xref: /dragonfly/sys/dev/disk/sdhci/sdhci.c (revision 678e8cc6)
1 /*-
2  * Copyright (c) 2008 Alexander Motin <mav@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
15  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
16  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
17  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
18  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
19  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
20  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
21  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
22  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
23  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
24  *
25  * $FreeBSD: src/sys/dev/sdhci/sdhci.c,v 1.8 2009/02/17 19:12:15 mav Exp $
26  */
27 
28 #include <stdarg.h>
29 
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/bus.h>
33 #include <sys/conf.h>
34 #include <sys/kernel.h>
35 #include <sys/lock.h>
36 #include <sys/module.h>
37 #include <sys/spinlock.h>
38 #include <sys/resource.h>
39 #include <sys/rman.h>
40 #include <sys/sysctl.h>
41 #include <sys/taskqueue.h>
42 
43 #include <bus/pci/pcireg.h>
44 #include <bus/pci/pcivar.h>
45 
46 #include <bus/mmc/bridge.h>
47 #include <bus/mmc/mmcreg.h>
48 #include <bus/mmc/mmcbrvar.h>
49 
50 #include "mmcbr_if.h"
51 #include "sdhci.h"
52 
53 #define DMA_BLOCK_SIZE	4096
54 #define DMA_BOUNDARY	0	/* DMA reload every 4K */
55 
56 /* Controller doesn't honor resets unless we touch the clock register */
57 #define SDHCI_QUIRK_CLOCK_BEFORE_RESET			(1<<0)
58 /* Controller really supports DMA */
59 #define SDHCI_QUIRK_FORCE_DMA				(1<<1)
60 /* Controller has unusable DMA engine */
61 #define SDHCI_QUIRK_BROKEN_DMA				(1<<2)
62 /* Controller doesn't like to be reset when there is no card inserted. */
63 #define SDHCI_QUIRK_NO_CARD_NO_RESET			(1<<3)
64 /* Controller has flaky internal state so reset it on each ios change */
65 #define SDHCI_QUIRK_RESET_ON_IOS			(1<<4)
66 /* Controller can only DMA chunk sizes that are a multiple of 32 bits */
67 #define SDHCI_QUIRK_32BIT_DMA_SIZE			(1<<5)
68 /* Controller needs to be reset after each request to stay stable */
69 #define SDHCI_QUIRK_RESET_AFTER_REQUEST			(1<<6)
70 /* Controller has an off-by-one issue with timeout value */
71 #define SDHCI_QUIRK_INCR_TIMEOUT_CONTROL		(1<<7)
72 /* Controller has broken read timings */
73 #define SDHCI_QUIRK_BROKEN_TIMINGS			(1<<8)
74 
75 static const struct sdhci_device {
76 	uint32_t	model;
77 	uint16_t	subvendor;
78 	char		*desc;
79 	u_int		quirks;
80 } sdhci_devices[] = {
81 	{ 0x08221180,	0xffff,	"RICOH R5C822 SD",
82 	    SDHCI_QUIRK_FORCE_DMA },
83 	{ 0x8034104c,	0xffff, "TI XX21/XX11 SD",
84 	    SDHCI_QUIRK_FORCE_DMA },
85 	{ 0x05501524,	0xffff, "ENE CB712 SD",
86 	    SDHCI_QUIRK_BROKEN_TIMINGS },
87 	{ 0x05511524,	0xffff, "ENE CB712 SD 2",
88 	    SDHCI_QUIRK_BROKEN_TIMINGS },
89 	{ 0x07501524,	0xffff, "ENE CB714 SD",
90 	    SDHCI_QUIRK_RESET_ON_IOS |
91 	    SDHCI_QUIRK_BROKEN_TIMINGS },
92 	{ 0x07511524,	0xffff, "ENE CB714 SD 2",
93 	    SDHCI_QUIRK_RESET_ON_IOS |
94 	    SDHCI_QUIRK_BROKEN_TIMINGS },
95 	{ 0x410111ab,	0xffff, "Marvell CaFe SD",
96 	    SDHCI_QUIRK_INCR_TIMEOUT_CONTROL },
97 	{ 0x2381197B,	0xffff,	"JMicron JMB38X SD",
98 	    SDHCI_QUIRK_32BIT_DMA_SIZE |
99 	    SDHCI_QUIRK_RESET_AFTER_REQUEST },
100 	{ 0,		0xffff,	NULL,
101 	    0 }
102 };
103 
104 struct sdhci_softc;
105 
106 struct sdhci_slot {
107 	struct sdhci_softc	*sc;
108 	device_t	dev;		/* Slot device */
109 	u_char		num;		/* Slot number */
110 	u_char		opt;		/* Slot options */
111 #define SDHCI_HAVE_DMA		1
112 	uint32_t	max_clk;	/* Max possible freq */
113 	uint32_t	timeout_clk;	/* Timeout freq */
114 	struct resource	*mem_res;	/* Memory resource */
115 	int		mem_rid;
116 	bus_dma_tag_t	dmatag;
117 	bus_dmamap_t	dmamap;
118 	u_char		*dmamem;
119 	bus_addr_t	paddr;		/* DMA buffer address */
120 	struct task	card_task;	/* Card presence check task */
121 	struct callout	card_callout;	/* Card insert delay callout */
122 	struct mmc_host host;		/* Host parameters */
123 	struct mmc_request *req;	/* Current request */
124 	struct mmc_command *curcmd;	/* Current command of current request */
125 
126 	uint32_t	intmask;	/* Current interrupt mask */
127 	uint32_t	clock;		/* Current clock freq. */
128 	size_t		offset;		/* Data buffer offset */
129 	uint8_t		hostctrl;	/* Current host control register */
130 	u_char		power;		/* Current power */
131 	u_char		bus_busy;	/* Bus busy status */
132 	u_char		cmd_done;	/* CMD command part done flag */
133 	u_char		data_done;	/* DAT command part done flag */
134 	u_char		flags;		/* Request execution flags */
135 #define CMD_STARTED		1
136 #define STOP_STARTED		2
137 #define SDHCI_USE_DMA		4	/* Use DMA for this req. */
138 	struct lock	lock;		/* Slot lock */
139 };
140 
141 struct sdhci_softc {
142 	device_t	dev;		/* Controller device */
143 	u_int		quirks;		/* Chip specific quirks */
144 	struct resource *irq_res;	/* IRQ resource */
145 	int		irq_rid;
146 	void		*intrhand;	/* Interrupt handle */
147 
148 	int		num_slots;	/* Number of slots on this controller */
149 	struct sdhci_slot slots[6];
150 };
151 
152 SYSCTL_NODE(_hw, OID_AUTO, sdhci, CTLFLAG_RD, 0, "sdhci driver");
153 
154 int	sdhci_debug;
155 TUNABLE_INT("hw.sdhci.debug", &sdhci_debug);
156 SYSCTL_INT(_hw_sdhci, OID_AUTO, debug, CTLFLAG_RW, &sdhci_debug, 0, "Debug level");
157 
158 static inline uint8_t
159 RD1(struct sdhci_slot *slot, bus_size_t off)
160 {
161 	bus_barrier(slot->mem_res, 0, 0xFF,
162 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
163 	return bus_read_1(slot->mem_res, off);
164 }
165 
166 static inline void
167 WR1(struct sdhci_slot *slot, bus_size_t off, uint8_t val)
168 {
169 	bus_barrier(slot->mem_res, 0, 0xFF,
170 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
171 	bus_write_1(slot->mem_res, off, val);
172 }
173 
174 static inline uint16_t
175 RD2(struct sdhci_slot *slot, bus_size_t off)
176 {
177 	bus_barrier(slot->mem_res, 0, 0xFF,
178 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
179 	return bus_read_2(slot->mem_res, off);
180 }
181 
182 static inline void
183 WR2(struct sdhci_slot *slot, bus_size_t off, uint16_t val)
184 {
185 	bus_barrier(slot->mem_res, 0, 0xFF,
186 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
187 	bus_write_2(slot->mem_res, off, val);
188 }
189 
190 static inline uint32_t
191 RD4(struct sdhci_slot *slot, bus_size_t off)
192 {
193 	bus_barrier(slot->mem_res, 0, 0xFF,
194 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
195 	return bus_read_4(slot->mem_res, off);
196 }
197 
198 static inline void
199 WR4(struct sdhci_slot *slot, bus_size_t off, uint32_t val)
200 {
201 	bus_barrier(slot->mem_res, 0, 0xFF,
202 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
203 	bus_write_4(slot->mem_res, off, val);
204 }
205 
206 static int slot_printf(struct sdhci_slot *, const char *, ...)
