metalsvm/arch/x86/mm/page.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. 
 */

#include <metalsvm/stddef.h>
#include <metalsvm/stdio.h>
#include <metalsvm/stdlib.h>
#include <metalsvm/mmu.h>
#include <metalsvm/string.h>
#include <metalsvm/page.h>
#include <metalsvm/spinlock.h>
#include <metalsvm/processor.h>
#include <metalsvm/tasks.h>
#include <metalsvm/errno.h>
#include <asm/irq.h>
#include <asm/multiboot.h>

/*
 * Virtual Memory Layout of the standard configuration
 * (1 GB kernel space)
 *
 * 0x00000000 - 0x000FFFFF: reserved for IO devices
 * 0x00100000 - 0x0DEADFFF: Kernel (size depends on the configuration)
 * 0xDEAE0000 - 0x3FFFEFFF: Kernel heap
 * 0x3FFFF000 - 0x3FFFFFFF: Page Table are mapped in this region
 *			   (The first 256 entries belongs to kernel space)
 */

/* 
 * Note that linker symbols are not variables, they have no memory allocated for
 * maintaining a value, rather their address is their value.
 */
extern const void kernel_start;
extern const void kernel_end;

// boot task's page directory and page directory lock
static page_dir_t boot_pgd = {{[0 ... 1023] = 0}};
static spinlock_t boot_pgd_lock = SPINLOCK_INIT;
static int paging_enabled = 0;

int get_kernel_pgd(task_t* task)
{
	if (BUILTIN_EXPECT(!task, 0))
		return -EINVAL;

	task->pgd = &boot_pgd;
	task->pgd_lock = &boot_pgd_lock;

	return 0;
}

size_t virt_to_phys(task_t* task, size_t viraddr)
{
	uint32_t index1, index2;
	page_table_t* pgt;
	size_t ret = 0;

	if (BUILTIN_EXPECT(!task || !task->pgd || !paging_enabled, 0))
		return 0;

	index1 = viraddr >> 22;
	index2 = (viraddr >> 12) & 0x3FF;

	if (!(task->pgd->entries[index1] & 0xFFFFF000))
		goto out;

	pgt = (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*4) & 0xFFFFF000);
	if (!pgt || !(pgt->entries[index2]))
		goto out;

	ret = pgt->entries[index2] & 0xFFFFF000; 	// determine page frame
	ret = ret | (viraddr & 0xFFF);			// add page offset
out:
	//kprintf("vir %p to phy %p\n", viraddr, ret);
	
	return ret;
}

size_t map_region(task_t* task, size_t viraddr, size_t phyaddr, uint32_t npages, uint32_t flags)
{
	page_table_t* pgt;
	size_t index, i;
	size_t ret;

	if (BUILTIN_EXPECT(!task || !task->pgd || !phyaddr, 0))
		return 0;

	if (BUILTIN_EXPECT(!paging_enabled && (viraddr != phyaddr), 0))
		return 0;

	if (!(flags & MAP_KERNEL_SPACE))
		return 0;

	if (!viraddr) {
		viraddr = vm_alloc(task, npages, flags);
		if (BUILTIN_EXPECT(!viraddr, 0)) {
			kputs("map_adress: found no valid virtual address\n");
			return 0;
		}
	}

	ret = viraddr;
	//kprintf("map %d pages from %p to %p\n", npages, phyaddr, ret);
	for(i=0; i<npages; i++, viraddr+=PAGE_SIZE, phyaddr+=PAGE_SIZE) {
		index = viraddr >> 22;

		if (!(task->pgd->entries[index])) {
			page_table_t* pgt_container;

			pgt = (page_table_t*) get_pages(1);
			if (BUILTIN_EXPECT(!pgt, 0)) {
				spinlock_unlock(task->pgd_lock);
				kputs("map_address: out of memory\n");
				return 0;
			}

			// set the new page table into the directory
			task->pgd->entries[index] = (uint32_t)pgt|KERN_TABLE;

			// if paging is already enabled, we need to use the virtual address
			if (paging_enabled)
				// we already know the virtual address of the "page table container"
				// (see file header)
				pgt_container = (page_table_t*) ((KERNEL_SPACE - PAGE_SIZE) & 0xFFFFF000);
			else
				pgt_container = (page_table_t*) (task->pgd->entries[(KERNEL_SPACE - PAGE_SIZE) >> 22] & 0xFFFFF000);

			if (BUILTIN_EXPECT(!pgt_container, 0)) {
				spinlock_unlock(task->pgd_lock);
                                kputs("map_address: internal error\n");
                                return 0;
			}

