Re: [PATCH] skge: fix broken driver

Mikulas Patocka [off-list ref] :

On Thu, 19 Sep 2013, Francois Romieu wrote:

[...]

quoted

Both patches don't behave exactly the same wrt pci_unmap_single.

[...]

I see, my patch passes a wrong value to pci_unmap_single. So I made this 
change to make it pass the correct value. Do you agree with this patch ?

Yes. I did not report it. Igor did.

You may "struct skge_element ee = *e;" and save a line. Who cares about
the extra copy when netdev_alloc_skb_ip_align fails ?

Something less ugly for the longer term
- use netdev_alloc_skb_ip_align in skge_rx_fill
- have skge_rx_setup return previouly stored sk_buff * - NULL if it was so -
  and ERR_PTR when it fails for whatever reason
- move netdev_alloc_skb_ip_align into skge_rx_setup
- pci_unmap in skge_rx_setup
- profit

Or isolate the struct sk_buff * + DEFINE_DMA_ part in skge_element then
save it as a whole in skge_rx_setup before initializing a new one as a
(netdev_alloc_skb_ip_align + pci_map).

Does someone volunteer to write it for net-next once the fix has been
merged and later pulled into net-next ?

-- 
Ueimor