xref: /illumos-gate/usr/src/uts/common/io/sdcard/adapters/sdhost/sdhost.c (revision 20a7641f9918de8574b8b3b47dbe35c4bfc78df1)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 #include "sdhost.h"
26 
27 typedef	struct sdstats	sdstats_t;
28 typedef	struct sdslot	sdslot_t;
29 typedef	struct sdhost	sdhost_t;
30 
31 struct sdstats {
32 	kstat_named_t	ks_ncmd;
33 	kstat_named_t	ks_ixfr;
34 	kstat_named_t	ks_oxfr;
35 	kstat_named_t	ks_ibytes;
36 	kstat_named_t	ks_obytes;
37 	kstat_named_t	ks_npio;
38 	kstat_named_t	ks_ndma;
39 	kstat_named_t	ks_nmulti;
40 	kstat_named_t	ks_baseclk;
41 	kstat_named_t	ks_cardclk;
42 	kstat_named_t	ks_tmusecs;
43 	kstat_named_t	ks_width;
44 	kstat_named_t	ks_flags;
45 	kstat_named_t	ks_capab;
46 };
47 
48 #define	SDFLAG_FORCE_PIO		(1U << 0)
49 #define	SDFLAG_FORCE_DMA		(1U << 1)
50 
51 /*
52  * Per slot state.
53  */
54 struct sdslot {
55 	sda_host_t		*ss_host;
56 	int			ss_num;
57 	ddi_acc_handle_t	ss_acch;
58 	caddr_t 		ss_regva;
59 	kmutex_t		ss_lock;
60 	uint8_t			ss_tmoutclk;
61 	uint32_t		ss_ocr;		/* OCR formatted voltages */
62 	uint16_t		ss_mode;
63 	boolean_t		ss_suspended;
64 	sdstats_t		ss_stats;
65 #define	ss_ncmd			ss_stats.ks_ncmd.value.ui64
66 #define	ss_ixfr			ss_stats.ks_ixfr.value.ui64
67 #define	ss_oxfr			ss_stats.ks_oxfr.value.ui64
68 #define	ss_ibytes		ss_stats.ks_ibytes.value.ui64
69 #define	ss_obytes		ss_stats.ks_obytes.value.ui64
70 #define	ss_ndma			ss_stats.ks_ndma.value.ui64
71 #define	ss_npio			ss_stats.ks_npio.value.ui64
72 #define	ss_nmulti		ss_stats.ks_nmulti.value.ui64
73 
74 #define	ss_baseclk		ss_stats.ks_baseclk.value.ui32
75 #define	ss_cardclk		ss_stats.ks_cardclk.value.ui32
76 #define	ss_tmusecs		ss_stats.ks_tmusecs.value.ui32
77 #define	ss_width		ss_stats.ks_width.value.ui32
78 #define	ss_flags		ss_stats.ks_flags.value.ui32
79 #define	ss_capab		ss_stats.ks_capab.value.ui32
80 	kstat_t			*ss_ksp;
81 
82 	/*
83 	 * Command in progress
84 	 */
85 	uint8_t			*ss_kvaddr;
86 	int			ss_blksz;
87 	uint16_t		ss_resid;	/* in blocks */
88 	int			ss_rcnt;
89 
90 	/* scratch buffer, to receive extra PIO data */
91 	caddr_t			ss_bounce;
92 	ddi_dma_handle_t	ss_bufdmah;
93 	ddi_acc_handle_t	ss_bufacch;
94 	ddi_dma_cookie_t	ss_bufdmac;
95 };
96 
97 /*
98  * This allocates a rather large chunk of contiguous memory for DMA.
99  * But doing so means that we'll almost never have to resort to PIO.
100  */
101 #define	SDHOST_BOUNCESZ		65536
102 
103 /*
104  * Per controller state.
