ath9k: FFT fixes and improvements
cover ath9k: FFT fixes and improvements
During FFT evaluation on an AR9462 adapter, we noticed that there were way less FFT samples received than we would expect. This patchset adds some counters to be able to check for received (and discarded) FFT samples, and fixes a variety of issues I found while debugging.
Cheers,
Simon
Simon Wunderlich (5):
ath9k: add counters for good and errorneous FFT/spectral frames
ath9k: return when short FFT frame was handled
ath9k: fix and simplify FFT max index retrieval
ath9k: FFT magnitude check: don't consider lower 3 data bits
ath9k: fix reporting calculated new FFT upper max
drivers/net/wireless/ath/ath9k/common-debug.c | 2 + drivers/net/wireless/ath/ath9k/common-debug.h | 4 ++ drivers/net/wireless/ath/ath9k/common-spectral.c | 83 +++++++++--------------- drivers/net/wireless/ath/ath9k/common-spectral.h | 17 +++++ 4 files changed, 55 insertions(+), 51 deletions(-)
Signature · 1 line
2.11.0
PATCH 1/5 ath9k: add counters for good and errorneous FFT/spectral frames
This is helpful to see whether spectral samples get discarded.
--- drivers/net/wireless/ath/ath9k/common-debug.c | 2 ++ drivers/net/wireless/ath/ath9k/common-debug.h | 4 ++++ drivers/net/wireless/ath/ath9k/common-spectral.c | 15 +++++++++++++-- 3 files changed, 19 insertions(+), 2 deletions(-)
drivers/net/wireless/ath/ath9k/common-debug.c +2 -0
@@ -144,6 +144,8 @@ static ssize_t read_file_recv(struct file *file, char __user *user_buf,RXS_ERR ("BEACONS" , rx_beacons);RXS_ERR ("FRAGS" , rx_frags);RXS_ERR ("SPECTRAL" , rx_spectral); +RXS_ERR ("SPECTRAL SMPL GOOD" , rx_spectral_sample_good); +RXS_ERR ("SPECTRAL SMPL ERR" , rx_spectral_sample_err);RXS_ERR ("CRC ERR" , crc_err);RXS_ERR ("DECRYPT CRC ERR" , decrypt_crc_err);
drivers/net/wireless/ath/ath9k/common-debug.h +4 -0
@@ -39,6 +39,8 @@* @rx_beacons: No. of beacons received. * @rx_frags: No. of rx-fragements received. * @rx_spectral: No of spectral packets received. +* @rx_spectral_sample_good: No. of good spectral samples +* @rx_spectral_sample_err: No. of good spectral samples */ struct ath_rx_stats {u32 rx_pkts_all;
@@ -58,6 +60,8 @@ struct ath_rx_stats {u32 rx_beacons;u32 rx_frags;u32 rx_spectral; +u32 rx_spectral_sample_good; +u32 rx_spectral_sample_err; };#ifdef CONFIG_ATH9K_COMMON_DEBUG
drivers/net/wireless/ath/ath9k/common-spectral.c +13 -2
@@ -501,6 +501,7 @@ int ath_cmn_process_fft(struct ath_spec_scan_priv *spec_priv, struct ieee80211_hu8 sample_buf[SPECTRAL_SAMPLE_MAX_LEN ] = {0 };struct ath_hw *ah = spec_priv->ah;struct ath_common *common =ath9k_hw_common (spec_priv->ah); +struct ath_softc *sc = (struct ath_softc *)common->priv;u8 num_bins, *vdata = (u8 *)hdr;struct ath_radar_info *radar_info;int len = rs->rs_datalen;
@@ -649,8 +650,13 @@ int ath_cmn_process_fft(struct ath_spec_scan_priv *spec_priv, struct ieee80211_h sample_buf, sample_len, sample_bytes); -fft_handler (rs, spec_priv, sample_buf, - tsf, freq, chan_type); + ret =fft_handler (rs, spec_priv, sample_buf, + tsf, freq, chan_type); + +if (ret ==0 ) +RX_STAT_INC (rx_spectral_sample_good); +else +RX_STAT_INC (rx_spectral_sample_err);memset (sample_buf,0 ,SPECTRAL_SAMPLE_MAX_LEN );
@@ -665,6 +671,11 @@ int ath_cmn_process_fft(struct ath_spec_scan_priv *spec_priv, struct ieee80211_h ret =fft_handler (rs, spec_priv, sample_start, tsf, freq, chan_type); +if (ret ==0 ) +RX_STAT_INC (rx_spectral_sample_good); +else +RX_STAT_INC (rx_spectral_sample_err); +/* Mix the received bins to the /dev/random * pool */
Signature · 1 line
2.11.0
Re: [PATCH 1/5] ath9k: add counters for good and errorneous FFT/spectral frames
Quoted 4 lines · Simon Wunderlich
This is helpful to see whether spectral samples get discarded.