207 	       __printflike(2, 3);
208 
209 /* bus entry points */
210 static int sdhci_probe(device_t dev);
211 static int sdhci_attach(device_t dev);
212 static int sdhci_detach(device_t dev);
213 static void sdhci_intr(void *);
214 
215 static void sdhci_set_clock(struct sdhci_slot *slot, uint32_t clock);
216 static void sdhci_start(struct sdhci_slot *slot);
217 static void sdhci_start_data(struct sdhci_slot *slot, struct mmc_data *data);
218 
219 static void sdhci_card_task(void *, int);
220 
221 /* helper routines */
222 #define SDHCI_LOCK(_slot)		lockmgr(&(_slot)->lock, LK_EXCLUSIVE)
223 #define	SDHCI_UNLOCK(_slot)		lockmgr(&(_slot)->lock, LK_RELEASE)
224 #define SDHCI_LOCK_INIT(_slot)		lockinit(&(_slot)->lock, "sdhci", 0, LK_CANRECURSE)
225 #define SDHCI_LOCK_DESTROY(_slot)	lockuninit(&(_slot)->lock);
226 #define SDHCI_ASSERT_LOCKED(_slot)	KKASSERT(lockstatus(&(_slot)->lock, curthread) != 0);
227 #define SDHCI_ASSERT_UNLOCKED(_slot)	KKASSERT(lockstatus(&(_slot)->lock, curthread) == 0);
228 
229 static int
230 slot_printf(struct sdhci_slot *slot, const char * fmt, ...)
231 {
232 	va_list ap;
233 	int retval;
234 
235 	retval = kprintf("%s-slot%d: ",
236 	    device_get_nameunit(slot->sc->dev), slot->num);
237 
238 	va_start(ap, fmt);
239 	retval += kvprintf(fmt, ap);
240 	va_end(ap);
241 	return (retval);
242 }
243 
244 static void
245 sdhci_getaddr(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
246 {
247 	if (error != 0) {
248 		kprintf("getaddr: error %d\n", error);
249 		return;
250 	}
251 	*(bus_addr_t *)arg = segs[0].ds_addr;
252 }
253 
254 static void
255 sdhci_dumpregs(struct sdhci_slot *slot)
256 {
257 	slot_printf(slot,
258 	    "============== REGISTER DUMP ==============\n");
259 
260 	slot_printf(slot, "Sys addr: 0x%08x | Version:  0x%08x\n",
261 	    RD4(slot, SDHCI_DMA_ADDRESS), RD2(slot, SDHCI_HOST_VERSION));
262 	slot_printf(slot, "Blk size: 0x%08x | Blk cnt:  0x%08x\n",
263 	    RD2(slot, SDHCI_BLOCK_SIZE), RD2(slot, SDHCI_BLOCK_COUNT));
264 	slot_printf(slot, "Argument: 0x%08x | Trn mode: 0x%08x\n",
265 	    RD4(slot, SDHCI_ARGUMENT), RD2(slot, SDHCI_TRANSFER_MODE));
266 	slot_printf(slot, "Present:  0x%08x | Host ctl: 0x%08x\n",
267 	    RD4(slot, SDHCI_PRESENT_STATE), RD1(slot, SDHCI_HOST_CONTROL));
268 	slot_printf(slot, "Power:    0x%08x | Blk gap:  0x%08x\n",
269 	    RD1(slot, SDHCI_POWER_CONTROL), RD1(slot, SDHCI_BLOCK_GAP_CONTROL));
270 	slot_printf(slot, "Wake-up:  0x%08x | Clock:    0x%08x\n",
271 	    RD1(slot, SDHCI_WAKE_UP_CONTROL), RD2(slot, SDHCI_CLOCK_CONTROL));
272 	slot_printf(slot, "Timeout:  0x%08x | Int stat: 0x%08x\n",
273 	    RD1(slot, SDHCI_TIMEOUT_CONTROL), RD4(slot, SDHCI_INT_STATUS));
274 	slot_printf(slot, "Int enab: 0x%08x | Sig enab: 0x%08x\n",
275 	    RD4(slot, SDHCI_INT_ENABLE), RD4(slot, SDHCI_SIGNAL_ENABLE));
276 	slot_printf(slot, "AC12 err: 0x%08x | Slot int: 0x%08x\n",
277 	    RD2(slot, SDHCI_ACMD12_ERR), RD2(slot, SDHCI_SLOT_INT_STATUS));
278 	slot_printf(slot, "Caps:     0x%08x | Max curr: 0x%08x\n",
279 	    RD4(slot, SDHCI_CAPABILITIES), RD4(slot, SDHCI_MAX_CURRENT));
280 
281 	slot_printf(slot,
282 	    "===========================================\n");
283 }
284 
285 static void
286 sdhci_reset(struct sdhci_slot *slot, uint8_t mask)
287 {
288 	int timeout;
289 	uint8_t res;
290 
291 	if (slot->sc->quirks & SDHCI_QUIRK_NO_CARD_NO_RESET) {
292 		if (!(RD4(slot, SDHCI_PRESENT_STATE) &
293 			SDHCI_CARD_PRESENT))
294 			return;
295 	}
296 
297 	/* Some controllers need this kick or reset won't work. */
298 	if ((mask & SDHCI_RESET_ALL) == 0 &&
299 	    (slot->sc->quirks & SDHCI_QUIRK_CLOCK_BEFORE_RESET)) {
300 		uint32_t clock;
301 
302 		/* This is to force an update */
303 		clock = slot->clock;
304 		slot->clock = 0;
305 		sdhci_set_clock(slot, clock);
306 	}
307 
308 	WR1(slot, SDHCI_SOFTWARE_RESET, mask);
309 
310 	if (mask & SDHCI_RESET_ALL) {
311 		slot->clock = 0;
312 		slot->power = 0;
313 	}
314 
315 	/* Wait max 100 ms */
316 	timeout = 100;
317 	/* Controller clears the bits when it's done */
318 	while ((res = RD1(slot, SDHCI_SOFTWARE_RESET)) & mask) {
319 		if (timeout == 0) {
320 			slot_printf(slot,
321 			    "Reset 0x%x never completed - 0x%x.\n",
322 			    (int)mask, (int)res);
323 			sdhci_dumpregs(slot);
324 			return;
325 		}
326 		timeout--;
327 		DELAY(1000);
328 	}
329 }
330 
331 static void
332 sdhci_init(struct sdhci_slot *slot)
333 {
334 
335 	sdhci_reset(slot, SDHCI_RESET_ALL);
336 
337 	/* Enable interrupts. */
338 	slot->intmask = SDHCI_INT_BUS_POWER | SDHCI_INT_DATA_END_BIT |
339 	    SDHCI_INT_DATA_CRC | SDHCI_INT_DATA_TIMEOUT | SDHCI_INT_INDEX |
340 	    SDHCI_INT_END_BIT | SDHCI_INT_CRC | SDHCI_INT_TIMEOUT |
341 	    SDHCI_INT_CARD_REMOVE | SDHCI_INT_CARD_INSERT |
342 	    SDHCI_INT_DATA_AVAIL | SDHCI_INT_SPACE_AVAIL |
343 	    SDHCI_INT_DMA_END | SDHCI_INT_DATA_END | SDHCI_INT_RESPONSE |
344 	    SDHCI_INT_ACMD12ERR;
345 	WR4(slot, SDHCI_INT_ENABLE, slot->intmask);
346 	WR4(slot, SDHCI_SIGNAL_ENABLE, slot->intmask);
347 }
348 
349 static void
350 sdhci_set_clock(struct sdhci_slot *slot, uint32_t clock)
351 {
352 	uint32_t res;
353 	uint16_t clk;
354 	int timeout;
355 
356 	if (clock == slot->clock)
357 		return;
358 	slot->clock = clock;
359 
360 	/* Turn off the clock. */
361 	WR2(slot, SDHCI_CLOCK_CONTROL, 0);
362 	/* If no clock requested - left it so. */
363 	if (clock == 0)
364 		return;
365 	/* Looking for highest freq <= clock. */
366 	res = slot->max_clk;
367 	for (clk = 1; clk < 256; clk <<= 1) {
368 		if (res <= clock)
369 			break;
370 		res >>= 1;
371 	}
372 	/* Divider 1:1 is 0x00, 2:1 is 0x01, 256:1 is 0x80 ... */
373 	clk >>= 1;
374 	/* Now we have got divider, set it. */
375 	clk <<= SDHCI_DIVIDER_SHIFT;
376 	WR2(slot, SDHCI_CLOCK_CONTROL, clk);
377 	/* Enable clock. */
378 	clk |= SDHCI_CLOCK_INT_EN;
379 	WR2(slot, SDHCI_CLOCK_CONTROL, clk);
380 	/* Wait up to 10 ms until it stabilize. */
381 	timeout = 10;
382 	while (!((clk = RD2(slot, SDHCI_CLOCK_CONTROL))
383 		& SDHCI_CLOCK_INT_STABLE)) {
384 		if (timeout == 0) {
385 			slot_printf(slot,
