blob: 9ec39b0fac55599ad3fd1988e0e68a2d6cdad31d (
plain) (
tree)
|
|
#include <slab_alloc.h>
typedef struct object {
struct object *next;
} object_t;
typedef struct cache {
struct { // pack this so that it takes less space...
uint32_t is_a_cache : 1; // if not, this is a region allocated alone
uint32_t n_free_objs : 31;
};
union {
struct { // when this is a cache
object_t* first_free_obj;
struct cache *next_cache;
} c;
struct { // when this is a region allocated alone
size_t region_size;
} sr;
};
void* region_addr;
struct cache *next_region;
} cache_t;
typedef struct slab {
cache_t *first_cache; // linked list of caches
} slab_t;
struct mem_allocator {
const slab_type_t *types;
slab_t *slabs;
cache_t *first_free_region_descriptor;
cache_t *all_regions;
page_alloc_fun_t alloc_fun;
page_free_fun_t free_fun;
};
// ============================================== //
// Helper functions for the manipulation of lists //
// ============================================== //
void add_free_region_descriptor(mem_allocator_t *a, cache_t *c) {
c->next_region = a->first_free_region_descriptor;
a->first_free_region_descriptor = c;
}
cache_t *take_region_descriptor(mem_allocator_t *a) {
if (a->first_free_region_descriptor == 0) {
// TODO : allocate more descriptors (not complicated)
return 0;
}
cache_t *x = a->first_free_region_descriptor;
a->first_free_region_descriptor = x->next_region;
return x;
}
// ============================== //
// The actual allocator functions //
// ============================== //
mem_allocator_t* create_slab_allocator(const slab_type_t *types, page_alloc_fun_t af, page_free_fun_t ff) {
union {
void* addr;
mem_allocator_t *a;
slab_t *s;
cache_t *c;
} ptr;
ptr.addr = af(PAGE_SIZE);
if (ptr.addr == 0) return 0; // could not allocate
void* end_addr = ptr.addr + PAGE_SIZE;
mem_allocator_t *a = ptr.a;
ptr.a++;
a->all_regions = 0;
a->alloc_fun = af;
a->free_fun = ff;
a->types = types;
a->slabs = ptr.s;
for (const slab_type_t *t = types; t->obj_size != 0; t++) {
ptr.s->first_cache = 0;
ptr.s++;
}
a->first_free_region_descriptor = 0;
while ((void*)(ptr.c + 1) <= end_addr) {
add_free_region_descriptor(a, ptr.c);
ptr.c++;
}
return a;
}
static void stack_and_destroy_regions(page_free_fun_t ff, cache_t *r) {
if (r == 0) return;
void* addr = r->region_addr;
stack_and_destroy_regions(ff, r->next_region);
ff(addr);
}
void destroy_slab_allocator(mem_allocator_t *a) {
stack_and_destroy_regions(a->free_fun, a->all_regions);
a->free_fun(a);
}
void* slab_alloc(mem_allocator_t* a, size_t sz) {
for (int i = 0; a->types[i].obj_size != 0; i++) {
const size_t obj_size = a->types[i].obj_size;
if (sz <= obj_size) {
// find a cache with free space
cache_t *fc = a->slabs[i].first_cache;
while (fc != 0 && fc->n_free_objs == 0) {
ASSERT(fc->c.first_free_obj == 0); // make sure n_free == 0 iff no object in the free stack
fc = fc->c.next_cache;
}
// if none found, try to allocate a new one
if (fc == 0) {
fc = take_region_descriptor(a);
if (fc == 0) return 0;
const size_t cache_size = a->types[i].pages_per_cache * PAGE_SIZE;
fc->region_addr = a->alloc_fun(cache_size);
if (fc->region_addr == 0) {
add_free_region_descriptor(a, fc);
return 0;
}
fc->is_a_cache = 1;
fc->n_free_objs = 0;
fc->c.first_free_obj = 0;
for (void* i = fc->region_addr; i + obj_size <= fc->region_addr + cache_size; i += obj_size) {
object_t *x = (object_t*)i;
x->next = fc->c.first_free_obj;
fc->c.first_free_obj = x;
fc->n_free_objs++;
}
ASSERT(fc->n_free_objs == cache_size / obj_size);
fc->next_region = a->all_regions;
a->all_regions = fc;
fc->c.next_cache = a->slabs[i].first_cache;
a->slabs[i].first_cache = fc;
}
// allocate on fc
ASSERT(fc != 0 && fc->n_free_objs > 0);
object_t *x = fc->c.first_free_obj;
fc->c.first_free_obj = x->next;
fc->n_free_objs--;
// TODO : if fc is full, put it at the end
return x;
}
}
// otherwise directly allocate using a->alloc_fun
cache_t *r = take_region_descriptor(a);
if (r == 0) return 0;
r->region_addr = a->alloc_fun(sz);
if (r->region_addr == 0) {
add_free_region_descriptor(a, r);
return 0;
} else {
r->is_a_cache = 0;
r->sr.region_size = sz;
r->next_region = a->all_regions;
a->all_regions = r;
return (void*)r->region_addr;
}
}
void slab_free(mem_allocator_t* a, void* addr) {
for (int i = 0; a->types[i].obj_size != 0; i++) {
size_t region_size = PAGE_SIZE * a->types[i].pages_per_cache;
for (cache_t *r = a->slabs[i].first_cache; r != 0; r = r->c.next_cache) {
if (addr >= r->region_addr && addr < r->region_addr + region_size) {
ASSERT((addr - r->region_addr) % a->types[i].obj_size == 0);
ASSERT(r->is_a_cache);
object_t *o = (object_t*)addr;
o->next = r->c.first_free_obj;
r->c.first_free_obj = o;
r->n_free_objs++;
// TODO : if cache is empty, free it
return;
}
}
}
// otherwise the block was directly allocated : look for it in regions.
a->free_fun(addr);
ASSERT(a->all_regions != 0);
if (a->all_regions->region_addr == addr) {
cache_t *r = a->all_regions;
ASSERT(r->is_a_cache == 0);
a->all_regions = r->next_region;
add_free_region_descriptor(a, r);
} else {
for (cache_t *i = a->all_regions; i->next_region != 0; i = i->next_region) {
if (i->next_region->region_addr == addr) {
cache_t *r = i->next_region;
ASSERT(r->is_a_cache == 0);
i->next_region = r->next_region;
add_free_region_descriptor(a, r);
return;
}
}
ASSERT(false);
}
}
|