xref: /freebsd/sys/dev/sdhci/sdhci.c (revision fdafd315)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2008 Alexander Motin <mav@FreeBSD.org>
5  * Copyright (c) 2017 Marius Strobl <marius@FreeBSD.org>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/bus.h>
32 #include <sys/callout.h>
33 #include <sys/conf.h>
34 #include <sys/kernel.h>
35 #include <sys/kobj.h>
36 #include <sys/libkern.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/module.h>
40 #include <sys/mutex.h>
41 #include <sys/resource.h>
42 #include <sys/rman.h>
43 #include <sys/sysctl.h>
44 #include <sys/taskqueue.h>
45 #include <sys/sbuf.h>
46 
47 #include <machine/bus.h>
48 #include <machine/resource.h>
49 #include <machine/stdarg.h>
50 
51 #include <dev/mmc/bridge.h>
52 #include <dev/mmc/mmcreg.h>
53 #include <dev/mmc/mmcbrvar.h>
54 
55 #include <dev/sdhci/sdhci.h>
56 
57 #include <cam/cam.h>
58 #include <cam/cam_ccb.h>
59 #include <cam/cam_debug.h>
60 #include <cam/cam_sim.h>
61 #include <cam/cam_xpt_sim.h>
62 
63 #include "mmcbr_if.h"
64 #include "sdhci_if.h"
65 
66 #include "opt_mmccam.h"
67 
68 SYSCTL_NODE(_hw, OID_AUTO, sdhci, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
69     "sdhci driver");
70 
71 static int sdhci_debug = 0;
72 SYSCTL_INT(_hw_sdhci, OID_AUTO, debug, CTLFLAG_RWTUN, &sdhci_debug, 0,
73     "Debug level");
74 u_int sdhci_quirk_clear = 0;
75 SYSCTL_INT(_hw_sdhci, OID_AUTO, quirk_clear, CTLFLAG_RWTUN, &sdhci_quirk_clear,
76     0, "Mask of quirks to clear");
77 u_int sdhci_quirk_set = 0;
78 SYSCTL_INT(_hw_sdhci, OID_AUTO, quirk_set, CTLFLAG_RWTUN, &sdhci_quirk_set, 0,
79     "Mask of quirks to set");
80 
81 #define	RD1(slot, off)	SDHCI_READ_1((slot)->bus, (slot), (off))
82 #define	RD2(slot, off)	SDHCI_READ_2((slot)->bus, (slot), (off))
83 #define	RD4(slot, off)	SDHCI_READ_4((slot)->bus, (slot), (off))
84 #define	RD_MULTI_4(slot, off, ptr, count)	\
85     SDHCI_READ_MULTI_4((slot)->bus, (slot), (off), (ptr), (count))
86 
87 #define	WR1(slot, off, val)	SDHCI_WRITE_1((slot)->bus, (slot), (off), (val))
88 #define	WR2(slot, off, val)	SDHCI_WRITE_2((slot)->bus, (slot), (off), (val))
89 #define	WR4(slot, off, val)	SDHCI_WRITE_4((slot)->bus, (slot), (off), (val))
90 #define	WR_MULTI_4(slot, off, ptr, count)	\
91     SDHCI_WRITE_MULTI_4((slot)->bus, (slot), (off), (ptr), (count))
92 
93 static void sdhci_acmd_irq(struct sdhci_slot *slot, uint16_t acmd_err);
94 static void sdhci_card_poll(void *arg);
95 static void sdhci_card_task(void *arg, int pending);
96 static void sdhci_cmd_irq(struct sdhci_slot *slot, uint32_t intmask);
97 static void sdhci_data_irq(struct sdhci_slot *slot, uint32_t intmask);
98 static int sdhci_exec_tuning(struct sdhci_slot *slot, bool reset);
99 static void sdhci_handle_card_present_locked(struct sdhci_slot *slot,
100     bool is_present);
101 static void sdhci_finish_command(struct sdhci_slot *slot);
102 static void sdhci_init(struct sdhci_slot *slot);
103 static void sdhci_read_block_pio(struct sdhci_slot *slot);
104 static void sdhci_req_done(struct sdhci_slot *slot);
105 static void sdhci_req_wakeup(struct mmc_request *req);
106 static void sdhci_retune(void *arg);
107 static void sdhci_set_clock(struct sdhci_slot *slot, uint32_t clock);
108 static void sdhci_set_power(struct sdhci_slot *slot, u_char power);
109 static void sdhci_set_transfer_mode(struct sdhci_slot *slot,
110    const struct mmc_data *data);
111 static void sdhci_start(struct sdhci_slot *slot);
112 static void sdhci_timeout(void *arg);
113 static void sdhci_start_command(struct sdhci_slot *slot,
114    struct mmc_command *cmd);
115 static void sdhci_start_data(struct sdhci_slot *slot,
116    const struct mmc_data *data);
117 static void sdhci_write_block_pio(struct sdhci_slot *slot);
118 static void sdhci_transfer_pio(struct sdhci_slot *slot);
119 
120 #ifdef MMCCAM
121 /* CAM-related */
122 static void sdhci_cam_action(struct cam_sim *sim, union ccb *ccb);
123 static int sdhci_cam_get_possible_host_clock(const struct sdhci_slot *slot,
124     int proposed_clock);
125 static void sdhci_cam_poll(struct cam_sim *sim);
126 static int sdhci_cam_request(struct sdhci_slot *slot, union ccb *ccb);
127 static int sdhci_cam_settran_settings(struct sdhci_slot *slot, union ccb *ccb);
128 static int sdhci_cam_update_ios(struct sdhci_slot *slot);
129 #endif
130 
131 /* helper routines */
132 static int sdhci_dma_alloc(struct sdhci_slot *slot);
133 static void sdhci_dma_free(struct sdhci_slot *slot);
134 static void sdhci_dumpcaps(struct sdhci_slot *slot);
135 static void sdhci_dumpcaps_buf(struct sdhci_slot *slot, struct sbuf *s);
136 static void sdhci_dumpregs(struct sdhci_slot *slot);
137 static void sdhci_dumpregs_buf(struct sdhci_slot *slot, struct sbuf *s);
138 static int sdhci_syctl_dumpcaps(SYSCTL_HANDLER_ARGS);
139 static int sdhci_syctl_dumpregs(SYSCTL_HANDLER_ARGS);
140 static void sdhci_getaddr(void *arg, bus_dma_segment_t *segs, int nsegs,
141     int error);
142 static int slot_printf(const struct sdhci_slot *slot, const char * fmt, ...)
143     __printflike(2, 3);
144 static int slot_sprintf(const struct sdhci_slot *slot, struct sbuf *s,
145     const char * fmt, ...) __printflike(3, 4);
146 static uint32_t sdhci_tuning_intmask(const struct sdhci_slot *slot);
147 
148 #define	SDHCI_LOCK(_slot)		mtx_lock(&(_slot)->mtx)
149 #define	SDHCI_UNLOCK(_slot)		mtx_unlock(&(_slot)->mtx)
150 #define	SDHCI_LOCK_INIT(_slot) \
151 	mtx_init(&_slot->mtx, "SD slot mtx", "sdhci", MTX_DEF)
152 #define	SDHCI_LOCK_DESTROY(_slot)	mtx_destroy(&_slot->mtx);
153 #define	SDHCI_ASSERT_LOCKED(_slot)	mtx_assert(&_slot->mtx, MA_OWNED);
154 #define	SDHCI_ASSERT_UNLOCKED(_slot)	mtx_assert(&_slot->mtx, MA_NOTOWNED);
155 
156 #define	SDHCI_DEFAULT_MAX_FREQ	50
157 
158 #define	SDHCI_200_MAX_DIVIDER	256
159 #define	SDHCI_300_MAX_DIVIDER	2046
160 
161 #define	SDHCI_CARD_PRESENT_TICKS	(hz / 5)
162 #define	SDHCI_INSERT_DELAY_TICKS	(hz / 2)
163 
164 /*
165  * Broadcom BCM577xx Controller Constants
166  */
167 /* Maximum divider supported by the default clock source. */
168 #define	BCM577XX_DEFAULT_MAX_DIVIDER	256
169 /* Alternative clock's base frequency. */
170 #define	BCM577XX_ALT_CLOCK_BASE		63000000
171 
172 #define	BCM577XX_HOST_CONTROL		0x198
173 #define	BCM577XX_CTRL_CLKSEL_MASK	0xFFFFCFFF
174 #define	BCM577XX_CTRL_CLKSEL_SHIFT	12
175 #define	BCM577XX_CTRL_CLKSEL_DEFAULT	0x0
176 #define	BCM577XX_CTRL_CLKSEL_64MHZ	0x3
177 
178 static void
sdhci_getaddr(void * arg,bus_dma_segment_t * segs,int nsegs,int error)179 sdhci_getaddr(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
180 {
181 
182 	if (error != 0) {
183 		printf("getaddr: error %d\n", error);
184 		return;
185 	}
186 	*(bus_addr_t *)arg = segs[0].ds_addr;
187 }
188 
189 static int
slot_printf(const struct sdhci_slot * slot,const char * fmt,...)190 slot_printf(const struct sdhci_slot *slot, const char * fmt, ...)
191 {
192 	char buf[128];
193 	va_list ap;
194 	int retval;
195 
196 	/*
197 	 * Make sure we print a single line all together rather than in two
198 	 * halves to avoid console gibberish bingo.
199 	 */
200 	va_start(ap, fmt);
201 	retval = vsnprintf(buf, sizeof(buf), fmt, ap);
202 	va_end(ap);
203 
204 	retval += printf("%s-slot%d: %s",
205 	    device_get_nameunit(slot->bus), slot->num, buf);
206 	return (retval);
207 }
208 
209 static int
slot_sprintf(const struct sdhci_slot * slot,struct sbuf * s,const char * fmt,...)210 slot_sprintf(const struct sdhci_slot *slot, struct sbuf *s,
211     const char * fmt, ...)
212 {
213 	va_list ap;
214 	int retval;
215 
216 	retval = sbuf_printf(s, "%s-slot%d: ", device_get_nameunit(slot->bus), slot->num);
217 
218 	va_start(ap, fmt);
219 	retval += sbuf_vprintf(s, fmt, ap);
220 	va_end(ap);
221 
222 	return (retval);
223 }
224 
225 static void
sdhci_dumpregs_buf(struct sdhci_slot * slot,struct sbuf * s)226 sdhci_dumpregs_buf(struct sdhci_slot *slot, struct sbuf *s)
227 {
228 	slot_sprintf(slot, s,  "============== REGISTER DUMP ==============\n");
229 
230 	slot_sprintf(slot, s,  "Sys addr: 0x%08x | Version:  0x%08x\n",
231 	    RD4(slot, SDHCI_DMA_ADDRESS), RD2(slot, SDHCI_HOST_VERSION));
232 	slot_sprintf(slot, s,  "Blk size: 0x%08x | Blk cnt:  0x%08x\n",
233 	    RD2(slot, SDHCI_BLOCK_SIZE), RD2(slot, SDHCI_BLOCK_COUNT));
234 	slot_sprintf(slot, s,  "Argument: 0x%08x | Trn mode: 0x%08x\n",
235 	    RD4(slot, SDHCI_ARGUMENT), RD2(slot, SDHCI_TRANSFER_MODE));
236 	slot_sprintf(slot, s,  "Present:  0x%08x | Host ctl: 0x%08x\n",
237 	    RD4(slot, SDHCI_PRESENT_STATE), RD1(slot, SDHCI_HOST_CONTROL));
238 	slot_sprintf(slot, s,  "Power:    0x%08x | Blk gap:  0x%08x\n",
239 	    RD1(slot, SDHCI_POWER_CONTROL), RD1(slot, SDHCI_BLOCK_GAP_CONTROL));
240 	slot_sprintf(slot, s,  "Wake-up:  0x%08x | Clock:    0x%08x\n",
241 	    RD1(slot, SDHCI_WAKE_UP_CONTROL), RD2(slot, SDHCI_CLOCK_CONTROL));
242 	slot_sprintf(slot, s,  "Timeout:  0x%08x | Int stat: 0x%08x\n",
243 	    RD1(slot, SDHCI_TIMEOUT_CONTROL), RD4(slot, SDHCI_INT_STATUS));
244 	slot_sprintf(slot, s,  "Int enab: 0x%08x | Sig enab: 0x%08x\n",
245 	    RD4(slot, SDHCI_INT_ENABLE), RD4(slot, SDHCI_SIGNAL_ENABLE));
246 	slot_sprintf(slot, s,  "AC12 err: 0x%08x | Host ctl2:0x%08x\n",
247 	    RD2(slot, SDHCI_ACMD12_ERR), RD2(slot, SDHCI_HOST_CONTROL2));
248 	slot_sprintf(slot, s,  "Caps:     0x%08x | Caps2:    0x%08x\n",
249 	    RD4(slot, SDHCI_CAPABILITIES), RD4(slot, SDHCI_CAPABILITIES2));
250 	slot_sprintf(slot, s,  "Max curr: 0x%08x | ADMA err: 0x%08x\n",
251 	    RD4(slot, SDHCI_MAX_CURRENT), RD1(slot, SDHCI_ADMA_ERR));
252 	slot_sprintf(slot, s,  "ADMA addr:0x%08x | Slot int: 0x%08x\n",
253 	    RD4(slot, SDHCI_ADMA_ADDRESS_LO), RD2(slot, SDHCI_SLOT_INT_STATUS));
254 
255 	slot_sprintf(slot, s,  "===========================================\n");
256 }
257 
258 static void
sdhci_dumpregs(struct sdhci_slot * slot)259 sdhci_dumpregs(struct sdhci_slot *slot)
260 {
261 	struct sbuf s;
262 
263 	if (sbuf_new(&s, NULL, 1024, SBUF_NOWAIT | SBUF_AUTOEXTEND) == NULL) {
264 		slot_printf(slot, "sdhci_dumpregs: Failed to allocate memory for sbuf\n");
265 		return;
266 	}
267 
268 	sbuf_set_drain(&s, &sbuf_printf_drain, NULL);
269 	sdhci_dumpregs_buf(slot, &s);
270 	sbuf_finish(&s);
271 	sbuf_delete(&s);
272 }
273 
274 static int
sdhci_syctl_dumpregs(SYSCTL_HANDLER_ARGS)275 sdhci_syctl_dumpregs(SYSCTL_HANDLER_ARGS)
276 {
277 	struct sdhci_slot *slot = arg1;
278 	struct sbuf s;
279 
280 	sbuf_new_for_sysctl(&s, NULL, 1024, req);
281 	sbuf_putc(&s, '\n');
282 	sdhci_dumpregs_buf(slot, &s);
283 	sbuf_finish(&s);
284 	sbuf_delete(&s);
285 
286 	return (0);
287 }
288 
289 static void
sdhci_dumpcaps_buf(struct sdhci_slot * slot,struct sbuf * s)290 sdhci_dumpcaps_buf(struct sdhci_slot *slot, struct sbuf *s)
291 {
292 	int host_caps = slot->host.caps;
293 	int caps = slot->caps;
294 
295 	slot_sprintf(slot, s,
296 	    "%uMHz%s %s VDD:%s%s%s VCCQ: 3.3V%s%s DRV: B%s%s%s %s %s\n",
297 	    slot->max_clk / 1000000,
298 	    (caps & SDHCI_CAN_DO_HISPD) ? " HS" : "",
299 	    (host_caps & MMC_CAP_8_BIT_DATA) ? "8bits" :
300 	    ((host_caps & MMC_CAP_4_BIT_DATA) ? "4bits" : "1bit"),
301 	    (caps & SDHCI_CAN_VDD_330) ? " 3.3V" : "",
302 	    (caps & SDHCI_CAN_VDD_300) ? " 3.0V" : "",
303 	    ((caps & SDHCI_CAN_VDD_180) &&
304 	    (slot->opt & SDHCI_SLOT_EMBEDDED)) ? " 1.8V" : "",
305 	    (host_caps & MMC_CAP_SIGNALING_180) ? " 1.8V" : "",
306 	    (host_caps & MMC_CAP_SIGNALING_120) ? " 1.2V" : "",
307 	    (host_caps & MMC_CAP_DRIVER_TYPE_A) ? "A" : "",
308 	    (host_caps & MMC_CAP_DRIVER_TYPE_C) ? "C" : "",
309 	    (host_caps & MMC_CAP_DRIVER_TYPE_D) ? "D" : "",
310 	    (slot->opt & SDHCI_HAVE_DMA) ? "DMA" : "PIO",
311 	    (slot->opt & SDHCI_SLOT_EMBEDDED) ? "embedded" :
312 	    (slot->opt & SDHCI_NON_REMOVABLE) ? "non-removable" :
313 	    "removable");
314 	if (host_caps & (MMC_CAP_MMC_DDR52 | MMC_CAP_MMC_HS200 |
315 	    MMC_CAP_MMC_HS400 | MMC_CAP_MMC_ENH_STROBE))
316 		slot_sprintf(slot, s, "eMMC:%s%s%s%s\n",
317 		    (host_caps & MMC_CAP_MMC_DDR52) ? " DDR52" : "",
318 		    (host_caps & MMC_CAP_MMC_HS200) ? " HS200" : "",
319 		    (host_caps & MMC_CAP_MMC_HS400) ? " HS400" : "",
320 		    ((host_caps &
321 		    (MMC_CAP_MMC_HS400 | MMC_CAP_MMC_ENH_STROBE)) ==
322 		    (MMC_CAP_MMC_HS400 | MMC_CAP_MMC_ENH_STROBE)) ?
