[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <59e18374-0941-1298-4119-366452375404@suse.com>
Date: Wed, 2 Feb 2022 10:14:24 +0100
From: Oliver Neukum <oneukum@...e.com>
To: Oleksij Rempel <o.rempel@...gutronix.de>,
Greg KH <gregkh@...uxfoundation.org>
CC: Oliver Neukum <oneukum@...e.com>, netdev@...r.kernel.org,
linux-usb@...r.kernel.org, linux-kernel@...r.kernel.org,
kernel@...gutronix.de, Jakub Kicinski <kuba@...nel.org>,
"David S. Miller" <davem@...emloft.net>
Subject: Re: [PATCH net-next v1 1/1] usbnet: add devlink support
On 28.01.22 12:12, Oleksij Rempel wrote:
> The user space should get an event in case there is a problem with the
> USB transfers, i.e. the URB status is != 0.
That would need filtering for unlinks, but that is a detail
> The use space then can decide if the USB device needs to be reset, power
> cycled and so on.
This is highly problematic. Not necessarily impossible but problematic.
Executing URBs is part of the
* device removal and addition IO paths
* block IO path
* SCSI error handling
* PM transition IO paths
While we can send messages to user space, we cannot allocate a lot
of memory and we cannot wait for responses.
> What about calling a to-be-written devlink function that reports the USB
> status if the URB status is not 0:
>
> diff --git a/drivers/usb/core/hcd.c b/drivers/usb/core/hcd.c
> index d0f45600b669..a90134854f32 100644
> --- a/drivers/usb/core/hcd.c
> +++ b/drivers/usb/core/hcd.c
> @@ -1648,6 +1648,8 @@ static void __usb_hcd_giveback_urb(struct urb *urb)
> usb_unanchor_urb(urb);
> if (likely(status == 0))
> usb_led_activity(USB_LED_EVENT_HOST);
> + else
> + devlink_report_usb_status(urb, status);
>
>
It seems to me that you are approaching this issue on too low
a level. Drivers generally do at least rudimentary error handling
or rather detection.
This would be easy to use if you gave them a nice API. Something
like:
devlink_report_urb_error(usb_interface * intf, int reason, bool handling);
Maybe also report resets and hub events from the hub driver and
I think you'd get what you need. You need to be aware of a need for
some kind of limiting logic for disconnects.
The change over from logging something not very helpful to
properly reporting is easy.
Regards
Oliver
Powered by blists - more mailing lists