xref: /freebsd/sys/dev/sdhci/fsl_sdhci.c (revision aa1a8ff2d6dbc51ef058f46f3db5a8bb77967145)
1 /*-
2  * Copyright (c) 2013 Ian Lepore <ian@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 AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  */
27 #include <sys/cdefs.h>
28 /*
29  * SDHCI driver glue for Freescale i.MX SoC and QorIQ families.
30  *
31  * This supports both eSDHC (earlier SoCs) and uSDHC (more recent SoCs).
32  */
33 
34 #include "opt_mmccam.h"
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/types.h>
39 #include <sys/bus.h>
40 #include <sys/callout.h>
41 #include <sys/kernel.h>
42 #include <sys/libkern.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/module.h>
46 #include <sys/mutex.h>
47 #include <sys/resource.h>
48 #include <sys/rman.h>
49 #include <sys/sysctl.h>
50 #include <sys/taskqueue.h>
51 #include <sys/time.h>
52 
53 #include <machine/bus.h>
54 #include <machine/resource.h>
55 #ifdef __arm__
56 #include <machine/intr.h>
57 
58 #include <arm/freescale/imx/imx_ccmvar.h>
59 #endif
60 
61 #ifdef __powerpc__
62 #include <powerpc/mpc85xx/mpc85xx.h>
63 #endif
64 
65 #include <dev/gpio/gpiobusvar.h>
66 
67 #include <dev/ofw/ofw_bus.h>
68 #include <dev/ofw/ofw_bus_subr.h>
69 
70 #include <dev/mmc/bridge.h>
71 
72 #include <dev/sdhci/sdhci.h>
73 #include <dev/sdhci/sdhci_fdt_gpio.h>
74 
75 #include "mmcbr_if.h"
76 #include "sdhci_if.h"
77 
78 struct fsl_sdhci_softc {
79 	device_t		dev;
80 	struct resource *	mem_res;
81 	struct resource *	irq_res;
82 	void *			intr_cookie;
83 	struct sdhci_slot	slot;
84 	struct callout		r1bfix_callout;
85 	sbintime_t		r1bfix_timeout_at;
86 	struct sdhci_fdt_gpio * gpio;
87 	uint32_t		baseclk_hz;
88 	uint32_t		cmd_and_mode;
89 	uint32_t		r1bfix_intmask;
90 	uint16_t		sdclockreg_freq_bits;
91 	uint8_t			r1bfix_type;
92 	uint8_t			hwtype;
93 	bool			slot_init_done;
94 };
95 
96 #define	R1BFIX_NONE	0	/* No fix needed at next interrupt. */
97 #define	R1BFIX_NODATA	1	/* Synthesize DATA_END for R1B w/o data. */
98 #define	R1BFIX_AC12	2	/* Wait for busy after auto command 12. */
99 
100 #define	HWTYPE_NONE	0	/* Hardware not recognized/supported. */
101 #define	HWTYPE_ESDHC	1	/* fsl5x and earlier. */
102 #define	HWTYPE_USDHC	2	/* fsl6. */
103 
104 /*
105  * Freescale-specific registers, or in some cases the layout of bits within the
106  * sdhci-defined register is different on Freescale.  These names all begin with
107  * SDHC_ (not SDHCI_).
108  */
109 
110 #define	SDHC_WTMK_LVL		0x44	/* Watermark Level register. */
111 #define	USDHC_MIX_CONTROL	0x48	/* Mix(ed) Control register. */
112 #define	SDHC_VEND_SPEC		0xC0	/* Vendor-specific register. */
113 #define	 SDHC_VEND_FRC_SDCLK_ON	(1 <<  8)
114 #define	 SDHC_VEND_IPGEN	(1 << 11)
115 #define	 SDHC_VEND_HCKEN	(1 << 12)
116 #define	 SDHC_VEND_PEREN	(1 << 13)
117 
118 #define	SDHC_PRES_STATE		0x24
119 #define	  SDHC_PRES_CIHB	  (1 <<  0)
120 #define	  SDHC_PRES_CDIHB	  (1 <<  1)
121 #define	  SDHC_PRES_DLA		  (1 <<  2)
122 #define	  SDHC_PRES_SDSTB	  (1 <<  3)
123 #define	  SDHC_PRES_IPGOFF	  (1 <<  4)
124 #define	  SDHC_PRES_HCKOFF	  (1 <<  5)
125 #define	  SDHC_PRES_PEROFF	  (1 <<  6)
126 #define	  SDHC_PRES_SDOFF	  (1 <<  7)
127 #define	  SDHC_PRES_WTA		  (1 <<  8)
128 #define	  SDHC_PRES_RTA		  (1 <<  9)
129 #define	  SDHC_PRES_BWEN	  (1 << 10)
130 #define	  SDHC_PRES_BREN	  (1 << 11)
131 #define	  SDHC_PRES_RTR		  (1 << 12)
132 #define	  SDHC_PRES_CINST	  (1 << 16)
133 #define	  SDHC_PRES_CDPL	  (1 << 18)
134 #define	  SDHC_PRES_WPSPL	  (1 << 19)
135 #define	  SDHC_PRES_CLSL	  (1 << 23)
136 #define	  SDHC_PRES_DLSL_SHIFT	  24
137 #define	  SDHC_PRES_DLSL_MASK	  (0xffU << SDHC_PRES_DLSL_SHIFT)
138 
139 #define	SDHC_PROT_CTRL		0x28
140 #define	 SDHC_PROT_LED		(1 << 0)
141 #define	 SDHC_PROT_WIDTH_1BIT	(0 << 1)
142 #define	 SDHC_PROT_WIDTH_4BIT	(1 << 1)
143 #define	 SDHC_PROT_WIDTH_8BIT	(2 << 1)
144 #define	 SDHC_PROT_WIDTH_MASK	(3 << 1)
145 #define	 SDHC_PROT_D3CD		(1 << 3)
146 #define	 SDHC_PROT_EMODE_BIG	(0 << 4)
147 #define	 SDHC_PROT_EMODE_HALF	(1 << 4)
148 #define	 SDHC_PROT_EMODE_LITTLE	(2 << 4)
149 #define	 SDHC_PROT_EMODE_MASK	(3 << 4)
150 #define	 SDHC_PROT_SDMA		(0 << 8)
151 #define	 SDHC_PROT_ADMA1	(1 << 8)
152 #define	 SDHC_PROT_ADMA2	(2 << 8)
153 #define	 SDHC_PROT_ADMA264	(3 << 8)
154 #define	 SDHC_PROT_DMA_MASK	(3 << 8)
155 #define	 SDHC_PROT_CDTL		(1 << 6)
156 #define	 SDHC_PROT_CDSS		(1 << 7)
157 
158 #define	SDHC_SYS_CTRL		0x2c
159 
160 /*
161  * The clock enable bits exist in different registers for ESDHC vs USDHC, but
162  * they are the same bits in both cases.  The divisor values go into the
163  * standard sdhci clock register, but in different bit positions and meanings
164    than the sdhci spec values.
