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
21void 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}
69EXPORT_SYMBOL(ar9003_paprd_enable);
70
71static 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
78	if (AR_SREV_9330(ah) || AR_SREV_9340(ah) ||
79	    AR_SREV_9462(ah) || AR_SREV_9565(ah)) {
80		power = ah->paprd_target_power + 2;
81	} else if (AR_SREV_9485(ah)) {
82		power = 25;
83	} else {
84		power = REG_READ_FIELD(ah, AR_PHY_POWERTX_RATE5,
85				       AR_PHY_POWERTX_RATE5_POWERTXHT20_0);
86
87		delta = abs((int) ah->paprd_target_power - (int) power);
88		if (delta > scale)
89			return -1;
90
91		if (delta < 4)
92			power -= 4 - delta;
93	}
94
95	return power;
96}
97
98static int ar9003_get_training_power_5g(struct ath_hw *ah)
99{
100	struct ath_common *common = ath9k_hw_common(ah);
101	struct ath9k_channel *chan = ah->curchan;
102	unsigned int power, scale, delta;
103
104	scale = ar9003_get_paprd_scale_factor(ah, chan);
105
106	if (IS_CHAN_HT40(chan))
107		power = REG_READ_FIELD(ah, AR_PHY_POWERTX_RATE8,
108			AR_PHY_POWERTX_RATE8_POWERTXHT40_5);
109	else
110		power = REG_READ_FIELD(ah, AR_PHY_POWERTX_RATE6,
111			AR_PHY_POWERTX_RATE6_POWERTXHT20_5);
112
113	power += scale;
114	delta = abs((int) ah->paprd_target_power - (int) power);
115	if (delta > scale)
116		return -1;
117
118	switch (get_streams(ah->txchainmask)) {
119	case 1:
120		delta = 6;
121		break;
122	case 2:
123		delta = 4;
124		break;
125	case 3:
126		delta = 2;
127		break;
128	default:
129		delta = 0;
130		ath_dbg(common, CALIBRATE, "Invalid tx-chainmask: %u\n",
131			ah->txchainmask);
132	}
133
134	power += delta;
135	return power;
136}
137
138static int ar9003_paprd_setup_single_table(struct ath_hw *ah)
139{
140	struct ath_common *common = ath9k_hw_common(ah);
141	static const u32 ctrl0[3] = {
142		AR_PHY_PAPRD_CTRL0_B0,
143		AR_PHY_PAPRD_CTRL0_B1,
144		AR_PHY_PAPRD_CTRL0_B2
145	};
146	static const u32 ctrl1[3] = {
147		AR_PHY_PAPRD_CTRL1_B0,
148		AR_PHY_PAPRD_CTRL1_B1,
149		AR_PHY_PAPRD_CTRL1_B2
150	};
151	int training_power;
152	int i, val;
153	u32 am2pm_mask = ah->paprd_ratemask;
154
155	if (IS_CHAN_2GHZ(ah->curchan))
156		training_power = ar9003_get_training_power_2g(ah);
157	else
158		training_power = ar9003_get_training_power_5g(ah);
159
160	ath_dbg(common, CALIBRATE, "Training power: %d, Target power: %d\n",
161		training_power, ah->paprd_target_power);
162
163	if (training_power < 0) {
164		ath_dbg(common, CALIBRATE,
165			"PAPRD target power delta out of range\n");
166		return -ERANGE;
167	}
168	ah->paprd_training_power = training_power;
169
170	if (AR_SREV_9330(ah))
171		am2pm_mask = 0;
172
173	REG_RMW_FIELD(ah, AR_PHY_PAPRD_AM2AM, AR_PHY_PAPRD_AM2AM_MASK,
174		      ah->paprd_ratemask);
175	REG_RMW_FIELD(ah, AR_PHY_PAPRD_AM2PM, AR_PHY_PAPRD_AM2PM_MASK,
176		      am2pm_mask);
177	REG_RMW_FIELD(ah, AR_PHY_PAPRD_HT40, AR_PHY_PAPRD_HT40_MASK,
178		      ah->paprd_ratemask_ht40);
179
180	ath_dbg(common, CALIBRATE, "PAPRD HT20 mask: 0x%x, HT40 mask: 0x%x\n",
181		ah->paprd_ratemask, ah->paprd_ratemask_ht40);
182
183	for (i = 0; i < ah->caps.max_txchains; i++) {
184		REG_RMW_FIELD(ah, ctrl0[i],
185			      AR_PHY_PAPRD_CTRL0_USE_SINGLE_TABLE_MASK, 1);
186		REG_RMW_FIELD(ah, ctrl1[i],
187			      AR_PHY_PAPRD_CTRL1_ADAPTIVE_AM2PM_ENABLE, 1);
188		REG_RMW_FIELD(ah, ctrl1[i],
189			      AR_PHY_PAPRD_CTRL1_ADAPTIVE_AM2AM_ENABLE, 1);
190		REG_RMW_FIELD(ah, ctrl1[i],
191			      AR_PHY_PAPRD_CTRL1_ADAPTIVE_SCALING_ENA, 0);
192		REG_RMW_FIELD(ah, ctrl1[i],
193			      AR_PHY_PAPRD_CTRL1_PA_GAIN_SCALE_FACT_MASK, 181);
194		REG_RMW_FIELD(ah, ctrl1[i],
195			      AR_PHY_PAPRD_CTRL1_PAPRD_MAG_SCALE_FACT, 361);
196		REG_RMW_FIELD(ah, ctrl1[i],
197			      AR_PHY_PAPRD_CTRL1_ADAPTIVE_SCALING_ENA, 0);
