-----Original Message-----
From: Dmitry Torokhov [mailto:dmitry.torokhov@gmail.com]
Sent: Wednesday, October 26, 2011 8:09 PM
To: KY Srinivasan
Cc: gregkh@suse.de; linux-kernel@vger.kernel.org;
devel@linuxdriverproject.org; virtualization@lists.osdl.org; linux-
input@vger.kernel.org; Haiyang Zhang; Jiri Kosina
Subject: Re: [PATCH 1/1] Staging: hv: mousevsc: Move the mouse driver out of
staging

On Sun, Oct 23, 2011 at 03:45:14PM +0000, KY Srinivasan wrote:

quoted

quoted

quoted

+
+	t = wait_for_completion_timeout(&input_dev->wait_event, 5*HZ);
+	if (t == 0) {
+		ret = -ETIMEDOUT;
+		goto cleanup;
+	}
+
+	response = &input_dev->protocol_resp;
+
+	if (!response->response.approved) {
+		pr_err("synthhid protocol request failed (version %d)",
+		       SYNTHHID_INPUT_VERSION);
+		ret = -ENODEV;
+		goto cleanup;
+	}
+
+	t = wait_for_completion_timeout(&input_dev->wait_event, 5*HZ);

We just completed the wait for this completion, why are we waiting on
the same completion again?

In response to our initial query, we expect the host to respond back with two
distinct pieces of information; we wait for both these responses.

I think you misunderstand how completion works in Linux. IIRC about
Windows events they are different ;) You can not signal completion
several times and then expect to wait corrsponding number of times. Once
you signal completion is it, well, complete.

Looking at the code for complete(), it looks like the "done" state is incremented
each time complete() is invoked and  the  code for do_wait_for_common() decrements the 
done state each time it is invoked (if the completion is properly signaled and we are not dealing
with a timeout. So, what am I missing here.

quoted

quoted

quoted

+	if (t == 0) {
+		ret = -ETIMEDOUT;
+		goto cleanup;
+	}
+
+	/*
+	 * We should have gotten the device attr, hid desc and report
+	 * desc at this point
+	 */
+	if (input_dev->dev_info_status)
+		ret = -ENOMEM;

-ENOMEM seems wrong.

There are many failures here and not being able to allocate memory is the
primary one; and so I chose to capture that.

Any chance that these failures have their own exit paths?

This is called from the probe functions and these errors are percolated up. On a different
note, I cleaned up this code based on the feedback I got from you. The patches have been 
sent out.

Regards,

K. Y