Signed-off-by Simon Wunderlich <[email protected]>Signed-off-by Kalle Valo <[email protected]>
5 patches applied to ath-next branch of ath.git, thanks.
03224678c013 ath9k: add counters for good and errorneous FFT/spectral frames b796a6c04e65 ath9k: return when short FFT frame was handled 2f85786b8a57 ath9k: fix and simplify FFT max index retrieval 4e7a3fa5394e ath9k: FFT magnitude check: don't consider lower 3 data bits 4fb5837ac2bd ath9k: fix reporting calculated new FFT upper max
Signature · 3 lines
https://patchwork.kernel.org/patch/10605607/ https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
PATCH 2/5 ath9k: return when short FFT frame was handled
In reply to Simon WunderlichWith the loop break like this, there are false "FFT report truncated" messages because the iterator is not advanced as the check expects.
Instead, just return, for a single frame there is nothing left to be done anyways.
--- drivers/net/wireless/ath/ath9k/common-spectral.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/wireless/ath/ath9k/common-spectral.c +1 -1
@@ -686,7 +686,7 @@ int ath_cmn_process_fft(struct ath_spec_scan_priv *spec_priv, struct ieee80211_h* loop. */ if (len <= fft_len +2 ) -break ; +return 1 ; sample_start = &vdata[i +1 ];
Signature · 1 line
2.11.0
PATCH 3/5 ath9k: fix and simplify FFT max index retrieval
In reply to Simon WunderlichFFT max index retrieval was not retrieved correctly for HT20/HT40 FFT frames. Fixing the retrieval allows us to remove the fixup function as well. While at it, split the spectral_max_index function into versions for ht20 and ht40 to simplify the code.
--- drivers/net/wireless/ath/ath9k/common-spectral.c | 45 ++++-------------------- drivers/net/wireless/ath/ath9k/common-spectral.h | 17 +++++++++ 2 files changed, 23 insertions(+), 39 deletions(-)
drivers/net/wireless/ath/ath9k/common-spectral.c +6 -39
@@ -59,8 +59,7 @@ ath_cmn_max_idx_verify_ht20_fft(u8 *sample_end, int bytes_read) sample = sample_end -SPECTRAL_HT20_SAMPLE_LEN +1 ; - max_index =spectral_max_index (mag_info->all_bins, -SPECTRAL_HT20_NUM_BINS ); + max_index =spectral_max_index_ht20 (mag_info->all_bins); max_magnitude =spectral_max_magnitude (mag_info->all_bins); max_exp = mag_info->max_exp &0xf ;
@@ -100,12 +99,10 @@ ath_cmn_max_idx_verify_ht20_40_fft(u8 *sample_end, int bytes_read) sample = sample_end -SPECTRAL_HT20_40_SAMPLE_LEN +1 ; lower_mag =spectral_max_magnitude (mag_info->lower_bins); - lower_max_index =spectral_max_index (mag_info->lower_bins, -SPECTRAL_HT20_40_NUM_BINS ); + lower_max_index =spectral_max_index_ht40 (mag_info->lower_bins); upper_mag =spectral_max_magnitude (mag_info->upper_bins); - upper_max_index =spectral_max_index (mag_info->upper_bins, -SPECTRAL_HT20_40_NUM_BINS ); + upper_max_index =spectral_max_index_ht40 (mag_info->upper_bins); max_exp = mag_info->max_exp &0xf ;
