[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <3b8ed7fbe4fd879b34b40c74485e2871849ddc93.camel@ew.tq-group.com>
Date: Tue, 17 Dec 2024 14:44:12 +0100
From: Matthias Schiffer <matthias.schiffer@...tq-group.com>
To: Nícolas "F. R. A. Prado" <nfraprado@...labora.com>,
Saravana Kannan <saravanak@...gle.com>
Cc: kernel@...labora.com, linux-kernel@...r.kernel.org, Greg Kroah-Hartman
<gregkh@...uxfoundation.org>, "Rafael J.Wysocki" <rafael@...nel.org>,
linux@...tq-group.com
Subject: Re: [PATCH] driver core: Don't log intentional skip of device link
creation as error
On Mon, 2024-06-24 at 11:20 -0400, Nícolas F. R. A. Prado wrote:
> Commit ac66c5bbb437 ("driver core: Allow only unprobed consumers for
> SYNC_STATE_ONLY device links") introduced an early return in
> device_link_add() to prevent useless links from being created. However
> the calling function fw_devlink_create_devlink() unconditionally prints
> an error if device_link_add() didn't create a link, even in this case
> where it is intentionally skipping the link creation.
>
> Add a check to detect if the link wasn't created intentionally and in
> that case don't log an error.
While it was concluded that this patch was not needed to fix the author's issue,
I have come across a similar problem that I have not been able to solve:
I'm currently working on a TI AM335x, and after updating to a recent kernel
(currently 6.12.4) and fixing the ti-sysc driver to probe in a few cases where
it was broken before, I'm now seeing the following messages:
ti-sysc 44e31000.target-module: Failed to create device link (0x180) with ocp
ti-sysc 48040000.target-module: Failed to create device link (0x180) with ocp
The direct cause of the failure is the same as noted in this patch: The
44e31000.target-module and 48040000.target-module devices have already been
probed at this point, so applying it would hide the error message.
The attempt to create these device links results from a call to
__fw_devlink_pickup_dangling_consumers() moving consumers from
/ocp/interrupt-controller@...00000 to /ocp. This happens because the interrupt
controller driver (irq-omap-intc) is declared using IRQCHIP_DECLARE(), so its
fwnode doesn't have an associated device.
As the existence of fwnodes without device is the whole point of
__fw_devlink_pickup_dangling_consumers(), I imagine that this is a valid case
that should be supported properly. Would it make sense to reconsider this patch,
or is there a better solution? The condition for hiding the error message could
also be made stricter by introducing a new fwnode_link flag to mark links that
have been moved by __fw_devlink_pickup_dangling_consumers().
Best regards,
Matthias
>
> Fixes: ac66c5bbb437 ("driver core: Allow only unprobed consumers for SYNC_STATE_ONLY device links")
> Signed-off-by: Nícolas F. R. A. Prado <nfraprado@...labora.com>
> Tested-by: Nícolas F. R. A. Prado <nfraprado@...labora.com>
> ---
> drivers/base/core.c | 9 +++++++--
> 1 file changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/base/core.c b/drivers/base/core.c
> index 2b4c0624b704..5eaafe3a280c 100644
> --- a/drivers/base/core.c
> +++ b/drivers/base/core.c
> @@ -2187,8 +2187,13 @@ static int fw_devlink_create_devlink(struct device *con,
> }
>
> if (con != sup_dev && !device_link_add(con, sup_dev, flags)) {
> - dev_err(con, "Failed to create device link (0x%x) with %s\n",
> - flags, dev_name(sup_dev));
> + if (flags & DL_FLAG_SYNC_STATE_ONLY &&
> + con->links.status != DL_DEV_NO_DRIVER &&
> + con->links.status != DL_DEV_PROBING)
> + dev_dbg(con, "Skipping device link creation for probed device\n");
> + else
> + dev_err(con, "Failed to create device link (0x%x) with %s\n",
> + flags, dev_name(sup_dev));
> ret = -EINVAL;
> }
>
>
> ---
> base-commit: b992b79ca8bc336fa8e2c80990b5af80ed8f36fd
> change-id: 20240624-fwdevlink-probed-no-err-45d21feb05fd
>
> Best regards,
--
TQ-Systems GmbH | Mühlstraße 2, Gut Delling | 82229 Seefeld, Germany
Amtsgericht München, HRB 105018
Geschäftsführer: Detlef Schneider, Rüdiger Stahl, Stefan Schneider
https://www.tq-group.com/
Powered by blists - more mailing lists