1 /*
2  * Copyright (c) 2010-2011 Atheros Communications Inc.
3  *
4  * Permission to use, copy, modify, and/or distribute this software for any
5  * purpose with or without fee is hereby granted, provided that the above
6  * copyright notice and this permission notice appear in all copies.
7  *
8  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
9  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
10  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
11  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
12  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
13  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
14  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
15  */
16 
17 #include <linux/export.h>
18 #include "hw.h"
19 #include "ar9003_phy.h"
20 
21 void ar9003_paprd_enable(struct ath_hw *ah, bool val)
22 {
23 	struct ath9k_channel *chan = ah->curchan;
24 	struct ar9300_eeprom *eep = &ah->eeprom.ar9300_eep;
25 
26 	/*
27 	 * 3 bits for modalHeader5G.papdRateMaskHt20
28 	 * is used for sub-band disabling of PAPRD.
29 	 * 5G band is divided into 3 sub-bands -- upper,
30 	 * middle, lower.
31 	 * if bit 30 of modalHeader5G.papdRateMaskHt20 is set
32 	 * -- disable PAPRD for upper band 5GHz
33 	 * if bit 29 of modalHeader5G.papdRateMaskHt20 is set
34 	 * -- disable PAPRD for middle band 5GHz
35 	 * if bit 28 of modalHeader5G.papdRateMaskHt20 is set
36 	 * -- disable PAPRD for lower band 5GHz
37 	 */
38 
39 	if (IS_CHAN_5GHZ(chan)) {
40 		if (chan->channel >= UPPER_5G_SUB_BAND_START) {
41 			if (le32_to_cpu(eep->modalHeader5G.papdRateMaskHt20)
42 								  & BIT(30))
43 				val = false;
44 		} else if (chan->channel >= MID_5G_SUB_BAND_START) {
45 			if (le32_to_cpu(eep->modalHeader5G.papdRateMaskHt20)
46 								  & BIT(29))
47 				val = false;
48 		} else {
49 			if (le32_to_cpu(eep->modalHeader5G.papdRateMaskHt20)
50 								  & BIT(28))
51 				val = false;
52 		}
53 	}
54 
55 	if (val) {
56 		ah->paprd_table_write_done = true;
57 		ath9k_hw_apply_txpower(ah, chan, false);
58 	}
59 
60 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_CTRL0_B0,
61 		      AR_PHY_PAPRD_CTRL0_PAPRD_ENABLE, !!val);
62 	if (ah->caps.tx_chainmask & BIT(1))
63 		REG_RMW_FIELD(ah, AR_PHY_PAPRD_CTRL0_B1,
64 			      AR_PHY_PAPRD_CTRL0_PAPRD_ENABLE, !!val);
65 	if (ah->caps.tx_chainmask & BIT(2))
66 		REG_RMW_FIELD(ah, AR_PHY_PAPRD_CTRL0_B2,
67 			      AR_PHY_PAPRD_CTRL0_PAPRD_ENABLE, !!val);
68 }
69 EXPORT_SYMBOL(ar9003_paprd_enable);
70 
71 static int ar9003_get_training_power_2g(struct ath_hw *ah)
72 {
73 	struct ath9k_channel *chan = ah->curchan;
74 	unsigned int power, scale, delta;
75 
76 	scale = ar9003_get_paprd_scale_factor(ah, chan);
77 	power = REG_READ_FIELD(ah, AR_PHY_POWERTX_RATE5,
78 			       AR_PHY_POWERTX_RATE5_POWERTXHT20_0);
79 
80 	delta = abs((int) ah->paprd_target_power - (int) power);
81 	if (delta > scale)
82 		return -1;
83 
84 	if (delta < 4)
85 		power -= 4 - delta;
86 
87 	return power;
88 }
89 
90 static int ar9003_get_training_power_5g(struct ath_hw *ah)
91 {
92 	struct ath_common *common = ath9k_hw_common(ah);
93 	struct ath9k_channel *chan = ah->curchan;
94 	unsigned int power, scale, delta;
95 
96 	scale = ar9003_get_paprd_scale_factor(ah, chan);
97 
98 	if (IS_CHAN_HT40(chan))
99 		power = REG_READ_FIELD(ah, AR_PHY_POWERTX_RATE8,
100 			AR_PHY_POWERTX_RATE8_POWERTXHT40_5);
101 	else
102 		power = REG_READ_FIELD(ah, AR_PHY_POWERTX_RATE6,
103 			AR_PHY_POWERTX_RATE6_POWERTXHT20_5);
104 
105 	power += scale;
106 	delta = abs((int) ah->paprd_target_power - (int) power);
107 	if (delta > scale)
108 		return -1;
109 
110 	switch (get_streams(ah->txchainmask)) {
111 	case 1:
112 		delta = 6;
113 		break;
114 	case 2:
115 		delta = 4;
116 		break;
117 	case 3:
118 		delta = 2;
119 		break;
120 	default:
121 		delta = 0;
122 		ath_dbg(common, CALIBRATE, "Invalid tx-chainmask: %u\n",
123 			ah->txchainmask);
124 	}
125 
126 	power += delta;
127 	return power;
128 }
129 
130 static int ar9003_paprd_setup_single_table(struct ath_hw *ah)
131 {
132 	struct ath_common *common = ath9k_hw_common(ah);
133 	static const u32 ctrl0[3] = {
134 		AR_PHY_PAPRD_CTRL0_B0,
135 		AR_PHY_PAPRD_CTRL0_B1,
136 		AR_PHY_PAPRD_CTRL0_B2
137 	};
138 	static const u32 ctrl1[3] = {
139 		AR_PHY_PAPRD_CTRL1_B0,
140 		AR_PHY_PAPRD_CTRL1_B1,
141 		AR_PHY_PAPRD_CTRL1_B2
142 	};
143 	int training_power;
144 	int i, val;
145 
146 	if (IS_CHAN_2GHZ(ah->curchan))
147 		training_power = ar9003_get_training_power_2g(ah);
148 	else
149 		training_power = ar9003_get_training_power_5g(ah);
150 
151 	ath_dbg(common, CALIBRATE, "Training power: %d, Target power: %d\n",
152 		training_power, ah->paprd_target_power);
153 
154 	if (training_power < 0) {
155 		ath_dbg(common, CALIBRATE,
156 			"PAPRD target power delta out of range\n");
157 		return -ERANGE;
158 	}
159 	ah->paprd_training_power = training_power;
160 
161 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_AM2AM, AR_PHY_PAPRD_AM2AM_MASK,
162 		      ah->paprd_ratemask);
163 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_AM2PM, AR_PHY_PAPRD_AM2PM_MASK,
164 		      ah->paprd_ratemask);
165 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_HT40, AR_PHY_PAPRD_HT40_MASK,
166 		      ah->paprd_ratemask_ht40);
167 
168 	for (i = 0; i < ah->caps.max_txchains; i++) {
169 		REG_RMW_FIELD(ah, ctrl0[i],
170 			      AR_PHY_PAPRD_CTRL0_USE_SINGLE_TABLE_MASK, 1);
171 		REG_RMW_FIELD(ah, ctrl1[i],
172 			      AR_PHY_PAPRD_CTRL1_ADAPTIVE_AM2PM_ENABLE, 1);
173 		REG_RMW_FIELD(ah, ctrl1[i],
174 			      AR_PHY_PAPRD_CTRL1_ADAPTIVE_AM2AM_ENABLE, 1);
175 		REG_RMW_FIELD(ah, ctrl1[i],
176 			      AR_PHY_PAPRD_CTRL1_ADAPTIVE_SCALING_ENA, 0);
177 		REG_RMW_FIELD(ah, ctrl1[i],
178 			      AR_PHY_PAPRD_CTRL1_PA_GAIN_SCALE_FACT_MASK, 181);
179 		REG_RMW_FIELD(ah, ctrl1[i],
180 			      AR_PHY_PAPRD_CTRL1_PAPRD_MAG_SCALE_FACT, 361);
181 		REG_RMW_FIELD(ah, ctrl1[i],
182 			      AR_PHY_PAPRD_CTRL1_ADAPTIVE_SCALING_ENA, 0);
183 		REG_RMW_FIELD(ah, ctrl0[i],
184 			      AR_PHY_PAPRD_CTRL0_PAPRD_MAG_THRSH, 3);
185 	}
186 
187 	ar9003_paprd_enable(ah, false);
188 
189 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
190 		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_PAPRD_LB_SKIP, 0x30);
191 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
192 		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_PAPRD_LB_ENABLE, 1);
193 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
194 		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_PAPRD_TX_GAIN_FORCE, 1);
195 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
196 		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_PAPRD_RX_BB_GAIN_FORCE, 0);
197 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
198 		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_PAPRD_IQCORR_ENABLE, 0);
199 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
200 		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_PAPRD_AGC2_SETTLING, 28);
201 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
202 		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_CF_PAPRD_TRAIN_ENABLE, 1);
203 	val = AR_SREV_9462(ah) ? 0x91 : 147;
204 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL2,
205 		      AR_PHY_PAPRD_TRAINER_CNTL2_CF_PAPRD_INIT_RX_BB_GAIN, val);
206 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
