On 05/23/2012 11:04 PM, Christoph Hellwig wrote:
>> +void resume_wait_epoch_requests(void)
>> +{
>> + struct request *req, *t;
>> +
>> + /* Request must have result SD_RES_OLD_NODE_VER
>> + * or SD_RES_NEW_NODE_VER in sys->wait_epoch_queue. */
>> + list_for_each_entry_safe(req, t, &sys->wait_rw_queue,
>> + request_list) {
>> +
>> + list_del(&req->request_list);
>> + list_add_tail(&req->request_list, &sys->request_queue);
>> + process_request_event_queues();
>> + }
>
> I think this should simply use list_splice_tail_init (which still needs
> to be ported from the Linux Kernel) to move all entries efficiently in
> one go, and then just call process_request_event_queues once.
>
Ah, I've pushed 'wait recovery' requests which should not be processed here
into this queue in my later patch, so I think list_splice_tail_init is not
suitable, but I can move process_request_event_queues() out of the loop, since
there's already a loop inside process_request_event_queues().
thanks,
levin
|
|