xref: /openbsd-src/sys/dev/ic/ar9003.c (revision f2da64fbbbf1b03f09f390ab01267c93dfd77c4c)
1 /*	$OpenBSD: ar9003.c,v 1.39 2016/01/05 18:41:15 stsp Exp $	*/
2 
3 /*-
4  * Copyright (c) 2010 Damien Bergamini <damien.bergamini@free.fr>
5  * Copyright (c) 2010 Atheros Communications Inc.
6  *
7  * Permission to use, copy, modify, and/or distribute this software for any
8  * purpose with or without fee is hereby granted, provided that the above
9  * copyright notice and this permission notice appear in all copies.
10  *
11  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
12  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
13  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
14  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
15  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
16  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
17  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
18  */
19 
20 /*
21  * Driver for Atheros 802.11a/g/n chipsets.
22  * Routines for AR9003 family.
23  */
24 
25 #include "bpfilter.h"
26 
27 #include <sys/param.h>
28 #include <sys/sockio.h>
29 #include <sys/mbuf.h>
30 #include <sys/kernel.h>
31 #include <sys/socket.h>
32 #include <sys/systm.h>
33 #include <sys/malloc.h>
34 #include <sys/queue.h>
35 #include <sys/timeout.h>
36 #include <sys/conf.h>
37 #include <sys/device.h>
38 #include <sys/stdint.h>	/* uintptr_t */
39 #include <sys/endian.h>
40 
41 #include <machine/bus.h>
42 
43 #if NBPFILTER > 0
44 #include <net/bpf.h>
45 #endif
46 #include <net/if.h>
47 #include <net/if_media.h>
48 
49 #include <netinet/in.h>
50 #include <netinet/if_ether.h>
51 
52 #include <net80211/ieee80211_var.h>
53 #include <net80211/ieee80211_amrr.h>
54 #include <net80211/ieee80211_radiotap.h>
55 
56 #include <dev/ic/athnreg.h>
57 #include <dev/ic/athnvar.h>
58 
59 #include <dev/ic/ar9003reg.h>
60 
61 int	ar9003_attach(struct athn_softc *);
62 int	ar9003_read_eep_word(struct athn_softc *, uint32_t, uint16_t *);
63 int	ar9003_read_eep_data(struct athn_softc *, uint32_t, void *, int);
64 int	ar9003_read_otp_word(struct athn_softc *, uint32_t, uint32_t *);
65 int	ar9003_read_otp_data(struct athn_softc *, uint32_t, void *, int);
66 int	ar9003_find_rom(struct athn_softc *);
67 int	ar9003_restore_rom_block(struct athn_softc *, uint8_t, uint8_t,
68 	    const uint8_t *, int);
69 int	ar9003_read_rom(struct athn_softc *);
70 int	ar9003_gpio_read(struct athn_softc *, int);
71 void	ar9003_gpio_write(struct athn_softc *, int, int);
72 void	ar9003_gpio_config_input(struct athn_softc *, int);
73 void	ar9003_gpio_config_output(struct athn_softc *, int, int);
74 void	ar9003_rfsilent_init(struct athn_softc *);
75 int	ar9003_dma_alloc(struct athn_softc *);
76 void	ar9003_dma_free(struct athn_softc *);
77 int	ar9003_tx_alloc(struct athn_softc *);
78 void	ar9003_tx_free(struct athn_softc *);
79 int	ar9003_rx_alloc(struct athn_softc *, int, int);
80 void	ar9003_rx_free(struct athn_softc *, int);
81 void	ar9003_reset_txsring(struct athn_softc *);
82 void	ar9003_rx_enable(struct athn_softc *);
83 void	ar9003_rx_radiotap(struct athn_softc *, struct mbuf *,
84 	    struct ar_rx_status *);
85 int	ar9003_rx_process(struct athn_softc *, int);
86 void	ar9003_rx_intr(struct athn_softc *, int);
87 int	ar9003_tx_process(struct athn_softc *);
88 void	ar9003_tx_intr(struct athn_softc *);
89 int	ar9003_swba_intr(struct athn_softc *);
90 int	ar9003_intr(struct athn_softc *);
91 int	ar9003_tx(struct athn_softc *, struct mbuf *, struct ieee80211_node *,
92 	    int);
93 void	ar9003_set_rf_mode(struct athn_softc *, struct ieee80211_channel *);
94 int	ar9003_rf_bus_request(struct athn_softc *);
95 void	ar9003_rf_bus_release(struct athn_softc *);
96 void	ar9003_set_phy(struct athn_softc *, struct ieee80211_channel *,
97 	    struct ieee80211_channel *);
98 void	ar9003_set_delta_slope(struct athn_softc *, struct ieee80211_channel *,
99 	    struct ieee80211_channel *);
100 void	ar9003_enable_antenna_diversity(struct athn_softc *);
101 void	ar9003_init_baseband(struct athn_softc *);
102 void	ar9003_disable_phy(struct athn_softc *);
103 void	ar9003_init_chains(struct athn_softc *);
104 void	ar9003_set_rxchains(struct athn_softc *);
105 void	ar9003_read_noisefloor(struct athn_softc *, int16_t *, int16_t *);
106 void	ar9003_write_noisefloor(struct athn_softc *, int16_t *, int16_t *);
107 void	ar9003_get_noisefloor(struct athn_softc *, struct ieee80211_channel *);
108 void	ar9003_bb_load_noisefloor(struct athn_softc *);
109 void	ar9300_noisefloor_calib(struct athn_softc *);
110 void	ar9003_do_noisefloor_calib(struct athn_softc *);
111 int	ar9003_init_calib(struct athn_softc *);
112 void	ar9003_do_calib(struct athn_softc *);
113 void	ar9003_next_calib(struct athn_softc *);
114 void	ar9003_calib_iq(struct athn_softc *);
115 int	ar9003_get_iq_corr(struct athn_softc *, int32_t[], int32_t[]);
116 int	ar9003_calib_tx_iq(struct athn_softc *);
117 void	ar9003_paprd_calib(struct athn_softc *, struct ieee80211_channel *);
118 int	ar9003_get_desired_txgain(struct athn_softc *, int, int);
119 void	ar9003_force_txgain(struct athn_softc *, uint32_t);
120 void	ar9003_set_training_gain(struct athn_softc *, int);
121 int	ar9003_paprd_tx_tone(struct athn_softc *);
122 int	ar9003_compute_predistortion(struct athn_softc *, const uint32_t *,
123 	    const uint32_t *);
124 void	ar9003_enable_predistorter(struct athn_softc *, int);
125 void	ar9003_paprd_enable(struct athn_softc *);
126 void	ar9003_paprd_tx_tone_done(struct athn_softc *);
127 void	ar9003_write_txpower(struct athn_softc *, int16_t power[]);
128 void	ar9003_reset_rx_gain(struct athn_softc *, struct ieee80211_channel *);
129 void	ar9003_reset_tx_gain(struct athn_softc *, struct ieee80211_channel *);
130 void	ar9003_hw_init(struct athn_softc *, struct ieee80211_channel *,
131 	    struct ieee80211_channel *);
132 void	ar9003_get_lg_tpow(struct athn_softc *, struct ieee80211_channel *,
133 	    uint8_t, const uint8_t *, const struct ar_cal_target_power_leg *,
134 	    int, uint8_t[]);
135 void	ar9003_get_ht_tpow(struct athn_softc *, struct ieee80211_channel *,
136 	    uint8_t, const uint8_t *, const struct ar_cal_target_power_ht *,
137 	    int, uint8_t[]);
138 void	ar9003_set_noise_immunity_level(struct athn_softc *, int);
139 void	ar9003_enable_ofdm_weak_signal(struct athn_softc *);
140 void	ar9003_disable_ofdm_weak_signal(struct athn_softc *);
141 void	ar9003_set_cck_weak_signal(struct athn_softc *, int);
142 void	ar9003_set_firstep_level(struct athn_softc *, int);
143 void	ar9003_set_spur_immunity_level(struct athn_softc *, int);
144 
145 /* Extern functions. */
146 void	athn_stop(struct ifnet *, int);
147 int	athn_interpolate(int, int, int, int, int);
148 int	athn_txtime(struct athn_softc *, int, int, u_int);
149 void	athn_inc_tx_trigger_level(struct athn_softc *);
150 int	athn_tx_pending(struct athn_softc *, int);
151 void	athn_stop_tx_dma(struct athn_softc *, int);
152 void	athn_get_delta_slope(uint32_t, uint32_t *, uint32_t *);
153 void	athn_config_pcie(struct athn_softc *);
154 void	athn_config_nonpcie(struct athn_softc *);
155 uint8_t	athn_chan2fbin(struct ieee80211_channel *);
156 
157 
158 int
159 ar9003_attach(struct athn_softc *sc)
160 {
161 	struct athn_ops *ops = &sc->ops;
162 	int error;
163 
164 	/* Set callbacks for AR9003 family. */
165 	ops->gpio_read = ar9003_gpio_read;
166 	ops->gpio_write = ar9003_gpio_write;
167 	ops->gpio_config_input = ar9003_gpio_config_input;
168 	ops->gpio_config_output = ar9003_gpio_config_output;
169 	ops->rfsilent_init = ar9003_rfsilent_init;
170 
171 	ops->dma_alloc = ar9003_dma_alloc;
172 	ops->dma_free = ar9003_dma_free;
173 	ops->rx_enable = ar9003_rx_enable;
174 	ops->intr = ar9003_intr;
175 	ops->tx = ar9003_tx;
176 
177 	ops->set_rf_mode = ar9003_set_rf_mode;
178 	ops->rf_bus_request = ar9003_rf_bus_request;
179 	ops->rf_bus_release = ar9003_rf_bus_release;
180 	ops->set_phy = ar9003_set_phy;
181 	ops->set_delta_slope = ar9003_set_delta_slope;
182 	ops->enable_antenna_diversity = ar9003_enable_antenna_diversity;
183 	ops->init_baseband = ar9003_init_baseband;
184 	ops->disable_phy = ar9003_disable_phy;
185 	ops->set_rxchains = ar9003_set_rxchains;
186 	ops->noisefloor_calib = ar9003_do_noisefloor_calib;
187 	ops->do_calib = ar9003_do_calib;
188 	ops->next_calib = ar9003_next_calib;
189 	ops->hw_init = ar9003_hw_init;
190 
191 	ops->set_noise_immunity_level = ar9003_set_noise_immunity_level;
192 	ops->enable_ofdm_weak_signal = ar9003_enable_ofdm_weak_signal;
193 	ops->disable_ofdm_weak_signal = ar9003_disable_ofdm_weak_signal;
194 	ops->set_cck_weak_signal = ar9003_set_cck_weak_signal;
195 	ops->set_firstep_level = ar9003_set_firstep_level;
196 	ops->set_spur_immunity_level = ar9003_set_spur_immunity_level;
197 
198 	/* Set MAC registers offsets. */
199 	sc->obs_off = AR_OBS;
200 	sc->gpio_input_en_off = AR_GPIO_INPUT_EN_VAL;
201 
202 	if (!(sc->flags & ATHN_FLAG_PCIE))
203 		athn_config_nonpcie(sc);
204 	else
205 		athn_config_pcie(sc);
206 
207 	/* Determine ROM type and location. */
208 	if ((error = ar9003_find_rom(sc)) != 0) {
209 		printf("%s: could not find ROM\n", sc->sc_dev.dv_xname);
210 		return (error);
211 	}
212 	/* Read entire ROM content in memory. */
213 	if ((error = ar9003_read_rom(sc)) != 0) {
214 		printf("%s: could not read ROM\n", sc->sc_dev.dv_xname);
215 		return (error);
216 	}
217 
218 	/* Determine if it is a non-enterprise AR9003 card. */
219 	if (AR_READ(sc, AR_ENT_OTP) & AR_ENT_OTP_MPSD)
220 		sc->flags |= ATHN_FLAG_NON_ENTERPRISE;
221 
222 	ops->setup(sc);
223 	return (0);
224 }
225 
226 /*
227  * Read 16-bit word from EEPROM.
228  */
229 int
230 ar9003_read_eep_word(struct athn_softc *sc, uint32_t addr, uint16_t *val)
231 {
232 	uint32_t reg;
233 	int ntries;
234 
235 	reg = AR_READ(sc, AR_EEPROM_OFFSET(addr));
236 	for (ntries = 0; ntries < 1000; ntries++) {
237 		reg = AR_READ(sc, AR_EEPROM_STATUS_DATA);
238 		if (!(reg & (AR_EEPROM_STATUS_DATA_BUSY |
239 		    AR_EEPROM_STATUS_DATA_PROT_ACCESS))) {
240 			*val = MS(reg, AR_EEPROM_STATUS_DATA_VAL);
241 			return (0);
242 		}
243 		DELAY(10);
244 	}
245 	*val = 0xffff;
246 	return (ETIMEDOUT);
247 }
248 
249 /*
250  * Read an arbitrary number of bytes at a specified address in EEPROM.
251  * NB: The address may not be 16-bit aligned.
252  */
253 int
254 ar9003_read_eep_data(struct athn_softc *sc, uint32_t addr, void *buf, int len)
255 {
256 	uint8_t *dst = buf;
257 	uint16_t val;
258 	int error;
259 
260 	if (len > 0 && (addr & 1)) {
261 		/* Deal with non-aligned reads. */
262 		addr >>= 1;
263 		error = ar9003_read_eep_word(sc, addr, &val);
264 		if (error != 0)
265 			return (error);
266 		*dst++ = val & 0xff;
267 		addr--;
268 		len--;
269 	} else
270 		addr >>= 1;
271 	for (; len >= 2; addr--, len -= 2) {
272 		error = ar9003_read_eep_word(sc, addr, &val);
273 		if (error != 0)
274 			return (error);
275 		*dst++ = val >> 8;
276 		*dst++ = val & 0xff;
277 	}
278 	if (len > 0) {
279 		error = ar9003_read_eep_word(sc, addr, &val);
280 		if (error != 0)
281 			return (error);
282 		*dst++ = val >> 8;
283 	}
284 	return (0);
285 }
286 
287 /*
288  * Read 32-bit word from OTPROM.
289  */
290 int
291 ar9003_read_otp_word(struct athn_softc *sc, uint32_t addr, uint32_t *val)
292 {
293 	uint32_t reg;
294 	int ntries;
295 
296 	reg = AR_READ(sc, AR_OTP_BASE(addr));
297 	for (ntries = 0; ntries < 1000; ntries++) {
298 		reg = AR_READ(sc, AR_OTP_STATUS);
299 		if (MS(reg, AR_OTP_STATUS_TYPE) == AR_OTP_STATUS_VALID) {
300 			*val = AR_READ(sc, AR_OTP_READ_DATA);
301 			return (0);
302 		}
303 		DELAY(10);
304 	}
305 	return (ETIMEDOUT);
306 }
307 
308 /*
309  * Read an arbitrary number of bytes at a specified address in OTPROM.
310  * NB: The address may not be 32-bit aligned.
311  */
312 int
313 ar9003_read_otp_data(struct athn_softc *sc, uint32_t addr, void *buf, int len)
314 {
315 	uint8_t *dst = buf;
316 	uint32_t val;
317 	int error;
318 
319 	/* NB: not optimal for non-aligned reads, but correct. */
320 	for (; len > 0; addr--, len--) {
321 		error = ar9003_read_otp_word(sc, addr >> 2, &val);
322 		if (error != 0)
323 			return (error);
324 		*dst++ = (val >> ((addr & 3) * 8)) & 0xff;
325 	}
326 	return (0);
327 }
328 
329 /*
330  * Determine if the chip has an external EEPROM or an OTPROM and its size.
331  */
332 int
333 ar9003_find_rom(struct athn_softc *sc)
334 {
335 	struct athn_ops *ops = &sc->ops;
336 	uint32_t hdr;
337 	int error;
338 
339 	/* Try EEPROM. */
340 	ops->read_rom_data = ar9003_read_eep_data;
341 
342 	sc->eep_size = AR_SREV_9485(sc) ? 4096 : 1024;
343 	sc->eep_base = sc->eep_size - 1;
344 	error = ops->read_rom_data(sc, sc->eep_base, &hdr, sizeof(hdr));
345 	if (error == 0 && hdr != 0 && hdr != 0xffffffff)
346 		return (0);
347 
348 	sc->eep_size = 512;
349 	sc->eep_base = sc->eep_size - 1;
350 	error = ops->read_rom_data(sc, sc->eep_base, &hdr, sizeof(hdr));
351 	if (error == 0 && hdr != 0 && hdr != 0xffffffff)
352 		return (0);
353 
354 	/* Try OTPROM. */
355 	ops->read_rom_data = ar9003_read_otp_data;
356 
357 	sc->eep_size = 1024;
358 	sc->eep_base = sc->eep_size - 1;
359 	error = ops->read_rom_data(sc, sc->eep_base, &hdr, sizeof(hdr));
360 	if (error == 0 && hdr != 0 && hdr != 0xffffffff)
361 		return (0);
362 
363 	sc->eep_size = 512;
364 	sc->eep_base = sc->eep_size - 1;
365 	error = ops->read_rom_data(sc, sc->eep_base, &hdr, sizeof(hdr));
366 	if (error == 0 && hdr != 0 && hdr != 0xffffffff)
367 		return (0);
368 
369 	return (EIO);	/* Not found. */
370 }
371 
372 int
373 ar9003_restore_rom_block(struct athn_softc *sc, uint8_t alg, uint8_t ref,
374     const uint8_t *buf, int len)
375 {
376 	const uint8_t *def, *ptr, *end;
377 	uint8_t *eep = sc->eep;
378 	int off, clen;
379 
380 	if (alg == AR_EEP_COMPRESS_BLOCK) {
381 		/* Block contains chunks that shadow ROM template. */
382 		def = sc->ops.get_rom_template(sc, ref);
383 		if (def == NULL) {
384 			DPRINTF(("unknown template image %d\n", ref));
385 			return (EINVAL);
386 		}
387 		/* Start with template. */
388 		memcpy(eep, def, sc->eep_size);
389 		/* Shadow template with chunks. */
390 		off = 0;	/* Offset in ROM image. */
391 		ptr = buf;	/* Offset in block. */
392 		end = buf + len;
393 		/* Process chunks. */
394 		while (ptr + 2 <= end) {
395 			off += *ptr++;	/* Gap with previous chunk. */
396 			clen = *ptr++;	/* Chunk length. */
397 			/* Make sure block is large enough. */
398 			if (ptr + clen > end)
399 				return (EINVAL);
400 			/* Make sure chunk fits in ROM image. */
401 			if (off + clen > sc->eep_size)
402 				return (EINVAL);
403 			/* Restore chunk. */
404 			DPRINTFN(2, ("ROM chunk @%d/%d\n", off, clen));
405 			memcpy(&eep[off], ptr, clen);
406 			ptr += clen;
407 			off += clen;
408 		}
409 	} else if (alg == AR_EEP_COMPRESS_NONE) {
410 		/* Block contains full ROM image. */
411 		if (len != sc->eep_size) {
412 			DPRINTF(("block length mismatch %d\n", len));
413 			return (EINVAL);
414 		}
415 		memcpy(eep, buf, len);
416 	}
417 	return (0);
418 }
419 
420 int
421 ar9003_read_rom(struct athn_softc *sc)
422 {
423 	struct athn_ops *ops = &sc->ops;
424 	uint8_t *buf, *ptr, alg, ref;
425 	uint16_t sum, rsum;
426 	uint32_t hdr;
427 	int error, addr, len, i, j;
428 
429 	/* Allocate space to store ROM in host memory. */
430 	sc->eep = malloc(sc->eep_size, M_DEVBUF, M_NOWAIT);
431 	if (sc->eep == NULL)
432 		return (ENOMEM);
433 
434 	/* Allocate temporary buffer to store ROM blocks. */
435 	buf = malloc(2048, M_DEVBUF, M_NOWAIT);
436 	if (buf == NULL)
437 		return (ENOMEM);
438 
439 	/* Restore vendor-specified ROM blocks. */
440 	addr = sc->eep_base;
441 	for (i = 0; i < 100; i++) {
442 		/* Read block header. */
443 		error = ops->read_rom_data(sc, addr, &hdr, sizeof(hdr));
444 		if (error != 0)
445 			break;
446 		if (hdr == 0 || hdr == 0xffffffff)
447 			break;
448 		addr -= sizeof(hdr);
449 
450 		/* Extract bits from header. */
451 		ptr = (uint8_t *)&hdr;
452 		alg = (ptr[0] & 0xe0) >> 5;
453 		ref = (ptr[1] & 0x80) >> 2 | (ptr[0] & 0x1f);
454 		len = (ptr[1] & 0x7f) << 4 | (ptr[2] & 0xf0) >> 4;
455 		DPRINTFN(2, ("ROM block %d: alg=%d ref=%d len=%d\n",
456 		    i, alg, ref, len));
457 
458 		/* Read block data (len <= 0x7ff). */
459 		error = ops->read_rom_data(sc, addr, buf, len);
460 		if (error != 0)
461 			break;
462 		addr -= len;
463 
464 		/* Read block checksum. */
465 		error = ops->read_rom_data(sc, addr, &sum, sizeof(sum));
466 		if (error != 0)
467 			break;
468 		addr -= sizeof(sum);
469 
470 		/* Compute block checksum. */
471 		rsum = 0;
472 		for (j = 0; j < len; j++)
473 			rsum += buf[j];
474 		/* Compare to that in ROM. */
475 		if (letoh16(sum) != rsum) {
476 			DPRINTF(("bad block checksum 0x%x/0x%x\n",
477 			    letoh16(sum), rsum));
478 			continue;	/* Skip bad block. */
479 		}
480 		/* Checksum is correct, restore block. */
481 		ar9003_restore_rom_block(sc, alg, ref, buf, len);
482 	}
483 #if BYTE_ORDER == BIG_ENDIAN
484 	/* NB: ROM is always little endian. */
485 	if (error == 0)
486 		ops->swap_rom(sc);
487 #endif
488 	free(buf, M_DEVBUF, 0);
489 	return (error);
490 }
491 
492 /*
493  * Access to General Purpose Input/Output ports.