386 			    "Internal clock never stabilised.\n");
387 			sdhci_dumpregs(slot);
388 			return;
389 		}
390 		timeout--;
391 		DELAY(1000);
392 	}
393 	/* Pass clock signal to the bus. */
394 	clk |= SDHCI_CLOCK_CARD_EN;
395 	WR2(slot, SDHCI_CLOCK_CONTROL, clk);
396 }
397 
398 static void
399 sdhci_set_power(struct sdhci_slot *slot, u_char power)
400 {
401 	uint8_t pwr;
402 
403 	if (slot->power == power)
404 		return;
405 	slot->power = power;
406 
407 	/* Turn off the power. */
408 	pwr = 0;
409 	WR1(slot, SDHCI_POWER_CONTROL, pwr);
410 	/* If power down requested - left it so. */
411 	if (power == 0)
412 		return;
413 	/* Set voltage. */
414 	switch (1 << power) {
415 	case MMC_OCR_LOW_VOLTAGE:
416 		pwr |= SDHCI_POWER_180;
417 		break;
418 	case MMC_OCR_290_300:
419 	case MMC_OCR_300_310:
420 		pwr |= SDHCI_POWER_300;
421 		break;
422 	case MMC_OCR_320_330:
423 	case MMC_OCR_330_340:
424 		pwr |= SDHCI_POWER_330;
425 		break;
426 	}
427 	WR1(slot, SDHCI_POWER_CONTROL, pwr);
428 	/* Turn on the power. */
429 	pwr |= SDHCI_POWER_ON;
430 	WR1(slot, SDHCI_POWER_CONTROL, pwr);
431 }
432 
433 static void
434 sdhci_read_block_pio(struct sdhci_slot *slot)
435 {
436 	uint32_t data;
437 	char *buffer;
438 	size_t left;
439 
440 	buffer = slot->curcmd->data->data;
441 	buffer += slot->offset;
442 	/* Transfer one block at a time. */
443 	left = min(512, slot->curcmd->data->len - slot->offset);
444 	slot->offset += left;
445 
446 	/* If we are too fast, broken controllers return zeroes. */
447 	if (slot->sc->quirks & SDHCI_QUIRK_BROKEN_TIMINGS)
448 		DELAY(10);
449 	/* Handle unalligned and alligned buffer cases. */
450 	if ((intptr_t)buffer & 3) {
451 		while (left > 3) {
452 			data = RD4(slot, SDHCI_BUFFER);
453 			buffer[0] = data;
454 			buffer[1] = (data >> 8);
455 			buffer[2] = (data >> 16);
456 			buffer[3] = (data >> 24);
457 			buffer += 4;
458 			left -= 4;
459 		}
460 	} else {
461 		bus_read_multi_stream_4(slot->mem_res, SDHCI_BUFFER,
462 		    (uint32_t *)buffer, left >> 2);
463 		left &= 3;
464 	}
465 	/* Handle uneven size case. */
466 	if (left > 0) {
467 		data = RD4(slot, SDHCI_BUFFER);
468 		while (left > 0) {
469 			*(buffer++) = data;
470 			data >>= 8;
471 			left--;
472 		}
473 	}
474 }
475 
476 static void
477 sdhci_write_block_pio(struct sdhci_slot *slot)
478 {
479 	uint32_t data = 0;
480 	char *buffer;
481 	size_t left;
482 
483 	buffer = slot->curcmd->data->data;
484 	buffer += slot->offset;
485 	/* Transfer one block at a time. */
486 	left = min(512, slot->curcmd->data->len - slot->offset);
487 	slot->offset += left;
488 
489 	/* Handle unalligned and alligned buffer cases. */
490 	if ((intptr_t)buffer & 3) {
491 		while (left > 3) {
492 			data = buffer[0] +
493 			    (buffer[1] << 8) +
494 			    (buffer[2] << 16) +
495 			    (buffer[3] << 24);
496 			left -= 4;
497 			buffer += 4;
498 			WR4(slot, SDHCI_BUFFER, data);
499 		}
500 	} else {
501 		bus_write_multi_stream_4(slot->mem_res, SDHCI_BUFFER,
502 		    (uint32_t *)buffer, left >> 2);
503 		left &= 3;
504 	}
505 	/* Handle uneven size case. */
506 	if (left > 0) {
507 		while (left > 0) {
508 			data <<= 8;
509 			data += *(buffer++);
510 			left--;
511 		}
512 		WR4(slot, SDHCI_BUFFER, data);
513 	}
514 }
515 
516 static void
517 sdhci_transfer_pio(struct sdhci_slot *slot)
518 {
519 
520 	/* Read as many blocks as possible. */
521 	if (slot->curcmd->data->flags & MMC_DATA_READ) {
522 		while (RD4(slot, SDHCI_PRESENT_STATE) &
523 		    SDHCI_DATA_AVAILABLE) {
524 			sdhci_read_block_pio(slot);
525 			if (slot->offset >= slot->curcmd->data->len)
526 				break;
527 		}
528 	} else {
529 		while (RD4(slot, SDHCI_PRESENT_STATE) &
530 		    SDHCI_SPACE_AVAILABLE) {
531 			sdhci_write_block_pio(slot);
532 			if (slot->offset >= slot->curcmd->data->len)
533 				break;
534 		}
535 	}
536 }
537 
538 static void
539 sdhci_card_delay(void *arg)
540 {
541 	struct sdhci_slot *slot = arg;
542 
543 	taskqueue_enqueue(taskqueue_swi, &slot->card_task);
544 }
545 
546 static void
547 sdhci_card_task(void *arg, int pending)
548 {
549 	struct sdhci_slot *slot = arg;
550 
551 	SDHCI_LOCK(slot);
552 	if (RD4(slot, SDHCI_PRESENT_STATE) & SDHCI_CARD_PRESENT) {
553 		if (slot->dev == NULL) {
554 			/* If card is present - attach mmc bus. */
555 			slot->dev = device_add_child(slot->sc->dev, "mmc", -1);
556 			device_set_ivars(slot->dev, slot);
557 			SDHCI_UNLOCK(slot);
558 			device_probe_and_attach(slot->dev);
559 		} else
560 			SDHCI_UNLOCK(slot);
561 	} else {
562 		if (slot->dev != NULL) {
563 			/* If no card present - detach mmc bus. */
564 			device_t d = slot->dev;
565 			slot->dev = NULL;
566 			SDHCI_UNLOCK(slot);
567 			device_delete_child(slot->sc->dev, d);
568 		} else
569 			SDHCI_UNLOCK(slot);
570 	}
571 }
572 
573 static int
574 sdhci_probe(device_t dev)
575 {
576 	uint32_t model;
577 	uint16_t subvendor;
578 	uint8_t class, subclass;
579 	int i, result;
580 
581 	model = (uint32_t)pci_get_device(dev) << 16;
582 	model |= (uint32_t)pci_get_vendor(dev) & 0x0000ffff;
583 	subvendor = pci_get_subvendor(dev);
584 	class = pci_get_class(dev);
585 	subclass = pci_get_subclass(dev);
586 
587 	result = ENXIO;
588 	for (i = 0; sdhci_devices[i].model != 0; i++) {
589 		if (sdhci_devices[i].model == model &&
590 		    (sdhci_devices[i].subvendor == 0xffff ||
591 		    sdhci_devices[i].subvendor == subvendor)) {
592 			device_set_desc(dev, sdhci_devices[i].desc);
593 			result = BUS_PROBE_DEFAULT;
594 			break;
595 		}
596 	}
597 	if (result == ENXIO && class == PCIC_BASEPERIPH &&
598 	    subclass == PCIS_BASEPERIPH_SDHC) {
599 		device_set_desc(dev, "Generic SD HCI");
600 		result = BUS_PROBE_GENERIC;
601 	}
602 
603 	return (result);
604 }
605 
606 static int
607 sdhci_attach(device_t dev)
608 {
609 	struct sdhci_softc *sc = device_get_softc(dev);
610 	uint32_t model;
611 	uint16_t subvendor;
612 	uint8_t class, subclass, progif;
613 	int err, slots, bar, i;
614 
615 	sc->dev = dev;
616 	model = (uint32_t)pci_get_device(dev) << 16;
617 	model |= (uint32_t)pci_get_vendor(dev) & 0x0000ffff;
618 	subvendor = pci_get_subvendor(dev);