@@ -117,17 +114,6 @@ ath_cmn_max_idx_verify_ht20_40_fft(u8 *sample_end, int bytes_read) ((upper_max_index < 1) || (lower_max_index < 1)))return -1 ; -/* Some time hardware messes up the index and adds -* the index of the middle point (dc_pos). Try to fix it. -*/ -if ((upper_max_index - dc_pos >0 ) && - (sample[upper_max_index] == (upper_mag >> max_exp))) - upper_max_index -= dc_pos; - -if ((lower_max_index - dc_pos >0 ) && - (sample[lower_max_index - dc_pos] == (lower_mag >> max_exp))) - lower_max_index -= dc_pos; -if ((sample[upper_max_index + dc_pos] != (upper_mag >> max_exp)) || (sample[lower_max_index] != (lower_mag >> max_exp)))return -1 ;
@@ -169,8 +155,7 @@ ath_cmn_process_ht20_fft(struct ath_rx_status *rs, magnitude =spectral_max_magnitude (mag_info->all_bins); fft_sample_20.max_magnitude =__cpu_to_be16 (magnitude); - max_index =spectral_max_index (mag_info->all_bins, -SPECTRAL_HT20_NUM_BINS ); + max_index =spectral_max_index_ht20 (mag_info->all_bins); fft_sample_20.max_index = max_index; bitmap_w =spectral_bitmap_weight (mag_info->all_bins);
@@ -302,12 +287,10 @@ ath_cmn_process_ht20_40_fft(struct ath_rx_status *rs, upper_mag =spectral_max_magnitude (mag_info->upper_bins); fft_sample_40.upper_max_magnitude =__cpu_to_be16 (upper_mag); - lower_max_index =spectral_max_index (mag_info->lower_bins, -SPECTRAL_HT20_40_NUM_BINS ); + lower_max_index =spectral_max_index_ht40 (mag_info->lower_bins); fft_sample_40.lower_max_index = lower_max_index; - upper_max_index =spectral_max_index (mag_info->upper_bins, -SPECTRAL_HT20_40_NUM_BINS ); + upper_max_index =spectral_max_index_ht40 (mag_info->upper_bins); fft_sample_40.upper_max_index = upper_max_index; lower_bitmap_w =spectral_bitmap_weight (mag_info->lower_bins);
@@ -331,22 +314,6 @@ ath_cmn_process_ht20_40_fft(struct ath_rx_status *rs, upper_mag >> max_exp, upper_max_index); -/* Some time hardware messes up the index and adds -* the index of the middle point (dc_pos). Try to fix it. -*/ -if ((upper_max_index - dc_pos >0 ) && - (fft_sample_40.data[upper_max_index] == (upper_mag >> max_exp))) { - upper_max_index -= dc_pos; - fft_sample_40.upper_max_index = upper_max_index; - } - -if ((lower_max_index - dc_pos >0 ) && - (fft_sample_40.data[lower_max_index - dc_pos] == - (lower_mag >> max_exp))) { - lower_max_index -= dc_pos; - fft_sample_40.lower_max_index = lower_max_index; - } -/* Check if we got the expected magnitude values at * the expected bins */
drivers/net/wireless/ath/ath9k/common-spectral.h +17 -0
@@ -145,6 +145,23 @@ static inline u8 spectral_max_index(u8 *bins, int num_bins)return m; } +static inline u8 spectral_max_index_ht40 (u8 *bins) +{ +u8 idx; + + idx =spectral_max_index (bins,SPECTRAL_HT20_40_NUM_BINS ); + +/* positive values and zero are starting at the beginning +* of the data field. +*/ +return idx % (SPECTRAL_HT20_40_NUM_BINS /2 ); +} + +static inline u8 spectral_max_index_ht20 (u8 *bins) +{ +return spectral_max_index (bins,SPECTRAL_HT20_NUM_BINS ); +} +/* return the bitmap weight from the all/upper/lower bins */ static inline u8 spectral_bitmap_weight (u8 *bins) {
