helpless/oogabooga/memory.c

#define KB(x) (x*1024ull)
#define MB(x) ((KB(x))*1024ull)
#define GB(x) ((MB(x))*1024ull)


void* program_memory = 0;
u64 program_memory_size = 0;

#ifndef INIT_MEMORY_SIZE
	#define INIT_MEMORY_SIZE KB(50)
#endif
// We may need to allocate stuff in initialization time before the heap is ready.
// That's what this is for.
u8 init_memory_arena[INIT_MEMORY_SIZE];
u8 *init_memory_head = init_memory_arena;

void* initialization_allocator_proc(u64 size, void *p, Allocator_Message message, void *data) {
	switch (message) {
		case ALLOCATOR_ALLOCATE: {
			p = init_memory_head;
			init_memory_head += size;
			
			if (init_memory_head >= ((u8*)init_memory_arena+INIT_MEMORY_SIZE)) {
				os_write_string_to_stdout(STR("Out of initialization memory! Please provide more by increasing INIT_MEMORY_SIZE"));
				os_break();
			}
			return p;
			break;
		}
		case ALLOCATOR_DEALLOCATE: {
			return 0;
		}
		case ALLOCATOR_REALLOCATE: {
			return 0;
		}
	}
	return 0;
}

Allocator get_initialization_allocator() {
	Allocator a;
	a.proc = initialization_allocator_proc;
	return a;
}

///
///
// Basic general heap allocator, free list
///
// Technically thread safe but synchronization is horrible.
// Fragmentation is catastrophic.
// We could fix it by merging free nodes every now and then
// BUT: We aren't really supposed to allocate/deallocate directly on the heap too much anyways...

#define MAX_HEAP_BLOCK_SIZE ((MB(500)+os.page_size)& ~(os.page_size-1))
#define DEFAULT_HEAP_BLOCK_SIZE (min(MAX_HEAP_BLOCK_SIZE, program_memory_size))
#define HEAP_ALIGNMENT (sizeof(Heap_Free_Node))
typedef struct Heap_Free_Node Heap_Free_Node;
typedef struct Heap_Block Heap_Block;

typedef struct Heap_Free_Node {
	u64 size;
	Heap_Free_Node *next;
} Heap_Free_Node;

typedef struct Heap_Block {
	u64 size;
	Heap_Free_Node *free_head;
	void* start;
	Heap_Block *next;
	// 32 bytes !!
} Heap_Block;

typedef struct {
	u64 size;
	Heap_Block *block;
} Heap_Allocation_Metadata;

Heap_Block *heap_head;
bool heap_initted = false;
Spinlock *heap_lock; // This is terrible but I don't care for now
	

u64 get_heap_block_size_excluding_metadata(Heap_Block *block) {
	return block->size - sizeof(Heap_Block);
}
u64 get_heap_block_size_including_metadata(Heap_Block *block) {
	return block->size;
}

bool is_pointer_in_program_memory(void *p) {
	return (u8*)p >= (u8*)program_memory && (u8*)p<((u8*)program_memory+program_memory_size);
}
bool is_pointer_in_stack(void* p) {
    void* stack_base = os_get_stack_base();
    void* stack_limit = os_get_stack_limit();
    return (uintptr_t)p >= (uintptr_t)stack_limit && (uintptr_t)p < (uintptr_t)stack_base;
}
bool is_pointer_in_static_memory(void* p) {
    return (uintptr_t)p >= (uintptr_t)os.static_memory_start && (uintptr_t)p < (uintptr_t)os.static_memory_end;
}
bool is_pointer_valid(void *p) {
	return is_pointer_in_program_memory(p) || is_pointer_in_stack(p) || is_pointer_in_static_memory(p);
}

// Meant for debug
void santiy_check_free_node_tree(Heap_Block *block) {
	Heap_Free_Node *node = block->free_head;
	
	u64 total_free = 0;
	while (node != 0) {
		Heap_Free_Node *other_node = node->next;
		
		while (other_node != 0) {
			assert(other_node != node, "Circular reference in heap free node tree. That's bad.");
			other_node = other_node->next;
		}
		total_free += node->size;
		assert(total_free <= block->size, "Free nodes are fucky wucky");
		node = node->next;
	}
	