619 	class = pci_get_class(dev);
620 	subclass = pci_get_subclass(dev);
621 	progif = pci_get_progif(dev);
622 	/* Apply chip specific quirks. */
623 	for (i = 0; sdhci_devices[i].model != 0; i++) {
624 		if (sdhci_devices[i].model == model &&
625 		    (sdhci_devices[i].subvendor == 0xffff ||
626 		    sdhci_devices[i].subvendor == subvendor)) {
627 			sc->quirks = sdhci_devices[i].quirks;
628 			break;
629 		}
630 	}
631 	/* Read slots info from PCI registers. */
632 	slots = pci_read_config(dev, PCI_SLOT_INFO, 1);
633 	bar = PCI_SLOT_INFO_FIRST_BAR(slots);
634 	slots = PCI_SLOT_INFO_SLOTS(slots);
635 	if (slots > 6 || bar > 5) {
636 		device_printf(dev, "Incorrect slots information (%d, %d).\n",
637 		    slots, bar);
638 		return (EINVAL);
639 	}
640 	/* Allocate IRQ. */
641 	sc->irq_rid = 0;
642 	sc->irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &sc->irq_rid,
643 	    RF_SHAREABLE | RF_ACTIVE);
644 	if (sc->irq_res == NULL) {
645 		device_printf(dev, "Can't allocate IRQ\n");
646 		return (ENOMEM);
647 	}
648 	/* Scan all slots. */
649 	for (i = 0; i < slots; i++) {
650 		struct sdhci_slot *slot = &sc->slots[sc->num_slots];
651 		uint32_t caps;
652 
653 		SDHCI_LOCK_INIT(slot);
654 		slot->sc = sc;
655 		slot->num = sc->num_slots;
656 		/* Allocate memory. */
657 		slot->mem_rid = PCIR_BAR(bar + i);
658 		slot->mem_res = bus_alloc_resource(dev,
659 		    SYS_RES_MEMORY, &slot->mem_rid, 0ul, ~0ul, 0x100, RF_ACTIVE);
660 		if (slot->mem_res == NULL) {
661 			device_printf(dev, "Can't allocate memory\n");
662 			SDHCI_LOCK_DESTROY(slot);
663 			continue;
664 		}
665 		/* Allocate DMA tag. */
666 		err = bus_dma_tag_create(NULL,
667 		    DMA_BLOCK_SIZE, 0, BUS_SPACE_MAXADDR_32BIT,
668 		    BUS_SPACE_MAXADDR, NULL, NULL,
669 		    DMA_BLOCK_SIZE, 1, DMA_BLOCK_SIZE,
670 		    BUS_DMA_ALLOCNOW, &slot->dmatag);
671 		if (err != 0) {
672 			device_printf(dev, "Can't create DMA tag\n");
673 			SDHCI_LOCK_DESTROY(slot);
674 			continue;
675 		}
676 		/* Allocate DMA memory. */
677 		err = bus_dmamem_alloc(slot->dmatag, (void **)&slot->dmamem,
678 		    BUS_DMA_NOWAIT, &slot->dmamap);
679 		if (err != 0) {
680 			device_printf(dev, "Can't alloc DMA memory\n");
681 			SDHCI_LOCK_DESTROY(slot);
682 			continue;
683 		}
684 		/* Map the memory. */
685 		err = bus_dmamap_load(slot->dmatag, slot->dmamap,
686 		    (void *)slot->dmamem, DMA_BLOCK_SIZE,
687 		    sdhci_getaddr, &slot->paddr, 0);
688 		if (err != 0 || slot->paddr == 0) {
689 			device_printf(dev, "Can't load DMA memory\n");
690 			SDHCI_LOCK_DESTROY(slot);
691 			continue;
692 		}
693 		/* Initialize slot. */
694 		sdhci_init(slot);
695 		caps = RD4(slot, SDHCI_CAPABILITIES);
696 		/* Calculate base clock frequency. */
697 		slot->max_clk =
698 			(caps & SDHCI_CLOCK_BASE_MASK) >> SDHCI_CLOCK_BASE_SHIFT;
699 		if (slot->max_clk == 0) {
700 			device_printf(dev, "Hardware doesn't specify base clock "
701 			    "frequency.\n");
702 		}
703 		slot->max_clk *= 1000000;
704 		/* Calculate timeout clock frequency. */
705 		slot->timeout_clk =
706 			(caps & SDHCI_TIMEOUT_CLK_MASK) >> SDHCI_TIMEOUT_CLK_SHIFT;
707 		if (slot->timeout_clk == 0) {
708 			device_printf(dev, "Hardware doesn't specify timeout clock "
709 			    "frequency.\n");
710 		}
711 		if (caps & SDHCI_TIMEOUT_CLK_UNIT)
712 			slot->timeout_clk *= 1000;
713 
714 		slot->host.f_min = slot->max_clk / 256;
715 		slot->host.f_max = slot->max_clk;
716 		slot->host.host_ocr = 0;
717 		if (caps & SDHCI_CAN_VDD_330)
718 		    slot->host.host_ocr |= MMC_OCR_320_330 | MMC_OCR_330_340;
719 		if (caps & SDHCI_CAN_VDD_300)
720 		    slot->host.host_ocr |= MMC_OCR_290_300 | MMC_OCR_300_310;
721 		if (caps & SDHCI_CAN_VDD_180)
722 		    slot->host.host_ocr |= MMC_OCR_LOW_VOLTAGE;
723 		if (slot->host.host_ocr == 0) {
724 			device_printf(dev, "Hardware doesn't report any "
725 			    "support voltages.\n");
726 		}
727 		slot->host.caps = MMC_CAP_4_BIT_DATA;
728 		if (caps & SDHCI_CAN_DO_HISPD)
729 			slot->host.caps |= MMC_CAP_HSPEED;
730 		/* Decide if we have usable DMA. */
731 		if (caps & SDHCI_CAN_DO_DMA)
732 			slot->opt |= SDHCI_HAVE_DMA;
733 		if (class == PCIC_BASEPERIPH &&
734 		    subclass == PCIS_BASEPERIPH_SDHC &&
735 		    progif != PCI_SDHCI_IFDMA)
736 			slot->opt &= ~SDHCI_HAVE_DMA;
737 		if (sc->quirks & SDHCI_QUIRK_BROKEN_DMA)
738 			slot->opt &= ~SDHCI_HAVE_DMA;
739 		if (sc->quirks & SDHCI_QUIRK_FORCE_DMA)
740 			slot->opt |= SDHCI_HAVE_DMA;
741 
742 		if (bootverbose || sdhci_debug) {
743 			slot_printf(slot, "%uMHz%s 4bits%s%s%s %s\n",
744 			    slot->max_clk / 1000000,
745 			    (caps & SDHCI_CAN_DO_HISPD) ? " HS" : "",
746 			    (caps & SDHCI_CAN_VDD_330) ? " 3.3V" : "",
747 			    (caps & SDHCI_CAN_VDD_300) ? " 3.0V" : "",
748 			    (caps & SDHCI_CAN_VDD_180) ? " 1.8V" : "",
749 			    (slot->opt & SDHCI_HAVE_DMA) ? "DMA" : "PIO");
750 			sdhci_dumpregs(slot);
751 		}
752 
753 		TASK_INIT(&slot->card_task, 0, sdhci_card_task, slot);
754 		callout_init(&slot->card_callout);
755 		sc->num_slots++;
756 	}
757 	device_printf(dev, "%d slot(s) allocated\n", sc->num_slots);
758 	/* Activate the interrupt */
759 	err = bus_setup_intr(dev, sc->irq_res, INTR_MPSAFE, sdhci_intr, sc, &sc->intrhand, NULL);
760 	if (err)
761 		device_printf(dev, "Can't setup IRQ\n");
762 	pci_enable_busmaster(dev);
763 	/* Process cards detection. */
764 	for (i = 0; i < sc->num_slots; i++) {
765 		struct sdhci_slot *slot = &sc->slots[i];
766 
767 		sdhci_card_task(slot, 0);
768 	}
769 
770 	return (0);
771 }
772 
773 static int
774 sdhci_detach(device_t dev)
775 {
776 	struct sdhci_softc *sc = device_get_softc(dev);
777 	int i;
778 
779 	bus_teardown_intr(dev, sc->irq_res, sc->intrhand);
780 	bus_release_resource(dev, SYS_RES_IRQ,
781 	    sc->irq_rid, sc->irq_res);
782 
783 	for (i = 0; i < sc->num_slots; i++) {
784 		struct sdhci_slot *slot = &sc->slots[i];
785 		device_t d;
786 
787 #if 0
788 		/* XXX: need implementation of callout_drain or workaround */
789 		callout_drain(&slot->card_callout);
790 #endif
791 		taskqueue_drain(taskqueue_swi, &slot->card_task);
