[openib-general] Re: [PATCH][26/26] IB: MAD cancel callbacks fromthread

[Sean Hefty]

>Roland Dreier wrote:
>> +void cancel_sends(void *data)
>> +{
>> +	struct ib_mad_agent_private *mad_agent_priv;
>> +	struct ib_mad_send_wr_private *mad_send_wr;
>> +	struct ib_mad_send_wc mad_send_wc;
>> +	unsigned long flags;
>> +
>> +	mad_agent_priv = (struct ib_mad_agent_private *)data;
>
>don't add casts to a void pointer, that's silly.

This is my bad.

>> +	mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
>> +	mad_send_wc.vendor_err = 0;
>> +
>> +	spin_lock_irqsave(&mad_agent_priv->lock, flags);
>> +	while (!list_empty(&mad_agent_priv->canceled_list)) {
>> +		mad_send_wr = list_entry(mad_agent_priv->canceled_list.next,
>> +					 struct ib_mad_send_wr_private,
>> +					 agent_list);
>> +
>> +		list_del(&mad_send_wr->agent_list);
>> +		spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
>> +
>> +		mad_send_wc.wr_id = mad_send_wr->wr_id;
>> +		mad_agent_priv->agent.send_handler(&mad_agent_priv->agent,
>> +						   &mad_send_wc);
>> +
>> +		kfree(mad_send_wr);
>> +		if (atomic_dec_and_test(&mad_agent_priv->refcount))
>> +			wake_up(&mad_agent_priv->wait);
>> +		spin_lock_irqsave(&mad_agent_priv->lock, flags);
>> +	}
>> +	spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
>
>dumb question... why is the lock dropped?  is it just for the
>send_handler(), or also for wr_id assigned, kfree, and wake_up() ?

The lock is dropped to avoid calling the user back with it held.  The if
statement / wake_up call near the bottom of the loop can be replaced with a
simple atomic_dec.  The test should always fail.  The lock is to protect
access to the canceled_list.

(Sorry about the mailer...)

- Sean