Message ID | 20090731181020.GE7963@makis (mailing list archive) |
---|---|
State | Not Applicable, archived |
Headers | show |
On Fri, Jul 31, 2009 at 11:10 AM, Nick Kossifidis<mick@madwifi-project.org> wrote: > Â * Get rid of calibration timer, instead use a software interrupt > Â to schedule the calibration tasklet. An example of itemizing one change. Explain the why, not the how. Luis -- To unsubscribe from this list: send the line "unsubscribe linux-wireless" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
2009/7/31 Luis R. Rodriguez <mcgrof@gmail.com>: > On Fri, Jul 31, 2009 at 11:10 AM, Nick > Kossifidis<mick@madwifi-project.org> wrote: >> Â * Get rid of calibration timer, instead use a software interrupt >> Â to schedule the calibration tasklet. > > An example of itemizing one change. Explain the why, not the how. > We 've discussed this with various people on wireless summit. a) We don't need a timer for this, there is no need for accuracy even with round_jiffies i think this is a waste of resources. Also we don't need to run calibration if we are idle (no interrupts). b) When we add ANI support we 'll just extend the poll function and calibration tasklet and handle all periodic phy calibration on one place (much cleaner). c) Having calibration on a tasklet is better since during calibration we can't transmit or receive (antennas are detached to measure noise floor), previously calibration could run in parallel with tx/rx and interfere (packet loss).
On Fri, Jul 31, 2009 at 11:25 AM, Nick Kossifidis<mickflemm@gmail.com> wrote: > 2009/7/31 Luis R. Rodriguez <mcgrof@gmail.com>: >> On Fri, Jul 31, 2009 at 11:10 AM, Nick >> Kossifidis<mick@madwifi-project.org> wrote: >>> Â * Get rid of calibration timer, instead use a software interrupt >>> Â to schedule the calibration tasklet. >> >> An example of itemizing one change. Explain the why, not the how. >> > > We 've discussed this with various people on wireless summit. Not everyone attended. > a) We don't need a timer for this, there is no need for accuracy > Â Â even with round_jiffies i think this is a waste of resources. > Â Â Also we don't need to run calibration if we are idle (no interrupts). > > b) When we add ANI support we 'll just extend the poll function > Â Â and calibration tasklet and handle all periodic phy calibration > Â Â on one place (much cleaner). > > c) Having calibration on a tasklet is better since during calibration > Â Â we can't transmit or receive (antennas are detached to measure > Â Â noise floor), previously calibration could run in parallel with tx/rx > Â Â and interfere (packet loss). It seems this good explanation can be mangled into the commit log entry. Luis -- To unsubscribe from this list: send the line "unsubscribe linux-wireless" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Fri, 2009-07-31 at 21:25 +0300, Nick Kossifidis wrote: > a) We don't need a timer for this, there is no need for accuracy > even with round_jiffies i think this is a waste of resources. > Also we don't need to run calibration if we are idle (no interrupts). It doesn't sound right to me. Can it be that the device gets so miscalibrated during the silence that it won't be able to receive signal once it's there? Say, the device is set to channel 1 but it actually listens to channel 3, and there are some weak beacons on channel 1. And then there is an issue of the frequency straying outside the permitted spectrum. Perhaps the transmission will trigger interrupts and thus recalibration, but in case of scanning, the first probe requests could be on wrong frequencies. I would probably leave the timer and allow the interrupts make the delays shorter.
On Fri, Jul 31, 2009 at 2:25 PM, Nick Kossifidis<mickflemm@gmail.com> wrote: > Â Â Also we don't need to run calibration if we are idle (no interrupts). I think this is the big win right now... > c) Having calibration on a tasklet is better since during calibration > Â Â we can't transmit or receive (antennas are detached to measure > Â Â noise floor), previously calibration could run in parallel with tx/rx > Â Â and interfere (packet loss). This can still happen, no? Two tasklets can run in parallel on different processors, as long as they are different tasklets. In practice, this won't happen much because tasklets run on the cpu that scheduled them, and irq affinity is such (at least on my hardware) that it's almost always the same CPU. But I think to make the above true it needs to stop the queues etc when doing calibration. Also we are missing tasklet_kill()?