494  */
495 int
496 ar9003_gpio_read(struct athn_softc *sc, int pin)
497 {
498 	KASSERT(pin < sc->ngpiopins);
499 	return (((AR_READ(sc, AR_GPIO_IN) & AR9300_GPIO_IN_VAL) &
500 	    (1 << pin)) != 0);
501 }
502 
503 void
504 ar9003_gpio_write(struct athn_softc *sc, int pin, int set)
505 {
506 	uint32_t reg;
507 
508 	KASSERT(pin < sc->ngpiopins);
509 	reg = AR_READ(sc, AR_GPIO_IN_OUT);
510 	if (set)
511 		reg |= 1 << pin;
512 	else
513 		reg &= ~(1 << pin);
514 	AR_WRITE(sc, AR_GPIO_IN_OUT, reg);
515 	AR_WRITE_BARRIER(sc);
516 }
517 
518 void
519 ar9003_gpio_config_input(struct athn_softc *sc, int pin)
520 {
521 	uint32_t reg;
522 
523 	reg = AR_READ(sc, AR_GPIO_OE_OUT);
524 	reg &= ~(AR_GPIO_OE_OUT_DRV_M << (pin * 2));
525 	reg |= AR_GPIO_OE_OUT_DRV_NO << (pin * 2);
526 	AR_WRITE(sc, AR_GPIO_OE_OUT, reg);
527 	AR_WRITE_BARRIER(sc);
528 }
529 
530 void
531 ar9003_gpio_config_output(struct athn_softc *sc, int pin, int type)
532 {
533 	uint32_t reg;
534 	int mux, off;
535 
536 	mux = pin / 6;
537 	off = pin % 6;
538 
539 	reg = AR_READ(sc, AR_GPIO_OUTPUT_MUX(mux));
540 	reg &= ~(0x1f << (off * 5));
541 	reg |= (type & 0x1f) << (off * 5);
542 	AR_WRITE(sc, AR_GPIO_OUTPUT_MUX(mux), reg);
543 
544 	reg = AR_READ(sc, AR_GPIO_OE_OUT);
545 	reg &= ~(AR_GPIO_OE_OUT_DRV_M << (pin * 2));
546 	reg |= AR_GPIO_OE_OUT_DRV_ALL << (pin * 2);
547 	AR_WRITE(sc, AR_GPIO_OE_OUT, reg);
548 	AR_WRITE_BARRIER(sc);
549 }
550 
551 void
552 ar9003_rfsilent_init(struct athn_softc *sc)
553 {
554 	uint32_t reg;
555 
556 	/* Configure hardware radio switch. */
557 	AR_SETBITS(sc, AR_GPIO_INPUT_EN_VAL, AR_GPIO_INPUT_EN_VAL_RFSILENT_BB);
558 	reg = AR_READ(sc, AR_GPIO_INPUT_MUX2);
559 	reg = RW(reg, AR_GPIO_INPUT_MUX2_RFSILENT, 0);
560 	AR_WRITE(sc, AR_GPIO_INPUT_MUX2, reg);
561 	ar9003_gpio_config_input(sc, sc->rfsilent_pin);
562 	AR_SETBITS(sc, AR_PHY_TEST, AR_PHY_TEST_RFSILENT_BB);
563 	if (!(sc->flags & ATHN_FLAG_RFSILENT_REVERSED)) {
564 		AR_SETBITS(sc, AR_GPIO_INTR_POL,
565 		    AR_GPIO_INTR_POL_PIN(sc->rfsilent_pin));
566 	}
567 	AR_WRITE_BARRIER(sc);
568 }
569 
570 int
571 ar9003_dma_alloc(struct athn_softc *sc)
572 {
573 	int error;
574 
575 	error = ar9003_tx_alloc(sc);
576 	if (error != 0)
577 		return (error);
578 
579 	error = ar9003_rx_alloc(sc, ATHN_QID_LP, AR9003_RX_LP_QDEPTH);
580 	if (error != 0)
581 		return (error);
582 
583 	error = ar9003_rx_alloc(sc, ATHN_QID_HP, AR9003_RX_HP_QDEPTH);
584 	if (error != 0)
585 		return (error);
586 
587 	return (0);
588 }
589 
590 void
591 ar9003_dma_free(struct athn_softc *sc)
592 {
593 	ar9003_tx_free(sc);
594 	ar9003_rx_free(sc, ATHN_QID_LP);
595 	ar9003_rx_free(sc, ATHN_QID_HP);
596 }
597 
598 int
599 ar9003_tx_alloc(struct athn_softc *sc)
600 {
601 	struct athn_tx_buf *bf;
602 	bus_size_t size;
603 	int error, nsegs, i;
604 
605 	/*
606 	 * Allocate Tx status ring.
607 	 */
608 	size = AR9003_NTXSTATUS * sizeof(struct ar_tx_status);
609 
610 	error = bus_dmamap_create(sc->sc_dmat, size, 1, size, 0,
611 	    BUS_DMA_NOWAIT, &sc->txsmap);
612 	if (error != 0)
613 		goto fail;
614 
615 	error = bus_dmamem_alloc(sc->sc_dmat, size, 4, 0, &sc->txsseg, 1,
616 	    &nsegs, BUS_DMA_NOWAIT | BUS_DMA_ZERO);
617 	if (error != 0)
618 		goto fail;
619 
620 	error = bus_dmamem_map(sc->sc_dmat, &sc->txsseg, 1, size,
621 	    (caddr_t *)&sc->txsring, BUS_DMA_NOWAIT | BUS_DMA_COHERENT);
622 	if (error != 0)
623 		goto fail;
624 
625 	error = bus_dmamap_load_raw(sc->sc_dmat, sc->txsmap, &sc->txsseg,
626 	    1, size, BUS_DMA_NOWAIT | BUS_DMA_READ);
627 	if (error != 0)
628 		goto fail;
629 
630 	/*
631 	 * Allocate a pool of Tx descriptors shared between all Tx queues.
632 	 */
633 	size = ATHN_NTXBUFS * sizeof(struct ar_tx_desc);
634 
635 	error = bus_dmamap_create(sc->sc_dmat, size, 1, size, 0,
636 	    BUS_DMA_NOWAIT, &sc->map);
637 	if (error != 0)
638 		goto fail;
639 
640 	error = bus_dmamem_alloc(sc->sc_dmat, size, 4, 0, &sc->seg, 1,
641 	    &nsegs, BUS_DMA_NOWAIT | BUS_DMA_ZERO);
642 	if (error != 0)
643 		goto fail;
644 
645 	error = bus_dmamem_map(sc->sc_dmat, &sc->seg, 1, size,
646 	    (caddr_t *)&sc->descs, BUS_DMA_NOWAIT | BUS_DMA_COHERENT);
647 	if (error != 0)
648 		goto fail;
649 
650 	error = bus_dmamap_load_raw(sc->sc_dmat, sc->map, &sc->seg, 1, size,
651 	    BUS_DMA_NOWAIT | BUS_DMA_WRITE);
652 	if (error != 0)
653 		goto fail;
654 
655 	SIMPLEQ_INIT(&sc->txbufs);
656 	for (i = 0; i < ATHN_NTXBUFS; i++) {
657 		bf = &sc->txpool[i];
658 
659 		error = bus_dmamap_create(sc->sc_dmat, ATHN_TXBUFSZ,
660 		    AR9003_MAX_SCATTER, ATHN_TXBUFSZ, 0, BUS_DMA_NOWAIT,
661 		    &bf->bf_map);
662 		if (error != 0) {
663 			printf("%s: could not create Tx buf DMA map\n",
664 			    sc->sc_dev.dv_xname);
665 			goto fail;
666 		}
667 
668 		bf->bf_descs = &((struct ar_tx_desc *)sc->descs)[i];
669 		bf->bf_daddr = sc->map->dm_segs[0].ds_addr +
670 		    i * sizeof(struct ar_tx_desc);
671 
672 		SIMPLEQ_INSERT_TAIL(&sc->txbufs, bf, bf_list);
673 	}
674 	return (0);
675  fail:
676 	ar9003_tx_free(sc);
677 	return (error);
678 }
679 
680 void
681 ar9003_tx_free(struct athn_softc *sc)
682 {
683 	struct athn_tx_buf *bf;
684 	int i;
685 
686 	for (i = 0; i < ATHN_NTXBUFS; i++) {
687 		bf = &sc->txpool[i];
688 
689 		if (bf->bf_map != NULL)
690 			bus_dmamap_destroy(sc->sc_dmat, bf->bf_map);
691 	}
692 	/* Free Tx descriptors. */
693 	if (sc->map != NULL) {
694 		if (sc->descs != NULL) {
695 			bus_dmamap_unload(sc->sc_dmat, sc->map);
696 			bus_dmamem_unmap(sc->sc_dmat, (caddr_t)sc->descs,
697 			    ATHN_NTXBUFS * sizeof(struct ar_tx_desc));
698 			bus_dmamem_free(sc->sc_dmat, &sc->seg, 1);
699 		}
700 		bus_dmamap_destroy(sc->sc_dmat, sc->map);
701 	}
702 	/* Free Tx status ring. */
703 	if (sc->txsmap != NULL) {
704 		if (sc->txsring != NULL) {
705 			bus_dmamap_unload(sc->sc_dmat, sc->txsmap);
706 			bus_dmamem_unmap(sc->sc_dmat, (caddr_t)sc->txsring,
707 			     AR9003_NTXSTATUS * sizeof(struct ar_tx_status));
708 			bus_dmamem_free(sc->sc_dmat, &sc->txsseg, 1);
709 		}
710 		bus_dmamap_destroy(sc->sc_dmat, sc->txsmap);
711 	}
712 }
713 
714 int
715 ar9003_rx_alloc(struct athn_softc *sc, int qid, int count)
716 {
717 	struct athn_rxq *rxq = &sc->rxq[qid];
718 	struct athn_rx_buf *bf;
719 	struct ar_rx_status *ds;
720 	int error, i;
721 
722 	rxq->bf = mallocarray(count, sizeof(*bf), M_DEVBUF,
723 	    M_NOWAIT | M_ZERO);
724 	if (rxq->bf == NULL)
725 		return (ENOMEM);
726 
727 	rxq->count = count;
728 
729 	for (i = 0; i < rxq->count; i++) {
730 		bf = &rxq->bf[i];
731 
732 		error = bus_dmamap_create(sc->sc_dmat, ATHN_RXBUFSZ, 1,
733 		    ATHN_RXBUFSZ, 0, BUS_DMA_NOWAIT | BUS_DMA_ALLOCNOW,
734 		    &bf->bf_map);
735 		if (error != 0) {
736 			printf("%s: could not create Rx buf DMA map\n",
737 			    sc->sc_dev.dv_xname);
738 			goto fail;
739 		}
740 		/*
741 		 * Assumes MCLGETI returns cache-line-size aligned buffers.
742 		 */
743 		bf->bf_m = MCLGETI(NULL, M_DONTWAIT, NULL, ATHN_RXBUFSZ);
744 		if (bf->bf_m == NULL) {
745 			printf("%s: could not allocate Rx mbuf\n",
746 			    sc->sc_dev.dv_xname);
747 			error = ENOBUFS;
748 			goto fail;
749 		}
750 
751 		error = bus_dmamap_load(sc->sc_dmat, bf->bf_map,
752 		    mtod(bf->bf_m, void *), ATHN_RXBUFSZ, NULL,
753 		    BUS_DMA_NOWAIT);
754 		if (error != 0) {
755 			printf("%s: could not DMA map Rx buffer\n",
756 			    sc->sc_dev.dv_xname);
757 			goto fail;
758 		}
759 
760 		ds = mtod(bf->bf_m, struct ar_rx_status *);
761 		memset(ds, 0, sizeof(*ds));
762 		bf->bf_desc = ds;
763 		bf->bf_daddr = bf->bf_map->dm_segs[0].ds_addr;
764 
765 		bus_dmamap_sync(sc->sc_dmat, bf->bf_map, 0, ATHN_RXBUFSZ,
766 		    BUS_DMASYNC_PREREAD);
767 	}
768 	return (0);
769  fail:
770 	ar9003_rx_free(sc, qid);
771 	return (error);
772 }
773 
774 void
775 ar9003_rx_free(struct athn_softc *sc, int qid)
776 {
777 	struct athn_rxq *rxq = &sc->rxq[qid];
778 	struct athn_rx_buf *bf;
779 	int i;
780 
781 	if (rxq->bf == NULL)
782 		return;
783 	for (i = 0; i < rxq->count; i++) {
784 		bf = &rxq->bf[i];
785 
786 		if (bf->bf_map != NULL)
787 			bus_dmamap_destroy(sc->sc_dmat, bf->bf_map);
788 		if (bf->bf_m != NULL)
789 			m_freem(bf->bf_m);
790 	}
791 	free(rxq->bf, M_DEVBUF, 0);
792 }
793 
794 void
795 ar9003_reset_txsring(struct athn_softc *sc)
796 {
797 	sc->txscur = 0;
798 	memset(sc->txsring, 0, AR9003_NTXSTATUS * sizeof(struct ar_tx_status));
799 	AR_WRITE(sc, AR_Q_STATUS_RING_START,
800 	    sc->txsmap->dm_segs[0].ds_addr);
801 	AR_WRITE(sc, AR_Q_STATUS_RING_END,
802 	    sc->txsmap->dm_segs[0].ds_addr + sc->txsmap->dm_segs[0].ds_len);
803 	AR_WRITE_BARRIER(sc);
804 }
805 
806 void
807 ar9003_rx_enable(struct athn_softc *sc)
808 {
809 	struct athn_rxq *rxq;
810 	struct athn_rx_buf *bf;
811 	struct ar_rx_status *ds;
812 	uint32_t reg;
813 	int qid, i;
814 
815 	reg = AR_READ(sc, AR_RXBP_THRESH);
816 	reg = RW(reg, AR_RXBP_THRESH_HP, 1);
817 	reg = RW(reg, AR_RXBP_THRESH_LP, 1);
818 	AR_WRITE(sc, AR_RXBP_THRESH, reg);
819 
820 	/* Set Rx buffer size. */
821 	AR_WRITE(sc, AR_DATABUF_SIZE, ATHN_RXBUFSZ - sizeof(*ds));
822 
823 	for (qid = 0; qid < 2; qid++) {
824 		rxq = &sc->rxq[qid];
825 
826 		/* Setup Rx status descriptors. */
827 		SIMPLEQ_INIT(&rxq->head);
828 		for (i = 0; i < rxq->count; i++) {
829 			bf = &rxq->bf[i];
830 			ds = bf->bf_desc;
831 
832 			memset(ds, 0, sizeof(*ds));
833 			if (qid == ATHN_QID_LP)
834 				AR_WRITE(sc, AR_LP_RXDP, bf->bf_daddr);
835 			else
836 				AR_WRITE(sc, AR_HP_RXDP, bf->bf_daddr);
837 			AR_WRITE_BARRIER(sc);
838 			SIMPLEQ_INSERT_TAIL(&rxq->head, bf, bf_list);
839 		}
840 	}
841 	/* Enable Rx. */
842 	AR_WRITE(sc, AR_CR, 0);
843 	AR_WRITE_BARRIER(sc);
844 }
845 
846 #if NBPFILTER > 0
847 void
848 ar9003_rx_radiotap(struct athn_softc *sc, struct mbuf *m,
849     struct ar_rx_status *ds)
850 {
851 #define IEEE80211_RADIOTAP_F_SHORTGI	0x80	/* XXX from FBSD */
852 
853 	struct athn_rx_radiotap_header *tap = &sc->sc_rxtap;
854 	struct ieee80211com *ic = &sc->sc_ic;
855 	struct mbuf mb;
856 	uint64_t tsf;
857 	uint32_t tstamp;
858 	uint8_t rate;
859 
860 	/* Extend the 15-bit timestamp from Rx status to 64-bit TSF. */
861 	tstamp = ds->ds_status3;
862 	tsf = AR_READ(sc, AR_TSF_U32);
863 	tsf = tsf << 32 | AR_READ(sc, AR_TSF_L32);
864 	if ((tsf & 0x7fff) < tstamp)
865 		tsf -= 0x8000;
866 	tsf = (tsf & ~0x7fff) | tstamp;
867 
868 	tap->wr_flags = IEEE80211_RADIOTAP_F_FCS;
869 	tap->wr_tsft = htole64(tsf);
870 	tap->wr_chan_freq = htole16(ic->ic_bss->ni_chan->ic_freq);
871 	tap->wr_chan_flags = htole16(ic->ic_bss->ni_chan->ic_flags);
872 	tap->wr_dbm_antsignal = MS(ds->ds_status5, AR_RXS5_RSSI_COMBINED);
873 	/* XXX noise. */
874 	tap->wr_antenna = MS(ds->ds_status4, AR_RXS4_ANTENNA);
875 	tap->wr_rate = 0;	/* In case it can't be found below. */
876 	rate = MS(ds->ds_status1, AR_RXS1_RATE);
877 	if (rate & 0x80) {		/* HT. */
878 		/* Bit 7 set means HT MCS instead of rate. */
879 		tap->wr_rate = rate;
880 		if (!(ds->ds_status4 & AR_RXS4_GI))
881 			tap->wr_flags |= IEEE80211_RADIOTAP_F_SHORTGI;
882 
883 	} else if (rate & 0x10) {	/* CCK. */
884 		if (rate & 0x04)
885 			tap->wr_flags |= IEEE80211_RADIOTAP_F_SHORTPRE;
886 		switch (rate & ~0x14) {
887 		case 0xb: tap->wr_rate =   2; break;
888 		case 0xa: tap->wr_rate =   4; break;
889 		case 0x9: tap->wr_rate =  11; break;
890 		case 0x8: tap->wr_rate =  22; break;
891 		}
892 	} else {			/* OFDM. */
893 		switch (rate) {
894 		case 0xb: tap->wr_rate =  12; break;
895 		case 0xf: tap->wr_rate =  18; break;
896 		case 0xa: tap->wr_rate =  24; break;
897 		case 0xe: tap->wr_rate =  36; break;
898 		case 0x9: tap->wr_rate =  48; break;
899 		case 0xd: tap->wr_rate =  72; break;
900 		case 0x8: tap->wr_rate =  96; break;
901 		case 0xc: tap->wr_rate = 108; break;
902 		}
903 	}
904 	mb.m_data = (caddr_t)tap;
905 	mb.m_len = sc->sc_rxtap_len;
906 	mb.m_next = m;
907 	mb.m_nextpkt = NULL;
908 	mb.m_type = 0;
909 	mb.m_flags = 0;
910 	bpf_mtap(sc->sc_drvbpf, &mb, BPF_DIRECTION_IN);
911 }
912 #endif
913 
914 int
915 ar9003_rx_process(struct athn_softc *sc, int qid)
916 {
917 	struct ieee80211com *ic = &sc->sc_ic;
918 	struct ifnet *ifp = &ic->ic_if;
919 	struct athn_rxq *rxq = &sc->rxq[qid];
920 	struct athn_rx_buf *bf;
921 	struct ar_rx_status *ds;
922 	struct ieee80211_frame *wh;
923 	struct ieee80211_rxinfo rxi;
924 	struct ieee80211_node *ni;
925 	struct mbuf *m, *m1;
926 	int error, len;
927 
928 	bf = SIMPLEQ_FIRST(&rxq->head);
929 	if (__predict_false(bf == NULL)) {	/* Should not happen. */
930 		printf("%s: Rx queue is empty!\n", sc->sc_dev.dv_xname);
931 		return (ENOENT);
932 	}
933 	bus_dmamap_sync(sc->sc_dmat, bf->bf_map, 0, ATHN_RXBUFSZ,
934 	    BUS_DMASYNC_POSTREAD);
935 
936 	ds = mtod(bf->bf_m, struct ar_rx_status *);
937 	if (!(ds->ds_status1 & AR_RXS1_DONE))
938 		return (EBUSY);
939 
940 	/* Check that it is a valid Rx status descriptor. */
941 	if ((ds->ds_info & (AR_RXI_DESC_ID_M | AR_RXI_DESC_TX |
942 	    AR_RXI_CTRL_STAT)) != SM(AR_RXI_DESC_ID, AR_VENDOR_ATHEROS))
943 		goto skip;
944 
945 	if (!(ds->ds_status11 & AR_RXS11_FRAME_OK)) {
946 		if (ds->ds_status11 & AR_RXS11_CRC_ERR)
947 			DPRINTFN(6, ("CRC error\n"));
948 		else if (ds->ds_status11 & AR_RXS11_PHY_ERR)
949 			DPRINTFN(6, ("PHY error=0x%x\n",
950 			    MS(ds->ds_status11, AR_RXS11_PHY_ERR_CODE)));
951 		else if (ds->ds_status11 & AR_RXS11_DECRYPT_CRC_ERR)
952 			DPRINTFN(6, ("Decryption CRC error\n"));
953 		else if (ds->ds_status11 & AR_RXS11_MICHAEL_ERR) {
954 			DPRINTFN(2, ("Michael MIC failure\n"));
955 			/* Report Michael MIC failures to net80211. */
956 			ic->ic_stats.is_rx_locmicfail++;
957 			ieee80211_michael_mic_failure(ic, 0);
958 			/*
959 			 * XXX Check that it is not a control frame
960 			 * (invalid MIC failures on valid ctl frames).
