metalsvm/mm/vma.c

/* 
 * Copyright 2011 Steffen Vogel, 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/vma.h>
#include <metalsvm/stdlib.h>
#include <metalsvm/stdio.h>
#include <metalsvm/tasks_types.h>
#include <metalsvm/spinlock.h>
#include <metalsvm/errno.h>

#ifdef CONFIG_MULTIBOOT
#include <asm/multiboot.h>
#endif
#include <asm/apic.h>

extern apic_mp_t* apic_mp;

/* 
 * 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;

/*
 * Kernel space VMA list and lock
 * 
 * For bootstrapping we initialize the VMA list with one empty VMA
 * (start == end) and expand this VMA by calls to vma_alloc()
 */
static vma_t vma_boot = { VMA_KERN_MIN, VMA_KERN_MIN, VMA_HEAP };
static vma_t* vma_list = &vma_boot;
static spinlock_t vma_lock = SPINLOCK_INIT;

// TODO: we might move the architecture specific VMA regions to a
//       seperate function arch_vma_init()
int vma_init()
{
	int ret;

	// add Kernel
	ret  = vma_add(PAGE_CEIL((size_t) &kernel_start),
		PAGE_FLOOR((size_t) &kernel_end),
		VMA_READ|VMA_WRITE|VMA_EXECUTE|VMA_CACHEABLE);
	if (ret)
		goto out;

	// add MP Table
	ret = vma_add(PAGE_CEIL((size_t) apic_mp),
		PAGE_FLOOR((size_t) apic_mp + sizeof(apic_mp_t)),
		VMA_READ|VMA_WRITE|VMA_EXECUTE|VMA_CACHEABLE);
	if (ret)
		goto out;

	// add IOAPIC and LAPIC memory mapped registers
	ret = vma_add(LAPIC_ADDR, LAPIC_ADDR + PAGE_SIZE, VMA_READ|VMA_WRITE);
	if (ret)
		goto out;

	ret = vma_add(IOAPIC_ADDR, IOAPIC_ADDR + PAGE_SIZE, VMA_READ|VMA_WRITE);
	if (ret)
		goto out;

#ifdef CONFIG_VGA
	// add VGA video memory
	ret = vma_add(VIDEO_MEM_ADDR, VIDEO_MEM_ADDR + PAGE_SIZE, VMA_READ|VMA_WRITE);
	if (ret)
		goto out;
#endif

#if MAX_CORES > 1
	// add SMP boot page
	ret = vma_add(SMP_SETUP_ADDR, SMP_SETUP_ADDR + PAGE_SIZE,
		VMA_READ|VMA_WRITE|VMA_EXECUTE|VMA_CACHEABLE);
	if (ret)
		goto out;
#endif

#ifdef CONFIG_MULTIBOOT
	// add Multiboot structures as modules
	if (mb_info) {
		ret = vma_add(PAGE_CEIL((size_t) mb_info),
			PAGE_FLOOR((size_t) mb_info + sizeof(multiboot_info_t)),
			VMA_READ|VMA_CACHEABLE);
		if (ret)
			goto out;

		if (mb_info->flags & MULTIBOOT_INFO_MEM_MAP) {
			ret = vma_add(PAGE_CEIL((size_t) mb_info->mmap_addr),
				PAGE_FLOOR((size_t) mb_info->mmap_addr + mb_info->mmap_length),
				VMA_READ|VMA_CACHEABLE);
		}

		if (mb_info->flags & MULTIBOOT_INFO_MODS) {
			multiboot_module_t* mmodule = (multiboot_module_t*) ((size_t) mb_info->mods_addr);

			ret = vma_add(PAGE_CEIL((size_t) mb_info->mods_addr),
				PAGE_FLOOR((size_t) mb_info->mods_addr + mb_info->mods_count*sizeof(multiboot_module_t)),
				VMA_READ|VMA_CACHEABLE);