2009/7/31 Pavel Roskin <proski@gnu.org>: > On Fri, 2009-07-31 at 21:25 +0300, Nick Kossifidis wrote: > >> a) We don't need a timer for this, there is no need for accuracy >> Â Â even with round_jiffies i think this is a waste of resources. >> Â Â Also we don't need to run calibration if we are idle (no interrupts). > > It doesn't sound right to me. > > Can it be that the device gets so miscalibrated during the silence that > it won't be able to receive signal once it's there? Â Say, the device is > set to channel 1 but it actually listens to channel 3, and there are > some weak beacons on channel 1. > Nope calibration deals mostly with noise immunity, it calculates the noise floor and fixes QAM constellation. It's not related to the synthesizer so it's not related to the channel frequency. > And then there is an issue of the frequency straying outside the > permitted spectrum. Â Perhaps the transmission will trigger interrupts > and thus recalibration, but in case of scanning, the first probe > requests could be on wrong frequencies. > On each reset we trigger calibration so no worries there.
On Fri, 2009-07-31 at 22:35 +0300, Nick Kossifidis wrote: > Nope calibration deals mostly with noise immunity, it calculates the > noise floor and fixes QAM constellation. It's not related to the synthesizer > so it's not related to the channel frequency. OK then, good to know.
2009/7/31 Bob Copeland <me@bobcopeland.com>: > On Fri, Jul 31, 2009 at 2:25 PM, Nick Kossifidis<mickflemm@gmail.com> wrote: >> Â Â Also we don't need to run calibration if we are idle (no interrupts). > > I think this is the big win right now... > >> c) Having calibration on a tasklet is better since during calibration >> Â Â we can't transmit or receive (antennas are detached to measure >> Â Â noise floor), previously calibration could run in parallel with tx/rx >> Â Â and interfere (packet loss). > > This can still happen, no? Â Two tasklets can run in parallel on > different processors, as long as they are different tasklets. > > In practice, this won't happen much because tasklets run on the > cpu that scheduled them, and irq affinity is such (at least on my > hardware) that it's almost always the same CPU. Â But I think to > make the above true it needs to stop the queues etc when doing > calibration. > ACK, how do we do that ? ieee80211_stop_queues / ieee80211_wake_queues > Also we are missing tasklet_kill()? > Yup, resending...
On Fri, Jul 31, 2009 at 10:52:42PM +0300, Nick Kossifidis wrote: > > make the above true it needs to stop the queues etc when doing > > calibration. > > ACK, how do we do that ? > ieee80211_stop_queues / ieee80211_wake_queues Yup. And disabling (at least rx and swba) interrupts on the device is also probably a good idea.
2009/8/1 Bob Copeland <me@bobcopeland.com>: > On Fri, Jul 31, 2009 at 10:52:42PM +0300, Nick Kossifidis wrote: >> > make the above true it needs to stop the queues etc when doing >> > calibration. >> >> ACK, how do we do that ? >> ieee80211_stop_queues / ieee80211_wake_queues > > Yup. Â And disabling (at least rx and swba) interrupts on the device > is also probably a good idea. > I thought about calling ath5k_hw_stop_rx_pcu but i don't think it's necessary i mean it won't receive anyway, why mess with pcu ?
On Sat, Aug 01, 2009 at 04:31:27PM +0300, Nick Kossifidis wrote: > I thought about calling ath5k_hw_stop_rx_pcu but i don't think it's > necessary i mean it won't receive anyway, why mess with pcu ? Yeah, agreed, we can just ignore rx.