198		REG_RMW_FIELD(ah, ctrl0[i],
199			      AR_PHY_PAPRD_CTRL0_PAPRD_MAG_THRSH, 3);
200	}
201
202	ar9003_paprd_enable(ah, false);
203
204	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
205		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_PAPRD_LB_SKIP, 0x30);
206	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
207		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_PAPRD_LB_ENABLE, 1);
208	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
209		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_PAPRD_TX_GAIN_FORCE, 1);
210	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
211		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_PAPRD_RX_BB_GAIN_FORCE, 0);
212	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
213		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_PAPRD_IQCORR_ENABLE, 0);
214	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
215		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_PAPRD_AGC2_SETTLING, 28);
216	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL1,
217		      AR_PHY_PAPRD_TRAINER_CNTL1_CF_CF_PAPRD_TRAIN_ENABLE, 1);
218
219	if (AR_SREV_9485(ah)) {
220		val = 148;
221	} else {
222		if (IS_CHAN_2GHZ(ah->curchan)) {
223			if (AR_SREV_9462(ah) || AR_SREV_9565(ah))
224				val = 145;
225			else
226				val = 147;
227		} else {
228			val = 137;
229		}
230	}
231
232	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL2,
233		      AR_PHY_PAPRD_TRAINER_CNTL2_CF_PAPRD_INIT_RX_BB_GAIN, val);
234	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
235		      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_FINE_CORR_LEN, 4);
236	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
237		      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_COARSE_CORR_LEN, 4);
238	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
239		      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_NUM_CORR_STAGES, 7);
240	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
241		      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_MIN_LOOPBACK_DEL, 1);
242
243	if (AR_SREV_9485(ah) ||
244	    AR_SREV_9462(ah) ||
245	    AR_SREV_9565(ah) ||
246	    AR_SREV_9550(ah) ||
247	    AR_SREV_9330(ah) ||
248	    AR_SREV_9340(ah))
249		REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
250			      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_QUICK_DROP, -3);
251	else
252		REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
253			      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_QUICK_DROP, -6);
254
255	val = -10;
256
257	if (IS_CHAN_2GHZ(ah->curchan) && !AR_SREV_9462(ah) && !AR_SREV_9565(ah))
258		val = -15;
259
260	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
261		      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_ADC_DESIRED_SIZE,
262		      val);
263	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
264		      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_BBTXMIX_DISABLE, 1);
265	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL4,
266		      AR_PHY_PAPRD_TRAINER_CNTL4_CF_PAPRD_SAFETY_DELTA, 0);
267	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL4,
268		      AR_PHY_PAPRD_TRAINER_CNTL4_CF_PAPRD_MIN_CORR, 400);
269	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL4,
270		      AR_PHY_PAPRD_TRAINER_CNTL4_CF_PAPRD_NUM_TRAIN_SAMPLES,
271		      100);
272	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_0_B0,
273		      AR_PHY_PAPRD_PRE_POST_SCALING, 261376);
274	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_1_B0,
275		      AR_PHY_PAPRD_PRE_POST_SCALING, 248079);
276	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_2_B0,
277		      AR_PHY_PAPRD_PRE_POST_SCALING, 233759);
278	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_3_B0,