792 
793 		SDHCI_LOCK(slot);
794 		d = slot->dev;
795 		slot->dev = NULL;
796 		SDHCI_UNLOCK(slot);
797 		if (d != NULL)
798 			device_delete_child(dev, d);
799 
800 		SDHCI_LOCK(slot);
801 		sdhci_reset(slot, SDHCI_RESET_ALL);
802 		SDHCI_UNLOCK(slot);
803 		bus_dmamap_unload(slot->dmatag, slot->dmamap);
804 		bus_dmamem_free(slot->dmatag, slot->dmamem, slot->dmamap);
805 		bus_dma_tag_destroy(slot->dmatag);
806 		bus_release_resource(dev, SYS_RES_MEMORY,
807 		    slot->mem_rid, slot->mem_res);
808 		SDHCI_LOCK_DESTROY(slot);
809 	}
810 	return (0);
811 }
812 
813 static int
814 sdhci_suspend(device_t dev)
815 {
816 	struct sdhci_softc *sc = device_get_softc(dev);
817 	int i, err;
818 
819 	err = bus_generic_suspend(dev);
820 	if (err)
821 		return (err);
822 	for (i = 0; i < sc->num_slots; i++)
823 		sdhci_reset(&sc->slots[i], SDHCI_RESET_ALL);
824 	return (0);
825 }
826 
827 static int
828 sdhci_resume(device_t dev)
829 {
830 	struct sdhci_softc *sc = device_get_softc(dev);
831 	int i;
832 
833 	for (i = 0; i < sc->num_slots; i++)
834 		sdhci_init(&sc->slots[i]);
835 	return (bus_generic_resume(dev));
836 }
837 
838 static int
839 sdhci_update_ios(device_t brdev, device_t reqdev)
840 {
841 	struct sdhci_slot *slot = device_get_ivars(reqdev);
842 	struct mmc_ios *ios = &slot->host.ios;
843 
844 	SDHCI_LOCK(slot);
845 	/* Do full reset on bus power down to clear from any state. */
846 	if (ios->power_mode == power_off) {
847 		WR4(slot, SDHCI_SIGNAL_ENABLE, 0);
848 		sdhci_init(slot);
849 	}
850 	/* Configure the bus. */
851 	sdhci_set_clock(slot, ios->clock);
852 	sdhci_set_power(slot, (ios->power_mode == power_off)?0:ios->vdd);
853 	if (ios->bus_width == bus_width_4)
854 		slot->hostctrl |= SDHCI_CTRL_4BITBUS;
855 	else
856 		slot->hostctrl &= ~SDHCI_CTRL_4BITBUS;
857 	if (ios->timing == bus_timing_hs)
858 		slot->hostctrl |= SDHCI_CTRL_HISPD;
859 	else
860 		slot->hostctrl &= ~SDHCI_CTRL_HISPD;
861 	WR1(slot, SDHCI_HOST_CONTROL, slot->hostctrl);
862 	/* Some controllers like reset after bus changes. */
863 	if(slot->sc->quirks & SDHCI_QUIRK_RESET_ON_IOS)
864 		sdhci_reset(slot, SDHCI_RESET_CMD | SDHCI_RESET_DATA);
865 
866 	SDHCI_UNLOCK(slot);
867 	return (0);
868 }
869 
870 static void
871 sdhci_set_transfer_mode(struct sdhci_slot *slot,
872 	struct mmc_data *data)
873 {
874 	uint16_t mode;
875 
876 	if (data == NULL)
877 		return;
878 
879 	mode = SDHCI_TRNS_BLK_CNT_EN;
880 	if (data->len > 512)
881 		mode |= SDHCI_TRNS_MULTI;
882 	if (data->flags & MMC_DATA_READ)
883 		mode |= SDHCI_TRNS_READ;
884 	if (slot->req->stop)
885 		mode |= SDHCI_TRNS_ACMD12;
886 	if (slot->flags & SDHCI_USE_DMA)
887 		mode |= SDHCI_TRNS_DMA;
888 
889 	WR2(slot, SDHCI_TRANSFER_MODE, mode);
890 }
891 
892 static void
893 sdhci_start_command(struct sdhci_slot *slot, struct mmc_command *cmd)
894 {
895 	struct mmc_request *req = slot->req;
896 	int flags, timeout;
897 	uint32_t mask, state;
898 
899 	slot->curcmd = cmd;
900 	slot->cmd_done = 0;
901 
902 	cmd->error = MMC_ERR_NONE;
903 
904 	/* This flags combination is not supported by controller. */
905 	if ((cmd->flags & MMC_RSP_136) && (cmd->flags & MMC_RSP_BUSY)) {
906 		slot_printf(slot, "Unsupported response type!\n");
907 		cmd->error = MMC_ERR_FAILED;
908 		slot->req = NULL;
909 		slot->curcmd = NULL;
910 		req->done(req);
911 		return;
912 	}
913 
914 	/* Read controller present state. */
915 	state = RD4(slot, SDHCI_PRESENT_STATE);
916 	/* Do not issue command if there is no card, clock or power.
917 	 * Controller will not detect timeout without clock active. */
918 	if ((state & SDHCI_CARD_PRESENT) == 0 ||
919 	    slot->power == 0 ||
920 	    slot->clock == 0) {
921 		cmd->error = MMC_ERR_FAILED;
922 		slot->req = NULL;
923 		slot->curcmd = NULL;
924 		req->done(req);
925 		return;
926 	}
927 	/* Always wait for free CMD bus. */
928 	mask = SDHCI_CMD_INHIBIT;
929 	/* Wait for free DAT if we have data or busy signal. */
930 	if (cmd->data || (cmd->flags & MMC_RSP_BUSY))
931 		mask |= SDHCI_DAT_INHIBIT;
932 	/* We shouldn't wait for DAT for stop commands. */
933 	if (cmd == slot->req->stop)
934 		mask &= ~SDHCI_DAT_INHIBIT;
935 	/* Wait for bus no more then 10 ms. */
936 	timeout = 10;
937 	while (state & mask) {
938 		if (timeout == 0) {
939 			slot_printf(slot, "Controller never released "
940 			    "inhibit bit(s).\n");
941 			sdhci_dumpregs(slot);
942 			cmd->error = MMC_ERR_FAILED;
943 			slot->req = NULL;
944 			slot->curcmd = NULL;
945 			req->done(req);
946 			return;
947 		}
948 		timeout--;
949 		DELAY(1000);
950 		state = RD4(slot, SDHCI_PRESENT_STATE);
951 	}
952 
953 	/* Prepare command flags. */
954 	if (!(cmd->flags & MMC_RSP_PRESENT))
955 		flags = SDHCI_CMD_RESP_NONE;
956 	else if (cmd->flags & MMC_RSP_136)
957 		flags = SDHCI_CMD_RESP_LONG;
958 	else if (cmd->flags & MMC_RSP_BUSY)
959 		flags = SDHCI_CMD_RESP_SHORT_BUSY;
960 	else
961 		flags = SDHCI_CMD_RESP_SHORT;
962 	if (cmd->flags & MMC_RSP_CRC)
963 		flags |= SDHCI_CMD_CRC;
964 	if (cmd->flags & MMC_RSP_OPCODE)
965 		flags |= SDHCI_CMD_INDEX;
966 	if (cmd->data)
967 		flags |= SDHCI_CMD_DATA;
968 	if (cmd->opcode == MMC_STOP_TRANSMISSION)
969 		flags |= SDHCI_CMD_TYPE_ABORT;
970 	/* Prepare data. */
971 	sdhci_start_data(slot, cmd->data);
972 	/*
973 	 * Interrupt aggregation: To reduce total number of interrupts
974 	 * group response interrupt with data interrupt when possible.
975 	 * If there going to be data interrupt, mask response one.
976 	 */
977 	if (slot->data_done == 0) {
978 		WR4(slot, SDHCI_SIGNAL_ENABLE,
979 		    slot->intmask &= ~SDHCI_INT_RESPONSE);
980 	}
981 	/* Set command argument. */
982 	WR4(slot, SDHCI_ARGUMENT, cmd->arg);
983 	/* Set data transfer mode. */
984 	sdhci_set_transfer_mode(slot, cmd->data);
985 	/* Set command flags. */
986 	WR1(slot, SDHCI_COMMAND_FLAGS, flags);
987 	/* Start command. */
988 	WR1(slot, SDHCI_COMMAND, cmd->opcode);
989 }
990 
991 static void
992 sdhci_finish_command(struct sdhci_slot *slot)
993 {
994 	int i;
995 
996 	slot->cmd_done = 1;
997 	/* Interrupt aggregation: Restore command interrupt.
