Re: [PATCH] ipc/msg.c: mitigate the lock contention with percpu counter

[Manfred Spraul <manfred@colorfullife.com>]

Hi Jiebin,

On 9/2/22 17:22, Jiebin Sun wrote:
> The msg_bytes and msg_hdrs atomic counters are frequently
> updated when IPC msg queue is in heavy use, causing heavy
> cache bounce and overhead. Change them to percpu_counters
> greatly improve the performance. Since there is one unique
> ipc namespace, additional memory cost is minimal.

With ipc namespaces, there is one struct per namespace, correct?

The cost is probably still ok, but the change log should be correct.


> @@ -1303,14 +1305,16 @@ void msg_init_ns(struct ipc_namespace *ns)
>   	ns->msg_ctlmnb = MSGMNB;
>   	ns->msg_ctlmni = MSGMNI;
>   
> -	atomic_set(&ns->msg_bytes, 0);
> -	atomic_set(&ns->msg_hdrs, 0);
> +	percpu_counter_init(&ns->percpu_msg_bytes, 0, GFP_KERNEL);
> +	percpu_counter_init(&ns->percpu_msg_hdrs, 0, GFP_KERNEL);
>   	ipc_init_ids(&ns->ids[IPC_MSG_IDS]);

These calls can fail. You must add error handling.

--

     Manfred