279		      AR_PHY_PAPRD_PRE_POST_SCALING, 220464);
280	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_4_B0,
281		      AR_PHY_PAPRD_PRE_POST_SCALING, 208194);
282	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_5_B0,
283		      AR_PHY_PAPRD_PRE_POST_SCALING, 196949);
284	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_6_B0,
285		      AR_PHY_PAPRD_PRE_POST_SCALING, 185706);
286	REG_RMW_FIELD(ah, AR_PHY_PAPRD_PRE_POST_SCALE_7_B0,
287		      AR_PHY_PAPRD_PRE_POST_SCALING, 175487);
288	return 0;
289}
290
291static void ar9003_paprd_get_gain_table(struct ath_hw *ah)
292{
293	u32 *entry = ah->paprd_gain_table_entries;
294	u8 *index = ah->paprd_gain_table_index;
295	u32 reg = AR_PHY_TXGAIN_TABLE;
296	int i;
297
298	for (i = 0; i < PAPRD_GAIN_TABLE_ENTRIES; i++) {
299		entry[i] = REG_READ(ah, reg);
300		index[i] = (entry[i] >> 24) & 0xff;
301		reg += 4;
302	}
303}
304
305static unsigned int ar9003_get_desired_gain(struct ath_hw *ah, int chain,
306					    int target_power)
307{
308	int olpc_gain_delta = 0, cl_gain_mod;
309	int alpha_therm, alpha_volt;
310	int therm_cal_value, volt_cal_value;
311	int therm_value, volt_value;
312	int thermal_gain_corr, voltage_gain_corr;
313	int desired_scale, desired_gain = 0;
314	u32 reg_olpc  = 0, reg_cl_gain  = 0;
315
316	REG_CLR_BIT(ah, AR_PHY_PAPRD_TRAINER_STAT1,
317		    AR_PHY_PAPRD_TRAINER_STAT1_PAPRD_TRAIN_DONE);
318	desired_scale = REG_READ_FIELD(ah, AR_PHY_TPC_12,
319				       AR_PHY_TPC_12_DESIRED_SCALE_HT40_5);
320	alpha_therm = REG_READ_FIELD(ah, AR_PHY_TPC_19,
321				     AR_PHY_TPC_19_ALPHA_THERM);
322	alpha_volt = REG_READ_FIELD(ah, AR_PHY_TPC_19,
323				    AR_PHY_TPC_19_ALPHA_VOLT);
324	therm_cal_value = REG_READ_FIELD(ah, AR_PHY_TPC_18,
325					 AR_PHY_TPC_18_THERM_CAL_VALUE);
326	volt_cal_value = REG_READ_FIELD(ah, AR_PHY_TPC_18,
327					AR_PHY_TPC_18_VOLT_CAL_VALUE);
328	therm_value = REG_READ_FIELD(ah, AR_PHY_BB_THERM_ADC_4,
329				     AR_PHY_BB_THERM_ADC_4_LATEST_THERM_VALUE);
330	volt_value = REG_READ_FIELD(ah, AR_PHY_BB_THERM_ADC_4,
331				    AR_PHY_BB_THERM_ADC_4_LATEST_VOLT_VALUE);
332
333	switch (chain) {
334	case 0:
335		reg_olpc = AR_PHY_TPC_11_B0;
336		reg_cl_gain = AR_PHY_CL_TAB_0;
337		break;
338	case 1:
339		reg_olpc = AR_PHY_TPC_11_B1;
340		reg_cl_gain = AR_PHY_CL_TAB_1;
341		break;
342	case 2:
343		reg_olpc = AR_PHY_TPC_11_B2;
344		reg_cl_gain = AR_PHY_CL_TAB_2;
345		break;
346	default:
347		ath_dbg(ath9k_hw_common(ah), CALIBRATE,
348			"Invalid chainmask: %d\n", chain);
349		break;
350	}
351
352	olpc_gain_delta = REG_READ_FIELD(ah, reg_olpc,
353					 AR_PHY_TPC_11_OLPC_GAIN_DELTA);
354	cl_gain_mod = REG_READ_FIELD(ah, reg_cl_gain,
355					 AR_PHY_CL_TAB_CL_GAIN_MOD);
356
357	if (olpc_gain_delta >= 128)
358		olpc_gain_delta = olpc_gain_delta - 256;
359
360	thermal_gain_corr = (alpha_therm * (therm_value - therm_cal_value) +
361			     (256 / 2)) / 256;
362	voltage_gain_corr = (alpha_volt * (volt_value - volt_cal_value) +
363			     (128 / 2)) / 128;
364	desired_gain = target_power - olpc_gain_delta - thermal_gain_corr -
365	    voltage_gain_corr + desired_scale + cl_gain_mod;
366
367	return desired_gain;
368}
369
370static void ar9003_tx_force_gain(struct ath_hw *ah, unsigned int gain_index)
371{
372	int selected_gain_entry, txbb1dbgain, txbb6dbgain, txmxrgain;
373	int padrvgnA, padrvgnB, padrvgnC, padrvgnD;
374	u32 *gain_table_entries = ah->paprd_gain_table_entries;
375
376	selected_gain_entry = gain_table_entries[gain_index];
377	txbb1dbgain = selected_gain_entry & 0x7;
378	txbb6dbgain = (selected_gain_entry >> 3) & 0x3;
379	txmxrgain = (selected_gain_entry >> 5) & 0xf;
380	padrvgnA = (selected_gain_entry >> 9) & 0xf;
381	padrvgnB = (selected_gain_entry >> 13) & 0xf;
382	padrvgnC = (selected_gain_entry >> 17) & 0xf;
