X-Git-Url: https://git.ucc.asn.au/?a=blobdiff_plain;f=Kernel%2Farch%2Fx86%2Fproc.c;h=82a3f40908fb821f08d893ac86cb6d94e10a4cec;hb=a2495c6ea4f4cab16b5d339ae511428e92e89e73;hp=45c5a56126ce89fb453b42e54ff9369d3ace3216;hpb=9f407c493c33928e0f19b834699d9694036ca42e;p=tpg%2Facess2.git diff --git a/Kernel/arch/x86/proc.c b/Kernel/arch/x86/proc.c index 45c5a561..82a3f409 100644 --- a/Kernel/arch/x86/proc.c +++ b/Kernel/arch/x86/proc.c @@ -3,6 +3,7 @@ * proc.c */ #include +#include #include #include #include @@ -10,75 +11,103 @@ #if USE_MP # include #endif +#include +#include // === FLAGS === #define DEBUG_TRACE_SWITCH 0 +#define DEBUG_DISABLE_DOUBLEFAULT 1 +#define DEBUG_VERY_SLOW_PERIOD 0 // === CONSTANTS === -#define SWITCH_MAGIC 0xFFFACE55 // There is no code in this area // Base is 1193182 -#define TIMER_DIVISOR 11931 //~100Hz +#define TIMER_BASE 1193182 +#if DEBUG_VERY_SLOW_PERIOD +# define TIMER_DIVISOR 1193 //~10Hz switch, with 10 quantum = 1s per thread +#else +# define TIMER_DIVISOR 11932 //~100Hz +#endif + +// === TYPES === +typedef struct sCPU +{ + Uint8 APICID; + Uint8 State; // 0: Unavaliable, 1: Idle, 2: Active + Uint16 Resvd; + tThread *Current; +} tCPU; // === IMPORTS === extern tGDT gGDT[]; extern tIDT gIDT[]; -extern void APStartup(); // 16-bit AP startup code -extern Uint GetEIP(); // start.asm +extern void APWait(void); // 16-bit AP pause code +extern void APStartup(void); // 16-bit AP startup code +extern Uint GetEIP(void); // start.asm +extern Uint GetEIP_Sched(void); // proc.asm +extern void NewTaskHeader(tThread *Thread, void *Fcn, int nArgs, ...); // Actually takes cdecl args +extern Uint Proc_CloneInt(Uint *ESP, Uint32 *CR3, int bNoUserClone); extern Uint32 gaInitPageDir[1024]; // start.asm -extern void Kernel_Stack_Top; -extern tSpinlock glThreadListLock; +extern char Kernel_Stack_Top[]; extern int giNumCPUs; extern int giNextTID; -extern int giTotalTickets; -extern int giNumActiveThreads; extern tThread gThreadZero; -extern tThread *gActiveThreads; -extern tThread *gSleepingThreads; -extern tThread *gDeleteThreads; -extern tThread *Threads_GetNextToRun(int CPU); -extern void Threads_Dump(); -extern tThread *Threads_CloneTCB(Uint *Err, Uint Flags); -extern void Isr8(); // Double Fault -extern void Proc_ReturnToUser(); +extern tProcess gProcessZero; +extern void Isr8(void); // Double Fault +extern void Proc_ReturnToUser(tVAddr Handler, Uint Argument, tVAddr KernelStack); +extern char scheduler_return[]; // Return address in SchedulerBase +extern char IRQCommon[]; // Common IRQ handler code +extern char IRQCommon_handled[]; // IRQCommon call return location +extern char GetEIP_Sched_ret[]; // GetEIP call return location +extern void SwitchTasks(Uint NewSP, Uint *OldSP, Uint NewIP, Uint *OldIO, Uint CR3); +extern void Proc_InitialiseSSE(void); +extern void Proc_SaveSSE(Uint DestPtr); +extern void Proc_DisableSSE(void); // === PROTOTYPES === -void ArchThreads_Init(); +//void ArchThreads_Init(void); #if USE_MP void MP_StartAP(int CPU); +void MP_SendIPIVector(int CPU, Uint8 Vector); void MP_SendIPI(Uint8 APICID, int Vector, int DeliveryMode); #endif -void Proc_Start(); -tThread *Proc_GetCurThread(); -void Proc_ChangeStack(); - int Proc_Clone(Uint *Err, Uint Flags); -void Proc_StartProcess(Uint16 SS, Uint Stack, Uint Flags, Uint16 CS, Uint IP); -void Proc_CallFaultHandler(tThread *Thread); -void Proc_Scheduler(); +void Proc_IdleThread(void *Ptr); +//void Proc_Start(void); +//tThread *Proc_GetCurThread(void); +void Proc_ChangeStack(void); +// int Proc_NewKThread(void (*Fcn)(void*), void *Data); +// int Proc_Clone(Uint *Err, Uint