metalsvm/arch/x86/mm/page64.c

/* 
 * Copyright 2012 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/vma.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>
#include <asm/apic.h>

/*
 * Virtual Memory Layout of the standard configuration
 * (1 GB kernel space)
 *
 * 0x000000000000 - 0x0000000FFFFF: reserved for IO devices (16MB)
 * 0x000000100000 - 0x00000DEADFFF: Kernel (size depends on the configuration) (221MB)
 * 0x00000DEAE000 - 0x00003FFFFFFF: Kernel heap
 * 0xFF8000000000 - 0xFFFFFFFFFFFF: Paging structures are mapped in this region (1GB)
 */

/* 
 * 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
extern page_map_t boot_pml4;
static spinlock_t kslock = SPINLOCK_INIT;
static int paging_enabled = 0;

/*static page_map_t boot_pml4 = {{[0 ... MAP_ENTRIES-1] = 0}};
static page_map_t boot_pdpt = {{[0 ... MAP_ENTRIES-1] = 0}};
static page_map_t boot_pgd = {{[0 ... MAP_ENTRIES-1] = 0}};
static page_map_t boot_pgt = {{[0 ... MAP_ENTRIES-1] = 0}};*/

page_map_t* get_boot_page_map(void)
{
	return &boot_pml4;
}

int create_page_map(task_t* task, int copy)
{
	// TODO: Currently, we support only kernel tasks
	// => all tasks are able to use the same pgd

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

	task->page_map = get_boot_page_map();

	return 0;
}

int drop_page_map(void)
{
#if 0
	page_map_t* pgd = per_core(current_task)->page_map;
	size_t phy_pgd = virt_to_phys((size_t) pgd);
	task_t* task = per_core(current_task);
	uint32_t i;

	if (BUILTIN_EXPECT(pgd == &boot_pgd, 0))
		return -EINVAL;

	spinlock_lock(&task->page_lock);
	
	for(i=0; i<1024; i++) {
		if (pgd->entries[i] & PG_USER) {
			put_page(pgd->entries[i] & PAGE_MASK);
			pgd->entries[i] = 0;
		}
	}

	// freeing the page directory
	put_page(phy_pgd);

	task->page_map = NULL;

	spinlock_unlock(&task->page_lock);
#endif

	return 0;
}

size_t virt_to_phys(size_t viraddr)
{
	task_t* task = per_core(current_task);
	page_map_t* pdpt, * pgd , * pgt;
	uint16_t index_pml4 = (viraddr >> 39) & 0x1FF;
	uint16_t index_pdpt = (viraddr >> 30) & 0x1FF;
	uint16_t index_pgd = (viraddr >> 21) & 0x1FF;
	uint16_t index_pgt = (viraddr >> 12) & 0x1FF;
	size_t ret = 0;

	if (!paging_enabled)
		return viraddr;

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

	spinlock_irqsave_lock(&task->page_lock);

	// TODO: Currently, we allocate pages only in kernel space.
	// => physical address of the page table is identical of the virtual address
	pdpt = (page_map_t*) (task->page_map->entries[index_pml4] & PAGE_MASK);
	if (!pdpt)
		goto out;

	pgd = (page_map_t*) (pdpt->entries[index_pdpt] & PAGE_MASK);
	if (!pgd)
		goto out;

	pgt = (page_map_t*) (pgd->entries[index_pgd] & PAGE_MASK);
	if (!pgt)
		goto out;

	ret = (size_t) (pgt->entries[index_pgt] & PAGE_MASK);
	if (!ret)
		goto out;

	ret = ret | (viraddr & 0xFFF);			// add page offset
out:
	//kprintf("vir %p to phy %p\n", viraddr, ret);

	spinlock_irqsave_unlock(&task->page_lock);

	return ret;
}

size_t map_region(size_t viraddr, size_t phyaddr, uint32_t npages, uint32_t flags)
{
	task_t* task = per_core(current_task);
	page_map_t* pdpt, * pgd, * pgt;
	uint16_t index_pml4, index_pdpt;
	uint16_t index_pgd, index_pgt;
	size_t i, ret;