383	padrvgnD = (selected_gain_entry >> 21) & 0x3;
384
385	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
386		      AR_PHY_TX_FORCED_GAIN_FORCED_TXBB1DBGAIN, txbb1dbgain);
387	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
388		      AR_PHY_TX_FORCED_GAIN_FORCED_TXBB6DBGAIN, txbb6dbgain);
389	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
390		      AR_PHY_TX_FORCED_GAIN_FORCED_TXMXRGAIN, txmxrgain);
391	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
392		      AR_PHY_TX_FORCED_GAIN_FORCED_PADRVGNA, padrvgnA);
393	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
394		      AR_PHY_TX_FORCED_GAIN_FORCED_PADRVGNB, padrvgnB);
395	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
396		      AR_PHY_TX_FORCED_GAIN_FORCED_PADRVGNC, padrvgnC);
397	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
398		      AR_PHY_TX_FORCED_GAIN_FORCED_PADRVGND, padrvgnD);
399	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
400		      AR_PHY_TX_FORCED_GAIN_FORCED_ENABLE_PAL, 0);
401	REG_RMW_FIELD(ah, AR_PHY_TX_FORCED_GAIN,
402		      AR_PHY_TX_FORCED_GAIN_FORCE_TX_GAIN, 0);
403	REG_RMW_FIELD(ah, AR_PHY_TPC_1, AR_PHY_TPC_1_FORCED_DAC_GAIN, 0);
404	REG_RMW_FIELD(ah, AR_PHY_TPC_1, AR_PHY_TPC_1_FORCE_DAC_GAIN, 0);
405}
406
407static inline int find_expn(int num)
408{
409	return fls(num) - 1;
410}
411
412static inline int find_proper_scale(int expn, int N)
413{
414	return (expn > N) ? expn - 10 : 0;
415}
416
417#define NUM_BIN 23
418
419static bool create_pa_curve(u32 *data_L, u32 *data_U, u32 *pa_table, u16 *gain)
420{
421	unsigned int thresh_accum_cnt;
422	int x_est[NUM_BIN + 1], Y[NUM_BIN + 1], theta[NUM_BIN + 1];
423	int PA_in[NUM_BIN + 1];
424	int B1_tmp[NUM_BIN + 1], B2_tmp[NUM_BIN + 1];
425	unsigned int B1_abs_max, B2_abs_max;
426	int max_index, scale_factor;
427	int y_est[NUM_BIN + 1];
428	int x_est_fxp1_nonlin, x_tilde[NUM_BIN + 1];
429	unsigned int x_tilde_abs;
430	int G_fxp, Y_intercept, order_x_by_y, M, I, L, sum_y_sqr, sum_y_quad;
431	int Q_x, Q_B1, Q_B2, beta_raw, alpha_raw, scale_B;
432	int Q_scale_B, Q_beta, Q_alpha, alpha, beta, order_1, order_2;
433	int order1_5x, order2_3x, order1_5x_rem, order2_3x_rem;
434	int y5, y3, tmp;
435	int theta_low_bin = 0;
436	int i;
437
438	/* disregard any bin that contains <= 16 samples */
439	thresh_accum_cnt = 16;
440	scale_factor = 5;
441	max_index = 0;
442	memset(theta, 0, sizeof(theta));
443	memset(x_est, 0, sizeof(x_est));
444	memset(Y, 0, sizeof(Y));
445	memset(y_est, 0, sizeof(y_est));
446	memset(x_tilde, 0, sizeof(x_tilde));
447
448	for (i = 0; i < NUM_BIN; i++) {
449		s32 accum_cnt, accum_tx, accum_rx, accum_ang;
450
451		/* number of samples */
452		accum_cnt = data_L[i] & 0xffff;
453
454		if (accum_cnt <= thresh_accum_cnt)
455			continue;
456
457		max_index++;
458
459		/* sum(tx amplitude) */
460		accum_tx = ((data_L[i] >> 16) & 0xffff) |
461		    ((data_U[i] & 0x7ff) << 16);
462
463		/* sum(rx amplitude distance to lower bin edge) */
464		accum_rx = ((data_U[i] >> 11) & 0x1f) |
465		    ((data_L[i + 23] & 0xffff) << 5);
466
467		/* sum(angles) */
468		accum_ang = ((data_L[i + 23] >> 16) & 0xffff) |
469		    ((data_U[i + 23] & 0x7ff) << 16);
470
471		accum_tx <<= scale_factor;
472		accum_rx <<= scale_factor;
473		x_est[max_index] =
474			(((accum_tx + accum_cnt) / accum_cnt) + 32) >>
475			scale_factor;
476
477		Y[max_index] =
478			((((accum_rx + accum_cnt) / accum_cnt) + 32) >>
479			    scale_factor) +
480			(1 << scale_factor) * i + 16;
481
482		if (accum_ang >= (1 << 26))
483			accum_ang -= 1 << 27;
484
485		theta[max_index] =
486			((accum_ang * (1 << scale_factor)) + accum_cnt) /
487			accum_cnt;
488	}
489
490	/*
491	 * Find average theta of first 5 bin and all of those to same value.
492	 * Curve is linear at that range.
