[Devel] Re: [PATCH 2/6] BC: beancounters core (API)

[Kirill Korotaev]

Andrew Morton wrote:
[... snip ...]

>>+#define bc_charge_locked(bc, r, v, s)			(0)
>>> +#define bc_charge(bc, r, v)				(0)
>
>akpm:/home/akpm> cat t.c
>void foo(void)
>{
>	(0);
>}
>akpm:/home/akpm> gcc -c -Wall t.c
>t.c: In function 'foo':
>t.c:4: warning: statement with no effect

these functions return value should always be checked (!).
i.e. it is never called like:
  ub_charge(bc, r, v);

>>+struct beancounter *beancounter_findcreate(uid_t uid, int mask)
>>+{
>>+	struct beancounter *new_bc, *bc;
>>+	unsigned long flags;
>>+	struct hlist_head *slot;
>>+	struct hlist_node *pos;
>>+
>>+	slot = &bc_hash[bc_hash_fun(uid)];
>>+	new_bc = NULL;
>>+
>>+retry:
>>+	spin_lock_irqsave(&bc_hash_lock, flags);
>>+	hlist_for_each_entry (bc, pos, slot, hash)
>>+		if (bc->bc_id == uid)
>>+			break;
>>+
>>+	if (pos != NULL) {
>>+		get_beancounter(bc);
>>+		spin_unlock_irqrestore(&bc_hash_lock, flags);
>>+
>>+		if (new_bc != NULL)
>>+			kmem_cache_free(bc_cachep, new_bc);
>>+		return bc;
>>+	}
>>+
>>+	if (!(mask & BC_ALLOC))
>>+		goto out_unlock;
>>+
>>+	if (new_bc != NULL)
>>+		goto out_install;
>>+
>>+	spin_unlock_irqrestore(&bc_hash_lock, flags);
>>+
>>+	new_bc = kmem_cache_alloc(bc_cachep,
>>+			mask & BC_ALLOC_ATOMIC ? GFP_ATOMIC : GFP_KERNEL);
>>+	if (new_bc == NULL)
>>+		goto out;
>>+
>>+	memcpy(new_bc, &default_beancounter, sizeof(*new_bc));
>>+	init_beancounter_struct(new_bc, uid);
>>+	goto retry;
>>+
>>+out_install:
>>+	hlist_add_head(&new_bc->hash, slot);
>>+out_unlock:
>>+	spin_unlock_irqrestore(&bc_hash_lock, flags);
>>+out:
>>+	return new_bc;
>>+}
> 
> 
> Can remove the global bc_hash_lock and make the locking per-hash-bucket.
it is not performance critical path IMHO.
this lock is taken on container create/change/destroy/user interfaces only.

>>+static inline void verify_held(struct beancounter *bc)
>>+{
>>+	int i;
>>+
>>+	for (i = 0; i < BC_RESOURCES; i++)
>>+		if (bc->bc_parms[i].held != 0)
>>+			bc_print_resource_warning(bc, i,
>>+					"resource is held on put", 0, 0);
>>+}
>>+
>>+void __put_beancounter(struct beancounter *bc)
>>+{
>>+	unsigned long flags;
>>+
>>+	/* equivalent to atomic_dec_and_lock_irqsave() */
>>+	local_irq_save(flags);
>>+	if (likely(!atomic_dec_and_lock(&bc->bc_refcount, &bc_hash_lock))) {
>>+		local_irq_restore(flags);
>>+		if (unlikely(atomic_read(&bc->bc_refcount) < 0))
>>+			printk(KERN_ERR "BC: Bad refcount: bc=%p, "
>>+					"luid=%d, ref=%d\n",
>>+					bc, bc->bc_id,
>>+					atomic_read(&bc->bc_refcount));
>>+		return;
>>+	}
>>+
>>+	BUG_ON(bc == &init_bc);
>>+	verify_held(bc);
>>+	hlist_del(&bc->hash);
>>+	spin_unlock_irqrestore(&bc_hash_lock, flags);
>>+	kmem_cache_free(bc_cachep, bc);
>>+}
> 
> I wonder if it's safe and worthwhile to optimise away the local_irq_save():
> 
> 	if (atomic_dec_and_test(&bc->bc_refcount)) {
> 		spin_lock_irqsave(&bc_hash_lock, flags);
> 		if (atomic_read(&bc->bc_refcount) == 0) {
> 			free it
put_beancounter can happen from IRQ context.
so we need something like atomic_dec_and_lock_irqsave()
Oleg Nesterov proposed more details.

Thanks,
Kirill