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

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

	if (flags & MAP_KERNEL_SPACE)
		spinlock_lock(&kslock);
	else 
		spinlock_irqsave_lock(&task->page_lock);

	if (!viraddr) {
		viraddr = vm_alloc(npages, flags);
		if (BUILTIN_EXPECT(!viraddr, 0)) {
			kputs("map_region: found no valid virtual address\n");
			ret = 0;
			goto out;
		}
	}

	ret = viraddr;
	for(i=0; i<npages; i++, viraddr+=PAGE_SIZE, phyaddr+=PAGE_SIZE) {
		index_pml4 = (viraddr >> 39) & 0x1FF;
		index_pdpt = (viraddr >> 30) & 0x1FF;
		index_pgd = (viraddr >> 21) & 0x1FF;
		index_pgt = (viraddr >> 12) & 0x1FF;

		pdpt = (page_map_t*) (task->page_map->entries[index_pml4] & PAGE_MASK);
		if (!pgt) {
			kputs("map_region: out of memory\n");
			ret = 0;
			goto out;
		}

		pgd = (page_map_t*) (pdpt->entries[index_pdpt] & PAGE_MASK);
		if (!pgd) {
			kputs("map_region: out of memory\n");
			ret = 0;
			goto out;
		}

		pgt = (page_map_t*) (pgd->entries[index_pgd] & PAGE_MASK);
		if (!pgt) {
			kputs("map_region: out of memory\n");
			ret = 0;
			goto out;
		}

		/* convert physical address to virtual */
		// Currently, we allocate pages only in kernel space.
		// => physical address of the page table is identical of the virtual address
		//if (paging_enabled)
		//	pgt = (page_map_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index*PAGE_SIZE) & PAGE_MASK);

		if (pgt->entries[index_pgt] && !(flags & MAP_REMAP)) {
			kprintf("0x%x is already mapped\n", viraddr);
			ret = 0;
			goto out;
		}

		if (flags & MAP_USER_SPACE)
			pgt->entries[index_pgt] = USER_PAGE|(phyaddr & PAGE_MASK);
		else
			pgt->entries[index_pgt] = KERN_PAGE|(phyaddr & PAGE_MASK);

		if (flags & MAP_NO_CACHE)
			pgt->entries[index_pgt] |= PG_PCD;

		if (flags & MAP_NO_ACCESS)
			pgt->entries[index_pgt] &= ~PG_PRESENT;

		if (flags & MAP_WT)
			pgt->entries[index_pgt] |= PG_PWT;

		if (flags & MAP_USER_SPACE)
			atomic_int32_inc(&task->user_usage);

		if (flags & MAP_NO_EXECUTION)
			pgt->entries[index_pgt] |= PG_XD;

		tlb_flush_one_page(viraddr);
	}

out:
	if (flags & MAP_KERNEL_SPACE)
		spinlock_unlock(&kslock);
	else
		spinlock_irqsave_unlock(&task->page_lock);

	return ret;
}

int change_page_permissions(size_t start, size_t end, uint32_t flags)
{
#if 0
	uint32_t index1, index2, newflags;
	size_t viraddr = start & PAGE_MASK;
	size_t phyaddr;
	page_map_t* pgt;
	page_map_t* pgd;
	task_t* task = per_core(current_task);

