metalsvm/kernel/tasks.c

/* 
 * Copyright 2010 Stefan Lankes, Chair for Operating Systems,
 *                               RWTH Aachen University
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 *
 * This file is part of MetalSVM. 
 */

/**
 * @author Stefan Lankes
 * @file kernel/tasks.c
 * @brief Implementations of task loading, killing, scheduling.
 *
 * This files contains all the implementations of different functions
 * to start tasks with, wake them up, schedule them, etc.
 */

#include <metalsvm/stdio.h>
#include <metalsvm/stdlib.h>
#include <metalsvm/string.h>
#include <metalsvm/errno.h>
#include <metalsvm/mmu.h>
#include <metalsvm/page.h>
#include <metalsvm/tasks.h>
#include <metalsvm/processor.h>
#include <metalsvm/spinlock.h>
#include <metalsvm/mailbox.h>
#include <metalsvm/syscall.h>
#include <metalsvm/fs.h>
#include <asm/apic.h>
#include <asm/elf.h>

DEFINE_PER_CORE(task_t*, current_task, NULL);

/** @brief Array of task structures 
 *
 * A task's id will be its position in this array.
 */
static task_t task_table[MAX_TASKS] = {[0 ... MAX_TASKS-1] = {0, TASK_INVALID, ATOMIC_INIT(0), \
			 SPINLOCK_INIT, NULL, SPINLOCK_INIT, NULL, 0}};
static spinlock_irqsave_t table_lock = SPINLOCK_IRQSAVE_INIT;

/** @brief helper function for the assembly code to determine the current task
 * @return Pointer to the task_t structure of current task
 */
task_t* get_current_task(void) {
	return per_core(current_task);
}

int multitasking_init(void) {
	if (task_table[0].status == TASK_INVALID) {
		task_table[0].id = 0;
		task_table[0].status = TASK_RUNNING;
		atomic_int32_set(&task_table[0].user_usage, 0);
		mailbox_wait_msg_init(&task_table[0].inbox);
		memset(task_table[0].outbox, 0x00, sizeof(mailbox_wait_msg_t*)*MAX_TASKS);
		per_core(current_task) = task_table+0;
		per_core(current_task)->pgd = get_boot_pgd();
		task_table[0].fpu_used = 0;
		return 0;
	}

	return -ENOMEM;
}

/** @brief Wakeup tasks which are waiting for a message from the current one
 *
 * @param result Current task's resulting return value 
 */
static void wakeup_blocked_tasks(int result)
{
	wait_msg_t tmp = { per_core(current_task)->id, result };
	unsigned int i;

	spinlock_irqsave_lock(&table_lock);

	/* wake up blocked tasks */
	for(i=0; i<MAX_TASKS; i++) {
		if (per_core(current_task)->outbox[i]) {
			mailbox_wait_msg_post(per_core(current_task)->outbox[i], tmp);
			per_core(current_task)->outbox[i] = NULL;
		}
	}

	spinlock_irqsave_unlock(&table_lock);
}

/** @brief A procedure to be called by 
 * procedures which are called by exiting tasks. */
static void NORETURN do_exit(int arg) {
	vma_t* tmp;

	kprintf("Terminate task: %u, return value %d\n", per_core(current_task)->id, arg);

	wakeup_blocked_tasks(arg);

	//vma_dump(per_core(current_task));
	spinlock_lock(&(per_core(current_task)->vma_lock));

	// remove memory regions
	while((tmp = per_core(current_task)->vma_list) != NULL) {
		kfree((void*) tmp->start, tmp->end - tmp->start + 1);
		per_core(current_task)->vma_list = tmp->next;
		kfree((void*) tmp, sizeof(vma_t));
	}

	spinlock_unlock(&(per_core(current_task)->vma_lock));

	drop_pgd(); // delete page directory and its page tables
	
	if (atomic_int32_read(&per_core(current_task)->user_usage))
		kprintf("Memory leak! Task %d did not release %d pages\n", 
				per_core(current_task)->id, atomic_int32_read(&per_core(current_task)->user_usage));
	per_core(current_task)->status = TASK_FINISHED;
	reschedule();
	
	kputs("Kernel panic: scheduler found no valid task\n");
	while(1) {
		HALT;
	}
}

/** @brief A procedure to be called by kernel tasks */
void NORETURN leave_kernel_task(void) {
        int result;

        result = get_return_value();
        do_exit(result);
}

/** @brief To be called by the systemcall to exit tasks */
void NORETURN sys_exit(int arg) {
	do_exit(arg);
}