165  */
166 #define	SDHC_CLK_IPGEN		(1 << 0)
167 #define	SDHC_CLK_HCKEN		(1 << 1)
168 #define	SDHC_CLK_PEREN		(1 << 2)
169 #define	SDHC_CLK_SDCLKEN	(1 << 3)
170 #define	SDHC_CLK_ENABLE_MASK	0x0000000f
171 #define	SDHC_CLK_DIVISOR_MASK	0x000000f0
172 #define	SDHC_CLK_DIVISOR_SHIFT	4
173 #define	SDHC_CLK_PRESCALE_MASK	0x0000ff00
174 #define	SDHC_CLK_PRESCALE_SHIFT	8
175 
176 static struct ofw_compat_data compat_data[] = {
177 	{"fsl,imx6q-usdhc",	HWTYPE_USDHC},
178 	{"fsl,imx6sl-usdhc",	HWTYPE_USDHC},
179 	{"fsl,imx53-esdhc",	HWTYPE_ESDHC},
180 	{"fsl,imx51-esdhc",	HWTYPE_ESDHC},
181 	{"fsl,esdhc",		HWTYPE_ESDHC},
182 	{NULL,			HWTYPE_NONE},
183 };
184 
185 static uint16_t fsl_sdhc_get_clock(struct fsl_sdhci_softc *sc);
186 static void fsl_sdhc_set_clock(struct fsl_sdhci_softc *sc, uint16_t val);
187 static void fsl_sdhci_r1bfix_func(void *arg);
188 
189 static inline uint32_t
190 RD4(struct fsl_sdhci_softc *sc, bus_size_t off)
191 {
192 
193 	return (bus_read_4(sc->mem_res, off));
194 }
195 
196 static inline void
197 WR4(struct fsl_sdhci_softc *sc, bus_size_t off, uint32_t val)
198 {
199 
200 	bus_write_4(sc->mem_res, off, val);
201 }
202 
203 static uint8_t
204 fsl_sdhci_read_1(device_t dev, struct sdhci_slot *slot, bus_size_t off)
205 {
206 	struct fsl_sdhci_softc *sc = device_get_softc(dev);
207 	uint32_t val32, wrk32;
208 
209 	/*
210 	 * Most of the things in the standard host control register are in the
211 	 * hardware's wider protocol control register, but some of the bits are
212 	 * moved around.
213 	 */
214 	if (off == SDHCI_HOST_CONTROL) {
215 		wrk32 = RD4(sc, SDHC_PROT_CTRL);
216 		val32 = wrk32 & (SDHCI_CTRL_LED | SDHCI_CTRL_CARD_DET |
217 		    SDHCI_CTRL_FORCE_CARD);
218 		switch (wrk32 & SDHC_PROT_WIDTH_MASK) {
219 		case SDHC_PROT_WIDTH_1BIT:
220 			/* Value is already 0. */
221 			break;
222 		case SDHC_PROT_WIDTH_4BIT:
223 			val32 |= SDHCI_CTRL_4BITBUS;
224 			break;
225 		case SDHC_PROT_WIDTH_8BIT:
226 			val32 |= SDHCI_CTRL_8BITBUS;
227 			break;
228 		}
229 		switch (wrk32 & SDHC_PROT_DMA_MASK) {
230 		case SDHC_PROT_SDMA:
231 			/* Value is already 0. */
232 			break;
233 		case SDHC_PROT_ADMA1:
234 			/* This value is deprecated, should never appear. */
235 			break;
236 		case SDHC_PROT_ADMA2:
237 			val32 |= SDHCI_CTRL_ADMA2;
238 			break;
239 		case SDHC_PROT_ADMA264:
240 			val32 |= SDHCI_CTRL_ADMA264;
241 			break;
242 		}
243 		return val32;
244 	}
245 
246 	/*
247 	 * XXX can't find the bus power on/off knob.  For now we have to say the
248 	 * power is always on and always set to the same voltage.
