working on smp scheduler

1 files changed, 192 insertions, 169 deletions

diff --git a/kernel/scheduler.c b/kernel/scheduler.c
index cac5ba7..d0b4b68 100644
--- a/kernel/scheduler.c
+++ b/kernel/scheduler.c
@@ -2,6 +2,7 @@
 #include "kernel.h"
 #include "gdt.h"
 
+#include "smp.h"
 #include "mem.h"
 #include "fs/elf.h"
 #include "asm_x86.h"
@@ -16,9 +17,10 @@
 
 #define NO_TASK 0xffffffff
 
-static volatile uint32_t current_task=NO_TASK;
+// we hold this stuff per cpu
+static volatile uint32_t current_task[SMP_MAX_PROC];
 
-// TODO: per cpu!
+// we hold this stuff per cpu
 static volatile struct task_list_struct
 {
     volatile bool active;      // is this slot used (Y/N)
@@ -37,11 +39,142 @@ static volatile struct task_list_struct
     volatile uint32_t ecx;
     volatile uint32_t edx;
     
-}task_list[MAX_TASKS];
+}task_list[SMP_MAX_PROC][MAX_TASKS];
+
+// init tasks //
+volatile void scheduler_init(uint32_t cpu, void *dir)
+{
+    for(int i=0;i<MAX_TASKS;i++)
+    {
+	task_list[cpu][i].active=false;
+    }
+
+    current_task[cpu]=0;
+   
+    // need to make space on the esp stacks for pushing vals vias task_pusha
+    
+    // this is our main kernel task at slot 0 (per cpu)
+    task_list[cpu][0].parent=0;
+    task_list[cpu][0].active=true;
+    task_list[cpu][0].wait=false;
+    task_list[cpu][0].vmem=dir;
+    task_list[cpu][0].esp = VMEM_CPU_STACK_TOP-0x200;
+    task_list[cpu][0].esp0 = 0; // esp0 not needed by kernel space tasks
+
+    task_pusha(task_list[cpu][0].esp);
+
+    // this is our main kernel task at slot 0 (per cpu)
+    task_list[cpu][1].parent=0;
+    task_list[cpu][1].active=true;
+    task_list[cpu][1].wait=false;
+    task_list[cpu][1].vmem=dir;
+    task_list[cpu][1].esp = kballoc(4)+4*4096-0x200; // 4 pages stack
+    task_list[cpu][1].esp0 =kballoc(4)+4*4096;  // esp0 not needed by kernel space tasks
+
+    task_pusha(task_list[cpu][0].esp);
+    task_pusha(task_list[cpu][1].esp);
+}
+
+//
+// REMEMBER WE ARE INSIDE AN INTERRUPT HERE - DON'T WASTE TIME!
+//
+// oldesp - is the adress of the stack pointer when pit_interrupt_handler was entered.
+// registers have been pushed with pusha to this old stack.
+//
+// stack pointer was moved to the 16kb stack we have from multiboot.s
+//
+// we need to return a NEW stack pointer where popa will get the registers the new task requires
+//
+volatile uint32_t scheduler_run(uint32_t oldesp,uint32_t force_pid)
+{
+    uint32_t cpu=smp_get(SMP_APIC_ID);
+    uint32_t init=smp_get(SMP_SCHEDULER_INIT);
+    if(init){
+	scheduler_init(cpu,x86_get_page_directory());
+        smp_set(SMP_SCHEDULER_INIT,0);
+    }
+    else task_list[cpu][current_task[cpu]].esp=oldesp;
+
+    for(int i=0;i<MAX_TASKS;i++)
+    {
+	int idx=(current_task[cpu]+1+i)%MAX_TASKS; // schedule round robin style
+
+	if(task_list[cpu][idx].active && !task_list[cpu][idx].wait) // find active non-blocked task
+	{
+	    current_task[cpu]=idx;
+	    
+	    install_tss(cpu,task_list[cpu][idx].esp0);
+	    x86_set_page_directory(task_list[cpu][idx].vmem);
+	    return task_list[cpu][idx].esp;
+	}
+    }
+
+    kpanic("nothing to schedule!");
+}
+
+
+void scheduler_func()
+{
+    // we need enable here again (since the pushed eflags have it disabled)? TODO: why??
+    x86_sti(); 
+
+    uint32_t cpu=smp_get(SMP_APIC_ID);
+
+    if(task_get_current_pid()==0)
+    while(1)
+    {
+        task_syscall_worker();
+    }
+
+    if(task_get_current_pid()==1)
+
+    while(1)
+    {
+	if(cpu==0)
+	{
+	    uint32_t alloc;
+	    uint32_t entry_global=load_elf(BIN_INIT,&alloc);
+	    task_set_brk(alloc);
+	    klog("breakpoint: 0x%08x",alloc);
+	    asm_usermode(entry_global); 
+        while(1);
+	}
+//	else syscall_write(1, "x",1); // stdout
+	    
+    }
+}
+    //asm("hlt");
+    /*
+
+    // if we are pid 0, replace ourselves with /bin/init and enter usermode
+    if(task_get_current_pid()==0)
+    {
+	    uint32_t alloc;
+	    uint32_t entry_global=load_elf(BIN_INIT,&alloc);
+	    task_set_brk(alloc);
+	    asm_usermode(entry_global); 
+    }
+
+    // kernel worker thread: SLEEPER
+    if(task_get_current_pid()==1)
+    {
+	while(1)
+	{
+	    __asm__("hlt");
+	}
+    }
+
+    // kernel worker thread: SYSCALL CHECKER
+    if(task_get_current_pid()==2)
+    {
+        task_syscall_worker();
+    }
+    */
 