/** @brief Aborting a task is like exiting it with result -1 */
void NORETURN abort(void) {
	do_exit(-1);
}

/** @brief Create a task with a specific entry point
 *
 * @param id Pointer to a tid_t struct were the id shall be set
 * @param ep Pointer to the function the task shall start with
 * @param arg Arguments list
 * @return
 * - 0 on success
 * - -ENOMEM (-12) or -EINVAL (-22) on failure
 */
static int create_task(tid_t* id, entry_point_t ep, void* arg)
{
	int ret = -ENOMEM;
	unsigned int i;

	if (BUILTIN_EXPECT(!ep, 0))
		return -EINVAL;

	spinlock_irqsave_lock(&table_lock);

	for(i=0; i<MAX_TASKS; i++) {
		if (task_table[i].status == TASK_INVALID) {
			atomic_int32_set(&task_table[i].user_usage, 0);

			ret = create_pgd(task_table+i, 0);		
			if (ret < 0) {
				ret = -ENOMEM;
				goto create_task_out;
			}

			task_table[i].id = i;
			spinlock_init(&task_table[i].vma_lock);
			task_table[i].vma_list = NULL;
			mailbox_wait_msg_init(&task_table[i].inbox);
			memset(task_table[i].outbox, 0x00, sizeof(mailbox_wait_msg_t*)*MAX_TASKS);
			task_table[i].outbox[per_core(current_task)->id] = &per_core(current_task)->inbox; 

			if (id)
				*id = i;	

			ret = create_default_frame(task_table+i, ep, arg);

			task_table[i].fpu_used = 0;
			task_table[i].status = TASK_READY;
			break;
		}
	}

create_task_out:
	spinlock_irqsave_unlock(&table_lock);

	return ret;
}


int sys_fork(void)
{
	int ret = -ENOMEM;
	unsigned int i;
	task_t* parent_task = per_core(current_task);
	vma_t** child;
	vma_t* parent;
	vma_t* tmp;

	spinlock_lock(&per_core(current_task)->vma_lock);
	spinlock_irqsave_lock(&table_lock);

	for(i=0; i<MAX_TASKS; i++) {
		if (task_table[i].status == TASK_INVALID) {
			atomic_int32_set(&task_table[i].user_usage, 0);

			ret = create_pgd(task_table+i, 1);		
			if (ret < 0) {
				ret = -ENOMEM;
				goto create_task_out;
			}

			task_table[i].id = i;
			spinlock_init(&task_table[i].vma_lock);

			// copy VMA list
			child = &task_table[i].vma_list;
			parent = per_core(current_task)->vma_list;
			tmp = NULL;

			while(parent) {
				*child = (vma_t*) kmalloc(sizeof(vma_t));
				if (BUILTIN_EXPECT(!child, 0))
					break;

				(*child)->start = parent->start;
				(*child)->end = parent->end;
				(*child)->type = parent->type;
				(*child)->prev = tmp;
				(*child)->next = NULL;

				parent = parent->next;
				tmp = *child;
				child = &((*child)->next);
			}

			mailbox_wait_msg_init(&task_table[i].inbox);
			memset(task_table[i].outbox, 0x00, sizeof(mailbox_wait_msg_t*)*MAX_TASKS);
			task_table[i].outbox[per_core(current_task)->id] = &per_core(current_task)->inbox; 
			task_table[i].fpu_used = 0x00;

			ret = arch_fork(task_table+i);

			if (parent_task != per_core(current_task)) {
				// Oh, the current task is the new child task!
				// Leave the function without releasing the locks
				// because the locks are already released 
				// by the parent task!
				return 0; 
			}

			if (!ret) {
				task_table[i].status = TASK_READY;
				ret = i;
			}
			break;
		}
	}

create_task_out:
	spinlock_irqsave_unlock(&table_lock);
	spinlock_unlock(&per_core(current_task)->vma_lock);

	return ret;
}

int create_kernel_task(tid_t* id, entry_point_t ep, void* arg)
{
	return create_task(id, ep, arg);
}

#define MAX_ARGS	(PAGE_SIZE - 2*sizeof(int) - sizeof(vfs_node_t*))

/** @brief Structure which keeps all 
 * relevant data for a new task to start */
typedef struct {
	/// Points to the node with the executable in the file system
	vfs_node_t* node;
	/// Argument count
	int argc;
	/// Environment var count
	int envc;
	/// Buffer for env and argv values
	char buffer[MAX_ARGS];
} load_args_t;