Signature · 1 line
2.11.0
PATCH 4/5 ath9k: FFT magnitude check: don't consider lower 3 data bits
In reply to Simon WunderlichThere were a lot of Magnitude Mismatch while getting FFT samples on my hardware (Atheros AR9462. I've compared the reported magnitude with the data in the FFT bin, and the FFT bin was less accurate:
[ 5395.193030] ath: phy0: FFT HT20 frame: max mag 0x89,max_mag_idx 28, ,magnitude 0x89 max_exp 0, data[28] = 0x88 [ 5395.194525] ath: phy0: FFT HT20 frame: max mag 0x89,max_mag_idx 28, ,magnitude 0x89 max_exp 0, data[28] = 0x88 [ 5395.196012] ath: phy0: FFT HT20 frame: max mag 0x88,max_mag_idx 28, ,magnitude 0x88 max_exp 0, data[28] = 0x88 [ 5395.197509] ath: phy0: FFT HT20 frame: max mag 0x6C,max_mag_idx 28, ,magnitude 0x6C max_exp 0, data[28] = 0x68 [ 5395.199015] ath: phy0: FFT HT20 frame: max mag 0x78,max_mag_idx 28, ,magnitude 0x78 max_exp 0, data[28] = 0x78 [ 5395.200497] ath: phy0: FFT HT20 frame: max mag 0xA1,max_mag_idx 28, ,magnitude 0xA1 max_exp 0, data[28] = 0xA0 [ 5395.202011] ath: phy0: FFT HT20 frame: max mag 0x91,max_mag_idx 28, ,magnitude 0x91 max_exp 0, data[28] = 0x90 [ 5395.203482] ath: phy0: FFT HT20 frame: max mag 0x89,max_mag_idx 28, ,magnitude 0x89 max_exp 0, data[28] = 0x88 [ 5395.204999] ath: phy0: FFT HT20 frame: max mag 0x27,max_mag_idx 4, ,magnitude 0x27 max_exp 0, data[4] = 0x20 [ 5395.206461] ath: phy0: FFT HT20 frame: max mag 0x41,max_mag_idx 28, ,magnitude 0x41 max_exp 0, data[28] = 0x40 [ 5395.207977] ath: phy0: FFT HT20 frame: max mag 0x51,max_mag_idx 28, ,magnitude 0x51 max_exp 0, data[28] = 0x50 [ 5395.209454] ath: phy0: FFT HT20 frame: max mag 0x53,max_mag_idx 28, ,magnitude 0x53 max_exp 0, data[28] = 0x50 [ 5395.210940] ath: phy0: FFT HT20 frame: max mag 0x40,max_mag_idx 28, ,magnitude 0x40 max_exp 0, data[28] = 0x40 [ 5395.212441] ath: phy0: FFT HT20 frame: max mag 0x59,max_mag_idx 28, ,magnitude 0x59 max_exp 0, data[28] = 0x58 [ 5395.213932] ath: phy0: FFT HT20 frame: max mag 0x53,max_mag_idx 28, ,magnitude 0x53 max_exp 0, data[28] = 0x50 [ 5395.215428] ath: phy0: FFT HT20 frame: max mag 0x7D,max_mag_idx 28, ,magnitude 0x7D max_exp 0, data[28] = 0x78 [ 5395.216910] ath: phy0: FFT HT20 frame: max mag 0x8C,max_mag_idx 28, ,magnitude 0x8C max_exp 0, data[28] = 0x88 [ 5395.218413] ath: phy0: FFT HT20 frame: max mag 0x7B,max_mag_idx 28, ,magnitude 0x7B max_exp 0, data[28] = 0x78 [ 5395.219900] ath: phy0: FFT HT20 frame: max mag 0x43,max_mag_idx 28,
,magnitude 0x43 max_exp 0, data[28] = 0x40
It seems like the lower 3 bits on my hardware are always zeroed, but the magnitude matches otherwise. Therefore, let's not make the magnitude check so strict so we can get those samples released to userspace.
