Re: [PATCH 6/6] mac80211: Allow extra TX headroom to be consumed by drivers.

[Gertjan van Wingerde <gwingerde@gmail.com>]

On 11/23/09 11:41, Johannes Berg wrote:
> On Mon, 2009-11-23 at 08:16 +0100, Gertjan van Wingerde wrote:
>> Allow drivers to consume the extra TX headroom they are requesting instead
>> of assuming they will give it back. This specifically is the case for
>> the rt2x00 driver, as a large part of the requested extra TX headroom is
>> used to properly align the frame for DMA usage, and the frame is never
>> moved back to the original location.
>> Fix this by reserving TX headroom accounting for both the driver requested
>> amount and the special monitor interface header that needs to be added,
>> instead of just the maximum of the two.
>>
>> See http://marc.info/?l=linux-kernel&m=125892467801662&w=2 for details.
> 
>> -	local->tx_headroom = max_t(unsigned int , local->hw.extra_tx_headroom,
>> -				   sizeof(struct ieee80211_tx_status_rtap_hdr));
>> +	local->tx_headroom = local->hw.extra_tx_headroom +
>> +				   sizeof(struct ieee80211_tx_status_rtap_hdr);
> 
> Even though the radiotap header currently is only 13 bytes long, I don't
> really like this, you're not consuming all of extra_tx_headroom! I'd
> rather have a definition somewhere
> 
> #define MAC80211_TX_STATUS_HEADROOM		13
> 
> (with a BUILD_BUG_ON(sizeof(rtap_hdr) == MAC80211_TX_STATUS_HEADROOM))
> so that you can make the driver set
> 	extra_tx_headroom = max_t(MAC80211_TX_STATUS_HEADROOM, txi) + align;
> 

We're just talking about a small amount of bytes (max 8 in the rt2x00 case),
but I'll look into this direction.
Probably will have to combine patches 5 & 6 of the series then, though.

---
Gertjan.