diff --git a/drivers/net/wireless/ath/ath5k/ath5k.h b/drivers/net/wireless/ath/ath5k/ath5k.h index 1047a6c..76cf5b2 100644 --- a/drivers/net/wireless/ath/ath5k/ath5k.h +++ b/drivers/net/wireless/ath/ath5k/ath5k.h @@ -919,6 +919,12 @@ enum ath5k_int { AR5K_INT_NOCARD = 0xffffffff }; +/* Software interrupts used for calibration */ +enum ath5k_software_interrupt { + AR5K_SWI_FULL_CALIBRATION = 0x01, + AR5K_SWI_SHORT_CALIBRATION = 0x02, +}; + /* * Power management */ @@ -1123,6 +1129,15 @@ struct ath5k_hw { /* noise floor from last periodic calibration */ s32 ah_noise_floor; + /* Calibration timestamp */ + u32 ah_cal_tstamp; + + /* Calibration interval (secs) */ + u8 ah_cal_intval; + + /* Software interrupt mask */ + u8 ah_swi_mask; + /* * Function pointers */ @@ -1276,6 +1291,7 @@ extern int ath5k_hw_channel(struct ath5k_hw *ah, struct ieee80211_channel *chann /* PHY calibration */ extern int ath5k_hw_phy_calibrate(struct ath5k_hw *ah, struct ieee80211_channel *channel); extern int ath5k_hw_noise_floor_calibration(struct ath5k_hw *ah, short freq); +extern void ath5k_hw_calibration_poll(struct ath5k_hw *ah); /* Spur mitigation */ bool ath5k_hw_chan_has_spur_noise(struct ath5k_hw *ah, struct ieee80211_channel *channel); diff --git a/drivers/net/wireless/ath/ath5k/base.c b/drivers/net/wireless/ath/ath5k/base.c index b64731b..02fa71a 100644 --- a/drivers/net/wireless/ath/ath5k/base.c +++ b/drivers/net/wireless/ath/ath5k/base.c @@ -59,7 +59,7 @@ #include "reg.h" #include "debug.h" -static int ath5k_calinterval = 10; /* Calibrate PHY every 10 secs (TODO: Fixme) */ +static u8 ath5k_calinterval = 10; /* Calibrate PHY every 10 secs (TODO: Fixme) */ static int modparam_nohwcrypt; module_param_named(nohwcrypt, modparam_nohwcrypt, bool, S_IRUGO); MODULE_PARM_DESC(nohwcrypt, "Disable hardware encryption."); @@ -376,7 +376,7 @@ static int ath5k_stop_hw(struct ath5k_softc *sc); static irqreturn_t ath5k_intr(int irq, void *dev_id); static void ath5k_tasklet_reset(unsigned long data); -static void ath5k_calibrate(unsigned long data); +static void ath5k_tasklet_calibrate(unsigned long data); /* * Module init/exit functions @@ -799,8 +799,8 @@ ath5k_attach(struct pci_dev *pdev, struct ieee80211_hw *hw) tasklet_init(&sc->rxtq, ath5k_tasklet_rx, (unsigned long)sc); tasklet_init(&sc->txtq, ath5k_tasklet_tx, (unsigned long)sc); tasklet_init(&sc->restq, ath5k_tasklet_reset, (unsigned long)sc); + tasklet_init(&sc->calib, ath5k_tasklet_calibrate, (unsigned long)sc); tasklet_init(&sc->beacontq, ath5k_tasklet_beacon, (unsigned long)sc); - setup_timer(&sc->calib_tim, ath5k_calibrate, (unsigned long)sc); ret = ath5k_eeprom_read_mac(ah, mac); if (ret) { @@ -2366,7 +2366,7 @@ ath5k_init(struct ath5k_softc *sc) sc->curband = &sc->sbands[sc->curchan->band]; sc->imask = AR5K_INT_RXOK | AR5K_INT_RXERR | AR5K_INT_RXEOL | AR5K_INT_RXORN | AR5K_INT_TXDESC | AR5K_INT_TXEOL | - AR5K_INT_FATAL | AR5K_INT_GLOBAL; + AR5K_INT_FATAL | AR5K_INT_GLOBAL | AR5K_INT_SWI; ret = ath5k_reset(sc, NULL); if (ret) goto done; @@ -2383,8 +2383,8 @@ ath5k_init(struct ath5k_softc *sc) /* Set ack to be sent at low bit-rates */ ath5k_hw_set_ack_bitrate_high(ah, false); - mod_timer(&sc->calib_tim, round_jiffies(jiffies + - msecs_to_jiffies(ath5k_calinterval * 1000))); + /* Set