998 	 * Main restore point for the case when command interrupt
999 	 * happened first. */
1000 	WR4(slot, SDHCI_SIGNAL_ENABLE, slot->intmask |= SDHCI_INT_RESPONSE);
1001 	/* In case of error - reset host and return. */
1002 	if (slot->curcmd->error) {
1003 		sdhci_reset(slot, SDHCI_RESET_CMD);
1004 		sdhci_reset(slot, SDHCI_RESET_DATA);
1005 		sdhci_start(slot);
1006 		return;
1007 	}
1008 	/* If command has response - fetch it. */
1009 	if (slot->curcmd->flags & MMC_RSP_PRESENT) {
1010 		if (slot->curcmd->flags & MMC_RSP_136) {
1011 			/* CRC is stripped so we need one byte shift. */
1012 			uint8_t extra = 0;
1013 			for (i = 0; i < 4; i++) {
1014 				uint32_t val = RD4(slot, SDHCI_RESPONSE + i * 4);
1015 				slot->curcmd->resp[3 - i] = (val << 8) + extra;
1016 				extra = val >> 24;
1017 			}
1018 		} else
1019 			slot->curcmd->resp[0] = RD4(slot, SDHCI_RESPONSE);
1020 	}
1021 	/* If data ready - finish. */
1022 	if (slot->data_done)
1023 		sdhci_start(slot);
1024 }
1025 
1026 static void
1027 sdhci_start_data(struct sdhci_slot *slot, struct mmc_data *data)
1028 {
1029 	uint32_t target_timeout, current_timeout;
1030 	uint8_t div;
1031 
1032 	if (data == NULL && (slot->curcmd->flags & MMC_RSP_BUSY) == 0) {
1033 		slot->data_done = 1;
1034 		return;
1035 	}
1036 
1037 	slot->data_done = 0;
1038 
1039 	/* Calculate and set data timeout.*/
1040 	/* XXX: We should have this from mmc layer, now assume 1 sec. */
1041 	target_timeout = 1000000;
1042 	div = 0;
1043 	current_timeout = (1 << 13) * 1000 / slot->timeout_clk;
1044 	while (current_timeout < target_timeout) {
1045 		div++;
1046 		current_timeout <<= 1;
1047 		if (div >= 0xF)
1048 			break;
1049 	}
1050 	/* Compensate for an off-by-one error in the CaFe chip.*/
1051 	if (slot->sc->quirks & SDHCI_QUIRK_INCR_TIMEOUT_CONTROL)
1052 		div++;
1053 	if (div >= 0xF) {
1054 		slot_printf(slot, "Timeout too large!\n");
1055 		div = 0xE;
1056 	}
1057 	WR1(slot, SDHCI_TIMEOUT_CONTROL, div);
1058 
1059 	if (data == NULL)
1060 		return;
1061 
1062 	/* Use DMA if possible. */
1063 	if ((slot->opt & SDHCI_HAVE_DMA))
1064 		slot->flags |= SDHCI_USE_DMA;
1065 	/* If data is small, broken DMA may return zeroes instead of data, */
1066 	if ((slot->sc->quirks & SDHCI_QUIRK_BROKEN_TIMINGS) &&
1067 	    (data->len <= 512))
1068 		slot->flags &= ~SDHCI_USE_DMA;
1069 	/* Some controllers require even block sizes. */
1070 	if ((slot->sc->quirks & SDHCI_QUIRK_32BIT_DMA_SIZE) &&
1071 	    ((data->len) & 0x3))
1072 		slot->flags &= ~SDHCI_USE_DMA;
1073 	/* Load DMA buffer. */
1074 	if (slot->flags & SDHCI_USE_DMA) {
1075 		if (data->flags & MMC_DATA_READ) {
1076 			bus_dmamap_sync(slot->dmatag, slot->dmamap, BUS_DMASYNC_PREREAD);
1077 		} else {
1078 			memcpy(slot->dmamem, data->data,
1079 			    (data->len < DMA_BLOCK_SIZE)?data->len:DMA_BLOCK_SIZE);
1080 			bus_dmamap_sync(slot->dmatag, slot->dmamap, BUS_DMASYNC_PREWRITE);
1081 		}
1082 		WR4(slot, SDHCI_DMA_ADDRESS, slot->paddr);
1083 		/* Interrupt aggregation: Mask border interrupt
1084 		 * for the last page and unmask else. */
1085 		if (data->len == DMA_BLOCK_SIZE)
1086 			slot->intmask &= ~SDHCI_INT_DMA_END;
1087 		else
1088 			slot->intmask |= SDHCI_INT_DMA_END;
1089 		WR4(slot, SDHCI_SIGNAL_ENABLE, slot->intmask);
1090 	}
1091 	/* Current data offset for both PIO and DMA. */
1092 	slot->offset = 0;
1093 	/* Set block size and request IRQ on 4K border. */
1094 	WR2(slot, SDHCI_BLOCK_SIZE,
1095 	    SDHCI_MAKE_BLKSZ(DMA_BOUNDARY, (data->len < 512)?data->len:512));
1096 	/* Set block count. */
1097 	WR2(slot, SDHCI_BLOCK_COUNT, (data->len + 511) / 512);
1098 }
1099 
1100 static void
1101 sdhci_finish_data(struct sdhci_slot *slot)
1102 {
1103 	struct mmc_data *data = slot->curcmd->data;
1104 
1105 	slot->data_done = 1;
1106 	/* Interrupt aggregation: Restore command interrupt.
1107 	 * Auxillary restore point for the case when data interrupt
1108 	 * happened first. */
1109 	if (!slot->cmd_done) {
1110 		WR4(slot, SDHCI_SIGNAL_ENABLE,
1111 		    slot->intmask |= SDHCI_INT_RESPONSE);
1112 	}
1113 	/* Unload rest of data from DMA buffer. */
1114 	if (slot->flags & SDHCI_USE_DMA) {
1115 		if (data->flags & MMC_DATA_READ) {
1116 			size_t left = data->len - slot->offset;
1117 			bus_dmamap_sync(slot->dmatag, slot->dmamap, BUS_DMASYNC_POSTREAD);
1118 			memcpy((u_char*)data->data + slot->offset, slot->dmamem,
1119 			    (left < DMA_BLOCK_SIZE)?left:DMA_BLOCK_SIZE);
1120 		} else
1121 			bus_dmamap_sync(slot->dmatag, slot->dmamap, BUS_DMASYNC_POSTWRITE);
1122 	}
1123 	/* If there was error - reset the host. */
1124 	if (slot->curcmd->error) {
1125 		sdhci_reset(slot, SDHCI_RESET_CMD);
1126 		sdhci_reset(slot, SDHCI_RESET_DATA);
1127 		sdhci_start(slot);
1128 		return;
1129 	}
1130 	/* If we already have command response - finish. */
1131 	if (slot->cmd_done)
1132 		sdhci_start(slot);
1133 }
1134 
1135 static void
1136 sdhci_start(struct sdhci_slot *slot)
1137 {
1138 	struct mmc_request *req;
1139 
1140 	req = slot->req;
1141 	if (req == NULL)
1142 		return;
1143 
1144 	if (!(slot->flags & CMD_STARTED)) {
1145 		slot->flags |= CMD_STARTED;
1146 		sdhci_start_command(slot, req->cmd);
1147 		return;
1148 	}
1149 /*	We don't need this until using Auto-CMD12 feature
1150 	if (!(slot->flags & STOP_STARTED) && req->stop) {
1151 		slot->flags |= STOP_STARTED;
1152 		sdhci_start_command(slot, req->stop);
1153 		return;
1154 	}
1155 */
1156 	if (sdhci_debug > 1)
1157 		slot_printf(slot, "result: %d\n", req->cmd->error);
1158 	if (!req->cmd->error &&
1159 	    (slot->sc->quirks & SDHCI_QUIRK_RESET_AFTER_REQUEST)) {
1160 		sdhci_reset(slot, SDHCI_RESET_CMD);
1161 		sdhci_reset(slot, SDHCI_RESET_DATA);
1162 	}
1163 
1164 	/* We must be done -- bad idea to do this while locked? */
1165 	slot->req = NULL;
1166 	slot->curcmd = NULL;
1167 	req->done(req);
1168 }
1169 
1170 static int
1171 sdhci_request(device_t brdev, device_t reqdev, struct mmc_request *req)
1172 {
1173 	struct sdhci_slot *slot = device_get_ivars(reqdev);
1174 
1175 	SDHCI_LOCK(slot);
1176 	if (slot->req != NULL) {
1177 		SDHCI_UNLOCK(slot);
1178 		return (EBUSY);
1179 	}
