Re: [PATCH v4 04/13] xen/pvcalls: implement socket command and handle events

[Stefano Stabellini]

On Thu, 21 Sep 2017, Boris Ostrovsky wrote:
> > +
> > +static inline int get_request(struct pvcalls_bedata *bedata, int *req_id)
> > +{
> > +	*req_id = bedata->ring.req_prod_pvt & (RING_SIZE(&bedata->ring) - 1);
> > +	if (RING_FULL(&bedata->ring) ||
> > +	    READ_ONCE(bedata->rsp[*req_id].req_id) != PVCALLS_INVALID_ID)
> > +		return -EAGAIN;
> > +	return 0;
> > +}
> > +
> >   static irqreturn_t pvcalls_front_event_handler(int irq, void *dev_id)
> >   {
> > +	struct xenbus_device *dev = dev_id;
> > +	struct pvcalls_bedata *bedata;
> > +	struct xen_pvcalls_response *rsp;
> > +	uint8_t *src, *dst;
> > +	int req_id = 0, more = 0, done = 0;
> > +
> > +	if (dev == NULL)
> > +		return IRQ_HANDLED;
> > +
> > +	pvcalls_enter;
> > +	bedata = dev_get_drvdata(&dev->dev);
> > +	if (bedata == NULL) {
> > +		pvcalls_exit;
> > +		return IRQ_HANDLED;
> > +	}
> > +
> > +again:
> > +	while (RING_HAS_UNCONSUMED_RESPONSES(&bedata->ring)) {
> > +		rsp = RING_GET_RESPONSE(&bedata->ring, bedata->ring.rsp_cons);
> > +
> > +		req_id = rsp->req_id;
> > +		dst = (uint8_t *)&bedata->rsp[req_id] + sizeof(rsp->req_id);
> > +		src = (uint8_t *)rsp + sizeof(rsp->req_id);
> > +		memcpy(dst, src, sizeof(*rsp) - sizeof(rsp->req_id));
> > +		/*
> > +		 * First copy the rest of the data, then req_id. It is
> > +		 * paired with the barrier when accessing bedata->rsp.
> > +		 */
> > +		smp_wmb();
> > +		WRITE_ONCE(bedata->rsp[req_id].req_id, rsp->req_id);
> > +
> > +		done = 1;
> > +		bedata->ring.rsp_cons++;
> > +	}
> > +
> > +	RING_FINAL_CHECK_FOR_RESPONSES(&bedata->ring, more);
> > +	if (more)
> > +		goto again;
> > +	if (done)
> > +		wake_up(&bedata->inflight_req);
> > +	pvcalls_exit;
> >   	return IRQ_HANDLED;
> >   }
> >   +int pvcalls_front_socket(struct socket *sock)
> > +{
> > +	struct pvcalls_bedata *bedata;
> > +	struct sock_mapping *map = NULL;
> > +	struct xen_pvcalls_request *req;
> > +	int notify, req_id, ret;
> > +
> > +	pvcalls_enter;
> 
> 
> Can you define enter/exit macros with parentheses (i.e.
> "pvcalls_enter/exit()')?

Sure


> > +	if (!pvcalls_front_dev) {
> > +		pvcalls_exit;
> > +		return -EACCES;
> > +	}
> > +	/*
> > +	 * PVCalls only supports domain AF_INET,
> > +	 * type SOCK_STREAM and protocol 0 sockets for now.
> > +	 *
> > +	 * Check socket type here, AF_INET and protocol checks are done
> > +	 * by the caller.
> > +	 */
> > +	if (sock->type != SOCK_STREAM) {
> > +		pvcalls_exit;
> > +		return -ENOTSUPP;
> > +	}
> 
> 
> This can be done without entering pvcalls. I think you can enter pvcalls at
> least right before grabbing socket lock (you may need to move
> dev_get_drvdata() lower).
> 
> (possibly in later patches too)

OK


> > +
> > +	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
> > +
> > +	map = kzalloc(sizeof(*map), GFP_KERNEL);
> > +	if (map == NULL) {
> > +		pvcalls_exit;
> > +		return -ENOMEM;
> > +	}
> > +
> > +	spin_lock(&bedata->socket_lock);
> > +
> > +	ret = get_request(bedata, &req_id);
> > +	if (ret < 0) {
> > +		kfree(map);
> > +		spin_unlock(&bedata->socket_lock);
> > +		pvcalls_exit;
> > +		return ret;
> > +	}
> > +
> > +	/*
> > +	 * sock->sk->sk_send_head is not used for ip sockets: reuse the
> > +	 * field to store a pointer to the struct sock_mapping
> > +	 * corresponding to the socket. This way, we can easily get the
> > +	 * struct sock_mapping from the struct socket.
> > +	 */
> > +	sock->sk->sk_send_head = (void *)map;
> > +	list_add_tail(&map->list, &bedata->socket_mappings);
> > +
> > +	req = RING_GET_REQUEST(&bedata->ring, req_id);
> > +	req->req_id = req_id;
> > +	req->cmd = PVCALLS_SOCKET;
> > +	req->u.socket.id = (uint64_t) map;
> > +	req->u.socket.domain = AF_INET;
> > +	req->u.socket.type = SOCK_STREAM;
> > +	req->u.socket.protocol = IPPROTO_IP;
> > +
> > +	bedata->ring.req_prod_pvt++;
> > +	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
> > +	spin_unlock(&bedata->socket_lock);
> > +	if (notify)
> > +		notify_remote_via_irq(bedata->irq);
> > +
> > +	wait_event(bedata->inflight_req,
> > +		   READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
> > +
> > +	ret = bedata->rsp[req_id].ret;
> > +	/* read ret, then set this rsp slot to be reused */
> > +	smp_mb();
> > +	WRITE_ONCE(bedata->rsp[req_id].req_id, PVCALLS_INVALID_ID);
> 
> Now that I looked at what *_ONCE macros do I am not sure this is needed, given
> the smp_mb() above. In pvcalls_front_event_handler() too. And, in fact, in
> get_request() as well.
> 
> (Again, this will probably be also applicable to subsequent patches)

Yes, I think you are right, although I think is much easier to use
*_ONCE everywhere we don't want reads/writes to be mangled. Nonetheless,
I removed WRITE_ONCE from places where the write is done immediately
after a barrier. I kept *_ONCE elsewhere.