			int i;
			for(i=0; i<mb_info->mods_count; i++) {
				ret = vma_add(PAGE_CEIL(mmodule[i].mod_start),
					PAGE_FLOOR(mmodule[i].mod_end),
					VMA_READ|VMA_WRITE|VMA_CACHEABLE);
				if (ret)
					goto out;
			}
		}
	}
#endif

out:
	return ret;
}

size_t vma_alloc(size_t size, uint32_t flags)
{
	task_t* task = per_core(current_task);
	spinlock_t* lock;
	vma_t** list;

	kprintf("vma_alloc(0x%lx, 0x%x)\n", size, flags);

	size_t base, limit; // boundaries for search
	size_t start, end;

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

	if (flags & VMA_USER) {
		base = VMA_KERN_MAX;
		limit = VMA_USER_MAX;
		list = &task->vma_list;
		lock = &task->vma_lock;
	}
	else {
		base = VMA_KERN_MIN;
		limit = VMA_KERN_MAX;
		list = &vma_list;
		lock = &vma_lock;
	}

	spinlock_lock(lock);

	// first fit search for free memory area
	vma_t* pred = NULL;  // vma before current gap
	vma_t* succ = *list; // vma after current gap
	do {
		start = (pred) ? pred->end : base;
		end = (succ) ? succ->start : limit;

		if (end > start && end - start > size)
			break; // we found a gap

		pred = succ;
		succ = (succ) ? succ->next : NULL;
	} while (pred || succ);

	if (BUILTIN_EXPECT(end > limit || end < start || end - start < size, 0)) {
		spinlock_unlock(lock);
		return 0;
	}


	if (pred && pred->flags == flags) {
		pred->end = start+size;
	}
	else {
		// insert new VMA
		vma_t* new = kmalloc(sizeof(vma_t));
		if (BUILTIN_EXPECT(!new, 0)) {
			spinlock_unlock(lock);
			return 0;
		}

		new->start = start;
		new->end = start+size;
		new->flags = flags;
		new->next = succ;
		new->prev = pred;

		if (succ)
			succ->prev = new;
		if (pred)
			pred->next = new;
		else
			*list = new;
	}

	spinlock_unlock(lock);

	return start;
}

int vma_free(size_t start, size_t end)
{
	task_t* task = per_core(current_task);
	spinlock_t* lock;
	vma_t* vma;
	vma_t** list;

	if (BUILTIN_EXPECT(start >= end, 0))
		return -EINVAL;

	if (end < VMA_KERN_MAX) {
		lock = &vma_lock;
		list = &vma_list;
	}
	else if (start >= VMA_KERN_MAX) {
		lock = &task->vma_lock;
		list = &task->vma_list;
	}

	if (BUILTIN_EXPECT(!*list, 0))
		return -EINVAL;

	spinlock_lock(lock);

	// search vma
	vma = *list;
	while (vma) {
		if (start >= vma->start && end <= vma->end) break;
		vma = vma->next;
	}

	if (BUILTIN_EXPECT(!vma, 0)) {
		spinlock_unlock(lock);
		return -EINVAL;
	}

	// free/resize vma
	if (start == vma->start && end == vma->end) {
		if (vma == *list)
			*list = vma->next; // update list head
		if (vma->prev)
			vma->prev->next = vma->next;
		if (vma->next)
			vma->next->prev = vma->prev;
		kfree(vma);
	}
	else if (start == vma->start)
		vma->start = end;
	else if (end == vma->end)
		vma->end = start;
	else {
		vma_t* new = kmalloc(sizeof(vma_t));
		if (BUILTIN_EXPECT(!new, 0)) {
			spinlock_unlock(lock);
			return -ENOMEM;
		}

		new->end = vma->end;
		vma->end = start;
		new->start = end;

		new->next = vma->next;
		vma->next = new;
		new->prev = vma;
	}

	spinlock_unlock(lock);

	return 0;
}

int vma_add(size_t start, size_t end, uint32_t flags)
{
	task_t* task = per_core(current_task);
	spinlock_t* lock;
	vma_t** list;

	if (BUILTIN_EXPECT(start >= end, 0))
		return -EINVAL;

	if (flags & VMA_USER) {
		list = &task->vma_list;
		lock = &task->vma_lock;