493	 */
494	for (i = 1; i < 6; i++)
495		theta_low_bin += theta[i];
496
497	theta_low_bin = theta_low_bin / 5;
498	for (i = 1; i < 6; i++)
499		theta[i] = theta_low_bin;
500
501	/* Set values at origin */
502	theta[0] = theta_low_bin;
503	for (i = 0; i <= max_index; i++)
504		theta[i] -= theta_low_bin;
505
506	x_est[0] = 0;
507	Y[0] = 0;
508	scale_factor = 8;
509
510	/* low signal gain */
511	if (x_est[6] == x_est[3])
512		return false;
513
514	G_fxp =
515	    (((Y[6] - Y[3]) * 1 << scale_factor) +
516	     (x_est[6] - x_est[3])) / (x_est[6] - x_est[3]);
517
518	/* prevent division by zero */
519	if (G_fxp == 0)
520		return false;
521
522	Y_intercept =
523	    (G_fxp * (x_est[0] - x_est[3]) +
524	     (1 << scale_factor)) / (1 << scale_factor) + Y[3];
525
526	for (i = 0; i <= max_index; i++)
527		y_est[i] = Y[i] - Y_intercept;
528
529	for (i = 0; i <= 3; i++) {
530		y_est[i] = i * 32;
531		x_est[i] = ((y_est[i] * 1 << scale_factor) + G_fxp) / G_fxp;
532	}
533
534	if (y_est[max_index] == 0)
535		return false;
536
537	x_est_fxp1_nonlin =
538	    x_est[max_index] - ((1 << scale_factor) * y_est[max_index] +
539				G_fxp) / G_fxp;
540
541	order_x_by_y =
542	    (x_est_fxp1_nonlin + y_est[max_index]) / y_est[max_index];
543
544	if (order_x_by_y == 0)
545		M = 10;
546	else if (order_x_by_y == 1)
547		M = 9;
548	else
549		M = 8;
550
551	I = (max_index > 15) ? 7 : max_index >> 1;
552	L = max_index - I;
553	scale_factor = 8;
554	sum_y_sqr = 0;
555	sum_y_quad = 0;
556	x_tilde_abs = 0;
557
558	for (i = 0; i <= L; i++) {
559		unsigned int y_sqr;
560		unsigned int y_quad;
561		unsigned int tmp_abs;
562
563		/* prevent division by zero */
564		if (y_est[i + I] == 0)
565			return false;
566
567		x_est_fxp1_nonlin =
568		    x_est[i + I] - ((1 << scale_factor) * y_est[i + I] +
569				    G_fxp) / G_fxp;
570
571		x_tilde[i] =
572		    (x_est_fxp1_nonlin * (1 << M) + y_est[i + I]) / y_est[i +
573									  I];
574		x_tilde[i] =
575		    (x_tilde[i] * (1 << M) + y_est[i + I]) / y_est[i + I];
576		x_tilde[i] =
577		    (x_tilde[i] * (1 << M) + y_est[i + I]) / y_est[i + I];
578		y_sqr =
579		    (y_est[i + I] * y_est[i + I] +
580		     (scale_factor * scale_factor)) / (scale_factor *
581						       scale_factor);
582		tmp_abs = abs(x_tilde[i]);
583		if (tmp_abs > x_tilde_abs)
584			x_tilde_abs = tmp_abs;
585
586		y_quad = y_sqr * y_sqr;
587		sum_y_sqr = sum_y_sqr + y_sqr;
588		sum_y_quad = sum_y_quad + y_quad;
589		B1_tmp[i] = y_sqr * (L + 1);
590		B2_tmp[i] = y_sqr;
591	}
592
593	B1_abs_max = 0;
594	B2_abs_max = 0;
595	for (i = 0; i <= L; i++) {
596		int abs_val;
597
598		B1_tmp[i] -= sum_y_sqr;
599		B2_tmp[i] = sum_y_quad - sum_y_sqr * B2_tmp[i];
600
601		abs_val = abs(B1_tmp[i]);
602		if (abs_val > B1_abs_max)
603			B1_abs_max = abs_val;
604
605		abs_val = abs(B2_tmp[i]);
606		if (abs_val > B2_abs_max)
607			B2_abs_max = abs_val;
608	}
609
610	Q_x = find_proper_scale(find_expn(x_tilde_abs), 10);
611	Q_B1 = find_proper_scale(find_expn(B1_abs_max), 10);
612	Q_B2 = find_proper_scale(find_expn(B2_abs_max), 10);
613
614	beta_raw = 0;
615	alpha_raw = 0;
616	for (i = 0; i <= L; i++) {
617		x_tilde[i] = x_tilde[i] / (1 << Q_x);
618		B1_tmp[i] = B1_tmp[i] / (1 << Q_B1);
619		B2_tmp[i] = B2_tmp[i] / (1 << Q_B2);
620		beta_raw = beta_raw + B1_tmp[i] * x_tilde[i];
621		alpha_raw = alpha_raw + B2_tmp[i] * x_tilde[i];
622	}
623
624	scale_B =
625	    ((sum_y_quad / scale_factor) * (L + 1) -
626	     (sum_y_sqr / scale_factor) * sum_y_sqr) * scale_factor;
627
628	Q_scale_B = find_proper_scale(find_expn(abs(scale_B)), 10);
629	scale_B = scale_B / (1 << Q_scale_B);
630	if (scale_B == 0)
631		return false;