961 			 */
962 		}
963 		ifp->if_ierrors++;
964 		goto skip;
965 	}
966 
967 	len = MS(ds->ds_status2, AR_RXS2_DATA_LEN);
968 	if (__predict_false(len < IEEE80211_MIN_LEN ||
969 	    len > ATHN_RXBUFSZ - sizeof(*ds))) {
970 		DPRINTF(("corrupted descriptor length=%d\n", len));
971 		ifp->if_ierrors++;
972 		goto skip;
973 	}
974 
975 	/* Allocate a new Rx buffer. */
976 	m1 = MCLGETI(NULL, M_DONTWAIT, NULL, ATHN_RXBUFSZ);
977 	if (__predict_false(m1 == NULL)) {
978 		ic->ic_stats.is_rx_nombuf++;
979 		ifp->if_ierrors++;
980 		goto skip;
981 	}
982 
983 	/* Unmap the old Rx buffer. */
984 	bus_dmamap_unload(sc->sc_dmat, bf->bf_map);
985 
986 	/* Map the new Rx buffer. */
987 	error = bus_dmamap_load(sc->sc_dmat, bf->bf_map, mtod(m1, void *),
988 	    ATHN_RXBUFSZ, NULL, BUS_DMA_NOWAIT | BUS_DMA_READ);
989 	if (__predict_false(error != 0)) {
990 		m_freem(m1);
991 
992 		/* Remap the old Rx buffer or panic. */
993 		error = bus_dmamap_load(sc->sc_dmat, bf->bf_map,
994 		    mtod(bf->bf_m, void *), ATHN_RXBUFSZ, NULL,
995 		    BUS_DMA_NOWAIT | BUS_DMA_READ);
996 		KASSERT(error != 0);
997 		bf->bf_daddr = bf->bf_map->dm_segs[0].ds_addr;
998 		ifp->if_ierrors++;
999 		goto skip;
1000 	}
1001 	bf->bf_desc = mtod(m1, struct ar_rx_status *);
1002 	bf->bf_daddr = bf->bf_map->dm_segs[0].ds_addr;
1003 
1004 	m = bf->bf_m;
1005 	bf->bf_m = m1;
1006 
1007 	/* Finalize mbuf. */
1008 	/* Strip Rx status descriptor from head. */
1009 	m->m_data = (caddr_t)&ds[1];
1010 	m->m_pkthdr.len = m->m_len = len;
1011 
1012 	/* Grab a reference to the source node. */
1013 	wh = mtod(m, struct ieee80211_frame *);
1014 	ni = ieee80211_find_rxnode(ic, wh);
1015 
1016 	/* Remove any HW padding after the 802.11 header. */
1017 	if (!(wh->i_fc[0] & IEEE80211_FC0_TYPE_CTL)) {
1018 		u_int hdrlen = ieee80211_get_hdrlen(wh);
1019 		if (hdrlen & 3) {
1020 			memmove((caddr_t)wh + 2, wh, hdrlen);
1021 			m_adj(m, 2);
1022 		}
1023 	}
1024 #if NBPFILTER > 0
1025 	if (__predict_false(sc->sc_drvbpf != NULL))
1026 		ar9003_rx_radiotap(sc, m, ds);
1027 #endif
1028 	/* Trim 802.11 FCS after radiotap. */
1029 	m_adj(m, -IEEE80211_CRC_LEN);
1030 
1031 	/* Send the frame to the 802.11 layer. */
1032 	rxi.rxi_flags = 0;	/* XXX */
1033 	rxi.rxi_rssi = MS(ds->ds_status5, AR_RXS5_RSSI_COMBINED);
1034 	rxi.rxi_tstamp = ds->ds_status3;
1035 	ieee80211_input(ifp, m, ni, &rxi);
1036 
1037 	/* Node is no longer needed. */
1038 	ieee80211_release_node(ic, ni);
1039 
1040  skip:
1041 	/* Unlink this descriptor from head. */
1042 	SIMPLEQ_REMOVE_HEAD(&rxq->head, bf_list);
1043 	memset(bf->bf_desc, 0, sizeof(*ds));
1044 
1045 	/* Re-use this descriptor and link it to tail. */
1046 	bus_dmamap_sync(sc->sc_dmat, bf->bf_map, 0, ATHN_RXBUFSZ,
1047 	    BUS_DMASYNC_PREREAD);
1048 
1049 	if (qid == ATHN_QID_LP)
1050 		AR_WRITE(sc, AR_LP_RXDP, bf->bf_daddr);
1051 	else
1052 		AR_WRITE(sc, AR_HP_RXDP, bf->bf_daddr);
1053 	AR_WRITE_BARRIER(sc);
1054 	SIMPLEQ_INSERT_TAIL(&rxq->head, bf, bf_list);
1055 
1056 	/* Re-enable Rx. */
1057 	AR_WRITE(sc, AR_CR, 0);
1058 	AR_WRITE_BARRIER(sc);
1059 	return (0);
1060 }
1061 
1062 void
1063 ar9003_rx_intr(struct athn_softc *sc, int qid)
1064 {
1065 	while (ar9003_rx_process(sc, qid) == 0);
1066 }
1067 
1068 int
1069 ar9003_tx_process(struct athn_softc *sc)
1070 {
1071 	struct ieee80211com *ic = &sc->sc_ic;
1072 	struct ifnet *ifp = &ic->ic_if;
1073 	struct athn_txq *txq;
1074 	struct athn_node *an;
1075 	struct athn_tx_buf *bf;
1076 	struct ar_tx_status *ds;
1077 	uint8_t qid, failcnt;
1078 
1079 	ds = &((struct ar_tx_status *)sc->txsring)[sc->txscur];
1080 	if (!(ds->ds_status8 & AR_TXS8_DONE))
1081 		return (EBUSY);
1082 
1083 	sc->txscur = (sc->txscur + 1) % AR9003_NTXSTATUS;
1084 
1085 	/* Check that it is a valid Tx status descriptor. */
1086 	if ((ds->ds_info & (AR_TXI_DESC_ID_M | AR_TXI_DESC_TX)) !=
1087 	    (SM(AR_TXI_DESC_ID, AR_VENDOR_ATHEROS) | AR_TXI_DESC_TX)) {
1088 		memset(ds, 0, sizeof(*ds));
1089 		return (0);
1090 	}
1091 	/* Retrieve the queue that was used to send this PDU. */
1092 	qid = MS(ds->ds_info, AR_TXI_QCU_NUM);
1093 	txq = &sc->txq[qid];
1094 
1095 	bf = SIMPLEQ_FIRST(&txq->head);
1096 	if (bf == NULL || bf == txq->wait) {
1097 		memset(ds, 0, sizeof(*ds));
1098 		return (0);
1099 	}
1100 	SIMPLEQ_REMOVE_HEAD(&txq->head, bf_list);
1101 	ifp->if_opackets++;
1102 
1103 	sc->sc_tx_timer = 0;
1104 
1105 	if (ds->ds_status3 & AR_TXS3_EXCESSIVE_RETRIES)
1106 		ifp->if_oerrors++;
1107 
1108 	if (ds->ds_status3 & AR_TXS3_UNDERRUN)
1109 		athn_inc_tx_trigger_level(sc);
1110 
1111 	/* Wakeup PA predistortion state machine. */
1112 	if (bf->bf_txflags & ATHN_TXFLAG_PAPRD)
1113 		ar9003_paprd_tx_tone_done(sc);
1114 
1115 	an = (struct athn_node *)bf->bf_ni;
1116 	/*
1117 	 * NB: the data fail count contains the number of un-acked tries
1118 	 * for the final series used.  We must add the number of tries for
1119 	 * each series that was fully processed.
1120 	 */
1121 	failcnt  = MS(ds->ds_status3, AR_TXS3_DATA_FAIL_CNT);
1122 	/* NB: Assume two tries per series. */
1123 	failcnt += MS(ds->ds_status8, AR_TXS8_FINAL_IDX) * 2;
1124 
1125 	/* Update rate control statistics. */
1126 	an->amn.amn_txcnt++;
1127 	if (failcnt > 0)
1128 		an->amn.amn_retrycnt++;
1129 
1130 	DPRINTFN(5, ("Tx done qid=%d status3=%d fail count=%d\n",
1131 	    qid, ds->ds_status3, failcnt));
1132 
1133 	/* Reset Tx status descriptor. */
1134 	memset(ds, 0, sizeof(*ds));
1135 
1136 	/* Unmap Tx buffer. */
1137 	bus_dmamap_sync(sc->sc_dmat, bf->bf_map, 0, bf->bf_map->dm_mapsize,
1138 	    BUS_DMASYNC_POSTWRITE);
1139 	bus_dmamap_unload(sc->sc_dmat, bf->bf_map);
1140 
1141 	m_freem(bf->bf_m);
1142 	bf->bf_m = NULL;
1143 	ieee80211_release_node(ic, bf->bf_ni);
1144 	bf->bf_ni = NULL;
1145 
1146 	/* Link Tx buffer back to global free list. */
1147 	SIMPLEQ_INSERT_TAIL(&sc->txbufs, bf, bf_list);
1148 
1149 	/* Queue buffers that are waiting if there is new room. */
1150 	if (--txq->queued < AR9003_TX_QDEPTH && txq->wait != NULL) {
1151 		AR_WRITE(sc, AR_QTXDP(qid), txq->wait->bf_daddr);
1152 		AR_WRITE_BARRIER(sc);
1153 		txq->wait = SIMPLEQ_NEXT(txq->wait, bf_list);
1154 	}
1155 	return (0);
1156 }
1157 
1158 void
1159 ar9003_tx_intr(struct athn_softc *sc)
1160 {
1161 	struct ieee80211com *ic = &sc->sc_ic;
1162 	struct ifnet *ifp = &ic->ic_if;
1163 
1164 	while (ar9003_tx_process(sc) == 0);
1165 
1166 	if (!SIMPLEQ_EMPTY(&sc->txbufs)) {
1167 		ifq_clr_oactive(&ifp->if_snd);
1168 		ifp->if_start(ifp);
1169 	}
1170 }
1171 
1172 #ifndef IEEE80211_STA_ONLY
1173 /*
1174  * Process Software Beacon Alert interrupts.
1175  */
1176 int
1177 ar9003_swba_intr(struct athn_softc *sc)
1178 {
1179 	struct ieee80211com *ic = &sc->sc_ic;
1180 	struct ifnet *ifp = &ic->ic_if;
1181 	struct ieee80211_node *ni = ic->ic_bss;
1182 	struct athn_tx_buf *bf = sc->bcnbuf;
1183 	struct ieee80211_frame *wh;
1184 	struct ar_tx_desc *ds;
1185 	struct mbuf *m;
1186 	uint32_t sum;
1187 	uint8_t ridx, hwrate;
1188 	int error, totlen;
1189 
1190 	if (ic->ic_tim_mcast_pending &&
1191 	    mq_empty(&ni->ni_savedq) &&
1192 	    SIMPLEQ_EMPTY(&sc->txq[ATHN_QID_CAB].head))
1193 		ic->ic_tim_mcast_pending = 0;
1194 
1195 	if (ic->ic_dtim_count == 0)
1196 		ic->ic_dtim_count = ic->ic_dtim_period - 1;
1197 	else
1198 		ic->ic_dtim_count--;
1199 
1200 	/* Make sure previous beacon has been sent. */
1201 	if (athn_tx_pending(sc, ATHN_QID_BEACON)) {
1202 		DPRINTF(("beacon stuck\n"));
1203 		return (EBUSY);
1204 	}
1205 	/* Get new beacon. */
1206 	m = ieee80211_beacon_alloc(ic, ic->ic_bss);
1207 	if (__predict_false(m == NULL))
1208 		return (ENOBUFS);
1209 	/* Assign sequence number. */
1210 	wh = mtod(m, struct ieee80211_frame *);
1211 	*(uint16_t *)&wh->i_seq[0] =
1212 	    htole16(ic->ic_bss->ni_txseq << IEEE80211_SEQ_SEQ_SHIFT);
1213 	ic->ic_bss->ni_txseq++;
1214 
1215 	/* Unmap and free old beacon if any. */
1216 	if (__predict_true(bf->bf_m != NULL)) {
1217 		bus_dmamap_sync(sc->sc_dmat, bf->bf_map, 0,
1218 		    bf->bf_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1219 		bus_dmamap_unload(sc->sc_dmat, bf->bf_map);
1220 		m_freem(bf->bf_m);
1221 		bf->bf_m = NULL;
1222 	}
1223 	/* DMA map new beacon. */
1224 	error = bus_dmamap_load_mbuf(sc->sc_dmat, bf->bf_map, m,
1225 	    BUS_DMA_NOWAIT | BUS_DMA_WRITE);
1226 	if (__predict_false(error != 0)) {
1227 		m_freem(m);
1228 		return (error);
1229 	}
1230 	bf->bf_m = m;
1231 
1232 	/* Setup Tx descriptor (simplified ar9003_tx()). */
1233 	ds = bf->bf_descs;
1234 	memset(ds, 0, sizeof(*ds));
1235 
1236 	ds->ds_info =
1237 	    SM(AR_TXI_DESC_ID, AR_VENDOR_ATHEROS) |
1238 	    SM(AR_TXI_DESC_NDWORDS, 23) |
1239 	    SM(AR_TXI_QCU_NUM, ATHN_QID_BEACON) |
1240 	    AR_TXI_DESC_TX | AR_TXI_CTRL_STAT;
1241 
1242 	totlen = m->m_pkthdr.len + IEEE80211_CRC_LEN;
1243 	ds->ds_ctl11 = SM(AR_TXC11_FRAME_LEN, totlen);
1244 	ds->ds_ctl11 |= SM(AR_TXC11_XMIT_POWER, AR_MAX_RATE_POWER);
1245 	ds->ds_ctl12 = SM(AR_TXC12_FRAME_TYPE, AR_FRAME_TYPE_BEACON);
1246 	ds->ds_ctl12 |= AR_TXC12_NO_ACK;
1247 	ds->ds_ctl17 = SM(AR_TXC17_ENCR_TYPE, AR_ENCR_TYPE_CLEAR);
1248 
1249 	/* Write number of tries. */
1250 	ds->ds_ctl13 = SM(AR_TXC13_XMIT_DATA_TRIES0, 1);
1251 
1252 	/* Write Tx rate. */
1253 	ridx = (ic->ic_curmode == IEEE80211_MODE_11A) ?
1254 	    ATHN_RIDX_OFDM6 : ATHN_RIDX_CCK1;
1255 	hwrate = athn_rates[ridx].hwrate;
1256 	ds->ds_ctl14 = SM(AR_TXC14_XMIT_RATE0, hwrate);
1257 
1258 	/* Write Tx chains. */
1259 	ds->ds_ctl18 = SM(AR_TXC18_CHAIN_SEL0, sc->txchainmask);
1260 
1261 	ds->ds_segs[0].ds_data = bf->bf_map->dm_segs[0].ds_addr;
1262 	/* Segment length must be a multiple of 4. */
1263 	ds->ds_segs[0].ds_ctl |= SM(AR_TXC_BUF_LEN,
1264 	    (bf->bf_map->dm_segs[0].ds_len + 3) & ~3);
1265 	/* Compute Tx descriptor checksum. */
1266 	sum = ds->ds_info;
1267 	sum += ds->ds_segs[0].ds_data;
1268 	sum += ds->ds_segs[0].ds_ctl;
1269 	sum = (sum >> 16) + (sum & 0xffff);
1270 	ds->ds_ctl10 = SM(AR_TXC10_PTR_CHK_SUM, sum);
1271 
1272 	bus_dmamap_sync(sc->sc_dmat, bf->bf_map, 0, bf->bf_map->dm_mapsize,
1273 	    BUS_DMASYNC_PREWRITE);
1274 
1275 	/* Stop Tx DMA before putting the new beacon on the queue. */
1276 	athn_stop_tx_dma(sc, ATHN_QID_BEACON);
1277 
1278 	AR_WRITE(sc, AR_QTXDP(ATHN_QID_BEACON), bf->bf_daddr);
1279 
1280 	for(;;) {
1281 		if (SIMPLEQ_EMPTY(&sc->txbufs))
1282 			break;
1283 
1284 		m = mq_dequeue(&ni->ni_savedq);
1285 		if (m == NULL)
1286 			break;
1287 		if (!mq_empty(&ni->ni_savedq)) {
1288 			/* more queued frames, set the more data bit */
1289 			wh = mtod(m, struct ieee80211_frame *);
1290 			wh->i_fc[1] |= IEEE80211_FC1_MORE_DATA;
1291 		}
1292 
1293 		if (sc->ops.tx(sc, m, ni, ATHN_TXFLAG_CAB) != 0) {
1294 			ieee80211_release_node(ic, ni);
1295 			ifp->if_oerrors++;
1296 			break;
1297 		}
1298 	}
1299 
1300 	/* Kick Tx. */
1301 	AR_WRITE(sc, AR_Q_TXE, 1 << ATHN_QID_BEACON);
1302 	AR_WRITE_BARRIER(sc);
1303 	return (0);
1304 }
1305 #endif
1306 
1307 int
1308 ar9003_intr(struct athn_softc *sc)
1309 {
1310 	uint32_t intr, intr2, intr5, sync;
1311 
1312 	/* Get pending interrupts. */
1313 	intr = AR_READ(sc, AR_INTR_ASYNC_CAUSE);
1314 	if (!(intr & AR_INTR_MAC_IRQ) || intr == AR_INTR_SPURIOUS) {
1315 		intr = AR_READ(sc, AR_INTR_SYNC_CAUSE);
1316 		if (intr == AR_INTR_SPURIOUS || (intr & sc->isync) == 0)
1317 			return (0);	/* Not for us. */
1318 	}
1319 
1320 	if ((AR_READ(sc, AR_INTR_ASYNC_CAUSE) & AR_INTR_MAC_IRQ) &&
1321 	    (AR_READ(sc, AR_RTC_STATUS) & AR_RTC_STATUS_M) == AR_RTC_STATUS_ON)
1322 		intr = AR_READ(sc, AR_ISR);
1323 	else
1324 		intr = 0;
1325 	sync = AR_READ(sc, AR_INTR_SYNC_CAUSE) & sc->isync;
1326 	if (intr == 0 && sync == 0)
1327 		return (0);	/* Not for us. */
1328 
1329 	if (intr != 0) {
1330 		if (intr & AR_ISR_BCNMISC) {
1331 			intr2 = AR_READ(sc, AR_ISR_S2);
1332 			if (intr2 & AR_ISR_S2_TIM)
1333 				/* TBD */;
1334 			if (intr2 & AR_ISR_S2_TSFOOR)
1335 				/* TBD */;
1336 			if (intr2 & AR_ISR_S2_BB_WATCHDOG)
1337 				/* TBD */;
1338 		}
1339 		intr = AR_READ(sc, AR_ISR_RAC);
1340 		if (intr == AR_INTR_SPURIOUS)
1341 			return (1);
1342 
1343 #ifndef IEEE80211_STA_ONLY
1344 		if (intr & AR_ISR_SWBA)
1345 			ar9003_swba_intr(sc);
1346 #endif
1347 		if (intr & (AR_ISR_RXMINTR | AR_ISR_RXINTM))
1348 			ar9003_rx_intr(sc, ATHN_QID_LP);
1349 		if (intr & (AR_ISR_LP_RXOK | AR_ISR_RXERR))
1350 			ar9003_rx_intr(sc, ATHN_QID_LP);
1351 		if (intr & AR_ISR_HP_RXOK)
1352 			ar9003_rx_intr(sc, ATHN_QID_HP);
1353 
1354 		if (intr & (AR_ISR_TXMINTR | AR_ISR_TXINTM))
1355 			ar9003_tx_intr(sc);
1356 		if (intr & (AR_ISR_TXOK | AR_ISR_TXERR | AR_ISR_TXEOL))
1357 			ar9003_tx_intr(sc);
1358 
1359 		if (intr & AR_ISR_GENTMR) {
1360 			intr5 = AR_READ(sc, AR_ISR_S5_S);
1361 			DPRINTF(("GENTMR trigger=%d thresh=%d\n",
1362 			    MS(intr5, AR_ISR_S5_GENTIMER_TRIG),
1363 			    MS(intr5, AR_ISR_S5_GENTIMER_THRESH)));
1364 		}
1365 	}
1366 	if (sync != 0) {
1367 		if (sync & AR_INTR_SYNC_RADM_CPL_TIMEOUT) {
1368 			AR_WRITE(sc, AR_RC, AR_RC_HOSTIF);
1369 			AR_WRITE(sc, AR_RC, 0);
1370 		}
1371 
1372 		if ((sc->flags & ATHN_FLAG_RFSILENT) &&
1373 		    (sync & AR_INTR_SYNC_GPIO_PIN(sc->rfsilent_pin))) {
1374 			struct ifnet *ifp = &sc->sc_ic.ic_if;
1375 
1376 			printf("%s: radio switch turned off\n",
1377 			    sc->sc_dev.dv_xname);
1378 			/* Turn the interface down. */
1379 			ifp->if_flags &= ~IFF_UP;
1380 			athn_stop(ifp, 1);
1381 			return (1);
1382 		}
1383 
1384 		AR_WRITE(sc, AR_INTR_SYNC_CAUSE, sync);
1385 		(void)AR_READ(sc, AR_INTR_SYNC_CAUSE);
1386 	}
1387 	return (1);
1388 }
1389 
1390 int
1391 ar9003_tx(struct athn_softc *sc, struct mbuf *m, struct ieee80211_node *ni,
1392     int txflags)
1393 {
1394 	struct ieee80211com *ic = &sc->sc_ic;
1395 	struct ieee80211_key *k = NULL;
1396 	struct ieee80211_frame *wh;
1397 	struct athn_series series[4];
1398 	struct ar_tx_desc *ds;
1399 	struct athn_txq *txq;
1400 	struct athn_tx_buf *bf;
1401 	struct athn_node *an = (void *)ni;
1402 	struct mbuf *m1;
1403 	uintptr_t entry;
1404 	uint32_t sum;
1405 	uint16_t qos = 0;
1406 	uint8_t txpower, type, encrtype, tid, ridx[4];
1407 	int i, error, totlen, hasqos, qid;
1408 
1409 	/* Grab a Tx buffer from our global free list. */
1410 	bf = SIMPLEQ_FIRST(&sc->txbufs);
1411 	KASSERT(bf != NULL);
1412 
1413 	/* Map 802.11 frame type to hardware frame type. */
1414 	wh = mtod(m, struct ieee80211_frame *);
1415 	if ((wh->i_fc[0] & IEEE80211_FC0_TYPE_MASK) ==
1416 	    IEEE80211_FC0_TYPE_MGT) {
1417 		/* NB: Beacons do not use ar9003_tx(). */
1418 		if ((wh->i_fc[0] & IEEE80211_FC0_SUBTYPE_MASK) ==
1419 		    IEEE80211_FC0_SUBTYPE_PROBE_RESP)
1420 			type = AR_FRAME_TYPE_PROBE_RESP;
1421 		else if ((wh->i_fc[0] & IEEE80211_FC0_SUBTYPE_MASK) ==
1422 		    IEEE80211_FC0_SUBTYPE_ATIM)
1423 			type = AR_FRAME_TYPE_ATIM;
1424 		else
1425 			type = AR_FRAME_TYPE_NORMAL;
1426 	} else if ((wh->i_fc[0] &
1427 	    (IEEE80211_FC0_TYPE_MASK | IEEE80211_FC0_SUBTYPE_MASK)) ==
1428 	    (IEEE80211_FC0_TYPE_CTL  | IEEE80211_FC0_SUBTYPE_PS_POLL)) {
1429 		type = AR_FRAME_TYPE_PSPOLL;
1430 	} else
1431 		type = AR_FRAME_TYPE_NORMAL;
1432 
1433 	if (wh->i_fc[1] & IEEE80211_FC1_PROTECTED) {
1434 		k = ieee80211_get_txkey(ic, wh, ni);
1435 		if ((m = ieee80211_encrypt(ic, m, k)) == NULL)
1436 			return (ENOBUFS);
1437 		wh = mtod(m, struct ieee80211_frame *);
1438 	}
1439 
1440 	/* XXX 2-byte padding for QoS and 4-addr headers. */
1441 
1442 	/* Select the HW Tx queue to use for this frame. */
1443 	if ((hasqos = ieee80211_has_qos(wh))) {
1444 		qos = ieee80211_get_qos(wh);
1445 		tid = qos & IEEE80211_QOS_TID;
1446 		qid = athn_ac2qid[ieee80211_up_to_ac(ic, tid)];
1447 	} else if (type == AR_FRAME_TYPE_PSPOLL) {
1448 		qid = ATHN_QID_PSPOLL;
1449 	} else if (txflags & ATHN_TXFLAG_CAB) {
1450 		qid = ATHN_QID_CAB;
1451 	} else
1452 		qid = ATHN_QID_AC_BE;
1453 	txq = &sc->txq[qid];
1454 
1455 	/* Select the transmit rates to use for this frame. */
1456 	if (IEEE80211_IS_MULTICAST(wh->i_addr1) ||
1457 	    (wh->i_fc[0] & IEEE80211_FC0_TYPE_MASK) !=
1458 	    IEEE80211_FC0_TYPE_DATA) {
1459 		/* Use lowest rate for all tries. */
1460 		ridx[0] = ridx[1] = ridx[2] = ridx[3] =
1461 		    (ic->ic_curmode == IEEE80211_MODE_11A) ?