249 	 */
250 	if (off == SDHCI_POWER_CONTROL) {
251 		return (SDHCI_POWER_ON | SDHCI_POWER_300);
252 	}
253 
254 	return ((RD4(sc, off & ~3) >> (off & 3) * 8) & 0xff);
255 }
256 
257 static uint16_t
258 fsl_sdhci_read_2(device_t dev, struct sdhci_slot *slot, bus_size_t off)
259 {
260 	struct fsl_sdhci_softc *sc = device_get_softc(dev);
261 	uint32_t val32;
262 
263 	if (sc->hwtype == HWTYPE_USDHC) {
264 		/*
265 		 * The USDHC hardware has nothing in the version register, but
266 		 * it's v3 compatible with all our translation code.
267 		 */
268 		if (off == SDHCI_HOST_VERSION) {
269 			return (SDHCI_SPEC_300 << SDHCI_SPEC_VER_SHIFT);
270 		}
271 		/*
272 		 * The USDHC hardware moved the transfer mode bits to the mixed
273 		 * control register, fetch them from there.
274 		 */
275 		if (off == SDHCI_TRANSFER_MODE)
276 			return (RD4(sc, USDHC_MIX_CONTROL) & 0x37);
277 
278 	} else if (sc->hwtype == HWTYPE_ESDHC) {
279 		/*
280 		 * The ESDHC hardware has the typical 32-bit combined "command
281 		 * and mode" register that we have to cache so that command
282 		 * isn't written until after mode.  On a read, just retrieve the
283 		 * cached values last written.
284 		 */
285 		if (off == SDHCI_TRANSFER_MODE) {
286 			return (sc->cmd_and_mode & 0x0000ffff);
287 		} else if (off == SDHCI_COMMAND_FLAGS) {
288 			return (sc->cmd_and_mode >> 16);
289 		}
290 	}
291 
292 	/*
293 	 * This hardware only manages one slot.  Synthesize a slot interrupt
294 	 * status register... if there are any enabled interrupts active they
295 	 * must be coming from our one and only slot.
296 	 */
297 	if (off == SDHCI_SLOT_INT_STATUS) {
298 		val32  = RD4(sc, SDHCI_INT_STATUS);
299 		val32 &= RD4(sc, SDHCI_SIGNAL_ENABLE);
300 		return (val32 ? 1 : 0);
301 	}
302 
303 	/*
304 	 * Clock bits are scattered into various registers which differ by
305 	 * hardware type, complex enough to have their own function.
306 	 */
307 	if (off == SDHCI_CLOCK_CONTROL) {
308 		return (fsl_sdhc_get_clock(sc));
309 	}
310 
311 	return ((RD4(sc, off & ~3) >> (off & 3) * 8) & 0xffff);
312 }
313 
314 static uint32_t
315 fsl_sdhci_read_4(device_t dev, struct sdhci_slot *slot, bus_size_t off)
316 {
317 	struct fsl_sdhci_softc *sc = device_get_softc(dev);
318 	uint32_t val32, wrk32;
319 
320 	val32 = RD4(sc, off);
321 
322 	/*
323 	 * The hardware leaves the base clock frequency out of the capabilities
324 	 * register, but we filled it in by setting slot->max_clk at attach time
325 	 * rather than here, because we can't represent frequencies above 63MHz
326 	 * in an sdhci 2.0 capabliities register.  The timeout clock is the same
327 	 * as the active output sdclock; we indicate that with a quirk setting
328 	 * so don't populate the timeout frequency bits.
329 	 *
330 	 * XXX Turn off (for now) features the hardware can do but this driver
331 	 * doesn't yet handle (1.8v, suspend/resume, etc).
332 	 */
333 	if (off == SDHCI_CAPABILITIES) {
334 		val32 &= ~SDHCI_CAN_VDD_180;
335 		val32 &= ~SDHCI_CAN_DO_SUSPEND;
336 		val32 |= SDHCI_CAN_DO_8BITBUS;
337 		return (val32);
338 	}
339 
340 	/*
341 	 * The hardware moves bits around in the present state register to make
342 	 * room for all 8 data line state bits.  To translate, mask out all the
343 	 * bits which are not in the same position in both registers (this also
344 	 * masks out some Freescale-specific bits in locations defined as
345 	 * reserved by sdhci), then shift the data line and retune request bits
346 	 * down to their standard locations.
347 	 */
348 	if (off == SDHCI_PRESENT_STATE) {
349 		wrk32 = val32;
350 		val32 &= 0x000F0F07;
351 		val32 |= (wrk32 >> 4) & SDHCI_STATE_DAT_MASK;
352 		val32 |= (wrk32 >> 9) & SDHCI_RETUNE_REQUEST;
353 		return (val32);
354 	}
355 
356 	/*
357 	 * fsl_sdhci_intr() can synthesize a DATA_END interrupt following a
358 	 * command with an R1B response, mix it into the hardware status.
359 	 */
360 	if (off == SDHCI_INT_STATUS) {
361 		return (val32 | sc->r1bfix_intmask);
362 	}
363 
364 	return val32;
365 }
366 
367 static void
368 fsl_sdhci_read_multi_4(device_t dev, struct sdhci_slot *slot, bus_size_t off,
369     uint32_t *data, bus_size_t count)
370 {
371 	struct fsl_sdhci_softc *sc = device_get_softc(dev);
372 
373 	bus_read_multi_4(sc->mem_res, off, data, count);
374 }
375 
376 static void
377 fsl_sdhci_write_1(device_t dev, struct sdhci_slot *slot, bus_size_t off, uint8_t val)
378 {
379 	struct fsl_sdhci_softc *sc = device_get_softc(dev);
380 	uint32_t val32;
381 
382 	/*
383 	 * Most of the things in the standard host control register are in the
384 	 * hardware's wider protocol control register, but some of the bits are
385 	 * moved around.
