On Wed, May 16, 2018 at 08:47:30PM +0300, Oleksandr Andrushchenko wrote:

On 05/16/2018 08:15 PM, Dmitry Torokhov wrote:

quoted

Hi Oleksandr,

On Mon, May 14, 2018 at 05:40:29PM +0300, Oleksandr Andrushchenko wrote:

quoted

@@ -211,93 +220,114 @@ static int xenkbd_probe(struct xenbus_device *dev,
  	if (!info->page)
  		goto error_nomem;
-	/* Set input abs params to match backend screen res */
-	abs = xenbus_read_unsigned(dev->otherend,
-				   XENKBD_FIELD_FEAT_ABS_POINTER, 0);
-	ptr_size[KPARAM_X] = xenbus_read_unsigned(dev->otherend,
-						  XENKBD_FIELD_WIDTH,
-						  ptr_size[KPARAM_X]);
-	ptr_size[KPARAM_Y] = xenbus_read_unsigned(dev->otherend,
-						  XENKBD_FIELD_HEIGHT,
-						  ptr_size[KPARAM_Y]);
-	if (abs) {
-		ret = xenbus_write(XBT_NIL, dev->nodename,
-				   XENKBD_FIELD_REQ_ABS_POINTER, "1");
-		if (ret) {
-			pr_warn("xenkbd: can't request abs-pointer\n");
-			abs = 0;
-		}
-	}
+	/*
+	 * The below are reverse logic, e.g. if the feature is set, then
+	 * do not expose the corresponding virtual device.
+	 */
+	with_kbd = !xenbus_read_unsigned(dev->nodename,
+					 XENKBD_FIELD_FEAT_DSBL_KEYBRD, 0);
-	touch = xenbus_read_unsigned(dev->nodename,
-				     XENKBD_FIELD_FEAT_MTOUCH, 0);
-	if (touch) {
+	with_ptr = !xenbus_read_unsigned(dev->nodename,
+					 XENKBD_FIELD_FEAT_DSBL_POINTER, 0);
+
+	/* Direct logic: if set, then create multi-touch device. */
+	with_mtouch = xenbus_read_unsigned(dev->nodename,
+					   XENKBD_FIELD_FEAT_MTOUCH, 0);
+	if (with_mtouch) {
  		ret = xenbus_write(XBT_NIL, dev->nodename,
  				   XENKBD_FIELD_REQ_MTOUCH, "1");
  		if (ret) {
  			pr_warn("xenkbd: can't request multi-touch");
-			touch = 0;
+			with_mtouch = 0;
  		}
  	}

Does it make sense to still end up calling xenkbd_connect_backend() when
all interfaces (keyboard, pointer, and multitouch) are disabled? Should
we do:

	if (!(with_kbd || || with_ptr || with_mtouch))
		return -ENXIO;

?

It does make sense. Then we probably need to move all xenbus_read_unsigned
calls to the very beginning of the .probe, so no memory allocations are made
which will be useless if we return -ENXIO, e.g. something like

static int xenkbd_probe(struct xenbus_device *dev,
                  const struct xenbus_device_id *id)
{
    int ret, i;
    bool with_mtouch, with_kbd, with_ptr;
    struct xenkbd_info *info;
    struct input_dev *kbd, *ptr, *mtouch;

<read with_mtouch, with_kbd, with_ptr here>

if (!(with_kbd | with_ptr | with_mtouch))
        return -ENXIO;

Does the above looks ok?

Yes. Another option is to keep the check where I suggested and do

	if (...) {
		ret = -ENXIO;
		goto error;
	}

Whichever you prefer is fine with me.

Thanks.

-- 
Dmitry