1462 			ATHN_RIDX_OFDM6 : ATHN_RIDX_CCK1;
1463 	} else if (ic->ic_fixed_rate != -1) {
1464 		/* Use same fixed rate for all tries. */
1465 		ridx[0] = ridx[1] = ridx[2] = ridx[3] =
1466 		    sc->fixed_ridx;
1467 	} else {
1468 		int txrate = ni->ni_txrate;
1469 		/* Use fallback table of the node. */
1470 		for (i = 0; i < 4; i++) {
1471 			ridx[i] = an->ridx[txrate];
1472 			txrate = an->fallback[txrate];
1473 		}
1474 	}
1475 
1476 #if NBPFILTER > 0
1477 	if (__predict_false(sc->sc_drvbpf != NULL)) {
1478 		struct athn_tx_radiotap_header *tap = &sc->sc_txtap;
1479 		struct mbuf mb;
1480 
1481 		tap->wt_flags = 0;
1482 		/* Use initial transmit rate. */
1483 		tap->wt_rate = athn_rates[ridx[0]].rate;
1484 		tap->wt_chan_freq = htole16(ic->ic_bss->ni_chan->ic_freq);
1485 		tap->wt_chan_flags = htole16(ic->ic_bss->ni_chan->ic_flags);
1486 		tap->wt_hwqueue = qid;
1487 		if (ridx[0] != ATHN_RIDX_CCK1 &&
1488 		    (ic->ic_flags & IEEE80211_F_SHPREAMBLE))
1489 			tap->wt_flags |= IEEE80211_RADIOTAP_F_SHORTPRE;
1490 		mb.m_data = (caddr_t)tap;
1491 		mb.m_len = sc->sc_txtap_len;
1492 		mb.m_next = m;
1493 		mb.m_nextpkt = NULL;
1494 		mb.m_type = 0;
1495 		mb.m_flags = 0;
1496 		bpf_mtap(sc->sc_drvbpf, &mb, BPF_DIRECTION_OUT);
1497 	}
1498 #endif
1499 
1500 	/* DMA map mbuf. */
1501 	error = bus_dmamap_load_mbuf(sc->sc_dmat, bf->bf_map, m,
1502 	    BUS_DMA_NOWAIT | BUS_DMA_WRITE);
1503 	if (__predict_false(error != 0)) {
1504 		if (error != EFBIG) {
1505 			printf("%s: can't map mbuf (error %d)\n",
1506 			    sc->sc_dev.dv_xname, error);
1507 			m_freem(m);
1508 			return (error);
1509 		}
1510 		/*
1511 		 * DMA mapping requires too many DMA segments; linearize
1512 		 * mbuf in kernel virtual address space and retry.
1513 		 */
1514 		MGETHDR(m1, M_DONTWAIT, MT_DATA);
1515 		if (m1 == NULL) {
1516 			m_freem(m);
1517 			return (ENOBUFS);
1518 		}
1519 		if (m->m_pkthdr.len > MHLEN) {
1520 			MCLGET(m1, M_DONTWAIT);
1521 			if (!(m1->m_flags & M_EXT)) {
1522 				m_freem(m);
1523 				m_freem(m1);
1524 				return (ENOBUFS);
1525 			}
1526 		}
1527 		m_copydata(m, 0, m->m_pkthdr.len, mtod(m1, caddr_t));
1528 		m1->m_pkthdr.len = m1->m_len = m->m_pkthdr.len;
1529 		m_freem(m);
1530 		m = m1;
1531 
1532 		error = bus_dmamap_load_mbuf(sc->sc_dmat, bf->bf_map, m,
1533 		    BUS_DMA_NOWAIT | BUS_DMA_WRITE);
1534 		if (error != 0) {
1535 			printf("%s: can't map mbuf (error %d)\n",
1536 			    sc->sc_dev.dv_xname, error);
1537 			m_freem(m);
1538 			return (error);
1539 		}
1540 	}
1541 	bf->bf_m = m;
1542 	bf->bf_ni = ni;
1543 	bf->bf_txflags = txflags;
1544 
1545 	wh = mtod(m, struct ieee80211_frame *);
1546 
1547 	totlen = m->m_pkthdr.len + IEEE80211_CRC_LEN;
1548 
1549 	/* Setup Tx descriptor. */
1550 	ds = bf->bf_descs;
1551 	memset(ds, 0, sizeof(*ds));
1552 
1553 	ds->ds_info =
1554 	    SM(AR_TXI_DESC_ID, AR_VENDOR_ATHEROS) |
1555 	    SM(AR_TXI_DESC_NDWORDS, 23) |
1556 	    SM(AR_TXI_QCU_NUM, qid) |
1557 	    AR_TXI_DESC_TX | AR_TXI_CTRL_STAT;
1558 
1559 	ds->ds_ctl11 = AR_TXC11_CLR_DEST_MASK;
1560 	txpower = AR_MAX_RATE_POWER;	/* Get from per-rate registers. */
1561 	ds->ds_ctl11 |= SM(AR_TXC11_XMIT_POWER, txpower);
1562 
1563 	ds->ds_ctl12 = SM(AR_TXC12_FRAME_TYPE, type);
1564 
1565 	if (IEEE80211_IS_MULTICAST(wh->i_addr1) ||
1566 	    (hasqos && (qos & IEEE80211_QOS_ACK_POLICY_MASK) ==
1567 	     IEEE80211_QOS_ACK_POLICY_NOACK))
1568 		ds->ds_ctl12 |= AR_TXC12_NO_ACK;
1569 
1570 	if (0 && k != NULL) {
1571 		/*
1572 		 * Map 802.11 cipher to hardware encryption type and
1573 		 * compute MIC+ICV overhead.
1574 		 */
1575 		switch (k->k_cipher) {
1576 		case IEEE80211_CIPHER_WEP40:
1577 		case IEEE80211_CIPHER_WEP104:
1578 			encrtype = AR_ENCR_TYPE_WEP;
1579 			totlen += 4;
1580 			break;
1581 		case IEEE80211_CIPHER_TKIP:
1582 			encrtype = AR_ENCR_TYPE_TKIP;
1583 			totlen += 12;
1584 			break;
1585 		case IEEE80211_CIPHER_CCMP:
1586 			encrtype = AR_ENCR_TYPE_AES;
1587 			totlen += 8;
1588 			break;
1589 		default:
1590 			panic("unsupported cipher");
1591 		}
1592 		/*
1593 		 * NB: The key cache entry index is stored in the key
1594 		 * private field when the key is installed.
1595 		 */
1596 		entry = (uintptr_t)k->k_priv;
1597 		ds->ds_ctl12 |= SM(AR_TXC12_DEST_IDX, entry);
1598 		ds->ds_ctl11 |= AR_TXC11_DEST_IDX_VALID;
1599 	} else
1600 		encrtype = AR_ENCR_TYPE_CLEAR;
1601 	ds->ds_ctl17 = SM(AR_TXC17_ENCR_TYPE, encrtype);
1602 
1603 	/* Check if frame must be protected using RTS/CTS or CTS-to-self. */
1604 	if (!IEEE80211_IS_MULTICAST(wh->i_addr1)) {
1605 		/* NB: Group frames are sent using CCK in 802.11b/g. */
1606 		if (totlen > ic->ic_rtsthreshold) {
1607 			ds->ds_ctl11 |= AR_TXC11_RTS_ENABLE;
1608 		} else if ((ic->ic_flags & IEEE80211_F_USEPROT) &&
1609 		    athn_rates[ridx[0]].phy == IEEE80211_T_OFDM) {
1610 			if (ic->ic_protmode == IEEE80211_PROT_RTSCTS)
1611 				ds->ds_ctl11 |= AR_TXC11_RTS_ENABLE;
1612 			else if (ic->ic_protmode == IEEE80211_PROT_CTSONLY)
1613 				ds->ds_ctl11 |= AR_TXC11_CTS_ENABLE;
1614 		}
1615 	}
1616 	if (ds->ds_ctl11 & (AR_TXC11_RTS_ENABLE | AR_TXC11_CTS_ENABLE)) {
1617 		/* Disable multi-rate retries when protection is used. */
1618 		ridx[1] = ridx[2] = ridx[3] = ridx[0];
1619 	}
1620 	/* Setup multi-rate retries. */
1621 	for (i = 0; i < 4; i++) {
1622 		series[i].hwrate = athn_rates[ridx[i]].hwrate;
1623 		if (athn_rates[ridx[i]].phy == IEEE80211_T_DS &&
1624 		    ridx[i] != ATHN_RIDX_CCK1 &&
1625 		    (ic->ic_flags & IEEE80211_F_SHPREAMBLE))
1626 			series[i].hwrate |= 0x04;
1627 		series[i].dur = 0;
1628 	}
1629 	if (!(ds->ds_ctl12 & AR_TXC12_NO_ACK)) {
1630 		/* Compute duration for each series. */
1631 		for (i = 0; i < 4; i++) {
1632 			series[i].dur = athn_txtime(sc, IEEE80211_ACK_LEN,
1633 			    athn_rates[ridx[i]].rspridx, ic->ic_flags);
1634 		}
1635 	}
1636 	/* If this is a PA training frame, select the Tx chain to use. */
1637 	if (__predict_false(txflags & ATHN_TXFLAG_PAPRD)) {
1638 		ds->ds_ctl12 |= SM(AR_TXC12_PAPRD_CHAIN_MASK,
1639 		    1 << sc->paprd_curchain);
1640 	}
1641 
1642 	/* Write number of tries for each series. */
1643 	ds->ds_ctl13 =
1644 	    SM(AR_TXC13_XMIT_DATA_TRIES0, 2) |
1645 	    SM(AR_TXC13_XMIT_DATA_TRIES1, 2) |
1646 	    SM(AR_TXC13_XMIT_DATA_TRIES2, 2) |
1647 	    SM(AR_TXC13_XMIT_DATA_TRIES3, 4);
1648 
1649 	/* Tell HW to update duration field in 802.11 header. */
1650 	if (type != AR_FRAME_TYPE_PSPOLL)
1651 		ds->ds_ctl13 |= AR_TXC13_DUR_UPDATE_ENA;
1652 
1653 	/* Write Tx rate for each series. */
1654 	ds->ds_ctl14 =
1655 	    SM(AR_TXC14_XMIT_RATE0, series[0].hwrate) |
1656 	    SM(AR_TXC14_XMIT_RATE1, series[1].hwrate) |
1657 	    SM(AR_TXC14_XMIT_RATE2, series[2].hwrate) |
1658 	    SM(AR_TXC14_XMIT_RATE3, series[3].hwrate);
1659 
1660 	/* Write duration for each series. */
1661 	ds->ds_ctl15 =
1662 	    SM(AR_TXC15_PACKET_DUR0, series[0].dur) |
1663 	    SM(AR_TXC15_PACKET_DUR1, series[1].dur);
1664 	ds->ds_ctl16 =
1665 	    SM(AR_TXC16_PACKET_DUR2, series[2].dur) |
1666 	    SM(AR_TXC16_PACKET_DUR3, series[3].dur);
1667 
1668 	if ((sc->flags & ATHN_FLAG_3TREDUCE_CHAIN) &&
1669 	    ic->ic_curmode == IEEE80211_MODE_11A) {
1670 		/*
1671 		 * In order to not exceed PCIe power requirements, we only
1672 		 * use two Tx chains for MCS0~15 on 5GHz band on these chips.
1673 		 */
1674 		ds->ds_ctl18 =
1675 		    SM(AR_TXC18_CHAIN_SEL0,
1676 			(ridx[0] <= ATHN_RIDX_MCS15) ? 0x3 : sc->txchainmask) |
1677 		    SM(AR_TXC18_CHAIN_SEL1,
1678 			(ridx[1] <= ATHN_RIDX_MCS15) ? 0x3 : sc->txchainmask) |
1679 		    SM(AR_TXC18_CHAIN_SEL2,
1680 			(ridx[2] <= ATHN_RIDX_MCS15) ? 0x3 : sc->txchainmask) |
1681 		    SM(AR_TXC18_CHAIN_SEL3,
1682 			(ridx[3] <= ATHN_RIDX_MCS15) ? 0x3 : sc->txchainmask);
1683 	} else {
1684 		/* Use the same Tx chains for all tries. */
1685 		ds->ds_ctl18 =
1686 		    SM(AR_TXC18_CHAIN_SEL0, sc->txchainmask) |
1687 		    SM(AR_TXC18_CHAIN_SEL1, sc->txchainmask) |
1688 		    SM(AR_TXC18_CHAIN_SEL2, sc->txchainmask) |
1689 		    SM(AR_TXC18_CHAIN_SEL3, sc->txchainmask);
1690 	}
1691 #ifdef notyet
1692 	/* Use the same short GI setting for all tries. */
1693 	if (ic->ic_flags & IEEE80211_F_SHGI)
1694 		ds->ds_ctl18 |= AR_TXC18_GI0123;
1695 	/* Use the same channel width for all tries. */
1696 	if (ic->ic_flags & IEEE80211_F_CBW40)
1697 		ds->ds_ctl18 |= AR_TXC18_2040_0123;
1698 #endif
1699 
1700 	if (ds->ds_ctl11 & (AR_TXC11_RTS_ENABLE | AR_TXC11_CTS_ENABLE)) {
1701 		uint8_t protridx, hwrate;
1702 		uint16_t dur = 0;
1703 
1704 		/* Use the same protection mode for all tries. */
1705 		if (ds->ds_ctl11 & AR_TXC11_RTS_ENABLE) {
1706 			ds->ds_ctl15 |= AR_TXC15_RTSCTS_QUAL01;
1707 			ds->ds_ctl16 |= AR_TXC16_RTSCTS_QUAL23;
1708 		}
1709 		/* Select protection rate (suboptimal but ok). */
1710 		protridx = (ic->ic_curmode == IEEE80211_MODE_11A) ?
1711 		    ATHN_RIDX_OFDM6 : ATHN_RIDX_CCK2;
1712 		if (ds->ds_ctl11 & AR_TXC11_RTS_ENABLE) {
1713 			/* Account for CTS duration. */
1714 			dur += athn_txtime(sc, IEEE80211_ACK_LEN,
1715 			    athn_rates[protridx].rspridx, ic->ic_flags);
1716 		}
1717 		dur += athn_txtime(sc, totlen, ridx[0], ic->ic_flags);
1718 		if (!(ds->ds_ctl12 & AR_TXC12_NO_ACK)) {
1719 			/* Account for ACK duration. */
1720 			dur += athn_txtime(sc, IEEE80211_ACK_LEN,
1721 			    athn_rates[ridx[0]].rspridx, ic->ic_flags);
1722 		}
1723 		/* Write protection frame duration and rate. */
1724 		ds->ds_ctl13 |= SM(AR_TXC13_BURST_DUR, dur);
1725 		hwrate = athn_rates[protridx].hwrate;
1726 		if (protridx == ATHN_RIDX_CCK2 &&
1727 		    (ic->ic_flags & IEEE80211_F_SHPREAMBLE))
1728 			hwrate |= 0x04;
1729 		ds->ds_ctl18 |= SM(AR_TXC18_RTSCTS_RATE, hwrate);
1730 	}
1731 
1732 	ds->ds_ctl11 |= SM(AR_TXC11_FRAME_LEN, totlen);
1733 	ds->ds_ctl19 = AR_TXC19_NOT_SOUNDING;
1734 
1735 	for (i = 0; i < bf->bf_map->dm_nsegs; i++) {
1736 		ds->ds_segs[i].ds_data = bf->bf_map->dm_segs[i].ds_addr;
1737 		ds->ds_segs[i].ds_ctl = SM(AR_TXC_BUF_LEN,
1738 		    bf->bf_map->dm_segs[i].ds_len);
1739 	}
1740 	/* Compute Tx descriptor checksum. */
1741 	sum = ds->ds_info + ds->ds_link;
1742 	for (i = 0; i < 4; i++) {
1743 		sum += ds->ds_segs[i].ds_data;
1744 		sum += ds->ds_segs[i].ds_ctl;
1745 	}
1746 	sum = (sum >> 16) + (sum & 0xffff);
1747 	ds->ds_ctl10 = SM(AR_TXC10_PTR_CHK_SUM, sum);
1748 
1749 	bus_dmamap_sync(sc->sc_dmat, bf->bf_map, 0, bf->bf_map->dm_mapsize,
1750 	    BUS_DMASYNC_PREWRITE);
1751 
1752 	DPRINTFN(6, ("Tx qid=%d nsegs=%d ctl11=0x%x ctl12=0x%x ctl14=0x%x\n",
1753 	    qid, bf->bf_map->dm_nsegs, ds->ds_ctl11, ds->ds_ctl12,
1754 	    ds->ds_ctl14));
1755 
1756 	SIMPLEQ_REMOVE_HEAD(&sc->txbufs, bf_list);
1757 	SIMPLEQ_INSERT_TAIL(&txq->head, bf, bf_list);
1758 
1759 	/* Queue buffer unless hardware FIFO is already full. */
1760 	if (++txq->queued <= AR9003_TX_QDEPTH) {
1761 		AR_WRITE(sc, AR_QTXDP(qid), bf->bf_daddr);
1762 		AR_WRITE_BARRIER(sc);
1763 	} else if (txq->wait == NULL)
1764 		txq->wait = bf;
1765 	return (0);
1766 }
1767 
1768 void
1769 ar9003_set_rf_mode(struct athn_softc *sc, struct ieee80211_channel *c)
1770 {
1771 	uint32_t reg;
1772 
1773 	reg = IEEE80211_IS_CHAN_2GHZ(c) ?
