[PATCH v2] ath10k: support msdu chaining.

[Michal Kazior]

On 28 February 2014 02:20,  <greearb at candelatech.com> wrote:
> From: Ben Greear <greearb at candelatech.com>
>
> Consolidate the list of msdu skbs into the msdu-head skb, delete the
> rest of the skbs, pass the msdu-head skb on up the stack as normal.
>
> Tested with high-speed TCP and UDP traffic on modified firmware that
> supports raw-rx.

As for an umodified firmware I expect ath10k will just re-assemble
corrupted frames just to have them dropped by mac80211. It might be
worth checking:

 if (chaining && fcs_error)
    drop();

..to avoid processing junk and flushing d-cache (this matters for less
powerful host systems).


>
> Signed-off-by: Ben Greear <greearb at candelatech.com>
> ---
>
> v2:  Add note about skb_try_coalesce, though I am not sure it will
>      work or not.
>      Only grow skb once, which should save on some memory (re)allocation
>      and copying when we have more than one chained skb.
>
>  drivers/net/wireless/ath/ath10k/htt_rx.c | 60 +++++++++++++++++++++++++++++---
>  1 file changed, 56 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/net/wireless/ath/ath10k/htt_rx.c b/drivers/net/wireless/ath/ath10k/htt_rx.c
> index 9f3def7..a5070a9 100644
> --- a/drivers/net/wireless/ath/ath10k/htt_rx.c
> +++ b/drivers/net/wireless/ath/ath10k/htt_rx.c
> @@ -401,6 +401,7 @@ static int ath10k_htt_rx_amsdu_pop(struct ath10k_htt *htt,
>                 msdu_len = MS(__le32_to_cpu(rx_desc->msdu_start.info0),
>                               RX_MSDU_START_INFO0_MSDU_LENGTH);
>                 msdu_chained = rx_desc->frag_info.ring2_more_count;
> +               msdu_chaining = msdu_chained;
>
>                 if (msdu_len_invalid)
>                         msdu_len = 0;
> @@ -428,7 +429,6 @@ static int ath10k_htt_rx_amsdu_pop(struct ath10k_htt *htt,
>
>                         msdu->next = next;
>                         msdu = next;
> -                       msdu_chaining = 1;
>                 }
>
>                 last_msdu = __le32_to_cpu(rx_desc->msdu_end.info0) &
> @@ -1008,14 +1008,66 @@ continue_on:
>                                 continue;
>                         }
>
> -                       /* FIXME: we do not support chaining yet.
> -                        * this needs investigation */
>                         if (msdu_chaining) {
> -                               ath10k_warn("htt rx msdu_chaining is true\n");
> +                               struct sk_buff *next = msdu_head->next;
> +                               struct sk_buff *to_free = next;
> +                               static int do_once = 1;
> +                               int space;
> +                               int total_len = 0;
> +
> +                               /* TODO:  Might could optimize this by using
> +                                * skb_try_coalesce or similar method to]
> +                                * decrease copying.
> +                                */

It'd be great if we could use kernel-provided function to merge skbs
instead of rolling out our own ;-)


> +
> +                               msdu_head->next = NULL;
> +
> +                               if (unlikely(do_once)) {
> +                                       ath10k_warn("htt rx msdu_chaining detected %d\n",
> +                                                   msdu_chaining);
> +                                       do_once = 0;
> +                               }

I don't really like this do_once thing. I wouldn't even bother with a
warning, since, well, chaining parsing is implemented. This should be
a simple debug print if anything.


> +
> +                               /* Allocate total length all at once. */
> +                               while (next) {
> +                                       total_len += next->len;
> +                                       next = next->next;
> +                               }
> +
> +                               space = total_len - skb_tailroom(msdu_head);
> +                               if ((space > 0) &&
> +                                   (pskb_expand_head(msdu_head, 0, space, GFP_ATOMIC) < 0)) {
> +                                       /* TODO:  bump some rx-oom error stat */
> +                                       goto outside_continue;
> +                               }
> +
> +                               /* Walk list again, copying contents into
> +                                * msdu_head
> +                                */
> +                               next = to_free;
> +                               while (next) {
> +                                       skb_copy_from_linear_data(next, skb_put(msdu_head, next->len),
> +                                                                 next->len);
> +                                       next = next->next;
> +                               }
> +
> +                               /* If here, we have consolidated skb.  Free the
> +                                * fragments and pass the main skb on up the
> +                                * stack.
> +                                */
> +                               ath10k_htt_rx_free_msdu_chain(to_free);
> +                               goto have_good_msdu;
> +
> +                       outside_continue:
> +                               /* put it back together so we can free the
> +                                * whole list at once.
> +                                */
> +                               msdu_head->next = to_free;
>                                 ath10k_htt_rx_free_msdu_chain(msdu_head);
>                                 continue;
>                         }
>
> +               have_good_msdu:
>                         info.skb     = msdu_head;
>                         info.fcs_err = ath10k_htt_rx_has_fcs_err(msdu_head);
>                         info.mic_err = ath10k_htt_rx_has_mic_err(msdu_head);

If anything, please, make it into a separate function. The rx handler
is already huge and it needs shrinking down, not expanding it more.


Michał