323 		    " HS400ES" : "");
324 	if (host_caps & (MMC_CAP_UHS_SDR12 | MMC_CAP_UHS_SDR25 |
325 	    MMC_CAP_UHS_SDR50 | MMC_CAP_UHS_SDR104))
326 		slot_sprintf(slot, s, "UHS-I:%s%s%s%s%s\n",
327 		    (host_caps & MMC_CAP_UHS_SDR12) ? " SDR12" : "",
328 		    (host_caps & MMC_CAP_UHS_SDR25) ? " SDR25" : "",
329 		    (host_caps & MMC_CAP_UHS_SDR50) ? " SDR50" : "",
330 		    (host_caps & MMC_CAP_UHS_SDR104) ? " SDR104" : "",
331 		    (host_caps & MMC_CAP_UHS_DDR50) ? " DDR50" : "");
332 	if (slot->opt & SDHCI_TUNING_SUPPORTED)
333 		slot_sprintf(slot, s,
334 		    "Re-tuning count %d secs, mode %d\n",
335 		    slot->retune_count, slot->retune_mode + 1);
336 }
337 
338 static void
sdhci_dumpcaps(struct sdhci_slot * slot)339 sdhci_dumpcaps(struct sdhci_slot *slot)
340 {
341 	struct sbuf s;
342 
343 	if (sbuf_new(&s, NULL, 1024, SBUF_NOWAIT | SBUF_AUTOEXTEND) == NULL) {
344 		slot_printf(slot, "sdhci_dumpcaps: Failed to allocate memory for sbuf\n");
345 		return;
346 	}
347 
348 	sbuf_set_drain(&s, &sbuf_printf_drain, NULL);
349 	sdhci_dumpcaps_buf(slot, &s);
350 	sbuf_finish(&s);
351 	sbuf_delete(&s);
352 }
353 
354 static int
sdhci_syctl_dumpcaps(SYSCTL_HANDLER_ARGS)355 sdhci_syctl_dumpcaps(SYSCTL_HANDLER_ARGS)
356 {
357 	struct sdhci_slot *slot = arg1;
358 	struct sbuf s;
359 
360 	sbuf_new_for_sysctl(&s, NULL, 1024, req);
361 	sbuf_putc(&s, '\n');
362 	sdhci_dumpcaps_buf(slot, &s);
363 	sbuf_finish(&s);
364 	sbuf_delete(&s);
365 
366 	return (0);
367 }
368 
369 static uint32_t
sdhci_tuning_intmask(const struct sdhci_slot * slot)370 sdhci_tuning_intmask(const struct sdhci_slot *slot)
371 {
372 	uint32_t intmask;
373 
374 	intmask = 0;
375 	if (slot->opt & SDHCI_TUNING_ENABLED) {
376 		intmask |= SDHCI_INT_TUNEERR;
377 		if (slot->retune_mode == SDHCI_RETUNE_MODE_2 ||
378 		    slot->retune_mode == SDHCI_RETUNE_MODE_3)
379 			intmask |= SDHCI_INT_RETUNE;
380 	}
381 	return (intmask);
382 }
383 
384 static void
sdhci_init(struct sdhci_slot * slot)385 sdhci_init(struct sdhci_slot *slot)
386 {
387 
388 	SDHCI_RESET(slot->bus, slot, SDHCI_RESET_ALL);
389 
390 	/* Enable interrupts. */
391 	slot->intmask = SDHCI_INT_BUS_POWER | SDHCI_INT_DATA_END_BIT |
392 	    SDHCI_INT_DATA_CRC | SDHCI_INT_DATA_TIMEOUT | SDHCI_INT_INDEX |
393 	    SDHCI_INT_END_BIT | SDHCI_INT_CRC | SDHCI_INT_TIMEOUT |
394 	    SDHCI_INT_DATA_AVAIL | SDHCI_INT_SPACE_AVAIL |
395 	    SDHCI_INT_DMA_END | SDHCI_INT_DATA_END | SDHCI_INT_RESPONSE |
396 	    SDHCI_INT_ACMD12ERR;
397 
398 	if (!(slot->quirks & SDHCI_QUIRK_POLL_CARD_PRESENT) &&
399 	    !(slot->opt & SDHCI_NON_REMOVABLE)) {
400 		slot->intmask |= SDHCI_INT_CARD_REMOVE | SDHCI_INT_CARD_INSERT;
401 	}
402 
403 	WR4(slot, SDHCI_INT_ENABLE, slot->intmask);
404 	WR4(slot, SDHCI_SIGNAL_ENABLE, slot->intmask);
405 }
406 
407 static void
sdhci_set_clock(struct sdhci_slot * slot,uint32_t clock)408 sdhci_set_clock(struct sdhci_slot *slot, uint32_t clock)
409 {
410 	uint32_t clk_base;
411 	uint32_t clk_sel;
412 	uint32_t res;
413 	uint16_t clk;
414 	uint16_t div;
415 	int timeout;
416 
417 	if (clock == slot->clock)
418 		return;
419 	clock = SDHCI_SET_CLOCK(slot->bus, slot, clock);
420 	slot->clock = clock;
421 
422 	/* Turn off the clock. */
423 	clk = RD2(slot, SDHCI_CLOCK_CONTROL);
424 	WR2(slot, SDHCI_CLOCK_CONTROL, clk & ~SDHCI_CLOCK_CARD_EN);
425 	/* If no clock requested - leave it so. */
426 	if (clock == 0)
427 		return;
428 
429 	/* Determine the clock base frequency */
430 	clk_base = slot->max_clk;
431 	if (slot->quirks & SDHCI_QUIRK_BCM577XX_400KHZ_CLKSRC) {
432 		clk_sel = RD2(slot, BCM577XX_HOST_CONTROL) &
433 		    BCM577XX_CTRL_CLKSEL_MASK;
434 
435 		/*
436 		 * Select clock source appropriate for the requested frequency.
437 		 */
438 		if ((clk_base / BCM577XX_DEFAULT_MAX_DIVIDER) > clock) {
439 			clk_base = BCM577XX_ALT_CLOCK_BASE;
440 			clk_sel |= (BCM577XX_CTRL_CLKSEL_64MHZ <<
441 			    BCM577XX_CTRL_CLKSEL_SHIFT);
442 		} else {
443 			clk_sel |= (BCM577XX_CTRL_CLKSEL_DEFAULT <<
444 			    BCM577XX_CTRL_CLKSEL_SHIFT);
445 		}
446 
447 		WR2(slot, BCM577XX_HOST_CONTROL, clk_sel);
448 	}
449 
450 	/* Recalculate timeout clock frequency based on the new sd clock. */
451 	if (slot->quirks & SDHCI_QUIRK_DATA_TIMEOUT_USES_SDCLK)
452 		slot->timeout_clk = slot->clock / 1000;
453 
454 	if (slot->version < SDHCI_SPEC_300) {
455 		/* Looking for highest freq <= clock. */
456 		res = clk_base;
457 		for (div = 1; div < SDHCI_200_MAX_DIVIDER; div <<= 1) {
458 			if (res <= clock)
459 				break;
460 			res >>= 1;
461 		}
462 		/* Divider 1:1 is 0x00, 2:1 is 0x01, 256:1 is 0x80 ... */
463 		div >>= 1;
464 	} else {
465 		/* Version 3.0 divisors are multiples of two up to 1023 * 2 */
466 		if (clock >= clk_base)
467 			div = 0;
468 		else {
469 			for (div = 2; div < SDHCI_300_MAX_DIVIDER; div += 2) {
470 				if ((clk_base / div) <= clock)
471 					break;
472 			}
473 		}
474 		div >>= 1;
475 	}
476 
477 	if (bootverbose || sdhci_debug)
478 		slot_printf(slot, "Divider %d for freq %d (base %d)\n",
479 			div, clock, clk_base);
480 
481 	/* Now we have got divider, set it. */
482 	clk = (div & SDHCI_DIVIDER_MASK) << SDHCI_DIVIDER_SHIFT;
483 	clk |= ((div >> SDHCI_DIVIDER_MASK_LEN) & SDHCI_DIVIDER_HI_MASK)
484 		<< SDHCI_DIVIDER_HI_SHIFT;
485 
486 	WR2(slot, SDHCI_CLOCK_CONTROL, clk);
487 	/* Enable clock. */
488 	clk |= SDHCI_CLOCK_INT_EN;
489 	WR2(slot, SDHCI_CLOCK_CONTROL, clk);
490 	/* Wait up to 10 ms until it stabilize. */
491 	timeout = 10;
492 	while (!((clk = RD2(slot, SDHCI_CLOCK_CONTROL))
493 		& SDHCI_CLOCK_INT_STABLE)) {
494 		if (timeout == 0) {
495 			slot_printf(slot,
496 			    "Internal clock never stabilised.\n");
497 			sdhci_dumpregs(slot);
498 			return;
499 		}
500 		timeout--;
501 		DELAY(1000);
502 	}
503 	/* Pass clock signal to the bus. */
504 	clk |= SDHCI_CLOCK_CARD_EN;
505 	WR2(slot, SDHCI_CLOCK_CONTROL, clk);
506 }
507 
508 static void
sdhci_set_power(struct sdhci_slot * slot,u_char power)509 sdhci_set_power(struct sdhci_slot *slot, u_char power)
510 {
511 	int i;
512 	uint8_t pwr;
513 
514 	if (slot->power == power)
515 		return;
516 
517 	slot->power = power;
518 
519 	/* Turn off the power. */
520 	pwr = 0;
521 	WR1(slot, SDHCI_POWER_CONTROL, pwr);
522 	/* If power down requested - leave it so. */
523 	if (power == 0)
524 		return;
525 	/* Set voltage. */
526 	switch (1 << power) {
527 	case MMC_OCR_LOW_VOLTAGE:
528 		pwr |= SDHCI_POWER_180;
529 		break;
530 	case MMC_OCR_290_300:
531 	case MMC_OCR_300_310:
532 		pwr |= SDHCI_POWER_300;
533 		break;
534 	case MMC_OCR_320_330:
535 	case MMC_OCR_330_340:
536 		pwr |= SDHCI_POWER_330;
537 		break;
538 	}
539 	WR1(slot, SDHCI_POWER_CONTROL, pwr);
540 	/*
541 	 * Turn on VDD1 power.  Note that at least some Intel controllers can
542 	 * fail to enable bus power on the first try after transiting from D3
543 	 * to D0, so we give them up to 2 ms.
544 	 */
545 	pwr |= SDHCI_POWER_ON;
546 	for (i = 0; i < 20; i++) {
547 		WR1(slot, SDHCI_POWER_CONTROL, pwr);
548 		if (RD1(slot, SDHCI_POWER_CONTROL) & SDHCI_POWER_ON)
549 			break;
550 		DELAY(100);
551 	}
552 	if (!(RD1(slot, SDHCI_POWER_CONTROL) & SDHCI_POWER_ON))
553 		slot_printf(slot, "Bus power failed to enable\n");
554 
555 	if (slot->quirks & SDHCI_QUIRK_INTEL_POWER_UP_RESET) {
556 		WR1(slot, SDHCI_POWER_CONTROL, pwr | 0x10);
557 		DELAY(10);
558 		WR1(slot, SDHCI_POWER_CONTROL, pwr);
559 		DELAY(300);
560 	}
561 }
562 
563 static void
sdhci_read_block_pio(struct sdhci_slot * slot)564 sdhci_read_block_pio(struct sdhci_slot *slot)
565 {
566 	uint32_t data;
567 	char *buffer;
568 	size_t left;
569 
570 	buffer = slot->curcmd->data->data;
571 	buffer += slot->offset;
572 	/* Transfer one block at a time. */
573 #ifdef MMCCAM
574 	if (slot->curcmd->data->flags & MMC_DATA_BLOCK_SIZE)
575 		left = min(slot->curcmd->data->block_size,
576 		    slot->curcmd->data->len - slot->offset);
577 	else
578 #endif
579 		left = min(512, slot->curcmd->data->len - slot->offset);
580 	slot->offset += left;
581 
582 	/* If we are too fast, broken controllers return zeroes. */
583 	if (slot->quirks & SDHCI_QUIRK_BROKEN_TIMINGS)
584 		DELAY(10);
585 	/* Handle unaligned and aligned buffer cases. */
586 	if ((intptr_t)buffer & 3) {
587 		while (left > 3) {
588 			data = RD4(slot, SDHCI_BUFFER);
589 			buffer[0] = data;
590 			buffer[1] = (data >> 8);
591 			buffer[2] = (data >> 16);
592 			buffer[3] = (data >> 24);
593 			buffer += 4;
594 			left -= 4;
595 		}
596 	} else {
597 		RD_MULTI_4(slot, SDHCI_BUFFER,
598 		    (uint32_t *)buffer, left >> 2);
599 		left &= 3;
600 	}
601 	/* Handle uneven size case. */
602 	if (left > 0) {
603 		data = RD4(slot, SDHCI_BUFFER);
604 		while (left > 0) {
605 			*(buffer++) = data;
606 			data >>= 8;
607 			left--;
608 		}
609 	}
610 }
611 
612 static void
sdhci_write_block_pio(struct sdhci_slot * slot)613 sdhci_write_block_pio(struct sdhci_slot *slot)
614 {
615 	uint32_t data = 0;
616 	char *buffer;
617 	size_t left;
618 
619 	buffer = slot->curcmd->data->data;
620 	buffer += slot->offset;
621 	/* Transfer one block at a time. */
622 #ifdef MMCCAM
623 	if (slot->curcmd->data->flags & MMC_DATA_BLOCK_SIZE) {
624 		left = min(slot->curcmd->data->block_size,
625 		    slot->curcmd->data->len - slot->offset);
626 	} else
627 #endif
628 		left = min(512, slot->curcmd->data->len - slot->offset);
629 	slot->offset += left;
630 
631 	/* Handle unaligned and aligned buffer cases. */
632 	if ((intptr_t)buffer & 3) {
633 		while (left > 3) {
634 			data = buffer[0] +
635 			    (buffer[1] << 8) +
636 			    (buffer[2] << 16) +
637 			    (buffer[3] << 24);
638 			left -= 4;
639 			buffer += 4;
640 			WR4(slot, SDHCI_BUFFER, data);
641 		}
642 	} else {
643 		WR_MULTI_4(slot, SDHCI_BUFFER,
644 		    (uint32_t *)buffer, left >> 2);
645 		left &= 3;
646 	}
647 	/* Handle uneven size case. */
648 	if (left > 0) {
649 		while (left > 0) {
650 			data <<= 8;
651 			data += *(buffer++);
652 			left--;
653 		}
654 		WR4(slot, SDHCI_BUFFER, data);
655 	}
656 }
657 
658 static void
sdhci_transfer_pio(struct sdhci_slot * slot)659 sdhci_transfer_pio(struct sdhci_slot *slot)
660 {
661 
662 	/* Read as many blocks as possible. */
663 	if (slot->curcmd->data->flags & MMC_DATA_READ) {
664 		while (RD4(slot, SDHCI_PRESENT_STATE) &
665 		    SDHCI_DATA_AVAILABLE) {
666 			sdhci_read_block_pio(slot);
667 			if (slot->offset >= slot->curcmd->data->len)
668 				break;
669 		}
670 	} else {
671 		while (RD4(slot, SDHCI_PRESENT_STATE) &
672 		    SDHCI_SPACE_AVAILABLE) {
673 			sdhci_write_block_pio(slot);
674 			if (slot->offset >= slot->curcmd->data->len)
675 				break;
676 		}
677 	}
678 }
679 
680 static void
sdhci_card_task(void * arg,int pending __unused)681 sdhci_card_task(void *arg, int pending __unused)
682 {
683 	struct sdhci_slot *slot = arg;
684 #ifndef MMCCAM
685 	device_t d;
686 #endif
687 
688 	SDHCI_LOCK(slot);
689 	if (SDHCI_GET_CARD_PRESENT(slot->bus, slot)) {
690 #ifdef MMCCAM
691 		if (slot->card_present == 0) {
692 #else
693 		if (slot->dev == NULL) {
694 #endif
695 			/* If card is present - attach mmc bus. */
696 			if (bootverbose || sdhci_debug)
697 				slot_printf(slot, "Card inserted\n");
698 #ifdef MMCCAM
699 			slot->card_present = 1;
700 			mmccam_start_discovery(slot->sim);
701 			SDHCI_UNLOCK(slot);
702 #else
703 			d = slot->dev = device_add_child(slot->bus, "mmc", -1);
704 			SDHCI_UNLOCK(slot);
705 			if (d) {
706 				device_set_ivars(d, slot);
707 				(void)device_probe_and_attach(d);
708 			}
709 #endif
710 		} else
711 			SDHCI_UNLOCK(slot);
712 	} else {
713 #ifdef MMCCAM
714 		if (slot->card_present == 1) {
715 #else
716 		if (slot->dev != NULL) {
717 			d = slot->dev;
718 #endif
719 			/* If no card present - detach mmc bus. */
720 			if (bootverbose || sdhci_debug)
721 				slot_printf(slot, "Card removed\n");
722 			slot->dev = NULL;
723 #ifdef MMCCAM
724 			slot->card_present = 0;
725 			mmccam_start_discovery(slot->sim);
726 			SDHCI_UNLOCK(slot);
727 #else
728 			slot->intmask &= ~sdhci_tuning_intmask(slot);
729 			WR4(slot, SDHCI_INT_ENABLE, slot->intmask);
730 			WR4(slot, SDHCI_SIGNAL_ENABLE, slot->intmask);
731 			slot->opt &= ~SDHCI_TUNING_ENABLED;
732 			SDHCI_UNLOCK(slot);
733 			callout_drain(&slot->retune_callout);
734 			device_delete_child(slot->bus, d);
735 #endif
736 		} else
737 			SDHCI_UNLOCK(slot);
738 	}
739 }
740 
741 static void
742 sdhci_handle_card_present_locked(struct sdhci_slot *slot, bool is_present)
743 {
744 	bool was_present;
745 
746 	/*
747 	 * If there was no card and now there is one, schedule the task to
748 	 * create the child device after a short delay.  The delay is to
749 	 * debounce the card insert (sometimes the card detect pin stabilizes
750 	 * before the other pins have made good contact).