1774 	    AR_PHY_MODE_DYNAMIC : AR_PHY_MODE_OFDM;
1775 	if (IEEE80211_IS_CHAN_5GHZ(c) &&
1776 	    (sc->flags & ATHN_FLAG_FAST_PLL_CLOCK)) {
1777 		reg |= AR_PHY_MODE_DYNAMIC | AR_PHY_MODE_DYN_CCK_DISABLE;
1778 	}
1779 	AR_WRITE(sc, AR_PHY_MODE, reg);
1780 	AR_WRITE_BARRIER(sc);
1781 }
1782 
1783 static __inline uint32_t
1784 ar9003_synth_delay(struct athn_softc *sc)
1785 {
1786 	uint32_t delay;
1787 
1788 	delay = MS(AR_READ(sc, AR_PHY_RX_DELAY), AR_PHY_RX_DELAY_DELAY);
1789 	if (sc->sc_ic.ic_curmode == IEEE80211_MODE_11B)
1790 		delay = (delay * 4) / 22;
1791 	else
1792 		delay = delay / 10;	/* in 100ns steps */
1793 	return (delay);
1794 }
1795 
1796 int
1797 ar9003_rf_bus_request(struct athn_softc *sc)
1798 {
1799 	int ntries;
1800 
1801 	/* Request RF Bus grant. */
1802 	AR_WRITE(sc, AR_PHY_RFBUS_REQ, AR_PHY_RFBUS_REQ_EN);
1803 	for (ntries = 0; ntries < 10000; ntries++) {
1804 		if (AR_READ(sc, AR_PHY_RFBUS_GRANT) & AR_PHY_RFBUS_GRANT_EN)
1805 			return (0);
1806 		DELAY(10);
1807 	}
1808 	DPRINTF(("could not kill baseband Rx"));
1809 	return (ETIMEDOUT);
1810 }
1811 
1812 void
1813 ar9003_rf_bus_release(struct athn_softc *sc)
1814 {
1815 	/* Wait for the synthesizer to settle. */
1816 	DELAY(AR_BASE_PHY_ACTIVE_DELAY + ar9003_synth_delay(sc));
1817 
1818 	/* Release the RF Bus grant. */
1819 	AR_WRITE(sc, AR_PHY_RFBUS_REQ, 0);
1820 	AR_WRITE_BARRIER(sc);
1821 }
1822 
1823 void
1824 ar9003_set_phy(struct athn_softc *sc, struct ieee80211_channel *c,
1825     struct ieee80211_channel *extc)
1826 {
1827 	uint32_t phy;
1828 
1829 	phy = AR_READ(sc, AR_PHY_GEN_CTRL);
1830 	phy |= AR_PHY_GC_HT_EN | AR_PHY_GC_SHORT_GI_40 |
1831 	    AR_PHY_GC_SINGLE_HT_LTF1 | AR_PHY_GC_WALSH;
1832 	if (extc != NULL) {
1833 		phy |= AR_PHY_GC_DYN2040_EN;
1834 		if (extc > c)	/* XXX */
1835 			phy |= AR_PHY_GC_DYN2040_PRI_CH;
1836 	}
1837 	/* Turn off Green Field detection for now. */
1838 	phy &= ~AR_PHY_GC_GF_DETECT_EN;
1839 	AR_WRITE(sc, AR_PHY_GEN_CTRL, phy);
1840 
1841 	AR_WRITE(sc, AR_2040_MODE,
1842 	    (extc != NULL) ? AR_2040_JOINED_RX_CLEAR : 0);
1843 
1844 	/* Set global transmit timeout. */
1845 	AR_WRITE(sc, AR_GTXTO, SM(AR_GTXTO_TIMEOUT_LIMIT, 25));
1846 	/* Set carrier sense timeout. */
1847 	AR_WRITE(sc, AR_CST, SM(AR_CST_TIMEOUT_LIMIT, 15));
1848 	AR_WRITE_BARRIER(sc);
1849 }
1850 
1851 void
1852 ar9003_set_delta_slope(struct athn_softc *sc, struct ieee80211_channel *c,
1853     struct ieee80211_channel *extc)
1854 {
1855 	uint32_t coeff, exp, man, reg;
1856 
1857 	/* Set Delta Slope (exponent and mantissa). */
1858 	coeff = (100 << 24) / c->ic_freq;
1859 	athn_get_delta_slope(coeff, &exp, &man);
1860 	DPRINTFN(5, ("delta slope coeff exp=%u man=%u\n", exp, man));
1861 
1862 	reg = AR_READ(sc, AR_PHY_TIMING3);
1863 	reg = RW(reg, AR_PHY_TIMING3_DSC_EXP, exp);
1864 	reg = RW(reg, AR_PHY_TIMING3_DSC_MAN, man);
1865 	AR_WRITE(sc, AR_PHY_TIMING3, reg);
1866 
1867 	/* For Short GI, coeff is 9/10 that of normal coeff. */
1868 	coeff = (9 * coeff) / 10;
1869 	athn_get_delta_slope(coeff, &exp, &man);
1870 	DPRINTFN(5, ("delta slope coeff exp=%u man=%u\n", exp, man));
1871 
1872 	reg = AR_READ(sc, AR_PHY_SGI_DELTA);
1873 	reg = RW(reg, AR_PHY_SGI_DSC_EXP, exp);
1874 	reg = RW(reg, AR_PHY_SGI_DSC_MAN, man);
1875 	AR_WRITE(sc, AR_PHY_SGI_DELTA, reg);
1876 	AR_WRITE_BARRIER(sc);
1877 }
1878 
1879 void
1880 ar9003_enable_antenna_diversity(struct athn_softc *sc)
1881 {
1882 	AR_SETBITS(sc, AR_PHY_CCK_DETECT,
1883 	    AR_PHY_CCK_DETECT_BB_ENABLE_ANT_FAST_DIV);
1884 	AR_WRITE_BARRIER(sc);
1885 }
1886 
1887 void
1888 ar9003_init_baseband(struct athn_softc *sc)
1889 {
1890 	uint32_t synth_delay;
1891 
1892 	synth_delay = ar9003_synth_delay(sc);
1893 	/* Activate the PHY (includes baseband activate and synthesizer on). */
1894 	AR_WRITE(sc, AR_PHY_ACTIVE, AR_PHY_ACTIVE_EN);
1895 	AR_WRITE_BARRIER(sc);
1896 	DELAY(AR_BASE_PHY_ACTIVE_DELAY + synth_delay);
1897 }
1898 
1899 void
1900 ar9003_disable_phy(struct athn_softc *sc)
1901 {
1902 	AR_WRITE(sc, AR_PHY_ACTIVE, AR_PHY_ACTIVE_DIS);
1903 	AR_WRITE_BARRIER(sc);
1904 }
1905 
1906 void
1907 ar9003_init_chains(struct athn_softc *sc)
1908 {
1909 	if (sc->rxchainmask == 0x5 || sc->txchainmask == 0x5)
1910 		AR_SETBITS(sc, AR_PHY_ANALOG_SWAP, AR_PHY_SWAP_ALT_CHAIN);
1911 
1912 	/* Setup chain masks. */
1913 	AR_WRITE(sc, AR_PHY_RX_CHAINMASK,  sc->rxchainmask);
1914 	AR_WRITE(sc, AR_PHY_CAL_CHAINMASK, sc->rxchainmask);
1915 
1916 	if (sc->flags & ATHN_FLAG_3TREDUCE_CHAIN) {
1917 		/*
1918 		 * All self-generated frames are sent using two Tx chains
1919 		 * on these chips to not exceed PCIe power requirements.
1920 		 */
1921 		AR_WRITE(sc, AR_SELFGEN_MASK, 0x3);
1922 	} else
1923 		AR_WRITE(sc, AR_SELFGEN_MASK, sc->txchainmask);
1924 	AR_WRITE_BARRIER(sc);
1925 }
1926 
1927 void
1928 ar9003_set_rxchains(struct athn_softc *sc)
1929 {
1930 	if (sc->rxchainmask == 0x3 || sc->rxchainmask == 0x5) {
1931 		AR_WRITE(sc, AR_PHY_RX_CHAINMASK,  sc->rxchainmask);
1932 		AR_WRITE(sc, AR_PHY_CAL_CHAINMASK, sc->rxchainmask);
1933 		AR_WRITE_BARRIER(sc);
1934 	}
1935 }
1936 
1937 void
1938 ar9003_read_noisefloor(struct athn_softc *sc, int16_t *nf, int16_t *nf_ext)
1939 {
1940 /* Sign-extends 9-bit value (assumes upper bits are zeroes). */
1941 #define SIGN_EXT(v)	(((v) ^ 0x100) - 0x100)
1942 	uint32_t reg;
1943 	int i;
1944 
1945 	for (i = 0; i < sc->nrxchains; i++) {
1946 		reg = AR_READ(sc, AR_PHY_CCA(i));
1947 		nf[i] = MS(reg, AR_PHY_MINCCA_PWR);
1948 		nf[i] = SIGN_EXT(nf[i]);
1949 
1950 		reg = AR_READ(sc, AR_PHY_EXT_CCA(i));
1951 		nf_ext[i] = MS(reg, AR_PHY_EXT_MINCCA_PWR);
1952 		nf_ext[i] = SIGN_EXT(nf_ext[i]);
1953 	}
1954 #undef SIGN_EXT
1955 }
1956 
1957 void
1958 ar9003_write_noisefloor(struct athn_softc *sc, int16_t *nf, int16_t *nf_ext)
1959 {
1960 	uint32_t reg;
1961 	int i;
1962 
1963 	for (i = 0; i < sc->nrxchains; i++) {
1964 		reg = AR_READ(sc, AR_PHY_CCA(i));
1965 		reg = RW(reg, AR_PHY_MAXCCA_PWR, nf[i]);
1966 		AR_WRITE(sc, AR_PHY_CCA(i), reg);
1967 
1968 		reg = AR_READ(sc, AR_PHY_EXT_CCA(i));
1969 		reg = RW(reg, AR_PHY_EXT_MAXCCA_PWR, nf_ext[i]);
1970 		AR_WRITE(sc, AR_PHY_EXT_CCA(i), reg);
1971 	}
1972 	AR_WRITE_BARRIER(sc);
1973 }
1974 
1975 void
1976 ar9003_get_noisefloor(struct athn_softc *sc, struct ieee80211_channel *c)
1977 {
1978 	int16_t nf[AR_MAX_CHAINS], nf_ext[AR_MAX_CHAINS];
1979 	int16_t min, max;
1980 	int i;
1981 
1982 	if (AR_READ(sc, AR_PHY_AGC_CONTROL) & AR_PHY_AGC_CONTROL_NF) {
1983 		/* Noisefloor calibration not finished. */
1984 		return;
1985 	}
1986 	/* Noisefloor calibration is finished. */
1987 	ar9003_read_noisefloor(sc, nf, nf_ext);
1988 
1989 	if (IEEE80211_IS_CHAN_2GHZ(c)) {
1990 		min = sc->cca_min_2g;
1991 		max = sc->cca_max_2g;
1992 	} else {
1993 		min = sc->cca_min_5g;
1994 		max = sc->cca_max_5g;
1995 	}
1996 	/* Update noisefloor history. */
1997 	for (i = 0; i < sc->nrxchains; i++) {
1998 		if (nf[i] < min)
1999 			nf[i] = min;
2000 		else if (nf[i] > max)
2001 			nf[i] = max;
2002 		if (nf_ext[i] < min)
2003 			nf_ext[i] = min;
2004 		else if (nf_ext[i] > max)
2005 			nf_ext[i] = max;
2006 
2007 		sc->nf_hist[sc->nf_hist_cur].nf[i] = nf[i];
2008 		sc->nf_hist[sc->nf_hist_cur].nf_ext[i] = nf_ext[i];
2009 	}
2010 	if (++sc->nf_hist_cur >= ATHN_NF_CAL_HIST_MAX)
2011 		sc->nf_hist_cur = 0;
2012 }
2013 
2014 void
2015 ar9003_bb_load_noisefloor(struct athn_softc *sc)
2016 {
2017 	int16_t nf[AR_MAX_CHAINS], nf_ext[AR_MAX_CHAINS];
2018 	int i, ntries;
2019 
2020 	/* Write filtered noisefloor values. */
2021 	for (i = 0; i < sc->nrxchains; i++) {
2022 		nf[i] = sc->nf_priv[i] * 2;
2023 		nf_ext[i] = sc->nf_ext_priv[i] * 2;
2024 	}
2025 	ar9003_write_noisefloor(sc, nf, nf_ext);
2026 
2027 	/* Load filtered noisefloor values into baseband. */
2028 	AR_CLRBITS(sc, AR_PHY_AGC_CONTROL, AR_PHY_AGC_CONTROL_ENABLE_NF);
2029 	AR_CLRBITS(sc, AR_PHY_AGC_CONTROL, AR_PHY_AGC_CONTROL_NO_UPDATE_NF);
2030 	AR_SETBITS(sc, AR_PHY_AGC_CONTROL, AR_PHY_AGC_CONTROL_NF);
2031 	/* Wait for load to complete. */
2032 	for (ntries = 0; ntries < 1000; ntries++) {
2033 		if (!(AR_READ(sc, AR_PHY_AGC_CONTROL) & AR_PHY_AGC_CONTROL_NF))
2034 			break;
2035 		DELAY(10);
2036 	}
2037 	if (ntries == 1000) {
2038 		DPRINTF(("failed to load noisefloor values\n"));
2039 		return;
2040 	}
2041 
2042 	/* Restore noisefloor values to initial (max) values. */
2043 	for (i = 0; i < AR_MAX_CHAINS; i++)
2044 		nf[i] = nf_ext[i] = -50 * 2;
2045 	ar9003_write_noisefloor(sc, nf, nf_ext);
2046 }
2047 
2048 void
2049 ar9300_noisefloor_calib(struct athn_softc *sc)
2050 {
2051 	AR_SETBITS(sc, AR_PHY_AGC_CONTROL, AR_PHY_AGC_CONTROL_ENABLE_NF);
2052 	AR_SETBITS(sc, AR_PHY_AGC_CONTROL, AR_PHY_AGC_CONTROL_NO_UPDATE_NF);
2053 	AR_SETBITS(sc, AR_PHY_AGC_CONTROL, AR_PHY_AGC_CONTROL_NF);
2054 }
2055 
2056 void
2057 ar9003_do_noisefloor_calib(struct athn_softc *sc)
2058 {
2059 	AR_SETBITS(sc, AR_PHY_AGC_CONTROL, AR_PHY_AGC_CONTROL_NF);
2060 }
2061 
2062 int
2063 ar9003_init_calib(struct athn_softc *sc)
2064 {
2065 	uint8_t txchainmask, rxchainmask;
2066 	uint32_t reg;
2067 	int ntries;
2068 
2069 	/* Save chains masks. */
2070 	txchainmask = sc->txchainmask;
2071 	rxchainmask = sc->rxchainmask;
2072 	/* Configure hardware before calibration. */
2073 	if (AR_READ(sc, AR_ENT_OTP) & AR_ENT_OTP_CHAIN2_DISABLE)
2074 		txchainmask = rxchainmask = 0x3;
2075 	else
2076 		txchainmask = rxchainmask = 0x7;
2077 	ar9003_init_chains(sc);
2078 
2079 	/* Perform Tx IQ calibration. */
2080 	ar9003_calib_tx_iq(sc);
2081 	/* Disable and re-enable the PHY chips. */
2082 	AR_WRITE(sc, AR_PHY_ACTIVE, AR_PHY_ACTIVE_DIS);
2083 	AR_WRITE_BARRIER(sc);
2084 	DELAY(5);
2085 	AR_WRITE(sc, AR_PHY_ACTIVE, AR_PHY_ACTIVE_EN);
2086 
2087 	/* Calibrate the AGC. */
2088 	AR_SETBITS(sc, AR_PHY_AGC_CONTROL, AR_PHY_AGC_CONTROL_CAL);
2089 	/* Poll for offset calibration completion. */
2090 	for (ntries = 0; ntries < 10000; ntries++) {
2091 		reg = AR_READ(sc, AR_PHY_AGC_CONTROL);
2092 		if (!(reg & AR_PHY_AGC_CONTROL_CAL))
2093 			break;
2094 		DELAY(10);
2095 	}
2096 	if (ntries == 10000)
2097 		return (ETIMEDOUT);
2098 
2099 	/* Restore chains masks. */
2100 	sc->txchainmask = txchainmask;
2101 	sc->rxchainmask = rxchainmask;
2102 	ar9003_init_chains(sc);
2103 
2104 	return (0);
2105 }
2106 
2107 void
2108 ar9003_do_calib(struct athn_softc *sc)
2109 {
2110 	uint32_t reg;
2111 
2112 	if (sc->cur_calib_mask & ATHN_CAL_IQ) {
2113 		reg = AR_READ(sc, AR_PHY_TIMING4);
2114 		reg = RW(reg, AR_PHY_TIMING4_IQCAL_LOG_COUNT_MAX, 10);
2115 		AR_WRITE(sc, AR_PHY_TIMING4, reg);
2116 		AR_WRITE(sc, AR_PHY_CALMODE, AR_PHY_CALMODE_IQ);
2117 		AR_SETBITS(sc, AR_PHY_TIMING4, AR_PHY_TIMING4_DO_CAL);
2118 		AR_WRITE_BARRIER(sc);
2119 	} else if (sc->cur_calib_mask & ATHN_CAL_TEMP) {
2120 		AR_SETBITS(sc, AR_PHY_65NM_CH0_THERM,
2121 		    AR_PHY_65NM_CH0_THERM_LOCAL);
2122 		AR_SETBITS(sc, AR_PHY_65NM_CH0_THERM,
2123 		    AR_PHY_65NM_CH0_THERM_START);
2124 		AR_WRITE_BARRIER(sc);
2125 	}
2126 }
2127 
2128 void
2129 ar9003_next_calib(struct athn_softc *sc)
2130 {
2131 	/* Check if we have any calibration in progress. */
2132 	if (sc->cur_calib_mask != 0) {
2133 		if (!(AR_READ(sc, AR_PHY_TIMING4) & AR_PHY_TIMING4_DO_CAL)) {
2134 			/* Calibration completed for current sample. */
2135 			ar9003_calib_iq(sc);
2136 		}
2137 	}
2138 }
2139 
2140 void
2141 ar9003_calib_iq(struct athn_softc *sc)
2142 {
2143 	struct athn_iq_cal *cal;
2144 	uint32_t reg, i_coff_denom, q_coff_denom;
2145 	int32_t i_coff, q_coff;
2146 	int i, iq_corr_neg;
2147 
2148 	for (i = 0; i < AR_MAX_CHAINS; i++) {
2149 		cal = &sc->calib.iq[i];
2150 
2151 		/* Read IQ calibration measures (clear on read). */
2152 		cal->pwr_meas_i = AR_READ(sc, AR_PHY_IQ_ADC_MEAS_0_B(i));
2153 		cal->pwr_meas_q = AR_READ(sc, AR_PHY_IQ_ADC_MEAS_1_B(i));
2154 		cal->iq_corr_meas =
2155 		    (int32_t)AR_READ(sc, AR_PHY_IQ_ADC_MEAS_2_B(i));
2156 	}
2157 
2158 	for (i = 0; i < sc->nrxchains; i++) {
2159 		cal = &sc->calib.iq[i];
2160 
2161 		if (cal->pwr_meas_q == 0)
2162 			continue;
2163 
2164 		if ((iq_corr_neg = cal->iq_corr_meas < 0))
2165 			cal->iq_corr_meas = -cal->iq_corr_meas;
2166 
2167 		i_coff_denom =
2168 		    (cal->pwr_meas_i / 2 + cal->pwr_meas_q / 2) / 256;
2169 		q_coff_denom = cal->pwr_meas_q / 64;
2170 
2171 		if (i_coff_denom == 0 || q_coff_denom == 0)
