFn(x) + except: + threadTrouble() + finally: + afterThreadRuns() + when hasAllocStack: + deallocThreadStorage(thrd.rawStack) + +proc threadProcWrapStackFrame[TArg](thrd: ptr Thread[TArg]) {.raises: [].} = + when defined(boehmgc): + boehmGC_call_with_stack_base(threadProcWrapDispatch[TArg], thrd) + elif not defined(nogc) and not defined(gogc) and not defined(gcRegions) and not usesDestructors: + var p {.volatile.}: pointer + # init the GC for refc/markandsweep + nimGC_setStackBottom(addr(p)) + when declared(initGC): + initGC() + when declared(threadType): + threadType = ThreadType.NimThread + threadProcWrapDispatch[TArg](thrd) + when declared(deallocOsPages): deallocOsPages() + else: + threadProcWrapDispatch(thrd) + +template nimThreadProcWrapperBody*(closure: untyped): untyped = + var thrd = cast[ptr Thread[TArg]](closure) + var core = thrd.core + when declared(globalsSlot): threadVarSetValue(globalsSlot, thrd.core) + threadProcWrapStackFrame(thrd) + # Since an unhandled exception terminates the whole process (!), there is + # no need for a ``try finally`` here, nor would it be correct: The current + # exception is tried to be re-raised by the code-gen after the ``finally``! + # However this is doomed to fail, because we already unmapped every heap + # page! + + # mark as not running anymore: + thrd.core = nil + thrd.dataFn = nil + deallocThreadStorage(cast[pointer](core)) diff --git a/lib/system/threadlocalstorage.nim b/lib/system/threadlocalstorage.nim index b62c903c3..977f42e72 100644 --- a/lib/system/threadlocalstorage.nim +++ b/lib/system/threadlocalstorage.nim @@ -1,83 +1,33 @@ +import std/private/threadtypes when defined(windows): - type - SysThread* = Handle - WinThreadProc = proc (x: pointer): int32 {.stdcall.} - - proc createThread(lpThreadAttributes: pointer, dwStackSize: int32, - lpStartAddress: WinThreadProc, - lpParameter: pointer, - dwCreationFlags: int32, - lpThreadId: var int32): SysThread {. - stdcall, dynlib: "kernel32", importc: "CreateThread".} - - proc winSuspendThread(hThread: SysThread): int32 {. - stdcall, dynlib: "kernel32", importc: "SuspendThread".} - - proc winResumeThread(hThread: SysThread): int32 {. - stdcall, dynlib: "kernel32", importc: "ResumeThread".} - - proc waitForSingleObject(hHandle: SysThread, dwMilliseconds: int32): int32 {. - stdcall, dynlib: "kernel32", importc: "WaitForSingleObject".} - - proc waitForMultipleObjects(nCount: int32, - lpHandles: ptr SysThread, - bWaitAll: int32, - dwMilliseconds: int32): int32 {. - stdcall, dynlib: "kernel32", importc: "WaitForMultipleObjects".} - - proc terminateThread(hThread: SysThread, dwExitCode: int32): int32 {. - stdcall, dynlib: "kernel32", importc: "TerminateThread".} - type ThreadVarSlot = distinct int32 - when true: - proc threadVarAlloc(): ThreadVarSlot {. - importc: "TlsAlloc", stdcall, header: "".} - proc threadVarSetValue(dwTlsIndex: ThreadVarSlot, lpTlsValue: pointer) {. - importc: "TlsSetValue", stdcall, header: "".} - proc tlsGetValue(dwTlsIndex: ThreadVarSlot): pointer {. - importc: "TlsGetValue", stdcall, header: "".} + proc threadVarAlloc(): ThreadVarSlot {. + importc: "TlsAlloc", stdcall, header: "".} + proc threadVarSetValue(dwTlsIndex: ThreadVarSlot, lpTlsValue: pointer) {. + importc: "TlsSetValue", stdcall, header: "".} + proc tlsGetValue(dwTlsIndex: ThreadVarSlot): pointer {. + importc: "TlsGetValue", stdcall, header: "".} - proc getLastError(): uint32 {. - importc: "GetLastError", stdcall, header: "".} - proc setLastError(x: uint32) {. - importc: "SetLastError", stdcall, header: "".} + proc getLastError(): uint32 {. + importc: "GetLastError", stdcall, header: "".} + proc setLastError(x: uint32) {. + importc: "SetLastError", stdcall, header: "".