		// check if address is in userspace
		if (BUILTIN_EXPECT(start < VMA_KERN_MAX, 0))
			return -EINVAL;
	}
	else {
		list = &vma_list;
		lock = &vma_lock;

		// check if address is in kernelspace
		if (BUILTIN_EXPECT(end >= VMA_KERN_MAX, 0))
			return -EINVAL;
	}

	spinlock_lock(lock);

	// search gap
	vma_t* pred = NULL;
	vma_t* succ = *list;

	while (pred || succ) {
		if ((!pred || pred->end <= start) &&
		    (!succ || succ->start >= end))
			break;

		pred = succ;
		succ = (succ) ? succ->next : NULL;
	}

	if (BUILTIN_EXPECT(*list && !pred && !succ, 0)) {
		spinlock_unlock(lock);
		return -EINVAL;
	}

	// insert new VMA
	vma_t* new = kmalloc(sizeof(vma_t));
	if (BUILTIN_EXPECT(!new, 0)) {
		spinlock_unlock(lock);
		return -ENOMEM;
	}

	new->start = start;
	new->end = end;
	new->flags = flags;
	new->next = succ;
	new->prev = pred;

	if (succ)
		succ->prev = new;
	if (pred)
		pred->next = new;
	else
		*list = new;

	spinlock_unlock(lock);

	return 0;
}

int copy_vma_list(task_t* task)
{
	task_t* parent_task = per_core(current_task);

	spinlock_init(&task->vma_lock);

	spinlock_lock(&parent_task->vma_lock);
	spinlock_lock(&task->vma_lock);

	vma_t* last = NULL;

	vma_t* parent;
	for (parent=parent_task->vma_list; parent; parent=parent->next) {
		vma_t *new = kmalloc(sizeof(vma_t));
		if (BUILTIN_EXPECT(!new, 0)) {
			spinlock_unlock(&task->vma_lock);
			spinlock_unlock(&parent_task->vma_lock);
			return -ENOMEM;
		}

		new->start = parent->start;
		new->end = parent->end;
		new->flags = parent->flags;
		new->prev = last;

		if (last)
			last->next = new;
		else
			task->vma_list = new;

		last = new;
	}

	spinlock_unlock(&task->vma_lock);
	spinlock_unlock(&parent_task->vma_lock);

	return 0;
}

int drop_vma_list()
{
	task_t* task = per_core(current_task);
	vma_t* vma;

	spinlock_lock(&task->vma_lock);

	while ((vma = task->vma_list)) {
		task->vma_list = vma->next;
		kfree(vma);
	}

	spinlock_unlock(&task->vma_lock);

	return 0;
}

void vma_dump()
{
	void print_vma(vma_t *vma) {
		while (vma) {
			kprintf("0x%lx - 0x%lx: size=%x, flags=%c%c%c\n", vma->start, vma->end, vma->end - vma->start,
				(vma->flags & VMA_READ) ? 'r' : '-',
				(vma->flags & VMA_WRITE) ? 'w' : '-',
				(vma->flags & VMA_EXECUTE) ? 'x' : '-');
			vma = vma->next;
		}
	}

	task_t* task = per_core(current_task);

	kputs("Kernelspace VMAs:\n");
	spinlock_lock(&vma_lock);
	print_vma(vma_list);
	spinlock_unlock(&vma_lock);

	kputs("Userspace VMAs:\n");
	spinlock_lock(&task->vma_lock);
	print_vma(task->vma_list);
	spinlock_unlock(&task->vma_lock);
}