632	Q_beta = find_proper_scale(find_expn(abs(beta_raw)), 10);
633	Q_alpha = find_proper_scale(find_expn(abs(alpha_raw)), 10);
634	beta_raw = beta_raw / (1 << Q_beta);
635	alpha_raw = alpha_raw / (1 << Q_alpha);
636	alpha = (alpha_raw << 10) / scale_B;
637	beta = (beta_raw << 10) / scale_B;
638	order_1 = 3 * M - Q_x - Q_B1 - Q_beta + 10 + Q_scale_B;
639	order_2 = 3 * M - Q_x - Q_B2 - Q_alpha + 10 + Q_scale_B;
640	order1_5x = order_1 / 5;
641	order2_3x = order_2 / 3;
642	order1_5x_rem = order_1 - 5 * order1_5x;
643	order2_3x_rem = order_2 - 3 * order2_3x;
644
645	for (i = 0; i < PAPRD_TABLE_SZ; i++) {
646		tmp = i * 32;
647		y5 = ((beta * tmp) >> 6) >> order1_5x;
648		y5 = (y5 * tmp) >> order1_5x;
649		y5 = (y5 * tmp) >> order1_5x;
650		y5 = (y5 * tmp) >> order1_5x;
651		y5 = (y5 * tmp) >> order1_5x;
652		y5 = y5 >> order1_5x_rem;
653		y3 = (alpha * tmp) >> order2_3x;
654		y3 = (y3 * tmp) >> order2_3x;
655		y3 = (y3 * tmp) >> order2_3x;
656		y3 = y3 >> order2_3x_rem;
657		PA_in[i] = y5 + y3 + (256 * tmp) / G_fxp;
658
659		if (i >= 2) {
660			tmp = PA_in[i] - PA_in[i - 1];
661			if (tmp < 0)
662				PA_in[i] =
663				    PA_in[i - 1] + (PA_in[i - 1] -
664						    PA_in[i - 2]);
665		}
666
667		PA_in[i] = (PA_in[i] < 1400) ? PA_in[i] : 1400;
668	}
669
670	beta_raw = 0;
671	alpha_raw = 0;
672
673	for (i = 0; i <= L; i++) {
674		int theta_tilde =
675		    ((theta[i + I] << M) + y_est[i + I]) / y_est[i + I];
676		theta_tilde =
677		    ((theta_tilde << M) + y_est[i + I]) / y_est[i + I];
678		theta_tilde =
679		    ((theta_tilde << M) + y_est[i + I]) / y_est[i + I];
680		beta_raw = beta_raw + B1_tmp[i] * theta_tilde;
681		alpha_raw = alpha_raw + B2_tmp[i] * theta_tilde;
682	}
683
684	Q_beta = find_proper_scale(find_expn(abs(beta_raw)), 10);
685	Q_alpha = find_proper_scale(find_expn(abs(alpha_raw)), 10);
686	beta_raw = beta_raw / (1 << Q_beta);
687	alpha_raw = alpha_raw / (1 << Q_alpha);
688
689	alpha = (alpha_raw << 10) / scale_B;
690	beta = (beta_raw << 10) / scale_B;
691	order_1 = 3 * M - Q_x - Q_B1 - Q_beta + 10 + Q_scale_B + 5;
692	order_2 = 3 * M - Q_x - Q_B2 - Q_alpha + 10 + Q_scale_B + 5;
693	order1_5x = order_1 / 5;
694	order2_3x = order_2 / 3;
695	order1_5x_rem = order_1 - 5 * order1_5x;
696	order2_3x_rem = order_2 - 3 * order2_3x;
697
698	for (i = 0; i < PAPRD_TABLE_SZ; i++) {
699		int PA_angle;
700
701		/* pa_table[4] is calculated from PA_angle for i=5 */
702		if (i == 4)
703			continue;
704
705		tmp = i * 32;
706		if (beta > 0)
707			y5 = (((beta * tmp - 64) >> 6) -
708			      (1 << order1_5x)) / (1 << order1_5x);
709		else
710			y5 = ((((beta * tmp - 64) >> 6) +
711			       (1 << order1_5x)) / (1 << order1_5x));
712
713		y5 = (y5 * tmp) / (1 << order1_5x);
714		y5 = (y5 * tmp) / (1 << order1_5x);
715		y5 = (y5 * tmp) / (1 << order1_5x);
716		y5 = (y5 * tmp) / (1 << order1_5x);
717		y5 = y5 / (1 << order1_5x_rem);
718
719		if (beta > 0)
720			y3 = (alpha * tmp -
721			      (1 << order2_3x)) / (1 << order2_3x);
722		else
723			y3 = (alpha * tmp +
724			      (1 << order2_3x)) / (1 << order2_3x);
725		y3 = (y3 * tmp) / (1 << order2_3x);
726		y3 = (y3 * tmp) / (1 << order2_3x);
727		y3 = y3 / (1 << order2_3x_rem);
728
729		if (i < 4) {
730			PA_angle = 0;
731		} else {
732			PA_angle = y5 + y3;
733			if (PA_angle < -150)
734				PA_angle = -150;
735			else if (PA_angle > 150)
736				PA_angle = 150;
737		}
738
739		pa_table[i] = ((PA_in[i] & 0x7ff) << 11) + (PA_angle & 0x7ff);
740		if (i == 5) {
741			PA_angle = (PA_angle + 2) >> 1;
742			pa_table[i - 1] = ((PA_in[i - 1] & 0x7ff) << 11) +
743			    (PA_angle & 0x7ff);
744		}
745	}
746
747	*gain = G_fxp;