} - proc threadVarGetValue(dwTlsIndex: ThreadVarSlot): pointer = - let realLastError = getLastError() - result = tlsGetValue(dwTlsIndex) - setLastError(realLastError) - else: - proc threadVarAlloc(): ThreadVarSlot {. - importc: "TlsAlloc", stdcall, dynlib: "kernel32".} - proc threadVarSetValue(dwTlsIndex: ThreadVarSlot, lpTlsValue: pointer) {. - importc: "TlsSetValue", stdcall, dynlib: "kernel32".} - proc threadVarGetValue(dwTlsIndex: ThreadVarSlot): pointer {. - importc: "TlsGetValue", stdcall, dynlib: "kernel32".} - - proc setThreadAffinityMask(hThread: SysThread, dwThreadAffinityMask: uint) {. - importc: "SetThreadAffinityMask", stdcall, header: "".} + proc threadVarGetValue(dwTlsIndex: ThreadVarSlot): pointer = + let realLastError = getLastError() + result = tlsGetValue(dwTlsIndex) + setLastError(realLastError) elif defined(genode): - import genode/env const GenodeHeader = "genode_cpp/threads.h" + type - SysThread* {.importcpp: "Nim::SysThread", - header: GenodeHeader, final, pure.} = object - GenodeThreadProc = proc (x: pointer) {.noconv.} ThreadVarSlot = int - proc initThread(s: var SysThread, - env: GenodeEnv, - stackSize: culonglong, - entry: GenodeThreadProc, - arg: pointer, - affinity: cuint) {. - importcpp: "#.initThread(@)".} - proc threadVarAlloc(): ThreadVarSlot = 0 proc offMainThread(): bool {. @@ -113,62 +63,18 @@ else: when not defined(haiku): {.passc: "-pthread".} - const - schedh = "#define _GNU_SOURCE\n#include " - pthreadh = "#define _GNU_SOURCE\n#include " - - when not declared(Time): - when defined(linux): - type Time = clong - else: - type Time = int - when (defined(linux) or defined(nintendoswitch)) and defined(amd64): type - SysThread* {.importc: "pthread_t", - header: "" .} = distinct culong - Pthread_attr {.importc: "pthread_attr_t", - header: "".} = object - abi: array[56 div sizeof(clong), clong] ThreadVarSlot {.importc: "pthread_key_t", header: "".} = distinct cuint elif defined(openbsd) and defined(amd64): type - SysThread* {.importc: "pthread_t", header: "".} = object - Pthread_attr {.importc: "pthread_attr_t", - header: "".} = object ThreadVarSlot {.importc: "pthread_key_t", header: "".} = cint else: type - SysThread* {.importc: "pthread_t", header: "".} = int - Pthread_attr {.importc: "pthread_attr_t", - header: "".} = object ThreadVarSlot {.importc: "pthread_key_t", header: "".} = object - type - Timespec {.importc: "struct timespec", header: "".} = object - tv_sec: Time - tv_nsec: clong - - proc pthread_attr_init(a1: var Pthread_attr): cint {. - importc, header: pthreadh.} - proc pthread_attr_setstack*(a1: ptr Pthread_attr, a2: pointer, a3: int): cint {. - importc, header: pthreadh.} - proc pthread_attr_setstacksize(a1: var Pthread_attr, a2: int): cint {. - importc, header: pthreadh.} - proc pthread_attr_destroy(a1: var Pthread_attr): cint {. - importc, header: pthreadh.} - - proc pthread_create(a1: var SysThread, a2: var Pthread_attr, - a3: proc (x: pointer): pointer {.noconv.}, - a4: pointer): cint {.importc: "pthread_create", - header: pthreadh.} - proc pthread_join(a1: SysThread, a2: ptr pointer): cint {. - importc, header: pthreadh.} - - proc pthread_cancel(a1: SysThread): cint {. - importc: "pthread_cancel", header: pthreadh.} proc pthread_getspecific(a1: ThreadVarSlot): pointer {. importc: "pthread_getspecific", header: pthreadh.} @@ -188,59 +94,13 @@ else: proc threadVarGetValue(s: ThreadVarSlot): pointer {.inline.} = result = pthread_getspecific(s) - type CpuSet {.importc: "cpu_set_t", header: schedh.} = object - when defined(linux) and defined(amd64): - abi: array[1024 div (8 * sizeof(culong)), culong] - - proc cpusetZero(s: var CpuSet) {.importc: "CPU_ZERO", header: schedh.