207 		      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_FINE_CORR_LEN, 4);
208 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
209 		      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_COARSE_CORR_LEN, 4);
210 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
211 		      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_NUM_CORR_STAGES, 7);
212 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
213 		      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_MIN_LOOPBACK_DEL, 1);
214 	if (AR_SREV_9485(ah) || AR_SREV_9462(ah) || AR_SREV_9550(ah))
215 		REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
216 			      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_QUICK_DROP,
217 			      -3);
218 	else
219 		REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
220 			      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_QUICK_DROP,
221 			      -6);
222 	val = AR_SREV_9462(ah) ? -10 : -15;
223 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
224 		      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_ADC_DESIRED_SIZE,
225 		      val);
226 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
227 		      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_BBTXMIX_DISABLE, 1);
228 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL4,
229 		      AR_PHY_PAPRD_TRAINER_CNTL4_CF_PAPRD_SAFETY_DELTA, 0);
230 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL4,
231 		      AR_PHY_PAPRD_TRAINER_CNTL4_CF_PAPRD_MIN_CORR, 400);
232 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL4,
233 		      AR_PHY_PAPRD_TRAINER_CNTL4_CF_PAPRD_NUM_TRAIN_SAMPLES,
234 		      100);
235 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_0_B0,
236 		      AR_PHY_PAPRD_PRE_POST_SCALING, 261376);
237 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_1_B0,
238 		      AR_PHY_PAPRD_PRE_POST_SCALING, 248079);
239 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_2_B0,
240 		      AR_PHY_PAPRD_PRE_POST_SCALING, 233759);
241 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_3_B0,
242 		      AR_PHY_PAPRD_PRE_POST_SCALING, 220464);
243 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_4_B0,
244 		      AR_PHY_PAPRD_PRE_POST_SCALING, 208194);
245 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_5_B0,
246 		      AR_PHY_PAPRD_PRE_POST_SCALING, 196949);
247 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_6_B0,
248 		      AR_PHY_PAPRD_PRE_POST_SCALING, 185706);
249 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_7_B0,
250 		      AR_PHY_PAPRD_PRE_POST_SCALING, 175487);
251 	return 0;
252 }
253 
254 static void ar9003_paprd_get_gain_table(struct ath_hw *ah)
255 {
256 	u32 *entry = ah->paprd_gain_table_entries;
257 	u8 *index = ah->paprd_gain_table_index;
258 	u32 reg = AR_PHY_TXGAIN_TABLE;
259 	int i;
260 
261 	memset(entry, 0, sizeof(ah->paprd_gain_table_entries));
262 	memset(index, 0, sizeof(ah->paprd_gain_table_index));
263 
264 	for (i = 0; i < PAPRD_GAIN_TABLE_ENTRIES; i++) {
265 		entry[i] = REG_READ(ah, reg);
266 		index[i] = (entry[i] >> 24) & 0xff;
267 		reg += 4;
268 	}
269 }
270 
271 static unsigned int ar9003_get_desired_gain(struct ath_hw *ah, int chain,
272 					    int target_power)
273 {
274 	int olpc_gain_delta = 0, cl_gain_mod;
275 	int alpha_therm, alpha_volt;
276 	int therm_cal_value, volt_cal_value;
277 	int therm_value, volt_value;
278 	int thermal_gain_corr, voltage_gain_corr;
279 	int desired_scale, desired_gain = 0;
280 	u32 reg_olpc  = 0, reg_cl_gain  = 0;
281 
282 	REG_CLR_BIT(ah, AR_PHY_PAPRD_TRAINER_STAT1,
283 		    AR_PHY_PAPRD_TRAINER_STAT1_PAPRD_TRAIN_DONE);
284 	desired_scale = REG_READ_FIELD(ah, AR_PHY_TPC_12,