386 	 */
387 	if (off == SDHCI_HOST_CONTROL) {
388 		val32 = RD4(sc, SDHC_PROT_CTRL);
389 		val32 &= ~(SDHC_PROT_LED | SDHC_PROT_DMA_MASK |
390 		    SDHC_PROT_WIDTH_MASK | SDHC_PROT_CDTL | SDHC_PROT_CDSS);
391 		val32 |= (val & SDHCI_CTRL_LED);
392 		if (val & SDHCI_CTRL_8BITBUS)
393 			val32 |= SDHC_PROT_WIDTH_8BIT;
394 		else
395 			val32 |= (val & SDHCI_CTRL_4BITBUS);
396 		val32 |= (val & (SDHCI_CTRL_SDMA | SDHCI_CTRL_ADMA2)) << 4;
397 		val32 |= (val & (SDHCI_CTRL_CARD_DET | SDHCI_CTRL_FORCE_CARD));
398 		WR4(sc, SDHC_PROT_CTRL, val32);
399 		return;
400 	}
401 
402 	/* XXX I can't find the bus power on/off knob; do nothing. */
403 	if (off == SDHCI_POWER_CONTROL) {
404 		return;
405 	}
406 #ifdef __powerpc__
407 	/* XXX Reset doesn't seem to work as expected.  Do nothing for now. */
408 	if (off == SDHCI_SOFTWARE_RESET)
409 		return;
410 #endif
411 
412 	val32 = RD4(sc, off & ~3);
413 	val32 &= ~(0xff << (off & 3) * 8);
414 	val32 |= (val << (off & 3) * 8);
415 
416 	WR4(sc, off & ~3, val32);
417 }
418 
419 static void
420 fsl_sdhci_write_2(device_t dev, struct sdhci_slot *slot, bus_size_t off, uint16_t val)
421 {
422 	struct fsl_sdhci_softc *sc = device_get_softc(dev);
423 	uint32_t val32;
424 
425 	/*
426 	 * The clock control stuff is complex enough to have its own function
427 	 * that can handle the ESDHC versus USDHC differences.
428 	 */
429 	if (off == SDHCI_CLOCK_CONTROL) {
430 		fsl_sdhc_set_clock(sc, val);
431 		return;
432 	}
433 
434 	/*
435 	 * Figure out whether we need to check the DAT0 line for busy status at
436 	 * interrupt time.  The controller should be doing this, but for some
437 	 * reason it doesn't.  There are two cases:
438 	 *  - R1B response with no data transfer should generate a DATA_END (aka
439 	 *    TRANSFER_COMPLETE) interrupt after waiting for busy, but if
440 	 *    there's no data transfer there's no DATA_END interrupt.  This is
441 	 *    documented; they seem to think it's a feature.
442 	 *  - R1B response after Auto-CMD12 appears to not work, even though
443 	 *    there's a control bit for it (bit 3) in the vendor register.
444 	 * When we're starting a command that needs a manual DAT0 line check at
445 	 * interrupt time, we leave ourselves a note in r1bfix_type so that we
446 	 * can do the extra work in fsl_sdhci_intr().
447 	 */
448 	if (off == SDHCI_COMMAND_FLAGS) {
449 		if (val & SDHCI_CMD_DATA) {
450 			const uint32_t MBAUTOCMD = SDHCI_TRNS_ACMD12 | SDHCI_TRNS_MULTI;
451 			val32 = RD4(sc, USDHC_MIX_CONTROL);
452 			if ((val32 & MBAUTOCMD) == MBAUTOCMD)
453 				sc->r1bfix_type = R1BFIX_AC12;
454 		} else {
455 			if ((val & SDHCI_CMD_RESP_MASK) == SDHCI_CMD_RESP_SHORT_BUSY) {
456 				WR4(sc, SDHCI_INT_ENABLE, slot->intmask | SDHCI_INT_RESPONSE);
457 				WR4(sc, SDHCI_SIGNAL_ENABLE, slot->intmask | SDHCI_INT_RESPONSE);
458 				sc->r1bfix_type = R1BFIX_NODATA;
459 			}
460 		}
461 	}
462 
463 	/*
464 	 * The USDHC hardware moved the transfer mode bits to mixed control; we
465 	 * just write them there and we're done.  The ESDHC hardware has the
466 	 * typical combined cmd-and-mode register that allows only 32-bit
467 	 * access, so when writing the mode bits just save them, then later when
468 	 * writing the command bits, add in the saved mode bits.
