On 08/01/2012 02:06 PM, Oliver Hartkopp wrote:
> Sorry for this potentially mangled mail from my webmail access ...
> 
> Fabio Baltieri <fabio.baltieri@gmail.com> hat am 1. August 2012 um 13:49
> geschrieben:
> 
>> +void devm_can_led_init(struct net_device *netdev)
>> +{
>> +  struct can_priv *priv = netdev_priv(netdev);
>> +  void *res;
>> +
>> +  res = devres_alloc(can_led_release, 0, GFP_KERNEL);
>> +  if (!res)
>> +          goto err_out;
>> +
>> +  priv->tx_led_trig_name = kasprintf(GFP_KERNEL, "%s-tx", netdev->name);
> 
> IMO putting a string with 8 or 9 bytes into a separate kmalloc memory sniplet is
> pretty oversized.
> Ok - these functions provide to hide the complexitiy for allocating and storing
> strings, which is
> definitely fine for path names and these kind of strings that are not known in
> length and probably
> more than 100 bytes long.
> 
> But in this case i would suggest to allocate a fixed space in can_priv, as we
> know the string length
> very good (IFNAMSZ + strlen("-tx")) and there's no reason to get all the
> overhead from three kmallocs
> instead of one for that small memory allocations.
> 
> So i would suggest:
> 
>> @@ -52,6 +53,13 @@ struct can_priv {
>>
>>    unsigned int echo_skb_max;
>>    struct sk_buff **echo_skb;
>> +
>> +#ifdef CONFIG_CAN_LEDS
>> +  struct led_trigger *tx_led_trig;
>> +  char *tx_led_trig_name;
> 
> char tx_led_trig_name[IFNAMSZ+4];
> 
>> +  struct led_trigger *rx_led_trig;
>> +  char *rx_led_trig_name;
> 
> char rx_led_trig_name[IFNAMSZ+4];
> 
>> +#endif
>>  };

Sounds reasonable. Please Introduce a #define for "IFNAMSZ + 4" and
document where the 4 comes from. Send a v6 patch, I'll force update the
git repo, after you have Oliver's Acked-by.

Marc

-- 
Pengutronix e.K.                  | Marc Kleine-Budde           |
Industrial Linux Solutions        | Phone: +49-231-2826-924     |
Vertretung West/Dortmund          | Fax:   +49-5121-206917-5555 |
Amtsgericht Hildesheim, HRA 2686  | http://www.pengutronix.de   |