Flags); +Uint Proc_MakeUserStack(void); +//void Proc_StartUser(Uint Entrypoint, Uint *Bases, int ArgC, char **ArgV, char **EnvP, int DataSize); +void Proc_StartProcess(Uint16 SS, Uint Stack, Uint Flags, Uint16 CS, Uint IP) NORETURN; + int Proc_Demote(Uint *Err, int Dest, tRegs *Regs); +//void Proc_CallFaultHandler(tThread *Thread); +//void Proc_DumpThreadCPUState(tThread *Thread); +void Proc_Scheduler(int CPU); // === GLOBALS === // --- Multiprocessing --- #if USE_MP volatile int giNumInitingCPUs = 0; tMPInfo *gMPFloatPtr = NULL; +volatile Uint32 giMP_TimerCount; // Start Count for Local APIC Timer tAPIC *gpMP_LocalAPIC = NULL; Uint8 gaAPIC_to_CPU[256] = {0}; -tCPU gaCPUs[MAX_CPUS]; -#else -tThread *gCurrentThread = NULL; -#endif -#if USE_PAE -Uint32 *gPML4s[4] = NULL; + int giProc_BootProcessorID = 0; +tTSS gaTSSs[MAX_CPUS]; // TSS Array #endif -tTSS *gTSSs = NULL; +tCPU gaCPUs[MAX_CPUS] = { + {.Current = &gThreadZero} + }; +tTSS *gTSSs = NULL; // Pointer to TSS array tTSS gTSS0 = {0}; // --- Error Recovery --- -char gaDoubleFaultStack[1024]; +char gaDoubleFaultStack[1024] __attribute__ ((section(".padata"))); tTSS gDoubleFault_TSS = { - .ESP0 = (Uint)&gaDoubleFaultStack[1023], + .ESP0 = (Uint)&gaDoubleFaultStack[1024], .SS0 = 0x10, .CR3 = (Uint)gaInitPageDir - KERNEL_BASE, .EIP = (Uint)Isr8, - .ESP = (Uint)&gaDoubleFaultStack[1023], + .ESP = (Uint)&gaDoubleFaultStack[1024], .CS = 0x08, .SS = 0x10, .DS = 0x10, .ES = 0x10, .FS = 0x10, .GS = 0x10, @@ -86,10 +115,10 @@ tTSS gDoubleFault_TSS = { // === CODE === /** - * \fn void ArchThreads_Init() + * \fn void ArchThreads_Init(void) * \brief Starts the process scheduler */ -void ArchThreads_Init() +void ArchThreads_Init(void) { Uint pos = 0; @@ -138,6 +167,7 @@ void ArchThreads_Init() { int i; tMPTable_Ent *ents; + #if DUMP_MP_TABLE Log("gMPFloatPtr = %p", gMPFloatPtr); Log("*gMPFloatPtr = {"); Log("\t.Sig = 0x%08x", gMPFloatPtr->Sig); @@ -151,8 +181,10 @@ void ArchThreads_Init() gMPFloatPtr->Features[4] ); Log("}"); - + #endif + mptable = (void*)( KERNEL_BASE|gMPFloatPtr->MPConfig ); + #if DUMP_MP_TABLE Log("mptable = %p", mptable); Log("*mptable = {"); Log("\t.Sig = 0x%08x", mptable->Sig); @@ -168,8 +200,9 @@ void ArchThreads_Init() Log("\t.ExtendedTableLen = 0x%04x", mptable->ExtendedTableLen); Log("\t.ExtendedTableChecksum = 0x%02x", mptable->ExtendedTableChecksum); Log("}"); + #endif - gpMP_LocalAPIC = (void*)MM_MapHWPage(mptable->LocalAPICMemMap, 1); + gpMP_LocalAPIC = (void*)MM_MapHWPages(mptable->LocalAPICMemMap, 1); ents = mptable->Entries; giNumCPUs = 0; @@ -181,13 +214,14 @@ void ArchThreads_Init() { case 0: // Processor entSize = 20; + #if DUMP_MP_TABLE Log("%i: Processor", i); Log("\t.APICID = %i", ents->Proc.APICID); Log("\t.APICVer = 0x%02x", ents->Proc.APICVer); Log("\t.CPUFlags = 0x%02x", ents->Proc.CPUFlags); Log("\t.CPUSignature = 0x%08x", ents->Proc.CPUSignature); Log("\t.FeatureFlags = 0x%08x", ents->Proc.FeatureFlags); - + #endif if( !(ents->Proc.CPUFlags & 1) ) { Log("DISABLED"); @@ -206,18 +240,19 @@ void ArchThreads_Init() gaCPUs[giNumCPUs].State = 0; giNumCPUs ++; - // Send IPI - if( !