/** @brief Internally used function to load tasks with a load_args_t structure
 * keeping all the information needed to launch.
 *
 * This is where the serious loading action is done.
 */
static int load_task(load_args_t* largs)
{
	uint32_t i, offset, idx;
	uint32_t addr, npages, flags, stack = 0;
	elf_header_t header;
	elf_program_header_t prog_header;
	//elf_section_header_t sec_header;
	vfs_node_t* node;

	if (!largs)
		return -EINVAL;

	node = largs->node;
	if (!node)
		return -EINVAL;

	read_fs(node, (uint8_t*)&header, sizeof(elf_header_t), 0);
	if (BUILTIN_EXPECT(header.ident.magic != ELF_MAGIC, 0))
		goto invalid;

	if (BUILTIN_EXPECT(header.type != ELF_ET_EXEC, 0))
		goto invalid;

	if (BUILTIN_EXPECT(header.machine != ELF_EM_386, 0))
		goto invalid;

	if (BUILTIN_EXPECT(header.ident._class != ELF_CLASS_32, 0))
		goto invalid;

	if (BUILTIN_EXPECT(header.ident.data != ELF_DATA_2LSB, 0))
		goto invalid;

	if (header.entry <= KERNEL_SPACE)
		goto invalid;

        // interpret program header table
	for (i=0; i<header.ph_entry_count; i++) {
		if (read_fs(node, (uint8_t*)&prog_header, sizeof(elf_program_header_t), header.ph_offset+i*header.ph_entry_size) == 0) {
			kprintf("Could not read programm header!\n");
			continue;
		}

		switch(prog_header.type)
		{
		case  ELF_PT_LOAD:  // load program segment
			if (!prog_header.virt_addr)
				continue;

			npages = (prog_header.mem_size / PAGE_SIZE);
			if (prog_header.mem_size % PAGE_SIZE)
				npages++;

			addr = get_pages(npages);

			flags = MAP_USER_SPACE;
			if (prog_header.flags & PF_X)
				flags |= MAP_CODE;

			// map page frames in the address space of the current task
			if (!map_region(prog_header.virt_addr, addr, npages, flags))
				kprintf("Could not map 0x%x at 0x%x\n", addr, prog_header.virt_addr);

			// clear pages
			memset((void*) prog_header.virt_addr, 0, npages*PAGE_SIZE);

			// load program
			read_fs(node, (uint8_t*)prog_header.virt_addr, prog_header.file_size, prog_header.offset);

			flags = VMA_CACHEABLE;
			if (prog_header.flags & PF_R)
				flags |= VMA_READ;
			if (prog_header.flags & PF_W)
				flags |= VMA_WRITE;
			if (prog_header.flags & PF_X)
				flags |= VMA_EXECUTE;
			vma_add(per_core(current_task), prog_header.virt_addr, prog_header.virt_addr+npages*PAGE_SIZE-1, flags);

			if (!(prog_header.flags & PF_W)) 
				change_page_permissions(prog_header.virt_addr, prog_header.virt_addr+npages*PAGE_SIZE-1, flags);
			break;

		case ELF_PT_GNU_STACK: // Indicates stack executability
			// create user-level stack
			npages = DEFAULT_STACK_SIZE / PAGE_SIZE;
			if (DEFAULT_STACK_SIZE % PAGE_SIZE)
				npages++;

			addr = get_pages(npages); 
			stack = header.entry*2; // virtual address of the stack

			if (!map_region(stack, addr, npages, MAP_USER_SPACE)) {
				kprintf("Could not map stack at 0x%x\n", stack);
				return -ENOMEM;
			}
			memset((void*) stack, 0, npages*PAGE_SIZE);

			// create vma regions for the user-level stack
			flags = VMA_CACHEABLE;
			if (prog_header.flags & PF_R)
				flags |= VMA_READ;
			if (prog_header.flags & PF_W)
				flags |= VMA_WRITE;
			if (prog_header.flags & PF_X)
				flags |= VMA_EXECUTE;
			vma_add(per_core(current_task), stack, stack+npages*PAGE_SIZE-1, flags);
			break;
		 }
	}

#if 0
	// interpret section header table
	for (i=0; i<header.sh_entry_count; i++) {
		if (read_fs(node, (uint8_t*)&sec_header, sizeof(elf_section_header_t), header.sh_offset+i*header.sh_entry_size) == 0) {
			kprintf("Could not read section header!\n");
			continue;
		}