PHY calibration timestamp and inteval */ + ah->ah_cal_intval = ath5k_calinterval; ret = 0; done: @@ -2477,7 +2477,6 @@ ath5k_stop_hw(struct ath5k_softc *sc) mmiowb(); mutex_unlock(&sc->lock); - del_timer_sync(&sc->calib_tim); tasklet_kill(&sc->rxtq); tasklet_kill(&sc->txtq); tasklet_kill(&sc->restq); @@ -2536,6 +2535,9 @@ ath5k_intr(int irq, void *dev_id) if (status & AR5K_INT_BMISS) { /* TODO */ } + if (status & AR5K_INT_SWI) { + tasklet_schedule(&sc->calib); + } if (status & AR5K_INT_MIB) { /* * These stats are also used for ANI i think @@ -2552,6 +2554,8 @@ ath5k_intr(int irq, void *dev_id) if (unlikely(!counter)) ATH5K_WARN(sc, "too many interrupts, giving up for now\n"); + ath5k_hw_calibration_poll(ah); + return IRQ_HANDLED; } @@ -2568,11 +2572,15 @@ ath5k_tasklet_reset(unsigned long data) * for temperature/environment changes. */ static void -ath5k_calibrate(unsigned long data) +ath5k_tasklet_calibrate(unsigned long data) { struct ath5k_softc *sc = (void *)data; struct ath5k_hw *ah = sc->ah; + /* Only full calibration for now */ + if (ah->ah_swi_mask != AR5K_SWI_FULL_CALIBRATION) + return; + ATH5K_DBG(sc, ATH5K_DEBUG_CALIBRATE, "channel %u/%x\n", ieee80211_frequency_to_channel(sc->curchan->center_freq), sc->curchan->hw_value); @@ -2590,8 +2598,8 @@ ath5k_calibrate(unsigned long data) ieee80211_frequency_to_channel( sc->curchan->center_freq)); - mod_timer(&sc->calib_tim, round_jiffies(jiffies + - msecs_to_jiffies(ath5k_calinterval * 1000))); + ah->ah_swi_mask = 0; + } diff --git a/drivers/net/wireless/ath/ath5k/base.h b/drivers/net/wireless/ath/ath5k/base.h index 778e422..667bd9d 100644 --- a/drivers/net/wireless/ath/ath5k/base.h +++ b/drivers/net/wireless/ath/ath5k/base.h @@ -177,6 +177,8 @@ struct ath5k_softc { struct ath5k_rfkill rf_kill; + struct tasklet_struct calib; /* calibration tasklet */ + spinlock_t block; /* protects beacon */ struct tasklet_struct beacontq; /* beacon intr tasklet */ struct ath5k_buf *bbuf; /* beacon buffer */ @@ -187,7 +189,6 @@ struct ath5k_softc { unsigned int nexttbtt; /* next beacon time in TU */ struct ath5k_txq *cabq; /* content after beacon */ - struct timer_list calib_tim; /* calibration timer */ int power_level; /* Requested tx power in dbm */ bool assoc; /* assocate state */ bool enable_beacon; /* true if beacons are on */ diff --git a/drivers/net/wireless/ath/ath5k/phy.c b/drivers/net/wireless/ath/ath5k/phy.c index 6afba98..d30bc3b 100644 --- a/drivers/net/wireless/ath/ath5k/phy.c +++ b/drivers/net/wireless/ath/ath5k/phy.c @@ -1104,6 +1104,26 @@ int ath5k_hw_channel(struct ath5k_hw *ah, struct ieee80211_channel *channel) PHY calibration \*****************/ +void +ath5k_hw_calibration_poll(struct ath5k_hw *ah) +{ + u32 current_time = (jiffies / HZ); + u32 cal_intval = ah->ah_cal_intval; + + if (!ah->ah_cal_tstamp) + ah->ah_cal_tstamp = current_time; + + /* For now we always do full calibration + * Mark software interrupt mask and fire software + * interrupt (bit gets auto-cleared) */ + if ((current_time - ah->ah_cal_tstamp) >= cal_intval) { + ah->ah_cal_tstamp = current_time; + ah->ah_swi_mask = AR5K_SWI_FULL_CALIBRATION; + AR5K_REG_ENABLE_BITS(ah, AR5K_CR, AR5K_CR_SWI); + } + +} + /** * ath5k_hw_noise_floor_calibration - perform PHY noise floor calibration *