--- drivers/net/wireless/ath/ath9k/common-spectral.c | 19 +++++++++++-------- 1 file changed, 11 insertions(+), 8 deletions(-)
drivers/net/wireless/ath/ath9k/common-spectral.c +11 -8
@@ -71,7 +71,7 @@ ath_cmn_max_idx_verify_ht20_fft(u8 *sample_end, int bytes_read)if (bytes_read <SPECTRAL_HT20_SAMPLE_LEN && max_index <1 )return -1 ; -if (sample[max_index] != (max_magnitude >> max_exp)) +if ((sample[max_index] &0xf8 ) != ((max_magnitude >> max_exp) &0xf8 ))return -1 ;else return 0 ;
@@ -114,8 +114,10 @@ ath_cmn_max_idx_verify_ht20_40_fft(u8 *sample_end, int bytes_read) ((upper_max_index < 1) || (lower_max_index < 1)))return -1 ; -if ((sample[upper_max_index + dc_pos] != (upper_mag >> max_exp)) || - (sample[lower_max_index] != (lower_mag >> max_exp))) +if (((sample[upper_max_index + dc_pos] &0xf8 ) != + ((upper_mag >> max_exp) &0xf8 )) || + ((sample[lower_max_index] &0xf8 ) != + ((lower_mag >> max_exp) &0xf8 )))return -1 ;else return 0 ;
@@ -173,7 +175,8 @@ ath_cmn_process_ht20_fft(struct ath_rx_status *rs, magnitude >> max_exp, max_index); -if (fft_sample_20.data[max_index] != (magnitude >> max_exp)) { +if ((fft_sample_20.data[max_index] &0xf8 ) != + ((magnitude >> max_exp) &0xf8 )) {ath_dbg (common,SPECTRAL_SCAN ,"Magnitude mismatch !\n" ); ret =-1 ; }
@@ -317,10 +320,10 @@ ath_cmn_process_ht20_40_fft(struct ath_rx_status *rs,/* Check if we got the expected magnitude values at * the expected bins */ -if ((fft_sample_40.data[upper_max_index + dc_pos] - != (upper_mag >> max_exp)) || - (fft_sample_40.data[lower_max_index] - != (lower_mag >> max_exp))) { +if (((fft_sample_40.data[upper_max_index + dc_pos] &0xf8 ) + != ((upper_mag >> max_exp) &0xf8 )) || + ((fft_sample_40.data[lower_max_index] &0xf8 ) + != ((lower_mag >> max_exp) &0xf8 ))) {ath_dbg (common,SPECTRAL_SCAN ,"Magnitude mismatch !\n" ); ret =-1 ; }
Signature · 1 line
2.11.0
PATCH 5/5 ath9k: fix reporting calculated new FFT upper max
In reply to Simon Wunderlich--- drivers/net/wireless/ath/ath9k/common-spectral.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/wireless/ath/ath9k/common-spectral.c +1 -1
@@ -381,7 +381,7 @@ ath_cmn_process_ht20_40_fft(struct ath_rx_status *rs,ath_dbg (common,SPECTRAL_SCAN ,"Calculated new upper max 0x%X at %i\n" , - tmp_mag, i); + tmp_mag, fft_sample_40.upper_max_index); }else for (i = dc_pos; i <SPECTRAL_HT20_40_NUM_BINS ; i++) {if (fft_sample_40.data[i] == (upper_mag >> max_exp))
Signature · 1 line
2.11.0
Re: [PATCH 5/5] ath9k: fix reporting calculated new FFT upper max
Quoted 3 lines · Simon Wunderlich
Cc Nick Kossifidis <[email protected]>Signed-off-by Simon Wunderlich <[email protected]>Signed-off-by Kalle Valo <[email protected]>
No empty commit logs, please. But I can add that, just tell me what to add.