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

	pgd = per_core(current_task)->page_map;
	if (BUILTIN_EXPECT(!pgd, 0))
		return -EINVAL;

	spinlock_lock(&task->page_lock);

	while (viraddr < end)
	{
		index1 = viraddr >> 22;
		index2 = (viraddr >> 12) & 0x3FF;

		while ((viraddr < end) && (index2 < 1024)) {
			pgt = (page_map_t*) (page_map_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
			if (pgt && pgt->entries[index2]) {
				phyaddr = pgt->entries[index2] & PAGE_MASK;
				newflags = pgt->entries[index2] & 0xFFF;  // get old flags

				if (!(newflags & PG_SVM_INIT)) {
					if ((newflags & PG_SVM_STRONG) && !(newflags & PG_PRESENT) && (flags & (VMA_READ|VMA_WRITE) && !(flags & VMA_NOACCESS)))
						newflags |= PG_PRESENT;
					else if ((newflags & PG_SVM_STRONG) && (newflags & PG_PRESENT) && (flags & VMA_NOACCESS))
						newflags &= ~PG_PRESENT;
				}

				// update flags
				if (!(flags & VMA_WRITE)) {
					newflags &= ~PG_RW;
				} else  {
					newflags |= PG_RW;
				}

				pgt->entries[index2] = (newflags & 0xFFF) | (phyaddr & PAGE_MASK);

				tlb_flush_one_page(viraddr);
			}

			index2++;
			viraddr += PAGE_SIZE;
		}
	}

	spinlock_unlock(&task->page_lock);
#endif

	return -EINVAL;
}

/* 
 * 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(uint32_t npages, uint32_t flags)
{
	task_t* task = per_core(current_task);
	size_t viraddr, i, j, ret = 0;
	size_t start, end;
	page_map_t* pdpt, * pgd, * pgt;
	uint16_t index_pml4, index_pdpt;
	uint16_t index_pgd, index_pgt;

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

	if (flags & MAP_KERNEL_SPACE) {
		start = (((size_t) &kernel_end) + 10*PAGE_SIZE) & PAGE_MASK;
		end = (KERNEL_SPACE - PAGE_SIZE) & PAGE_MASK; 
	} else {
		start = KERNEL_SPACE & PAGE_MASK;
		end = PAGE_MASK;
	}

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

	if (flags & MAP_KERNEL_SPACE)
		spinlock_lock(&kslock);
	else
		spinlock_irqsave_lock(&task->page_lock);

	viraddr = i = start;
	j = 0;
	do {
		index_pml4 = (viraddr >> 39) & 0x1FF;
		index_pdpt = (viraddr >> 30) & 0x1FF;
		index_pgd = (viraddr >> 21) & 0x1FF;
		index_pgt = (viraddr >> 12) & 0x1FF;

		// Currently, we allocate pages only in kernel space.
		// => physical address of the page table is identical of the virtual address
		pdpt = (page_map_t*) (task->page_map->entries[index_pml4] & PAGE_MASK);
		if (!pdpt) {
			i += (size_t)MAP_ENTRIES*MAP_ENTRIES*MAP_ENTRIES*PAGE_SIZE;
			j += MAP_ENTRIES*MAP_ENTRIES*MAP_ENTRIES;
			continue;
		}

		pgd = (page_map_t*) (pdpt->entries[index_pdpt] & PAGE_MASK);
		if (!pgd) {
			i += MAP_ENTRIES*MAP_ENTRIES*PAGE_SIZE;
			j += MAP_ENTRIES*MAP_ENTRIES;
			continue;
		}

		pgt = (page_map_t*) (pgd->entries[index_pgd] & PAGE_MASK);
                if (!pgt) {
                        i += MAP_ENTRIES*PAGE_SIZE;
                        j += MAP_ENTRIES;
                        continue;
                }

		if (!(pgt->entries[index_pgt])) {
			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))
		ret = viraddr;

	if (flags & MAP_KERNEL_SPACE)
		spinlock_unlock(&kslock);
	else
		spinlock_irqsave_unlock(&task->page_lock);

	return ret;
}

int unmap_region(size_t viraddr, uint32_t npages)
{
	task_t* task = per_core(current_task);
	page_map_t* pdpt, * pgd, * pgt;
	size_t i;
	uint16_t index_pml4, index_pdpt;
	uint16_t index_pgd, index_pgt;