469 	 */
470 	if (sc->hwtype == HWTYPE_USDHC) {
471 		if (off == SDHCI_TRANSFER_MODE) {
472 			val32 = RD4(sc, USDHC_MIX_CONTROL);
473 			val32 &= ~0x3f;
474 			val32 |= val & 0x37;
475 			// XXX acmd23 not supported here (or by sdhci driver)
476 			WR4(sc, USDHC_MIX_CONTROL, val32);
477 			return;
478 		}
479 	} else if (sc->hwtype == HWTYPE_ESDHC) {
480 		if (off == SDHCI_TRANSFER_MODE) {
481 			sc->cmd_and_mode =
482 			    (sc->cmd_and_mode & 0xffff0000) | val;
483 			return;
484 		} else if (off == SDHCI_COMMAND_FLAGS) {
485 			sc->cmd_and_mode =
486 			    (sc->cmd_and_mode & 0xffff) | (val << 16);
487 			WR4(sc, SDHCI_TRANSFER_MODE, sc->cmd_and_mode);
488 			return;
489 		}
490 	}
491 
492 	val32 = RD4(sc, off & ~3);
493 	val32 &= ~(0xffff << (off & 3) * 8);
494 	val32 |= ((val & 0xffff) << (off & 3) * 8);
495 	WR4(sc, off & ~3, val32);
496 }
497 
498 static void
499 fsl_sdhci_write_4(device_t dev, struct sdhci_slot *slot, bus_size_t off, uint32_t val)
500 {
501 	struct fsl_sdhci_softc *sc = device_get_softc(dev);
502 
503 	/* Clear synthesized interrupts, then pass the value to the hardware. */
504 	if (off == SDHCI_INT_STATUS) {
505 		sc->r1bfix_intmask &= ~val;
506 	}
507 
508 	WR4(sc, off, val);
509 }
510 
511 static void
512 fsl_sdhci_write_multi_4(device_t dev, struct sdhci_slot *slot, bus_size_t off,
513     uint32_t *data, bus_size_t count)
514 {
515 	struct fsl_sdhci_softc *sc = device_get_softc(dev);
516 
517 	bus_write_multi_4(sc->mem_res, off, data, count);
518 }
519 
520 static uint16_t
521 fsl_sdhc_get_clock(struct fsl_sdhci_softc *sc)
522 {
523 	uint16_t val;
524 
525 	/*
526 	 * Whenever the sdhci driver writes the clock register we save a
527 	 * snapshot of just the frequency bits, so that we can play them back
528 	 * here on a register read without recalculating the frequency from the
529 	 * prescalar and divisor bits in the real register.  We'll start with
530 	 * those bits, and mix in the clock status and enable bits that come
531 	 * from different places depending on which hardware we've got.
532 	 */
533 	val = sc->sdclockreg_freq_bits;
534 
535 	/*
536 	 * The internal clock is always enabled (actually, the hardware manages
537 	 * it).  Whether the internal clock is stable yet after a frequency
538 	 * change comes from the present-state register on both hardware types.
539 	 */
540 	val |= SDHCI_CLOCK_INT_EN;
541 	if (RD4(sc, SDHC_PRES_STATE) & SDHC_PRES_SDSTB)
542 	    val |= SDHCI_CLOCK_INT_STABLE;
543 
544 	/*
545 	 * On i.MX ESDHC hardware the card bus clock enable is in the usual
546 	 * sdhci register but it's a different bit, so transcribe it (note the
547 	 * difference between standard SDHCI_ and Freescale SDHC_ prefixes
548 	 * here). On USDHC and QorIQ ESDHC hardware there is a force-on bit, but
549 	 * no force-off for the card bus clock (the hardware runs the clock when
550 	 * transfers are active no matter what), so we always say the clock is
551 	 * on.
552 	 * XXX Maybe we should say it's in whatever state the sdhci driver last
553 	 * set it to.
554 	 */
555 	if (sc->hwtype == HWTYPE_ESDHC) {
556 #ifdef __arm__
557 		if (RD4(sc, SDHC_SYS_CTRL) & SDHC_CLK_SDCLKEN)
558 #endif
559 			val |= SDHCI_CLOCK_CARD_EN;
560 	} else {
561 		val |= SDHCI_CLOCK_CARD_EN;
562 	}
563 
564 	return (val);
565 }
566 
567 static void
568 fsl_sdhc_set_clock(struct fsl_sdhci_softc *sc, uint16_t val)
569 {
570 	uint32_t divisor, freq, prescale, val32;
571 
572 	val32 = RD4(sc, SDHCI_CLOCK_CONTROL);
573 
574 	/*
575 	 * Save the frequency-setting bits in SDHCI format so that we can play
576 	 * them back in get_clock without complex decoding of hardware regs,
577 	 * then deal with the freqency part of the value based on hardware type.
578 	 */
579 	sc->sdclockreg_freq_bits = val & SDHCI_DIVIDERS_MASK;
580 	if (sc->hwtype == HWTYPE_ESDHC) {
581 		/*
582 		 * The i.MX5 ESDHC hardware requires the driver to manually
583 		 * start and stop the sd bus clock.  If the enable bit is not
584 		 * set, turn off the clock in hardware and we're done, otherwise
585 		 * decode the requested frequency.  ESDHC hardware is sdhci 2.0;
586 		 * the sdhci driver will use the original 8-bit divisor field
587 		 * and the "base / 2^N" divisor scheme.
588 		 */
589 		if ((val & SDHCI_CLOCK_CARD_EN) == 0) {
590 #ifdef __arm__
591 			/* On QorIQ, this is a reserved bit. */
592 			WR4(sc, SDHCI_CLOCK_CONTROL, val32 & ~SDHC_CLK_SDCLKEN);
593 #endif
594 			return;
595 		}
596 		divisor = (val >> SDHCI_DIVIDER_SHIFT) & SDHCI_DIVIDER_MASK;
597 		freq = sc->baseclk_hz >> ffs(divisor);
598 	} else {
599 		/*
600 		 * The USDHC hardware provides only "force always on" control
601 		 * over the sd bus clock, but no way to turn it off.  (If a cmd
602 		 * or data transfer is in progress the clock is on, otherwise it
603 		 * is off.)  If the clock is being disabled, we can just return
604 		 * now, otherwise we decode the requested frequency.  USDHC
605 		 * hardware is sdhci 3.0; the sdhci driver will use a 10-bit
606 		 * divisor using the "base / 2*N" divisor scheme.
