RE: [PATCH] net: stmmac: synchronize stmmac_open and stmmac_dvr_probe

["Kweh, Hock Leong" <hock.leong.kweh@intel.com>]

> -----Original Message-----
> From: Florian Fainelli [mailto:f.fainelli@gmail.com]
> Sent: Tuesday, December 27, 2016 1:14 PM
> To: Kweh, Hock Leong <hock.leong.kweh@intel.com>; David S. Miller
> <davem@davemloft.net>; Joao Pinto <Joao.Pinto@synopsys.com>; Giuseppe
> CAVALLARO <peppe.cavallaro@st.com>; seraphin.bonnaffe@st.com
> Cc: Alexandre TORGUE <alexandre.torgue@gmail.com>; Joachim Eastwood
> <manabian@gmail.com>; Niklas Cassel <niklas.cassel@axis.com>; Johan Hovold
> <johan@kernel.org>; pavel@ucw.cz; Ong, Boon Leong
> <boon.leong.ong@intel.com>; netdev <netdev@vger.kernel.org>; LKML <linux-
> kernel@vger.kernel.org>; Voon, Weifeng <weifeng.voon@intel.com>; Lars
> Persson <lars.persson@axis.com>
> Subject: Re: [PATCH] net: stmmac: synchronize stmmac_open and
> stmmac_dvr_probe
> 
> 
> 
> On 12/26/2016 09:10 PM, Florian Fainelli wrote:
> >
> >
> > On 12/27/2016 03:44 AM, Kweh, Hock Leong wrote:
> >> From: "Kweh, Hock Leong" <hock.leong.kweh@intel.com>
> >>
> >> If kernel module stmmac driver being loaded after OS booted, there is a
> >> race condition between stmmac_open() and stmmac_mdio_register(), which
> is
> >> invoked inside stmmac_dvr_probe(), and the error is showed in dmesg log as
> >> PHY not found and stmmac_open() failed:
> >> [  473.919358] stmmaceth 0000:01:00.0 (unnamed net_device) (uninitialized):
> >> 		stmmac_dvr_probe: warning: cannot get CSR clock
> >> [  473.919382] stmmaceth 0000:01:00.0: no reset control found
> >> [  473.919412] stmmac - user ID: 0x10, Synopsys ID: 0x42
> >> [  473.919429] stmmaceth 0000:01:00.0: DMA HW capability register
> supported
> >> [  473.919436] stmmaceth 0000:01:00.0: RX Checksum Offload Engine
> supported
> >> [  473.919443] stmmaceth 0000:01:00.0: TX Checksum insertion supported
> >> [  473.919451] stmmaceth 0000:01:00.0 (unnamed net_device) (uninitialized):
> >> 		Enable RX Mitigation via HW Watchdog Timer
> >> [  473.921395] libphy: PHY stmmac-1:00 not found
> >> [  473.921417] stmmaceth 0000:01:00.0 eth0: Could not attach to PHY
> >> [  473.921427] stmmaceth 0000:01:00.0 eth0: stmmac_open: Cannot attach
> to
> >> 		PHY (error: -19)
> >> [  473.959710] libphy: stmmac: probed
> >> [  473.959724] stmmaceth 0000:01:00.0 eth0: PHY ID 01410cc2 at 0 IRQ POLL
> >> 		(stmmac-1:00) active
> >> [  473.959728] stmmaceth 0000:01:00.0 eth0: PHY ID 01410cc2 at 1 IRQ POLL
> >> 		(stmmac-1:01)
> >> [  473.959731] stmmaceth 0000:01:00.0 eth0: PHY ID 01410cc2 at 2 IRQ POLL
> >> 		(stmmac-1:02)
> >> [  473.959734] stmmaceth 0000:01:00.0 eth0: PHY ID 01410cc2 at 3 IRQ POLL
> >> 		(stmmac-1:03)
> >>
> >> The resolution used wait_for_completion_interruptible() to synchronize
> >> stmmac_open() and stmmac_dvr_probe() to prevent the race condition
> >> happening.
> >
> > The proper fix for this would be to have register_netdev() be the last
> > thing done in stmmac_drv_probe(), whereas right now, the last thing done
> > is stmmac_mdio_register(), leading the window you are seeing here, where
> > the network interface can be open prior to all resources being set up,
> > including, but not limited to MDIO devices.
> 
> Something like the following untested patch should plug this race:
> 
> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> index bb40382e205d..5910ea51f8f6 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> @@ -3339,13 +3339,6 @@ int stmmac_dvr_probe(struct device *device,
> 
>         spin_lock_init(&priv->lock);
> 
> -       ret = register_netdev(ndev);
> -       if (ret) {
> -               netdev_err(priv->dev, "%s: ERROR %i registering the
> device\n",
> -                          __func__, ret);
> -               goto error_netdev_register;
> -       }
> -
>         /* If a specific clk_csr value is passed from the platform
>          * this means that the CSR Clock Range selection cannot be
>          * changed at run-time and it is fixed. Viceversa the driver'll
> try to
> @@ -3372,11 +3365,14 @@ int stmmac_dvr_probe(struct device *device,
>                 }
>         }
> 
> -       return 0;
> +       ret = register_netdev(ndev);
> +       if (ret)
> +               netdev_err(priv->dev, "%s: ERROR %i registering the
> device\n",
> +                          __func__, ret);
> +
> +       return ret;
> 
>  error_mdio_register:
> -       unregister_netdev(ndev);
> -error_netdev_register:
>         netif_napi_del(&priv->napi);
>  error_hw_init:
>         clk_disable_unprepare(priv->pclk);
> 
> --
> Florian

Thanks. Will try out to confirm.

Regards,
Wilson