}
inline void check_meta(Heap_Allocation_Metadata *meta) {
// If > 256GB then prolly not legit lol
	assert(meta->size < 1024ULL*1024ULL*1024ULL*256ULL, "Garbage pointer passed to heap_dealloc !!! Or could be corrupted memory.");	
	assert(is_pointer_in_program_memory(meta->block), "Garbage pointer passed to heap_dealloc !!! Or could be corrupted memory."); 
}

typedef struct {
	Heap_Free_Node *best_fit;
	Heap_Free_Node *previous;
	u64 delta;
} Heap_Search_Result;
Heap_Search_Result search_heap_block(Heap_Block *block, u64 size) {
	
	if (block->free_head == 0)  return (Heap_Search_Result){0, 0, 0};
		
	Heap_Free_Node *node = block->free_head;
	Heap_Free_Node *previous = 0;
	
	Heap_Free_Node *best_fit = 0;
	Heap_Free_Node *before_best_fit = 0;
	u64 best_fit_delta = 0;
	
	while (node != 0) {
		
		if (node->size == size) {
			Heap_Search_Result result;
			result.best_fit = node;
			result.previous = previous;
			result.delta = 0;
			assert(result.previous != result.best_fit);
			return result;
		}
		
		if (node->size >= size) {
			u64 delta = node->size - size;
			
			if (delta < best_fit_delta || !best_fit) {
				before_best_fit = previous;
				best_fit = node;
				best_fit_delta = delta;
			}
		}
		
		if (node->next) previous = node;
		node = node->next;
	}
	
	if (!best_fit)  return (Heap_Search_Result){0, 0, 0};

	Heap_Search_Result result;
	result.best_fit = best_fit;
	result.previous = before_best_fit;
	result.delta = best_fit_delta;
	assert(result.previous != result.best_fit);
	return result;
}

Heap_Block *make_heap_block(Heap_Block *parent, u64 size) {

	size = (size) & ~(HEAP_ALIGNMENT-1);	

	Heap_Block *block;
	if (parent) {
		block = (Heap_Block*)(((u8*)parent)+get_heap_block_size_including_metadata(parent));
		parent->next = block;
	} else {
		block = (Heap_Block*)program_memory;
	}
	
	
	
	// #Speed #Cleanup
	if (((u8*)block)+size >= ((u8*)program_memory)+program_memory_size) {
		u64 minimum_size = ((u8*)block+size) - (u8*)program_memory + 1;
		u64 new_program_size = max((cast(u64)(minimum_size*2)), program_memory_size*2);
		assert(new_program_size >= minimum_size, "Bröd");
		const u64 ATTEMPTS = 1000;
		for (u64 i = 0; i <= ATTEMPTS; i++) {
			if (program_memory_size >= new_program_size) break; // Another thread might have resized already, causing it to fail here.
			assert(i < ATTEMPTS, "OS is not letting us allocate more memory. Maybe we are out of memory?");
			if (os_grow_program_memory(new_program_size))
				break;
		}
	}
	
	block->start = ((u8*)block)+sizeof(Heap_Block);
	block->size = size;
	block->next = 0;
	block->free_head = (Heap_Free_Node*)block->start;
	block->free_head->size = get_heap_block_size_excluding_metadata(block);
	block->free_head->next = 0;
	
	return block;
}

void heap_init() {
	if (heap_initted) return;
	assert(HEAP_ALIGNMENT == 16);
	assert(sizeof(Heap_Allocation_Metadata) % HEAP_ALIGNMENT == 0);
	heap_initted = true;
	heap_head = make_heap_block(0, DEFAULT_HEAP_BLOCK_SIZE);
	heap_lock = os_make_spinlock(get_initialization_allocator());
}



void *heap_alloc(u64 size) {

	if (!heap_initted) heap_init();

	// #Sync #Speed oof
	os_spinlock_lock(heap_lock);
	
	size += sizeof(Heap_Allocation_Metadata);
	
	size = (size+HEAP_ALIGNMENT) & ~(HEAP_ALIGNMENT-1);
	
	assert(size < MAX_HEAP_BLOCK_SIZE, "Past Charlie has been lazy and did not handle large allocations like this. I apologize on behalf of past Charlie. A quick fix could be to increase the heap block size for now.");
	