(ents->Proc.CPUFlags & 2) ) - { - MP_StartAP( giNumCPUs-1 ); + // Set BSP Variable + if( ents->Proc.CPUFlags & 2 ) { + giProc_BootProcessorID = giNumCPUs-1; } break; + + #if DUMP_MP_TABLE >= 2 case 1: // Bus entSize = 8; Log("%i: Bus", i); Log("\t.ID = %i", ents->Bus.ID); - Log("\t.TypeString = '%6c'", ents->Bus.TypeString); + Log("\t.TypeString = '%6C'", ents->Bus.TypeString); break; case 2: // I/O APIC entSize = 8; @@ -250,6 +285,7 @@ void ArchThreads_Init() default: Log("%i: Unknown (%i)", i, ents->Type); break; + #endif } ents = (void*)( (Uint)ents + entSize ); } @@ -258,11 +294,7 @@ void ArchThreads_Init() Warning("Too many CPUs detected (%i), only using %i of them", giNumCPUs, MAX_CPUS); giNumCPUs = MAX_CPUS; } - - while( giNumInitingCPUs ) - MM_FinishVirtualInit(); - - Panic("Uh oh... MP Table Parsing is unimplemented\n"); + gTSSs = gaTSSs; } else { Log("No MP Table was found, assuming uniprocessor\n"); @@ -272,32 +304,52 @@ void ArchThreads_Init() #else giNumCPUs = 1; gTSSs = &gTSS0; - MM_FinishVirtualInit(); #endif + #if !DEBUG_DISABLE_DOUBLEFAULT // Initialise Double Fault TSS - /* - gGDT[5].LimitLow = sizeof(tTSS); - gGDT[5].LimitHi = 0; - gGDT[5].Access = 0x89; // Type - gGDT[5].Flags = 0x4; - */ gGDT[5].BaseLow = (Uint)&gDoubleFault_TSS & 0xFFFF; gGDT[5].BaseMid = (Uint)&gDoubleFault_TSS >> 16; gGDT[5].BaseHi = (Uint)&gDoubleFault_TSS >> 24; - Log_Debug("Proc", "gIDT[8] = {OffsetLo:%04x, CS:%04x, Flags:%04x, OffsetHi:%04x}", - gIDT[8].OffsetLo, gIDT[8].CS, gIDT[8].Flags, gIDT[8].OffsetHi); + // Set double fault IDT to use the new TSS gIDT[8].OffsetLo = 0; gIDT[8].CS = 5<<3; gIDT[8].Flags = 0x8500; gIDT[8].OffsetHi = 0; - Log_Debug("Proc", "gIDT[8] = {OffsetLo:%04x, CS:%04x, Flags:%04x, OffsetHi:%04x}", - gIDT[8].OffsetLo, gIDT[8].CS, gIDT[8].Flags, gIDT[8].OffsetHi); + #endif + + // Set timer frequency + outb(0x43, 0x34); // Set Channel 0, Low/High, Rate Generator + outb(0x40, TIMER_DIVISOR&0xFF); // Low Byte of Divisor + outb(0x40, (TIMER_DIVISOR>>8)&0xFF); // High Byte - //__asm__ __volatile__ ("xchg %bx, %bx"); + Log_Debug("Proc", "PIT Frequency %i.%03i Hz", + TIMER_BASE/TIMER_DIVISOR, + ((Uint64)TIMER_BASE*1000/TIMER_DIVISOR)%1000 + ); #if USE_MP + // Get the count setting for APIC timer + Log("Determining APIC Count"); + __asm__ __volatile__ ("sti"); + while( giMP_TimerCount == 0 ) __asm__ __volatile__ ("hlt"); + __asm__ __volatile__ ("cli"); + Log("APIC Count %i", giMP_TimerCount); + { + Uint64 freq = giMP_TimerCount; + freq *= TIMER_BASE; + freq /= TIMER_DIVISOR; + if( (freq /= 1000) < 2*1000) + Log("Bus Frequency %i KHz", freq); + else if( (freq /= 1000) < 2*1000) + Log("Bus Frequency %i MHz", freq); + else if( (freq /= 1000) < 2*1000) + Log("Bus Frequency %i GHz", freq); + else + Log("Bus Frequency %i THz", freq); + } + // Initialise Normal TSS(s) for(pos=0;pos> 16; + gGDT[6+pos].BaseMid = ((Uint)(&gTSSs[pos]) >> 16) & 0xFFFF; gGDT[6+pos].BaseHi = ((Uint)(&gTSSs[pos])) >> 24; #if USE_MP } - for(pos=0;pos>8)&0xFF); // High Byte + gProcessZero.MemState.CR3 = (Uint)gaInitPageDir - KERNEL_BASE; // Create Per-Process Data Block - MM_Allocate(MM_PPD_CFG); + if( !MM_Allocate(MM_PPD_CFG) ) + { + Panic("OOM - No space for initial Per-Process Config"); + } + + // Initialise SSE support + Proc_InitialiseSSE(); // Change Stacks Proc_ChangeStack(); } #if USE_MP +/** + * \brief Start an AP + */ void MP_StartAP(int CPU) { - Log("Starting AP %i (APIC %i)", CPU, gaCPUs[CPU].APICID); + Log_Log("Proc", "Starting AP %i (APIC %i)", CPU, gaCPUs[CPU].APICID); + // Set location of AP startup code and mark for a warm restart - *(Uint16*)(KERNEL_BASE|0x467) = (Uint)&APStartup - (KERNEL_BASE|0xFFFF0); + *(Uint16*)(KERNEL_BASE|0x467) = (Uint)&APWait - (KERNEL_BASE|0xFFFF0); *(Uint16*)(KERNEL_BASE|0x469) = 0xFFFF; - outb(0x70, 0x0F); outb(0x71, 0x0A); // Warm Reset - MP_SendIPI(gaCPUs[CPU].APICID, 0, 5); + outb(0x70, 0x0F); outb(0x71, 0x0A); // Set warm reset flag + MP_SendIPI(gaCPUs[CPU].APICID, 0, 5); // Init IPI + + // Delay + inb(0x80); inb(0x80); inb(0x80); inb(0x80); + + // TODO: Use a better address, preferably registered with the MM + // - MM_AllocDMA mabye? + // Create a far jump + *(Uint8*)(KERNEL_BASE|0x11000) = 0xEA; // Far JMP + *(Uint16*)(KERNEL_BASE|0x11001) = (Uint)&APStartup - (KERNEL_BASE|0xFFFF0); // IP + *(Uint16*)(KERNEL_BASE|0x11003) = 0xFFFF; // CS + // Send a Startup-IPI to make the CPU execute at 0x11000 (which we + // just filled) + MP_SendIPI(gaCPUs[CPU].APICID, 0x11, 6); // StartupIPI + giNumInitingCPUs ++; } +void MP_SendIPIVector(int CPU, Uint8 Vector) +{ + MP_SendIPI(gaCPUs[CPU].APICID, Vector, 0); +} + +/** + * \brief Send an Inter-Processor Interrupt + * \param APICID Processor's Local APIC ID + * \param Vector Argument of some kind + * \param DeliveryMode Type of signal + * \note 3A 10.5 "APIC/Handling Local Interrupts" + */ void MP_SendIPI(Uint8 APICID, int Vector, int DeliveryMode) { - Uint32 addr = (Uint)gpMP_LocalAPIC + 0x300; Uint32 val; // Hi val = (Uint)APICID << 24; - Log("*%p = 0x%08x", addr+0x10, val); - *(Uint32*)(addr+0x10) = val; +// Log("%p = 0x%08x", &gpMP_LocalAPIC->ICR[1], val); + gpMP_LocalAPIC->ICR[1].Val = val; // Low (and send) val = ((DeliveryMode & 7) << 8) | (Vector & 0xFF); - Log("*%p = 0x%08x", addr, val); - *(Uint32*)addr = val; +// Log("%p = 0x%08x", &gpMP_LocalAPIC->ICR[0], val); + gpMP_LocalAPIC->ICR[0].Val = val; } #endif +void Proc_IdleThread(void *Ptr) +{ + tCPU *cpu = &gaCPUs[GetCPUNum()]; + cpu->Current->ThreadName = strdup("Idle Thread"); + Threads_SetPriority( cpu->Current, -1 ); // Never called randomly + cpu->Current->Quantum = 1; // 1 slice quantum + for(;;) { + __asm__ __volatile__ ("sti"); // Make sure interrupts are enabled + __asm__ __volatile__ ("hlt"); // Make sure interrupts are enabled + Proc_Reschedule(); + } +} + /** - * \fn void Proc_Start() + * \fn void Proc_Start(void) * \brief Start process scheduler */ -void Proc_Start() +void Proc_Start(void) { + int tid; + #if USE_MP + int i; + #endif + + #if USE_MP + // Start APs + for( i = 0; i < giNumCPUs; i ++ ) + { + if(i) gaCPUs[i].Current = NULL; + + // Create Idle Task + tid = Proc_NewKThread(Proc_IdleThread, &gaCPUs[i]); + + // Start the AP + if( i != giProc_BootProcessorID ) { + MP_StartAP( i ); + } + } + + // BSP still should run the current task + gaCPUs[0].Current = &gThreadZero; + + // Start interrupts and wait for APs to come up + Log_Debug("Proc", "Waiting for APs to come up"); + __asm__ __volatile__ ("sti"); + while( giNumInitingCPUs ) __asm__ __volatile__ ("hlt"); + #else + // Create Idle Task + tid = Proc_NewKThread(Proc_IdleThread, &gaCPUs[0]); +// gaCPUs[0].IdleThread = Threads_GetThread(tid); + + // Set current task + gaCPUs[0].Current = &gThreadZero; + // Start Interrupts (and hence scheduler) __asm__ __volatile__("sti"); + #endif + MM_FinishVirtualInit(); } /** - * \fn tThread *Proc_GetCurThread() + * \fn tThread *Proc_GetCurThread(void) * \brief Gets the current thread */ -tThread *Proc_GetCurThread() +tThread *Proc_GetCurThread(void) { #if USE_MP - return gaCPUs[ gaAPIC_to_CPU[gpMP_LocalAPIC->ID.Val&0xFF] ].Current; + return gaCPUs[ GetCPUNum() ].Current; #else - return gCurrentThread; + return gaCPUs[ 0 ].Current; #endif } /** - * \fn void Proc_ChangeStack() + * \fn void Proc_ChangeStack(void) * \brief Swaps the current stack for a new one (in the proper stack reigon) */ -void Proc_ChangeStack() +void Proc_ChangeStack(void) { Uint esp, ebp; Uint tmpEbp, oldEsp; @@ -446,129 +570,139 @@ void Proc_ChangeStack() __asm__ __volatile__ ("mov %0, %%ebp"::"r"(ebp)); } +void Proc_ClearProcess(tProcess *Process) +{ + MM_ClearSpace(Process->MemState.CR3); +} + +void Proc_ClearThread(tThread *Thread) +{ + if(Thread->SavedState.SSE) { + free(Thread->SavedState.SSE); + Thread->SavedState.SSE = NULL; + } +} + +tTID Proc_NewKThread(void (*Fcn)(void*), void *Data) +{ + Uint esp; + tThread *newThread, *cur; + + cur = Proc_GetCurThread(); + newThread = Threads_CloneTCB(0); + if(!newThread) return -1; + + // Create new KStack + newThread->KernelStack = MM_NewKStack(); + // Check for errors + if(newThread->KernelStack == 0) { + free(newThread); + return -1; + } + + esp = newThread->KernelStack; + *(Uint*)(esp-=4) = (Uint)Data; // Data (shadowed) + *(Uint*)(esp-=4) = 1; // Number of params + *(Uint*)(esp-=4) = (Uint)Fcn; // Function to call + *(Uint*)(esp-=4) = (Uint)newThread; // Thread ID + + newThread->SavedState.ESP = esp; + newThread->SavedState.EIP = (Uint)&NewTaskHeader; + newThread->SavedState.SSE = NULL; +// Log("New (KThread) %p, esp = %p", newThread->SavedState.EIP, newThread->SavedState.ESP); + +// MAGIC_BREAK(); + Threads_AddActive(newThread); + + return newThread->TID; +} + /** * \fn int Proc_Clone(Uint *Err, Uint Flags) * \brief Clone the current process */ -int Proc_Clone(Uint *Err, Uint Flags) +tPID Proc_Clone(Uint Flags) { tThread *newThread; tThread *cur = Proc_GetCurThread(); - Uint eip, esp, ebp; - - __asm__ __volatile__ ("mov %%esp, %0": "=r"(esp)); - __asm__ __volatile__ ("mov %%ebp, %0": "=r"(ebp)); - - newThread = Threads_CloneTCB(Err, Flags); - if(!newThread) return -1; - - // Initialise Memory Space (New Addr space or kernel stack) - if(Flags & CLONE_VM) { - newThread->MemState.CR3 = MM_Clone(); - newThread->KernelStack = cur->KernelStack; - } else { - Uint tmpEbp, oldEsp = esp; - - // Set CR3 - newThread->MemState.CR3 = cur->MemState.CR3; - - // Create new KStack - newThread->KernelStack = MM_NewKStack(); - // Check for errors - if(newThread->KernelStack == 0) { - free(newThread); - return -1; - } + Uint eip; - // Get ESP as a used size - esp = cur->KernelStack - esp; - // Copy used stack - memcpy( (void*)(newThread->KernelStack - esp), (void*)(cur->KernelStack - esp), esp ); - // Get ESP as an offset in the new stack - esp = newThread->KernelStack - esp; - // Adjust EBP - ebp = newThread->KernelStack - (cur->KernelStack - ebp); - - // Repair EBPs & Stack Addresses - // Catches arguments also, but may trash stack-address-like values - for(tmpEbp = esp; tmpEbp < newThread->KernelStack; tmpEbp += 4) - { - if(oldEsp < *(Uint*)tmpEbp && *(Uint*)tmpEbp < cur->KernelStack) - *(Uint*)tmpEbp += newThread->KernelStack - cur->KernelStack; - } + // Sanity, please + if( !(Flags & CLONE_VM) ) { + Log_Error("Proc", "Proc_Clone: Don't leave CLONE_VM unset, use Proc_NewKThread instead"); + return -1; } - // Save core machine state - newThread->SavedState.ESP = esp; - newThread->SavedState.EBP = ebp; - eip = GetEIP(); - if(eip == SWITCH_MAGIC) { - outb(0x20, 0x20); // ACK Timer and return as child + // New thread + newThread = Threads_CloneTCB(Flags); + if(!newThread) return -1; + + newThread->KernelStack = cur->KernelStack; + + // Clone state + eip = Proc_CloneInt(&newThread->SavedState.ESP, &newThread->Process->MemState.CR3, Flags & CLONE_NOUSER); + if( eip == 0 ) { return 0; } - - // Set EIP as parent newThread->SavedState.EIP = eip; + newThread->SavedState.SSE = NULL; + newThread->SavedState.bSSEModified = 0; - // Lock list and add to active + // Check for errors + if( newThread->Process->MemState.CR3 == 0 ) { + Log_Error("Proc", "Proc_Clone: MM_Clone failed"); + Threads_Delete(newThread); + return -1; + } + + // Add the new thread to the run queue Threads_AddActive(newThread); - return newThread->TID; } /** - * \fn int Proc_SpawnWorker() + * \fn int Proc_SpawnWorker(void) * \brief Spawns a new worker thread */ -int Proc_SpawnWorker() +int Proc_SpawnWorker(void (*Fcn)(void*), void *Data) { - tThread *new, *cur; - Uint eip, esp, ebp; - - cur = Proc_GetCurThread(); + tThread *new; + Uint stack_contents[4]; // Create new thread - new = malloc( sizeof(tThread) ); + new = Threads_CloneThreadZero(); if(!new) { Warning("Proc_SpawnWorker - Out of heap space!