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

	if (viraddr <= KERNEL_SPACE)
		spinlock_lock(&kslock);
	else
		spinlock_irqsave_lock(&task->page_lock);

	i = 0;
	while(i<npages)
	{
		index_pml4 = (viraddr >> 39) & 0x1FF;
		index_pdpt = (viraddr >> 30) & 0x1FF;
		index_pgd = (viraddr >> 21) & 0x1FF;
		index_pgt = (viraddr >> 12) & 0x1FF;

		// Currently, we allocate pages only in kernel space.
		// => physical address of the page table is identical of the virtual address
		pdpt = (page_map_t*) (task->page_map->entries[index_pml4] & PAGE_MASK);
		if (!pdpt) {
			viraddr += (size_t) MAP_ENTRIES*MAP_ENTRIES*MAP_ENTRIES*PAGE_SIZE;
			i += MAP_ENTRIES*MAP_ENTRIES*MAP_ENTRIES;
			continue;
		}

		pgd = (page_map_t*) (pdpt->entries[index_pdpt] & PAGE_MASK);
		if (!pgd) {
			viraddr += MAP_ENTRIES*MAP_ENTRIES*PAGE_SIZE;
			i += MAP_ENTRIES*MAP_ENTRIES;
			continue;
		}

		pgt = (page_map_t*) (pgd->entries[index_pgd] & PAGE_MASK);
		if (!pgt) {
			viraddr += MAP_ENTRIES*PAGE_SIZE;
			i += MAP_ENTRIES;
                        continue;
                }

		if (pgt->entries[index_pgt])
			pgt->entries[index_pgt] &= ~PG_PRESENT;

		viraddr +=PAGE_SIZE;
		i++;

		if (viraddr > KERNEL_SPACE)
			atomic_int32_dec(&task->user_usage);

		tlb_flush_one_page(viraddr);
	}

	if (viraddr <= KERNEL_SPACE)
		spinlock_unlock(&kslock);
	else
		spinlock_irqsave_unlock(&task->page_lock);

	return 0;
}

int vm_free(size_t viraddr, uint32_t npages)
{
	task_t* task = per_core(current_task);
	page_map_t* pdpt, * pgd, * pgt;
	size_t i;
	uint16_t index_pml4, index_pdpt;
	uint16_t index_pgd, index_pgt;

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

	if (viraddr <= KERNEL_SPACE)
		spinlock_lock(&kslock);
	else
		spinlock_irqsave_lock(&task->page_lock);

	i = 0;
	while(i<npages)
	{
		index_pml4 = (viraddr >> 39) & 0x1FF;
		index_pdpt = (viraddr >> 30) & 0x1FF;
		index_pgd = (viraddr >> 21) & 0x1FF;
		index_pgt = (viraddr >> 12) & 0x1FF;

		// Currently, we allocate pages only in kernel space.
		// => physical address of the page table is identical of the virtual address
		pdpt = (page_map_t*) (task->page_map->entries[index_pml4] & PAGE_MASK);
		if (!pdpt) {
			viraddr += (size_t) MAP_ENTRIES*MAP_ENTRIES*MAP_ENTRIES*PAGE_SIZE;
			i += MAP_ENTRIES*MAP_ENTRIES*MAP_ENTRIES;
			continue;
		}

		pgd = (page_map_t*) (pdpt->entries[index_pdpt] & PAGE_MASK);
		if (!pgd) {
			viraddr += MAP_ENTRIES*MAP_ENTRIES*PAGE_SIZE;
			i += MAP_ENTRIES*MAP_ENTRIES;
			continue;
		}

		pgt = (page_map_t*) (pgd->entries[index_pgd] & PAGE_MASK);
		if (!pgt) {
			viraddr += MAP_ENTRIES*PAGE_SIZE;
			i += MAP_ENTRIES;
			continue;
		}

		if (pgt->entries[index_pgt])
			pgt->entries[index_pgt] = 0;

		viraddr +=PAGE_SIZE;
		i++;

		tlb_flush_one_page(viraddr);
	}

	if (viraddr <= KERNEL_SPACE)
		spinlock_unlock(&kslock);
	else
		spinlock_irqsave_unlock(&task->page_lock);

	return 0;
}

static void pagefault_handler(struct state *s)
{
	task_t* task = per_core(current_task);
	//page_map_t* pgd = task->page_map;
	//page_map_t* pgt = NULL;
	size_t viraddr = read_cr2();
	//size_t phyaddr;