748	return true;
749}
750
751void ar9003_paprd_populate_single_table(struct ath_hw *ah,
752					struct ath9k_hw_cal_data *caldata,
753					int chain)
754{
755	u32 *paprd_table_val = caldata->pa_table[chain];
756	u32 small_signal_gain = caldata->small_signal_gain[chain];
757	u32 training_power = ah->paprd_training_power;
758	u32 reg = 0;
759	int i;
760
761	if (chain == 0)
762		reg = AR_PHY_PAPRD_MEM_TAB_B0;
763	else if (chain == 1)
764		reg = AR_PHY_PAPRD_MEM_TAB_B1;
765	else if (chain == 2)
766		reg = AR_PHY_PAPRD_MEM_TAB_B2;
767
768	for (i = 0; i < PAPRD_TABLE_SZ; i++) {
769		REG_WRITE(ah, reg, paprd_table_val[i]);
770		reg = reg + 4;
771	}
772
773	if (chain == 0)
774		reg = AR_PHY_PA_GAIN123_B0;
775	else if (chain == 1)
776		reg = AR_PHY_PA_GAIN123_B1;
777	else
778		reg = AR_PHY_PA_GAIN123_B2;
779
780	REG_RMW_FIELD(ah, reg, AR_PHY_PA_GAIN123_PA_GAIN1, small_signal_gain);
781
782	REG_RMW_FIELD(ah, AR_PHY_PAPRD_CTRL1_B0,
783		      AR_PHY_PAPRD_CTRL1_PAPRD_POWER_AT_AM2AM_CAL,
784		      training_power);
785
786	if (ah->caps.tx_chainmask & BIT(1))
787		REG_RMW_FIELD(ah, AR_PHY_PAPRD_CTRL1_B1,
788			      AR_PHY_PAPRD_CTRL1_PAPRD_POWER_AT_AM2AM_CAL,
789			      training_power);
790
791	if (ah->caps.tx_chainmask & BIT(2))
792		/* val AR_PHY_PAPRD_CTRL1_PAPRD_POWER_AT_AM2AM_CAL correct? */
793		REG_RMW_FIELD(ah, AR_PHY_PAPRD_CTRL1_B2,
794			      AR_PHY_PAPRD_CTRL1_PAPRD_POWER_AT_AM2AM_CAL,
795			      training_power);
796}
797EXPORT_SYMBOL(ar9003_paprd_populate_single_table);
798
799void ar9003_paprd_setup_gain_table(struct ath_hw *ah, int chain)
800{
801	unsigned int i, desired_gain, gain_index;
802	unsigned int train_power = ah->paprd_training_power;
803
804	desired_gain = ar9003_get_desired_gain(ah, chain, train_power);
805
806	gain_index = 0;
807	for (i = 0; i < PAPRD_GAIN_TABLE_ENTRIES; i++) {
808		if (ah->paprd_gain_table_index[i] >= desired_gain)
809			break;
810		gain_index++;
811	}
812
813	ar9003_tx_force_gain(ah, gain_index);
814
815	REG_CLR_BIT(ah, AR_PHY_PAPRD_TRAINER_STAT1,
816			AR_PHY_PAPRD_TRAINER_STAT1_PAPRD_TRAIN_DONE);
817}
818EXPORT_SYMBOL(ar9003_paprd_setup_gain_table);
819
820static bool ar9003_paprd_retrain_pa_in(struct ath_hw *ah,
821				       struct ath9k_hw_cal_data *caldata,
822				       int chain)
823{
824	u32 *pa_in = caldata->pa_table[chain];
825	int capdiv_offset, quick_drop_offset;
826	int capdiv2g, quick_drop;
827	int count = 0;
828	int i;
829
830	if (!AR_SREV_9485(ah) && !AR_SREV_9330(ah))
831		return false;
832
833	capdiv2g = REG_READ_FIELD(ah, AR_PHY_65NM_CH0_TXRF3,
834				  AR_PHY_65NM_CH0_TXRF3_CAPDIV2G);
835
836	quick_drop = REG_READ_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
837				    AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_QUICK_DROP);
838
839	if (quick_drop)
840		quick_drop -= 0x40;
841
842	for (i = 0; i < NUM_BIN + 1; i++) {
843		if (pa_in[i] == 1400)
844			count++;
845	}
846
847	if (AR_SREV_9485(ah)) {
848		if (pa_in[23] < 800) {
849			capdiv_offset = (int)((1000 - pa_in[23] + 75) / 150);
850			capdiv2g += capdiv_offset;
851			if (capdiv2g > 7) {
852				capdiv2g = 7;
853				if (pa_in[23] < 600) {
854					quick_drop++;
855					if (quick_drop > 0)
856						quick_drop = 0;
857				}
858			}
859		} else if (pa_in[23] == 1400) {
860			quick_drop_offset = min_t(int, count / 3, 2);
861			quick_drop += quick_drop_offset;
862			capdiv2g += quick_drop_offset / 2;
863
864			if (capdiv2g > 7)
865				capdiv2g = 7;
866
867			if (quick_drop > 0) {
868				quick_drop = 0;
869				capdiv2g -= quick_drop_offset;
870				if (capdiv2g < 0)
871					capdiv2g = 0;
872			}
873		} else {
874			return false;
875		}