} - proc cpusetIncl(cpu: cint; s: var CpuSet) {. - importc: "CPU_SET", header: schedh.} - - when defined(android): - # libc of android doesn't implement pthread_setaffinity_np, - # it exposes pthread_gettid_np though, so we can use that in combination - # with sched_setaffinity to set the thread affinity. - type Pid {.importc: "pid_t", header: "".} = int32 # From posix_other.nim - - proc setAffinityTID(tid: Pid; setsize: csize_t; s: var CpuSet) {. - importc: "sched_setaffinity", header: schedh.} - - proc pthread_gettid_np(thread: SysThread): Pid {. - importc: "pthread_gettid_np", header: pthreadh.} - - proc setAffinity(thread: SysThread; setsize: csize_t; s: var CpuSet) = - setAffinityTID(pthread_gettid_np(thread), setsize, s) - else: - proc setAffinity(thread: SysThread; setsize: csize_t; s: var CpuSet) {. - importc: "pthread_setaffinity_np", header: pthreadh.} - - -const - emulatedThreadVars = compileOption("tlsEmulation") when emulatedThreadVars: # the compiler generates this proc for us, so that we can get the size of # the thread local var block; we use this only for sanity checking though proc nimThreadVarsSize(): int {.noconv, importc: "NimThreadVarsSize".} -# we preallocate a fixed size for thread local storage, so that no heap -# allocations are needed. Currently less than 16K are used on a 64bit machine. -# We use `float` for proper alignment: -const nimTlsSize {.intdefine.} = 16000 -type - ThreadLocalStorage = array[0..(nimTlsSize div sizeof(float)), float] - PGcThread = ptr GcThread - GcThread {.pure, inheritable.} = object - when emulatedThreadVars: - tls: ThreadLocalStorage - else: - nil - when hasSharedHeap: - next, prev: PGcThread - stackBottom, stackTop: pointer - stackSize: int - else: - nil + when emulatedThreadVars: var globalsSlot: ThreadVarSlot diff --git a/lib/system/threads.nim b/lib/system/threads.nim deleted file mode 100644 index 4e190e443..000000000 --- a/lib/system/threads.nim +++ /dev/null @@ -1,390 +0,0 @@ -# -# -# Nim's Runtime Library -# (c) Copyright 2012 Andreas Rumpf -# -# See the file "copying.txt", included in this -# distribution, for details about the copyright. -# - -## Thread support for Nim. -## -## **Note**: This is part of the system module. Do not import it directly. -## To activate thread support you need to compile -## with the `--threads:on`:option: command line switch. -## -## Nim's memory model for threads is quite different from other common -## programming languages (C, Pascal): Each thread has its own -## (garbage collected) heap and sharing of memory is restricted. This helps -## to prevent race conditions and improves efficiency. See `the manual for -## details of this memory model `_. -## -## Examples -## ======== -## -## .. code-block:: Nim -## -## import std/locks -## -## var -## thr: array[0..4, Thread[tuple[a,b: int]]] -## L: Lock -## -## proc threadFunc(interval: tuple[a,b: int]) {.thread.} = -## for i in interval.a..interval.b: -## acquire(L) # lock stdout -## echo i -## release(L) -## -## initLock(L) -## -## for i in 0..high(thr): -## createThread(thr[i], threadFunc, (i*10, i*10+5)) -## joinThreads(thr) -## -## deinitLock(L) - -when not declared(ThisIsSystem): - {.error: "You must not import this module explicitly".} - -when defined(nimPreviewSlimSystem): - import std/assertions - -const - hasAllocStack = defined(zephyr) # maybe freertos too? - -when defined(gcDestructors): - proc allocThreadStorage(size: int): pointer = - result = c_malloc(csize_t size) - zeroMem(result, size) - - proc deallocThreadStorage(p: pointer) = c_free(p) -else: - template allocThreadStorage(size: untyped): untyped = allocShared0(size) - template deallocThreadStorage(p: pointer) = deallocShared(p) - -when hasAllocStack or defined(zephyr) or defined(freertos): - const - nimThreadStackSize {.intdefine.