[v8] net: ethernet: add driver for Aurora VLSI NB8800 Ethernet controller

On 25/11/2015 17:16, Måns Rullgård wrote:

> Alexander Duyck writes:
> 
>> On Wed, Nov 25, 2015 at 5:04 AM, Måns Rullgård wrote:
>>
>>> Mason writes:
>>>
>>>> On 25/11/2015 13:45, Måns Rullgård wrote:
>>>>
>>>>> Mason wrote:
>>>>>
>>>>>> On 19/11/2015 14:02, Mans Rullgard wrote:
>>>>>>
>>>>>>> +  if (dma_mapping_error(&dev->dev, dma_addr)) {
>>>>>>> +          skb_free_frag(data);
>>>>>>> +          return -ENOMEM;
>>>>>>> +  }
>>>>>>
>>>>>> I'm back-porting this driver to 4.1
>>>>>>
>>>>>> skb_free_frag() was introduced in 4.2 by 181edb2bfa22b IIUC.
>>>>>>
>>>>>> +static inline void skb_free_frag(void *addr)
>>>>>> +{
>>>>>> +       __free_page_frag(addr);
>>>>>> +}
>>>>>>
>>>>>> Should I just copy the definition of __free_page_frag() ?
>>>>>
>>>>> Looks like it ought to work.  Try and find out.  Not that you'll ever
>>>>> hit that error condition unless you fake it.
>>>>
>>>> Turns out __free_pages_ok() is static and I'd rather not touch
>>>> mm/page_alloc.c in my back-port.
>>>>
>>>> Since you say the error condition is rare, I think I'll go with
>>>> the code that 181edb2bfa22b replaced (put_page, IIUC).
>>>>
>>>> #include <linux/version.h>
>>>> #if LINUX_VERSION_CODE < KERNEL_VERSION(4,2,0)
>>>> #define skb_free_frag(data) put_page(virt_to_head_page(data))
>>>> #else
>>>> #error DELETE ME NOW (see commit 181edb2bfa22b)
>>>> #endif
>>>
>>> You can simply put_page(page) instead since we already have the
>>> virt_to_head_page() a few lines up.
>>
>> What you could do is use __free_pages instead of __free_pages_ok.
>> Generally you will want to use __free_pages instead of put_page just
>> to avoid a bunch of unnecessary tests and function pointer accesses.
> 
> Note that this is on a should-never-happen error path, so there's no
> need to be counting cycles.

Per Mans' advice, I have locally committed this patch:


Regards.

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Mason <slash.tmp@free.fr> writes:

> On 25/11/2015 17:16, Måns Rullgård wrote:
>
>> Alexander Duyck writes:
>> 
>>> On Wed, Nov 25, 2015 at 5:04 AM, Måns Rullgård wrote:
>>>
>>>> Mason writes:
>>>>
>>>>> On 25/11/2015 13:45, Måns Rullgård wrote:
>>>>>
>>>>>> Mason wrote:
>>>>>>
>>>>>>> On 19/11/2015 14:02, Mans Rullgard wrote:
>>>>>>>
>>>>>>>> +  if (dma_mapping_error(&dev->dev, dma_addr)) {
>>>>>>>> +          skb_free_frag(data);
>>>>>>>> +          return -ENOMEM;
>>>>>>>> +  }
>>>>>>>
>>>>>>> I'm back-porting this driver to 4.1
>>>>>>>
>>>>>>> skb_free_frag() was introduced in 4.2 by 181edb2bfa22b IIUC.
>>>>>>>
>>>>>>> +static inline void skb_free_frag(void *addr)
>>>>>>> +{
>>>>>>> +       __free_page_frag(addr);
>>>>>>> +}
>>>>>>>
>>>>>>> Should I just copy the definition of __free_page_frag() ?
>>>>>>
>>>>>> Looks like it ought to work.  Try and find out.  Not that you'll ever
>>>>>> hit that error condition unless you fake it.
>>>>>
>>>>> Turns out __free_pages_ok() is static and I'd rather not touch
>>>>> mm/page_alloc.c in my back-port.
>>>>>
>>>>> Since you say the error condition is rare, I think I'll go with
>>>>> the code that 181edb2bfa22b replaced (put_page, IIUC).
>>>>>
>>>>> #include <linux/version.h>
>>>>> #if LINUX_VERSION_CODE < KERNEL_VERSION(4,2,0)
>>>>> #define skb_free_frag(data) put_page(virt_to_head_page(data))
>>>>> #else
>>>>> #error DELETE ME NOW (see commit 181edb2bfa22b)
>>>>> #endif
>>>>
>>>> You can simply put_page(page) instead since we already have the
>>>> virt_to_head_page() a few lines up.
>>>
>>> What you could do is use __free_pages instead of __free_pages_ok.
>>> Generally you will want to use __free_pages instead of put_page just
>>> to avoid a bunch of unnecessary tests and function pointer accesses.
>> 
>> Note that this is on a should-never-happen error path, so there's no
>> need to be counting cycles.
>
> Per Mans' advice, I have locally committed this patch:
>
> diff --git a/drivers/net/ethernet/aurora/nb8800.c b/drivers/net/ethernet/aurora/nb8800.c
> index ecc4a334c507..9f929be96b74 100644
> --- a/drivers/net/ethernet/aurora/nb8800.c
> +++ b/drivers/net/ethernet/aurora/nb8800.c
> @@ -39,6 +39,13 @@
>
>  #include "nb8800.h"
>
> +#include <linux/version.h>
> +#if LINUX_VERSION_CODE < KERNEL_VERSION(4,2,0)
> +#define skb_free_frag(data) put_page(page)
> +#else
> +#error DELETE ME NOW (see commit 181edb2bfa22)
> +#endif
> +
>  static void nb8800_tx_done(struct net_device *dev);
>  static int nb8800_dma_stop(struct net_device *dev);

That works only by accident due to how local variables are named.  If
you want a macro to emulate the new function, your first suggestion is
better.  I thought you were going to change the actual function.