\n"); return -1; } - memcpy(new, &gThreadZero, sizeof(tThread)); - // Set Thread ID - new->TID = giNextTID++; - // Create a new worker stack (in PID0's address space) - // The stack is relocated by this code - new->KernelStack = MM_NewWorkerStack(); - // Get ESP and EBP based in the new stack - __asm__ __volatile__ ("mov %%esp, %0": "=r"(esp)); - __asm__ __volatile__ ("mov %%ebp, %0": "=r"(ebp)); - esp = new->KernelStack - (cur->KernelStack - esp); - ebp = new->KernelStack - (cur->KernelStack - ebp); + // Create the stack contents + stack_contents[3] = (Uint)Data; + stack_contents[2] = 1; + stack_contents[1] = (Uint)Fcn; + stack_contents[0] = (Uint)new; + // Create a new worker stack (in PID0's address space) + new->KernelStack = MM_NewWorkerStack(stack_contents, sizeof(stack_contents)); + // Save core machine state - new->SavedState.ESP = esp; - new->SavedState.EBP = ebp; - eip = GetEIP(); - if(eip == SWITCH_MAGIC) { - outb(0x20, 0x20); // ACK Timer and return as child - return 0; - } + new->SavedState.ESP = new->KernelStack - sizeof(stack_contents); + new->SavedState.EIP = (Uint)NewTaskHeader; + new->SavedState.SSE = NULL; + new->SavedState.bSSEModified = 0; - // Set EIP as parent - new->SavedState.EIP = eip; // Mark as active - new->Status = THREAD_STAT_ACTIVE; + new->Status = THREAD_STAT_PREINIT; Threads_AddActive( new ); return new->TID; } /** - * \fn Uint Proc_MakeUserStack() + * \fn Uint Proc_MakeUserStack(void) * \brief Creates a new user stack */ -Uint Proc_MakeUserStack() +Uint Proc_MakeUserStack(void) { int i; Uint base = USER_STACK_TOP - USER_STACK_SZ; @@ -581,48 +715,49 @@ Uint Proc_MakeUserStack() if(i != -1) return 0; // Allocate Stack - Allocate incrementally to clean up MM_Dump output - for( i = 0; i < USER_STACK_SZ/4069; i++ ) - MM_Allocate( base + (i<<12) ); + for( i = 0; i < USER_STACK_SZ/0x1000; i++ ) + { + if( !MM_Allocate( base + (i<<12) ) ) + { + Warning("OOM: Proc_MakeUserStack"); + return 0; + } + } return base + USER_STACK_SZ; } - -/** - * \fn void Proc_StartUser(Uint Entrypoint, Uint *Bases, int ArgC, char **ArgV, char **EnvP, int DataSize) - * \brief Starts a user task - */ -void Proc_StartUser(Uint Entrypoint, Uint *Bases, int ArgC, char **ArgV, char **EnvP, int DataSize) +void Proc_StartUser(Uint Entrypoint, Uint Base, int ArgC, const char **ArgV, int DataSize) { - Uint *stack = (void*)Proc_MakeUserStack(); + Uint *stack; int i; - Uint delta; + const char **envp = NULL; Uint16 ss, cs; - LOG("stack = 0x%x", stack); - - // Copy Arguments - stack = (void*)( (Uint)stack - DataSize ); + // Copy data to the user stack and free original buffer + stack = (void*)Proc_MakeUserStack(); + stack -= (DataSize+sizeof(*stack)-1)/sizeof(*stack); memcpy( stack, ArgV, DataSize ); + free(ArgV); // Adjust Arguments and environment - delta = (Uint)stack - (Uint)ArgV; - ArgV = (char**)stack; - for( i = 0; ArgV[i]; i++ ) ArgV[i] += delta; - i ++; - EnvP = &ArgV[i]; - for( i = 0; EnvP[i]; i++ ) EnvP[i] += delta; + if( DataSize ) + { + Uint delta = (Uint)stack - (Uint)ArgV; + ArgV = (const char**)stack; + for( i = 0; ArgV[i]; i++ ) ArgV[i] += delta; + envp = &ArgV[i+1]; + for( i = 0; envp[i]; i++ ) envp[i] += delta; + } // User Mode Segments ss = 0x23; cs = 0x1B; // Arguments - *--stack = (Uint)EnvP; + *--stack = (Uint)envp; *--stack = (Uint)ArgV; *--stack = (Uint)ArgC; - while(*Bases) - *--stack = *Bases++; - *--stack = 0; // Return Address + *--stack = Base; Proc_StartProcess(ss, (Uint)stack, 0x202, cs, Entrypoint); } @@ -703,113 +838,187 @@ void Proc_CallFaultHandler(tThread *Thread) { // Rewinds the stack and calls the user function // Never returns - __asm__ __volatile__ ("mov %0, %%ebp;\n\tcall Proc_ReturnToUser" :: "r"(Thread->FaultHandler)); + Proc_ReturnToUser( Thread->FaultHandler, Thread->CurFaultNum, Thread->KernelStack ); for(;;); } -/** - * \fn void Proc_Scheduler(int CPU) - * \brief Swap current thread and clears dead threads - */ -void Proc_Scheduler(int CPU) +void Proc_DumpThreadCPUState(tThread *Thread) { - Uint esp, ebp, eip; - tThread *thread; - - // If the spinlock is set, let it complete - if(IS_LOCKED(&glThreadListLock)) return; - - // Clear Delete Queue - while(gDeleteThreads) + if( Thread->CurCPU > -1 ) { - thread = gDeleteThreads->Next; - if(gDeleteThreads->IsLocked) { // Only free if structure is unused - gDeleteThreads->Status = THREAD_STAT_NULL; - free( gDeleteThreads ); + int maxBacktraceDistance = 6; + tRegs *regs = NULL; + Uint32 *stack; + + if( Thread->CurCPU != GetCPUNum() ) { + Log(" Currently running"); + return ; } - gDeleteThreads = thread; - } - - // Check if there is any tasks running - if(giNumActiveThreads == 0) { - Log("No Active threads, sleeping"); - __asm__ __volatile__ ("hlt"); - return; + + // Backtrace to find the IRQ entrypoint + // - This will usually only be called by an IRQ, so this should + // work + __asm__ __volatile__ ("mov %%ebp, %0" : "=r" (stack)); + while( maxBacktraceDistance -- ) + { + // [ebp] = oldEbp + // [ebp+4] = retaddr + + if( stack[1] == (tVAddr)&IRQCommon_handled ) { + regs = (void*)stack[2]; + break; + } + + stack = (void*)stack[0]; + } + + if( !regs ) { + Log(" Unable to find IRQ Entry"); + return ; + } + + Log(" at %04x:%08x", regs->cs, regs->eip); + return ; } - // Get current thread - #if USE_MP - thread = gaCPUs[CPU].Current; - #else - thread = gCurrentThread; - #endif - - // Reduce remaining quantum and continue timeslice if non-zero - if(thread->Remaining--) return; - // Reset quantum for next call - thread->Remaining = thread->Quantum; - - // Get machine state - __asm__ __volatile__ ("mov %%esp, %0":"=r"(esp)); - __asm__ __volatile__ ("mov %%ebp, %0":"=r"(ebp)); - eip = GetEIP(); - if(eip == SWITCH_MAGIC) return; // Check if a switch happened + tVAddr diffFromScheduler = Thread->SavedState.EIP - (tVAddr)SwitchTasks; + tVAddr diffFromClone = Thread->SavedState.EIP - (tVAddr)Proc_CloneInt; + tVAddr diffFromSpawn = Thread->SavedState.EIP - (tVAddr)NewTaskHeader; - // Save machine state - thread->SavedState.ESP = esp; - thread->SavedState.EBP = ebp; - thread->SavedState.EIP = eip; + if( diffFromClone > 0 && diffFromClone < 40 ) // When I last checked, .newTask was at .+27 + { + Log(" Creating process"); + return ; + } - // Get next thread - thread = Threads_GetNextToRun(CPU); + if( diffFromSpawn == 0 ) + { + Log(" Creating thread"); + return ; + } - // Error Check - if(thread == NULL) { - Warning("Hmm... Threads_GetNextToRun returned NULL, I don't think this should happen.\n"); - return; + if( diffFromScheduler > 0 && diffFromScheduler < 128 ) // When I last checked, GetEIP was at .