876	} else if (AR_SREV_9330(ah)) {
877		if (pa_in[23] < 1000) {
878			capdiv_offset = (1000 - pa_in[23]) / 100;
879			capdiv2g += capdiv_offset;
880			if (capdiv_offset > 3) {
881				capdiv_offset = 1;
882				quick_drop--;
883			}
884
885			capdiv2g += capdiv_offset;
886			if (capdiv2g > 6)
887				capdiv2g = 6;
888			if (quick_drop < -4)
889				quick_drop = -4;
890		} else if (pa_in[23] == 1400) {
891			if (count > 3) {
892				quick_drop++;
893				capdiv2g -= count / 4;
894				if (quick_drop > -2)
895					quick_drop = -2;
896			} else {
897				capdiv2g--;
898			}
899
900			if (capdiv2g < 0)
901				capdiv2g = 0;
902		} else {
903			return false;
904		}
905	}
906
907	REG_RMW_FIELD(ah, AR_PHY_65NM_CH0_TXRF3,
908		      AR_PHY_65NM_CH0_TXRF3_CAPDIV2G, capdiv2g);
909	REG_RMW_FIELD(ah, AR_PHY_PAPRD_TRAINER_CNTL3,
910		      AR_PHY_PAPRD_TRAINER_CNTL3_CF_PAPRD_QUICK_DROP,
911		      quick_drop);
912
913	return true;
914}
915
916int ar9003_paprd_create_curve(struct ath_hw *ah,
917			      struct ath9k_hw_cal_data *caldata, int chain)
918{
919	u16 *small_signal_gain = &caldata->small_signal_gain[chain];
920	u32 *pa_table = caldata->pa_table[chain];
921	u32 *data_L, *data_U;
922	int i, status = 0;
923	u32 *buf;
924	u32 reg;
925
926	memset(caldata->pa_table[chain], 0, sizeof(caldata->pa_table[chain]));
927
928	buf = kmalloc(2 * 48 * sizeof(u32), GFP_KERNEL);
929	if (!buf)
930		return -ENOMEM;
931
932	data_L = &buf[0];
933	data_U = &buf[48];
934
935	REG_CLR_BIT(ah, AR_PHY_CHAN_INFO_MEMORY,
936		    AR_PHY_CHAN_INFO_MEMORY_CHANINFOMEM_S2_READ);
937
938	reg = AR_PHY_CHAN_INFO_TAB_0;
939	for (i = 0; i < 48; i++)
940		data_L[i] = REG_READ(ah, reg + (i << 2));
941
942	REG_SET_BIT(ah, AR_PHY_CHAN_INFO_MEMORY,
943		    AR_PHY_CHAN_INFO_MEMORY_CHANINFOMEM_S2_READ);
944
945	for (i = 0; i < 48; i++)
946		data_U[i] = REG_READ(ah, reg + (i << 2));
947
948	if (!create_pa_curve(data_L, data_U, pa_table, small_signal_gain))
949		status = -2;
950
951	if (ar9003_paprd_retrain_pa_in(ah, caldata, chain))
952		status = -EINPROGRESS;
953
954	REG_CLR_BIT(ah, AR_PHY_PAPRD_TRAINER_STAT1,
955		    AR_PHY_PAPRD_TRAINER_STAT1_PAPRD_TRAIN_DONE);
956
957	kfree(buf);
958
959	return status;
960}
961EXPORT_SYMBOL(ar9003_paprd_create_curve);
962
963int ar9003_paprd_init_table(struct ath_hw *ah)
964{
965	int ret;
966
967	ret = ar9003_paprd_setup_single_table(ah);
968	if (ret < 0)
969	    return ret;
970
971	ar9003_paprd_get_gain_table(ah);
972	return 0;
973}
974EXPORT_SYMBOL(ar9003_paprd_init_table);
975
976bool ar9003_paprd_is_done(struct ath_hw *ah)
977{
978	int paprd_done, agc2_pwr;
979
980	paprd_done = REG_READ_FIELD(ah, AR_PHY_PAPRD_TRAINER_STAT1,
981				AR_PHY_PAPRD_TRAINER_STAT1_PAPRD_TRAIN_DONE);
982
983	if (AR_SREV_9485(ah))
984		goto exit;
985
986	if (paprd_done == 0x1) {
987		agc2_pwr = REG_READ_FIELD(ah, AR_PHY_PAPRD_TRAINER_STAT1,
988				AR_PHY_PAPRD_TRAINER_STAT1_PAPRD_AGC2_PWR);
989
990		ath_dbg(ath9k_hw_common(ah), CALIBRATE,
991			"AGC2_PWR = 0x%x training done = 0x%x\n",
992			agc2_pwr, paprd_done);
993	/*
994	 * agc2_pwr range should not be less than 'IDEAL_AGC2_PWR_CHANGE'
995	 * when the training is completely done, otherwise retraining is
996	 * done to make sure the value is in ideal range
997	 */
998		if (agc2_pwr <= PAPRD_IDEAL_AGC2_PWR_RANGE)
999			paprd_done = 0;
1000	}
1001exit:
1002	return !!paprd_done;
1003}
1004EXPORT_SYMBOL(ar9003_paprd_is_done);
1005
1006bool ar9003_is_paprd_enabled(struct ath_hw *ah)
1007{
1008	if ((ah->caps.hw_caps & ATH9K_HW_CAP_PAPRD) && ah->config.enable_paprd)
1009		return true;
1010
1011	return false;
1012}
1013EXPORT_SYMBOL(ar9003_is_paprd_enabled);
1014