Re: [patch 6/13] signal/timer/event fds v7 - timerfd core ...

[Eric Dumazet <dada1@cosmosbay.com>]

Davide Libenzi a écrit :

> +struct timerfd_ctx {
> +	struct hrtimer tmr;
> +	ktime_t tintv;
> +	spinlock_t lock;
> +	wait_queue_head_t wqh;
> +	unsigned long ticks;
> +};

> +static struct kmem_cache *timerfd_ctx_cachep;

> +	timerfd_ctx_cachep = kmem_cache_create("timerfd_ctx_cache",
> +						sizeof(struct timerfd_ctx),
> +						0, SLAB_PANIC, NULL, NULL);


Do we really expect thousands of active timerfd_ctx ?

If not, using kmalloc()/kfree() would be fine, because sizeof(struct 
timerfd_ctx) is so small.

on SMP / NUMA platforms, each new kmem_cache is rather expensive. (memory 
allocated at kmem_cache_create(), but also memory used when cache is not 
empty, with slabs in freelist for each cpu/node)

Using a general cache might be cheaper : No memory overhead for yet another 
kmem_cache.

I know individual caches are good to spot memory leaks, but in timerfd case, 
you dont have mem leaks, do you ? :)