607 		 */
608 		if ((val & SDHCI_CLOCK_CARD_EN) == 0)
609 			return;
610 		divisor = ((val >> SDHCI_DIVIDER_SHIFT) & SDHCI_DIVIDER_MASK) |
611 		    ((val >> SDHCI_DIVIDER_HI_SHIFT) & SDHCI_DIVIDER_HI_MASK) <<
612 		    SDHCI_DIVIDER_MASK_LEN;
613 		if (divisor == 0)
614 			freq = sc->baseclk_hz;
615 		else
616 			freq = sc->baseclk_hz / (2 * divisor);
617 	}
618 
619 	/*
620 	 * Get a prescaler and final divisor to achieve the desired frequency.
621 	 */
622 	for (prescale = 2; freq < sc->baseclk_hz / (prescale * 16);)
623 		prescale <<= 1;
624 
625 	for (divisor = 1; freq < sc->baseclk_hz / (prescale * divisor);)
626 		++divisor;
627 
628 #ifdef DEBUG
629 	device_printf(sc->dev,
630 	    "desired SD freq: %d, actual: %d; base %d prescale %d divisor %d\n",
631 	    freq, sc->baseclk_hz / (prescale * divisor), sc->baseclk_hz,
632 	    prescale, divisor);
633 #endif
634 
635 	/*
636 	 * Adjust to zero-based values, and store them to the hardware.
637 	 */
638 	prescale >>= 1;
639 	divisor -= 1;
640 
641 	val32 &= ~(SDHC_CLK_DIVISOR_MASK | SDHC_CLK_PRESCALE_MASK);
642 	val32 |= divisor << SDHC_CLK_DIVISOR_SHIFT;
643 	val32 |= prescale << SDHC_CLK_PRESCALE_SHIFT;
644 	val32 |= SDHC_CLK_IPGEN;
645 	WR4(sc, SDHCI_CLOCK_CONTROL, val32);
646 }
647 
648 static boolean_t
649 fsl_sdhci_r1bfix_is_wait_done(struct fsl_sdhci_softc *sc)
650 {
651 	uint32_t inhibit;
652 
653 	mtx_assert(&sc->slot.mtx, MA_OWNED);
654 
655 	/*
656 	 * Check the DAT0 line status using both the DLA (data line active) and
657 	 * CDIHB (data inhibit) bits in the present state register.  In theory
658 	 * just DLA should do the trick,  but in practice it takes both.  If the
659 	 * DAT0 line is still being held and we're not yet beyond the timeout
660 	 * point, just schedule another callout to check again later.
661 	 */
662 	inhibit = RD4(sc, SDHC_PRES_STATE) & (SDHC_PRES_DLA | SDHC_PRES_CDIHB);
663 
664 	if (inhibit && getsbinuptime() < sc->r1bfix_timeout_at) {
665 		callout_reset_sbt(&sc->r1bfix_callout, SBT_1MS, 0,
666 		    fsl_sdhci_r1bfix_func, sc, 0);
667 		return (false);
668 	}
669 
670 	/*
671 	 * If we reach this point with the inhibit bits still set, we've got a
672 	 * timeout, synthesize a DATA_TIMEOUT interrupt.  Otherwise the DAT0
673 	 * line has been released, and we synthesize a DATA_END, and if the type
674 	 * of fix needed was on a command-without-data we also now add in the
675 	 * original INT_RESPONSE that we suppressed earlier.
676 	 */
677 	if (inhibit)
678 		sc->r1bfix_intmask |= SDHCI_INT_DATA_TIMEOUT;
679 	else {
680 		sc->r1bfix_intmask |= SDHCI_INT_DATA_END;
681 		if (sc->r1bfix_type == R1BFIX_NODATA)
682 			sc->r1bfix_intmask |= SDHCI_INT_RESPONSE;
683 	}
684 
685 	sc->r1bfix_type = R1BFIX_NONE;
686 	return (true);
687 }
688 
689 static void
690 fsl_sdhci_r1bfix_func(void * arg)
691 {
692 	struct fsl_sdhci_softc *sc = arg;
693 	boolean_t r1bwait_done;
694 
695 	mtx_lock(&sc->slot.mtx);
696 	r1bwait_done = fsl_sdhci_r1bfix_is_wait_done(sc);
697 	mtx_unlock(&sc->slot.mtx);
698 	if (r1bwait_done)
699 		sdhci_generic_intr(&sc->slot);
700 }
701 
702 static void
703 fsl_sdhci_intr(void *arg)
704 {
705 	struct fsl_sdhci_softc *sc = arg;
706 	uint32_t intmask;
707 
708 	mtx_lock(&sc->slot.mtx);
709 
710 	/*
711 	 * Manually check the DAT0 line for R1B response types that the
712 	 * controller fails to handle properly.  The controller asserts the done
713 	 * interrupt while the card is still asserting busy with the DAT0 line.
714 	 *
715 	 * We check DAT0 immediately because most of the time, especially on a
716 	 * read, the card will actually be done by time we get here.  If it's
717 	 * not, then the wait_done routine will schedule a callout to re-check
718 	 * periodically until it is done.  In that case we clear the interrupt
719 	 * out of the hardware now so that we can present it later when the DAT0
720 	 * line is released.
721 	 *
722 	 * If we need to wait for the DAT0 line to be released, we set up a
723 	 * timeout point 250ms in the future.  This number comes from the SD
724 	 * spec, which allows a command to take that long.  In the real world,
725 	 * cards tend to take 10-20ms for a long-running command such as a write
726 	 * or erase that spans two pages.
