Re: [PATCH v2 3/3] xen: optimize xenbus driver for multiple concurrent xenstore accesses

[Juergen Gross]

On 18/01/17 21:14, Boris Ostrovsky wrote:
> On 01/16/2017 09:15 AM, Juergen Gross wrote:
>> +
>> +static uint32_t xs_request_enter(struct xb_req_data *req)
>> +{
>> +	uint32_t rq_id;
>> +
>> +	req->type = req->msg.type;
>> +
>> +	spin_lock(&xs_state_lock);
>> +	for (;;) {
>> +		if (req->msg.tx_id != 0)
>> +			break;
>> +		if (xs_suspend_active) {
>> +			spin_unlock(&xs_state_lock);
>> +			wait_event(xs_state_enter_wq, xs_suspend_active == 0);
>> +			spin_lock(&xs_state_lock);
>> +			continue;
>> +		}
>> +		if (req->type == XS_TRANSACTION_START)
>> +			xs_state_users++;
>> +		break;
>> +	}
>> +	xs_state_users++;
>> +	rq_id = xs_request_id++;
>> +	spin_unlock(&xs_state_lock);
>> +
>> +	return rq_id;
>> +}
> 
> I should have noticed this last time but I've been looking at this code
> again and I don't think I understand why you are incrementing count for
> XS_TRANSACTION_START inside the loop.
> 
> In fact, why not just 'while(xs_suspend_active) {}' loop?

That's a valid question.

I'll change it. The reason to have the larger loop body isn't existing
any longer.


Juergen