 volatile int task_reset(uint32_t pid, uint32_t entry, uint32_t stack)
 {
-    uint32_t *stk=task_list[pid].esp;
+    uint32_t cpu=smp_get(SMP_APIC_ID);
+    uint32_t *stk=task_list[cpu][pid].esp;
     stk[14]=entry;
     stk[17]=stack;
     return 1;
@@ -49,24 +182,25 @@ volatile int task_reset(uint32_t pid, uint32_t entry, uint32_t stack)
 
 volatile int add_task(uint32_t parent,uint32_t vmem)
 {
+    uint32_t cpu=smp_get(SMP_APIC_ID);
     for(int i=0;i<MAX_TASKS;i++)
     {
-	if(task_list[i].active!=true)
+	if(task_list[cpu][i].active!=true)
 	{
-	    task_list[i].parent=parent;
+	    task_list[cpu][i].parent=parent;
 
 	    // TODO: do this without paging please!
-	    task_list[i].vmem=vmem; 
-	    task_list[i].esp = kballoc(4)+2*4096; // center
+	    task_list[cpu][i].vmem=vmem; 
+	    task_list[cpu][i].esp = kballoc(4)+2*4096; // center
 
 	    // TODO: use own page here and copy it instead of this!
-            task_list[i].esp0 = kballoc(4)+4*4096;
+            task_list[cpu][i].esp0 = kballoc(4)+4*4096;
 
-	    task_list[i].wait=false;
-	    task_list[i].brk=task_list[current_task].brk;
+	    task_list[cpu][i].wait=false;
+	    task_list[cpu][i].brk=task_list[cpu][current_task[cpu]].brk;
 
-            uint32_t *source=(uint32_t *)task_list[parent].esp;
-	    uint32_t *dst=(uint32_t *)task_list[i].esp;
+            uint32_t *source=(uint32_t *)task_list[cpu][parent].esp;
+	    uint32_t *dst=(uint32_t *)task_list[cpu][i].esp;
 
 	    for(int x=0;x<100;x++) //TODO:  better copy this page too instead of stack
 	    {
@@ -75,11 +209,11 @@ volatile int add_task(uint32_t parent,uint32_t vmem)
 		source++;
 	    }
 
-	    uint32_t *stack=task_list[i].esp;
+	    uint32_t *stack=task_list[cpu][i].esp;
 	    stack[12]=0x1; 
 	    stack[13]=0; // this task returns pid=0 to the caller
 
-	    task_list[i].active=true; //TODO: LOCK! (also other similar)
+	    task_list[cpu][i].active=true; //TODO: LOCK! (also other similar)
 	    return i;
 	}
     }
@@ -97,6 +231,7 @@ volatile int add_task(uint32_t parent,uint32_t vmem)
 
 void task_syscall_worker()
 {
+    uint32_t cpu=smp_get(SMP_APIC_ID);
     while(1)
     {
 	//klog("checking if any pending syscalls.");
@@ -104,40 +239,40 @@ void task_syscall_worker()
 	bool nowork=true;
 	for(int i=0;i<MAX_TASKS;i++)
 	{
-	    if(task_list[i].wait)
+	    if(task_list[cpu][i].wait)
 	    {
-    //	    klog("task %d waiting on syscall %d. processing...",i,task_list[i].eax);
-		task_list[2].vmem=task_list[i].vmem; // switch syscall worker to pagedir of calling userprog
-		x86_set_page_directory(task_list[2].vmem);
+    //	    klog("task %d waiting on syscall %d. processing...",i,task_list[cpu][i].eax);
+		task_list[cpu][2].vmem=task_list[cpu][i].vmem; // switch syscall worker to pagedir of calling userprog
+		x86_set_page_directory(task_list[cpu][2].vmem);
 
