Re: [PATCH v2] net/mlx5e: Use kvfree() in mlx5e_accel_fs_tcp_create()

[YueHaibing <yuehaibing@huawei.com>]

On 2022/11/10 15:31, Tariq Toukan wrote:
> 
> 
> On 11/8/2022 9:45 PM, Eric Dumazet wrote:
>> On Tue, Nov 8, 2022 at 9:58 AM Tariq Toukan <ttoukan.linux@gmail.com> wrote:
>>>
>>>
>>>
>>> On 11/8/2022 4:06 PM, YueHaibing wrote:
>>>> 'accel_tcp' is allocted by kvzalloc(), which should freed by kvfree().
>>>>
>>>> Fixes: f52f2faee581 ("net/mlx5e: Introduce flow steering API")
>>>> Signed-off-by: YueHaibing <yuehaibing@huawei.com>
>>>> ---
>>>> v2: fix the same issue in mlx5e_accel_fs_tcp_destroy() and a commit log typo
>>>> ---
>>>>    drivers/net/ethernet/mellanox/mlx5/core/en_accel/fs_tcp.c | 4 ++--
>>>>    1 file changed, 2 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/fs_tcp.c b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/fs_tcp.c
>>>> index 285d32d2fd08..d7c020f72401 100644
>>>> --- a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/fs_tcp.c
>>>> +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/fs_tcp.c
>>>> @@ -365,7 +365,7 @@ void mlx5e_accel_fs_tcp_destroy(struct mlx5e_flow_steering *fs)
>>>>        for (i = 0; i < ACCEL_FS_TCP_NUM_TYPES; i++)
>>>>                accel_fs_tcp_destroy_table(fs, i);
>>>>
>>>> -     kfree(accel_tcp);
>>>> +     kvfree(accel_tcp);
>>>>        mlx5e_fs_set_accel_tcp(fs, NULL);
>>>>    }
>>>>
>>>> @@ -397,7 +397,7 @@ int mlx5e_accel_fs_tcp_create(struct mlx5e_flow_steering *fs)
>>>>    err_destroy_tables:
>>>>        while (--i >= 0)
>>>>                accel_fs_tcp_destroy_table(fs, i);
>>>> -     kfree(accel_tcp);
>>>> +     kvfree(accel_tcp);
>>>>        mlx5e_fs_set_accel_tcp(fs, NULL);
>>>>        return err;
>>>>    }
>>>
>>> Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
>>>
>>> Thanks for your patch.
>>
>> Although this structure is 64 bytes... Not sure why kvmalloc() has
>> been used for this small chunk.
> 
> It's a small chunk indeed. Unnecessary usage of kvmalloc.
> 
> Although it's not critical (used only in slowpath), it'd be nice to clean it up and directly call kzalloc, instead of aligning the kfree().
> 
> YueHaibing, can you please submit a v3 for this?

Sure， will resend.
> 
> Regards,
> Tariq
> .