751 	 *
752 	 * If there was a card present and now it's gone, immediately schedule
753 	 * the task to delete the child device.  No debouncing -- gone is gone,
754 	 * because once power is removed, a full card re-init is needed, and
755 	 * that happens by deleting and recreating the child device.
756 	 */
757 #ifdef MMCCAM
758 	was_present = slot->card_present;
759 #else
760 	was_present = slot->dev != NULL;
761 #endif
762 	if (!was_present && is_present) {
763 		taskqueue_enqueue_timeout(taskqueue_swi_giant,
764 		    &slot->card_delayed_task, -SDHCI_INSERT_DELAY_TICKS);
765 	} else if (was_present && !is_present) {
766 		taskqueue_enqueue(taskqueue_swi_giant, &slot->card_task);
767 	}
768 }
769 
770 void
771 sdhci_handle_card_present(struct sdhci_slot *slot, bool is_present)
772 {
773 
774 	SDHCI_LOCK(slot);
775 	sdhci_handle_card_present_locked(slot, is_present);
776 	SDHCI_UNLOCK(slot);
777 }
778 
779 static void
780 sdhci_card_poll(void *arg)
781 {
782 	struct sdhci_slot *slot = arg;
783 
784 	sdhci_handle_card_present(slot,
785 	    SDHCI_GET_CARD_PRESENT(slot->bus, slot));
786 	callout_reset(&slot->card_poll_callout, SDHCI_CARD_PRESENT_TICKS,
787 	    sdhci_card_poll, slot);
788 }
789 
790 static int
791 sdhci_dma_alloc(struct sdhci_slot *slot)
792 {
793 	int err;
794 
795 	if (!(slot->quirks & SDHCI_QUIRK_BROKEN_SDMA_BOUNDARY)) {
796 		if (maxphys <= 1024 * 4)
797 			slot->sdma_boundary = SDHCI_BLKSZ_SDMA_BNDRY_4K;
798 		else if (maxphys <= 1024 * 8)
799 			slot->sdma_boundary = SDHCI_BLKSZ_SDMA_BNDRY_8K;
800 		else if (maxphys <= 1024 * 16)
801 			slot->sdma_boundary = SDHCI_BLKSZ_SDMA_BNDRY_16K;
802 		else if (maxphys <= 1024 * 32)
803 			slot->sdma_boundary = SDHCI_BLKSZ_SDMA_BNDRY_32K;
804 		else if (maxphys <= 1024 * 64)
805 			slot->sdma_boundary = SDHCI_BLKSZ_SDMA_BNDRY_64K;
806 		else if (maxphys <= 1024 * 128)
807 			slot->sdma_boundary = SDHCI_BLKSZ_SDMA_BNDRY_128K;
808 		else if (maxphys <= 1024 * 256)
809 			slot->sdma_boundary = SDHCI_BLKSZ_SDMA_BNDRY_256K;
810 		else
811 			slot->sdma_boundary = SDHCI_BLKSZ_SDMA_BNDRY_512K;
812 	}
813 	slot->sdma_bbufsz = SDHCI_SDMA_BNDRY_TO_BBUFSZ(slot->sdma_boundary);
814 
815 	/*
816 	 * Allocate the DMA tag for an SDMA bounce buffer.
817 	 * Note that the SDHCI specification doesn't state any alignment
818 	 * constraint for the SDMA system address.  However, controllers
819 	 * typically ignore the SDMA boundary bits in SDHCI_DMA_ADDRESS when
820 	 * forming the actual address of data, requiring the SDMA buffer to
821 	 * be aligned to the SDMA boundary.
822 	 */
823 	err = bus_dma_tag_create(bus_get_dma_tag(slot->bus), slot->sdma_bbufsz,
824 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
825 	    slot->sdma_bbufsz, 1, slot->sdma_bbufsz, BUS_DMA_ALLOCNOW,
826 	    NULL, NULL, &slot->dmatag);
827 	if (err != 0) {
828 		slot_printf(slot, "Can't create DMA tag for SDMA\n");
829 		return (err);
830 	}
831 	/* Allocate DMA memory for the SDMA bounce buffer. */
832 	err = bus_dmamem_alloc(slot->dmatag, (void **)&slot->dmamem,
833 	    BUS_DMA_NOWAIT, &slot->dmamap);
834 	if (err != 0) {
835 		slot_printf(slot, "Can't alloc DMA memory for SDMA\n");
836 		bus_dma_tag_destroy(slot->dmatag);
837 		return (err);
838 	}
839 	/* Map the memory of the SDMA bounce buffer. */
840 	err = bus_dmamap_load(slot->dmatag, slot->dmamap,
841 	    (void *)slot->dmamem, slot->sdma_bbufsz, sdhci_getaddr,
842 	    &slot->paddr, 0);
843 	if (err != 0 || slot->paddr == 0) {
844 		slot_printf(slot, "Can't load DMA memory for SDMA\n");
845 		bus_dmamem_free(slot->dmatag, slot->dmamem, slot->dmamap);
846 		bus_dma_tag_destroy(slot->dmatag);
847 		if (err)
848 			return (err);
849 		else
850 			return (EFAULT);
851 	}
852 
853 	return (0);
854 }
855 
856 static void
857 sdhci_dma_free(struct sdhci_slot *slot)
858 {
859 
860 	bus_dmamap_unload(slot->dmatag, slot->dmamap);
861 	bus_dmamem_free(slot->dmatag, slot->dmamem, slot->dmamap);
862 	bus_dma_tag_destroy(slot->dmatag);
863 }
864 
865 int
866 sdhci_init_slot(device_t dev, struct sdhci_slot *slot, int num)
867 {
868 	kobjop_desc_t kobj_desc;
869 	kobj_method_t *kobj_method;
870 	uint32_t caps, caps2, freq, host_caps;
871 	int err;
872 	char node_name[8];
873 	struct sysctl_oid *node_oid;
874 
875 	SDHCI_LOCK_INIT(slot);
876 
877 	slot->num = num;
878 	slot->bus = dev;
879 
880 	slot->version = (RD2(slot, SDHCI_HOST_VERSION)
881 		>> SDHCI_SPEC_VER_SHIFT) & SDHCI_SPEC_VER_MASK;
882 	if (slot->quirks & SDHCI_QUIRK_MISSING_CAPS) {
883 		caps = slot->caps;
884 		caps2 = slot->caps2;
885 	} else {
886 		caps = RD4(slot, SDHCI_CAPABILITIES);
887 		if (slot->version >= SDHCI_SPEC_300)
888 			caps2 = RD4(slot, SDHCI_CAPABILITIES2);
889 		else
890 			caps2 = 0;
891 	}
892 	if (slot->version >= SDHCI_SPEC_300) {
893 		if ((caps & SDHCI_SLOTTYPE_MASK) != SDHCI_SLOTTYPE_REMOVABLE &&
894 		    (caps & SDHCI_SLOTTYPE_MASK) != SDHCI_SLOTTYPE_EMBEDDED) {
895 			slot_printf(slot,
896 			    "Driver doesn't support shared bus slots\n");
897 			SDHCI_LOCK_DESTROY(slot);
898 			return (ENXIO);
899 		} else if ((caps & SDHCI_SLOTTYPE_MASK) ==
900 		    SDHCI_SLOTTYPE_EMBEDDED) {
901 			slot->opt |= SDHCI_SLOT_EMBEDDED | SDHCI_NON_REMOVABLE;
902 		}
903 	}
904 	/* Calculate base clock frequency. */
905 	if (slot->version >= SDHCI_SPEC_300)
906 		freq = (caps & SDHCI_CLOCK_V3_BASE_MASK) >>
907 		    SDHCI_CLOCK_BASE_SHIFT;
908 	else
909 		freq = (caps & SDHCI_CLOCK_BASE_MASK) >>
910 		    SDHCI_CLOCK_BASE_SHIFT;
911 	if (freq != 0)
912 		slot->max_clk = freq * 1000000;
913 	/*
914 	 * If the frequency wasn't in the capabilities and the hardware driver
915 	 * hasn't already set max_clk we're probably not going to work right
916 	 * with an assumption, so complain about it.
917 	 */
918 	if (slot->max_clk == 0) {
919 		slot->max_clk = SDHCI_DEFAULT_MAX_FREQ * 1000000;
920 		slot_printf(slot, "Hardware doesn't specify base clock "
921 		    "frequency, using %dMHz as default.\n",
922 		    SDHCI_DEFAULT_MAX_FREQ);
923 	}
924 	/* Calculate/set timeout clock frequency. */
925 	if (slot->quirks & SDHCI_QUIRK_DATA_TIMEOUT_USES_SDCLK) {
926 		slot->timeout_clk = slot->max_clk / 1000;
927 	} else if (slot->quirks & SDHCI_QUIRK_DATA_TIMEOUT_1MHZ) {
928 		slot->timeout_clk = 1000;
929 	} else {
930 		slot->timeout_clk = (caps & SDHCI_TIMEOUT_CLK_MASK) >>
931 		    SDHCI_TIMEOUT_CLK_SHIFT;
932 		if (caps & SDHCI_TIMEOUT_CLK_UNIT)
933 			slot->timeout_clk *= 1000;
934 	}
935 	/*
936 	 * If the frequency wasn't in the capabilities and the hardware driver
937 	 * hasn't already set timeout_clk we'll probably work okay using the
938 	 * max timeout, but still mention it.
939 	 */
940 	if (slot->timeout_clk == 0) {
941 		slot_printf(slot, "Hardware doesn't specify timeout clock "
942 		    "frequency, setting BROKEN_TIMEOUT quirk.\n");
943 		slot->quirks |= SDHCI_QUIRK_BROKEN_TIMEOUT_VAL;
944 	}
945 
946 	slot->host.f_min = SDHCI_MIN_FREQ(slot->bus, slot);
947 	slot->host.f_max = slot->max_clk;
948 	slot->host.host_ocr = 0;
949 	if (caps & SDHCI_CAN_VDD_330)
950 	    slot->host.host_ocr |= MMC_OCR_320_330 | MMC_OCR_330_340;
951 	if (caps & SDHCI_CAN_VDD_300)
952 	    slot->host.host_ocr |= MMC_OCR_290_300 | MMC_OCR_300_310;
953 	/*
954 	 * 1.8V VDD is not supposed to be used for removable cards.  Hardware
955 	 * prior to v3.0 had no way to indicate embedded slots, but did
956 	 * sometimes support 1.8v for non-removable devices.
957 	 */
958 	if ((caps & SDHCI_CAN_VDD_180) && (slot->version < SDHCI_SPEC_300 ||
959 	    (slot->opt & SDHCI_SLOT_EMBEDDED)))
960 	    slot->host.host_ocr |= MMC_OCR_LOW_VOLTAGE;
961 	if (slot->host.host_ocr == 0) {
962 		slot_printf(slot, "Hardware doesn't report any "
963 		    "support voltages.\n");
964 	}
965 
966 	host_caps = slot->host.caps;
967 	host_caps |= MMC_CAP_4_BIT_DATA;
968 	if (caps & SDHCI_CAN_DO_8BITBUS)
969 		host_caps |= MMC_CAP_8_BIT_DATA;
970 	if (caps & SDHCI_CAN_DO_HISPD)
971 		host_caps |= MMC_CAP_HSPEED;
972 	if (slot->quirks & SDHCI_QUIRK_BOOT_NOACC)
973 		host_caps |= MMC_CAP_BOOT_NOACC;
974 	if (slot->quirks & SDHCI_QUIRK_WAIT_WHILE_BUSY)
975 		host_caps |= MMC_CAP_WAIT_WHILE_BUSY;
976 
977 	/* Determine supported UHS-I and eMMC modes. */
978 	if (caps2 & (SDHCI_CAN_SDR50 | SDHCI_CAN_SDR104 | SDHCI_CAN_DDR50))
979 		host_caps |= MMC_CAP_UHS_SDR12 | MMC_CAP_UHS_SDR25;
980 	if (caps2 & SDHCI_CAN_SDR104) {
981 		host_caps |= MMC_CAP_UHS_SDR104 | MMC_CAP_UHS_SDR50;
982 		if (!(slot->quirks & SDHCI_QUIRK_BROKEN_MMC_HS200))
983 			host_caps |= MMC_CAP_MMC_HS200;
984 	} else if (caps2 & SDHCI_CAN_SDR50)
985 		host_caps |= MMC_CAP_UHS_SDR50;
986 	if (caps2 & SDHCI_CAN_DDR50 &&
987 	    !(slot->quirks & SDHCI_QUIRK_BROKEN_UHS_DDR50))
988 		host_caps |= MMC_CAP_UHS_DDR50;
989 	if (slot->quirks & SDHCI_QUIRK_MMC_DDR52)
990 		host_caps |= MMC_CAP_MMC_DDR52;
991 	if (slot->quirks & SDHCI_QUIRK_CAPS_BIT63_FOR_MMC_HS400 &&
992 	    caps2 & SDHCI_CAN_MMC_HS400)
993 		host_caps |= MMC_CAP_MMC_HS400;
994 	if (slot->quirks & SDHCI_QUIRK_MMC_HS400_IF_CAN_SDR104 &&
995 	    caps2 & SDHCI_CAN_SDR104)
996 		host_caps |= MMC_CAP_MMC_HS400;
997 
998 	/*
999 	 * Disable UHS-I and eMMC modes if the set_uhs_timing method is the
1000 	 * default NULL implementation.
1001 	 */
1002 	kobj_desc = &sdhci_set_uhs_timing_desc;
1003 	kobj_method = kobj_lookup_method(((kobj_t)dev)->ops->cls, NULL,
1004 	    kobj_desc);
1005 	if (kobj_method == &kobj_desc->deflt)
1006 		host_caps &= ~(MMC_CAP_UHS_SDR12 | MMC_CAP_UHS_SDR25 |
1007 		    MMC_CAP_UHS_SDR50 | MMC_CAP_UHS_DDR50 | MMC_CAP_UHS_SDR104 |
1008 		    MMC_CAP_MMC_DDR52 | MMC_CAP_MMC_HS200 | MMC_CAP_MMC_HS400);
1009 
1010 #define	SDHCI_CAP_MODES_TUNING(caps2)					\
1011     (((caps2) & SDHCI_TUNE_SDR50 ? MMC_CAP_UHS_SDR50 : 0) |		\
1012     MMC_CAP_UHS_DDR50 | MMC_CAP_UHS_SDR104 | MMC_CAP_MMC_HS200 |	\
1013     MMC_CAP_MMC_HS400)
1014 
1015 	/*
1016 	 * Disable UHS-I and eMMC modes that require (re-)tuning if either
1017 	 * the tune or re-tune method is the default NULL implementation.
1018 	 */
1019 	kobj_desc = &mmcbr_tune_desc;
1020 	kobj_method = kobj_lookup_method(((kobj_t)dev)->ops->cls, NULL,
1021 	    kobj_desc);
1022 	if (kobj_method == &kobj_desc->deflt)
1023 		goto no_tuning;
1024 	kobj_desc = &mmcbr_retune_desc;
1025 	kobj_method = kobj_lookup_method(((kobj_t)dev)->ops->cls, NULL,
1026 	    kobj_desc);
1027 	if (kobj_method == &kobj_desc->deflt) {
1028 no_tuning:
1029 		host_caps &= ~(SDHCI_CAP_MODES_TUNING(caps2));
1030 	}
1031 
1032 	/* Allocate tuning structures and determine tuning parameters. */
1033 	if (host_caps & SDHCI_CAP_MODES_TUNING(caps2)) {
1034 		slot->opt |= SDHCI_TUNING_SUPPORTED;
1035 		slot->tune_req = malloc(sizeof(*slot->tune_req), M_DEVBUF,
1036 		    M_WAITOK);
1037 		slot->tune_cmd = malloc(sizeof(*slot->tune_cmd), M_DEVBUF,
1038 		    M_WAITOK);
1039 		slot->tune_data = malloc(sizeof(*slot->tune_data), M_DEVBUF,
1040 		    M_WAITOK);
1041 		if (caps2 & SDHCI_TUNE_SDR50)
1042 			slot->opt |= SDHCI_SDR50_NEEDS_TUNING;
1043 		slot->retune_mode = (caps2 & SDHCI_RETUNE_MODES_MASK) >>
1044 		    SDHCI_RETUNE_MODES_SHIFT;
1045 		if (slot->retune_mode == SDHCI_RETUNE_MODE_1) {
1046 			slot->retune_count = (caps2 & SDHCI_RETUNE_CNT_MASK) >>
1047 			    SDHCI_RETUNE_CNT_SHIFT;
1048 			if (slot->retune_count > 0xb) {
1049 				slot_printf(slot, "Unknown re-tuning count "
1050 				    "%x, using 1 sec\n", slot->retune_count);
1051 				slot->retune_count = 1;
1052 			} else if (slot->retune_count != 0)
1053 				slot->retune_count =
1054 				    1 << (slot->retune_count - 1);
1055 		}
1056 	}
1057 
1058 #undef SDHCI_CAP_MODES_TUNING
1059 
1060 	/* Determine supported VCCQ signaling levels. */
1061 	host_caps |= MMC_CAP_SIGNALING_330;
1062 	if (host_caps & (MMC_CAP_UHS_SDR12 | MMC_CAP_UHS_SDR25 |
1063 	    MMC_CAP_UHS_SDR50 | MMC_CAP_UHS_DDR50 | MMC_CAP_UHS_SDR104 |
1064 	    MMC_CAP_MMC_DDR52_180 | MMC_CAP_MMC_HS200_180 |
1065 	    MMC_CAP_MMC_HS400_180))
1066 		host_caps |= MMC_CAP_SIGNALING_120 | MMC_CAP_SIGNALING_180;
1067 
1068 	/*
1069 	 * Disable 1.2 V and 1.8 V signaling if the switch_vccq method is the
1070 	 * default NULL implementation.  Disable 1.2 V support if it's the
1071 	 * generic SDHCI implementation.