2172 			continue;	/* Prevents division by zero. */
2173 
2174 		i_coff = cal->iq_corr_meas / i_coff_denom;
2175 		q_coff = (cal->pwr_meas_i / q_coff_denom) - 64;
2176 
2177 		if (i_coff > 63)
2178 			i_coff = 63;
2179 		else if (i_coff < -63)
2180 			i_coff = -63;
2181 		/* Negate i_coff if iq_corr_meas is positive. */
2182 		if (!iq_corr_neg)
2183 			i_coff = -i_coff;
2184 		if (q_coff > 63)
2185 			q_coff = 63;
2186 		else if (q_coff < -63)
2187 			q_coff = -63;
2188 
2189 		DPRINTFN(2, ("IQ calibration for chain %d\n", i));
2190 		reg = AR_READ(sc, AR_PHY_RX_IQCAL_CORR_B(i));
2191 		reg = RW(reg, AR_PHY_RX_IQCAL_CORR_IQCORR_Q_I_COFF, i_coff);
2192 		reg = RW(reg, AR_PHY_RX_IQCAL_CORR_IQCORR_Q_Q_COFF, q_coff);
2193 		AR_WRITE(sc, AR_PHY_RX_IQCAL_CORR_B(i), reg);
2194 	}
2195 
2196 	/* Apply new settings. */
2197 	AR_SETBITS(sc, AR_PHY_RX_IQCAL_CORR_B(0),
2198 	    AR_PHY_RX_IQCAL_CORR_IQCORR_ENABLE);
2199 	AR_WRITE_BARRIER(sc);
2200 
2201 	/* IQ calibration done. */
2202 	sc->cur_calib_mask &= ~ATHN_CAL_IQ;
2203 	memset(&sc->calib, 0, sizeof(sc->calib));
2204 }
2205 
2206 #define DELPT	32
2207 int
2208 ar9003_get_iq_corr(struct athn_softc *sc, int32_t res[6], int32_t coeff[2])
2209 {
2210 /* Sign-extends 12-bit value (assumes upper bits are zeroes). */
2211 #define SIGN_EXT(v)	(((v) ^ 0x800) - 0x800)
2212 #define SCALE		(1 << 15)
2213 #define SHIFT		(1 <<  8)
2214 	struct {
2215 		int32_t	m, p, c;
2216 	} val[2][2];
2217 	int32_t mag[2][2], phs[2][2], cos[2], sin[2];
2218 	int32_t min, max, div, f1, f2, f3, m, p, c;
2219 	int32_t txmag, txphs, rxmag, rxphs;
2220 	int32_t q_coff, i_coff;
2221 	int i, j;
2222 
2223 	/* Extract our twelve signed 12-bit values from res[] array. */
2224 	val[0][0].m = res[0] & 0xfff;
2225 	val[0][0].p = (res[0] >> 12) & 0xfff;
2226 	val[0][0].c = ((res[0] >> 24) & 0xff) | (res[1] & 0xf) << 8;
2227 
2228 	val[0][1].m = (res[1] >> 4) & 0xfff;
2229 	val[0][1].p = res[2] & 0xfff;
2230 	val[0][1].c = (res[2] >> 12) & 0xfff;
2231 
2232 	val[1][0].m = ((res[2] >> 24) & 0xff) | (res[3] & 0xf) << 8;
2233 	val[1][0].p = (res[3] >> 4) & 0xfff;
2234 	val[1][0].c = res[4] & 0xfff;
2235 
2236 	val[1][1].m = (res[4] >> 12) & 0xfff;
2237 	val[1][1].p = ((res[4] >> 24) & 0xff) | (res[5] & 0xf) << 8;
2238 	val[1][1].c = (res[5] >> 4) & 0xfff;
2239 
2240 	for (i = 0; i < 2; i++) {
2241 		for (j = 0; j < 2; j++) {
2242 			m = SIGN_EXT(val[i][j].m);
2243 			p = SIGN_EXT(val[i][j].p);
2244 			c = SIGN_EXT(val[i][j].c);
2245 
2246 			if (p == 0)
2247 				return (1);	/* Prevent division by 0. */
2248 
2249 			mag[i][j] = (m * SCALE) / p;
2250 			phs[i][j] = (c * SCALE) / p;
2251 		}
2252 		sin[i] = ((mag[i][0] - mag[i][1]) * SHIFT) / DELPT;
2253 		cos[i] = ((phs[i][0] - phs[i][1]) * SHIFT) / DELPT;
2254 		/* Find magnitude by approximation. */
2255 		min = MIN(abs(sin[i]), abs(cos[i]));
2256 		max = MAX(abs(sin[i]), abs(cos[i]));
2257 		div = max - (max / 32) + (min / 8) + (min / 4);
2258 		if (div == 0)
2259 			return (1);	/* Prevent division by 0. */
2260 		/* Normalize sin and cos by magnitude. */
2261 		sin[i] = (sin[i] * SCALE) / div;
2262 		cos[i] = (cos[i] * SCALE) / div;
2263 	}
2264 
2265 	/* Compute IQ mismatch (solve 4x4 linear equation). */
2266 	f1 = cos[0] - cos[1];
2267 	f3 = sin[0] - sin[1];
2268 	f2 = (f1 * f1 + f3 * f3) / SCALE;
2269 	if (f2 == 0)
2270 		return (1);	/* Prevent division by 0. */
2271 
2272 	/* Compute Tx magnitude mismatch. */
2273 	txmag = (f1 * ( mag[0][0] - mag[1][0]) +
2274 		 f3 * ( phs[0][0] - phs[1][0])) / f2;
2275 	/* Compute Tx phase mismatch. */
2276 	txphs = (f3 * (-mag[0][0] + mag[1][0]) +
2277 		 f1 * ( phs[0][0] - phs[1][0])) / f2;
2278 
2279 	if (txmag == SCALE)
2280 		return (1);	/* Prevent division by 0. */
2281 
2282 	/* Compute Rx magnitude mismatch. */
2283 	rxmag = mag[0][0] - (cos[0] * txmag + sin[0] * txphs) / SCALE;
2284 	/* Compute Rx phase mismatch. */
2285 	rxphs = phs[0][0] + (sin[0] * txmag - cos[0] * txphs) / SCALE;
2286 
2287 	if (-rxmag == SCALE)
2288 		return (1);	/* Prevent division by 0. */
2289 
2290 	txmag = (txmag * SCALE) / (SCALE - txmag);
2291 	txphs = -txphs;
2292 
2293 	q_coff = (txmag * 128) / SCALE;
2294 	if (q_coff < -63)
2295 		q_coff = -63;
2296 	else if (q_coff > 63)
2297 		q_coff = 63;
2298 	i_coff = (txphs * 256) / SCALE;
2299 	if (i_coff < -63)
2300 		i_coff = -63;
2301 	else if (i_coff > 63)
2302 		i_coff = 63;
2303 	coeff[0] = q_coff * 128 + i_coff;
2304 
2305 	rxmag = (-rxmag * SCALE) / (SCALE + rxmag);
2306 	rxphs = -rxphs;
2307 
2308 	q_coff = (rxmag * 128) / SCALE;
2309 	if (q_coff < -63)
2310 		q_coff = -63;
2311 	else if (q_coff > 63)
2312 		q_coff = 63;
2313 	i_coff = (rxphs * 256) / SCALE;
2314 	if (i_coff < -63)
2315 		i_coff = -63;
2316 	else if (i_coff > 63)
2317 		i_coff = 63;
2318 	coeff[1] = q_coff * 128 + i_coff;
2319 
2320 	return (0);
2321 #undef SHIFT
2322 #undef SCALE
2323 #undef SIGN_EXT
2324 }
2325 
2326 int
2327 ar9003_calib_tx_iq(struct athn_softc *sc)
2328 {
2329 	uint32_t reg;
2330 	int32_t res[6], coeff[2];
2331 	int i, j, ntries;
2332 
2333 	reg = AR_READ(sc, AR_PHY_TX_IQCAL_CONTROL_1);
2334 	reg = RW(reg, AR_PHY_TX_IQCAQL_CONTROL_1_IQCORR_I_Q_COFF_DELPT, DELPT);
2335 	AR_WRITE(sc, AR_PHY_TX_IQCAL_CONTROL_1, reg);
2336 
2337 	/* Start Tx IQ calibration. */
2338 	AR_SETBITS(sc, AR_PHY_TX_IQCAL_START, AR_PHY_TX_IQCAL_START_DO_CAL);
2339 	/* Wait for completion. */
2340 	for (ntries = 0; ntries < 10000; ntries++) {
2341 		reg = AR_READ(sc, AR_PHY_TX_IQCAL_START);
2342 		if (!(reg & AR_PHY_TX_IQCAL_START_DO_CAL))
2343 			break;
2344 		DELAY(10);
2345 	}
2346 	if (ntries == 10000)
2347 		return (ETIMEDOUT);
2348 
2349 	for (i = 0; i < sc->ntxchains; i++) {
2350 		/* Read Tx IQ calibration status for this chain. */
2351 		reg = AR_READ(sc, AR_PHY_TX_IQCAL_STATUS_B(i));
2352 		if (reg & AR_PHY_TX_IQCAL_STATUS_FAILED)
2353 			return (EIO);
2354 		/*
2355 		 * Read Tx IQ calibration results for this chain.
2356 		 * This consists in twelve signed 12-bit values.
2357 		 */
2358 		for (j = 0; j < 3; j++) {
2359 			AR_CLRBITS(sc, AR_PHY_CHAN_INFO_MEMORY,
2360 			    AR_PHY_CHAN_INFO_TAB_S2_READ);
2361 			reg = AR_READ(sc, AR_PHY_CHAN_INFO_TAB(i, j));
2362 			res[j * 2 + 0] = reg;
2363 
2364 			AR_SETBITS(sc, AR_PHY_CHAN_INFO_MEMORY,
2365 			    AR_PHY_CHAN_INFO_TAB_S2_READ);
2366 			reg = AR_READ(sc, AR_PHY_CHAN_INFO_TAB(i, j));
2367 			res[j * 2 + 1] = reg & 0xffff;
2368 		}
2369 
2370 		/* Compute Tx IQ correction. */
2371 		if (ar9003_get_iq_corr(sc, res, coeff) != 0)
2372 			return (EIO);
2373 
2374 		/* Write Tx IQ correction coefficients. */
2375 		reg = AR_READ(sc, AR_PHY_TX_IQCAL_CORR_COEFF_01_B(i));
2376 		reg = RW(reg, AR_PHY_TX_IQCAL_CORR_COEFF_01_COEFF_TABLE,
2377 		    coeff[0]);
2378 		AR_WRITE(sc, AR_PHY_TX_IQCAL_CORR_COEFF_01_B(i), reg);
2379 
2380 		reg = AR_READ(sc, AR_PHY_RX_IQCAL_CORR_B(i));
2381 		reg = RW(reg, AR_PHY_RX_IQCAL_CORR_LOOPBACK_IQCORR_Q_Q_COFF,
2382 		    coeff[1] >> 7);
2383 		reg = RW(reg, AR_PHY_RX_IQCAL_CORR_LOOPBACK_IQCORR_Q_I_COFF,
2384 		    coeff[1]);
2385 		AR_WRITE(sc, AR_PHY_RX_IQCAL_CORR_B(i), reg);
2386 		AR_WRITE_BARRIER(sc);
2387 	}
2388 
2389 	/* Enable Tx IQ correction. */
2390 	AR_SETBITS(sc, AR_PHY_TX_IQCAL_CONTROL_3,
2391 	    AR_PHY_TX_IQCAL_CONTROL_3_IQCORR_EN);
2392 	AR_SETBITS(sc, AR_PHY_RX_IQCAL_CORR_B(0),
2393 	    AR_PHY_RX_IQCAL_CORR_B0_LOOPBACK_IQCORR_EN);
2394 	AR_WRITE_BARRIER(sc);
2395 	return (0);
2396 }
2397 #undef DELPT
2398 
2399 /*-
2400  * The power amplifier predistortion state machine works as follows:
2401  * 1) Disable digital predistorters for all Tx chains
2402  * 2) Repeat steps 3~7 for all Tx chains
2403  * 3)   Force Tx gain to that of training signal
2404  * 4)   Send training signal (asynchronous)
2405  * 5)   Wait for training signal to complete (asynchronous)
2406  * 6)   Read PA measurements (input power, output power, output phase)
2407  * 7)   Compute the predistortion function that linearizes PA output
2408  * 8) Write predistortion functions to hardware tables for all Tx chains
2409  * 9) Enable digital predistorters for all Tx chains
2410  */
2411 void
2412 ar9003_paprd_calib(struct athn_softc *sc, struct ieee80211_channel *c)
2413 {
2414 	static const int scaling[] = {
2415 		261376, 248079, 233759, 220464,
2416 		208194, 196949, 185706, 175487
2417 	};
2418 	struct athn_ops *ops = &sc->ops;
2419 	uint32_t reg, ht20mask, ht40mask;
2420 	int i;
2421 
2422 	/* Read PA predistortion masks from ROM. */
2423 	ops->get_paprd_masks(sc, c, &ht20mask, &ht40mask);
2424 
2425 	/* AM-to-AM: amplifier's amplitude characteristic. */
2426 	reg = AR_READ(sc, AR_PHY_PAPRD_AM2AM);
2427 	reg = RW(reg, AR_PHY_PAPRD_AM2AM_MASK, ht20mask);
2428 	AR_WRITE(sc, AR_PHY_PAPRD_AM2AM, reg);
2429 
2430 	/* AM-to-PM: amplifier's phase transfer characteristic. */
2431 	reg = AR_READ(sc, AR_PHY_PAPRD_AM2PM);
2432 	reg = RW(reg, AR_PHY_PAPRD_AM2PM_MASK, ht20mask);
2433 	AR_WRITE(sc, AR_PHY_PAPRD_AM2PM, reg);
2434 
2435 	reg = AR_READ(sc, AR_PHY_PAPRD_HT40);
2436 	reg = RW(reg, AR_PHY_PAPRD_HT40_MASK, ht40mask);
2437 	AR_WRITE(sc, AR_PHY_PAPRD_HT40, reg);
2438 
2439 	for (i = 0; i < AR9003_MAX_CHAINS; i++) {
2440 		AR_SETBITS(sc, AR_PHY_PAPRD_CTRL0_B(i),
2441 		    AR_PHY_PAPRD_CTRL0_USE_SINGLE_TABLE);
2442 
2443 		reg = AR_READ(sc, AR_PHY_PAPRD_CTRL1_B(i));
2444 		reg = RW(reg, AR_PHY_PAPRD_CTRL1_PA_GAIN_SCALE_FACT, 181);
2445 		reg = RW(reg, AR_PHY_PAPRD_CTRL1_MAG_SCALE_FACT, 361);
2446 		reg &= ~AR_PHY_PAPRD_CTRL1_ADAPTIVE_SCALING_ENA;
2447 		reg |= AR_PHY_PAPRD_CTRL1_ADAPTIVE_AM2AM_ENA;
2448 		reg |= AR_PHY_PAPRD_CTRL1_ADAPTIVE_AM2PM_ENA;
2449 		AR_WRITE(sc, AR_PHY_PAPRD_CTRL1_B(i), reg);
2450 
2451 		reg = AR_READ(sc, AR_PHY_PAPRD_CTRL0_B(i));
2452 		reg = RW(reg, AR_PHY_PAPRD_CTRL0_PAPRD_MAG_THRSH, 3);
2453 		AR_WRITE(sc, AR_PHY_PAPRD_CTRL0_B(i), reg);
2454 	}
2455 
2456 	/* Disable all digital predistorters during calibration. */
2457 	for (i = 0; i < AR9003_MAX_CHAINS; i++) {
2458 		AR_CLRBITS(sc, AR_PHY_PAPRD_CTRL0_B(i),
2459 		    AR_PHY_PAPRD_CTRL0_PAPRD_ENABLE);
2460 	}
2461 	AR_WRITE_BARRIER(sc);
2462 
2463 	/*
2464 	 * Configure training signal.
2465 	 */
2466 	reg = AR_READ(sc, AR_PHY_PAPRD_TRAINER_CNTL1);
2467 	reg = RW(reg, AR_PHY_PAPRD_TRAINER_CNTL1_AGC2_SETTLING, 28);
2468 	reg = RW(reg, AR_PHY_PAPRD_TRAINER_CNTL1_LB_SKIP, 0x30);
2469 	reg &= ~AR_PHY_PAPRD_TRAINER_CNTL1_RX_BB_GAIN_FORCE;
2470 	reg &= ~AR_PHY_PAPRD_TRAINER_CNTL1_IQCORR_ENABLE;
2471 	reg |= AR_PHY_PAPRD_TRAINER_CNTL1_LB_ENABLE;
2472 	reg |= AR_PHY_PAPRD_TRAINER_CNTL1_TX_GAIN_FORCE;
2473 	reg |= AR_PHY_PAPRD_TRAINER_CNTL1_TRAIN_ENABLE;
2474 	AR_WRITE(sc, AR_PHY_PAPRD_TRAINER_CNTL1, reg);
2475 
2476 	AR_WRITE(sc, AR_PHY_PAPRD_TRAINER_CNTL2, 147);
2477 
2478 	reg = AR_READ(sc, AR_PHY_PAPRD_TRAINER_CNTL3);
2479 	reg = RW(reg, AR_PHY_PAPRD_TRAINER_CNTL3_FINE_CORR_LEN, 4);
2480 	reg = RW(reg, AR_PHY_PAPRD_TRAINER_CNTL3_COARSE_CORR_LEN, 4);
2481 	reg = RW(reg, AR_PHY_PAPRD_TRAINER_CNTL3_NUM_CORR_STAGES, 7);
2482 	reg = RW(reg, AR_PHY_PAPRD_TRAINER_CNTL3_MIN_LOOPBACK_DEL, 1);
2483 	if (AR_SREV_9485(sc))
2484 		reg = RW(reg, AR_PHY_PAPRD_TRAINER_CNTL3_QUICK_DROP, -3);
2485 	else
2486 		reg = RW(reg, AR_PHY_PAPRD_TRAINER_CNTL3_QUICK_DROP, -6);
2487 	reg = RW(reg, AR_PHY_PAPRD_TRAINER_CNTL3_ADC_DESIRED_SIZE, -15);
2488 	reg |= AR_PHY_PAPRD_TRAINER_CNTL3_BBTXMIX_DISABLE;
2489 	AR_WRITE(sc, AR_PHY_PAPRD_TRAINER_CNTL3, reg);
2490 
2491 	reg = AR_READ(sc, AR_PHY_PAPRD_TRAINER_CNTL4);
2492 	reg = RW(reg, AR_PHY_PAPRD_TRAINER_CNTL4_SAFETY_DELTA, 0);
2493 	reg = RW(reg, AR_PHY_PAPRD_TRAINER_CNTL4_MIN_CORR, 400);
2494 	reg = RW(reg, AR_PHY_PAPRD_TRAINER_CNTL4_NUM_TRAIN_SAMPLES, 100);
2495 	AR_WRITE(sc, AR_PHY_PAPRD_TRAINER_CNTL4, reg);
2496 
2497 	for (i = 0; i < nitems(scaling); i++) {
2498 		reg = AR_READ(sc, AR_PHY_PAPRD_PRE_POST_SCALE_B0(i));
2499 		reg = RW(reg, AR_PHY_PAPRD_PRE_POST_SCALING, scaling[i]);
2500 		AR_WRITE(sc, AR_PHY_PAPRD_PRE_POST_SCALE_B0(i), reg);
2501 	}
2502 
2503 	/* Save Tx gain table. */
2504 	for (i = 0; i < AR9003_TX_GAIN_TABLE_SIZE; i++)
2505 		sc->txgain[i] = AR_READ(sc, AR_PHY_TXGAIN_TABLE(i));
2506 
2507 	/* Set Tx power of training signal (use setting for MCS0). */
2508 	sc->trainpow = MS(AR_READ(sc, AR_PHY_PWRTX_RATE5),
2509 	    AR_PHY_PWRTX_RATE5_POWERTXHT20_0) - 4;
2510 
2511 	/*
2512 	 * Start PA predistortion calibration state machine.
