On Wed, Nov 19, 2008 at 2:28 PM, Johannes Berg <johannes@xxxxxxxxxxxxxxxx> wrote: > On Wed, 2008-11-19 at 14:24 +0200, Tomas Winkler wrote: >> On Wed, Nov 19, 2008 at 1:40 PM, Johannes Berg >> <johannes@xxxxxxxxxxxxxxxx> wrote: >> > On Wed, 2008-11-19 at 11:21 +0200, Tomas Winkler wrote: >> > >> >> > Indeed. One would seem to be the buffer list itself, which also needs to >> >> > fit within a 40-bit mask with 256 alignment (rxq->bd), the rb_stts needs >> >> > 16-byte alignment which probably is always satisfied, but I'm not sure, >> >> > you'll need to research the alignment guarantees. >> >> >> >> Note these are allocated using persistent calls and not kmalloc I hope >> >> at least does are page alignment. >> > >> > Good point. On powerpc at least that uses alloc_pages internally. >> >> Not sure if if wouldn't be better just work directly with pages in RX? > > Well, in theory yes, but it's not possible to have skb->data point to a > separate page rather than a kmalloc area. And while you could allocate > an skb with very little headroom and the rest of the skb data in the > fraglist with pages, mac80211 couldn't handle it. I will look how this is handled in Ethernet drivers there should be some example with jumbo packets. Tomas -- To unsubscribe from this list: send the line "unsubscribe linux-wireless" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
