Re: Handling incoming ZLP in cdc-wdm

[Oliver Neukum <oneukum@suse.com>]

[-- Attachment #1: Type: text/plain, Size: 766 bytes --]

Hi,

On 26.03.25 16:50, Robert Hodaszi wrote:

> The problem is, wdm_poll() always return with EPOLLIN even when wdm_in_callback() receives a ZLP, as it sets WDM_READ. So it makes sense for glib to think, there's a pending packet. In wdm_read(), if the packet's length is 0 (desc->length = 0) and WDM_READ is set, we reach
> 
>      if (!desc->length)
> 
> line, where it puts out another URB (as the resp_count is not 0), clear WDM_READ and go back to "retry". The second time we test WDM_READ, it is obviously not set yet, and as we are reading non-blocking, the function returns with EAGAIN.

Arguably the interrupt handler should set the flag for a readable result only
if indeed there is data in the buffer. Could you try the attached patch?

	Regards
		Oliver

[-- Attachment #2: 0001-usb-cdc-wdm-check-for-zero-length-response.patch --]
[-- Type: text/x-patch, Size: 834 bytes --]

From 47311e34cad1cc13ab5d2fbae239f188a6e82c1d Mon Sep 17 00:00:00 2001
From: Oliver Neukum <oneukum@suse.com>
Date: Thu, 27 Mar 2025 14:17:45 +0100
Subject: [PATCH] usb: cdc-wdm: check for zero length response

We should not indicate that there's a response if we have no
data in the buffer.

Signed-off-by: Oliver Neukum <oneukum@suse.com>
---
 drivers/usb/class/cdc-wdm.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/usb/class/cdc-wdm.c b/drivers/usb/class/cdc-wdm.c
index 86ee39db013f..856488a7cb6b 100644
--- a/drivers/usb/class/cdc-wdm.c
+++ b/drivers/usb/class/cdc-wdm.c
@@ -236,6 +236,8 @@ static void wdm_in_callback(struct urb *urb)
 		 */
 		schedule_work(&desc->service_outs_intr);
 	} else {
+		if (!desc->length)
+			goto out;
 		set_bit(WDM_READ, &desc->flags);
 		wake_up(&desc->wait);
 	}
-- 
2.49.0