		// TODO: interpret section header
	}
#endif

	if (BUILTIN_EXPECT(!stack, 0)) {
		kprintf("Stack is missing!\n");
		return -ENOMEM;
	}

	// push strings on the stack
	offset = DEFAULT_STACK_SIZE-8;
	memset((void*) (stack+offset), 0, 4);
	offset -= MAX_ARGS;
	memcpy((void*) (stack+offset), largs->buffer, MAX_ARGS);
	idx = offset;

	// push argv on the stack
	offset -= largs->argc * sizeof(char*);
	for(i=0; i<largs->argc; i++) {
		((char**) (stack+offset))[i] = (char*) (stack+idx);
		
		while(((char*) stack)[idx] != '\0')
			idx++;
		idx++;
	}

	// push env on the stack
	offset -= (largs->envc+1) * sizeof(char*);
	for(i=0; i<largs->envc; i++) {
		((char**) (stack+offset))[i] = (char*) (stack+idx);

		while(((char*) stack)[idx] != '\0')
			idx++;
		idx++;
	}
	((char**) (stack+offset))[largs->envc] = NULL;

	// push pointer to env
	offset -= sizeof(char**);
	if (!(largs->envc))
		*((char***) (stack+offset)) = NULL;
	else
		*((char***) (stack+offset)) = (char**) (stack + offset + sizeof(char**));

	// push pointer to argv
	offset -= sizeof(char**);
	*((char***) (stack+offset)) = (char**) (stack + offset + 2*sizeof(char**) + (largs->envc+1) * sizeof(char*));

	// push argc on the stack
	offset -= sizeof(int);
	*((int*) (stack+offset)) = largs->argc;

	kfree(largs, sizeof(load_args_t));

	jump_to_user_code(header.entry, stack+offset);

	return 0;

invalid:
	kprintf("Invalid executable!\n");

	return -EINVAL;
}

/** @brief This call is used to adapt create_task calls 
 * which want to have a start function and argument list */
static int STDCALL user_entry(void* arg)
{
	return load_task((load_args_t*) arg);
}

/** @brief Luxus-edition of create_user_task functions. Just call with an exe name
 *
 * @param id Pointer to the tid_t structure which shall be filles
 * @param fname Executable's path and filename
 * @param argv Arguments list
 * @return
 * - 0 on success
 * - -ENOMEM (-12) or -EINVAL (-22) on failure
 */
int create_user_task(tid_t* id, const char* fname, char** argv)
{
	vfs_node_t* node;
	int argc = 0;
	size_t i, buffer_size = 0;
	load_args_t* load_args = NULL;
	char *dest, *src;

	node = findnode_fs((char*) fname);
	if (!node || !(node->type == FS_FILE))
		return -EINVAL;

	// determine buffer size of argv
	if (argv) {
		while (argv[argc]) {
			buffer_size += (strlen(argv[argc]) + 1);
			argc++;
		}
	}

	if (argc <= 0)
		return -EINVAL;
	if (buffer_size >= MAX_ARGS)
		return -EINVAL;

	load_args = kmalloc(sizeof(load_args_t));
	if (BUILTIN_EXPECT(!load_args, 0))
		return -ENOMEM;
	load_args->node = node;
	load_args->argc = argc;
	load_args->envc = 0;
	dest = load_args->buffer;
	for (i=0; i<argc; i++) {
		src = argv[i];
		while ((*dest++ = *src++) != 0);
	}

	return create_task(id, user_entry, load_args);
}

/** @brief Used by the execve-Systemcall */
int sys_execve(const char* fname, char** argv, char** env)
{
	vfs_node_t* node;
	vma_t* tmp;
	size_t i, buffer_size = 0;
	load_args_t* load_args = NULL;
	char *dest, *src;
	int ret, argc = 0;
	int envc = 0;

	node = findnode_fs((char*) fname);
	if (!node || !(node->type == FS_FILE))
		return -EINVAL;

	// determine total buffer size of argv and env
	if (argv) {
		while (argv[argc]) {
			buffer_size += (strlen(argv[argc]) + 1);
			argc++;
		}
	}

	if (env) {
		while (env[envc]) {
			buffer_size += (strlen(env[envc]) + 1);
			envc++;
		}
	}

	if (argc <= 0)
		return -EINVAL;
	if (buffer_size >= MAX_ARGS)
		return -EINVAL;

	load_args = kmalloc(sizeof(load_args_t));
	if (BUILTIN_EXPECT(!load_args, 0))
		return -ENOMEM;

	load_args->node = node;
	load_args->argc = argc;
	load_args->envc = envc;
	dest = load_args->buffer;
	for (i=0; i<argc; i++) {
		src = argv[i];
		while ((*dest++ = *src++) != 0);
	}
	for (i=0; i<envc; i++) {
		src = env[i];
		while ((*dest++ = *src++) != 0);
	}

	spinlock_lock(&(per_core(current_task)->vma_lock));