1072 	 */
1073 	kobj_desc = &mmcbr_switch_vccq_desc;
1074 	kobj_method = kobj_lookup_method(((kobj_t)dev)->ops->cls, NULL,
1075 	    kobj_desc);
1076 	if (kobj_method == &kobj_desc->deflt)
1077 		host_caps &= ~(MMC_CAP_SIGNALING_120 | MMC_CAP_SIGNALING_180);
1078 	else if (kobj_method->func == (kobjop_t)sdhci_generic_switch_vccq)
1079 		host_caps &= ~MMC_CAP_SIGNALING_120;
1080 
1081 	/* Determine supported driver types (type B is always mandatory). */
1082 	if (caps2 & SDHCI_CAN_DRIVE_TYPE_A)
1083 		host_caps |= MMC_CAP_DRIVER_TYPE_A;
1084 	if (caps2 & SDHCI_CAN_DRIVE_TYPE_C)
1085 		host_caps |= MMC_CAP_DRIVER_TYPE_C;
1086 	if (caps2 & SDHCI_CAN_DRIVE_TYPE_D)
1087 		host_caps |= MMC_CAP_DRIVER_TYPE_D;
1088 	slot->host.caps = host_caps;
1089 
1090 	/* Decide if we have usable DMA. */
1091 	if (caps & SDHCI_CAN_DO_DMA)
1092 		slot->opt |= SDHCI_HAVE_DMA;
1093 
1094 	if (slot->quirks & SDHCI_QUIRK_BROKEN_DMA)
1095 		slot->opt &= ~SDHCI_HAVE_DMA;
1096 	if (slot->quirks & SDHCI_QUIRK_FORCE_DMA)
1097 		slot->opt |= SDHCI_HAVE_DMA;
1098 	if (slot->quirks & SDHCI_QUIRK_ALL_SLOTS_NON_REMOVABLE)
1099 		slot->opt |= SDHCI_NON_REMOVABLE;
1100 
1101 	/*
1102 	 * Use platform-provided transfer backend
1103 	 * with PIO as a fallback mechanism
1104 	 */
1105 	if (slot->opt & SDHCI_PLATFORM_TRANSFER)
1106 		slot->opt &= ~SDHCI_HAVE_DMA;
1107 
1108 	if (slot->opt & SDHCI_HAVE_DMA) {
1109 		err = sdhci_dma_alloc(slot);
1110 		if (err != 0) {
1111 			if (slot->opt & SDHCI_TUNING_SUPPORTED) {
1112 				free(slot->tune_req, M_DEVBUF);
1113 				free(slot->tune_cmd, M_DEVBUF);
1114 				free(slot->tune_data, M_DEVBUF);
1115 			}
1116 			SDHCI_LOCK_DESTROY(slot);
1117 			return (err);
1118 		}
1119 	}
1120 
1121 	if (bootverbose || sdhci_debug) {
1122 		sdhci_dumpcaps(slot);
1123 		sdhci_dumpregs(slot);
1124 	}
1125 
1126 	slot->timeout = 10;
1127 	SYSCTL_ADD_INT(device_get_sysctl_ctx(slot->bus),
1128 	    SYSCTL_CHILDREN(device_get_sysctl_tree(slot->bus)), OID_AUTO,
1129 	    "timeout", CTLFLAG_RWTUN, &slot->timeout, 0,
1130 	    "Maximum timeout for SDHCI transfers (in secs)");
1131 	TASK_INIT(&slot->card_task, 0, sdhci_card_task, slot);
1132 	TIMEOUT_TASK_INIT(taskqueue_swi_giant, &slot->card_delayed_task, 0,
1133 		sdhci_card_task, slot);
1134 	callout_init(&slot->card_poll_callout, 1);
1135 	callout_init_mtx(&slot->timeout_callout, &slot->mtx, 0);
1136 	callout_init_mtx(&slot->retune_callout, &slot->mtx, 0);
1137 
1138 	if ((slot->quirks & SDHCI_QUIRK_POLL_CARD_PRESENT) &&
1139 	    !(slot->opt & SDHCI_NON_REMOVABLE)) {
1140 		callout_reset(&slot->card_poll_callout,
1141 		    SDHCI_CARD_PRESENT_TICKS, sdhci_card_poll, slot);
1142 	}
1143 
1144 	sdhci_init(slot);
1145 
1146 	snprintf(node_name, sizeof(node_name), "slot%d", slot->num);
1147 
1148 	node_oid = SYSCTL_ADD_NODE(device_get_sysctl_ctx(dev),
1149 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
1150 	    OID_AUTO, node_name, CTLFLAG_RW, 0, "slot specific node");
1151 
1152 	node_oid = SYSCTL_ADD_NODE(device_get_sysctl_ctx(dev),
1153 	    SYSCTL_CHILDREN(node_oid), OID_AUTO, "debug", CTLFLAG_RW, 0,
1154 	    "Debugging node");
1155 
1156 	SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev), SYSCTL_CHILDREN(node_oid),
1157 	    OID_AUTO, "dumpregs", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
1158 	    slot, 0, &sdhci_syctl_dumpregs,
1159 	    "A", "Dump SDHCI registers");
1160 
1161 	SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev), SYSCTL_CHILDREN(node_oid),
1162 	    OID_AUTO, "dumpcaps", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
1163 	    slot, 0, &sdhci_syctl_dumpcaps,
1164 	    "A", "Dump SDHCI capabilites");
1165 
1166 	return (0);
1167 }
1168 
1169 #ifndef MMCCAM
1170 void
1171 sdhci_start_slot(struct sdhci_slot *slot)
1172 {
1173 
1174 	sdhci_card_task(slot, 0);
1175 }
1176 #endif
1177 
1178 int
1179 sdhci_cleanup_slot(struct sdhci_slot *slot)
1180 {
1181 	device_t d;
1182 
1183 	callout_drain(&slot->timeout_callout);
1184 	callout_drain(&slot->card_poll_callout);
1185 	callout_drain(&slot->retune_callout);
1186 	taskqueue_drain(taskqueue_swi_giant, &slot->card_task);
1187 	taskqueue_drain_timeout(taskqueue_swi_giant, &slot->card_delayed_task);
1188 
1189 	SDHCI_LOCK(slot);
1190 	d = slot->dev;
1191 	slot->dev = NULL;
1192 	SDHCI_UNLOCK(slot);
1193 	if (d != NULL)
1194 		device_delete_child(slot->bus, d);
1195 
1196 	SDHCI_LOCK(slot);
1197 	SDHCI_RESET(slot->bus, slot, SDHCI_RESET_ALL);
1198 	SDHCI_UNLOCK(slot);
1199 	if (slot->opt & SDHCI_HAVE_DMA)
1200 		sdhci_dma_free(slot);
1201 	if (slot->opt & SDHCI_TUNING_SUPPORTED) {
1202 		free(slot->tune_req, M_DEVBUF);
1203 		free(slot->tune_cmd, M_DEVBUF);
1204 		free(slot->tune_data, M_DEVBUF);
1205 	}
1206 
1207 	SDHCI_LOCK_DESTROY(slot);
1208 
1209 	return (0);
1210 }
1211 
1212 int
1213 sdhci_generic_suspend(struct sdhci_slot *slot)
1214 {
1215 
1216 	/*
1217 	 * We expect the MMC layer to issue initial tuning after resume.
1218 	 * Otherwise, we'd need to indicate re-tuning including circuit reset
1219 	 * being required at least for re-tuning modes 1 and 2 ourselves.
1220 	 */
1221 	callout_drain(&slot->retune_callout);
1222 	SDHCI_LOCK(slot);
1223 	slot->opt &= ~SDHCI_TUNING_ENABLED;
1224 	SDHCI_RESET(slot->bus, slot, SDHCI_RESET_ALL);
1225 	SDHCI_UNLOCK(slot);
1226 
1227 	return (0);
1228 }
1229 
1230 int
1231 sdhci_generic_resume(struct sdhci_slot *slot)
1232 {
1233 
1234 	SDHCI_LOCK(slot);
1235 	sdhci_init(slot);
1236 	SDHCI_UNLOCK(slot);
1237 
1238 	return (0);
1239 }
1240 
1241 void
1242 sdhci_generic_reset(device_t brdev __unused, struct sdhci_slot *slot,
1243     uint8_t mask)
1244 {
1245 	int timeout;
1246 	uint32_t clock;
1247 
1248 	if (slot->quirks & SDHCI_QUIRK_NO_CARD_NO_RESET) {
1249 		if (!SDHCI_GET_CARD_PRESENT(slot->bus, slot))
1250 			return;
1251 	}
1252 
1253 	/* Some controllers need this kick or reset won't work. */
1254 	if ((mask & SDHCI_RESET_ALL) == 0 &&
1255 	    (slot->quirks & SDHCI_QUIRK_CLOCK_BEFORE_RESET)) {
1256 		/* This is to force an update */
1257 		clock = slot->clock;
1258 		slot->clock = 0;
1259 		sdhci_set_clock(slot, clock);
1260 	}
1261 
1262 	if (mask & SDHCI_RESET_ALL) {
1263 		slot->clock = 0;
1264 		slot->power = 0;
1265 	}
1266 
1267 	WR1(slot, SDHCI_SOFTWARE_RESET, mask);
1268 
1269 	if (slot->quirks & SDHCI_QUIRK_WAITFOR_RESET_ASSERTED) {
1270 		/*
1271 		 * Resets on TI OMAPs and AM335x are incompatible with SDHCI
1272 		 * specification.  The reset bit has internal propagation delay,
1273 		 * so a fast read after write returns 0 even if reset process is
1274 		 * in progress.  The workaround is to poll for 1 before polling
1275 		 * for 0.  In the worst case, if we miss seeing it asserted the
1276 		 * time we spent waiting is enough to ensure the reset finishes.
1277 		 */
1278 		timeout = 10000;
1279 		while ((RD1(slot, SDHCI_SOFTWARE_RESET) & mask) != mask) {
1280 			if (timeout <= 0)
1281 				break;
1282 			timeout--;
1283 			DELAY(1);
1284 		}
1285 	}
1286 
1287 	/* Wait max 100 ms */
1288 	timeout = 10000;
1289 	/* Controller clears the bits when it's done */
1290 	while (RD1(slot, SDHCI_SOFTWARE_RESET) & mask) {
1291 		if (timeout <= 0) {
1292 			slot_printf(slot, "Reset 0x%x never completed.\n",
1293 			    mask);
1294 			sdhci_dumpregs(slot);
1295 			return;
1296 		}
1297 		timeout--;
1298 		DELAY(10);
1299 	}
1300 }
1301 
1302 uint32_t
1303 sdhci_generic_min_freq(device_t brdev __unused, struct sdhci_slot *slot)
1304 {
1305 
1306 	if (slot->version >= SDHCI_SPEC_300)
1307 		return (slot->max_clk / SDHCI_300_MAX_DIVIDER);
1308 	else
1309 		return (slot->max_clk / SDHCI_200_MAX_DIVIDER);
1310 }
1311 
1312 bool
1313 sdhci_generic_get_card_present(device_t brdev __unused, struct sdhci_slot *slot)
1314 {
1315 
1316 	if (slot->opt & SDHCI_NON_REMOVABLE)
1317 		return true;
1318 
1319 	return (RD4(slot, SDHCI_PRESENT_STATE) & SDHCI_CARD_PRESENT);
1320 }
1321 
1322 void
1323 sdhci_generic_set_uhs_timing(device_t brdev __unused, struct sdhci_slot *slot)
1324 {
1325 	const struct mmc_ios *ios;
1326 	uint16_t hostctrl2;
1327 
1328 	if (slot->version < SDHCI_SPEC_300)
1329 		return;
1330 
1331 	SDHCI_ASSERT_LOCKED(slot);
1332 	ios = &slot->host.ios;
1333 	sdhci_set_clock(slot, 0);
1334 	hostctrl2 = RD2(slot, SDHCI_HOST_CONTROL2);
1335 	hostctrl2 &= ~SDHCI_CTRL2_UHS_MASK;
1336 	if (ios->clock > SD_SDR50_MAX) {
1337 		if (ios->timing == bus_timing_mmc_hs400 ||
1338 		    ios->timing == bus_timing_mmc_hs400es)
1339 			hostctrl2 |= SDHCI_CTRL2_MMC_HS400;
1340 		else
1341 			hostctrl2 |= SDHCI_CTRL2_UHS_SDR104;
1342 	}
1343 	else if (ios->clock > SD_SDR25_MAX)
1344 		hostctrl2 |= SDHCI_CTRL2_UHS_SDR50;
1345 	else if (ios->clock > SD_SDR12_MAX) {
1346 		if (ios->timing == bus_timing_uhs_ddr50 ||
1347 		    ios->timing == bus_timing_mmc_ddr52)
1348 			hostctrl2 |= SDHCI_CTRL2_UHS_DDR50;
1349 		else
1350 			hostctrl2 |= SDHCI_CTRL2_UHS_SDR25;
1351 	} else if (ios->clock > SD_MMC_CARD_ID_FREQUENCY)
1352 		hostctrl2 |= SDHCI_CTRL2_UHS_SDR12;
1353 	WR2(slot, SDHCI_HOST_CONTROL2, hostctrl2);
1354 	sdhci_set_clock(slot, ios->clock);
1355 }
1356 
1357 int
1358 sdhci_generic_update_ios(device_t brdev, device_t reqdev)
1359 {
1360 	struct sdhci_slot *slot = device_get_ivars(reqdev);
1361 	struct mmc_ios *ios = &slot->host.ios;
1362 
1363 	SDHCI_LOCK(slot);
1364 	/* Do full reset on bus power down to clear from any state. */
1365 	if (ios->power_mode == power_off) {
1366 		WR4(slot, SDHCI_SIGNAL_ENABLE, 0);
1367 		sdhci_init(slot);
1368 	}
1369 	/* Configure the bus. */
1370 	sdhci_set_clock(slot, ios->clock);
1371 	sdhci_set_power(slot, (ios->power_mode == power_off) ? 0 : ios->vdd);
1372 	if (ios->bus_width == bus_width_8) {
1373 		slot->hostctrl |= SDHCI_CTRL_8BITBUS;
1374 		slot->hostctrl &= ~SDHCI_CTRL_4BITBUS;
1375 	} else if (ios->bus_width == bus_width_4) {
1376 		slot->hostctrl &= ~SDHCI_CTRL_8BITBUS;
1377 		slot->hostctrl |= SDHCI_CTRL_4BITBUS;
1378 	} else if (ios->bus_width == bus_width_1) {
1379 		slot->hostctrl &= ~SDHCI_CTRL_8BITBUS;
1380 		slot->hostctrl &= ~SDHCI_CTRL_4BITBUS;
1381 	} else {
1382 		panic("Invalid bus width: %d", ios->bus_width);
1383 	}
1384 	if (ios->clock > SD_SDR12_MAX &&
1385 	    !(slot->quirks & SDHCI_QUIRK_DONT_SET_HISPD_BIT))
1386 		slot->hostctrl |= SDHCI_CTRL_HISPD;
1387 	else
1388 		slot->hostctrl &= ~SDHCI_CTRL_HISPD;
1389 	WR1(slot, SDHCI_HOST_CONTROL, slot->hostctrl);
1390 	SDHCI_SET_UHS_TIMING(brdev, slot);
1391 	/* Some controllers like reset after bus changes. */
1392 	if (slot->quirks & SDHCI_QUIRK_RESET_ON_IOS)
1393 		SDHCI_RESET(slot->bus, slot,
1394 		    SDHCI_RESET_CMD | SDHCI_RESET_DATA);
1395 
1396 	SDHCI_UNLOCK(slot);
1397 	return (0);
1398 }
1399 
1400 int
1401 sdhci_generic_switch_vccq(device_t brdev __unused, device_t reqdev)
1402 {
1403 	struct sdhci_slot *slot = device_get_ivars(reqdev);
1404 	enum mmc_vccq vccq;
1405 	int err;
1406 	uint16_t hostctrl2;
1407 
1408 	if (slot->version < SDHCI_SPEC_300)
1409 		return (0);
1410 
1411 	err = 0;
1412 	vccq = slot->host.ios.vccq;
1413 	SDHCI_LOCK(slot);
1414 	sdhci_set_clock(slot, 0);
1415 	hostctrl2 = RD2(slot, SDHCI_HOST_CONTROL2);
1416 	switch (vccq) {
1417 	case vccq_330:
1418 		if (!(hostctrl2 & SDHCI_CTRL2_S18_ENABLE))
1419 			goto done;
1420 		hostctrl2 &= ~SDHCI_CTRL2_S18_ENABLE;
1421 		WR2(slot, SDHCI_HOST_CONTROL2, hostctrl2);
1422 		DELAY(5000);
1423 		hostctrl2 = RD2(slot, SDHCI_HOST_CONTROL2);
1424 		if (!(hostctrl2 & SDHCI_CTRL2_S18_ENABLE))
1425 			goto done;
1426 		err = EAGAIN;
1427 		break;
1428 	case vccq_180:
1429 		if (!(slot->host.caps & MMC_CAP_SIGNALING_180)) {
1430 			err = EINVAL;
1431 			goto done;
1432 		}
1433 		if (hostctrl2 & SDHCI_CTRL2_S18_ENABLE)
1434 			goto done;
1435 		hostctrl2 |= SDHCI_CTRL2_S18_ENABLE;
1436 		WR2(slot, SDHCI_HOST_CONTROL2, hostctrl2);
1437 		DELAY(5000);
1438 		hostctrl2 = RD2(slot, SDHCI_HOST_CONTROL2);
1439 		if (hostctrl2 & SDHCI_CTRL2_S18_ENABLE)
1440 			goto done;
1441 		err = EAGAIN;
1442 		break;
1443 	default:
1444 		slot_printf(slot,
1445 		    "Attempt to set unsupported signaling voltage\n");
1446 		err = EINVAL;
1447 		break;
1448 	}
1449 done:
1450 	sdhci_set_clock(slot, slot->host.ios.clock);
1451 	SDHCI_UNLOCK(slot);
1452 	return (err);
1453 }
1454 
1455 int
1456 sdhci_generic_tune(device_t brdev __unused, device_t reqdev, bool hs400)
1457 {
1458 	struct sdhci_slot *slot = device_get_ivars(reqdev);
1459 	const struct mmc_ios *ios = &slot->host.ios;
1460 	struct mmc_command *tune_cmd;
1461 	struct mmc_data *tune_data;
1462 	uint32_t opcode;
1463 	int err;
1464 
1465 	if (!(slot->opt & SDHCI_TUNING_SUPPORTED))
1466 		return (0);
1467 
1468 	slot->retune_ticks = slot->retune_count * hz;
1469 	opcode = MMC_SEND_TUNING_BLOCK;
1470 	SDHCI_LOCK(slot);
1471 	switch (ios->timing) {
1472 	case bus_timing_mmc_hs400:
1473 		slot_printf(slot, "HS400 must be tuned in HS200 mode\n");
1474 		SDHCI_UNLOCK(slot);
1475 		return (EINVAL);
1476 	case bus_timing_mmc_hs200:
1477 		/*
1478 		 * In HS400 mode, controllers use the data strobe line to
1479 		 * latch data from the devices so periodic re-tuning isn't
1480 		 * expected to be required.