#if 0
	if ((viraddr >= task->start_heap) && (viraddr <= task->end_heap) && (viraddr > KERNEL_SPACE)) {
		viraddr = viraddr & PAGE_MASK;

		phyaddr = get_page();
		if (BUILTIN_EXPECT(!phyaddr, 0))
			goto default_handler;

		if (map_region(viraddr, phyaddr, 1, MAP_USER_SPACE) == viraddr) {
			memset((void*) viraddr, 0x00, PAGE_SIZE);
			return;
		}
		
		kprintf("Could not map 0x%x at 0x%x\n", phyaddr, viraddr);
		put_page(phyaddr);
	}
#endif

//default_handler:
	kprintf("PAGE FAULT: Task %u got page fault at %p (irq %llu, cs:rip 0x%llx:0x%llx)\n", task->id, viraddr, s->int_no, s->cs, s->rip);
	kprintf("Register state: rax = 0x%llx, rbx = 0x%llx, rcx = 0x%llx, rdx = 0x%llx, rdi = 0x%llx, rsi = 0x%llx, rbp = 0x%llx, rsp = 0x%llx\n", 
		s->rax, s->rbx, s->rcx, s->rdx, s->rdi, s->rsi, s->rbp, s->rsp);

while(1);
	irq_enable();
	abort();
}

int arch_paging_init(void)
{
	uint32_t i, npages;

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

	// kernel is already maped into the kernel space (see entry64.asm)
	// this includes .data, .bss, .text, video memory and the multiboot structure

#if MAX_CORES > 1
	// Reserve page for smp boot code
	if (!map_region(SMP_SETUP_ADDR, SMP_SETUP_ADDR, 1, MAP_KERNEL_SPACE|MAP_NO_CACHE)) {
		kputs("could not reserve page for smp boot code\n");
		return -ENOMEM;
	}
#endif

#ifdef CONFIG_MULTIBOOT
#if 0
	/*
	 * Map reserved memory regions into the kernel space
	 */
	if (mb_info && (mb_info->flags & MULTIBOOT_INFO_MEM_MAP)) {
		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(mmap->addr, mmap->addr, npages, MAP_KERNEL_SPACE|MAP_NO_CACHE);
			}
			mmap++;
		}
	}
#endif

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

		npages = mb_info->mods_count * sizeof(multiboot_module_t) >> PAGE_SHIFT;
		if (mb_info->mods_count * sizeof(multiboot_module_t) & (PAGE_SIZE-1))
			npages++;
		map_region((size_t) (mb_info->mods_addr), (size_t) (mb_info->mods_addr), npages, MAP_REMAP|MAP_KERNEL_SPACE);

		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_SHIFT;
			if (mmodule->mod_end & (PAGE_SIZE-1))
				npages++;
			kprintf("Map module %s at 0x%x (%u pages)\n", (char*) mmodule->cmdline, mmodule->mod_start, npages);
			map_region((size_t) (mmodule->mod_start), (size_t) (mmodule->mod_start), npages, MAP_REMAP|MAP_KERNEL_SPACE);
		}
	}
#endif

	/* signalize that we are able to use paging */
	paging_enabled = 1;

	/*
	 * we turned on paging
	 * => now, we are able to register our task
	 */
	register_task();

	// APIC registers into the kernel address space
	map_apic();

	return 0;
}