Re: [PATCH net-next v2 1/2] net: stmmac: Return early if invalid in loongson_dwmac_fix_reset()

[Andrew Lunn]

On Thu, Jul 24, 2025 at 05:05:57PM +0800, Tiezhu Yang wrote:
> On 2025/7/24 上午10:26, Tiezhu Yang wrote:
> > On 2025/7/23 下午10:53, Andrew Lunn wrote:
> > > On Wed, Jul 23, 2025 at 06:00:55PM +0800, Tiezhu Yang wrote:
> > > > If the MAC controller does not connect to any PHY interface, there is a
> > > > missing clock, then the DMA reset fails.
> 
> ...
> 
> > > > +    if (value & DMA_BUS_MODE_SFT_RESET)
> > > > +        return -EINVAL;
> > > 
> > > What happens with this return value? Do you get an error message which
> > > gives a hint the PHY clock is missing? Would a netdev_err() make sense
> > > here?
> > 
> > Yes, I will use dev_err() rather than netdev_err() (because there is no
> > net_device member here) to do something like this:
> > 
> > diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac-loongson.c
> > b/drivers/net/ethernet/stmicro/stmmac/dwmac-loongson.c
> > index 6d10077666c7..4a7b2b11ecce 100644
> > --- a/drivers/net/ethernet/stmicro/stmmac/dwmac-loongson.c
> > +++ b/drivers/net/ethernet/stmicro/stmmac/dwmac-loongson.c
> > @@ -513,8 +513,11 @@ static int loongson_dwmac_fix_reset(void *priv,
> > void __iomem *ioaddr)
> >   {
> >          u32 value = readl(ioaddr + DMA_BUS_MODE);
> > 
> > -       if (value & DMA_BUS_MODE_SFT_RESET)
> > +       if (value & DMA_BUS_MODE_SFT_RESET) {
> > +               struct plat_stmmacenet_data *plat = priv;
> > +               dev_err(&plat->pdev->dev, "the PHY clock is missing\n");
> >                  return -EINVAL;
> > +       }
> > 
> >          value |= DMA_BUS_MODE_SFT_RESET;
> >          writel(value, ioaddr + DMA_BUS_MODE);
> 
> Oops, the above changes can not work well.
> 
> It can not use netdev_err() or dev_err() to print message with device info
> in loongson_dwmac_fix_reset() directly, this is because the type of "priv"
> argument is struct plat_stmmacenet_data and the "pdev" member of "priv" is
> NULL here, it will lead to the fatal error "Unable to handle kernel paging
> request at virtual address" when printing message.

Maybe it would be better to change fix_soc_reset() to have struct
stmmac_priv * as its first parameter. There are not too many user of
it, so it is not too big a change.

>         ret = stmmac_reset(priv, priv->ioaddr);
>         if (ret) {
> +               if (ret == -EINVAL)
> +                       netdev_err(priv->dev, "the PHY clock is missing\n");
> +

The problem with this is, you have no idea if EINVAL might of come
from somewhere else.

	Andrew