1481 		 */
1482 		if (hs400)
1483 			slot->retune_ticks = 0;
1484 		opcode = MMC_SEND_TUNING_BLOCK_HS200;
1485 		break;
1486 	case bus_timing_uhs_ddr50:
1487 	case bus_timing_uhs_sdr104:
1488 		break;
1489 	case bus_timing_uhs_sdr50:
1490 		if (slot->opt & SDHCI_SDR50_NEEDS_TUNING)
1491 			break;
1492 		SDHCI_UNLOCK(slot);
1493 		return (0);
1494 	default:
1495 		slot_printf(slot, "Tuning requested but not required.\n");
1496 		SDHCI_UNLOCK(slot);
1497 		return (EINVAL);
1498 	}
1499 
1500 	tune_cmd = slot->tune_cmd;
1501 	memset(tune_cmd, 0, sizeof(*tune_cmd));
1502 	tune_cmd->opcode = opcode;
1503 	tune_cmd->flags = MMC_RSP_R1 | MMC_CMD_ADTC;
1504 	tune_data = tune_cmd->data = slot->tune_data;
1505 	memset(tune_data, 0, sizeof(*tune_data));
1506 	tune_data->len = (opcode == MMC_SEND_TUNING_BLOCK_HS200 &&
1507 	    ios->bus_width == bus_width_8) ? MMC_TUNING_LEN_HS200 :
1508 	    MMC_TUNING_LEN;
1509 	tune_data->flags = MMC_DATA_READ;
1510 	tune_data->mrq = tune_cmd->mrq = slot->tune_req;
1511 
1512 	slot->opt &= ~SDHCI_TUNING_ENABLED;
1513 	err = sdhci_exec_tuning(slot, true);
1514 	if (err == 0) {
1515 		slot->opt |= SDHCI_TUNING_ENABLED;
1516 		slot->intmask |= sdhci_tuning_intmask(slot);
1517 		WR4(slot, SDHCI_INT_ENABLE, slot->intmask);
1518 		WR4(slot, SDHCI_SIGNAL_ENABLE, slot->intmask);
1519 		if (slot->retune_ticks) {
1520 			callout_reset(&slot->retune_callout, slot->retune_ticks,
1521 			    sdhci_retune, slot);
1522 		}
1523 	}
1524 	SDHCI_UNLOCK(slot);
1525 	return (err);
1526 }
1527 
1528 int
1529 sdhci_generic_retune(device_t brdev __unused, device_t reqdev, bool reset)
1530 {
1531 	struct sdhci_slot *slot = device_get_ivars(reqdev);
1532 	int err;
1533 
1534 	if (!(slot->opt & SDHCI_TUNING_ENABLED))
1535 		return (0);
1536 
1537 	/* HS400 must be tuned in HS200 mode. */
1538 	if (slot->host.ios.timing == bus_timing_mmc_hs400)
1539 		return (EINVAL);
1540 
1541 	SDHCI_LOCK(slot);
1542 	err = sdhci_exec_tuning(slot, reset);
1543 	/*
1544 	 * There are two ways sdhci_exec_tuning() can fail:
1545 	 * EBUSY should not actually happen when requests are only issued
1546 	 *	 with the host properly acquired, and
1547 	 * EIO   re-tuning failed (but it did work initially).
1548 	 *
1549 	 * In both cases, we should retry at later point if periodic re-tuning
1550 	 * is enabled.  Note that due to slot->retune_req not being cleared in
1551 	 * these failure cases, the MMC layer should trigger another attempt at
1552 	 * re-tuning with the next request anyway, though.
1553 	 */
1554 	if (slot->retune_ticks) {
1555 		callout_reset(&slot->retune_callout, slot->retune_ticks,
1556 		    sdhci_retune, slot);
1557 	}
1558 	SDHCI_UNLOCK(slot);
1559 	return (err);
1560 }
1561 
1562 static int
1563 sdhci_exec_tuning(struct sdhci_slot *slot, bool reset)
1564 {
1565 	struct mmc_request *tune_req;
1566 	struct mmc_command *tune_cmd;
1567 	int i;
1568 	uint32_t intmask;
1569 	uint16_t hostctrl2;
1570 	u_char opt;
1571 
1572 	SDHCI_ASSERT_LOCKED(slot);
1573 	if (slot->req != NULL)
1574 		return (EBUSY);
1575 
1576 	/* Tuning doesn't work with DMA enabled. */
1577 	opt = slot->opt;
1578 	slot->opt = opt & ~SDHCI_HAVE_DMA;
1579 
1580 	/*
1581 	 * Ensure that as documented, SDHCI_INT_DATA_AVAIL is the only
1582 	 * kind of interrupt we receive in response to a tuning request.
1583 	 */
1584 	intmask = slot->intmask;
1585 	slot->intmask = SDHCI_INT_DATA_AVAIL;
1586 	WR4(slot, SDHCI_INT_ENABLE, SDHCI_INT_DATA_AVAIL);
1587 	WR4(slot, SDHCI_SIGNAL_ENABLE, SDHCI_INT_DATA_AVAIL);
1588 
1589 	hostctrl2 = RD2(slot, SDHCI_HOST_CONTROL2);
1590 	if (reset)
1591 		hostctrl2 &= ~SDHCI_CTRL2_SAMPLING_CLOCK;
1592 	else
1593 		hostctrl2 |= SDHCI_CTRL2_SAMPLING_CLOCK;
1594 	WR2(slot, SDHCI_HOST_CONTROL2, hostctrl2 | SDHCI_CTRL2_EXEC_TUNING);
1595 
1596 	tune_req = slot->tune_req;
1597 	tune_cmd = slot->tune_cmd;
1598 	for (i = 0; i < MMC_TUNING_MAX; i++) {
1599 		memset(tune_req, 0, sizeof(*tune_req));
1600 		tune_req->cmd = tune_cmd;
1601 		tune_req->done = sdhci_req_wakeup;
1602 		tune_req->done_data = slot;
1603 		slot->req = tune_req;
1604 		slot->flags = 0;
1605 		sdhci_start(slot);
1606 		while (!(tune_req->flags & MMC_REQ_DONE))
1607 			msleep(tune_req, &slot->mtx, 0, "sdhciet", 0);
1608 		if (!(tune_req->flags & MMC_TUNE_DONE))
1609 			break;
1610 		hostctrl2 = RD2(slot, SDHCI_HOST_CONTROL2);
1611 		if (!(hostctrl2 & SDHCI_CTRL2_EXEC_TUNING))
1612 			break;
1613 		if (tune_cmd->opcode == MMC_SEND_TUNING_BLOCK)
1614 			DELAY(1000);
1615 	}
1616 
1617 	/*
1618 	 * Restore DMA usage and interrupts.
1619 	 * Note that the interrupt aggregation code might have cleared
1620 	 * SDHCI_INT_DMA_END and/or SDHCI_INT_RESPONSE in slot->intmask
1621 	 * and SDHCI_SIGNAL_ENABLE respectively so ensure SDHCI_INT_ENABLE
1622 	 * doesn't lose these.
1623 	 */
1624 	slot->opt = opt;
1625 	slot->intmask = intmask;
1626 	WR4(slot, SDHCI_INT_ENABLE, intmask | SDHCI_INT_DMA_END |
1627 	    SDHCI_INT_RESPONSE);
1628 	WR4(slot, SDHCI_SIGNAL_ENABLE, intmask);
1629 
1630 	if ((hostctrl2 & (SDHCI_CTRL2_EXEC_TUNING |
1631 	    SDHCI_CTRL2_SAMPLING_CLOCK)) == SDHCI_CTRL2_SAMPLING_CLOCK) {
1632 		slot->retune_req = 0;
1633 		return (0);
1634 	}
1635 
1636 	slot_printf(slot, "Tuning failed, using fixed sampling clock\n");
1637 	WR2(slot, SDHCI_HOST_CONTROL2, hostctrl2 & ~(SDHCI_CTRL2_EXEC_TUNING |
1638 	    SDHCI_CTRL2_SAMPLING_CLOCK));
1639 	SDHCI_RESET(slot->bus, slot, SDHCI_RESET_CMD | SDHCI_RESET_DATA);
1640 	return (EIO);
1641 }
1642 
1643 static void
1644 sdhci_retune(void *arg)
1645 {
1646 	struct sdhci_slot *slot = arg;
1647 
1648 	slot->retune_req |= SDHCI_RETUNE_REQ_NEEDED;
1649 }
1650 
1651 #ifdef MMCCAM
1652 static void
1653 sdhci_req_done(struct sdhci_slot *slot)
1654 {
1655 	union ccb *ccb;
1656 
1657 	if (__predict_false(sdhci_debug > 1))
1658 		slot_printf(slot, "%s\n", __func__);
1659 	if (slot->ccb != NULL && slot->curcmd != NULL) {
1660 		callout_stop(&slot->timeout_callout);
1661 		ccb = slot->ccb;
1662 		slot->ccb = NULL;
1663 		slot->curcmd = NULL;
1664 
1665 		/* Tell CAM the request is finished */
1666 		struct ccb_mmcio *mmcio;
1667 		mmcio = &ccb->mmcio;
1668 
1669 		ccb->ccb_h.status =
1670 		    (mmcio->cmd.error == 0 ? CAM_REQ_CMP : CAM_REQ_CMP_ERR);
1671 		xpt_done(ccb);
1672 	}
1673 }
1674 #else
1675 static void
1676 sdhci_req_done(struct sdhci_slot *slot)
1677 {
1678 	struct mmc_request *req;
1679 
1680 	if (slot->req != NULL && slot->curcmd != NULL) {
1681 		callout_stop(&slot->timeout_callout);
1682 		req = slot->req;
1683 		slot->req = NULL;
1684 		slot->curcmd = NULL;
1685 		req->done(req);
1686 	}
1687 }
1688 #endif
1689 
1690 static void
1691 sdhci_req_wakeup(struct mmc_request *req)
1692 {
1693 
1694 	req->flags |= MMC_REQ_DONE;
1695 	wakeup(req);
1696 }
1697 
1698 static void
1699 sdhci_timeout(void *arg)
1700 {
1701 	struct sdhci_slot *slot = arg;
1702 
1703 	if (slot->curcmd != NULL) {
1704 		slot_printf(slot, "Controller timeout\n");
1705 		sdhci_dumpregs(slot);
1706 		SDHCI_RESET(slot->bus, slot,
1707 		    SDHCI_RESET_CMD | SDHCI_RESET_DATA);
1708 		slot->curcmd->error = MMC_ERR_TIMEOUT;
1709 		sdhci_req_done(slot);
1710 	} else {
1711 		slot_printf(slot, "Spurious timeout - no active command\n");
1712 	}
1713 }
1714 
1715 static void
1716 sdhci_set_transfer_mode(struct sdhci_slot *slot, const struct mmc_data *data)
1717 {
1718 	uint16_t mode;
1719 
1720 	if (data == NULL)
1721 		return;
1722 
1723 	mode = SDHCI_TRNS_BLK_CNT_EN;
1724 	if (data->len > 512 || data->block_count > 1) {
1725 		mode |= SDHCI_TRNS_MULTI;
1726 		if (data->block_count == 0 && __predict_true(
1727 #ifdef MMCCAM
1728 		    slot->ccb->mmcio.stop.opcode == MMC_STOP_TRANSMISSION &&
1729 #else
1730 		    slot->req->stop != NULL &&
1731 #endif
1732 		    !(slot->quirks & SDHCI_QUIRK_BROKEN_AUTO_STOP)))
1733 			mode |= SDHCI_TRNS_ACMD12;
1734 	}
1735 	if (data->flags & MMC_DATA_READ)
1736 		mode |= SDHCI_TRNS_READ;
1737 	if (slot->flags & SDHCI_USE_DMA)
1738 		mode |= SDHCI_TRNS_DMA;
1739 
1740 	WR2(slot, SDHCI_TRANSFER_MODE, mode);
1741 }
1742 
1743 static void
1744 sdhci_start_command(struct sdhci_slot *slot, struct mmc_command *cmd)
1745 {
1746 	int flags, timeout;
1747 	uint32_t mask;
1748 
1749 	slot->curcmd = cmd;
1750 	slot->cmd_done = 0;
1751 
1752 	cmd->error = MMC_ERR_NONE;
1753 
1754 	/* This flags combination is not supported by controller. */
1755 	if ((cmd->flags & MMC_RSP_136) && (cmd->flags & MMC_RSP_BUSY)) {
1756 		slot_printf(slot, "Unsupported response type!\n");
1757 		cmd->error = MMC_ERR_FAILED;
1758 		sdhci_req_done(slot);
1759 		return;
1760 	}
1761 
1762 	/*
1763 	 * Do not issue command if there is no card, clock or power.
1764 	 * Controller will not detect timeout without clock active.
1765 	 */
1766 	if (!SDHCI_GET_CARD_PRESENT(slot->bus, slot) ||
1767 	    slot->power == 0 ||
1768 	    slot->clock == 0) {
1769 		slot_printf(slot,
1770 			    "Cannot issue a command (power=%d clock=%d)\n",
1771 			    slot->power, slot->clock);
1772 		cmd->error = MMC_ERR_FAILED;
1773 		sdhci_req_done(slot);
1774 		return;
1775 	}
1776 	/* Always wait for free CMD bus. */
1777 	mask = SDHCI_CMD_INHIBIT;
1778 	/* Wait for free DAT if we have data or busy signal. */
1779 	if (cmd->data != NULL || (cmd->flags & MMC_RSP_BUSY))
1780 		mask |= SDHCI_DAT_INHIBIT;
1781 	/*
1782 	 * We shouldn't wait for DAT for stop commands or CMD19/CMD21.  Note
1783 	 * that these latter are also special in that SDHCI_CMD_DATA should
1784 	 * be set below but no actual data is ever read from the controller.