1180 	if (sdhci_debug > 1) {
1181 		slot_printf(slot, "CMD%u arg %#x flags %#x dlen %u dflags %#x\n",
1182 		    req->cmd->opcode, req->cmd->arg, req->cmd->flags,
1183 		    (req->cmd->data)?(u_int)req->cmd->data->len:0,
1184 		    (req->cmd->data)?req->cmd->data->flags:0);
1185 	}
1186 	slot->req = req;
1187 	slot->flags = 0;
1188 	sdhci_start(slot);
1189 	SDHCI_UNLOCK(slot);
1190 	if (dumping) {
1191 		while (slot->req != NULL) {
1192 			sdhci_intr(slot->sc);
1193 			DELAY(10);
1194 		}
1195 	}
1196 	return (0);
1197 }
1198 
1199 static int
1200 sdhci_get_ro(device_t brdev, device_t reqdev)
1201 {
1202 	struct sdhci_slot *slot = device_get_ivars(reqdev);
1203 	uint32_t val;
1204 
1205 	SDHCI_LOCK(slot);
1206 	val = RD4(slot, SDHCI_PRESENT_STATE);
1207 	SDHCI_UNLOCK(slot);
1208 	return (!(val & SDHCI_WRITE_PROTECT));
1209 }
1210 
1211 static int
1212 sdhci_acquire_host(device_t brdev, device_t reqdev)
1213 {
1214 	struct sdhci_slot *slot = device_get_ivars(reqdev);
1215 	int err = 0;
1216 
1217 	SDHCI_LOCK(slot);
1218 	while (slot->bus_busy)
1219 		lksleep(slot, &slot->lock, 0, "sdhciah", 0);
1220 	slot->bus_busy++;
1221 	/* Activate led. */
1222 	WR1(slot, SDHCI_HOST_CONTROL, slot->hostctrl |= SDHCI_CTRL_LED);
1223 	SDHCI_UNLOCK(slot);
1224 	return (err);
1225 }
1226 
1227 static int
1228 sdhci_release_host(device_t brdev, device_t reqdev)
1229 {
1230 	struct sdhci_slot *slot = device_get_ivars(reqdev);
1231 
1232 	SDHCI_LOCK(slot);
1233 	/* Deactivate led. */
1234 	WR1(slot, SDHCI_HOST_CONTROL, slot->hostctrl &= ~SDHCI_CTRL_LED);
1235 	slot->bus_busy--;
1236 	SDHCI_UNLOCK(slot);
1237 	wakeup(slot);
1238 	return (0);
1239 }
1240 
1241 static void
1242 sdhci_cmd_irq(struct sdhci_slot *slot, uint32_t intmask)
1243 {
1244 
1245 	if (!slot->curcmd) {
1246 		slot_printf(slot, "Got command interrupt 0x%08x, but "
1247 		    "there is no active command.\n", intmask);
1248 		sdhci_dumpregs(slot);
1249 		return;
1250 	}
1251 	if (intmask & SDHCI_INT_TIMEOUT)
1252 		slot->curcmd->error = MMC_ERR_TIMEOUT;
1253 	else if (intmask & SDHCI_INT_CRC)
1254 		slot->curcmd->error = MMC_ERR_BADCRC;
1255 	else if (intmask & (SDHCI_INT_END_BIT | SDHCI_INT_INDEX))
1256 		slot->curcmd->error = MMC_ERR_FIFO;
1257 
1258 	sdhci_finish_command(slot);
1259 }
1260 
1261 static void
1262 sdhci_data_irq(struct sdhci_slot *slot, uint32_t intmask)
1263 {
1264 
1265 	if (!slot->curcmd) {
1266 		slot_printf(slot, "Got data interrupt 0x%08x, but "
1267 		    "there is no active command.\n", intmask);
1268 		sdhci_dumpregs(slot);
1269 		return;
1270 	}
1271 	if (slot->curcmd->data == NULL &&
1272 	    (slot->curcmd->flags & MMC_RSP_BUSY) == 0) {
1273 		slot_printf(slot, "Got data interrupt 0x%08x, but "
1274 		    "there is no active data operation.\n",
1275 		    intmask);
1276 		sdhci_dumpregs(slot);
1277 		return;
1278 	}
1279 	if (intmask & SDHCI_INT_DATA_TIMEOUT)
1280 		slot->curcmd->error = MMC_ERR_TIMEOUT;
1281 	else if (intmask & (SDHCI_INT_DATA_CRC | SDHCI_INT_DATA_END_BIT))
1282 		slot->curcmd->error = MMC_ERR_BADCRC;
1283 	if (slot->curcmd->data == NULL &&
1284 	    (intmask & (SDHCI_INT_DATA_AVAIL | SDHCI_INT_SPACE_AVAIL |
1285 	    SDHCI_INT_DMA_END))) {
1286 		slot_printf(slot, "Got data interrupt 0x%08x, but "
1287 		    "there is busy-only command.\n", intmask);
1288 		sdhci_dumpregs(slot);
1289 		slot->curcmd->error = MMC_ERR_INVALID;
1290 	}
1291 	if (slot->curcmd->error) {
1292 		/* No need to continue after any error. */
1293 		sdhci_finish_data(slot);
1294 		return;
1295 	}
1296 
1297 	/* Handle PIO interrupt. */
1298 	if (intmask & (SDHCI_INT_DATA_AVAIL | SDHCI_INT_SPACE_AVAIL))
1299 		sdhci_transfer_pio(slot);
1300 	/* Handle DMA border. */
1301 	if (intmask & SDHCI_INT_DMA_END) {
1302 		struct mmc_data *data = slot->curcmd->data;
1303 		size_t left;
1304 
1305 		/* Unload DMA buffer... */
1306 		left = data->len - slot->offset;
1307 		if (data->flags & MMC_DATA_READ) {
1308 			bus_dmamap_sync(slot->dmatag, slot->dmamap,
1309 			    BUS_DMASYNC_POSTREAD);
1310 			memcpy((u_char*)data->data + slot->offset, slot->dmamem,
1311 			    (left < DMA_BLOCK_SIZE)?left:DMA_BLOCK_SIZE);
1312 		} else {
1313 			bus_dmamap_sync(slot->dmatag, slot->dmamap,
1314 			    BUS_DMASYNC_POSTWRITE);
1315 		}
1316 		/* ... and reload it again. */
1317 		slot->offset += DMA_BLOCK_SIZE;
1318 		left = data->len - slot->offset;
1319 		if (data->flags & MMC_DATA_READ) {
1320 			bus_dmamap_sync(slot->dmatag, slot->dmamap,
1321 			    BUS_DMASYNC_PREREAD);
1322 		} else {
1323 			memcpy(slot->dmamem, (u_char*)data->data + slot->offset,
1324 			    (left < DMA_BLOCK_SIZE)?left:DMA_BLOCK_SIZE);
1325 			bus_dmamap_sync(slot->dmatag, slot->dmamap,
1326 			    BUS_DMASYNC_PREWRITE);
1327 		}
1328 		/* Interrupt aggregation: Mask border interrupt
1329 		 * for the last page. */
1330 		if (left == DMA_BLOCK_SIZE) {
1331 			slot->intmask &= ~SDHCI_INT_DMA_END;
1332 			WR4(slot, SDHCI_SIGNAL_ENABLE, slot->intmask);
1333 		}
1334 		/* Restart DMA. */
1335 		WR4(slot, SDHCI_DMA_ADDRESS, slot->paddr);
1336 	}
1337 	/* We have got all data. */
1338 	if (intmask & SDHCI_INT_DATA_END)
1339 		sdhci_finish_data(slot);
1340 }
1341 
1342 static void
1343 sdhci_acmd_irq(struct sdhci_slot *slot)
1344 {
1345 	uint16_t err;
1346 
1347 	err = RD4(slot, SDHCI_ACMD12_ERR);
1348 	if (!slot->curcmd) {
1349 		slot_printf(slot, "Got AutoCMD12 error 0x%04x, but "
1350 		    "there is no active command.\n", err);
1351 		sdhci_dumpregs(slot);
1352 		return;
1353 	}
1354 	slot_printf(slot, "Got AutoCMD12 error 0x%04x\n", err);
1355 	sdhci_reset(slot, SDHCI_RESET_CMD);
1356 }
1357 
1358 static void
1359 sdhci_intr(void *arg)
1360 {
1361 	struct sdhci_softc *sc = (struct sdhci_softc *)arg;
1362 	int i;
1363 
1364 	for (i = 0; i < sc->num_slots; i++) {
1365 		struct sdhci_slot *slot = &sc->slots[i];
1366 		uint32_t intmask;
1367 
1368 		SDHCI_LOCK(slot);
1369 		/* Read slot interrupt status. */
1370 		intmask = RD4(slot, SDHCI_INT_STATUS);
1371 		if (intmask == 0 || intmask == 0xffffffff) {
1372 			SDHCI_UNLOCK(slot);
1373 			continue;
1374 		}
1375 		if (sdhci_debug > 2)
