Setup kernel stack; but process is failing

src/config.h

@@ -27,6 +27,7 @@
 #define KERNEL_BASE 0xFFFFFFFF80000000ULL
 // 2 MB should be enough (as of now, the whole kernel ELF is around 75kb)
 #define KERNEL_SIZE 0x200000
+#define KERNEL_STACK_SIZE 65536
 
 /* heap */
 #define KHEAP_SIZE (16*1024*1024)

src/entry.S

@@ -0,0 +1,23 @@
+bits 64
+global _start
+
+extern kmain
+extern kernel_stack
+
+KERNEL_STACK_SIZE equ 65536
+
+section .text
+
+_start:
+    cli
+
+    ; load kernel stack
+    lea rsp, [kernel_stack+KERNEL_STACK_SIZE]
+
+    ; rbp=0 so last frame in stack trace
+    xor rbp, rbp
+
+    ; 16 byte align
+    and rsp, -16
+
+    call kmain

src/idt/idt.c

@@ -13,6 +13,7 @@
 #include <stdbool.h>
 #include "sched/scheduler.h"
 #include "config.h"
+#include "sched/process.h"
 
 struct interrupt_descriptor idt[256];
 struct idtr idt_reg;
@@ -118,6 +119,11 @@ static void gp_fault_handler(struct cpu_status_t* ctx)
 
 struct cpu_status_t* interrupt_dispatch(struct cpu_status_t* context)
 {
+    if (context == NULL)
+    {
+        panic(NULL, "Interrupt dispatch recieved NULL context!");
+    }
+
     switch(context->vector_number)
     {
         case 0:
@@ -140,6 +146,7 @@ struct cpu_status_t* interrupt_dispatch(struct cpu_status_t* context)
             break;
         case 6:
             DEBUG("Invalid Opcode!");
+            panic(context, "Invalid Opcode!");
             break;
         case 7:
             DEBUG("Device Not Available!");
@@ -194,7 +201,8 @@ struct cpu_status_t* interrupt_dispatch(struct cpu_status_t* context)
             if (ticks % SCHEDULER_QUANTUM == 0)
             {
                 CLEAR_INTERRUPTS;
-                scheduler_schedule();
+                struct cpu_status_t* current_ctx = scheduler_schedule(context);
+                process_switch(current_ctx->iret_rsp, current_ctx->iret_rip);
                 SET_INTERRUPTS;
             }
 

src/kernel.h

@@ -30,6 +30,7 @@ enum ErrorCodes
 
 void panic(struct cpu_status_t* ctx, const char* str);
 void hcf();
+void idle();
 #define assert(check) do { if(!(check)) hcf(); } while(0)
 
 struct boot_context

src/kmain.c

@@ -37,6 +37,8 @@ void hcf()
 // Doing nothing (can be interrupted)
 void idle() {for(;;)asm("hlt");}
 
+uint8_t kernel_stack[KERNEL_STACK_SIZE] __attribute__((aligned(16)));
+
 void panic(struct cpu_status_t* ctx, const char* str)
 {
 	CLEAR_INTERRUPTS;
@@ -68,17 +70,13 @@ extern struct process_t* current_process;
 
 void pedicel_main(void* arg)
 {
-	panic(NULL, "we did it!");
+	printf("Hello, world from a KERNEL PROCESS!");
+	//panic(NULL, "WE DID IT! Hello, world from a PROCESS!!!!");
 }
 
-void two_main(void* arg)
+void idle_main(void* arg)
 {
-
-}
-
-void three_main(void* arg)
-{
-
+	for(;;)asm("hlt");
 }
 
 // This is our entry point
@@ -113,8 +111,7 @@ void kmain()
 	vmm_init();
 	
 	struct process_t* pedicel = process_create("pedicel", (void*)pedicel_main, 0);
-	//struct process_t* two = process_create("two", (void*)two_main, 0);
-	//struct process_t* three = process_create("three", (void*)three_main, 0);
+	
 	process_display_list(processes_list);
 
 	scheduler_init();

src/mem/heap/kheap.c

@@ -142,6 +142,6 @@ void kfree(void* ptr)
 // Should return a pointer to top of the stack (as stack grows DOWNWARDS)
 void* kalloc_stack()
 {
-    uint8_t* ptr = kmalloc(PROCESS_STACK_SIZE);
+    uint8_t* ptr = kmalloc(PROCESS_STACK_SIZE); // As it's out of kmalloc, stack is already mapped into kernel space
     return ptr ? ptr+PROCESS_STACK_SIZE : NULL;
 }

src/mem/paging/paging.c

@@ -24,7 +24,7 @@ If we use 1GB huge pages: PML4 -> PDPT -> 1gb pages
 4KB (regular size): PML4 -> PDPT -> PD -> PT -> 4kb pages
 */
 
-static inline void load_cr3(uint64_t value) {
+void load_cr3(uint64_t value) {
     asm volatile ("mov %0, %%cr3" :: "r"(value) : "memory");
 }
 

src/mem/paging/paging.h

@@ -16,6 +16,9 @@
 void paging_init();
 void paging_map_page(uint64_t* root_table, uint64_t virt, uint64_t phys, uint64_t flags);
 
+// To swap root page tables
+void load_cr3(uint64_t value);
+
 extern uint64_t hhdm_off;
 