285 				       AR_PHY_TPC_12_DESIRED_SCALE_HT40_5);
286 	alpha_therm = REG_READ_FIELD(ah, AR_PHY_TPC_19,
287 				     AR_PHY_TPC_19_ALPHA_THERM);
288 	alpha_volt = REG_READ_FIELD(ah, AR_PHY_TPC_19,
289 				    AR_PHY_TPC_19_ALPHA_VOLT);
290 	therm_cal_value = REG_READ_FIELD(ah, AR_PHY_TPC_18,
291 					 AR_PHY_TPC_18_THERM_CAL_VALUE);
292 	volt_cal_value = REG_READ_FIELD(ah, AR_PHY_TPC_18,
293 					AR_PHY_TPC_18_VOLT_CAL_VALUE);
294 	therm_value = REG_READ_FIELD(ah, AR_PHY_BB_THERM_ADC_4,
295 				     AR_PHY_BB_THERM_ADC_4_LATEST_THERM_VALUE);
296 	volt_value = REG_READ_FIELD(ah, AR_PHY_BB_THERM_ADC_4,
297 				    AR_PHY_BB_THERM_ADC_4_LATEST_VOLT_VALUE);
298 
299 	switch (chain) {
300 	case 0:
301 		reg_olpc = AR_PHY_TPC_11_B0;
302 		reg_cl_gain = AR_PHY_CL_TAB_0;
303 		break;
304 	case 1:
305 		reg_olpc = AR_PHY_TPC_11_B1;
306 		reg_cl_gain = AR_PHY_CL_TAB_1;
307 		break;
308 	case 2:
309 		reg_olpc = AR_PHY_TPC_11_B2;
310 		reg_cl_gain = AR_PHY_CL_TAB_2;
311 		break;
312 	default:
313 		ath_dbg(ath9k_hw_common(ah), CALIBRATE,
314 			"Invalid chainmask: %d\n", chain);
315 		break;
316 	}
317 
318 	olpc_gain_delta = REG_READ_FIELD(ah, reg_olpc,
319 					 AR_PHY_TPC_11_OLPC_GAIN_DELTA);
320 	cl_gain_mod = REG_READ_FIELD(ah, reg_cl_gain,
321 					 AR_PHY_CL_TAB_CL_GAIN_MOD);
322 
323 	if (olpc_gain_delta >= 128)
324 		olpc_gain_delta = olpc_gain_delta - 256;
325 
326 	thermal_gain_corr = (alpha_therm * (therm_value - therm_cal_value) +
327 			     (256 / 2)) / 256;
328 	voltage_gain_corr = (alpha_volt * (volt_value - volt_cal_value) +
329 			     (128 / 2)) / 128;
330 	desired_gain = target_power - olpc_gain_delta - thermal_gain_corr -
331 	    voltage_gain_corr + desired_scale + cl_gain_mod;
332 
333 	return desired_gain;
334 }
335 
336 static void ar9003_tx_force_gain(struct ath_hw *ah, unsigned int gain_index)
337 {
338 	int selected_gain_entry, txbb1dbgain, txbb6dbgain, txmxrgain;
339 	int padrvgnA, padrvgnB, padrvgnC, padrvgnD;
340 	u32 *gain_table_entries = ah->paprd_gain_table_entries;
341 
342 	selected_gain_entry = gain_table_entries[gain_index];
343 	txbb1dbgain = selected_gain_entry & 0x7;
344 	txbb6dbgain = (selected_gain_entry >> 3) & 0x3;
345 	txmxrgain = (selected_gain_entry >> 5) & 0xf;
346 	padrvgnA = (selected_gain_entry >> 9) & 0xf;
347 	padrvgnB = (selected_gain_entry >> 13) & 0xf;
348 	padrvgnC = (selected_gain_entry >> 17) & 0xf;
349 	padrvgnD = (selected_gain_entry >> 21) & 0x3;
350 
351 	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
352 		      AR_PHY_TX_FORCED_GAIN_FORCED_TXBB1DBGAIN, txbb1dbgain);
353 	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
354 		      AR_PHY_TX_FORCED_GAIN_FORCED_TXBB6DBGAIN, txbb6dbgain);
355 	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
356 		      AR_PHY_TX_FORCED_GAIN_FORCED_TXMXRGAIN, txmxrgain);
357 	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
358 		      AR_PHY_TX_FORCED_GAIN_FORCED_PADRVGNA, padrvgnA);
359 	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
360 		      AR_PHY_TX_FORCED_GAIN_FORCED_PADRVGNB, padrvgnB);
361 	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
362 		      AR_PHY_TX_FORCED_GAIN_FORCED_PADRVGNC, padrvgnC);
363 	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
364 		      AR_PHY_TX_FORCED_GAIN_FORCED_PADRVGND, padrvgnD);
365 	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
366 		      AR_PHY_TX_FORCED_GAIN_FORCED_ENABLE_PAL, 0);
367 	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
368 		      AR_PHY_TX_FORCED_GAIN_FORCE_TX_GAIN, 0);
369 	REG_RMW_FIELD(ah, AR_PHY_TPC_1, AR_PHY_TPC_1_FORCED_DAC_GAIN, 0);
370 	REG_RMW_FIELD(ah, AR_PHY_TPC_1, AR_PHY_TPC_1_FORCE_DAC_GAIN, 0);
371 }
372 
