2022-12-20 22:02:56

by Christophe JAILLET

Subject: [PATCH net] fjes: Fix an error handling path in fjes_probe()

A netif_napi_add() call is hidden in fjes_sw_init(). It should be undone
by a corresponding netif_napi_del() call in the error handling path of the
probe, as already done inthe remove function.

Fixes: 265859309a76 ("fjes: NAPI polling function")
Signed-off-by: Christophe JAILLET <[email protected]>
---
drivers/net/fjes/fjes_main.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/net/fjes/fjes_main.c b/drivers/net/fjes/fjes_main.c
index 2513be6d4e11..01b4c9c6adbd 100644
--- a/drivers/net/fjes/fjes_main.c
+++ b/drivers/net/fjes/fjes_main.c
@@ -1370,7 +1370,7 @@ static int fjes_probe(struct platform_device *plat_dev)
adapter->txrx_wq = alloc_workqueue(DRV_NAME "/txrx", WQ_MEM_RECLAIM, 0);
if (unlikely(!adapter->txrx_wq)) {
err = -ENOMEM;
- goto err_free_netdev;
+ goto err_del_napi;
}

adapter->control_wq = alloc_workqueue(DRV_NAME "/control",
@@ -1431,6 +1431,8 @@ static int fjes_probe(struct platform_device *plat_dev)
destroy_workqueue(adapter->control_wq);
err_free_txrx_wq:
destroy_workqueue(adapter->txrx_wq);
+err_del_napi:
+ netif_napi_del(&adapter->napi);
err_free_netdev:
free_netdev(netdev);
err_out:
--
2.34.1

2022-12-21 10:17:59

by Michal Swiatkowski

Subject: Re: [PATCH net] fjes: Fix an error handling path in fjes_probe()

On Tue, Dec 20, 2022 at 09:57:06PM +0100, Christophe JAILLET wrote:
> A netif_napi_add() call is hidden in fjes_sw_init(). It should be undone
> by a corresponding netif_napi_del() call in the error handling path of the
> probe, as already done inthe remove function.
>
> Fixes: 265859309a76 ("fjes: NAPI polling function")
> Signed-off-by: Christophe JAILLET <[email protected]>
> ---
> drivers/net/fjes/fjes_main.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/net/fjes/fjes_main.c b/drivers/net/fjes/fjes_main.c
> index 2513be6d4e11..01b4c9c6adbd 100644
> --- a/drivers/net/fjes/fjes_main.c
> +++ b/drivers/net/fjes/fjes_main.c
> @@ -1370,7 +1370,7 @@ static int fjes_probe(struct platform_device *plat_dev)
> adapter->txrx_wq = alloc_workqueue(DRV_NAME "/txrx", WQ_MEM_RECLAIM, 0);
> if (unlikely(!adapter->txrx_wq)) {
> err = -ENOMEM;
> - goto err_free_netdev;
> + goto err_del_napi;
> }
>
> adapter->control_wq = alloc_workqueue(DRV_NAME "/control",
> @@ -1431,6 +1431,8 @@ static int fjes_probe(struct platform_device *plat_dev)
> destroy_workqueue(adapter->control_wq);
> err_free_txrx_wq:
> destroy_workqueue(adapter->txrx_wq);
> +err_del_napi:
> + netif_napi_del(&adapter->napi);
> err_free_netdev:
> free_netdev(netdev);
> err_out:

Reviewed-by: Michal Swiatkowski <[email protected]>

I wonder if it won't be better to have fjes_sw_deinit() instead or
change fjes_sw_init to only netif_napi_add(). You know, to avoid another
bug here when someone add sth to the fjes_sw_deinit(). This is only
suggestion, patch looks fine.

> --
> 2.34.1
>

2022-12-22 01:58:44

by Jakub Kicinski

Subject: Re: [PATCH net] fjes: Fix an error handling path in fjes_probe()

On Tue, 20 Dec 2022 21:57:06 +0100 Christophe JAILLET wrote:
> A netif_napi_add() call is hidden in fjes_sw_init(). It should be undone
> by a corresponding netif_napi_del() call in the error handling path of the
> probe, as already done inthe remove function.

Ah, now I see why Michal asked about netif_napi_del() in the other
driver :S free_netdev() cleans up NAPIs.

2022-12-29 16:46:18

by Christophe JAILLET

Subject: Re: [PATCH net] fjes: Fix an error handling path in fjes_probe()