 #define PHYS_TO_VIRT(x) ((void*)((uintptr_t)(x) + hhdm_off))

src/sched/process.c

@@ -11,10 +11,13 @@
 #include "string/string.h"
 #include "mem/gdt/gdt.h"
 #include "config.h"
+#include "io/serial/serial.h"
 
 struct process_t* processes_list;
 struct process_t* current_process;
 
+extern uint64_t *kernel_pml4;
+
 size_t next_free_pid = 0;
 
 void process_init()
@@ -52,14 +55,22 @@ struct process_t* process_create(char* name, void(*function)(void*), void* arg)
     proc->pid = next_free_pid++;
     proc->status = READY;
 
+    uint64_t* stack_top = (uint64_t*)kalloc_stack();
+    // push return address to the stack so when "ret" hits we jmp to exit instead of idk what
+    // stack grows DOWNWARDS!!
+    *(--stack_top) = (uint64_t)process_exit;
+
     proc->context = ctx;
     proc->context->iret_ss = KERNEL_DATA_SEGMENT; // process will live in kernel mode
-    proc->context->iret_rsp = (uint64_t)kalloc_stack();
+    proc->context->iret_rsp = (uint64_t)stack_top;
     proc->context->iret_flags = 0x202; //bit 2 and 9 set (Interrupt Flag)
     proc->context->iret_cs = KERNEL_CODE_SEGMENT;
     proc->context->iret_rip = (uint64_t)function; // beginning of executable code
     proc->context->rdi = (uint64_t)arg; // 1st arg is in rdi (as per x64 calling convention)
-    proc->context->rbp = 0;
+    proc->context->rbp = 0; 
+
+    // Kernel PML4 as it already maps code/stack (when switching to userland we'll have to change that)
+    proc->root_page_table = kernel_pml4;
 
     proc->next = 0;
 
@@ -126,4 +137,38 @@ struct process_t* process_get_next(struct process_t* process)
 {
     if (!process) return NULL;
     return process->next;
+}
+
+// (from gdt) This will switch tasks ONLY in ring 0
+// KERNEL CS = 0x08
+// KERNEL SS = 0x10
+__attribute__((naked, noreturn))
+void process_switch(uint64_t stack_addr, uint64_t code_addr)
+{
+    asm volatile(" \
+        push $0x10 \n\
+        push %0 \n\
+        push $0x202 \n\
+        push $0x08 \n\
+        push %1 \n\
+        iretq \n\
+        " :: "r"(stack_addr), "r"(code_addr));
+}
+
+
+// Will be used to clean up resources (if any, when we implement it)
+// Just mark as DEAD then idle. Scheduler will delete process at next timer interrupt % quantum.
+void process_exit()
+{
+    CLEAR_INTERRUPTS;
+    if (current_process)
+    {
+        current_process->status = DEAD;
+    }
+    SET_INTERRUPTS;
+    
+    for (;;)
+    {
+        asm("hlt");
+    }
 }

src/sched/process.h

@@ -9,6 +9,7 @@
 
 #include <stddef.h>
 #include "config.h"
+#include <stdint.h>
 
 typedef enum 
 {
@@ -33,6 +34,8 @@ struct process_t* process_create(char* name, void(*function)(void*), void* arg);
 void process_add(struct process_t** processes_list, struct process_t* process);
 void process_delete(struct process_t** processes_list, struct process_t* process);
 struct process_t* process_get_next(struct process_t* process);
+void process_switch(uint64_t stack_addr, uint64_t code_addr);
+void process_exit();
 
 void process_display_list(struct process_t* processes_list);
 

src/sched/scheduler.c

@@ -6,6 +6,8 @@
 
 #include "kernel.h"
 #include "process.h"
+#include "mem/paging/paging.h"
+#include <stdint.h>
 
 extern struct process_t* processes_list;
 extern struct process_t* current_process;
@@ -18,7 +20,7 @@ void scheduler_init()
 
 struct cpu_status_t* scheduler_schedule(struct cpu_status_t* context)
 {
-    current_process->context = context;
+    //current_process->context = context;
     current_process->status = READY;
 
     for (;;) {
@@ -41,6 +43,13 @@ struct cpu_status_t* scheduler_schedule(struct cpu_status_t* context)
         }
     }
     
-    DEBUG("current_process={pid=%u name='%s'}", current_process->pid, current_process->name);
+
+    // Current_process gets wrong context?? (iret_rip points to other stuff than process function; like putpixel() for example)
+    DEBUG("current_process={pid=%u, name='%s', root_page_table[virt]=%p}", current_process->pid, current_process->name, current_process->root_page_table);
+
+    load_cr3(VIRT_TO_PHYS((uint64_t)current_process->root_page_table));
+    DEBUG("loaded process pml4 into cr3");
+    /* process_switch(current_process->context->rbp, current_process->context->iret_rip);
+    DEBUG("switched to process rip!"); */
     return current_process->context;
 }

src/sched/scheduler.h

@@ -7,7 +7,7 @@
 #ifndef SCHEDULER_H
 #define SCHEDULER_H
 
-void scheduler_schedule();
+struct cpu_status_t* scheduler_schedule(struct cpu_status_t* context);
 void scheduler_init();
 
 #endif