Re: [PATCH net-next] atl1e: Atheros L1E Gigabit Ethernet driver

[Stephen Hemminger <shemminger@vyatta.com>]

On Mon, 14 Jul 2008 11:28:21 +0800
Jie Yang <Jie.Yang@Atheros.com> wrote:

> From: Jie Yang <jie.yang@atheros.com>
> 
> Full patch for the Atheros L1E Gigabit Ethernet driver.
> Supportring AR8121, AR8113 and AR8114
> 
> Signed-off-by: Jie Yang <jie.yang @atheros.com>
>
> +
> +struct atl1e_recv_ret_status {
> +       u16 seq_num;
> +       u16 hash_lo;
> +       __le32  word1;
> +       u16 pkt_flag;
> +       u16 err_flag;
> +       u16 hash_hi;
> +       u16 vtag;
> +} __attribute__((packed));

No need for packed if structure has no holes. And compiler is too stupid
to know that and generates worse code.

> +typedef enum {
> +       atl1e_10_half = 0,
> +       atl1e_10_full = 1,
> +       atl1e_100_half = 2,
> +       atl1e_100_full = 3
> +} atl1e_speed_duplex_type;

enum's are good, typedef's are bad.  Kernel style is not to use typedef's
except in a very few limited places like locking.


...

> +#ifdef module_param_array
> +               if (num_media_type > bd) {
> +#endif

shouldn't need to ifdef like that?? module_param_array is part of 2.6 always.

> +                       val = media_type[bd];
> +                       atl1e_validate_option(&val, &opt, pdev);
> +                       adapter->hw.media_type = (u16) val;
> +#ifdef module_param_array
> +               } else {
> +                       adapter->hw.media_type = (u16)(opt.def);
> +               }
> +#endif
> +       }
> +}
> +
> diff --git a/include/linux/pci_ids.h b/include/linux/pci_ids.h
> index e74b14a..9388130 100644
> --- a/include/linux/pci_ids.h
> +++ b/include/linux/pci_ids.h
> @@ -2181,6 +2181,7 @@
> 
>  #define PCI_VENDOR_ID_ATTANSIC         0x1969
>  #define PCI_DEVICE_ID_ATTANSIC_L1      0x1048
> +#define PCI_DEVICE_ID_ATTANSIC_L1E     0x1026

Don't add pci_ids just put in driver. Jeff made that decision because
the number of id's was just growing too large.