373 static inline int find_expn(int num)
374 {
375 	return fls(num) - 1;
376 }
377 
378 static inline int find_proper_scale(int expn, int N)
379 {
380 	return (expn > N) ? expn - 10 : 0;
381 }
382 
383 #define NUM_BIN 23
384 
385 static bool create_pa_curve(u32 *data_L, u32 *data_U, u32 *pa_table, u16 *gain)
386 {
387 	unsigned int thresh_accum_cnt;
388 	int x_est[NUM_BIN + 1], Y[NUM_BIN + 1], theta[NUM_BIN + 1];
389 	int PA_in[NUM_BIN + 1];
390 	int B1_tmp[NUM_BIN + 1], B2_tmp[NUM_BIN + 1];
391 	unsigned int B1_abs_max, B2_abs_max;
392 	int max_index, scale_factor;
393 	int y_est[NUM_BIN + 1];
394 	int x_est_fxp1_nonlin, x_tilde[NUM_BIN + 1];
395 	unsigned int x_tilde_abs;
396 	int G_fxp, Y_intercept, order_x_by_y, M, I, L, sum_y_sqr, sum_y_quad;
397 	int Q_x, Q_B1, Q_B2, beta_raw, alpha_raw, scale_B;
398 	int Q_scale_B, Q_beta, Q_alpha, alpha, beta, order_1, order_2;
399 	int order1_5x, order2_3x, order1_5x_rem, order2_3x_rem;
400 	int y5, y3, tmp;
401 	int theta_low_bin = 0;
402 	int i;
403 
404 	/* disregard any bin that contains <= 16 samples */
405 	thresh_accum_cnt = 16;
406 	scale_factor = 5;
407 	max_index = 0;
408 	memset(theta, 0, sizeof(theta));
409 	memset(x_est, 0, sizeof(x_est));
410 	memset(Y, 0, sizeof(Y));
411 	memset(y_est, 0, sizeof(y_est));
412 	memset(x_tilde, 0, sizeof(x_tilde));
413 
414 	for (i = 0; i < NUM_BIN; i++) {
415 		s32 accum_cnt, accum_tx, accum_rx, accum_ang;
416 
417 		/* number of samples */
418 		accum_cnt = data_L[i] & 0xffff;
419 
420 		if (accum_cnt <= thresh_accum_cnt)
421 			continue;
422 
423 		/* sum(tx amplitude) */
424 		accum_tx = ((data_L[i] >> 16) & 0xffff) |
425 		    ((data_U[i] & 0x7ff) << 16);
426 
427 		/* sum(rx amplitude distance to lower bin edge) */
428 		accum_rx = ((data_U[i] >> 11) & 0x1f) |
429 		    ((data_L[i + 23] & 0xffff) << 5);
430 
431 		/* sum(angles) */
432 		accum_ang = ((data_L[i + 23] >> 16) & 0xffff) |
433 		    ((data_U[i + 23] & 0x7ff) << 16);
434 
435 		accum_tx <<= scale_factor;
436 		accum_rx <<= scale_factor;
437 		x_est[i + 1] = (((accum_tx + accum_cnt) / accum_cnt) + 32) >>
438 		    scale_factor;
439 
440 		Y[i + 1] = ((((accum_rx + accum_cnt) / accum_cnt) + 32) >>
441 			    scale_factor) +
442 			    (1 << scale_factor) * max_index + 16;
443 
444 		if (accum_ang >= (1 << 26))
445 			accum_ang -= 1 << 27;
446 
447 		theta[i + 1] = ((accum_ang * (1 << scale_factor)) + accum_cnt) /
448 		    accum_cnt;
449 
450 		max_index++;
451 	}
452 
453 	/*
454 	 * Find average theta of first 5 bin and all of those to same value.
455 	 * Curve is linear at that range.
456 	 */
457 	for (i = 1; i < 6; i++)
458 		theta_low_bin += theta[i];
459 
460 	theta_low_bin = theta_low_bin / 5;
461 	for (i = 1; i < 6; i++)
462 		theta[i] = theta_low_bin;
463 
464 	/* Set values at origin */
465 	theta[0] = theta_low_bin;
466 	for (i = 0; i <= max_index; i++)
467 		theta[i] -= theta_low_bin;
468 
469 	x_est[0] = 0;
470 	Y[0] = 0;
471 	scale_factor = 8;
472 
473 	/* low signal gain */
474 	if (x_est[6] == x_est[3])
475 		return false;
476 
477 	G_fxp =
478 	    (((Y[6] - Y[3]) * 1 << scale_factor) +
479 	     (x_est[6] - x_est[3])) / (x_est[6] - x_est[3]);
480 
481 	/* prevent division by zero */
482 	if (G_fxp == 0)
483 		return false;
484 
485 	Y_intercept =
486 	    (G_fxp * (x_est[0] - x_est[3]) +
487 	     (1 << scale_factor)) / (1 << scale_factor) + Y[3];
488 
489 	for (i = 0; i <= max_index; i++)
490 		y_est[i] = Y[i] - Y_intercept;
491 
492 	for (i = 0; i <= 3; i++) {
493 		y_est[i] = i * 32;
494 		x_est[i] = ((y_est[i] * 1 << scale_factor) + G_fxp) / G_fxp;
495 	}
496 
497 	if (y_est[max_index] == 0)
498 		return false;
499 
500 	x_est_fxp1_nonlin =