			// map the new table into the address space of the kernel space
		        pgt_container->entries[index] = ((size_t) pgt)|KERN_PAGE;

			// clear the page table
			if (paging_enabled)
				memset((void*) (KERNEL_SPACE - 1024*PAGE_SIZE + index*4), 0, PAGE_SIZE);
			else
				memset(pgt, 0, PAGE_SIZE);
		} else pgt = (page_table_t*) (task->pgd->entries[index] & 0xFFFFF000);

		/* convert physical address to virtual */
		if (paging_enabled)
			pgt = (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index*4) & 0xFFFFF000);

		index = (viraddr >> 12) & 0x3FF;
		if (BUILTIN_EXPECT(pgt->entries[index], 0)) {
			spinlock_unlock(task->pgd_lock);
			kprintf("0x%x is already maped\n");
			return 0;
		}

		pgt->entries[index] = KERN_PAGE|(phyaddr & 0xFFFFF000);
		if (flags & MAP_NO_CACHE)
			pgt->entries[index] |= PG_PCD;

		tlb_flush_one_page(viraddr);
	}
	
	return ret;
}

/* 
 * Use the first fit algorithm to find a valid address range
 *
 * TODO: O(n) => bad performance, we need a better approach
 */
size_t vm_alloc(task_t* task, uint32_t npages, uint32_t flags)
{
	uint32_t index1, index2, j;
	size_t	viraddr, i;
	size_t start, end;
	page_table_t* pgt;

	if (BUILTIN_EXPECT(!task || !task->pgd || !paging_enabled, 0))
		return 0;

	if (flags & MAP_KERNEL_SPACE) {
		start = (((size_t) &kernel_end) + PAGE_SIZE) & 0xFFFFF000;
		end = (KERNEL_SPACE - 2*PAGE_SIZE) & 0xFFFFF000; // we need 1 PAGE for our PGTs
	} else {
		start = KERNEL_SPACE & 0xFFFFF000;
		end = 0xFFFFF000;
	}

	if (BUILTIN_EXPECT(!npages, 0))
		return 0;

	viraddr = i = start;
	j = 0;
	do {
		index1 = i >> 22;
		index2 =  (i >> 12) & 0x3FF;

		pgt = (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*4) & 0xFFFFF000);
		if (!pgt || !(pgt->entries[index2])) {
			i+=PAGE_SIZE;
			j++;
		} else {
			// restart search
			j = 0;
			viraddr = i + PAGE_SIZE;
			i = i + PAGE_SIZE;
		}
	} while((j < npages) && (i<=end));

	if ((j >= npages) && (viraddr < end))
		return viraddr;

	return 0;
}

int vm_free(task_t* task, size_t viraddr, uint32_t npages)
{
	uint32_t i;
	uint32_t index1, index2;
	page_table_t* pgt;

	if (BUILTIN_EXPECT(!task || !task->pgd || !paging_enabled, 0))
		return -EINVAL;

	for(i=0; i<npages; i++, viraddr+=PAGE_SIZE)
	{
		index1 = viraddr >> 22;
		index2 =  (viraddr >> 12) & 0x3FF;

                pgt = (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*4) & 0xFFFFF000);
		if (!pgt)
			continue;
		pgt->entries[index2] = 0;
	}

	return 0;
}

int print_paging_tree(size_t viraddr)
{
	uint32_t index1, index2;
	page_dir_t* pgd = NULL;
	page_table_t* pgt = NULL;

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

	index1 = viraddr >> 22;
	index2 = (viraddr >> 12) & 0x3FF;

	kprintf("Paging dump of address 0x%x\n", viraddr);
	pgd = per_core(current_task)->pgd;
	kprintf("\tPage directory entry %u: ", index1);
	if (pgd) {
		kprintf("0x%0x\n", pgd->entries[index1]);
		pgt = (page_table_t*) (pgd->entries[index1] & 0xFFFFF000);
	} else 
		kputs("invalid page directory\n");