Signature · 3 lines
https://patchwork.kernel.org/patch/10605615/ https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
A message that is not in the archive (sent off-list or not yet received): [email protected]
Re: [PATCH 5/5] ath9k: fix reporting calculated new FFT upper max
Quoted 6 lines · CEST Kalle Valo
Quoted 3 lines · Simon Wunderlich
Cc Nick Kossifidis <[email protected]>Signed-off-by Simon Wunderlich <[email protected]>Signed-off-by Kalle Valo <[email protected]>No empty commit logs, please. But I can add that, just tell me what to add.
How about:
Since the debug print code is outside of the loop, it shouldn't use the loop iterator anymore but instead print the found maximum index.
Let me know if you need me to resend or anything else.
Thank you,
Simon
Re: [PATCH 5/5] ath9k: fix reporting calculated new FFT upper max
Quoted 6 lines · Simon Wunderlich
Quoted 5 lines · CEST Kalle Valo
Quoted 3 lines · Simon Wunderlich
Cc Nick Kossifidis <[email protected]>Signed-off-by Simon Wunderlich <[email protected]>Signed-off-by Kalle Valo <[email protected]>
Quoted 6 lines
Quoted 1 line
No empty commit logs, please. But I can add that, just tell me what to add.
How about:
Since the debug print code is outside of the loop, it shouldn't use the loop iterator anymore but instead print the found maximum index.
Perfect, thanks.
Quoted 1 line
Let me know if you need me to resend or anything else.
No need, I added this to the patch in the (not yet pushed) pending branch.
Signature · 1 line
Kalle Valo
Re: [PATCH 5/5] ath9k: fix reporting calculated new FFT upper max
Quoted 15 lines · CEST Kalle Valo
Quoted 6 lines · Simon Wunderlich
Quoted 5 lines · CEST Kalle Valo
Quoted 3 lines · Simon Wunderlich
Cc Nick Kossifidis <[email protected]>Signed-off-by Simon Wunderlich <[email protected]>Signed-off-by Kalle Valo <[email protected]>Quoted 7 lines
Quoted 2 lines
No empty commit logs, please. But I can add that, just tell me what to add.
How about:
Since the debug print code is outside of the loop, it shouldn't use the loop iterator anymore but instead print the found maximum index.
Quoted 6 lines
Perfect, thanks.
Quoted 1 line
Let me know if you need me to resend or anything else.
No need, I added this to the patch in the (not yet pushed) pending branch.
Awesome, thank you!
Cheers,
Simon
compiles OK, no interfere with normal card operations (IBSS) so it should be merged.
Quoted 26 lines · Simon Wunderlich
During FFT evaluation on an AR9462 adapter, we noticed that there were way less FFT samples received than we would expect. This patchset adds some counters to be able to check for received (and discarded) FFT samples, and fixes a variety of issues I found while debugging.
Cheers,
SimonSimon Wunderlich (5):
ath9k: add counters for good and errorneous FFT/spectral frames
ath9k: return when short FFT frame was handled
ath9k: fix and simplify FFT max index retrieval
ath9k: FFT magnitude check: don't consider lower 3 data bits
ath9k: fix reporting calculated new FFT upper maxdrivers/net/wireless/ath/ath9k/common-debug.c | 2 + drivers/net/wireless/ath/ath9k/common-debug.h | 4 ++ drivers/net/wireless/ath/ath9k/common-spectral.c | 83 +++++++++--------------- drivers/net/wireless/ath/ath9k/common-spectral.h | 17 +++++ 4 files changed, 55 insertions(+), 51 deletions(-)--
2.11.0