From: Ivo van Doorn <ivdoorn@gmail.com>
To: Johannes Berg <johannes@sipsolutions.net>
Subject: Re: ieee80211_tx_status: headroom too small
Date: Sat, 31 Oct 2009 10:30:45 +0100
Cc: Nikolai ZHUBR <zhubr@mail.ru>, linux-wireless@vger.kernel.org
References: <1094614475.20091030230028@mail.ru> <200910310959.43260.IvDoorn@gmail.com> <1256981113.3555.92.camel@johannes.local>
In-Reply-To: <1256981113.3555.92.camel@johannes.local>
MIME-Version: 1.0
Content-Type: text/plain;
  charset="iso-8859-15"
Message-Id: <200910311030.45582.IvDoorn@gmail.com>
Sender: linux-wireless-owner@vger.kernel.org

On Saturday 31 October 2009, Johannes Berg wrote:
> On Sat, 2009-10-31 at 09:59 +0100, Ivo van Doorn wrote:
> > On Saturday 31 October 2009, Johannes Berg wrote:
> > > On Sat, 2009-10-31 at 09:51 +0100, Ivo van Doorn wrote:
> > > 
> > > > > Ok. I looked at the problem and it's a very odd problem -- mac80211
> > > > > should always reserve enough headroom in buffers it passes to the
> > > > > driver. Does rt61pci somehow realloc frames?
> > > > 
> > > > It does use skb_push. but will/should always stay within the limits
> > > > of rt2x00dev->hw->extra_tx_headroom. Although I just realize there
> > > > might be a situation where it needs to align the frame to a 4-byte boundary,
> > > > it could exceed that amount. :(
> > > 
> > > extra_tx_headroom is set to max_t(driver, mac80211) so that there's
> > > always some headroom. This doesn't take into account alignment though.
> > > Should we simply add 2 to that value?
> > 
> > 4 actually, wasn't there some cases where the header could have an odd
> > number of bytes and would require 3 bytes for alignment?
> 
> Hm, is there? I thought they changed the 11s draft to no longer have
> that.

Well 2 is fine in that case, (I was basing my comment on something said on this mailinglist
a long time ago). :)

Ivo