+0x30 + { + // Scheduled out + Log(" At %04x:%08x", Thread->SavedState.UserCS, Thread->SavedState.UserEIP); + return ; } - #if DEBUG_TRACE_SWITCH - Log("Switching to task %i, CR3 = 0x%x, EIP = %p", - thread->TID, - thread->MemState.CR3, - thread->SavedState.EIP - ); - #endif + Log(" Just created (unknown %p)", Thread->SavedState.EIP); +} + +void Proc_Reschedule(void) +{ + tThread *nextthread, *curthread; + int cpu = GetCPUNum(); + + // TODO: Wait for the lock? + if(IS_LOCKED(&glThreadListLock)) return; - // Set current thread - #if USE_MP - gaCPUs[CPU].Current = thread; - #else - gCurrentThread = thread; + curthread = Proc_GetCurThread(); + + nextthread = Threads_GetNextToRun(cpu, curthread); + + if(!nextthread || nextthread == curthread) + return ; + + #if DEBUG_TRACE_SWITCH + // HACK: Ignores switches to the idle threads + if( nextthread->TID == 0 || nextthread->TID > giNumCPUs ) + { + LogF("\nSwitching CPU %i to %p (%i %s) - CR3 = 0x%x, EIP = %p, ESP = %p\n", + GetCPUNum(), + nextthread, nextthread->TID, nextthread->ThreadName, + nextthread->Process->MemState.CR3, + nextthread->SavedState.EIP, + nextthread->SavedState.ESP + ); + LogF("OldCR3 = %P\n", curthread->Process->MemState.CR3); + } #endif + + // Update CPU state + gaCPUs[cpu].Current = nextthread; + gTSSs[cpu].ESP0 = nextthread->KernelStack-4; + __asm__ __volatile__("mov %0, %%db0\n\t" : : "r"(nextthread) ); + + // Save FPU/MMX/XMM/SSE state + if( curthread && curthread->SavedState.SSE ) + { + Proc_SaveSSE( ((Uint)curthread->SavedState.SSE + 0xF) & ~0xF ); + curthread->SavedState.bSSEModified = 0; + Proc_DisableSSE(); + } + + if( curthread ) + { + SwitchTasks( + nextthread->SavedState.ESP, &curthread->SavedState.ESP, + nextthread->SavedState.EIP, &curthread->SavedState.EIP, + nextthread->Process->MemState.CR3 + ); + } + else + { + SwitchTasks( + nextthread->SavedState.ESP, 0, + nextthread->SavedState.EIP, 0, + nextthread->Process->MemState.CR3 + ); + } + + return ; +} + +/** + * \fn void Proc_Scheduler(int CPU) + * \brief Swap current thread and clears dead threads + */ +void Proc_Scheduler(int CPU) +{ +#if 0 + tThread *thread; - // Update Kernel Stack pointer - gTSSs[CPU].ESP0 = thread->KernelStack-4; + // If the spinlock is set, let it complete + if(IS_LOCKED(&glThreadListLock)) return; - // Set address space - #if USE_PAE - # error "Todo: Implement PAE Address space switching" - #else - __asm__ __volatile__ ("mov %0, %%cr3"::"a"(thread->MemState.CR3)); - #endif + // Get current thread + thread = gaCPUs[CPU].Current; - #if 0 - if(thread->SavedState.ESP > 0xC0000000 - && thread->SavedState.ESP < thread->KernelStack-0x2000) { - Log_Warning("Proc", "Possible bad ESP %p (PID %i)", thread->SavedState.ESP); + if( thread ) + { + tRegs *regs; + Uint ebp; + // Reduce remaining quantum and continue timeslice if non-zero + if( thread->Remaining-- ) + return; + // Reset quantum for next call + thread->Remaining = thread->Quantum; + + // TODO: Make this more stable somehow + __asm__ __volatile__("mov %%ebp, %0" : "=r" (ebp)); + regs = (tRegs*)(ebp+(2+2)*4); // EBP,Ret + CPU,CurThread + thread->SavedState.UserCS = regs->cs; + thread->SavedState.UserEIP = regs->eip; + + if(thread->bInstrTrace) { + regs->eflags |= 0x100; // Set TF + Log("%p De-scheduled", thread); + } + else + regs->eflags &= ~0x100; // Clear TF } + + // TODO: Ack timer? + #if USE_MP + if( GetCPUNum() ) + gpMP_LocalAPIC->EOI.Val = 0; + else #endif - - // Switch threads - __asm__ __volatile__ ( - "mov %1, %%esp\n\t" // Restore ESP - "mov %2, %%ebp\n\t" // and EBP - "jmp *%3" : : // And return to where we saved state (Proc_Clone or Proc_Scheduler) - "a"(SWITCH_MAGIC), "b"(thread->SavedState.ESP), - "d"(thread->SavedState.EBP), "c"(thread->SavedState.EIP) - ); - for(;;); // Shouldn't reach here + outb(0x20, 0x20); + __asm__ __volatile__ ("sti"); + Proc_Reschedule(); +#endif } // === EXPORTS ===