2513 	 */
2514 	/* Find first available Tx chain. */
2515 	sc->paprd_curchain = 0;
2516 	while (!(sc->txchainmask & (1 << sc->paprd_curchain)))
2517 		sc->paprd_curchain++;
2518 
2519 	/* Make sure training done bit is clear. */
2520 	AR_CLRBITS(sc, AR_PHY_PAPRD_TRAINER_STAT1,
2521 	    AR_PHY_PAPRD_TRAINER_STAT1_TRAIN_DONE);
2522 	AR_WRITE_BARRIER(sc);
2523 
2524 	/* Transmit training signal. */
2525 	ar9003_paprd_tx_tone(sc);
2526 }
2527 
2528 int
2529 ar9003_get_desired_txgain(struct athn_softc *sc, int chain, int pow)
2530 {
2531 	int32_t scale, atemp, avolt, tempcal, voltcal, temp, volt;
2532 	int32_t tempcorr, voltcorr;
2533 	uint32_t reg;
2534 	int8_t delta;
2535 
2536 	scale = MS(AR_READ(sc, AR_PHY_TPC_12),
2537 	    AR_PHY_TPC_12_DESIRED_SCALE_HT40_5);
2538 
2539 	reg = AR_READ(sc, AR_PHY_TPC_19);
2540 	atemp = MS(reg, AR_PHY_TPC_19_ALPHA_THERM);
2541 	avolt = MS(reg, AR_PHY_TPC_19_ALPHA_VOLT);
2542 
2543 	reg = AR_READ(sc, AR_PHY_TPC_18);
2544 	tempcal = MS(reg, AR_PHY_TPC_18_THERM_CAL);
2545 	voltcal = MS(reg, AR_PHY_TPC_18_VOLT_CAL);
2546 
2547 	reg = AR_READ(sc, AR_PHY_BB_THERM_ADC_4);
2548 	temp = MS(reg, AR_PHY_BB_THERM_ADC_4_LATEST_THERM);
2549 	volt = MS(reg, AR_PHY_BB_THERM_ADC_4_LATEST_VOLT);
2550 
2551 	delta = (int8_t)MS(AR_READ(sc, AR_PHY_TPC_11_B(chain)),
2552 	    AR_PHY_TPC_11_OLPC_GAIN_DELTA);
2553 
2554 	/* Compute temperature and voltage correction. */
2555 	tempcorr = (atemp * (temp - tempcal) + 128) / 256;
2556 	voltcorr = (avolt * (volt - voltcal) + 64) / 128;
2557 
2558 	/* Compute desired Tx gain. */
2559 	return (pow - delta - tempcorr - voltcorr + scale);
2560 }
2561 
2562 void
2563 ar9003_force_txgain(struct athn_softc *sc, uint32_t txgain)
2564 {
2565 	uint32_t reg;
2566 
2567 	reg = AR_READ(sc, AR_PHY_TX_FORCED_GAIN);
2568 	reg = RW(reg, AR_PHY_TX_FORCED_GAIN_TXBB1DBGAIN,
2569 	    MS(txgain, AR_PHY_TXGAIN_TXBB1DBGAIN));
2570 	reg = RW(reg, AR_PHY_TX_FORCED_GAIN_TXBB6DBGAIN,
2571 	    MS(txgain, AR_PHY_TXGAIN_TXBB6DBGAIN));
2572 	reg = RW(reg, AR_PHY_TX_FORCED_GAIN_TXMXRGAIN,
2573 	    MS(txgain, AR_PHY_TXGAIN_TXMXRGAIN));
2574 	reg = RW(reg, AR_PHY_TX_FORCED_GAIN_PADRVGNA,
2575 	    MS(txgain, AR_PHY_TXGAIN_PADRVGNA));
2576 	reg = RW(reg, AR_PHY_TX_FORCED_GAIN_PADRVGNB,
2577 	    MS(txgain, AR_PHY_TXGAIN_PADRVGNB));
2578 	reg = RW(reg, AR_PHY_TX_FORCED_GAIN_PADRVGNC,
2579 	    MS(txgain, AR_PHY_TXGAIN_PADRVGNC));
2580 	reg = RW(reg, AR_PHY_TX_FORCED_GAIN_PADRVGND,
2581 	    MS(txgain, AR_PHY_TXGAIN_PADRVGND));
2582 	reg &= ~AR_PHY_TX_FORCED_GAIN_ENABLE_PAL;
2583 	reg &= ~AR_PHY_TX_FORCED_GAIN_FORCE_TX_GAIN;
2584 	AR_WRITE(sc, AR_PHY_TX_FORCED_GAIN, reg);
2585 
2586 	reg = AR_READ(sc, AR_PHY_TPC_1);
2587 	reg = RW(reg, AR_PHY_TPC_1_FORCED_DAC_GAIN, 0);
2588 	reg &= ~AR_PHY_TPC_1_FORCE_DAC_GAIN;
2589 	AR_WRITE(sc, AR_PHY_TPC_1, reg);
2590 	AR_WRITE_BARRIER(sc);
2591 }
2592 
2593 void
2594 ar9003_set_training_gain(struct athn_softc *sc, int chain)
2595 {
2596 	int i, gain;
2597 
2598 	/*
2599 	 * Get desired gain for training signal power (take into account
2600 	 * current temperature/voltage).
2601 	 */
2602 	gain = ar9003_get_desired_txgain(sc, chain, sc->trainpow);
2603 	/* Find entry in table. */
2604 	for (i = 0; i < AR9003_TX_GAIN_TABLE_SIZE - 1; i++)
2605 		if (MS(sc->txgain[i], AR_PHY_TXGAIN_INDEX) >= gain)
2606 			break;
2607 	ar9003_force_txgain(sc, sc->txgain[i]);
2608 }
2609 
2610 int
2611 ar9003_paprd_tx_tone(struct athn_softc *sc)
2612 {
2613 #define TONE_LEN	1800
2614 	struct ieee80211com *ic = &sc->sc_ic;
2615 	struct ieee80211_frame *wh;
2616 	struct ieee80211_node *ni;
2617 	struct mbuf *m;
2618 	int error;
2619 
2620 	/* Build a Null (no data) frame of TONE_LEN bytes. */
2621 	m = MCLGETI(NULL, M_DONTWAIT, NULL, TONE_LEN);
2622 	if (m == NULL)
2623 		return (ENOBUFS);
2624 	memset(mtod(m, caddr_t), 0, TONE_LEN);
2625 	wh = mtod(m, struct ieee80211_frame *);
2626 	wh->i_fc[0] = IEEE80211_FC0_TYPE_DATA | IEEE80211_FC0_SUBTYPE_NODATA;
2627 	wh->i_fc[1] = IEEE80211_FC1_DIR_NODS;
2628 	*(uint16_t *)wh->i_dur = htole16(10);	/* XXX */
2629 	IEEE80211_ADDR_COPY(wh->i_addr1, ic->ic_myaddr);
2630 	IEEE80211_ADDR_COPY(wh->i_addr2, ic->ic_myaddr);
2631 	IEEE80211_ADDR_COPY(wh->i_addr3, ic->ic_myaddr);
2632 	m->m_pkthdr.len = m->m_len = TONE_LEN;
2633 
2634 	/* Set gain of training signal. */
2635 	ar9003_set_training_gain(sc, sc->paprd_curchain);
2636 
2637 	/* Transmit training signal. */
2638 	ni = ieee80211_ref_node(ic->ic_bss);
2639 	if ((error = ar9003_tx(sc, m, ni, ATHN_TXFLAG_PAPRD)) != 0)
2640 		ieee80211_release_node(ic, ni);
2641 	return (error);
2642 #undef TONE_LEN
2643 }
2644 
2645 static __inline int
2646 get_scale(int val)
2647 {
2648 	int log = 0;
2649 
2650 	/* Find the log base 2 (position of highest bit set). */
2651 	while (val >>= 1)
2652 		log++;
2653 
2654 	return ((log > 10) ? log - 10 : 0);
2655 }
2656 
2657 /*
2658  * Compute predistortion function to linearize power amplifier output based
2659  * on feedback from training signal.
2660  */
2661 int
2662 ar9003_compute_predistortion(struct athn_softc *sc, const uint32_t *lo,
2663     const uint32_t *hi)
2664 {
2665 #define NBINS	23
2666 	int chain = sc->paprd_curchain;
2667 	int x[NBINS + 1], y[NBINS + 1], t[NBINS + 1];
2668 	int b1[NBINS + 1], b2[NBINS + 1], xtilde[NBINS + 1];
2669 	int nsamples, txsum, rxsum, rosum, maxidx;
2670 	int order, order5x, order5xrem, order3x, order3xrem, y5, y3;
2671 	int icept, G, I, L, M, angle, xnonlin, y2, y4, sumy2, sumy4;
2672 	int alpha, beta, scale, Qalpha, Qbeta, Qscale, Qx, Qb1, Qb2;
2673 	int tavg, ttilde, maxb1abs, maxb2abs, maxxtildeabs, in;
2674 	int tmp, i;
2675 
2676 	/* Set values at origin. */
2677 	x[0] = y[0] = t[0] = 0;
2678 
2679 #define SCALE	32
2680 	maxidx = 0;
2681 	for (i = 0; i < NBINS; i++) {
2682 		nsamples = lo[i] & 0xffff;
2683 		/* Skip bins that contain 16 or less samples. */
2684 		if (nsamples <= 16) {
2685 			x[i + 1] = y[i + 1] = t[i + 1] = 0;
2686 			continue;
2687 		}
2688 		txsum = (hi[i] & 0x7ff) << 16 | lo[i] >> 16;
2689 		rxsum = (lo[i + NBINS] & 0xffff) << 5 |
2690 		    ((hi[i] >> 11) & 0x1f);
2691 		rosum = (hi[i + NBINS] & 0x7ff) << 16 | hi[i + NBINS] >> 16;
2692 		/* Sign-extend 27-bit value. */
2693 		rosum = (rosum ^ 0x4000000) - 0x4000000;
2694 
2695 		txsum *= SCALE;
2696 		rxsum *= SCALE;
2697 		rosum *= SCALE;
2698 
2699 		x[i + 1] = ((txsum + nsamples) / nsamples + SCALE) / SCALE;
2700 		y[i + 1] = ((rxsum + nsamples) / nsamples + SCALE) / SCALE +
2701 		    SCALE * maxidx + SCALE / 2;
2702 		t[i + 1] = (rosum + nsamples) / nsamples;
2703 		maxidx++;
2704 	}
2705 #undef SCALE
2706 
2707 #define SCALE_LOG	8
2708 #define SCALE		(1 << SCALE_LOG)
2709 	if (x[6] == x[3])
2710 		return (1);	/* Prevent division by 0. */
2711 	G = ((y[6] - y[3]) * SCALE + (x[6] - x[3])) / (x[6] - x[3]);
2712 	if (G == 0)
2713 		return (1);	/* Prevent division by 0. */
2714 
2715 	sc->gain1[chain] = G;	/* Save low signal gain. */
2716 
2717 	/* Find interception point. */
2718 	icept = (G * (x[0] - x[3]) + SCALE) / SCALE + y[3];
2719 	for (i = 0; i <= 3; i++) {
2720 		y[i] = i * 32;
2721 		x[i] = (y[i] * SCALE + G) / G;
2722 	}
2723 	for (i = 4; i <= maxidx; i++)
2724 		y[i] -= icept;
2725 
2726 	xnonlin = x[maxidx] - (y[maxidx] * SCALE + G) / G;
2727 	order = (xnonlin + y[maxidx]) / y[maxidx];
2728 	if (order == 0)
2729 		M = 10;
2730 	else if (order == 1)
2731 		M = 9;
2732 	else
2733 		M = 8;
2734 
2735 	I = (maxidx >= 16) ? 7 : maxidx / 2;
2736 	L = maxidx - I;
2737 
2738 	sumy2 = sumy4 = y2 = y4 = 0;
2739 	for (i = 0; i <= L; i++) {
2740 		if (y[i + I] == 0)
2741 			return (1);	/* Prevent division by 0. */
2742 
2743 		xnonlin = x[i + I] - ((y[i + I] * SCALE) + G) / G;
2744 		xtilde[i] = ((xnonlin << M) + y[i + I]) / y[i + I];
2745 		xtilde[i] = ((xtilde[i] << M) + y[i + I]) / y[i + I];
2746 		xtilde[i] = ((xtilde[i] << M) + y[i + I]) / y[i + I];
2747 
2748 		y2 = (y[i + I] * y[i + I] + SCALE * SCALE) / (SCALE * SCALE);
2749 
2750 		sumy2 += y2;
2751 		sumy4 += y2 * y2;
2752 
2753 		b1[i] = y2 * (L + 1);
2754 		b2[i] = y2;
2755 	}
2756 	for (i = 0; i <= L; i++) {
2757 		b1[i] -= sumy2;
2758 		b2[i] = sumy4 - sumy2 * b2[i];
2759 	}
2760 
2761 	maxxtildeabs = maxb1abs = maxb2abs = 0;
2762 	for (i = 0; i <= L; i++) {
2763 		tmp = abs(xtilde[i]);
2764 		if (tmp > maxxtildeabs)
2765 			maxxtildeabs = tmp;
2766 
2767 		tmp = abs(b1[i]);
2768 		if (tmp > maxb1abs)
2769 			maxb1abs = tmp;
2770 
2771 		tmp = abs(b2[i]);
2772 		if (tmp > maxb2abs)
2773 			maxb2abs = tmp;
2774 	}
2775 	Qx  = get_scale(maxxtildeabs);
2776 	Qb1 = get_scale(maxb1abs);
2777 	Qb2 = get_scale(maxb2abs);
2778 	for (i = 0; i <= L; i++) {
2779 		xtilde[i] /= 1 << Qx;
2780 		b1[i] /= 1 << Qb1;
2781 		b2[i] /= 1 << Qb2;
2782 	}
2783 
2784 	alpha = beta = 0;
2785 	for (i = 0; i <= L; i++) {
2786 		alpha += b1[i] * xtilde[i];
2787 		beta  += b2[i] * xtilde[i];
2788 	}
2789 
2790 	scale = ((y4 / SCALE_LOG) * (L + 1) -
2791 		 (y2 / SCALE_LOG) * sumy2) * SCALE_LOG;
2792 
2793 	Qscale = get_scale(abs(scale));
2794 	scale /= 1 << Qscale;
2795 	Qalpha = get_scale(abs(alpha));
2796 	alpha /= 1 << Qalpha;
2797 	Qbeta  = get_scale(abs(beta));
2798 	beta  /= 1 << Qbeta;
2799 
2800 	order = 3 * M - Qx - Qb1 - Qbeta + 10 + Qscale;
2801 	order5x = 1 << (order / 5);
2802 	order5xrem = 1 << (order % 5);
2803 
2804 	order = 3 * M - Qx - Qb2 - Qalpha + 10 + Qscale;
2805 	order3x = 1 << (order / 3);
2806 	order3xrem = 1 << (order % 3);
2807 
2808 	for (i = 0; i < AR9003_PAPRD_MEM_TAB_SIZE; i++) {
2809 		tmp = i * 32;
2810 
2811 		/* Fifth order. */
2812 		y5 = ((beta * tmp) / 64) / order5x;
2813 		y5 = (y5 * tmp) / order5x;
2814 		y5 = (y5 * tmp) / order5x;
2815 		y5 = (y5 * tmp) / order5x;
2816 		y5 = (y5 * tmp) / order5x;
2817 		y5 = y5 / order5xrem;
2818 
2819 		/* Third oder. */
2820 		y3 = (alpha * tmp) / order3x;
2821 		y3 = (y3 * tmp) / order3x;
2822 		y3 = (y3 * tmp) / order3x;
2823 		y3 = y3 / order3xrem;
2824 
2825 		in = y5 + y3 + (SCALE * tmp) / G;
2826 		if (i >= 2 && in < sc->pa_in[chain][i - 1]) {
2827 			in = sc->pa_in[chain][i - 1] +
2828 			    (sc->pa_in[chain][i - 1] -
2829 			     sc->pa_in[chain][i - 2]);
2830 		}
2831 		if (in > 1400)
2832 			in = 1400;
2833 		sc->pa_in[chain][i] = in;
2834 	}
2835 
2836 	/* Compute average theta of first 5 bins (linear region). */
2837 	tavg = 0;
2838 	for (i = 1; i <= 5; i++)
2839 		tavg += t[i];
2840 	tavg /= 5;
2841 	for (i = 1; i <= 5; i++)
2842 		t[i] = 0;
2843 	for (i = 6; i <= maxidx; i++)
2844 		t[i] -= tavg;
2845 
2846 	alpha = beta = 0;
2847 	for (i = 0; i <= L; i++) {
2848 		ttilde = ((t[i + I] << M) + y[i + I]) / y[i + I];
2849 		ttilde = ((ttilde << M) +  y[i + I]) / y[i + I];
2850 		ttilde = ((ttilde << M) +  y[i + I]) / y[i + I];
2851 
2852 		alpha += b2[i] * ttilde;
2853 		beta  += b1[i] * ttilde;
2854 	}
2855 
2856 	Qalpha = get_scale(abs(alpha));
2857 	alpha /= 1 << Qalpha;
2858 	Qbeta  = get_scale(abs(beta));
2859 	beta  /= 1 << Qbeta;
2860 
2861 	order = 3 * M - Qx - Qb1 - Qbeta + 10 + Qscale + 5;
2862 	order5x = 1 << (order / 5);
2863 	order5xrem = 1 << (order % 5);
2864 
2865 	order = 3 * M - Qx - Qb2 - Qalpha + 10 + Qscale + 5;
2866 	order3x = 1 << (order / 3);
2867 	order3xrem = 1 << (order % 3);
2868 
2869 	for (i = 0; i <= 4; i++)
2870 		sc->angle[chain][i] = 0;	/* Linear at that range. */
2871 	for (i = 5; i < AR9003_PAPRD_MEM_TAB_SIZE; i++) {
2872 		tmp = i * 32;
2873 
2874 		/* Fifth order. */
2875 		if (beta > 0)
2876 			y5 = (((beta * tmp - 64) / 64) - order5x) / order5x;
2877 		else
2878 			y5 = (((beta * tmp - 64) / 64) + order5x) / order5x;
2879 		y5 = (y5 * tmp) / order5x;
2880 		y5 = (y5 * tmp) / order5x;
2881 		y5 = (y5 * tmp) / order5x;
2882 		y5 = (y5 * tmp) / order5x;
2883 		y5 = y5 / order5xrem;
2884 
2885 		/* Third oder. */
2886 		if (beta > 0)	/* XXX alpha? */
2887 			y3 = (alpha * tmp - order3x) / order3x;
2888 		else
2889 			y3 = (alpha * tmp + order3x) / order3x;
2890 		y3 = (y3 * tmp) / order3x;
2891 		y3 = (y3 * tmp) / order3x;
2892 		y3 = y3 / order3xrem;
2893 
2894 		angle = y5 + y3;
2895 		if (angle < -150)
2896 			angle = -150;
2897 		else if (angle > 150)
2898 			angle = 150;
2899 		sc->angle[chain][i] = angle;
2900 	}
2901 	/* Angle for entry 4 is derived from angle for entry 5. */
2902 	sc->angle[chain][4] = (sc->angle[chain][5] + 2) / 2;
2903 
2904 	return (0);
2905 #undef SCALE
2906 #undef SCALE_LOG
2907 #undef NBINS
2908 }
2909 
2910 void
2911 ar9003_enable_predistorter(struct athn_softc *sc, int chain)
2912 {
2913 	uint32_t reg;
2914 	int i;
2915 
2916 	/* Write digital predistorter lookup table. */
2917 	for (i = 0; i < AR9003_PAPRD_MEM_TAB_SIZE; i++) {
2918 		AR_WRITE(sc, AR_PHY_PAPRD_MEM_TAB_B(chain, i),
2919 		    SM(AR_PHY_PAPRD_PA_IN, sc->pa_in[chain][i]) |
2920 		    SM(AR_PHY_PAPRD_ANGLE, sc->angle[chain][i]));
2921 	}
2922 
2923 	reg = AR_READ(sc, AR_PHY_PA_GAIN123_B(chain));
2924 	reg = RW(reg, AR_PHY_PA_GAIN123_PA_GAIN1, sc->gain1[chain]);
2925 	AR_WRITE(sc, AR_PHY_PA_GAIN123_B(chain), reg);
2926 
2927 	/* Indicate Tx power used for calibration (training signal). */
2928 	reg = AR_READ(sc, AR_PHY_PAPRD_CTRL1_B(chain));
2929 	reg = RW(reg, AR_PHY_PAPRD_CTRL1_POWER_AT_AM2AM_CAL, sc->trainpow);
2930 	AR_WRITE(sc, AR_PHY_PAPRD_CTRL1_B(chain), reg);
2931 
2932 	/* Enable digital predistorter for this chain. */
2933 	AR_SETBITS(sc, AR_PHY_PAPRD_CTRL0_B(chain),
2934 	    AR_PHY_PAPRD_CTRL0_PAPRD_ENABLE);
2935 	AR_WRITE_BARRIER(sc);
2936 }
2937 
2938 void
2939 ar9003_paprd_enable(struct athn_softc *sc)
2940 {
2941 	int i;
2942 
2943 	/* Enable digital predistorters for all Tx chains. */
2944 	for (i = 0; i < AR9003_MAX_CHAINS; i++)
2945 		if (sc->txchainmask & (1 << i))
2946 			ar9003_enable_predistorter(sc, i);
2947 }
2948 
2949 /*
2950  * This function is called when our training signal has been sent.