Le 21/12/2022 à 10:59, Michal Swiatkowski a écrit :
> On Tue, Dec 20, 2022 at 09:57:06PM +0100, Christophe JAILLET wrote:
>> A netif_napi_add() call is hidden in fjes_sw_init(). It should be undone
>> by a corresponding netif_napi_del() call in the error handling path of the
>> probe, as already done inthe remove function.
>>
>> Fixes: 265859309a76 ("fjes: NAPI polling function")
>> Signed-off-by: Christophe JAILLET <[email protected]>
>> ---
>> drivers/net/fjes/fjes_main.c | 4 +++-
>> 1 file changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/net/fjes/fjes_main.c b/drivers/net/fjes/fjes_main.c
>> index 2513be6d4e11..01b4c9c6adbd 100644
>> --- a/drivers/net/fjes/fjes_main.c
>> +++ b/drivers/net/fjes/fjes_main.c
>> @@ -1370,7 +1370,7 @@ static int fjes_probe(struct platform_device *plat_dev)
>> adapter->txrx_wq = alloc_workqueue(DRV_NAME "/txrx", WQ_MEM_RECLAIM, 0);
>> if (unlikely(!adapter->txrx_wq)) {
>> err = -ENOMEM;
>> - goto err_free_netdev;
>> + goto err_del_napi;
>> }
>>
>> adapter->control_wq = alloc_workqueue(DRV_NAME "/control",
>> @@ -1431,6 +1431,8 @@ static int fjes_probe(struct platform_device *plat_dev)
>> destroy_workqueue(adapter->control_wq);
>> err_free_txrx_wq:
>> destroy_workqueue(adapter->txrx_wq);
>> +err_del_napi:
>> + netif_napi_del(&adapter->napi);
>> err_free_netdev:
>> free_netdev(netdev);
>> err_out:
>
> Reviewed-by: Michal Swiatkowski <[email protected]>
>
> I wonder if it won't be better to have fjes_sw_deinit() instead or
> change fjes_sw_init to only netif_napi_add(). You know, to avoid another
> bug here when someone add sth to the fjes_sw_deinit(). This is only
> suggestion, patch looks fine.

hi,

based on Jakub's comment [1], free_netdev() already cleans up NAPIs (see
[2]).

So would it make more sense to remove netif_napi_del() from the
.remove() function instead?
The call looks useless to me now.

CJ

[1]: https://lore.kernel.org/all/[email protected]/
[2]: https://elixir.bootlin.com/linux/v6.2-rc1/source/net/core/dev.c#L10710

>
>> --
>> 2.34.1
>>
>

2023-01-04 11:32:07

by Michal Swiatkowski

Subject: Re: [PATCH net] fjes: Fix an error handling path in fjes_probe()

On Thu, Dec 29, 2022 at 05:34:22PM +0100, Christophe JAILLET wrote:
> Le 21/12/2022 ? 10:59, Michal Swiatkowski a ?crit?:
> > On Tue, Dec 20, 2022 at 09:57:06PM +0100, Christophe JAILLET wrote:
> > > A netif_napi_add() call is hidden in fjes_sw_init(). It should be undone
> > > by a corresponding netif_napi_del() call in the error handling path of the
> > > probe, as already done inthe remove function.
> > >
> > > Fixes: 265859309a76 ("fjes: NAPI polling function")
> > > Signed-off-by: Christophe JAILLET <[email protected]>
> > > ---
> > > drivers/net/fjes/fjes_main.c | 4 +++-
> > > 1 file changed, 3 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/net/fjes/fjes_main.c b/drivers/net/fjes/fjes_main.c
> > > index 2513be6d4e11..01b4c9c6adbd 100644
> > > --- a/drivers/net/fjes/fjes_main.c
> > > +++ b/drivers/net/fjes/fjes_main.c
> > > @@ -1370,7 +1370,7 @@ static int fjes_probe(struct platform_device *plat_dev)
> > > adapter->txrx_wq = alloc_workqueue(DRV_NAME "/txrx", WQ_MEM_RECLAIM, 0);
> > > if (unlikely(!adapter->txrx_wq)) {
> > > err = -ENOMEM;
> > > - goto err_free_netdev;
> > > + goto err_del_napi;
> > > }
> > > adapter->control_wq = alloc_workqueue(DRV_NAME "/control",
> > > @@ -1431,6 +1431,8 @@ static int fjes_probe(struct platform_device *plat_dev)
> > > destroy_workqueue(adapter->control_wq);
> > > err_free_txrx_wq:
> > > destroy_workqueue(adapter->txrx_wq);
> > > +err_del_napi:
> > > + netif_napi_del(&adapter->napi);
> > > err_free_netdev:
> > > free_netdev(netdev);
> > > err_out:
> >
> > Reviewed-by: Michal Swiatkowski <[email protected]>
> >
> > I wonder if it won't be better to have fjes_sw_deinit() instead or
> > change fjes_sw_init to only netif_napi_add(). You know, to avoid another
> > bug here when someone add sth to the fjes_sw_deinit(). This is only
> > suggestion, patch looks fine.
>
> hi,
>
> based on Jakub's comment [1], free_netdev() already cleans up NAPIs (see
> [2]).
>
> So would it make more sense to remove netif_napi_del() from the .remove()
> function instead?
> The call looks useless to me now.
>
> CJ
>
> [1]: https://lore.kernel.org/all/[email protected]/
> [2]: https://elixir.bootlin.com/linux/v6.2-rc1/source/net/core/dev.c#L10710
>

Yeah, it make more sense.

Thanks, Michal
> >
> > > --
> > > 2.34.1
> > >
> >
>