minzkn/thread
|
문서의 수정
목차 ¶1.2. 개요 ¶
1.4.1. 컴파일러의 최적화에 따른 비의도적 결과 ¶최적화에 의한 Thread code의 의도되지 않은 현상이 발생할수있습니다.
일단 그러한 부분에 대해서는 불가피하게 Assembly로 구현하여 저는 해결방법을 모색하였습니다.
그러나 함수 자체의 Frame의 최적화가 이뤄질때면 방법이 난해합니다. 이에 대한 고민은 반드시 필요합니다.
이에 대한 충분한 고려가 되지 않으면 오직 자신의 PC에서만 실행되는 기형적인 Thread code가 될것입니다.
1.4.2. Stack의 가변적 검출 ¶각 Thread는 고유의 Stack을 소유하게 설계됩니다.
하지만 그것이 고정적이라면 지역변수로 커다란 배열을 선언하여 사용하는데 부담이 아닐수 없겠습니다.
때문에 Stack을 가변적으로 늘려줄수 있는 방법이 요구됩니다.
1.4.3. How to swap ¶이것은 Stack이 Swap되는것에 대한 고민이 필요하다는데 있습니다.
무턱대고 모두 Swap이 알아서 되면 좋겠지만 잘 생각해보시면 Thread의 Stack은 과연 Swap이 필요한가를 생각해봐야 합니다. 잘못하면 오히려 부작용이 만만치 않다는데 중점을 두고 생각해볼 문제임이 분명합니다.
필자의 소견은 "Stack은 Heap과 다르다" 입니다. 결코 Swap이 되지 않고 아예 고려되지 않는것이 좋다는 생각입니다.
여러분들의 생각은 어떠신지요? 각자 만들어보고 성능의 평가를 할수 있는 기회가 언젠가 주어졌으면 좋겠군요.
1.6. 예제 ¶아래의 예제 소스는 실제 필자가 만들고 있는 운영체제의 일부를 발췌한것입니다. 발췌하면서 약간의 불필요한 부분과 나름대로의 주요 소스는 발췌하지 않았습니다. 차후에는 완전한 Thread예제를 공개하도록 노력하겠습니다. 하지만 현시점에서는 이 이상의 소스는 공개하지 않을 방침입니다.
/*
Copyright (c) Information Equipment co.,LTD. Code by JaeHyuk Cho <mailto:minzkn@infoeq.co.kr> KOREA MZ Local Thread library v0.0.1b - Simple is best ! */ #if !defined(DEF_SOURCE_thread_c) #define DEF_SOURCE_thread_c "thread.c" typedef struct ts_STACK { void *Stack; int StackSize, StackPointer; }t_STACK; typedef struct ts_THREAD_TASK { struct ts_THREAD_TASK *Next; t_STACK *Stack; unsigned int TaskID, ESP, Tick, Active; void * (*Entry)(void *, void *); void *Argument; }t_THREAD_TASK; typedef struct ts_THREAD { t_THREAD_TASK *Task; t_THREAD_TASK *CurrentTask; unsigned int TaskCount, MakeID; }t_THREAD; t_THREAD *ML_CreateTHREAD(void); t_THREAD *ML_DestroyTHREAD(t_THREAD *s_THREAD); t_THREAD *ML_AddTHREAD(t_THREAD *s_THREAD, void * (*s_ThreadFunction)(void *, void *), void *s_Argument, int s_StackSize); t_THREAD *ML_RunTHREAD(t_THREAD *s_THREAD); int ML_SleepTHREAD(t_THREAD *s_THREAD); t_STACK *ML_CreateSTACK(int s_StackSize); t_STACK *ML_DestroySTACK(t_STACK *s_STACK); int ML_PushSTACK(t_STACK *s_STACK, int s_Value); int ML_PopSTACK(t_STACK *s_STACK, int *s_Value); int ML_SetSTACK(t_STACK *s_STACK, int s_StackPointer); void __ML_ReturnTHREAD__(void); static void *__ML_ManagerTHREAD__(void *s_ThreadHandle, void *s_Argument) { static t_THREAD *sg_THREAD = (t_THREAD *)0; if(sg_THREAD != (t_THREAD *)s_ThreadHandle)sg_THREAD = (t_THREAD *)s_ThreadHandle; ML_SleepTHREAD((t_THREAD *)s_ThreadHandle); if(((t_THREAD *)s_ThreadHandle)->Task->Active == 0)return(s_Argument); t_inline_asm( "__ML_ReturnTHREAD__:\n\t" "pushl $__ML_ReturnTHREAD__\n\t" /* Retry push return address */ ); t_inline_asm( "\n\t" : "=a"(((t_THREAD *)s_ThreadHandle)->CurrentTask->Argument) ); ((t_THREAD *)s_ThreadHandle)->CurrentTask->Active = 0; ML_SleepTHREAD((t_THREAD *)s_ThreadHandle); return(s_Argument); } t_THREAD *ML_CreateTHREAD(void) { t_THREAD *s_Return; s_Return = (t_THREAD *)ML_Alloc(sizeof(t_THREAD)); if(s_Return) { s_Return->Task = s_Return->CurrentTask = (t_THREAD_TASK *)0; s_Return->TaskCount = s_Return->MakeID = 0u; s_Return = ML_AddTHREAD(s_Return, __ML_ManagerTHREAD__, (void *)0, (4 << 10)); } return(s_Return); } t_THREAD *ML_DestroyTHREAD(t_THREAD *s_THREAD) { t_THREAD_TASK *s_THREAD_TASK; if(s_THREAD) { while(s_THREAD->Task && s_THREAD->TaskCount--) { s_THREAD_TASK = s_THREAD->Task; s_THREAD->Task = s_THREAD->Task->Next; if(s_THREAD_TASK->Stack)(void)ML_DestroySTACK(s_THREAD_TASK->Stack); (void)ML_Free(s_THREAD_TASK); } (void)ML_Free(s_THREAD); s_THREAD = (t_THREAD *)0; } return(s_THREAD); } t_THREAD *ML_AddTHREAD(t_THREAD *s_THREAD, void * (*s_ThreadFunction)(void *, void *), void *s_Argument, int s_StackSize) { t_THREAD_TASK *s_THREAD_TASK; if(s_THREAD == (t_THREAD *)0)s_THREAD = ML_CreateTHREAD(); if(s_THREAD) { if(s_THREAD->Task) { s_THREAD_TASK = s_THREAD->Task; while(s_THREAD_TASK->Next && s_THREAD_TASK->Next != s_THREAD->Task)s_THREAD_TASK = s_THREAD_TASK->Next; s_THREAD_TASK->Next = (t_THREAD_TASK *)ML_Alloc(sizeof(t_THREAD_TASK)); s_THREAD_TASK = s_THREAD_TASK->Next; if(s_THREAD->CurrentTask == (t_THREAD_TASK *)0)s_THREAD->CurrentTask = s_THREAD->Task; } else s_THREAD->Task = s_THREAD->CurrentTask = s_THREAD_TASK = (t_THREAD_TASK *)ML_Alloc(sizeof(t_THREAD_TASK)); if(s_THREAD_TASK) { if(s_StackSize < ( 4 << 10 ))s_StackSize = ( 4 << 10 ); s_THREAD_TASK->Next = s_THREAD->Task; s_THREAD_TASK->Stack = ML_CreateSTACK(s_StackSize); s_THREAD_TASK->TaskID = (s_THREAD->MakeID++); s_THREAD_TASK->Tick = 0; s_THREAD_TASK->Active = 1; s_THREAD_TASK->Entry = s_ThreadFunction; s_THREAD_TASK->Argument = s_Argument; s_THREAD_TASK->ESP = (unsigned int)s_THREAD_TASK->Stack->Stack + s_THREAD_TASK->Stack->StackPointer; s_THREAD->TaskCount++; } } return(s_THREAD); } t_THREAD *ML_RunTHREAD(t_THREAD *s_THREAD) { struct { unsigned int eax, ebx, ecx, edx, esi, edi, ebp, esp, flags; }s_Register; t_THREAD_TASK *s_THREAD_TASK; unsigned int s_RegisterAddress, s_TempEBX; if(s_THREAD) { if(s_THREAD->Task) { s_RegisterAddress = (unsigned int)(&s_Register); t_inline_asm( "\n\t" "movl %%ebx, %1\n\t" "movl %0, %%ebx\n\t" "movl %%eax, 0(%%ebx)\n\t" "movl %1, %%eax\n\t" "movl %%eax, 4(%%ebx)\n\t" "movl 0(%%ebx), %%eax\n\t" "movl %%ecx, 8(%%ebx)\n\t" "movl %%edx, 12(%%ebx)\n\t" "movl %%esi, 16(%%ebx)\n\t" "movl %%edi, 20(%%ebx)\n\t" "movl %%ebp, 24(%%ebx)\n\t" "movl %%esp, 28(%%ebx)\n\t" "pushfl\n\t" "popl 32(%%ebx)\n\t" "movl 4(%%ebx), %%ebx\n\t" "\n\t" : : "m"(s_RegisterAddress), "m"(s_TempEBX) ); s_THREAD_TASK = s_THREAD->Task; ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.flags); ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.esp); ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.ebp); ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.edi); ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.esi); ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.edx); ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.ecx); ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.ebx); ML_PushSTACK(s_THREAD_TASK->Stack, (int)s_THREAD_TASK->Argument); ML_PushSTACK(s_THREAD_TASK->Stack, (int)s_THREAD); s_THREAD_TASK->ESP = (unsigned int)s_THREAD_TASK->Stack->Stack + s_THREAD_TASK->Stack->StackPointer; s_THREAD_TASK = s_THREAD_TASK->Next; while(s_THREAD_TASK && s_THREAD_TASK != s_THREAD->Task) { ML_PushSTACK(s_THREAD_TASK->Stack, (int)s_THREAD_TASK->Argument); ML_PushSTACK(s_THREAD_TASK->Stack, (int)s_THREAD); ML_PushSTACK(s_THREAD_TASK->Stack, (int)__ML_ReturnTHREAD__); ML_PushSTACK(s_THREAD_TASK->Stack, (int)s_THREAD_TASK->Argument); ML_PushSTACK(s_THREAD_TASK->Stack, (int)s_THREAD); ML_PushSTACK(s_THREAD_TASK->Stack, (int)__ML_ReturnTHREAD__); ML_PushSTACK(s_THREAD_TASK->Stack, (int)s_THREAD_TASK->Entry); /* First swich entry */ ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.flags); ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.ebp); ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.edi); ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.esi); ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.edx); ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.ecx); ML_PushSTACK(s_THREAD_TASK->Stack, s_Register.ebx); s_THREAD_TASK->ESP = (unsigned int)s_THREAD_TASK->Stack->Stack + s_THREAD_TASK->Stack->StackPointer; s_THREAD_TASK = s_THREAD_TASK->Next; } t_inline_asm( "\n\t" "movl %1, %%ecx\n\t" "movl %0, %%ebp\n\t" "movl %%ebp, %%esp\n\t" "call *%%ecx\n\t" "addl $4 + 4, %%esp\n\t" "popl %%ebx\n\t" "popl %%ecx\n\t" "popl %%edx\n\t" "popl %%esi\n\t" "popl %%edi\n\t" "popl %%ebp\n\t" "popl %%eax\n\t" /* Change stack (x86) */ "popfl\n\t" "movl %%eax, %%esp\n\t" "\n\t" : : "m"(s_THREAD->Task->ESP), "m"(s_THREAD->Task->Entry) ); } } return(s_THREAD); } int ML_SleepTHREAD(t_THREAD *s_THREAD) { s_THREAD->CurrentTask->Tick++; t_inline_asm( "\n\t" "movl %%esp, %%eax\n\t" "subl $28, %%eax\n\t" "\n\t" : "=a"(s_THREAD->CurrentTask->ESP) ); do { s_THREAD->CurrentTask = s_THREAD->CurrentTask->Next; if(s_THREAD->CurrentTask == s_THREAD->Task) { if(s_THREAD->Task->Active == 1) { s_THREAD->CurrentTask->Active = 0; continue; } else break; } }while(s_THREAD->CurrentTask->Active == 0); if(s_THREAD->CurrentTask != s_THREAD->Task)s_THREAD->Task->Active = 1; t_inline_asm( "\n\t" "pushfl\n\t" "pushl %%ebp\n\t" "pushl %%edi\n\t" "pushl %%esi\n\t" "pushl %%edx\n\t" "pushl %%ecx\n\t" "pushl %%ebx\n\t" "movl %0, %%esp\n\t" "popl %%ebx\n\t" "popl %%ecx\n\t" "popl %%edx\n\t" "popl %%esi\n\t" "popl %%edi\n\t" "popl %%ebp\n\t" "popfl\n\t" "\n\t" : : "a"(s_THREAD->CurrentTask->ESP) ); return(1); } t_STACK *ML_CreateSTACK(int s_StackSize) { t_STACK *s_Return; if(s_StackSize < (4 << 10))s_StackSize = (4 << 10); s_Return = (t_STACK *)ML_Alloc(sizeof(t_STACK)); if(s_Return) { s_Return->Stack = (void *)ML_Alloc(s_StackSize); s_Return->StackSize = s_Return->StackPointer = s_StackSize; } return(s_Return); } t_STACK *ML_DestroySTACK(t_STACK *s_STACK) { if(s_STACK) { if(s_STACK->Stack && s_STACK->StackSize > 0)(void)ML_Free(s_STACK->Stack); (void)ML_Free(s_STACK); s_STACK = (t_STACK *)0; } return(s_STACK); } int ML_PushSTACK(t_STACK *s_STACK, int s_Value) { if(s_STACK) { if(s_STACK->Stack && s_STACK->StackSize >= sizeof(s_Value) && s_STACK->StackPointer >= sizeof(s_Value)) { s_STACK->StackPointer -= sizeof(s_Value); ML_PokeDoubleWord(s_STACK->Stack, s_STACK->StackPointer, s_Value); return(s_STACK->StackPointer); } } return(0); } int ML_PopSTACK(t_STACK *s_STACK, int *s_Value) { int s_Return = (-1); if(s_STACK) { if(s_STACK->Stack && s_STACK->StackSize >= sizeof(int) && s_STACK->StackPointer <= (s_STACK->StackSize - sizeof(int))) { s_Return = ML_PeekDoubleWord(s_STACK->Stack, s_STACK->StackPointer); s_STACK->StackPointer += sizeof(int); } } if(s_Value)*(s_Value) = s_Return; return(s_Return); } int ML_SetSTACK(t_STACK *s_STACK, int s_StackPointer) { if(s_STACK) { s_STACK->StackPointer = s_StackPointer; return(s_STACK->StackPointer); } return(0); } #endif /* End of source */ 1.8. 문서를 마치면서 ¶이 문서에서 틀린 부분이나 고쳐야 할부분이 있으면 꼭 알려주십시요. 이글의 최근 문서는 http://doc.kldp.org 에서 만나보실수 있습니다.
필자의 홈페이지:
![[http]](/imgs/http.png) http://minzkn.pe.ky
|
It is better to wear out than to rust out.    
|