} = 8192 - nimThreadStackGuard {.intdefine.} = 128 - - StackGuardSize = nimThreadStackGuard - ThreadStackSize = nimThreadStackSize - nimThreadStackGuard -else: - const - StackGuardSize = 4096 - ThreadStackMask = - when defined(genode): - 1024*64*sizeof(int)-1 - else: - 1024*256*sizeof(int)-1 - - ThreadStackSize = ThreadStackMask+1 - StackGuardSize - -#const globalsSlot = ThreadVarSlot(0) -#sysAssert checkSlot.int == globalsSlot.int - -# Zephyr doesn't include this properly without some help -when defined(zephyr): - {.emit: """/*INCLUDESECTION*/ - #include - """.} - -# create for the main thread. Note: do not insert this data into the list -# of all threads; it's not to be stopped etc. -when not defined(useNimRtl): - #when not defined(createNimRtl): initStackBottom() - when declared(initGC): - initGC() - when not emulatedThreadVars: - type ThreadType {.pure.} = enum - None = 0, - NimThread = 1, - ForeignThread = 2 - var - threadType {.rtlThreadVar.}: ThreadType - - threadType = ThreadType.NimThread - -# We jump through some hops here to ensure that Nim thread procs can have -# the Nim calling convention. This is needed because thread procs are -# ``stdcall`` on Windows and ``noconv`` on UNIX. Alternative would be to just -# use ``stdcall`` since it is mapped to ``noconv`` on UNIX anyway. - -type - Thread*[TArg] = object - core: PGcThread - sys: SysThread - when TArg is void: - dataFn: proc () {.nimcall, gcsafe.} - else: - dataFn: proc (m: TArg) {.nimcall, gcsafe.} - data: TArg - when hasAllocStack: - rawStack: pointer - -proc `=copy`*[TArg](x: var Thread[TArg], y: Thread[TArg]) {.error.} - -var - threadDestructionHandlers {.rtlThreadVar.}: seq[proc () {.closure, gcsafe, raises: [].}] - -proc onThreadDestruction*(handler: proc () {.closure, gcsafe, raises: [].}) = - ## Registers a *thread local* handler that is called at the thread's - ## destruction. - ## - ## A thread is destructed when the `.thread` proc returns - ## normally or when it raises an exception. Note that unhandled exceptions - ## in a thread nevertheless cause the whole process to die. - threadDestructionHandlers.add handler - -template afterThreadRuns() = - for i in countdown(threadDestructionHandlers.len-1, 0): - threadDestructionHandlers[i]() - -when not defined(boehmgc) and not hasSharedHeap and not defined(gogc) and not defined(gcRegions): - proc deallocOsPages() {.rtl, raises: [].} - -proc threadTrouble() {.raises: [], gcsafe.} - ## defined in system/excpt.nim - -when defined(boehmgc): - type GCStackBaseProc = proc(sb: pointer, t: pointer) {.noconv.} - proc boehmGC_call_with_stack_base(sbp: GCStackBaseProc, p: pointer) - {.importc: "GC_call_with_stack_base", boehmGC.} - proc boehmGC_register_my_thread(sb: pointer) - {.importc: "GC_register_my_thread", boehmGC.} - proc boehmGC_unregister_my_thread() - {.importc: "GC_unregister_my_thread", boehmGC.} - - proc threadProcWrapDispatch[TArg](sb: pointer, thrd: pointer) {.noconv, raises: [].} = - boehmGC_register_my_thread(sb) - try: - let thrd = cast[ptr Thread[TArg]](thrd) - when TArg is void: - thrd.dataFn() - else: - thrd.dataFn(thrd.data) - except: - threadTrouble() - finally: - afterThreadRuns() - boehmGC_unregister_my_thread() -else: - proc threadProcWrapDispatch[TArg](thrd: ptr Thread[TArg]) {.raises: [].} = - try: - when TArg is void: - thrd.dataFn() - else: - when defined(nimV2): - thrd.dataFn(thrd.data) - else: - var x: TArg - deepCopy(x, thrd.data) - thrd.dataFn(x) - except: - threadTrouble() - finally: - afterThreadRuns() - when hasAllocStack: - deallocThreadStorage(thrd.rawStack) - -proc threadProcWrapStackFrame[TArg](thrd: ptr Thread[TArg]) {.raises: [].