	/* convert physical address to virtual */
	if (paging_enabled && pgt)
		pgt = (page_table_t*) (KERNEL_SPACE - 1024*PAGE_SIZE + index1*4);

	kprintf("\tPage table entry %u: ", index2);
	if (pgt)
		kprintf("0x%x\n", pgt->entries[index2]);
	else 
		kputs("invalid page table\n");

	return 0;
}

static void pagefault_handler(struct state *s)
{
	kprintf("PAGE FAULT: Task %u got page fault at irq %u\n", per_core(current_task)->id, s->int_no);
	kprintf("Register state: eax = 0x%x, ebx = 0x%x, ecx = 0x%x, edx = 0x%x, edi = 0x%x, esi = 0x%x, ebp = 0x%x, esp = 0x%x\n", 
		s->eax, s->ebx, s->ecx, s->edx, s->edi, s->esi, s->ebp, s->esp);

	abort();
}

int arch_paging_init(void)
{
	uint32_t i, npages, index1, index2;
	page_table_t* pgt;
	size_t viraddr;

	// uninstall default handler and install our own
	irq_uninstall_handler(14);
	irq_install_handler(14, pagefault_handler);

	// Create a page table to reference to the other page tables
	pgt = (page_table_t*) get_pages(1);
	if (!pgt) {
		kputs("arch_paging_init: Not enough memory!\n");
		return -ENOMEM;
	}
	memset(pgt, 0, PAGE_SIZE);
	
	// map this table at the end of the kernel space
	viraddr = KERNEL_SPACE - PAGE_SIZE;
	index1 = viraddr >> 22;
	index2 = (viraddr >> 12) & 0x3FF;

	// now, we create a self reference
	per_core(current_task)->pgd->entries[index1] = (((size_t) pgt) & 0xFFFFF000)|USER_TABLE;
	pgt->entries[index2] = ((size_t) pgt & 0xFFFFF000)|KERN_PAGE;

	/* 
	 * Set the page table and page directory entries for the kernel. We map the kernel's physical address 
	 * to the same virtual address.
	 */
	npages = ((size_t) &kernel_end - (size_t) &kernel_start) / PAGE_SIZE;
	if ((size_t)&kernel_end % PAGE_SIZE)
		npages++;
	map_region(per_core(current_task), (size_t)&kernel_start, (size_t)&kernel_start, npages, MAP_KERNEL_SPACE);

#ifdef CONFIG_VGA
	// map the video memory into the kernel space
	map_region(per_core(current_task), VIDEO_MEM_ADDR, VIDEO_MEM_ADDR, 1, MAP_KERNEL_SPACE|MAP_NO_CACHE);
#endif

#ifdef CONFIG_MULTIBOOT
	/*
	 * of course, mb_info has to map into the kernel space
	 */
	if (mb_info)
		map_region(per_core(current_task), (size_t) mb_info, (size_t) mb_info, 1, MAP_KERNEL_SPACE);

	/*
	 * Map reserved memory regions into the kernel space
	 */
	if (mb_info && (mb_info->flags & (1 << 6))) {
		multiboot_memory_map_t* mmap = (multiboot_memory_map_t*) mb_info->mmap_addr;
		multiboot_memory_map_t* mmap_end = (void*) ((size_t) mb_info->mmap_addr + mb_info->mmap_length);

		while (mmap < mmap_end) {
			if (mmap->type != MULTIBOOT_MEMORY_AVAILABLE) {
				npages = mmap->len / PAGE_SIZE;
				if ((mmap->addr+mmap->len) % PAGE_SIZE)
					npages++;
				map_region(per_core(current_task), mmap->addr, mmap->addr, npages, MAP_KERNEL_SPACE|MAP_NO_CACHE);
			}
			mmap++;
		}
	}

	/* 
	 * Modules like the init ram disk are already loaded.
	 * Therefore, we map these moduels into the kernel space.
	 */
	if (mb_info && (mb_info->flags & (1 << 3))) {
		multiboot_module_t* mmodule = (multiboot_module_t*) mb_info->mods_addr;

		for(i=0; i<mb_info->mods_count; i++, mmodule++) {
			// map physical address to the same virtual address
			npages = (mmodule->mod_end - mmodule->mod_start) / PAGE_SIZE;
			if (mmodule->mod_end % PAGE_SIZE)
				npages++;
			map_region(per_core(current_task), (size_t) (mmodule->mod_start), (size_t) (mmodule->mod_start), npages, MAP_KERNEL_SPACE);
		}
	}
#endif

	/* enable paging */
	write_cr3((uint32_t) &boot_pgd);
	i = read_cr0();
	i = i | (1 << 31);
	write_cr0(i);
	paging_enabled = 1;

	/*
	 * we turned on paging
	 * => now, we are able to register our task for Task State Switching
	 */
	register_task(per_core(current_task));

	return 0;
}