501 	    x_est[max_index] - ((1 << scale_factor) * y_est[max_index] +
502 				G_fxp) / G_fxp;
503 
504 	order_x_by_y =
505 	    (x_est_fxp1_nonlin + y_est[max_index]) / y_est[max_index];
506 
507 	if (order_x_by_y == 0)
508 		M = 10;
509 	else if (order_x_by_y == 1)
510 		M = 9;
511 	else
512 		M = 8;
513 
514 	I = (max_index > 15) ? 7 : max_index >> 1;
515 	L = max_index - I;
516 	scale_factor = 8;
517 	sum_y_sqr = 0;
518 	sum_y_quad = 0;
519 	x_tilde_abs = 0;
520 
521 	for (i = 0; i <= L; i++) {
522 		unsigned int y_sqr;
523 		unsigned int y_quad;
524 		unsigned int tmp_abs;
525 
526 		/* prevent division by zero */
527 		if (y_est[i + I] == 0)
528 			return false;
529 
530 		x_est_fxp1_nonlin =
531 		    x_est[i + I] - ((1 << scale_factor) * y_est[i + I] +
532 				    G_fxp) / G_fxp;
533 
534 		x_tilde[i] =
535 		    (x_est_fxp1_nonlin * (1 << M) + y_est[i + I]) / y_est[i +
536 									  I];
537 		x_tilde[i] =
538 		    (x_tilde[i] * (1 << M) + y_est[i + I]) / y_est[i + I];
539 		x_tilde[i] =
540 		    (x_tilde[i] * (1 << M) + y_est[i + I]) / y_est[i + I];
541 		y_sqr =
542 		    (y_est[i + I] * y_est[i + I] +
543 		     (scale_factor * scale_factor)) / (scale_factor *
544 						       scale_factor);
545 		tmp_abs = abs(x_tilde[i]);
546 		if (tmp_abs > x_tilde_abs)
547 			x_tilde_abs = tmp_abs;
548 
549 		y_quad = y_sqr * y_sqr;
550 		sum_y_sqr = sum_y_sqr + y_sqr;
551 		sum_y_quad = sum_y_quad + y_quad;
552 		B1_tmp[i] = y_sqr * (L + 1);
553 		B2_tmp[i] = y_sqr;
554 	}
555 
556 	B1_abs_max = 0;
557 	B2_abs_max = 0;
558 	for (i = 0; i <= L; i++) {
559 		int abs_val;
560 
561 		B1_tmp[i] -= sum_y_sqr;
562 		B2_tmp[i] = sum_y_quad - sum_y_sqr * B2_tmp[i];
563 
564 		abs_val = abs(B1_tmp[i]);
565 		if (abs_val > B1_abs_max)
566 			B1_abs_max = abs_val;
567 
568 		abs_val = abs(B2_tmp[i]);
569 		if (abs_val > B2_abs_max)
570 			B2_abs_max = abs_val;
571 	}
572 
573 	Q_x = find_proper_scale(find_expn(x_tilde_abs), 10);
574 	Q_B1 = find_proper_scale(find_expn(B1_abs_max), 10);
575 	Q_B2 = find_proper_scale(find_expn(B2_abs_max), 10);
576 
577 	beta_raw = 0;
578 	alpha_raw = 0;
579 	for (i = 0; i <= L; i++) {
580 		x_tilde[i] = x_tilde[i] / (1 << Q_x);
581 		B1_tmp[i] = B1_tmp[i] / (1 << Q_B1);
582 		B2_tmp[i] = B2_tmp[i] / (1 << Q_B2);
583 		beta_raw = beta_raw + B1_tmp[i] * x_tilde[i];
584 		alpha_raw = alpha_raw + B2_tmp[i] * x_tilde[i];
585 	}
586 
587 	scale_B =
588 	    ((sum_y_quad / scale_factor) * (L + 1) -
589 	     (sum_y_sqr / scale_factor) * sum_y_sqr) * scale_factor;
590 
591 	Q_scale_B = find_proper_scale(find_expn(abs(scale_B)), 10);
592 	scale_B = scale_B / (1 << Q_scale_B);
593 	if (scale_B == 0)
594 		return false;
595 	Q_beta = find_proper_scale(find_expn(abs(beta_raw)), 10);
596 	Q_alpha = find_proper_scale(find_expn(abs(alpha_raw)), 10);
597 	beta_raw = beta_raw / (1 << Q_beta);
598 	alpha_raw = alpha_raw / (1 << Q_alpha);
599 	alpha = (alpha_raw << 10) / scale_B;
600 	beta = (beta_raw << 10) / scale_B;
601 	order_1 = 3 * M - Q_x - Q_B1 - Q_beta + 10 + Q_scale_B;
602 	order_2 = 3 * M - Q_x - Q_B2 - Q_alpha + 10 + Q_scale_B;
603 	order1_5x = order_1 / 5;
604 	order2_3x = order_2 / 3;
605 	order1_5x_rem = order_1 - 5 * order1_5x;
606 	order2_3x_rem = order_2 - 3 * order2_3x;
607 
608 	for (i = 0; i < PAPRD_TABLE_SZ; i++) {
609 		tmp = i * 32;
610 		y5 = ((beta * tmp) >> 6) >> order1_5x;
611 		y5 = (y5 * tmp) >> order1_5x;
612 		y5 = (y5 * tmp) >> order1_5x;
613 		y5 = (y5 * tmp) >> order1_5x;
614 		y5 = (y5 * tmp) >> order1_5x;
615 		y5 = y5 >> order1_5x_rem;
616 		y3 = (alpha * tmp) >> order2_3x;
617 		y3 = (y3 * tmp) >> order2_3x;
618 		y3 = (y3 * tmp) >> order2_3x;
619 		y3 = y3 >> order2_3x_rem;