1785 	*/
1786 #ifdef MMCCAM
1787 	if (cmd == &slot->ccb->mmcio.stop ||
1788 #else
1789 	if (cmd == slot->req->stop ||
1790 #endif
1791 	    __predict_false(cmd->opcode == MMC_SEND_TUNING_BLOCK ||
1792 	    cmd->opcode == MMC_SEND_TUNING_BLOCK_HS200))
1793 		mask &= ~SDHCI_DAT_INHIBIT;
1794 	/*
1795 	 *  Wait for bus no more then 250 ms.  Typically there will be no wait
1796 	 *  here at all, but when writing a crash dump we may be bypassing the
1797 	 *  host platform's interrupt handler, and in some cases that handler
1798 	 *  may be working around hardware quirks such as not respecting r1b
1799 	 *  busy indications.  In those cases, this wait-loop serves the purpose
1800 	 *  of waiting for the prior command and data transfers to be done, and
1801 	 *  SD cards are allowed to take up to 250ms for write and erase ops.
1802 	 *  (It's usually more like 20-30ms in the real world.)
1803 	 */
1804 	timeout = 250;
1805 	while (mask & RD4(slot, SDHCI_PRESENT_STATE)) {
1806 		if (timeout == 0) {
1807 			slot_printf(slot, "Controller never released "
1808 			    "inhibit bit(s).\n");
1809 			sdhci_dumpregs(slot);
1810 			cmd->error = MMC_ERR_FAILED;
1811 			sdhci_req_done(slot);
1812 			return;
1813 		}
1814 		timeout--;
1815 		DELAY(1000);
1816 	}
1817 
1818 	/* Prepare command flags. */
1819 	if (!(cmd->flags & MMC_RSP_PRESENT))
1820 		flags = SDHCI_CMD_RESP_NONE;
1821 	else if (cmd->flags & MMC_RSP_136)
1822 		flags = SDHCI_CMD_RESP_LONG;
1823 	else if (cmd->flags & MMC_RSP_BUSY)
1824 		flags = SDHCI_CMD_RESP_SHORT_BUSY;
1825 	else
1826 		flags = SDHCI_CMD_RESP_SHORT;
1827 	if (cmd->flags & MMC_RSP_CRC)
1828 		flags |= SDHCI_CMD_CRC;
1829 	if (cmd->flags & MMC_RSP_OPCODE)
1830 		flags |= SDHCI_CMD_INDEX;
1831 	if (cmd->data != NULL)
1832 		flags |= SDHCI_CMD_DATA;
1833 	if (cmd->opcode == MMC_STOP_TRANSMISSION)
1834 		flags |= SDHCI_CMD_TYPE_ABORT;
1835 	/* Prepare data. */
1836 	sdhci_start_data(slot, cmd->data);
1837 	/*
1838 	 * Interrupt aggregation: To reduce total number of interrupts
1839 	 * group response interrupt with data interrupt when possible.
1840 	 * If there going to be data interrupt, mask response one.
1841 	 */
1842 	if (slot->data_done == 0) {
1843 		WR4(slot, SDHCI_SIGNAL_ENABLE,
1844 		    slot->intmask &= ~SDHCI_INT_RESPONSE);
1845 	}
1846 	/* Set command argument. */
1847 	WR4(slot, SDHCI_ARGUMENT, cmd->arg);
1848 	/* Set data transfer mode. */
1849 	sdhci_set_transfer_mode(slot, cmd->data);
1850 	if (__predict_false(sdhci_debug > 1))
1851 		slot_printf(slot, "Starting command opcode %#04x flags %#04x\n",
1852 		    cmd->opcode, flags);
1853 
1854 	/* Start command. */
1855 	WR2(slot, SDHCI_COMMAND_FLAGS, (cmd->opcode << 8) | (flags & 0xff));
1856 	/* Start timeout callout. */
1857 	callout_reset(&slot->timeout_callout, slot->timeout * hz,
1858 	    sdhci_timeout, slot);
1859 }
1860 
1861 static void
1862 sdhci_finish_command(struct sdhci_slot *slot)
1863 {
1864 	int i;
1865 	uint32_t val;
1866 	uint8_t extra;
1867 
1868 	if (__predict_false(sdhci_debug > 1))
1869 		slot_printf(slot, "%s: called, err %d flags %#04x\n",
1870 		    __func__, slot->curcmd->error, slot->curcmd->flags);
1871 	slot->cmd_done = 1;
1872 	/*
1873 	 * Interrupt aggregation: Restore command interrupt.
1874 	 * Main restore point for the case when command interrupt
1875 	 * happened first.
1876 	 */
1877 	if (__predict_true(slot->curcmd->opcode != MMC_SEND_TUNING_BLOCK &&
1878 	    slot->curcmd->opcode != MMC_SEND_TUNING_BLOCK_HS200))
1879 		WR4(slot, SDHCI_SIGNAL_ENABLE, slot->intmask |=
1880 		    SDHCI_INT_RESPONSE);
1881 	/* In case of error - reset host and return. */
1882 	if (slot->curcmd->error) {
1883 		if (slot->curcmd->error == MMC_ERR_BADCRC)
1884 			slot->retune_req |= SDHCI_RETUNE_REQ_RESET;
1885 		SDHCI_RESET(slot->bus, slot, SDHCI_RESET_CMD);
1886 		SDHCI_RESET(slot->bus, slot, SDHCI_RESET_DATA);
1887 		sdhci_start(slot);
1888 		return;
1889 	}
1890 	/* If command has response - fetch it. */
1891 	if (slot->curcmd->flags & MMC_RSP_PRESENT) {
1892 		if (slot->curcmd->flags & MMC_RSP_136) {
1893 			/* CRC is stripped so we need one byte shift. */
1894 			extra = 0;
1895 			for (i = 0; i < 4; i++) {
1896 				val = RD4(slot, SDHCI_RESPONSE + i * 4);
1897 				if (slot->quirks &
1898 				    SDHCI_QUIRK_DONT_SHIFT_RESPONSE)
1899 					slot->curcmd->resp[3 - i] = val;
1900 				else {
1901 					slot->curcmd->resp[3 - i] =
1902 					    (val << 8) | extra;
1903 					extra = val >> 24;
1904 				}
1905 			}
1906 		} else
1907 			slot->curcmd->resp[0] = RD4(slot, SDHCI_RESPONSE);
1908 	}
1909 	if (__predict_false(sdhci_debug > 1))
1910 		slot_printf(slot, "Resp: %#04x %#04x %#04x %#04x\n",
1911 		    slot->curcmd->resp[0], slot->curcmd->resp[1],
1912 		    slot->curcmd->resp[2], slot->curcmd->resp[3]);
1913 
1914 	/* If data ready - finish. */
1915 	if (slot->data_done)
1916 		sdhci_start(slot);
1917 }
1918 
1919 static void
1920 sdhci_start_data(struct sdhci_slot *slot, const struct mmc_data *data)
1921 {
1922 	uint32_t blkcnt, blksz, current_timeout, sdma_bbufsz, target_timeout;
1923 	uint8_t div;
1924 
1925 	if (data == NULL && (slot->curcmd->flags & MMC_RSP_BUSY) == 0) {
1926 		slot->data_done = 1;
1927 		return;
1928 	}
1929 
1930 	slot->data_done = 0;
1931 
1932 	/* Calculate and set data timeout.*/
1933 	/* XXX: We should have this from mmc layer, now assume 1 sec. */
1934 	if (slot->quirks & SDHCI_QUIRK_BROKEN_TIMEOUT_VAL) {
1935 		div = 0xE;
1936 	} else {
1937 		target_timeout = 1000000;
1938 		div = 0;
1939 		current_timeout = (1 << 13) * 1000 / slot->timeout_clk;
1940 		while (current_timeout < target_timeout && div < 0xE) {
1941 			++div;
1942 			current_timeout <<= 1;
1943 		}
1944 		/* Compensate for an off-by-one error in the CaFe chip.*/
1945 		if (div < 0xE &&
1946 		    (slot->quirks & SDHCI_QUIRK_INCR_TIMEOUT_CONTROL)) {
1947 			++div;
1948 		}
1949 	}
1950 	WR1(slot, SDHCI_TIMEOUT_CONTROL, div);
1951 
1952 	if (data == NULL)
1953 		return;
1954 
1955 	/* Use DMA if possible. */
1956 	if ((slot->opt & SDHCI_HAVE_DMA))
1957 		slot->flags |= SDHCI_USE_DMA;
1958 	/* If data is small, broken DMA may return zeroes instead of data. */
1959 	if ((slot->quirks & SDHCI_QUIRK_BROKEN_TIMINGS) &&
1960 	    (data->len <= 512))
1961 		slot->flags &= ~SDHCI_USE_DMA;
1962 	/* Some controllers require even block sizes. */
1963 	if ((slot->quirks & SDHCI_QUIRK_32BIT_DMA_SIZE) &&
1964 	    ((data->len) & 0x3))
1965 		slot->flags &= ~SDHCI_USE_DMA;
1966 	/* Load DMA buffer. */
1967 	if (slot->flags & SDHCI_USE_DMA) {
1968 		sdma_bbufsz = slot->sdma_bbufsz;
1969 		if (data->flags & MMC_DATA_READ)
1970 			bus_dmamap_sync(slot->dmatag, slot->dmamap,
1971 			    BUS_DMASYNC_PREREAD);
1972 		else {
1973 			memcpy(slot->dmamem, data->data, ulmin(data->len,
1974 			    sdma_bbufsz));
1975 			bus_dmamap_sync(slot->dmatag, slot->dmamap,
1976 			    BUS_DMASYNC_PREWRITE);
1977 		}
1978 		WR4(slot, SDHCI_DMA_ADDRESS, slot->paddr);
1979 		/*
1980 		 * Interrupt aggregation: Mask border interrupt for the last
1981 		 * bounce buffer and unmask otherwise.
1982 		 */
1983 		if (data->len == sdma_bbufsz)
1984 			slot->intmask &= ~SDHCI_INT_DMA_END;
1985 		else
1986 			slot->intmask |= SDHCI_INT_DMA_END;
1987 		WR4(slot, SDHCI_SIGNAL_ENABLE, slot->intmask);
1988 	}
1989 	/* Current data offset for both PIO and DMA. */
1990 	slot->offset = 0;
1991 #ifdef MMCCAM
1992 	if (data->flags & MMC_DATA_BLOCK_SIZE) {
1993 		/* Set block size and request border interrupts on the SDMA boundary. */
1994 		blksz = SDHCI_MAKE_BLKSZ(slot->sdma_boundary, data->block_size);
1995 		blkcnt = data->block_count;
1996 		if (__predict_false(sdhci_debug > 0))
1997 			slot_printf(slot, "SDIO Custom block params: blksz: "
1998 			    "%#10x, blk cnt: %#10x\n", blksz, blkcnt);
1999 	} else
2000 #endif
2001 	{
2002 		/* Set block size and request border interrupts on the SDMA boundary. */
2003 		blksz = SDHCI_MAKE_BLKSZ(slot->sdma_boundary, ulmin(data->len, 512));
2004 		blkcnt = howmany(data->len, 512);
2005 	}
2006 
2007 	WR2(slot, SDHCI_BLOCK_SIZE, blksz);
2008 	WR2(slot, SDHCI_BLOCK_COUNT, blkcnt);
2009 	if (__predict_false(sdhci_debug > 1))
2010 		slot_printf(slot, "Blk size: 0x%08x | Blk cnt:  0x%08x\n",
2011 		    blksz, blkcnt);
2012 }
2013 
2014 void
2015 sdhci_finish_data(struct sdhci_slot *slot)
2016 {
2017 	struct mmc_data *data = slot->curcmd->data;
2018 	size_t left;
2019 
2020 	/* Interrupt aggregation: Restore command interrupt.
2021 	 * Auxiliary restore point for the case when data interrupt
2022 	 * happened first. */
2023 	if (!slot->cmd_done) {
2024 		WR4(slot, SDHCI_SIGNAL_ENABLE,
2025 		    slot->intmask |= SDHCI_INT_RESPONSE);
2026 	}
2027 	/* Unload rest of data from DMA buffer. */
2028 	if (!slot->data_done && (slot->flags & SDHCI_USE_DMA) &&
2029 	    slot->curcmd->data != NULL) {
2030 		if (data->flags & MMC_DATA_READ) {
2031 			left = data->len - slot->offset;
2032 			bus_dmamap_sync(slot->dmatag, slot->dmamap,
2033 			    BUS_DMASYNC_POSTREAD);
2034 			memcpy((u_char*)data->data + slot->offset, slot->dmamem,
2035 			    ulmin(left, slot->sdma_bbufsz));
2036 		} else
2037 			bus_dmamap_sync(slot->dmatag, slot->dmamap,
2038 			    BUS_DMASYNC_POSTWRITE);
2039 	}
2040 	slot->data_done = 1;
2041 	/* If there was error - reset the host. */
2042 	if (slot->curcmd->error) {
2043 		if (slot->curcmd->error == MMC_ERR_BADCRC)
2044 			slot->retune_req |= SDHCI_RETUNE_REQ_RESET;
2045 		SDHCI_RESET(slot->bus, slot, SDHCI_RESET_CMD);
2046 		SDHCI_RESET(slot->bus, slot, SDHCI_RESET_DATA);
2047 		sdhci_start(slot);
2048 		return;
2049 	}
2050 	/* If we already have command response - finish. */
2051 	if (slot->cmd_done)
2052 		sdhci_start(slot);
2053 }
2054 
2055 #ifdef MMCCAM
2056 static void
2057 sdhci_start(struct sdhci_slot *slot)
2058 {
2059 	union ccb *ccb;
2060 	struct ccb_mmcio *mmcio;
2061 
2062 	ccb = slot->ccb;
2063 	if (ccb == NULL)
2064 		return;
2065 
2066 	mmcio = &ccb->mmcio;
2067 	if (!(slot->flags & CMD_STARTED)) {
2068 		slot->flags |= CMD_STARTED;
2069 		sdhci_start_command(slot, &mmcio->cmd);
2070 		return;
2071 	}
2072 
2073 	/*
2074 	 * Old stack doesn't use this!
2075 	 * Enabling this code causes significant performance degradation
2076 	 * and IRQ storms on BBB, Wandboard behaves fine.
2077 	 * Not using this code does no harm...
