On Thu, Jul 26, 2012 at 7:31 PM, Thomas Huehn <tho...@net.t-labs.tu-berlin.de> wrote: > Hi Johannes, > > Johannes Berg schrieb: > >>> >>> /home/johannes/sys/wireless/drivers/net/wireless/ti/wlcore/tx.c: In >>> function ‘wl1271_skb_queue_head’: >>> /home/johannes/sys/wireless/drivers/net/wireless/ti/wlcore/tx.c:622:48: >>> warning: ‘hlid’ may be used uninitialized in this function >>> [-Wuninitialized] >> >> Those changes make no sense anyway -- you should be able to pass the >> station pointer through if it previously used info->control.sta, instead >> of doing an (expensive) lookup. > > > This is the call path > > in main.c ... INIT_WORK(&wl->tx_work, wl1271_tx_work); > jumps to tx.c...wl1271_tx_work(struct work_struct *work) > calls wl1271_tx_work(struct work_struct *work) > calls wlcore_tx_work_locked(struct wl1271 *wl) > calls wl1271_skb_queue_head() > calls wl12xx_tx_get_hlid() > calls wl12xx_tx_get_hlid_ap() .... where sta is needed. > > But I could not find any point in the call path where I could grab the > sta pointer and pass it through. I will move the rcu into > wl12xx_tx_get_hlid_ap(), but I do not see any cheaper way here as using rcu.
I agree with Johannes. The find_sta() in the Tx path is unacceptable. A relatively simple change here would be to put the sta pointer in the portion of the struct reserved for rate control during op_tx(). wlcore doesn't use software rate control anyway. This would avoid from changing the code too much. _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
