#define NCHUNKS ((PAGE_SIZE - ZHDR_SIZE_ALIGNED) >> CHUNK_SHIFT)
#define BUDDY_MASK (0x3)
+#define BUDDY_SHIFT 2
/**
* struct z3fold_pool - stores metadata for each z3fold pool
PAGE_HEADLESS = 0,
MIDDLE_CHUNK_MAPPED,
NEEDS_COMPACTING,
- PAGE_STALE
+ PAGE_STALE,
+ PAGE_CLAIMED, /* by either reclaim or free */
};
/*****************
clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
clear_bit(NEEDS_COMPACTING, &page->private);
clear_bit(PAGE_STALE, &page->private);
+ clear_bit(PAGE_CLAIMED, &page->private);
spin_lock_init(&zhdr->page_lock);
kref_init(&zhdr->refcount);
unsigned long handle;
handle = (unsigned long)zhdr;
- if (bud != HEADLESS)
- handle += (bud + zhdr->first_num) & BUDDY_MASK;
+ if (bud != HEADLESS) {
+ handle |= (bud + zhdr->first_num) & BUDDY_MASK;
+ if (bud == LAST)
+ handle |= (zhdr->last_chunks << BUDDY_SHIFT);
+ }
return handle;
}
return (struct z3fold_header *)(handle & PAGE_MASK);
}
+/* only for LAST bud, returns zero otherwise */
+static unsigned short handle_to_chunks(unsigned long handle)
+{
+ return (handle & ~PAGE_MASK) >> BUDDY_SHIFT;
+}
+
/*
* (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle
* but that doesn't matter. because the masking will result in the
spin_lock_init(&pool->lock);
spin_lock_init(&pool->stale_lock);
pool->unbuddied = __alloc_percpu(sizeof(struct list_head)*NCHUNKS, 2);
+ if (!pool->unbuddied)
+ goto out_pool;
for_each_possible_cpu(cpu) {
struct list_head *unbuddied =
per_cpu_ptr(pool->unbuddied, cpu);
pool->name = name;
pool->compact_wq = create_singlethread_workqueue(pool->name);
if (!pool->compact_wq)
- goto out;
+ goto out_unbuddied;
pool->release_wq = create_singlethread_workqueue(pool->name);
if (!pool->release_wq)
goto out_wq;
out_wq:
destroy_workqueue(pool->compact_wq);
-out:
+out_unbuddied:
+ free_percpu(pool->unbuddied);
+out_pool:
kfree(pool);
+out:
return NULL;
}
page = virt_to_page(zhdr);
if (test_bit(PAGE_HEADLESS, &page->private)) {
- /* HEADLESS page stored */
- bud = HEADLESS;
- } else {
- z3fold_page_lock(zhdr);
- bud = handle_to_buddy(handle);
-
- switch (bud) {
- case FIRST:
- zhdr->first_chunks = 0;
- break;
- case MIDDLE:
- zhdr->middle_chunks = 0;
- zhdr->start_middle = 0;
- break;
- case LAST:
- zhdr->last_chunks = 0;
- break;
- default:
- pr_err("%s: unknown bud %d\n", __func__, bud);
- WARN_ON(1);
- z3fold_page_unlock(zhdr);
- return;
+ /* if a headless page is under reclaim, just leave.
+ * NB: we use test_and_set_bit for a reason: if the bit
+ * has not been set before, we release this page
+ * immediately so we don't care about its value any more.
+ */
+ if (!test_and_set_bit(PAGE_CLAIMED, &page->private)) {
+ spin_lock(&pool->lock);
+ list_del(&page->lru);
+ spin_unlock(&pool->lock);
+ free_z3fold_page(page);
+ atomic64_dec(&pool->pages_nr);
}
+ return;
}
- if (bud == HEADLESS) {
- spin_lock(&pool->lock);
- list_del(&page->lru);
- spin_unlock(&pool->lock);
- free_z3fold_page(page);
- atomic64_dec(&pool->pages_nr);
+ /* Non-headless case */
+ z3fold_page_lock(zhdr);
+ bud = handle_to_buddy(handle);
+
+ switch (bud) {
+ case FIRST:
+ zhdr->first_chunks = 0;
+ break;
+ case MIDDLE:
+ zhdr->middle_chunks = 0;
+ break;
+ case LAST:
+ zhdr->last_chunks = 0;
+ break;
+ default:
+ pr_err("%s: unknown bud %d\n", __func__, bud);
+ WARN_ON(1);
+ z3fold_page_unlock(zhdr);
return;
}
atomic64_dec(&pool->pages_nr);
return;
}
+ if (test_bit(PAGE_CLAIMED, &page->private)) {
+ z3fold_page_unlock(zhdr);
+ return;
+ }
if (test_and_set_bit(NEEDS_COMPACTING, &page->private)) {
z3fold_page_unlock(zhdr);
return;
}
list_for_each_prev(pos, &pool->lru) {
page = list_entry(pos, struct page, lru);
+
+ /* this bit could have been set by free, in which case
+ * we pass over to the next page in the pool.
+ */
+ if (test_and_set_bit(PAGE_CLAIMED, &page->private))
+ continue;
+
+ zhdr = page_address(page);
if (test_bit(PAGE_HEADLESS, &page->private))
- /* candidate found */
break;
- zhdr = page_address(page);
- if (!z3fold_page_trylock(zhdr))
+ if (!z3fold_page_trylock(zhdr)) {
+ zhdr = NULL;
continue; /* can't evict at this point */
+ }
kref_get(&zhdr->refcount);
list_del_init(&zhdr->buddy);
zhdr->cpu = -1;
+ break;
}
+ if (!zhdr)
+ break;
+
list_del_init(&page->lru);
spin_unlock(&pool->lock);
goto next;
}
next:
- spin_lock(&pool->lock);
if (test_bit(PAGE_HEADLESS, &page->private)) {
if (ret == 0) {
- spin_unlock(&pool->lock);
free_z3fold_page(page);
+ atomic64_dec(&pool->pages_nr);
return 0;
}
- } else if (kref_put(&zhdr->refcount, release_z3fold_page)) {
- atomic64_dec(&pool->pages_nr);
+ spin_lock(&pool->lock);
+ list_add(&page->lru, &pool->lru);
+ spin_unlock(&pool->lock);
+ } else {
+ z3fold_page_lock(zhdr);
+ clear_bit(PAGE_CLAIMED, &page->private);
+ if (kref_put(&zhdr->refcount,
+ release_z3fold_page_locked)) {
+ atomic64_dec(&pool->pages_nr);
+ return 0;
+ }
+ /*
+ * if we are here, the page is still not completely
+ * free. Take the global pool lock then to be able
+ * to add it back to the lru list
+ */
+ spin_lock(&pool->lock);
+ list_add(&page->lru, &pool->lru);
spin_unlock(&pool->lock);
- return 0;
+ z3fold_page_unlock(zhdr);
}
- /*
- * Add to the beginning of LRU.
- * Pool lock has to be kept here to ensure the page has
- * not already been released
- */
- list_add(&page->lru, &pool->lru);
+ /* We started off locked to we need to lock the pool back */
+ spin_lock(&pool->lock);
}
spin_unlock(&pool->lock);
return -EAGAIN;
set_bit(MIDDLE_CHUNK_MAPPED, &page->private);
break;
case LAST:
- addr += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT);
+ addr += PAGE_SIZE - (handle_to_chunks(handle) << CHUNK_SHIFT);
break;
default:
pr_err("unknown buddy id %d\n", buddy);