| Message ID | 1248877744-7273-2-git-send-email-me@bobcopeland.com (mailing list archive) |
|---|---|
| State | Accepted, archived |
| Headers | show |
2009/7/29 Bob Copeland <me@bobcopeland.com>: > We need to process tx descriptors for all queues (currently main tx > queue and cabq) which may have triggered the TX completion interrupt. > Otherwise, the queues can get stuck after sending a few frames. > > Also, the CAB queue does need to be DBA (DMA beacon alert) gated, as > in ath9k and madwifi, instead of beacon sent gated or the CAB frames > may not be sent at the right time. > I think we are doing the wrong thing, CAB queue is supposed to fire after beacon queue. Ath9k fires up cab queue and beacon queue at the same time and uses ready time to hold cab queue for a fixed interval. Try to remove the readytime stuff from cab queue and i guess it'll work fine (now we fire it after the beacon queue AND wait for a beacon transmition period).
On Wed, Jul 29, 2009 at 4:06 PM, Nick Kossifidis<mickflemm@gmail.com> wrote: > 2009/7/29 Bob Copeland <me@bobcopeland.com>: > I think we are doing the wrong thing, CAB queue is supposed to fire > after beacon queue. Ath9k fires up cab queue and beacon queue at the > same time and uses ready time to hold cab queue for a fixed interval. > Try to remove the readytime stuff from cab queue and i guess it'll > work fine (now we fire it after the beacon queue AND wait for a beacon > transmition period). Yeah I was getting almost no delivery of buffered frames without that change, but I didn't check the delivery time to see if they were too late, which they probably are. So you are saying keep BCN_SENT_GT, but remove this: ath5k_hw_reg_write(ah, ((AR5K_TUNE_BEACON_INTERVAL - (AR5K_TUNE_SW_BEACON_RESP - AR5K_TUNE_DMA_BEACON_RESP) - AR5K_TUNE_ADDITIONAL_SWBA_BACKOFF) * 1024) | AR5K_QCU_RDYTIMECFG_ENABLE, AR5K_QUEUE_RDYTIMECFG(queue)); I'll give it ago. Maybe split that part out because the process tx queue fix is obvious enough. (Now I see that the similar-to-above-stuff from madwifi is only compat code and also uses a readyTime like ath9k. Ah.)
diff --git a/drivers/net/wireless/ath/ath5k/base.c b/drivers/net/wireless/ath/ath5k/base.c index 55ee976..3a1c156 100644 --- a/drivers/net/wireless/ath/ath5k/base.c +++ b/drivers/net/wireless/ath/ath5k/base.c @@ -2000,9 +2000,12 @@ ath5k_tx_processq(struct ath5k_softc *sc, struct ath5k_txq *txq) static void ath5k_tasklet_tx(unsigned long data) { + int i; struct ath5k_softc *sc = (void *)data; - ath5k_tx_processq(sc, sc->txq); + for (i=0; i < AR5K_NUM_TX_QUEUES; i++) + if (sc->txqs[i].setup && (sc->ah->ah_txq_isr & BIT(i))) + ath5k_tx_processq(sc, &sc->txqs[i]); } diff --git a/drivers/net/wireless/ath/ath5k/qcu.c b/drivers/net/wireless/ath/ath5k/qcu.c index 6d5aaf0..d7d5d11 100644 --- a/drivers/net/wireless/ath/ath5k/qcu.c +++ b/drivers/net/wireless/ath/ath5k/qcu.c @@ -409,7 +409,7 @@ int ath5k_hw_reset_tx_queue(struct ath5k_hw *ah, unsigned int queue) case AR5K_TX_QUEUE_CAB: AR5K_REG_ENABLE_BITS(ah, AR5K_QUEUE_MISC(queue), - AR5K_QCU_MISC_FRSHED_BCN_SENT_GT | + AR5K_QCU_MISC_FRSHED_DBA_GT | AR5K_QCU_MISC_CBREXP_DIS | AR5K_QCU_MISC_CBREXP_BCN_DIS);
We need to process tx descriptors for all queues (currently main tx queue and cabq) which may have triggered the TX completion interrupt. Otherwise, the queues can get stuck after sending a few frames. Also, the CAB queue does need to be DBA (DMA beacon alert) gated, as in ath9k and madwifi, instead of beacon sent gated or the CAB frames may not be sent at the right time. Signed-off-by: Bob Copeland <me@bobcopeland.com> --- drivers/net/wireless/ath/ath5k/base.c | 5 ++++- drivers/net/wireless/ath/ath5k/qcu.c | 2 +- 2 files changed, 5 insertions(+), 2 deletions(-)