} = - when defined(boehmgc): - boehmGC_call_with_stack_base(threadProcWrapDispatch[TArg], thrd) - elif not defined(nogc) and not defined(gogc) and not defined(gcRegions) and not usesDestructors: - var p {.volatile.}: pointer - # init the GC for refc/markandsweep - nimGC_setStackBottom(addr(p)) - initGC() - when declared(threadType): - threadType = ThreadType.NimThread - threadProcWrapDispatch[TArg](thrd) - when declared(deallocOsPages): deallocOsPages() - else: - threadProcWrapDispatch(thrd) - -template threadProcWrapperBody(closure: untyped): untyped = - var thrd = cast[ptr Thread[TArg]](closure) - var core = thrd.core - when declared(globalsSlot): threadVarSetValue(globalsSlot, thrd.core) - threadProcWrapStackFrame(thrd) - # Since an unhandled exception terminates the whole process (!), there is - # no need for a ``try finally`` here, nor would it be correct: The current - # exception is tried to be re-raised by the code-gen after the ``finally``! - # However this is doomed to fail, because we already unmapped every heap - # page! - - # mark as not running anymore: - thrd.core = nil - thrd.dataFn = nil - deallocThreadStorage(cast[pointer](core)) - -{.push stack_trace:off.} -when defined(windows): - proc threadProcWrapper[TArg](closure: pointer): int32 {.stdcall.} = - threadProcWrapperBody(closure) - # implicitly return 0 -elif defined(genode): - proc threadProcWrapper[TArg](closure: pointer) {.noconv.} = - threadProcWrapperBody(closure) -else: - proc threadProcWrapper[TArg](closure: pointer): pointer {.noconv.} = - threadProcWrapperBody(closure) -{.pop.} - -proc running*[TArg](t: Thread[TArg]): bool {.inline.} = - ## Returns true if `t` is running. - result = t.dataFn != nil - -proc handle*[TArg](t: Thread[TArg]): SysThread {.inline.} = - ## Returns the thread handle of `t`. - result = t.sys - -when hostOS == "windows": - const MAXIMUM_WAIT_OBJECTS = 64 - - proc joinThread*[TArg](t: Thread[TArg]) {.inline.} = - ## Waits for the thread `t` to finish. - discard waitForSingleObject(t.sys, -1'i32) - - proc joinThreads*[TArg](t: varargs[Thread[TArg]]) = - ## Waits for every thread in `t` to finish. - var a: array[MAXIMUM_WAIT_OBJECTS, SysThread] - var k = 0 - while k < len(t): - var count = min(len(t) - k, MAXIMUM_WAIT_OBJECTS) - for i in 0..(count - 1): a[i] = t[i + k].sys - discard waitForMultipleObjects(int32(count), - cast[ptr SysThread](addr(a)), 1, -1) - inc(k, MAXIMUM_WAIT_OBJECTS) - -elif defined(genode): - proc joinThread*[TArg](t: Thread[TArg]) {.importcpp.} - ## Waits for the thread `t` to finish. - - proc joinThreads*[TArg](t: varargs[Thread[TArg]]) = - ## Waits for every thread in `t` to finish. - for i in 0..t.high: joinThread(t[i]) - -else: - proc joinThread*[TArg](t: Thread[TArg]) {.inline.} = - ## Waits for the thread `t` to finish. - discard pthread_join(t.sys, nil) - - proc joinThreads*[TArg](t: varargs[Thread[TArg]]) = - ## Waits for every thread in `t` to finish. - for i in 0..t.high: joinThread(t[i]) - -when false: - # XXX a thread should really release its heap here somehow: - proc destroyThread*[TArg](t: var Thread[TArg]) = - ## Forces the thread `t` to terminate. This is potentially dangerous if - ## you don't have full control over `t` and its acquired resources. - when hostOS == "windows": - discard TerminateThread(t.sys, 1'i32) - else: - discard pthread_cancel(t.sys) - when declared(registerThread): unregisterThread(addr(t)) - t.dataFn = nil - ## if thread `t` already exited, `t.core` will be `null`. - if not isNil(t.core): - deallocThreadStorage(t.core) - t.core = nil - -when hostOS == "windows": - proc createThread*[TArg](t: var Thread[TArg], - tp: proc (arg: TArg) {.thread, nimcall.