-		if(task_list[i].eax==SYSCALL_WAIT)
+		if(task_list[cpu][i].eax==SYSCALL_WAIT)
 		{
 		    continue;
 		}
 
-		if(task_list[i].eax==SYSCALL_READ)
+		if(task_list[cpu][i].eax==SYSCALL_READ)
 		{
 		uint32_t ok= chk_syscall_read(
-			    task_list[i].edx,
-			    task_list[i].ecx,
-			    task_list[i].ebx
+			    task_list[cpu][i].edx,
+			    task_list[cpu][i].ecx,
+			    task_list[cpu][i].ebx
 			    );
 		if(!ok)continue;
 		}
 
 		nowork=false;
 
-		uint32_t ret= syscall_generic(task_list[i].eax,
-			    task_list[i].edx,
-			    task_list[i].ecx,
-			    task_list[i].ebx,
+		uint32_t ret= syscall_generic(task_list[cpu][i].eax,
+			    task_list[cpu][i].edx,
+			    task_list[cpu][i].ecx,
+			    task_list[cpu][i].ebx,
 			    i);
 
-		uint32_t *stack=task_list[i].esp;
+		uint32_t *stack=task_list[cpu][i].esp;
 		stack[12]=0x1; 
 		stack[13]=ret; 
 	    
-		task_list[i].wait=false;
+		task_list[cpu][i].wait=false;
 	    }
 	}
 
@@ -147,67 +282,17 @@ void task_syscall_worker()
     }
 }
 
-//
-// REMEMBER WE ARE INSIDE AN INTERRUPT HERE - DON'T WASTE TIME!
-//
-// oldesp - is the adress of the stack pointer when pit_interrupt_handler was entered.
-// registers have been pushed with pusha to this old stack.
-//
-// stack pointer was moved to the 16kb stack we have from multiboot.s
-//
-// we need to return a NEW stack pointer where popa will get the registers the new task requires
-//
-volatile uint32_t my_scheduler(uint32_t oldesp,uint32_t force_pid)
-{
-    uint32_t *apic_id=0x8000000; //TODO: test cpu private pages
-    //klog("scheduler 0x%x",*apic_id); // TODO: do not log we are inisde an interrupt!!
-
-    // 
-    static bool first=true;
-    if(current_task==NO_TASK)return oldesp;
-    if(!first) task_list[current_task].esp=oldesp;
-    first=false;
-    //
-
-    if(force_pid>-1)
-    {
-	    int pid=force_pid;
-	    current_task=pid;
-	    install_tss(0,task_list[pid].esp0);
-
-	    x86_set_page_directory(task_list[pid].vmem);
-	    return task_list[pid].esp;
-    }
-
-    for(int i=0;i<MAX_TASKS;i++)
-    {
-	int pid=(current_task+1+i)%MAX_TASKS; // schedule round robin style
-
-	if(task_list[pid].active && !task_list[pid].wait) // find active non-blocked task
-	{
-	    //if(current_task!=pid)klog("switch from %d to %d", current_task, pid);
-
-	    current_task=pid;
-	    install_tss(0,task_list[pid].esp0);
-
-	    x86_set_page_directory(task_list[pid].vmem);
-	    return task_list[pid].esp;
-	}
-
-    }
-
-    kpanic("nothing to schedule!");
-}
 
 volatile uint32_t task_syscall(uint32_t eax,uint32_t ebx, uint32_t ecx, uint32_t edx)
 {
-    task_list[current_task].wait=true;
-    task_list[current_task].eax=eax;
-    task_list[current_task].ebx=ebx;
-    task_list[current_task].ecx=ecx;
-    task_list[current_task].edx=edx;
-
-    task_list[2].wait=false;
+    uint32_t cpu=smp_get(SMP_APIC_ID);
+    task_list[cpu][current_task[cpu]].wait=true;
+    task_list[cpu][current_task[cpu]].eax=eax;
+    task_list[cpu][current_task[cpu]].ebx=ebx;
+    task_list[cpu][current_task[cpu]].ecx=ecx;
+    task_list[cpu][current_task[cpu]].edx=edx;
+
+    task_list[cpu][2].wait=false;
     return 1;
 }
 
