[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Tue, 04 Apr 2017 10:33:48 -0700 (PDT)
From: David Miller <davem@...emloft.net>
To: nsekhar@...com
Cc: grygorii.strashko@...com, linux-omap@...r.kernel.org,
netdev@...r.kernel.org
Subject: Re: [PATCH] net: ethernet: ti: cpsw: fix race condition during
open()
From: Sekhar Nori <nsekhar@...com>
Date: Mon, 3 Apr 2017 17:34:28 +0530
> TI's cpsw driver handles both OF and non-OF case for phy
> connect. Unfortunately of_phy_connect() returns NULL on
> error while phy_connect() returns ERR_PTR().
>
> To handle this, cpsw_slave_open() overrides the return value
> from phy_connect() to make it NULL or error.
>
> This leaves a small window, where cpsw_adjust_link() may be
> invoked for a slave while slave->phy pointer is temporarily
> set to -ENODEV (or some other error) before it is finally set
> to NULL.
>
> _cpsw_adjust_link() only handles the NULL case, and an oops
> results when ERR_PTR() is seen by it.
>
> Note that cpsw_adjust_link() checks PHY status for each
> slave whenever it is invoked. It can so happen that even
> though phy_connect() for a given slave returns error,
> _cpsw_adjust_link() is still called for that slave because
> the link status of another slave changed.
>
> Fix this by using a temporary pointer to store return value
> of {of_}phy_connect() and do a one-time write to slave->phy.
>
> Reviewed-by: Grygorii Strashko <grygorii.strashko@...com>
> Reported-by: Yan Liu <yan-liu@...com>
> Signed-off-by: Sekhar Nori <nsekhar@...com>
Applied, thank you.
Powered by blists - more mailing lists