}, - param: TArg) = - ## Creates a new thread `t` and starts its execution. - ## - ## Entry point is the proc `tp`. - ## `param` is passed to `tp`. `TArg` can be `void` if you - ## don't need to pass any data to the thread. - t.core = cast[PGcThread](allocThreadStorage(sizeof(GcThread))) - - when TArg isnot void: t.data = param - t.dataFn = tp - when hasSharedHeap: t.core.stackSize = ThreadStackSize - var dummyThreadId: int32 - t.sys = createThread(nil, ThreadStackSize, threadProcWrapper[TArg], - addr(t), 0'i32, dummyThreadId) - if t.sys <= 0: - raise newException(ResourceExhaustedError, "cannot create thread") - - proc pinToCpu*[Arg](t: var Thread[Arg]; cpu: Natural) = - ## Pins a thread to a `CPU`:idx:. - ## - ## In other words sets a thread's `affinity`:idx:. - ## If you don't know what this means, you shouldn't use this proc. - setThreadAffinityMask(t.sys, uint(1 shl cpu)) - -elif defined(genode): - var affinityOffset: cuint = 1 - ## CPU affinity offset for next thread, safe to roll-over. - - proc createThread*[TArg](t: var Thread[TArg], - tp: proc (arg: TArg) {.thread, nimcall.}, - param: TArg) = - t.core = cast[PGcThread](allocThreadStorage(sizeof(GcThread))) - - when TArg isnot void: t.data = param - t.dataFn = tp - when hasSharedHeap: t.stackSize = ThreadStackSize - t.sys.initThread( - runtimeEnv, - ThreadStackSize.culonglong, - threadProcWrapper[TArg], addr(t), affinityOffset) - inc affinityOffset - - proc pinToCpu*[Arg](t: var Thread[Arg]; cpu: Natural) = - {.hint: "cannot change Genode thread CPU affinity after initialization".} - discard - -else: - proc createThread*[TArg](t: var Thread[TArg], - tp: proc (arg: TArg) {.thread, nimcall.}, - param: TArg) = - ## Creates a new thread `t` and starts its execution. - ## - ## Entry point is the proc `tp`. `param` is passed to `tp`. - ## `TArg` can be `void` if you - ## don't need to pass any data to the thread. - t.core = cast[PGcThread](allocThreadStorage(sizeof(GcThread))) - - when TArg isnot void: t.data = param - t.dataFn = tp - when hasSharedHeap: t.core.stackSize = ThreadStackSize - var a {.noinit.}: Pthread_attr - doAssert pthread_attr_init(a) == 0 - when hasAllocStack: - var - rawstk = allocThreadStorage(ThreadStackSize + StackGuardSize) - stk = cast[pointer](cast[uint](rawstk) + StackGuardSize) - let setstacksizeResult = pthread_attr_setstack(addr a, stk, ThreadStackSize) - t.rawStack = rawstk - else: - let setstacksizeResult = pthread_attr_setstacksize(a, ThreadStackSize) - - when not defined(ios): - # This fails on iOS - doAssert(setstacksizeResult == 0) - if pthread_create(t.sys, a, threadProcWrapper[TArg], addr(t)) != 0: - raise newException(ResourceExhaustedError, "cannot create thread") - doAssert pthread_attr_destroy(a) == 0 - - proc pinToCpu*[Arg](t: var Thread[Arg]; cpu: Natural) = - ## Pins a thread to a `CPU`:idx:. - ## - ## In other words sets a thread's `affinity`:idx:. - ## If you don't know what this means, you shouldn't use this proc. - when not defined(macosx): - var s {.noinit.}: CpuSet - cpusetZero(s) - cpusetIncl(cpu.cint, s) - setAffinity(t.sys, csize_t(sizeof(s)), s) - -proc createThread*(t: var Thread[void], tp: proc () {.thread, nimcall.}) = - createThread[void](t, tp) - -when not defined(gcOrc): - include threadids diff --git a/tools/kochdocs.nim b/tools/kochdocs.nim index ee9c29499..c0bdb0503 100644 --- a/tools/kochdocs.nim +++ b/tools/kochdocs.nim @@ -138,7 +138,6 @@ mm.md """.splitWhitespace().mapIt("doc" / it) doc0 = """ -lib/system/threads.nim lib/system/channels_builtin.nim """.splitWhitespace() # ran by `nim doc0` instead of `nim doc` -- cgit 1.4.1-2-gfad0