2078 	if (!(slot->flags & STOP_STARTED) && mmcio->stop.opcode != 0) {
2079 		slot->flags |= STOP_STARTED;
2080 		sdhci_start_command(slot, &mmcio->stop);
2081 		return;
2082 	}
2083 	*/
2084 	if (__predict_false(sdhci_debug > 1))
2085 		slot_printf(slot, "result: %d\n", mmcio->cmd.error);
2086 	if (mmcio->cmd.error == 0 &&
2087 	    (slot->quirks & SDHCI_QUIRK_RESET_AFTER_REQUEST)) {
2088 		SDHCI_RESET(slot->bus, slot, SDHCI_RESET_CMD);
2089 		SDHCI_RESET(slot->bus, slot, SDHCI_RESET_DATA);
2090 	}
2091 
2092 	sdhci_req_done(slot);
2093 }
2094 #else
2095 static void
2096 sdhci_start(struct sdhci_slot *slot)
2097 {
2098 	const struct mmc_request *req;
2099 
2100 	req = slot->req;
2101 	if (req == NULL)
2102 		return;
2103 
2104 	if (!(slot->flags & CMD_STARTED)) {
2105 		slot->flags |= CMD_STARTED;
2106 		sdhci_start_command(slot, req->cmd);
2107 		return;
2108 	}
2109 	if ((slot->quirks & SDHCI_QUIRK_BROKEN_AUTO_STOP) &&
2110 	    !(slot->flags & STOP_STARTED) && req->stop) {
2111 		slot->flags |= STOP_STARTED;
2112 		sdhci_start_command(slot, req->stop);
2113 		return;
2114 	}
2115 	if (__predict_false(sdhci_debug > 1))
2116 		slot_printf(slot, "result: %d\n", req->cmd->error);
2117 	if (!req->cmd->error &&
2118 	    ((slot->curcmd == req->stop &&
2119 	     (slot->quirks & SDHCI_QUIRK_BROKEN_AUTO_STOP)) ||
2120 	     (slot->quirks & SDHCI_QUIRK_RESET_AFTER_REQUEST))) {
2121 		SDHCI_RESET(slot->bus, slot, SDHCI_RESET_CMD);
2122 		SDHCI_RESET(slot->bus, slot, SDHCI_RESET_DATA);
2123 	}
2124 
2125 	sdhci_req_done(slot);
2126 }
2127 #endif
2128 
2129 int
2130 sdhci_generic_request(device_t brdev __unused, device_t reqdev,
2131     struct mmc_request *req)
2132 {
2133 	struct sdhci_slot *slot = device_get_ivars(reqdev);
2134 
2135 	SDHCI_LOCK(slot);
2136 	if (slot->req != NULL) {
2137 		SDHCI_UNLOCK(slot);
2138 		return (EBUSY);
2139 	}
2140 	if (__predict_false(sdhci_debug > 1)) {
2141 		slot_printf(slot,
2142 		    "CMD%u arg %#x flags %#x dlen %u dflags %#x\n",
2143 		    req->cmd->opcode, req->cmd->arg, req->cmd->flags,
2144 		    (req->cmd->data)?(u_int)req->cmd->data->len:0,
2145 		    (req->cmd->data)?req->cmd->data->flags:0);
2146 	}
2147 	slot->req = req;
2148 	slot->flags = 0;
2149 	sdhci_start(slot);
2150 	SDHCI_UNLOCK(slot);
2151 	if (dumping) {
2152 		while (slot->req != NULL) {
2153 			sdhci_generic_intr(slot);
2154 			DELAY(10);
2155 		}
2156 	}
2157 	return (0);
2158 }
2159 
2160 int
2161 sdhci_generic_get_ro(device_t brdev __unused, device_t reqdev)
2162 {
2163 	struct sdhci_slot *slot = device_get_ivars(reqdev);
2164 	uint32_t val;
2165 
2166 	SDHCI_LOCK(slot);
2167 	val = RD4(slot, SDHCI_PRESENT_STATE);
2168 	SDHCI_UNLOCK(slot);
2169 	return (!(val & SDHCI_WRITE_PROTECT));
2170 }
2171 
2172 int
2173 sdhci_generic_acquire_host(device_t brdev __unused, device_t reqdev)
2174 {
2175 	struct sdhci_slot *slot = device_get_ivars(reqdev);
2176 	int err = 0;
2177 
2178 	SDHCI_LOCK(slot);
2179 	while (slot->bus_busy)
2180 		msleep(slot, &slot->mtx, 0, "sdhciah", 0);
2181 	slot->bus_busy++;
2182 	/* Activate led. */
2183 	WR1(slot, SDHCI_HOST_CONTROL, slot->hostctrl |= SDHCI_CTRL_LED);
2184 	SDHCI_UNLOCK(slot);
2185 	return (err);
2186 }
2187 
2188 int
2189 sdhci_generic_release_host(device_t brdev __unused, device_t reqdev)
2190 {
2191 	struct sdhci_slot *slot = device_get_ivars(reqdev);
2192 
2193 	SDHCI_LOCK(slot);
2194 	/* Deactivate led. */
2195 	WR1(slot, SDHCI_HOST_CONTROL, slot->hostctrl &= ~SDHCI_CTRL_LED);
2196 	slot->bus_busy--;
2197 	wakeup(slot);
2198 	SDHCI_UNLOCK(slot);
2199 	return (0);
2200 }
2201 
2202 static void
2203 sdhci_cmd_irq(struct sdhci_slot *slot, uint32_t intmask)
2204 {
2205 
2206 	if (!slot->curcmd) {
2207 		slot_printf(slot, "Got command interrupt 0x%08x, but "
2208 		    "there is no active command.\n", intmask);
2209 		sdhci_dumpregs(slot);
2210 		return;
2211 	}
2212 	if (intmask & SDHCI_INT_TIMEOUT)
2213 		slot->curcmd->error = MMC_ERR_TIMEOUT;
2214 	else if (intmask & SDHCI_INT_CRC)
2215 		slot->curcmd->error = MMC_ERR_BADCRC;
2216 	else if (intmask & (SDHCI_INT_END_BIT | SDHCI_INT_INDEX))
2217 		slot->curcmd->error = MMC_ERR_FIFO;
2218 
2219 	sdhci_finish_command(slot);
2220 }
2221 
2222 static void
2223 sdhci_data_irq(struct sdhci_slot *slot, uint32_t intmask)
2224 {
2225 	struct mmc_data *data;
2226 	size_t left;
2227 	uint32_t sdma_bbufsz;
2228 
2229 	if (!slot->curcmd) {
2230 		slot_printf(slot, "Got data interrupt 0x%08x, but "
2231 		    "there is no active command.\n", intmask);
2232 		sdhci_dumpregs(slot);
2233 		return;
2234 	}
2235 	if (slot->curcmd->data == NULL &&
2236 	    (slot->curcmd->flags & MMC_RSP_BUSY) == 0) {
2237 		slot_printf(slot, "Got data interrupt 0x%08x, but "
2238 		    "there is no active data operation.\n",
2239 		    intmask);
2240 		sdhci_dumpregs(slot);
2241 		return;
2242 	}
2243 	if (intmask & SDHCI_INT_DATA_TIMEOUT)
2244 		slot->curcmd->error = MMC_ERR_TIMEOUT;
2245 	else if (intmask & (SDHCI_INT_DATA_CRC | SDHCI_INT_DATA_END_BIT))
2246 		slot->curcmd->error = MMC_ERR_BADCRC;
2247 	if (slot->curcmd->data == NULL &&
2248 	    (intmask & (SDHCI_INT_DATA_AVAIL | SDHCI_INT_SPACE_AVAIL |
2249 	    SDHCI_INT_DMA_END))) {
2250 		slot_printf(slot, "Got data interrupt 0x%08x, but "
2251 		    "there is busy-only command.\n", intmask);
2252 		sdhci_dumpregs(slot);
2253 		slot->curcmd->error = MMC_ERR_INVALID;
2254 	}
2255 	if (slot->curcmd->error) {
2256 		/* No need to continue after any error. */
2257 		goto done;
2258 	}
2259 
2260 	/* Handle tuning completion interrupt. */
2261 	if (__predict_false((intmask & SDHCI_INT_DATA_AVAIL) &&
2262 	    (slot->curcmd->opcode == MMC_SEND_TUNING_BLOCK ||
2263 	    slot->curcmd->opcode == MMC_SEND_TUNING_BLOCK_HS200))) {
2264 		slot->req->flags |= MMC_TUNE_DONE;
2265 		sdhci_finish_command(slot);
2266 		sdhci_finish_data(slot);
2267 		return;
2268 	}
2269 	/* Handle PIO interrupt. */
2270 	if (intmask & (SDHCI_INT_DATA_AVAIL | SDHCI_INT_SPACE_AVAIL)) {
2271 		if ((slot->opt & SDHCI_PLATFORM_TRANSFER) &&
2272 		    SDHCI_PLATFORM_WILL_HANDLE(slot->bus, slot)) {
2273 			SDHCI_PLATFORM_START_TRANSFER(slot->bus, slot,
2274 			    &intmask);
2275 			slot->flags |= PLATFORM_DATA_STARTED;
2276 		} else
2277 			sdhci_transfer_pio(slot);
2278 	}
2279 	/* Handle DMA border. */
2280 	if (intmask & SDHCI_INT_DMA_END) {
2281 		data = slot->curcmd->data;
2282 		sdma_bbufsz = slot->sdma_bbufsz;
2283 
2284 		/* Unload DMA buffer ... */
2285 		left = data->len - slot->offset;
2286 		if (data->flags & MMC_DATA_READ) {
2287 			bus_dmamap_sync(slot->dmatag, slot->dmamap,
2288 			    BUS_DMASYNC_POSTREAD);
2289 			memcpy((u_char*)data->data + slot->offset, slot->dmamem,
2290 			    ulmin(left, sdma_bbufsz));
2291 		} else {
2292 			bus_dmamap_sync(slot->dmatag, slot->dmamap,
2293 			    BUS_DMASYNC_POSTWRITE);
2294 		}
2295 		/* ... and reload it again. */
2296 		slot->offset += sdma_bbufsz;
2297 		left = data->len - slot->offset;
2298 		if (data->flags & MMC_DATA_READ) {
2299 			bus_dmamap_sync(slot->dmatag, slot->dmamap,
2300 			    BUS_DMASYNC_PREREAD);
2301 		} else {
2302 			memcpy(slot->dmamem, (u_char*)data->data + slot->offset,
2303 			    ulmin(left, sdma_bbufsz));
2304 			bus_dmamap_sync(slot->dmatag, slot->dmamap,
2305 			    BUS_DMASYNC_PREWRITE);
2306 		}
2307 		/*
2308 		 * Interrupt aggregation: Mask border interrupt for the last
2309 		 * bounce buffer.
2310 		 */
2311 		if (left == sdma_bbufsz) {
2312 			slot->intmask &= ~SDHCI_INT_DMA_END;
2313 			WR4(slot, SDHCI_SIGNAL_ENABLE, slot->intmask);
2314 		}
2315 		/* Restart DMA. */
2316 		WR4(slot, SDHCI_DMA_ADDRESS, slot->paddr);
2317 	}
2318 	/* We have got all data. */
2319 	if (intmask & SDHCI_INT_DATA_END) {
2320 		if (slot->flags & PLATFORM_DATA_STARTED) {
2321 			slot->flags &= ~PLATFORM_DATA_STARTED;
2322 			SDHCI_PLATFORM_FINISH_TRANSFER(slot->bus, slot);
2323 		} else
2324 			sdhci_finish_data(slot);
2325 	}
2326 done:
2327 	if (slot->curcmd != NULL && slot->curcmd->error != 0) {
2328 		if (slot->flags & PLATFORM_DATA_STARTED) {
2329 			slot->flags &= ~PLATFORM_DATA_STARTED;
2330 			SDHCI_PLATFORM_FINISH_TRANSFER(slot->bus, slot);
2331 		} else
2332 			sdhci_finish_data(slot);
2333 	}
2334 }
2335 
2336 static void
2337 sdhci_acmd_irq(struct sdhci_slot *slot, uint16_t acmd_err)
2338 {
2339 
2340 	if (!slot->curcmd) {
2341 		slot_printf(slot, "Got AutoCMD12 error 0x%04x, but "
2342 		    "there is no active command.\n", acmd_err);
2343 		sdhci_dumpregs(slot);
2344 		return;
2345 	}
2346 	slot_printf(slot, "Got AutoCMD12 error 0x%04x\n", acmd_err);
2347 	SDHCI_RESET(slot->bus, slot, SDHCI_RESET_CMD);
2348 }
2349 
2350 void
2351 sdhci_generic_intr(struct sdhci_slot *slot)
2352 {
2353 	uint32_t intmask, present;
2354 	uint16_t val16;
2355 
2356 	SDHCI_LOCK(slot);
2357 	/* Read slot interrupt status. */
2358 	intmask = RD4(slot, SDHCI_INT_STATUS);
2359 	if (intmask == 0 || intmask == 0xffffffff) {
2360 		SDHCI_UNLOCK(slot);
2361 		return;
2362 	}
2363 	if (__predict_false(sdhci_debug > 2))
2364 		slot_printf(slot, "Interrupt %#x\n", intmask);
2365 
2366 	/* Handle tuning error interrupt. */
2367 	if (__predict_false(intmask & SDHCI_INT_TUNEERR)) {
2368 		WR4(slot, SDHCI_INT_STATUS, SDHCI_INT_TUNEERR);
2369 		slot_printf(slot, "Tuning error indicated\n");
2370 		slot->retune_req |= SDHCI_RETUNE_REQ_RESET;
2371 		if (slot->curcmd) {
2372 			slot->curcmd->error = MMC_ERR_BADCRC;
2373 			sdhci_finish_command(slot);
2374 		}
2375 	}
2376 	/* Handle re-tuning interrupt. */
2377 	if (__predict_false(intmask & SDHCI_INT_RETUNE))
2378 		slot->retune_req |= SDHCI_RETUNE_REQ_NEEDED;
2379 	/* Handle card presence interrupts. */
2380 	if (intmask & (SDHCI_INT_CARD_INSERT | SDHCI_INT_CARD_REMOVE)) {
2381 		present = (intmask & SDHCI_INT_CARD_INSERT) != 0;
2382 		slot->intmask &=
2383 		    ~(SDHCI_INT_CARD_INSERT | SDHCI_INT_CARD_REMOVE);
2384 		slot->intmask |= present ? SDHCI_INT_CARD_REMOVE :
2385 		    SDHCI_INT_CARD_INSERT;
2386 		WR4(slot, SDHCI_INT_ENABLE, slot->intmask);
2387 		WR4(slot, SDHCI_SIGNAL_ENABLE, slot->intmask);
2388 		WR4(slot, SDHCI_INT_STATUS, intmask &
2389 		    (SDHCI_INT_CARD_INSERT | SDHCI_INT_CARD_REMOVE));
2390 		sdhci_handle_card_present_locked(slot, present);
2391 	}
2392 	/* Handle command interrupts. */
2393 	if (intmask & SDHCI_INT_CMD_MASK) {
2394 		WR4(slot, SDHCI_INT_STATUS, intmask & SDHCI_INT_CMD_MASK);
2395 		sdhci_cmd_irq(slot, intmask & SDHCI_INT_CMD_MASK);
2396 	}
2397 	/* Handle data interrupts. */
2398 	if (intmask & SDHCI_INT_DATA_MASK) {
2399 		WR4(slot, SDHCI_INT_STATUS, intmask & SDHCI_INT_DATA_MASK);
2400 		/* Don't call data_irq in case of errored command. */
2401 		if ((intmask & SDHCI_INT_CMD_ERROR_MASK) == 0)
2402 			sdhci_data_irq(slot, intmask & SDHCI_INT_DATA_MASK);
2403 	}
2404 	/* Handle AutoCMD12 error interrupt. */
2405 	if (intmask & SDHCI_INT_ACMD12ERR) {
2406 		/* Clearing SDHCI_INT_ACMD12ERR may clear SDHCI_ACMD12_ERR. */
2407 		val16 = RD2(slot, SDHCI_ACMD12_ERR);
2408 		WR4(slot, SDHCI_INT_STATUS, SDHCI_INT_ACMD12ERR);
2409 		sdhci_acmd_irq(slot, val16);
2410 	}
2411 	/* Handle bus power interrupt. */
2412 	if (intmask & SDHCI_INT_BUS_POWER) {
2413 		WR4(slot, SDHCI_INT_STATUS, SDHCI_INT_BUS_POWER);
2414 		slot_printf(slot, "Card is consuming too much power!\n");
2415 	}
2416 	intmask &= ~(SDHCI_INT_ERROR | SDHCI_INT_TUNEERR | SDHCI_INT_RETUNE |
2417 	    SDHCI_INT_CARD_INSERT | SDHCI_INT_CARD_REMOVE | SDHCI_INT_CMD_MASK |
2418 	    SDHCI_INT_DATA_MASK | SDHCI_INT_ACMD12ERR | SDHCI_INT_BUS_POWER);
2419 	/* The rest is unknown. */
2420 	if (intmask) {
2421 		WR4(slot, SDHCI_INT_STATUS, intmask);
2422 		slot_printf(slot, "Unexpected interrupt 0x%08x.\n",
2423 		    intmask);
2424 		sdhci_dumpregs(slot);
2425 	}
2426 
2427 	SDHCI_UNLOCK(slot);
2428 }
2429 
2430 int
2431 sdhci_generic_read_ivar(device_t bus, device_t child, int which,
2432     uintptr_t *result)
2433 {
2434 	const struct sdhci_slot *slot = device_get_ivars(child);
2435 
2436 	switch (which) {
2437 	default:
2438 		return (EINVAL);
2439 	case MMCBR_IVAR_BUS_MODE:
2440 		*result = slot->host.ios.bus_mode;
2441 		break;
2442 	case MMCBR_IVAR_BUS_WIDTH:
2443 		*result = slot->host.ios.bus_width;
2444 		break;
2445 	case MMCBR_IVAR_CHIP_SELECT:
2446 		*result = slot->host.ios.chip_select;
2447 		break;
2448 	case MMCBR_IVAR_CLOCK:
2449 		*result = slot->host.ios.clock;
2450 		break;
2451 	case MMCBR_IVAR_F_MIN:
2452 		*result = slot->host.f_min;
2453 		break;
2454 	case MMCBR_IVAR_F_MAX:
2455 		*result = slot->host.f_max;
2456 		break;
2457 	case MMCBR_IVAR_HOST_OCR:
2458 		*result = slot->host.host_ocr;
2459 		break;
2460 	case MMCBR_IVAR_MODE:
2461 		*result = slot->host.mode;
2462 		break;
2463 	case MMCBR_IVAR_OCR:
2464 		*result = slot->host.ocr;
2465 		break;
2466 	case MMCBR_IVAR_POWER_MODE:
2467 		*result = slot->host.ios.power_mode;
2468 		break;
2469 	case MMCBR_IVAR_VDD:
2470 		*result = slot->host.ios.vdd;
2471 		break;
2472 	case MMCBR_IVAR_RETUNE_REQ:
2473 		if (slot->opt & SDHCI_TUNING_ENABLED) {
2474 			if (slot->retune_req & SDHCI_RETUNE_REQ_RESET) {
2475 				*result = retune_req_reset;
2476 				break;
2477 			}
2478 			if (slot->retune_req & SDHCI_RETUNE_REQ_NEEDED) {
2479 				*result = retune_req_normal;
2480 				break;
2481 			}
2482 		}
2483 		*result = retune_req_none;
2484 		break;
2485 	case MMCBR_IVAR_VCCQ:
2486 		*result = slot->host.ios.vccq;
2487 		break;
2488 	case MMCBR_IVAR_CAPS:
2489 		*result = slot->host.caps;
2490 		break;
2491 	case MMCBR_IVAR_TIMING:
2492 		*result = slot->host.ios.timing;
2493 		break;
2494 	case MMCBR_IVAR_MAX_DATA:
2495 		/*
2496 		 * Re-tuning modes 1 and 2 restrict the maximum data length
2497 		 * per read/write command to 4 MiB.
2498 		 */
2499 		if (slot->opt & SDHCI_TUNING_ENABLED &&
2500 		    (slot->retune_mode == SDHCI_RETUNE_MODE_1 ||
2501 		    slot->retune_mode == SDHCI_RETUNE_MODE_2)) {
2502 			*result = 4 * 1024 * 1024 / MMC_SECTOR_SIZE;
2503 			break;
2504 		}
2505 		*result = 65535;
2506 		break;
2507 	case MMCBR_IVAR_MAX_BUSY_TIMEOUT:
2508 		/*
2509 		 * Currently, sdhci_start_data() hardcodes 1 s for all CMDs.