	Heap_Block *block = heap_head;
	Heap_Block *last_block = 0;
	Heap_Free_Node *best_fit = 0;
	Heap_Block *best_fit_block = 0;
	Heap_Free_Node *previous = 0;
	u64 best_fit_delta = 0;
	// #Speed
	// Maybe instead of going through EVERY free node to find best fit we do a good-enough fit
	while (block != 0) {
	
		if (block->size < size) {
			last_block = block;
			block = block->next;
			continue;
		}
	
		Heap_Search_Result result = search_heap_block(block, size);
		Heap_Free_Node *node = result.best_fit;
		if (node) {
			if (node->size < size) continue;
			if (node->size == size) {
				best_fit = node;
				best_fit_block = block;
				previous = result.previous;
				best_fit_delta = 0;
				break;
			}
			
			u64 delta = node->size-size;
			if (delta < best_fit_delta || !best_fit) {
				best_fit = node;
				best_fit_block = block;
				previous = result.previous;
				best_fit_delta = delta;
			}
		}
		
		last_block = block;
		block = block->next;
	}
	
	if (!best_fit) {
		block = make_heap_block(last_block, max(DEFAULT_HEAP_BLOCK_SIZE, size));
		previous = 0;
		best_fit = block->free_head;
		best_fit_block = block;
	}
		
	
	// Ideally this should not be possible.
	// If we run out of program_memory, we should just grow it and if that fails
	// we crash because out of memory.
	assert(best_fit != 0, "Internal heap allocation failed");
	
	Heap_Free_Node *new_free_node = 0;
	if (size != best_fit->size) {
		u64 remainder = best_fit->size - size;
		new_free_node = (Heap_Free_Node*)(((u8*)best_fit)+size);
		new_free_node->size = remainder;
		new_free_node->next = best_fit->next;
	}
	
	if (previous && new_free_node) {
		assert(previous->next == best_fit, "Bro what");
		previous->next = new_free_node;
	} else if (previous) {
		assert(previous->next == best_fit, "Bro what");
		previous->next = best_fit->next;
	}
	
	if (best_fit_block->free_head == best_fit) {
		// If we allocated the first free node then replace with new free node or just
		// remove it if perfect fit.
		if (new_free_node) {
			new_free_node->next = best_fit_block->free_head->next;
			best_fit_block->free_head = new_free_node;
		} else best_fit_block->free_head = best_fit_block->free_head->next;
	}
	
	Heap_Allocation_Metadata *meta = (Heap_Allocation_Metadata*)best_fit;
	meta->size = size;
	meta->block = best_fit_block;

#if CONFIGURATION == VERY_DEBUG
	santiy_check_free_node_tree(meta->block);
#endif
	
	// #Sync #Speed oof
	os_spinlock_unlock(heap_lock);
	
	
	void *p = ((u8*)meta)+sizeof(Heap_Allocation_Metadata);
	assert((u64)p % HEAP_ALIGNMENT == 0);
	return p;
}
void heap_dealloc(void *p) {
	// #Sync #Speed oof
	
	if (!heap_initted) heap_init();

	os_spinlock_lock(heap_lock);
	
	assert(is_pointer_in_program_memory(p), "Garbage pointer; out of program memory bounds!"); 
	p = (u8*)p-sizeof(Heap_Allocation_Metadata);
	Heap_Allocation_Metadata *meta = (Heap_Allocation_Metadata*)(p);
	check_meta(meta);
	
	
	
	// Yoink meta data before we start overwriting it
	Heap_Block *block = meta->block;
	u64 size = meta->size;
	
	Heap_Free_Node *new_node = cast(Heap_Free_Node*)p;
	new_node->size = size;
	
	if (new_node < block->free_head) {
		if ((u8*)new_node+size == (u8*)block->free_head) {
			new_node->size = size + block->free_head->size;
			new_node->next = block->free_head->next;
			block->free_head = new_node;
		} else {
			new_node->next = block->free_head;
			block->free_head = new_node;
		}
	} else {
		Heap_Free_Node *node = block->free_head;
	
		
		while (true) {
		
			assert(node != 0, "We didn't find where the free node should be! uh oh");
			
			if (new_node > node) {
				u8* node_tail = (u8*)node + node->size;
				if (cast(u8*)new_node == node_tail) {
					node->size += new_node->size;
					break;
				} else {
					new_node->next = node->next;
					node->next = new_node;
					