1376 			slot_printf(slot, "Interrupt %#x\n", intmask);
1377 
1378 		/* Handle card presence interrupts. */
1379 		if (intmask & (SDHCI_INT_CARD_INSERT | SDHCI_INT_CARD_REMOVE)) {
1380 			WR4(slot, SDHCI_INT_STATUS, intmask &
1381 			    (SDHCI_INT_CARD_INSERT | SDHCI_INT_CARD_REMOVE));
1382 
1383 			if (intmask & SDHCI_INT_CARD_REMOVE) {
1384 				if (bootverbose || sdhci_debug)
1385 					slot_printf(slot, "Card removed\n");
1386 				callout_stop(&slot->card_callout);
1387 				taskqueue_enqueue(taskqueue_swi,
1388 				    &slot->card_task);
1389 			}
1390 			if (intmask & SDHCI_INT_CARD_INSERT) {
1391 				if (bootverbose || sdhci_debug)
1392 					slot_printf(slot, "Card inserted\n");
1393 				callout_reset(&slot->card_callout, hz / 2,
1394 				    sdhci_card_delay, slot);
1395 			}
1396 			intmask &= ~(SDHCI_INT_CARD_INSERT | SDHCI_INT_CARD_REMOVE);
1397 		}
1398 		/* Handle command interrupts. */
1399 		if (intmask & SDHCI_INT_CMD_MASK) {
1400 			WR4(slot, SDHCI_INT_STATUS, intmask & SDHCI_INT_CMD_MASK);
1401 			sdhci_cmd_irq(slot, intmask & SDHCI_INT_CMD_MASK);
1402 		}
1403 		/* Handle data interrupts. */
1404 		if (intmask & SDHCI_INT_DATA_MASK) {
1405 			WR4(slot, SDHCI_INT_STATUS, intmask & SDHCI_INT_DATA_MASK);
1406 			sdhci_data_irq(slot, intmask & SDHCI_INT_DATA_MASK);
1407 		}
1408 		/* Handle AutoCMD12 error interrupt. */
1409 		if (intmask & SDHCI_INT_ACMD12ERR) {
1410 			WR4(slot, SDHCI_INT_STATUS, SDHCI_INT_ACMD12ERR);
1411 			sdhci_acmd_irq(slot);
1412 		}
1413 		intmask &= ~(SDHCI_INT_CMD_MASK | SDHCI_INT_DATA_MASK);
1414 		intmask &= ~SDHCI_INT_ACMD12ERR;
1415 		intmask &= ~SDHCI_INT_ERROR;
1416 		/* Handle bus power interrupt. */
1417 		if (intmask & SDHCI_INT_BUS_POWER) {
1418 			WR4(slot, SDHCI_INT_STATUS, SDHCI_INT_BUS_POWER);
1419 			slot_printf(slot,
1420 			    "Card is consuming too much power!\n");
1421 			intmask &= ~SDHCI_INT_BUS_POWER;
1422 		}
1423 		/* The rest is unknown. */
1424 		if (intmask) {
1425 			WR4(slot, SDHCI_INT_STATUS, intmask);
1426 			slot_printf(slot, "Unexpected interrupt 0x%08x.\n",
1427 			    intmask);
1428 			sdhci_dumpregs(slot);
1429 		}
1430 
1431 		SDHCI_UNLOCK(slot);
1432 	}
1433 }
1434 
1435 static int
1436 sdhci_read_ivar(device_t bus, device_t child, int which, uintptr_t *result)
1437 {
1438 	struct sdhci_slot *slot = device_get_ivars(child);
1439 
1440 	switch (which) {
1441 	default:
1442 		return (EINVAL);
1443 	case MMCBR_IVAR_BUS_MODE:
1444 		*(int *)result = slot->host.ios.bus_mode;
1445 		break;
1446 	case MMCBR_IVAR_BUS_WIDTH:
1447 		*(int *)result = slot->host.ios.bus_width;
1448 		break;
1449 	case MMCBR_IVAR_CHIP_SELECT:
1450 		*(int *)result = slot->host.ios.chip_select;
1451 		break;
1452 	case MMCBR_IVAR_CLOCK:
1453 		*(int *)result = slot->host.ios.clock;
1454 		break;
1455 	case MMCBR_IVAR_F_MIN:
1456 		*(int *)result = slot->host.f_min;
1457 		break;
1458 	case MMCBR_IVAR_F_MAX:
1459 		*(int *)result = slot->host.f_max;
1460 		break;
1461 	case MMCBR_IVAR_HOST_OCR:
1462 		*(int *)result = slot->host.host_ocr;
1463 		break;
1464 	case MMCBR_IVAR_MODE:
1465 		*(int *)result = slot->host.mode;
1466 		break;
1467 	case MMCBR_IVAR_OCR:
1468 		*(int *)result = slot->host.ocr;
1469 		break;
1470 	case MMCBR_IVAR_POWER_MODE:
1471 		*(int *)result = slot->host.ios.power_mode;
1472 		break;
1473 	case MMCBR_IVAR_VDD:
1474 		*(int *)result = slot->host.ios.vdd;
1475 		break;
1476 	case MMCBR_IVAR_CAPS:
1477 		*(int *)result = slot->host.caps;
1478 		break;
1479 	case MMCBR_IVAR_TIMING:
1480 		*(int *)result = slot->host.ios.timing;
1481 		break;
1482 	case MMCBR_IVAR_MAX_DATA:
1483 		*(int *)result = 65535;
1484 		break;
1485 	}
1486 	return (0);
1487 }
1488 
1489 static int
1490 sdhci_write_ivar(device_t bus, device_t child, int which, uintptr_t value)
1491 {
1492 	struct sdhci_slot *slot = device_get_ivars(child);
1493 
1494 	switch (which) {
1495 	default:
1496 		return (EINVAL);
1497 	case MMCBR_IVAR_BUS_MODE:
1498 		slot->host.ios.bus_mode = value;
1499 		break;
1500 	case MMCBR_IVAR_BUS_WIDTH:
1501 		slot->host.ios.bus_width = value;
1502 		break;
1503 	case MMCBR_IVAR_CHIP_SELECT:
1504 		slot->host.ios.chip_select = value;
1505 		break;
1506 	case MMCBR_IVAR_CLOCK:
1507 		if (value > 0) {
1508 			uint32_t clock = slot->max_clk;
1509 			int i;
1510 
1511 			for (i = 0; i < 8; i++) {
1512 				if (clock <= value)
1513 					break;
1514 				clock >>= 1;
1515 			}
1516 			slot->host.ios.clock = clock;
1517 		} else
1518 			slot->host.ios.clock = 0;
1519 		break;
1520 	case MMCBR_IVAR_MODE:
1521 		slot->host.mode = value;
1522 		break;
1523 	case MMCBR_IVAR_OCR:
1524 		slot->host.ocr = value;
1525 		break;
1526 	case MMCBR_IVAR_POWER_MODE:
1527 		slot->host.ios.power_mode = value;
1528 		break;
1529 	case MMCBR_IVAR_VDD:
1530 		slot->host.ios.vdd = value;
1531 		break;
1532 	case MMCBR_IVAR_TIMING:
1533 		slot->host.ios.timing = value;
1534 		break;
1535 	case MMCBR_IVAR_CAPS:
1536 	case MMCBR_IVAR_HOST_OCR:
1537 	case MMCBR_IVAR_F_MIN:
1538 	case MMCBR_IVAR_F_MAX:
1539 	case MMCBR_IVAR_MAX_DATA:
1540 		return (EINVAL);
1541 	}
1542 	return (0);
1543 }
1544 
1545 static device_method_t sdhci_methods[] = {
1546 	/* device_if */
1547 	DEVMETHOD(device_probe, sdhci_probe),
1548 	DEVMETHOD(device_attach, sdhci_attach),
1549 	DEVMETHOD(device_detach, sdhci_detach),
1550 	DEVMETHOD(device_suspend, sdhci_suspend),
1551 	DEVMETHOD(device_resume, sdhci_resume),
1552 
1553 	/* Bus interface */
1554 	DEVMETHOD(bus_read_ivar,	sdhci_read_ivar),
1555 	DEVMETHOD(bus_write_ivar,	sdhci_write_ivar),
1556 
1557 	/* mmcbr_if */
1558 	DEVMETHOD(mmcbr_update_ios, sdhci_update_ios),
1559 	DEVMETHOD(mmcbr_request, sdhci_request),
1560 	DEVMETHOD(mmcbr_get_ro, sdhci_get_ro),
1561 	DEVMETHOD(mmcbr_acquire_host, sdhci_acquire_host),
1562 	DEVMETHOD(mmcbr_release_host, sdhci_release_host),
1563 
1564 	{0, 0},
1565 };
1566 
1567 static driver_t sdhci_driver = {
1568 	"sdhci",
1569 	sdhci_methods,
1570 	sizeof(struct sdhci_softc),
1571 };
1572 static devclass_t sdhci_devclass;
1573 
1574 
1575 DRIVER_MODULE(sdhci, pci, sdhci_driver, sdhci_devclass, NULL, NULL);
1576