@@ -215,118 +300,56 @@ volatile uint32_t task_syscall(uint32_t eax,uint32_t ebx, uint32_t ecx, uint32_t
 //TODO: notify waiting parent when child finished;
 volatile uint32_t task_exit(uint32_t pid)
 { 
-    task_list[pid].active=false;
-    int parent_pid=task_list[pid].parent;
-    if(task_list[parent_pid].wait&&task_list[parent_pid].eax==SYSCALL_WAIT)
-	    task_list[parent_pid].wait=false;
+    uint32_t cpu=smp_get(SMP_APIC_ID);
+    task_list[cpu][pid].active=false;
+    int parent_pid=task_list[cpu][pid].parent;
+    if(task_list[cpu][parent_pid].wait&&task_list[cpu][parent_pid].eax==SYSCALL_WAIT)
+	    task_list[cpu][parent_pid].wait=false;
     klog("[%d] exit", pid);
-    vmem_free_dir(task_list[pid].vmem);
+    vmem_free_dir(task_list[cpu][pid].vmem);
     return 1;
 }
 
 volatile uint32_t task_wait(uint32_t pid)
 { 
+    uint32_t cpu=smp_get(SMP_APIC_ID);
     klog("[%d] wait", pid);
-    task_list[pid].wait=true;
-    task_list[pid].eax=SYSCALL_WAIT;
+    task_list[cpu][pid].wait=true;
+    task_list[cpu][pid].eax=SYSCALL_WAIT;
     return 1;
 }
 
 volatile uint32_t task_fork(uint32_t pid)
 { 
+    uint32_t cpu=smp_get(SMP_APIC_ID);
 //TODO: what will happen if we get rescheduled!?!?!
-   int ret=add_task(pid,vmem_new_space_dir(task_list[pid].vmem,false));
+   int ret=add_task(pid,vmem_new_space_dir(task_list[cpu][pid].vmem,false));
    klog("[%d] forked -> [%d] (free blocks remaining: %d )", pid, ret,0);
    return ret;
 }
 volatile uint32_t task_clone(uint32_t pid)
 { 
+    uint32_t cpu=smp_get(SMP_APIC_ID);
 //TODO: what will happen if we get rescheduled!?!?!
-   int ret=add_task(pid,vmem_new_space_dir(task_list[pid].vmem,true));
+   int ret=add_task(pid,vmem_new_space_dir(task_list[cpu][pid].vmem,true));
    klog("[%d] cloned -> [%d] (free blocks remaining: %d )", pid, ret,0);
    return ret;
 }
 
-// init task (root of all other tasks / processes) //
-volatile void scheduler_init(void *dir)
-{
-    for(int i=0;i<MAX_TASKS;i++)
-    {
-	task_list[i].active=false;
-    }
-
-    current_task=0;
-    
-    // this is our main user task on slot 0 
-    task_list[0].parent=0;
-    task_list[0].active=true;
-    task_list[0].wait=false;
-    task_list[0].vmem=dir;
-    task_list[0].esp = kballoc(4)+3*4096;
-    task_list[0].esp0 = kballoc(4)+4*4096;
-
-//    task_list[1].parent=0;
-//    task_list[1].active=true;
-//    task_list[1].waiting=false;
-//    task_list[1].syscall=false;
-//    task_list[1].vmem=dir;
-//    task_list[1].esp = kballoc(4)+3*4096; 
-//    task_list[1].esp0 = 0; // not needed by kernel space tasks
-
-    task_list[2].parent=0;
-    task_list[2].active=true;
-    task_list[2].wait=false;
-    task_list[2].vmem=dir;
-    task_list[2].esp = kballoc(4)+3*4096; 
-    task_list[2].esp0 = 0; // not needed by kernel space tasks
-
-    task_pusha(task_list[2].esp);
-//    task_pusha(task_list[1].esp);
-    task_pusha(task_list[0].esp);
-}
-
 volatile int task_get_current_pid()
 {
-    return current_task;
+    uint32_t cpu=smp_get(SMP_APIC_ID);
+    return current_task[cpu];
 }
 
 volatile uint32_t task_get_brk()
 {
-    return task_list[current_task].brk;
+    uint32_t cpu=smp_get(SMP_APIC_ID);
+    return task_list[cpu][current_task[cpu]].brk;
 }
 
 volatile void task_set_brk(uint32_t brk)
 {
-    task_list[current_task].brk=brk;
-}
-
-void userfunc()
-{
-
-    // we need enable here again (since the pushed eflags have it disabled)!
-    x86_sti(); 
-
-    // if we are pid 0, replace ourselves with /bin/init and enter usermode
-    if(task_get_current_pid()==0)
-    {
-	    uint32_t alloc;
-	    uint32_t entry_global=load_elf(BIN_INIT,&alloc);
-	    task_set_brk(alloc);
-	    asm_usermode(entry_global); 
-    }
-
-    // kernel worker thread: SLEEPER
-    if(task_get_current_pid()==1)
-    {
-	while(1)
-	{
-	    __asm__("hlt");
-	}
-    }
-
-    // kernel worker thread: SYSCALL CHECKER
-    if(task_get_current_pid()==2)
-    {
-        task_syscall_worker();
-    }
+    uint32_t cpu=smp_get(SMP_APIC_ID);
+    task_list[cpu][current_task[cpu]].brk=brk;
 }