2951  */
2952 void
2953 ar9003_paprd_tx_tone_done(struct athn_softc *sc)
2954 {
2955 	uint32_t lo[48], hi[48];
2956 	int i;
2957 
2958 	/* Make sure training is complete. */
2959 	if (!(AR_READ(sc, AR_PHY_PAPRD_TRAINER_STAT1) &
2960 	    AR_PHY_PAPRD_TRAINER_STAT1_TRAIN_DONE))
2961 		return;
2962 
2963 	/* Read feedback from training signal. */
2964 	AR_CLRBITS(sc, AR_PHY_CHAN_INFO_MEMORY, AR_PHY_CHAN_INFO_TAB_S2_READ);
2965 	for (i = 0; i < nitems(lo); i++)
2966 		lo[i] = AR_READ(sc, AR_PHY_CHAN_INFO_TAB(0, i));
2967 	AR_SETBITS(sc, AR_PHY_CHAN_INFO_MEMORY, AR_PHY_CHAN_INFO_TAB_S2_READ);
2968 	for (i = 0; i < nitems(hi); i++)
2969 		hi[i] = AR_READ(sc, AR_PHY_CHAN_INFO_TAB(0, i));
2970 
2971 	AR_CLRBITS(sc, AR_PHY_PAPRD_TRAINER_STAT1,
2972 	    AR_PHY_PAPRD_TRAINER_STAT1_TRAIN_DONE);
2973 
2974 	/* Compute predistortion function based on this feedback. */
2975 	if (ar9003_compute_predistortion(sc, lo, hi) != 0)
2976 		return;
2977 
2978 	/* Get next available Tx chain. */
2979 	while (++sc->paprd_curchain < AR9003_MAX_CHAINS)
2980 		if (sc->txchainmask & (1 << sc->paprd_curchain))
2981 			break;
2982 	if (sc->paprd_curchain == AR9003_MAX_CHAINS) {
2983 		/* All Tx chains measured; enable digital predistortion. */
2984 		ar9003_paprd_enable(sc);
2985 	} else	/* Measure next Tx chain. */
2986 		ar9003_paprd_tx_tone(sc);
2987 }
2988 
2989 void
2990 ar9003_write_txpower(struct athn_softc *sc, int16_t power[ATHN_POWER_COUNT])
2991 {
2992 	/* Make sure forced gain is disabled. */
2993 	AR_WRITE(sc, AR_PHY_TX_FORCED_GAIN, 0);
2994 
2995 	AR_WRITE(sc, AR_PHY_PWRTX_RATE1,
2996 	    (power[ATHN_POWER_OFDM18  ] & 0x3f) << 24 |
2997 	    (power[ATHN_POWER_OFDM12  ] & 0x3f) << 16 |
2998 	    (power[ATHN_POWER_OFDM9   ] & 0x3f) <<  8 |
2999 	    (power[ATHN_POWER_OFDM6   ] & 0x3f));
3000 	AR_WRITE(sc, AR_PHY_PWRTX_RATE2,
3001 	    (power[ATHN_POWER_OFDM54  ] & 0x3f) << 24 |
3002 	    (power[ATHN_POWER_OFDM48  ] & 0x3f) << 16 |
3003 	    (power[ATHN_POWER_OFDM36  ] & 0x3f) <<  8 |
3004 	    (power[ATHN_POWER_OFDM24  ] & 0x3f));
3005 	AR_WRITE(sc, AR_PHY_PWRTX_RATE3,
3006 	    (power[ATHN_POWER_CCK2_SP ] & 0x3f) << 24 |
3007 	    (power[ATHN_POWER_CCK2_LP ] & 0x3f) << 16 |
3008 	    /* NB: No eXtended Range for AR9003. */
3009 	    (power[ATHN_POWER_CCK1_LP ] & 0x3f));
3010 	AR_WRITE(sc, AR_PHY_PWRTX_RATE4,
3011 	    (power[ATHN_POWER_CCK11_SP] & 0x3f) << 24 |
3012 	    (power[ATHN_POWER_CCK11_LP] & 0x3f) << 16 |
3013 	    (power[ATHN_POWER_CCK55_SP] & 0x3f) <<  8 |
3014 	    (power[ATHN_POWER_CCK55_LP] & 0x3f));
3015 	/*
3016 	 * NB: AR_PHY_PWRTX_RATE5 needs to be written even if HT is disabled
3017 	 * because it is read by PA predistortion functions.
3018 	 */
3019 	AR_WRITE(sc, AR_PHY_PWRTX_RATE5,
3020 	    (power[ATHN_POWER_HT20( 5)] & 0x3f) << 24 |
3021 	    (power[ATHN_POWER_HT20( 4)] & 0x3f) << 16 |
3022 	    (power[ATHN_POWER_HT20( 1)] & 0x3f) <<  8 |
3023 	    (power[ATHN_POWER_HT20( 0)] & 0x3f));
3024 	AR_WRITE(sc, AR_PHY_PWRTX_RATE6,
3025 	    (power[ATHN_POWER_HT20(13)] & 0x3f) << 24 |
3026 	    (power[ATHN_POWER_HT20(12)] & 0x3f) << 16 |
3027 	    (power[ATHN_POWER_HT20( 7)] & 0x3f) <<  8 |
3028 	    (power[ATHN_POWER_HT20( 6)] & 0x3f));
3029 	AR_WRITE(sc, AR_PHY_PWRTX_RATE7,
3030 	    (power[ATHN_POWER_HT40( 5)] & 0x3f) << 24 |
3031 	    (power[ATHN_POWER_HT40( 4)] & 0x3f) << 16 |
3032 	    (power[ATHN_POWER_HT40( 1)] & 0x3f) <<  8 |
3033 	    (power[ATHN_POWER_HT40( 0)] & 0x3f));
3034 	AR_WRITE(sc, AR_PHY_PWRTX_RATE8,
3035 	    (power[ATHN_POWER_HT40(13)] & 0x3f) << 24 |
3036 	    (power[ATHN_POWER_HT40(12)] & 0x3f) << 16 |
3037 	    (power[ATHN_POWER_HT40( 7)] & 0x3f) <<  8 |
3038 	    (power[ATHN_POWER_HT40( 6)] & 0x3f));
3039 	AR_WRITE(sc, AR_PHY_PWRTX_RATE10,
3040 	    (power[ATHN_POWER_HT20(21)] & 0x3f) << 24 |
3041 	    (power[ATHN_POWER_HT20(20)] & 0x3f) << 16 |
3042 	    (power[ATHN_POWER_HT20(15)] & 0x3f) <<  8 |
3043 	    (power[ATHN_POWER_HT20(14)] & 0x3f));
3044 	AR_WRITE(sc, AR_PHY_PWRTX_RATE11,
3045 	    (power[ATHN_POWER_HT40(23)] & 0x3f) << 24 |
3046 	    (power[ATHN_POWER_HT40(22)] & 0x3f) << 16 |
3047 	    (power[ATHN_POWER_HT20(23)] & 0x3f) <<  8 |
3048 	    (power[ATHN_POWER_HT20(22)] & 0x3f));
3049 	AR_WRITE(sc, AR_PHY_PWRTX_RATE12,
3050 	    (power[ATHN_POWER_HT40(21)] & 0x3f) << 24 |
3051 	    (power[ATHN_POWER_HT40(20)] & 0x3f) << 16 |
3052 	    (power[ATHN_POWER_HT40(15)] & 0x3f) <<  8 |
3053 	    (power[ATHN_POWER_HT40(14)] & 0x3f));
3054 	AR_WRITE_BARRIER(sc);
3055 }
3056 
3057 void
3058 ar9003_reset_rx_gain(struct athn_softc *sc, struct ieee80211_channel *c)
3059 {
3060 #define X(x)	((uint32_t)(x) << 2)
3061 	const struct athn_gain *prog = sc->rx_gain;
3062 	const uint32_t *pvals;
3063 	int i;
3064 
3065 	if (IEEE80211_IS_CHAN_2GHZ(c))
3066 		pvals = prog->vals_2g;
3067 	else
3068 		pvals = prog->vals_5g;
3069 	for (i = 0; i < prog->nregs; i++)
3070 		AR_WRITE(sc, X(prog->regs[i]), pvals[i]);
3071 	AR_WRITE_BARRIER(sc);
3072 #undef X
3073 }
3074 
3075 void
3076 ar9003_reset_tx_gain(struct athn_softc *sc, struct ieee80211_channel *c)
3077 {
3078 #define X(x)	((uint32_t)(x) << 2)
3079 	const struct athn_gain *prog = sc->tx_gain;
3080 	const uint32_t *pvals;
3081 	int i;
3082 
3083 	if (IEEE80211_IS_CHAN_2GHZ(c))
3084 		pvals = prog->vals_2g;
3085 	else
3086 		pvals = prog->vals_5g;
3087 	for (i = 0; i < prog->nregs; i++)
3088 		AR_WRITE(sc, X(prog->regs[i]), pvals[i]);
3089 	AR_WRITE_BARRIER(sc);
3090 #undef X
3091 }
3092 
3093 void
3094 ar9003_hw_init(struct athn_softc *sc, struct ieee80211_channel *c,
3095     struct ieee80211_channel *extc)
3096 {
3097 #define X(x)	((uint32_t)(x) << 2)
3098 	struct athn_ops *ops = &sc->ops;
3099 	const struct athn_ini *ini = sc->ini;
3100 	const uint32_t *pvals;
3101 	uint32_t reg;
3102 	int i;
3103 
3104 	/*
3105 	 * The common init values include the pre and core phases for the
3106 	 * SoC, MAC, BB and Radio subsystems.
3107 	 */
3108 	DPRINTFN(4, ("writing pre and core init vals\n"));
3109 	for (i = 0; i < ini->ncmregs; i++) {
3110 		AR_WRITE(sc, X(ini->cmregs[i]), ini->cmvals[i]);
3111 		if (AR_IS_ANALOG_REG(X(ini->cmregs[i])))
3112 			DELAY(100);
3113 		if ((i & 0x1f) == 0)
3114 			DELAY(1);
3115 	}
3116 
3117 	/*
3118 	 * The modal init values include the post phase for the SoC, MAC,
3119 	 * BB and Radio subsystems.
3120 	 */
3121 	if (extc != NULL) {
3122 		if (IEEE80211_IS_CHAN_2GHZ(c))
3123 			pvals = ini->vals_2g40;
3124 		else
3125 			pvals = ini->vals_5g40;
3126 	} else {
3127 		if (IEEE80211_IS_CHAN_2GHZ(c))
3128 			pvals = ini->vals_2g20;
3129 		else
3130 			pvals = ini->vals_5g20;
3131 	}
3132 	DPRINTFN(4, ("writing post init vals\n"));
3133 	for (i = 0; i < ini->nregs; i++) {
3134 		AR_WRITE(sc, X(ini->regs[i]), pvals[i]);
3135 		if (AR_IS_ANALOG_REG(X(ini->regs[i])))
3136 			DELAY(100);
3137 		if ((i & 0x1f) == 0)
3138 			DELAY(1);
3139 	}
3140 
3141 	if (sc->rx_gain != NULL)
3142 		ar9003_reset_rx_gain(sc, c);
3143 	if (sc->tx_gain != NULL)
3144 		ar9003_reset_tx_gain(sc, c);
3145 
3146 	if (IEEE80211_IS_CHAN_5GHZ(c) &&
3147 	    (sc->flags & ATHN_FLAG_FAST_PLL_CLOCK)) {
3148 		/* Update modal values for fast PLL clock. */
3149 		if (extc != NULL)
3150 			pvals = ini->fastvals_5g40;
3151 		else
3152 			pvals = ini->fastvals_5g20;
3153 		DPRINTFN(4, ("writing fast pll clock init vals\n"));
3154 		for (i = 0; i < ini->nfastregs; i++) {
3155 			AR_WRITE(sc, X(ini->fastregs[i]), pvals[i]);
3156 			if (AR_IS_ANALOG_REG(X(ini->fastregs[i])))
3157 				DELAY(100);
3158 			if ((i & 0x1f) == 0)
3159 				DELAY(1);
3160 		}
3161 	}
3162 
3163 	/*
3164 	 * Set the RX_ABORT and RX_DIS bits to prevent frames with corrupted
3165 	 * descriptor status.
3166 	 */
3167 	AR_SETBITS(sc, AR_DIAG_SW, AR_DIAG_RX_DIS | AR_DIAG_RX_ABORT);
3168 
3169 	reg = AR_READ(sc, AR_PCU_MISC_MODE2);
3170 	reg &= ~AR_PCU_MISC_MODE2_ADHOC_MCAST_KEYID_ENABLE;
3171 	reg |= AR_PCU_MISC_MODE2_AGG_WEP_ENABLE_FIX;
3172 	reg |= AR_PCU_MISC_MODE2_ENABLE_AGGWEP;
3173 	AR_WRITE(sc, AR_PCU_MISC_MODE2, reg);
3174 	AR_WRITE_BARRIER(sc);
3175 
3176 	ar9003_set_phy(sc, c, extc);
3177 	ar9003_init_chains(sc);
3178 
3179 	ops->set_txpower(sc, c, extc);
3180 #undef X
3181 }
3182 
3183 void
3184 ar9003_get_lg_tpow(struct athn_softc *sc, struct ieee80211_channel *c,
3185     uint8_t ctl, const uint8_t *fbins,
3186     const struct ar_cal_target_power_leg *tgt, int nchans, uint8_t tpow[4])
3187 {
3188 	uint8_t fbin;
3189 	int i, delta, lo, hi;
3190 
3191 	lo = hi = -1;
3192 	fbin = athn_chan2fbin(c);
3193 	for (i = 0; i < nchans; i++) {
3194 		delta = fbin - fbins[i];
3195 		/* Find the largest sample that is <= our frequency. */
3196 		if (delta >= 0 && (lo == -1 || delta < fbin - fbins[lo]))
3197 			lo = i;
3198 		/* Find the smallest sample that is >= our frequency. */
3199 		if (delta <= 0 && (hi == -1 || delta > fbin - fbins[hi]))
3200 			hi = i;
3201 	}
3202 	if (lo == -1)
3203 		lo = hi;
3204 	else if (hi == -1)
3205 		hi = lo;
3206 	/* Interpolate values. */
3207 	for (i = 0; i < 4; i++) {
3208 		tpow[i] = athn_interpolate(fbin,
3209 		    fbins[lo], tgt[lo].tPow2x[i],
3210 		    fbins[hi], tgt[hi].tPow2x[i]);
3211 	}
3212 	/* XXX Apply conformance test limit. */
3213 }
3214 
3215 void
3216 ar9003_get_ht_tpow(struct athn_softc *sc, struct ieee80211_channel *c,
3217     uint8_t ctl, const uint8_t *fbins,
3218     const struct ar_cal_target_power_ht *tgt, int nchans, uint8_t tpow[14])
3219 {
3220 	uint8_t fbin;
3221 	int i, delta, lo, hi;
3222 
3223 	lo = hi = -1;
3224 	fbin = athn_chan2fbin(c);
3225 	for (i = 0; i < nchans; i++) {
3226 		delta = fbin - fbins[i];
3227 		/* Find the largest sample that is <= our frequency. */
3228 		if (delta >= 0 && (lo == -1 || delta < fbin - fbins[lo]))
3229 			lo = i;
3230 		/* Find the smallest sample that is >= our frequency. */
3231 		if (delta <= 0 && (hi == -1 || delta > fbin - fbins[hi]))
3232 			hi = i;
3233 	}
3234 	if (lo == -1)
3235 		lo = hi;
3236 	else if (hi == -1)
3237 		hi = lo;
3238 	/* Interpolate values. */
3239 	for (i = 0; i < 14; i++) {
3240 		tpow[i] = athn_interpolate(fbin,
3241 		    fbins[lo], tgt[lo].tPow2x[i],
3242 		    fbins[hi], tgt[hi].tPow2x[i]);
3243 	}
3244 	/* XXX Apply conformance test limit. */
3245 }
3246 
3247 /*
3248  * Adaptive noise immunity.
3249  */
3250 void
3251 ar9003_set_noise_immunity_level(struct athn_softc *sc, int level)
3252 {
3253 	int high = level == 4;
3254 	uint32_t reg;
3255 
3256 	reg = AR_READ(sc, AR_PHY_DESIRED_SZ);
3257 	reg = RW(reg, AR_PHY_DESIRED_SZ_TOT_DES, high ? -62 : -55);
3258 	AR_WRITE(sc, AR_PHY_DESIRED_SZ, reg);
3259 
3260 	reg = AR_READ(sc, AR_PHY_AGC);
3261 	reg = RW(reg, AR_PHY_AGC_COARSE_LOW, high ? -70 : -64);
3262 	reg = RW(reg, AR_PHY_AGC_COARSE_HIGH, high ? -12 : -14);
3263 	AR_WRITE(sc, AR_PHY_AGC, reg);
3264 
3265 	reg = AR_READ(sc, AR_PHY_FIND_SIG);
3266 	reg = RW(reg, AR_PHY_FIND_SIG_FIRPWR, high ? -80 : -78);
3267 	AR_WRITE(sc, AR_PHY_FIND_SIG, reg);
3268 	AR_WRITE_BARRIER(sc);
3269 }
3270 
3271 void
3272 ar9003_enable_ofdm_weak_signal(struct athn_softc *sc)
3273 {
3274 	uint32_t reg;
3275 
3276 	reg = AR_READ(sc, AR_PHY_SFCORR_LOW);
3277 	reg = RW(reg, AR_PHY_SFCORR_LOW_M1_THRESH_LOW, 50);
3278 	reg = RW(reg, AR_PHY_SFCORR_LOW_M2_THRESH_LOW, 40);
3279 	reg = RW(reg, AR_PHY_SFCORR_LOW_M2COUNT_THR_LOW, 48);
3280 	AR_WRITE(sc, AR_PHY_SFCORR_LOW, reg);
3281 
3282 	reg = AR_READ(sc, AR_PHY_SFCORR);
3283 	reg = RW(reg, AR_PHY_SFCORR_M1_THRESH, 77);
3284 	reg = RW(reg, AR_PHY_SFCORR_M2_THRESH, 64);
3285 	reg = RW(reg, AR_PHY_SFCORR_M2COUNT_THR, 16);
3286 	AR_WRITE(sc, AR_PHY_SFCORR, reg);
3287 
3288 	reg = AR_READ(sc, AR_PHY_SFCORR_EXT);
3289 	reg = RW(reg, AR_PHY_SFCORR_EXT_M1_THRESH_LOW, 50);
3290 	reg = RW(reg, AR_PHY_SFCORR_EXT_M2_THRESH_LOW, 40);
3291 	reg = RW(reg, AR_PHY_SFCORR_EXT_M1_THRESH, 77);
3292 	reg = RW(reg, AR_PHY_SFCORR_EXT_M2_THRESH, 64);
3293 	AR_WRITE(sc, AR_PHY_SFCORR_EXT, reg);
3294 
3295 	AR_SETBITS(sc, AR_PHY_SFCORR_LOW,
3296 	    AR_PHY_SFCORR_LOW_USE_SELF_CORR_LOW);
3297 	AR_WRITE_BARRIER(sc);
3298 }
3299 
3300 void
3301 ar9003_disable_ofdm_weak_signal(struct athn_softc *sc)
3302 {
3303 	uint32_t reg;
3304 
3305 	reg = AR_READ(sc, AR_PHY_SFCORR_LOW);
3306 	reg = RW(reg, AR_PHY_SFCORR_LOW_M1_THRESH_LOW, 127);
3307 	reg = RW(reg, AR_PHY_SFCORR_LOW_M2_THRESH_LOW, 127);
3308 	reg = RW(reg, AR_PHY_SFCORR_LOW_M2COUNT_THR_LOW, 63);
3309 	AR_WRITE(sc, AR_PHY_SFCORR_LOW, reg);
3310 
3311 	reg = AR_READ(sc, AR_PHY_SFCORR);
3312 	reg = RW(reg, AR_PHY_SFCORR_M1_THRESH, 127);
3313 	reg = RW(reg, AR_PHY_SFCORR_M2_THRESH, 127);
3314 	reg = RW(reg, AR_PHY_SFCORR_M2COUNT_THR, 31);
3315 	AR_WRITE(sc, AR_PHY_SFCORR, reg);
3316 
3317 	reg = AR_READ(sc, AR_PHY_SFCORR_EXT);
3318 	reg = RW(reg, AR_PHY_SFCORR_EXT_M1_THRESH_LOW, 127);
3319 	reg = RW(reg, AR_PHY_SFCORR_EXT_M2_THRESH_LOW, 127);
3320 	reg = RW(reg, AR_PHY_SFCORR_EXT_M1_THRESH, 127);
3321 	reg = RW(reg, AR_PHY_SFCORR_EXT_M2_THRESH, 127);
3322 	AR_WRITE(sc, AR_PHY_SFCORR_EXT, reg);
3323 
3324 	AR_CLRBITS(sc, AR_PHY_SFCORR_LOW,
3325 	    AR_PHY_SFCORR_LOW_USE_SELF_CORR_LOW);
3326 	AR_WRITE_BARRIER(sc);
3327 }
3328 
3329 void
3330 ar9003_set_cck_weak_signal(struct athn_softc *sc, int high)
3331 {
3332 	uint32_t reg;
3333 
3334 	reg = AR_READ(sc, AR_PHY_CCK_DETECT);
3335 	reg = RW(reg, AR_PHY_CCK_DETECT_WEAK_SIG_THR_CCK, high ? 6 : 8);
3336 	AR_WRITE(sc, AR_PHY_CCK_DETECT, reg);
3337 	AR_WRITE_BARRIER(sc);
3338 }
3339 
3340 void
3341 ar9003_set_firstep_level(struct athn_softc *sc, int level)
3342 {
3343 	uint32_t reg;
3344 
3345 	reg = AR_READ(sc, AR_PHY_FIND_SIG);
3346 	reg = RW(reg, AR_PHY_FIND_SIG_FIRSTEP, level * 4);
3347 	AR_WRITE(sc, AR_PHY_FIND_SIG, reg);
3348 	AR_WRITE_BARRIER(sc);
3349 }
3350 
3351 void
3352 ar9003_set_spur_immunity_level(struct athn_softc *sc, int level)
3353 {
3354 	uint32_t reg;
3355 
3356 	reg = AR_READ(sc, AR_PHY_TIMING5);
3357 	reg = RW(reg, AR_PHY_TIMING5_CYCPWR_THR1, (level + 1) * 2);
3358 	AR_WRITE(sc, AR_PHY_TIMING5, reg);
3359 	AR_WRITE_BARRIER(sc);
3360 }
3361