727 	 */
728 	switch (sc->r1bfix_type) {
729 	case R1BFIX_NODATA:
730 		intmask = RD4(sc, SDHCI_INT_STATUS) & SDHCI_INT_RESPONSE;
731 		break;
732 	case R1BFIX_AC12:
733 		intmask = RD4(sc, SDHCI_INT_STATUS) & SDHCI_INT_DATA_END;
734 		break;
735 	default:
736 		intmask = 0;
737 		break;
738 	}
739 	if (intmask) {
740 		sc->r1bfix_timeout_at = getsbinuptime() + 250 * SBT_1MS;
741 		if (!fsl_sdhci_r1bfix_is_wait_done(sc)) {
742 			WR4(sc, SDHCI_INT_STATUS, intmask);
743 			bus_barrier(sc->mem_res, SDHCI_INT_STATUS, 4,
744 			    BUS_SPACE_BARRIER_WRITE);
745 		}
746 	}
747 
748 	mtx_unlock(&sc->slot.mtx);
749 	sdhci_generic_intr(&sc->slot);
750 }
751 
752 static int
753 fsl_sdhci_get_ro(device_t bus, device_t child)
754 {
755 	struct fsl_sdhci_softc *sc = device_get_softc(bus);
756 
757 	return (sdhci_fdt_gpio_get_readonly(sc->gpio));
758 }
759 
760 static bool
761 fsl_sdhci_get_card_present(device_t dev, struct sdhci_slot *slot)
762 {
763 	struct fsl_sdhci_softc *sc = device_get_softc(dev);
764 
765 	return (sdhci_fdt_gpio_get_present(sc->gpio));
766 }
767 
768 #ifdef __powerpc__
769 static uint32_t
770 fsl_sdhci_get_platform_clock(device_t dev)
771 {
772 	phandle_t node;
773 	uint32_t clock;
774 
775 	node = ofw_bus_get_node(dev);
776 
777 	/* Get sdhci node properties */
778 	if((OF_getprop(node, "clock-frequency", (void *)&clock,
779 	    sizeof(clock)) <= 0) || (clock == 0)) {
780 		clock = mpc85xx_get_system_clock();
781 
782 		if (clock == 0) {
783 			device_printf(dev,"Cannot acquire correct sdhci "
784 			    "frequency from DTS.\n");
785 
786 			return (0);
787 		}
788 	}
789 
790 	if (bootverbose)
791 		device_printf(dev, "Acquired clock: %d from DTS\n", clock);
792 
793 	return (clock);
794 }
795 #endif
796 
797 static int
798 fsl_sdhci_detach(device_t dev)
799 {
800 	struct fsl_sdhci_softc *sc = device_get_softc(dev);
801 
802 	if (sc->gpio != NULL)
803 		sdhci_fdt_gpio_teardown(sc->gpio);
804 
805 	callout_drain(&sc->r1bfix_callout);
806 
807 	if (sc->slot_init_done)
808 		sdhci_cleanup_slot(&sc->slot);
809 
810 	if (sc->intr_cookie != NULL)
811 		bus_teardown_intr(dev, sc->irq_res, sc->intr_cookie);
812 	if (sc->irq_res != NULL)
813 		bus_release_resource(dev, SYS_RES_IRQ,
814 		    rman_get_rid(sc->irq_res), sc->irq_res);
815 
816 	if (sc->mem_res != NULL) {
817 		bus_release_resource(dev, SYS_RES_MEMORY,
818 		    rman_get_rid(sc->mem_res), sc->mem_res);
819 	}
820 
821 	return (0);
822 }
823 
824 static int
825 fsl_sdhci_attach(device_t dev)
826 {
827 	struct fsl_sdhci_softc *sc = device_get_softc(dev);
828 	int rid, err;
829 #ifdef __powerpc__
830 	phandle_t node;
831 	uint32_t protctl;
832 #endif
833 
834 	sc->dev = dev;
835 
836 	callout_init(&sc->r1bfix_callout, 1);
837 
838 	sc->hwtype = ofw_bus_search_compatible(dev, compat_data)->ocd_data;
839 	if (sc->hwtype == HWTYPE_NONE)
840 		panic("Impossible: not compatible in fsl_sdhci_attach()");
841 
842 	rid = 0;
843 	sc->mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
844 	    RF_ACTIVE);
845 	if (!sc->mem_res) {
846 		device_printf(dev, "cannot allocate memory window\n");
847 		err = ENXIO;
848 		goto fail;
849 	}
850 
851 	rid = 0;
852 	sc->irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
853 	    RF_ACTIVE);
854 	if (!sc->irq_res) {
855 		device_printf(dev, "cannot allocate interrupt\n");
856 		err = ENXIO;
857 		goto fail;
858 	}
859 
860 	if (bus_setup_intr(dev, sc->irq_res, INTR_TYPE_BIO | INTR_MPSAFE,
861 	    NULL, fsl_sdhci_intr, sc, &sc->intr_cookie)) {
862 		device_printf(dev, "cannot setup interrupt handler\n");
863 		err = ENXIO;
864 		goto fail;
865 	}
866 
867 	sc->slot.quirks |= SDHCI_QUIRK_DATA_TIMEOUT_USES_SDCLK;
868 
869 	/*
870 	 * DMA is not really broken, I just haven't implemented it yet.
871 	 */
872 	sc->slot.quirks |= SDHCI_QUIRK_BROKEN_DMA;
873 
874 	/*
875 	 * Set the buffer watermark level to 128 words (512 bytes) for both read
876 	 * and write.  The hardware has a restriction that when the read or
877 	 * write ready status is asserted, that means you can read exactly the
878 	 * number of words set in the watermark register before you have to
879 	 * re-check the status and potentially wait for more data.  The main
880 	 * sdhci driver provides no hook for doing status checking on less than
881 	 * a full block boundary, so we set the watermark level to be a full
882 	 * block.  Reads and writes where the block size is less than the
883 	 * watermark size will work correctly too, no need to change the
884 	 * watermark for different size blocks.  However, 128 is the maximum
885 	 * allowed for the watermark, so PIO is limitted to 512 byte blocks
886 	 * (which works fine for SD cards, may be a problem for SDIO some day).
