Re: [PATCH 08/11] firewire-sbp-target: Add sbp_login.{c,h}

[Stefan Richter <stefanr@s5r6.in-berlin.de>]

On Apr 11 Chris Boot wrote:
> +static void session_check_for_reset(struct sbp_session *sess)
> +{
> +	bool card_valid = false;
> +
> +	spin_lock_bh(&sess->lock);
> +
> +	if (sess->card) {
> +		spin_lock_irq(&sess->card->lock);
> +		card_valid = (sess->card->local_node != NULL);
> +		spin_unlock_irq(&sess->card->lock);
> +
> +		if (!card_valid) {
> +			fw_card_put(sess->card);
> +			sess->card = NULL;
> +		}
> +	}
> +
> +	if (!card_valid || (sess->generation != sess->card->generation)) {
> +		pr_info("Waiting for reconnect from node: %016llx\n",
> +				sess->guid);
> +
> +		sess->node_id = -1;
> +		sess->reconnect_expires = get_jiffies_64() +
> +			((sess->reconnect_hold + 1) * HZ);
> +	}
> +
> +	spin_unlock_bh(&sess->lock);
> +}

The card->local_node != NULL test by itself is atomic, it does not benefit
from being wrapped by card->lock acquisition.

Well, OK, the lock effectively forces the compiler to determine the value
of card_valid only once.  If the lock weren't there I guess the compiler
might feel entitled to reload card->local_node in the second !card_valid
test.  But even if you lose ACCESS_ONCE behavior by removing the card->lock
acquisition, I can't see how that could be detrimental relative to the
current code.

I am wondering on the other hand if there isn't actually a dependency
between this local_node test and something else, e.g. the generation test.
I.e. might the locking be incomplete?  Not sure about that.  I think I
rather want to look at that again when I received the code through
mainline.

BTW "card_valid" sounds rather generic; maybe call it "topology_valid"?
Either way, it can turn valid or invalid any time when firewire-core gets
to handle a self-ID-complete event.
-- 
Stefan Richter
-=====-===-- -=-- -===-
http://arcgraph.de/sr/