2022-12-21 07:34:10

by Christophe JAILLET

Subject: [PATCH] ixp4xx_eth: Fix an error handling path in ixp4xx_eth_probe()

If an error occurs after a successful ixp4xx_mdio_register() call, it
should be undone by a corresponding ixp4xx_mdio_remove().

Add the missing call in the error handling path, as already done in the
remove function.

Fixes: 2098c18d6cf6 ("IXP4xx: Add PHYLIB support to Ethernet driver.")
Signed-off-by: Christophe JAILLET <[email protected]>
---
drivers/net/ethernet/xscale/ixp4xx_eth.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/xscale/ixp4xx_eth.c b/drivers/net/ethernet/xscale/ixp4xx_eth.c
index 3b0c5f177447..007d68b385a5 100644
--- a/drivers/net/ethernet/xscale/ixp4xx_eth.c
+++ b/drivers/net/ethernet/xscale/ixp4xx_eth.c
@@ -1490,8 +1490,10 @@ static int ixp4xx_eth_probe(struct platform_device *pdev)

netif_napi_add_weight(ndev, &port->napi, eth_poll, NAPI_WEIGHT);

- if (!(port->npe = npe_request(NPE_ID(port->id))))
- return -EIO;
+ if (!(port->npe = npe_request(NPE_ID(port->id)))) {
+ err = -EIO;
+ goto err_remove_mdio;
+ }

port->plat = plat;
npe_port_tab[NPE_ID(port->id)] = port;
@@ -1530,6 +1532,8 @@ static int ixp4xx_eth_probe(struct platform_device *pdev)
err_free_mem:
npe_port_tab[NPE_ID(port->id)] = NULL;
npe_release(port->npe);
+err_remove_mdio:
+ ixp4xx_mdio_remove();
return err;
}

--
2.34.1

2022-12-21 18:54:59

by Michal Swiatkowski

Subject: Re: [PATCH] ixp4xx_eth: Fix an error handling path in ixp4xx_eth_probe()

On Wed, Dec 21, 2022 at 08:17:52AM +0100, Christophe JAILLET wrote:
> If an error occurs after a successful ixp4xx_mdio_register() call, it
> should be undone by a corresponding ixp4xx_mdio_remove().

What about error when mdio_bus is 0? It means that mdio_register can
return no error, but sth happen and there is no need to call mdio_remove
in this case?

I mean:
/* If the instance with the MDIO bus has not yet appeared,
* defer probing until it gets probed.
*/
if (!mdio_bus)
return -EPROBE_DEFER;
>
> Add the missing call in the error handling path, as already done in the
> remove function.
>
> Fixes: 2098c18d6cf6 ("IXP4xx: Add PHYLIB support to Ethernet driver.")
> Signed-off-by: Christophe JAILLET <[email protected]>
> ---
> drivers/net/ethernet/xscale/ixp4xx_eth.c | 8 ++++++--
> 1 file changed, 6 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/ethernet/xscale/ixp4xx_eth.c b/drivers/net/ethernet/xscale/ixp4xx_eth.c
> index 3b0c5f177447..007d68b385a5 100644
> --- a/drivers/net/ethernet/xscale/ixp4xx_eth.c
> +++ b/drivers/net/ethernet/xscale/ixp4xx_eth.c
> @@ -1490,8 +1490,10 @@ static int ixp4xx_eth_probe(struct platform_device *pdev)
>
> netif_napi_add_weight(ndev, &port->napi, eth_poll, NAPI_WEIGHT);
netif_napi_add_weight() doesn't need to be unrolled in case of error
(call netif_napi_del() or something)?

Thanks

>
> - if (!(port->npe = npe_request(NPE_ID(port->id))))
> - return -EIO;
> + if (!(port->npe = npe_request(NPE_ID(port->id)))) {
> + err = -EIO;
> + goto err_remove_mdio;
> + }
>
> port->plat = plat;
> npe_port_tab[NPE_ID(port->id)] = port;
> @@ -1530,6 +1532,8 @@ static int ixp4xx_eth_probe(struct platform_device *pdev)
> err_free_mem:
> npe_port_tab[NPE_ID(port->id)] = NULL;
> npe_release(port->npe);
> +err_remove_mdio:
> + ixp4xx_mdio_remove();
> return err;
> }
>
> --
> 2.34.1
>

2022-12-22 02:17:26

by Jakub Kicinski

Subject: Re: [PATCH] ixp4xx_eth: Fix an error handling path in ixp4xx_eth_probe()

On Wed, 21 Dec 2022 19:50:23 +0100 Michal Swiatkowski wrote:
> On Wed, Dec 21, 2022 at 08:17:52AM +0100, Christophe JAILLET wrote:
> > If an error occurs after a successful ixp4xx_mdio_register() call, it
> > should be undone by a corresponding ixp4xx_mdio_remove().
>
> What about error when mdio_bus is 0? It means that mdio_register can
> return no error, but sth happen and there is no need to call mdio_remove
> in this case?
>
> I mean:
> /* If the instance with the MDIO bus has not yet appeared,
> * defer probing until it gets probed.
> */
> if (!mdio_bus)
> return -EPROBE_DEFER;

Indeed, I think a closer look is warranted, this code operates on
global variables, it's also not clear to me whether it's safe to always
call remove, even if has_mdio is false?

> > Add the missing call in the error handling path, as already done in the
> > remove function.
> >
> > Fixes: 2098c18d6cf6 ("IXP4xx: Add PHYLIB support to Ethernet driver.")
> > Signed-off-by: Christophe JAILLET <[email protected]>
> > ---
> > drivers/net/ethernet/xscale/ixp4xx_eth.c | 8 ++++++--
> > 1 file changed, 6 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/xscale/ixp4xx_eth.c b/drivers/net/ethernet/xscale/ixp4xx_eth.c
> > index 3b0c5f177447..007d68b385a5 100644
> > --- a/drivers/net/ethernet/xscale/ixp4xx_eth.c
> > +++ b/drivers/net/ethernet/xscale/ixp4xx_eth.c
> > @@ -1490,8 +1490,10 @@ static int ixp4xx_eth_probe(struct platform_device *pdev)
> >
> > netif_napi_add_weight(ndev, &port->napi, eth_poll, NAPI_WEIGHT);
> netif_napi_add_weight() doesn't need to be unrolled in case of error
> (call netif_napi_del() or something)?

free_netdev() cleans it up automatically