887 	 *
888 	 * XXX need named constants for this stuff.
889 	 */
890 	/* P1022 has the '*_BRST_LEN' fields as reserved, always reading 0x10 */
891 	if (ofw_bus_is_compatible(dev, "fsl,p1022-esdhc"))
892 		WR4(sc, SDHC_WTMK_LVL, 0x10801080);
893 	else
894 		WR4(sc, SDHC_WTMK_LVL, 0x08800880);
895 
896 	/*
897 	 * We read in native byte order in the main driver, but the register
898 	 * defaults to little endian.
899 	 */
900 #ifdef __powerpc__
901 	sc->baseclk_hz = fsl_sdhci_get_platform_clock(dev);
902 #else
903 	sc->baseclk_hz = imx_ccm_sdhci_hz();
904 #endif
905 	sc->slot.max_clk = sc->baseclk_hz;
906 
907 	/*
908 	 * Set up any gpio pin handling described in the FDT data. This cannot
909 	 * fail; see comments in sdhci_fdt_gpio.h for details.
910 	 */
911 	sc->gpio = sdhci_fdt_gpio_setup(dev, &sc->slot);
912 
913 #ifdef __powerpc__
914 	node = ofw_bus_get_node(dev);
915 	/* Default to big-endian on powerpc */
916 	protctl = RD4(sc, SDHC_PROT_CTRL);
917 	protctl &= ~SDHC_PROT_EMODE_MASK;
918 	if (OF_hasprop(node, "little-endian"))
919 		protctl |= SDHC_PROT_EMODE_LITTLE;
920 	else
921 		protctl |= SDHC_PROT_EMODE_BIG;
922 	WR4(sc, SDHC_PROT_CTRL, protctl);
923 #endif
924 
925 	sdhci_init_slot(dev, &sc->slot, 0);
926 	sc->slot_init_done = true;
927 
928 	bus_generic_probe(dev);
929 	bus_generic_attach(dev);
930 
931 	sdhci_start_slot(&sc->slot);
932 
933 	return (0);
934 
935 fail:
936 	fsl_sdhci_detach(dev);
937 	return (err);
938 }
939 
940 static int
941 fsl_sdhci_probe(device_t dev)
942 {
943 
944 	if (!ofw_bus_status_okay(dev))
945 		return (ENXIO);
946 
947 	switch (ofw_bus_search_compatible(dev, compat_data)->ocd_data) {
948 	case HWTYPE_ESDHC:
949 		device_set_desc(dev, "Freescale eSDHC controller");
950 		return (BUS_PROBE_DEFAULT);
951 	case HWTYPE_USDHC:
952 		device_set_desc(dev, "Freescale uSDHC controller");
953 		return (BUS_PROBE_DEFAULT);
954 	default:
955 		break;
956 	}
957 	return (ENXIO);
958 }
959 
960 static device_method_t fsl_sdhci_methods[] = {
961 	/* Device interface */
962 	DEVMETHOD(device_probe,		fsl_sdhci_probe),
963 	DEVMETHOD(device_attach,	fsl_sdhci_attach),
964 	DEVMETHOD(device_detach,	fsl_sdhci_detach),
965 
966 	/* Bus interface */
967 	DEVMETHOD(bus_read_ivar,	sdhci_generic_read_ivar),
968 	DEVMETHOD(bus_write_ivar,	sdhci_generic_write_ivar),
969 
970 	/* MMC bridge interface */
971 	DEVMETHOD(mmcbr_update_ios,	sdhci_generic_update_ios),
972 	DEVMETHOD(mmcbr_request,	sdhci_generic_request),
973 	DEVMETHOD(mmcbr_get_ro,		fsl_sdhci_get_ro),
974 	DEVMETHOD(mmcbr_acquire_host,	sdhci_generic_acquire_host),
975 	DEVMETHOD(mmcbr_release_host,	sdhci_generic_release_host),
976 
977 	/* SDHCI accessors */
978 	DEVMETHOD(sdhci_read_1,		fsl_sdhci_read_1),
979 	DEVMETHOD(sdhci_read_2,		fsl_sdhci_read_2),
980 	DEVMETHOD(sdhci_read_4,		fsl_sdhci_read_4),
981 	DEVMETHOD(sdhci_read_multi_4,	fsl_sdhci_read_multi_4),
982 	DEVMETHOD(sdhci_write_1,	fsl_sdhci_write_1),
983 	DEVMETHOD(sdhci_write_2,	fsl_sdhci_write_2),
984 	DEVMETHOD(sdhci_write_4,	fsl_sdhci_write_4),
985 	DEVMETHOD(sdhci_write_multi_4,	fsl_sdhci_write_multi_4),
986 	DEVMETHOD(sdhci_get_card_present,fsl_sdhci_get_card_present),
987 
988 	DEVMETHOD_END
989 };
990 
991 static driver_t fsl_sdhci_driver = {
992 	"sdhci_fsl",
993 	fsl_sdhci_methods,
994 	sizeof(struct fsl_sdhci_softc),
995 };
996 
997 DRIVER_MODULE(sdhci_fsl, simplebus, fsl_sdhci_driver, NULL, NULL);
998 SDHCI_DEPEND(sdhci_fsl);
999 
1000 #ifndef MMCCAM
1001 MMC_DECLARE_BRIDGE(sdhci_fsl);
1002 #endif
1003