620 		PA_in[i] = y5 + y3 + (256 * tmp) / G_fxp;
621 
622 		if (i >= 2) {
623 			tmp = PA_in[i] - PA_in[i - 1];
624 			if (tmp < 0)
625 				PA_in[i] =
626 				    PA_in[i - 1] + (PA_in[i - 1] -
627 						    PA_in[i - 2]);
628 		}
629 
630 		PA_in[i] = (PA_in[i] < 1400) ? PA_in[i] : 1400;
631 	}
632 
633 	beta_raw = 0;
634 	alpha_raw = 0;
635 
636 	for (i = 0; i <= L; i++) {
637 		int theta_tilde =
638 		    ((theta[i + I] << M) + y_est[i + I]) / y_est[i + I];
639 		theta_tilde =
640 		    ((theta_tilde << M) + y_est[i + I]) / y_est[i + I];
641 		theta_tilde =
642 		    ((theta_tilde << M) + y_est[i + I]) / y_est[i + I];
643 		beta_raw = beta_raw + B1_tmp[i] * theta_tilde;
644 		alpha_raw = alpha_raw + B2_tmp[i] * theta_tilde;
645 	}
646 
647 	Q_beta = find_proper_scale(find_expn(abs(beta_raw)), 10);
648 	Q_alpha = find_proper_scale(find_expn(abs(alpha_raw)), 10);
649 	beta_raw = beta_raw / (1 << Q_beta);
650 	alpha_raw = alpha_raw / (1 << Q_alpha);
651 
652 	alpha = (alpha_raw << 10) / scale_B;
653 	beta = (beta_raw << 10) / scale_B;
654 	order_1 = 3 * M - Q_x - Q_B1 - Q_beta + 10 + Q_scale_B + 5;
655 	order_2 = 3 * M - Q_x - Q_B2 - Q_alpha + 10 + Q_scale_B + 5;
656 	order1_5x = order_1 / 5;
657 	order2_3x = order_2 / 3;
658 	order1_5x_rem = order_1 - 5 * order1_5x;
659 	order2_3x_rem = order_2 - 3 * order2_3x;
660 
661 	for (i = 0; i < PAPRD_TABLE_SZ; i++) {
662 		int PA_angle;
663 
664 		/* pa_table[4] is calculated from PA_angle for i=5 */
665 		if (i == 4)
666 			continue;
667 
668 		tmp = i * 32;
669 		if (beta > 0)
670 			y5 = (((beta * tmp - 64) >> 6) -
671 			      (1 << order1_5x)) / (1 << order1_5x);
672 		else
673 			y5 = ((((beta * tmp - 64) >> 6) +
674 			       (1 << order1_5x)) / (1 << order1_5x));
675 
676 		y5 = (y5 * tmp) / (1 << order1_5x);
677 		y5 = (y5 * tmp) / (1 << order1_5x);
678 		y5 = (y5 * tmp) / (1 << order1_5x);
679 		y5 = (y5 * tmp) / (1 << order1_5x);
680 		y5 = y5 / (1 << order1_5x_rem);
681 
682 		if (beta > 0)
683 			y3 = (alpha * tmp -
684 			      (1 << order2_3x)) / (1 << order2_3x);
685 		else
686 			y3 = (alpha * tmp +
687 			      (1 << order2_3x)) / (1 << order2_3x);
688 		y3 = (y3 * tmp) / (1 << order2_3x);
689 		y3 = (y3 * tmp) / (1 << order2_3x);
690 		y3 = y3 / (1 << order2_3x_rem);
691 
692 		if (i < 4) {
693 			PA_angle = 0;
694 		} else {
695 			PA_angle = y5 + y3;
696 			if (PA_angle < -150)
697 				PA_angle = -150;
698 			else if (PA_angle > 150)
699 				PA_angle = 150;
700 		}
701 
702 		pa_table[i] = ((PA_in[i] & 0x7ff) << 11) + (PA_angle & 0x7ff);
703 		if (i == 5) {
704 			PA_angle = (PA_angle + 2) >> 1;
705 			pa_table[i - 1] = ((PA_in[i - 1] & 0x7ff) << 11) +
706 			    (PA_angle & 0x7ff);
707 		}
708 	}
709 
710 	*gain = G_fxp;
711 	return true;
712 }
713 
714 void ar9003_paprd_populate_single_table(struct ath_hw *ah,
715 					struct ath9k_hw_cal_data *caldata,
716 					int chain)
717 {
718 	u32 *paprd_table_val = caldata->pa_table[chain];
719 	u32 small_signal_gain = caldata->small_signal_gain[chain];
720 	u32 training_power = ah->paprd_training_power;
721 	u32 reg = 0;
722 	int i;
723 
724 	if (chain == 0)
725 		reg = AR_PHY_PAPRD_MEM_TAB_B0;
726 	else if (chain == 1)
727 		reg = AR_PHY_PAPRD_MEM_TAB_B1;
728 	else if (chain == 2)
729 		reg = AR_PHY_PAPRD_MEM_TAB_B2;
730 
731 	for (i = 0; i < PAPRD_TABLE_SZ; i++) {
732 		REG_WRITE(ah, reg, paprd_table_val[i]);
733 		reg = reg + 4;
734 	}
735 
736 	if (chain == 0)
737 		reg = AR_PHY_PA_GAIN123_B0;
738 	else if (chain == 1)
739 		reg = AR_PHY_PA_GAIN123_B1;
740 	else
741 		reg = AR_PHY_PA_GAIN123_B2;
742 
743 	REG_RMW_FIELD(ah, reg, AR_PHY_PA_GAIN123_PA_GAIN1, small_signal_gain);
744 