2510 		 */
2511 		*result = 1000000;
2512 		break;
2513 	}
2514 	return (0);
2515 }
2516 
2517 int
2518 sdhci_generic_write_ivar(device_t bus, device_t child, int which,
2519     uintptr_t value)
2520 {
2521 	struct sdhci_slot *slot = device_get_ivars(child);
2522 	uint32_t clock, max_clock;
2523 	int i;
2524 
2525 	if (sdhci_debug > 1)
2526 		slot_printf(slot, "%s: var=%d\n", __func__, which);
2527 	switch (which) {
2528 	default:
2529 		return (EINVAL);
2530 	case MMCBR_IVAR_BUS_MODE:
2531 		slot->host.ios.bus_mode = value;
2532 		break;
2533 	case MMCBR_IVAR_BUS_WIDTH:
2534 		slot->host.ios.bus_width = value;
2535 		break;
2536 	case MMCBR_IVAR_CHIP_SELECT:
2537 		slot->host.ios.chip_select = value;
2538 		break;
2539 	case MMCBR_IVAR_CLOCK:
2540 		if (value > 0) {
2541 			max_clock = slot->max_clk;
2542 			clock = max_clock;
2543 
2544 			if (slot->version < SDHCI_SPEC_300) {
2545 				for (i = 0; i < SDHCI_200_MAX_DIVIDER;
2546 				    i <<= 1) {
2547 					if (clock <= value)
2548 						break;
2549 					clock >>= 1;
2550 				}
2551 			} else {
2552 				for (i = 0; i < SDHCI_300_MAX_DIVIDER;
2553 				    i += 2) {
2554 					if (clock <= value)
2555 						break;
2556 					clock = max_clock / (i + 2);
2557 				}
2558 			}
2559 
2560 			slot->host.ios.clock = clock;
2561 		} else
2562 			slot->host.ios.clock = 0;
2563 		break;
2564 	case MMCBR_IVAR_MODE:
2565 		slot->host.mode = value;
2566 		break;
2567 	case MMCBR_IVAR_OCR:
2568 		slot->host.ocr = value;
2569 		break;
2570 	case MMCBR_IVAR_POWER_MODE:
2571 		slot->host.ios.power_mode = value;
2572 		break;
2573 	case MMCBR_IVAR_VDD:
2574 		slot->host.ios.vdd = value;
2575 		break;
2576 	case MMCBR_IVAR_VCCQ:
2577 		slot->host.ios.vccq = value;
2578 		break;
2579 	case MMCBR_IVAR_TIMING:
2580 		slot->host.ios.timing = value;
2581 		break;
2582 	case MMCBR_IVAR_CAPS:
2583 	case MMCBR_IVAR_HOST_OCR:
2584 	case MMCBR_IVAR_F_MIN:
2585 	case MMCBR_IVAR_F_MAX:
2586 	case MMCBR_IVAR_MAX_DATA:
2587 	case MMCBR_IVAR_RETUNE_REQ:
2588 		return (EINVAL);
2589 	}
2590 	return (0);
2591 }
2592 
2593 #ifdef MMCCAM
2594 void
2595 sdhci_start_slot(struct sdhci_slot *slot)
2596 {
2597 
2598 	if ((slot->devq = cam_simq_alloc(1)) == NULL)
2599 		goto fail;
2600 
2601 	mtx_init(&slot->sim_mtx, "sdhcisim", NULL, MTX_DEF);
2602 	slot->sim = cam_sim_alloc(sdhci_cam_action, sdhci_cam_poll,
2603 	    "sdhci_slot", slot, device_get_unit(slot->bus),
2604 	    &slot->sim_mtx, 1, 1, slot->devq);
2605 
2606 	if (slot->sim == NULL) {
2607 		cam_simq_free(slot->devq);
2608 		slot_printf(slot, "cannot allocate CAM SIM\n");
2609 		goto fail;
2610 	}
2611 
2612 	mtx_lock(&slot->sim_mtx);
2613 	if (xpt_bus_register(slot->sim, slot->bus, 0) != 0) {
2614 		slot_printf(slot, "cannot register SCSI pass-through bus\n");
2615 		cam_sim_free(slot->sim, FALSE);
2616 		cam_simq_free(slot->devq);
2617 		mtx_unlock(&slot->sim_mtx);
2618 		goto fail;
2619 	}
2620 	mtx_unlock(&slot->sim_mtx);
2621 
2622 	/* End CAM-specific init */
2623 	slot->card_present = 0;
2624 	sdhci_card_task(slot, 0);
2625 	return;
2626 
2627 fail:
2628 	if (slot->sim != NULL) {
2629 		mtx_lock(&slot->sim_mtx);
2630 		xpt_bus_deregister(cam_sim_path(slot->sim));
2631 		cam_sim_free(slot->sim, FALSE);
2632 		mtx_unlock(&slot->sim_mtx);
2633 	}
2634 
2635 	if (slot->devq != NULL)
2636 		cam_simq_free(slot->devq);
2637 }
2638 
2639 void
2640 sdhci_cam_action(struct cam_sim *sim, union ccb *ccb)
2641 {
2642 	struct sdhci_slot *slot;
2643 
2644 	slot = cam_sim_softc(sim);
2645 	if (slot == NULL) {
2646 		ccb->ccb_h.status = CAM_SEL_TIMEOUT;
2647 		xpt_done(ccb);
2648 		return;
2649 	}
2650 
2651 	mtx_assert(&slot->sim_mtx, MA_OWNED);
2652 
2653 	switch (ccb->ccb_h.func_code) {
2654 	case XPT_PATH_INQ:
2655 		mmc_path_inq(&ccb->cpi, "Deglitch Networks", sim, maxphys);
2656 		break;
2657 
2658 	case XPT_MMC_GET_TRAN_SETTINGS:
2659 	case XPT_GET_TRAN_SETTINGS:
2660 	{
2661 		struct ccb_trans_settings *cts = &ccb->cts;
2662 		uint32_t max_data;
2663 
2664 		if (sdhci_debug > 1)
2665 			slot_printf(slot, "Got XPT_GET_TRAN_SETTINGS\n");
2666 
2667 		cts->protocol = PROTO_MMCSD;
2668 		cts->protocol_version = 1;
2669 		cts->transport = XPORT_MMCSD;
2670 		cts->transport_version = 1;
2671 		cts->xport_specific.valid = 0;
2672 		cts->proto_specific.mmc.host_ocr = slot->host.host_ocr;
2673 		cts->proto_specific.mmc.host_f_min = slot->host.f_min;
2674 		cts->proto_specific.mmc.host_f_max = slot->host.f_max;
2675 		cts->proto_specific.mmc.host_caps = slot->host.caps;
2676 		/*
2677 		 * Re-tuning modes 1 and 2 restrict the maximum data length
2678 		 * per read/write command to 4 MiB.
2679 		 */
2680 		if (slot->opt & SDHCI_TUNING_ENABLED &&
2681 		    (slot->retune_mode == SDHCI_RETUNE_MODE_1 ||
2682 		    slot->retune_mode == SDHCI_RETUNE_MODE_2)) {
2683 			max_data = 4 * 1024 * 1024 / MMC_SECTOR_SIZE;
2684 		} else {
2685 			max_data = 65535;
2686 		}
2687 		cts->proto_specific.mmc.host_max_data = max_data;
2688 
2689 		memcpy(&cts->proto_specific.mmc.ios, &slot->host.ios, sizeof(struct mmc_ios));
2690 		ccb->ccb_h.status = CAM_REQ_CMP;
2691 		break;
2692 	}
2693 	case XPT_MMC_SET_TRAN_SETTINGS:
2694 	case XPT_SET_TRAN_SETTINGS:
2695 		if (sdhci_debug > 1)
2696 			slot_printf(slot, "Got XPT_SET_TRAN_SETTINGS\n");
2697 		sdhci_cam_settran_settings(slot, ccb);
2698 		ccb->ccb_h.status = CAM_REQ_CMP;
2699 		break;
2700 	case XPT_RESET_BUS:
2701 		if (sdhci_debug > 1)
2702 			slot_printf(slot, "Got XPT_RESET_BUS, ACK it...\n");
2703 		ccb->ccb_h.status = CAM_REQ_CMP;
2704 		break;
2705 	case XPT_MMC_IO:
2706 		/*
2707 		 * Here is the HW-dependent part of
2708 		 * sending the command to the underlying h/w
2709 		 * At some point in the future an interrupt comes.
2710 		 * Then the request will be marked as completed.
2711 		 */
2712 		if (__predict_false(sdhci_debug > 1))
2713 			slot_printf(slot, "Got XPT_MMC_IO\n");
2714 		ccb->ccb_h.status = CAM_REQ_INPROG;
2715 
2716 		sdhci_cam_request(cam_sim_softc(sim), ccb);
2717 		return;
2718 	default:
2719 		ccb->ccb_h.status = CAM_REQ_INVALID;
2720 		break;
2721 	}
2722 	xpt_done(ccb);
2723 	return;
2724 }
2725 
2726 void
2727 sdhci_cam_poll(struct cam_sim *sim)
2728 {
2729 	sdhci_generic_intr(cam_sim_softc(sim));
2730 }
2731 
2732 static int
2733 sdhci_cam_get_possible_host_clock(const struct sdhci_slot *slot,
2734     int proposed_clock)
2735 {
2736 	int max_clock, clock, i;
2737 
2738 	if (proposed_clock == 0)
2739 		return 0;
2740 	max_clock = slot->max_clk;
2741 	clock = max_clock;
2742 
2743 	if (slot->version < SDHCI_SPEC_300) {
2744 		for (i = 0; i < SDHCI_200_MAX_DIVIDER; i <<= 1) {
2745 			if (clock <= proposed_clock)
2746 				break;
2747 			clock >>= 1;
2748 		}
2749 	} else {
2750 		for (i = 0; i < SDHCI_300_MAX_DIVIDER; i += 2) {
2751 			if (clock <= proposed_clock)
2752 				break;
2753 			clock = max_clock / (i + 2);
2754 		}
2755 	}
2756 	return clock;
2757 }
2758 
2759 static int
2760 sdhci_cam_settran_settings(struct sdhci_slot *slot, union ccb *ccb)
2761 {
2762 	struct mmc_ios *ios;
2763 	const struct mmc_ios *new_ios;
2764 	const struct ccb_trans_settings_mmc *cts;
2765 
2766 	ios = &slot->host.ios;
2767 	cts = &ccb->cts.proto_specific.mmc;
2768 	new_ios = &cts->ios;
2769 
2770 	/* Update only requested fields */
2771 	if (cts->ios_valid & MMC_CLK) {
2772 		ios->clock = sdhci_cam_get_possible_host_clock(slot, new_ios->clock);
2773 		if (sdhci_debug > 1)
2774 			slot_printf(slot, "Clock => %d\n", ios->clock);
2775 	}
2776 	if (cts->ios_valid & MMC_VDD) {
2777 		ios->vdd = new_ios->vdd;
2778 		if (sdhci_debug > 1)
2779 			slot_printf(slot, "VDD => %d\n", ios->vdd);
2780 	}
2781 	if (cts->ios_valid & MMC_CS) {
2782 		ios->chip_select = new_ios->chip_select;
2783 		if (sdhci_debug > 1)
2784 			slot_printf(slot, "CS => %d\n", ios->chip_select);
2785 	}
2786 	if (cts->ios_valid & MMC_BW) {
2787 		ios->bus_width = new_ios->bus_width;
2788 		if (sdhci_debug > 1)
2789 			slot_printf(slot, "Bus width => %d\n", ios->bus_width);
2790 	}
2791 	if (cts->ios_valid & MMC_PM) {
2792 		ios->power_mode = new_ios->power_mode;
2793 		if (sdhci_debug > 1)
2794 			slot_printf(slot, "Power mode => %d\n", ios->power_mode);
2795 	}
2796 	if (cts->ios_valid & MMC_BT) {
2797 		ios->timing = new_ios->timing;
2798 		if (sdhci_debug > 1)
2799 			slot_printf(slot, "Timing => %d\n", ios->timing);
2800 	}
2801 	if (cts->ios_valid & MMC_BM) {
2802 		ios->bus_mode = new_ios->bus_mode;
2803 		if (sdhci_debug > 1)
2804 			slot_printf(slot, "Bus mode => %d\n", ios->bus_mode);
2805 	}
2806 	if (cts->ios_valid & MMC_VCCQ) {
2807 		ios->vccq = new_ios->vccq;
2808 		if (sdhci_debug > 1)
2809 			slot_printf(slot, "VCCQ => %d\n", ios->vccq);
2810 	}
2811 
2812 	/* XXX Provide a way to call a chip-specific IOS update, required for TI */
2813 	return (sdhci_cam_update_ios(slot));
2814 }
2815 
2816 static int
2817 sdhci_cam_update_ios(struct sdhci_slot *slot)
2818 {
2819 	struct mmc_ios *ios = &slot->host.ios;
2820 
2821 	if (sdhci_debug > 1)
2822 		slot_printf(slot, "%s: power_mode=%d, clk=%d, bus_width=%d, timing=%d\n",
2823 		    __func__, ios->power_mode, ios->clock, ios->bus_width, ios->timing);
2824 	SDHCI_LOCK(slot);
2825 	/* Do full reset on bus power down to clear from any state. */
2826 	if (ios->power_mode == power_off) {
2827 		WR4(slot, SDHCI_SIGNAL_ENABLE, 0);
2828 		sdhci_init(slot);
2829 	}
2830 	/* Configure the bus. */
2831 	sdhci_set_clock(slot, ios->clock);
2832 	sdhci_set_power(slot, (ios->power_mode == power_off) ? 0 : ios->vdd);
2833 	if (ios->bus_width == bus_width_8) {
2834 		slot->hostctrl |= SDHCI_CTRL_8BITBUS;
2835 		slot->hostctrl &= ~SDHCI_CTRL_4BITBUS;
2836 	} else if (ios->bus_width == bus_width_4) {
2837 		slot->hostctrl &= ~SDHCI_CTRL_8BITBUS;
2838 		slot->hostctrl |= SDHCI_CTRL_4BITBUS;
2839 	} else if (ios->bus_width == bus_width_1) {
2840 		slot->hostctrl &= ~SDHCI_CTRL_8BITBUS;
2841 		slot->hostctrl &= ~SDHCI_CTRL_4BITBUS;
2842 	} else {
2843 		panic("Invalid bus width: %d", ios->bus_width);
2844 	}
2845 	if (ios->timing == bus_timing_hs &&
2846 	    !(slot->quirks & SDHCI_QUIRK_DONT_SET_HISPD_BIT))
2847 		slot->hostctrl |= SDHCI_CTRL_HISPD;
2848 	else
2849 		slot->hostctrl &= ~SDHCI_CTRL_HISPD;
2850 	WR1(slot, SDHCI_HOST_CONTROL, slot->hostctrl);
2851 	/* Some controllers like reset after bus changes. */
2852 	if(slot->quirks & SDHCI_QUIRK_RESET_ON_IOS)
2853 		SDHCI_RESET(slot->bus, slot,
2854 		    SDHCI_RESET_CMD | SDHCI_RESET_DATA);
2855 
2856 	SDHCI_UNLOCK(slot);
2857 	return (0);
2858 }
2859 
2860 static int
2861 sdhci_cam_request(struct sdhci_slot *slot, union ccb *ccb)
2862 {
2863 	const struct ccb_mmcio *mmcio;
2864 
2865 	mmcio = &ccb->mmcio;
2866 
2867 	SDHCI_LOCK(slot);
2868 /*	if (slot->req != NULL) {
2869 		SDHCI_UNLOCK(slot);
2870 		return (EBUSY);
2871 	}
2872 */
2873 	if (__predict_false(sdhci_debug > 1)) {
2874 		slot_printf(slot, "CMD%u arg %#x flags %#x dlen %u dflags %#x "
2875 		    "blksz=%zu blkcnt=%zu\n",
2876 		    mmcio->cmd.opcode, mmcio->cmd.arg, mmcio->cmd.flags,
2877 		    mmcio->cmd.data != NULL ? (unsigned int) mmcio->cmd.data->len : 0,
2878 		    mmcio->cmd.data != NULL ? mmcio->cmd.data->flags : 0,
2879 		    mmcio->cmd.data != NULL ? mmcio->cmd.data->block_size : 0,
2880 		    mmcio->cmd.data != NULL ? mmcio->cmd.data->block_count : 0);
2881 	}
2882 	if (mmcio->cmd.data != NULL) {
2883 		if (mmcio->cmd.data->len == 0 || mmcio->cmd.data->flags == 0)
2884 			panic("data->len = %d, data->flags = %d -- something is b0rked",
2885 			    (int)mmcio->cmd.data->len, mmcio->cmd.data->flags);
2886 	}
2887 	slot->ccb = ccb;
2888 	slot->flags = 0;
2889 	sdhci_start(slot);
2890 	SDHCI_UNLOCK(slot);
2891 	return (0);
2892 }
2893 #endif /* MMCCAM */
2894 
2895 MODULE_VERSION(sdhci, SDHCI_VERSION);
2896