					u8* new_node_tail = (u8*)new_node + new_node->size;
					if (new_node->next && (u8*)new_node->next == new_node_tail) {
						new_node->size += new_node->next->size;
						new_node->next = new_node->next->next;
					}
					break;
				}
			}
			
			node = node->next;
		}
	}
	
#if CONFIGURATION == VERY_DEBUG
	santiy_check_free_node_tree(block);
#endif

	// #Sync #Speed oof
	os_spinlock_unlock(heap_lock);
}

void* heap_allocator_proc(u64 size, void *p, Allocator_Message message, void* data) {
	switch (message) {
		case ALLOCATOR_ALLOCATE: {
			return heap_alloc(size);
			break;
		}
		case ALLOCATOR_DEALLOCATE: {
			assert(is_pointer_valid(p), "Invalid pointer passed to heap allocator deallocate");
			Heap_Allocation_Metadata *meta = (Heap_Allocation_Metadata*)(((u64)p)-sizeof(Heap_Allocation_Metadata));
			check_meta(meta);
			heap_dealloc(p);
			return 0;
		}
		case ALLOCATOR_REALLOCATE: {
			if (!p) {
				return heap_alloc(size);
			}
			assert(is_pointer_valid(p), "Invalid pointer passed to heap allocator reallocate");
			Heap_Allocation_Metadata *meta = (Heap_Allocation_Metadata*)(((u64)p)-sizeof(Heap_Allocation_Metadata));
			check_meta(meta);
			void *new = heap_alloc(size);
			memcpy(new, p, min(size, meta->size));
			heap_dealloc(p);
			return new;
		}
	}
	return 0;
}

Allocator get_heap_allocator() {
	Allocator heap_allocator;
	
	heap_allocator.proc = heap_allocator_proc;
	heap_allocator.data = 0;
	
	return heap_allocator;
}

///
///
// Temporary storage
///

#ifndef TEMPORARY_STORAGE_SIZE
	#define TEMPORARY_STORAGE_SIZE (1024ULL*1024ULL*2ULL) // 2mb
#endif

void* talloc(u64);
void* temp_allocator_proc(u64 size, void *p, Allocator_Message message, void*);

thread_local void * temporary_storage = 0;
thread_local bool   temporary_storage_initted = false;
thread_local void * temporary_storage_pointer = 0;
thread_local bool   has_warned_temporary_storage_overflow = false;
thread_local Allocator temp;


void* temp_allocator_proc(u64 size, void *p, Allocator_Message message, void* data) {
	switch (message) {
		case ALLOCATOR_ALLOCATE: {
			return talloc(size);
			break;
		}
		case ALLOCATOR_DEALLOCATE: {
			return 0;
		}
		case ALLOCATOR_REALLOCATE: {
			return 0;
		}
	}
	return 0;
}

void temporary_storage_init() {
	if (temporary_storage_initted) return;
	
	temporary_storage = heap_alloc(TEMPORARY_STORAGE_SIZE);
	assert(temporary_storage, "Failed allocating temporary storage");
	temporary_storage_pointer = temporary_storage;

	temp.proc = temp_allocator_proc;
	temp.data = 0;
	
	temporary_storage_initted = true;
	
	temp.proc = temp_allocator_proc;
}

void* talloc(u64 size) {
	if (!temporary_storage_initted) temporary_storage_init();
	
	assert(size < TEMPORARY_STORAGE_SIZE, "Bruddah this is too large for temp allocator");
	
	void* p = temporary_storage_pointer;
	
	temporary_storage_pointer = (u8*)temporary_storage_pointer + size;
	
	if ((u8*)temporary_storage_pointer >= (u8*)temporary_storage+TEMPORARY_STORAGE_SIZE) {
		if (!has_warned_temporary_storage_overflow) {
			os_write_string_to_stdout(STR("WARNING: temporary storage was overflown, we wrap around at the start.\n"));
		}
		temporary_storage_pointer = temporary_storage;
		return talloc(size);;
	}
	
	return p;
}

void reset_temporary_storage() {
	if (!temporary_storage_initted) temporary_storage_init();
	
	temporary_storage_pointer = temporary_storage;
	
	has_warned_temporary_storage_overflow = true;
}