745 	REG_RMW_FIELD(ah, AR_PHY_PAPRD_CTRL1_B0,
746 		      AR_PHY_PAPRD_CTRL1_PAPRD_POWER_AT_AM2AM_CAL,
747 		      training_power);
748 
749 	if (ah->caps.tx_chainmask & BIT(1))
750 		REG_RMW_FIELD(ah, AR_PHY_PAPRD_CTRL1_B1,
751 			      AR_PHY_PAPRD_CTRL1_PAPRD_POWER_AT_AM2AM_CAL,
752 			      training_power);
753 
754 	if (ah->caps.tx_chainmask & BIT(2))
755 		/* val AR_PHY_PAPRD_CTRL1_PAPRD_POWER_AT_AM2AM_CAL correct? */
756 		REG_RMW_FIELD(ah, AR_PHY_PAPRD_CTRL1_B2,
757 			      AR_PHY_PAPRD_CTRL1_PAPRD_POWER_AT_AM2AM_CAL,
758 			      training_power);
759 }
760 EXPORT_SYMBOL(ar9003_paprd_populate_single_table);
761 
762 int ar9003_paprd_setup_gain_table(struct ath_hw *ah, int chain)
763 {
764 	unsigned int i, desired_gain, gain_index;
765 	unsigned int train_power = ah->paprd_training_power;
766 
767 	desired_gain = ar9003_get_desired_gain(ah, chain, train_power);
768 
769 	gain_index = 0;
770 	for (i = 0; i < PAPRD_GAIN_TABLE_ENTRIES; i++) {
771 		if (ah->paprd_gain_table_index[i] >= desired_gain)
772 			break;
773 		gain_index++;
774 	}
775 
776 	ar9003_tx_force_gain(ah, gain_index);
777 
778 	REG_CLR_BIT(ah, AR_PHY_PAPRD_TRAINER_STAT1,
779 			AR_PHY_PAPRD_TRAINER_STAT1_PAPRD_TRAIN_DONE);
780 
781 	return 0;
782 }
783 EXPORT_SYMBOL(ar9003_paprd_setup_gain_table);
784 
785 int ar9003_paprd_create_curve(struct ath_hw *ah,
786 			      struct ath9k_hw_cal_data *caldata, int chain)
787 {
788 	u16 *small_signal_gain = &caldata->small_signal_gain[chain];
789 	u32 *pa_table = caldata->pa_table[chain];
790 	u32 *data_L, *data_U;
791 	int i, status = 0;
792 	u32 *buf;
793 	u32 reg;
794 
795 	memset(caldata->pa_table[chain], 0, sizeof(caldata->pa_table[chain]));
796 
797 	buf = kmalloc(2 * 48 * sizeof(u32), GFP_ATOMIC);
798 	if (!buf)
799 		return -ENOMEM;
800 
801 	data_L = &buf[0];
802 	data_U = &buf[48];
803 
804 	REG_CLR_BIT(ah, AR_PHY_CHAN_INFO_MEMORY,
805 		    AR_PHY_CHAN_INFO_MEMORY_CHANINFOMEM_S2_READ);
806 
807 	reg = AR_PHY_CHAN_INFO_TAB_0;
808 	for (i = 0; i < 48; i++)
809 		data_L[i] = REG_READ(ah, reg + (i << 2));
810 
811 	REG_SET_BIT(ah, AR_PHY_CHAN_INFO_MEMORY,
812 		    AR_PHY_CHAN_INFO_MEMORY_CHANINFOMEM_S2_READ);
813 
814 	for (i = 0; i < 48; i++)
815 		data_U[i] = REG_READ(ah, reg + (i << 2));
816 
817 	if (!create_pa_curve(data_L, data_U, pa_table, small_signal_gain))
818 		status = -2;
819 
820 	REG_CLR_BIT(ah, AR_PHY_PAPRD_TRAINER_STAT1,
821 		    AR_PHY_PAPRD_TRAINER_STAT1_PAPRD_TRAIN_DONE);
822 
823 	kfree(buf);
824 
825 	return status;
826 }
827 EXPORT_SYMBOL(ar9003_paprd_create_curve);
828 
829 int ar9003_paprd_init_table(struct ath_hw *ah)
830 {
831 	int ret;
832 
833 	ret = ar9003_paprd_setup_single_table(ah);
834 	if (ret < 0)
835 	    return ret;
836 
837 	ar9003_paprd_get_gain_table(ah);
838 	return 0;
839 }
840 EXPORT_SYMBOL(ar9003_paprd_init_table);
841 
842 bool ar9003_paprd_is_done(struct ath_hw *ah)
843 {
844 	int paprd_done, agc2_pwr;
845 	paprd_done = REG_READ_FIELD(ah, AR_PHY_PAPRD_TRAINER_STAT1,
846 				AR_PHY_PAPRD_TRAINER_STAT1_PAPRD_TRAIN_DONE);
847 
848 	if (paprd_done == 0x1) {
849 		agc2_pwr = REG_READ_FIELD(ah, AR_PHY_PAPRD_TRAINER_STAT1,
850 				AR_PHY_PAPRD_TRAINER_STAT1_PAPRD_AGC2_PWR);
851 
852 		ath_dbg(ath9k_hw_common(ah), CALIBRATE,
853 			"AGC2_PWR = 0x%x training done = 0x%x\n",
854 			agc2_pwr, paprd_done);
855 	/*
856 	 * agc2_pwr range should not be less than 'IDEAL_AGC2_PWR_CHANGE'
857 	 * when the training is completely done, otherwise retraining is
858 	 * done to make sure the value is in ideal range
859 	 */
860 		if (agc2_pwr <= PAPRD_IDEAL_AGC2_PWR_RANGE)
861 			paprd_done = 0;
862 	}
863 
864 	return !!paprd_done;
865 }
866 EXPORT_SYMBOL(ar9003_paprd_is_done);
867