	// remove old program
	while((tmp = per_core(current_task)->vma_list) != NULL) {
		kfree((void*) tmp->start, tmp->end - tmp->start + 1);
		per_core(current_task)->vma_list = tmp->next;
		kfree((void*) tmp, sizeof(vma_t));
	}

	spinlock_unlock(&(per_core(current_task)->vma_lock));

	/*
	 * we use a trap gate to enter the kernel
	 * => eflags are not changed
	 * => interrupts are enabled
	 * => we could directly load the new task
	 */

	ret = load_task(load_args);

	kfree(load_args, sizeof(load_args_t));

	return ret;
}

/** @brief Called by tasks which are waiting for another task's
 * return value. */
tid_t wait(int32_t* result) 
{
	wait_msg_t tmp = { -1, -1};

	/* 
	 * idle tasks are not allowed to wait for another task 
	 * they should always run...
	 */
	if (BUILTIN_EXPECT(per_core(current_task)->status == TASK_IDLE, 0))
		return -EINVAL;

	mailbox_wait_msg_fetch(&per_core(current_task)->inbox, &tmp);

	if (result)
		*result = tmp.result;

	return tmp.id;
}

/** @brief Wakeup a blocked task
 * @param id The task's tid_t structure
 * @return
 * - 0 on success
 * - -EINVAL (-22) on failure
 */
int wakeup_task(tid_t id)
{
	int ret = -EINVAL;

	/* avoid nested locking */
	spinlock_irqsave_lock(&table_lock);

	if (task_table[id].status != TASK_BLOCKED) {
		kprintf("Task %d is not blocked!\n", id);
	} else {
		task_table[id].status = TASK_READY;
		ret = 0;
	}

	spinlock_irqsave_unlock(&table_lock);

	return ret;
}

/** @brief Block a running or ready task.
 * @param id The task's tid_t structure
 * @return
 * - 0 on success
 * - -EINVAL (-22) on failure
 */
int block_task(tid_t id)
{
	int ret = -EINVAL;

	spinlock_irqsave_lock(&table_lock);

        if ((task_table[id].status == TASK_RUNNING) || (task_table[id].status == TASK_READY)) {
		task_table[id].status = TASK_BLOCKED;
		ret = 0;
	} else kprintf("Unable to block task %d!\n", id);

        spinlock_irqsave_unlock(&table_lock);

        return ret;
}

/** @brief _The_ scheduler procedure
 *
 * Manages scheduling - right now this is just a round robin scheduler.
 */
void scheduler(void) 
{
	unsigned int i;
	unsigned int new_id;

#if MAX_CORES > 1
	spinlock_irqsave_lock(&table_lock);
#endif

	/* signalize that this task could be reused */
	if (per_core(current_task)->status == TASK_FINISHED)
		per_core(current_task)->status = TASK_INVALID; 

	for(i=1, new_id=(per_core(current_task)->id + 1) % MAX_TASKS; 
		i<MAX_TASKS; i++, new_id=(new_id+1) % MAX_TASKS) 
	{
		if (task_table[new_id].status == TASK_READY) {
			if (per_core(current_task)->status == TASK_RUNNING)
				per_core(current_task)->status = TASK_READY;
			task_table[new_id].status = TASK_RUNNING;

			if (per_core(current_task)->fpu_used)
				save_fpu_state(&(per_core(current_task)->fpu));	
			per_core(current_task) = task_table+new_id;
			goto get_task_out;
		}
	}

	if ((per_core(current_task)->status == TASK_RUNNING) || (per_core(current_task)->status == TASK_IDLE))
		goto get_task_out;

	/* 
	 * we switch to the idle task, if the current task terminates 
	 * and no other is ready
	 */
	for(i=0; i<MAX_TASKS; i++) {
		if (task_table[i].status == TASK_IDLE) {
			per_core(current_task) = task_table+i;
			goto get_task_out;
		}
	}

get_task_out:
#if MAX_CORES > 1
	spinlock_irqsave_unlock(&table_lock);
#else
	return;
#endif
}