105  */
106 struct sdhost {
107 	int			sh_numslots;
108 	ddi_dma_attr_t		sh_dmaattr;
109 	sdslot_t		sh_slots[SDHOST_MAXSLOTS];
110 	sda_host_t		*sh_host;
111 
112 	/*
113 	 * Interrupt related information.
114 	 */
115 	ddi_intr_handle_t	sh_ihandle;
116 	int			sh_icap;
117 	uint_t			sh_ipri;
118 };
119 
120 #define	PROPSET(x)							\
121 	(ddi_prop_get_int(DDI_DEV_T_ANY, dip,				\
122 	DDI_PROP_DONTPASS | DDI_PROP_NOTPROM, x, 0) != 0)
123 
124 
125 static int sdhost_attach(dev_info_t *, ddi_attach_cmd_t);
126 static int sdhost_detach(dev_info_t *, ddi_detach_cmd_t);
127 static int sdhost_quiesce(dev_info_t *);
128 static int sdhost_suspend(dev_info_t *);
129 static int sdhost_resume(dev_info_t *);
130 
131 static void sdhost_enable_interrupts(sdslot_t *);
132 static void sdhost_disable_interrupts(sdslot_t *);
133 static int sdhost_setup_intr(dev_info_t *, sdhost_t *);
134 static uint_t sdhost_intr(caddr_t, caddr_t);
135 static int sdhost_init_slot(dev_info_t *, sdhost_t *, int, int);
136 static void sdhost_uninit_slot(sdhost_t *, int);
137 static sda_err_t sdhost_soft_reset(sdslot_t *, uint8_t);
138 static sda_err_t sdhost_set_clock(sdslot_t *, uint32_t);
139 static void sdhost_xfer_done(sdslot_t *, sda_err_t);
140 static sda_err_t sdhost_wait_cmd(sdslot_t *, sda_cmd_t *);
141 static uint_t sdhost_slot_intr(sdslot_t *);
142 
143 static sda_err_t sdhost_cmd(void *, sda_cmd_t *);
144 static sda_err_t sdhost_getprop(void *, sda_prop_t, uint32_t *);
145 static sda_err_t sdhost_setprop(void *, sda_prop_t, uint32_t);
146 static sda_err_t sdhost_poll(void *);
147 static sda_err_t sdhost_reset(void *);
148 static sda_err_t sdhost_halt(void *);
149 
150 static struct dev_ops sdhost_dev_ops = {
151 	DEVO_REV,			/* devo_rev */
152 	0,				/* devo_refcnt */
153 	ddi_no_info,			/* devo_getinfo */
154 	nulldev,			/* devo_identify */
155 	nulldev,			/* devo_probe */
156 	sdhost_attach,			/* devo_attach */
157 	sdhost_detach,			/* devo_detach */
158 	nodev,				/* devo_reset */
159 	NULL,				/* devo_cb_ops */
160 	NULL,				/* devo_bus_ops */
161 	NULL,				/* devo_power */
162 	sdhost_quiesce,			/* devo_quiesce */
163 };
164 
165 static struct modldrv sdhost_modldrv = {
166 	&mod_driverops,			/* drv_modops */
167 	"Standard SD Host Controller",	/* drv_linkinfo */
168 	&sdhost_dev_ops			/* drv_dev_ops */
169 };
170 
171 static struct modlinkage modlinkage = {
172 	MODREV_1,			/* ml_rev */
173 	{ &sdhost_modldrv, NULL }	/* ml_linkage */
174 };
175 
176 static struct sda_ops sdhost_ops = {
177 	SDA_OPS_VERSION,
178 	sdhost_cmd,			/* so_cmd */
179 	sdhost_getprop,			/* so_getprop */
180 	sdhost_setprop,			/* so_setprop */
181 	sdhost_poll,			/* so_poll */
182 	sdhost_reset,			/* so_reset */
183 	sdhost_halt,			/* so_halt */
184 };
185 
186 static ddi_device_acc_attr_t sdhost_regattr = {
187 	DDI_DEVICE_ATTR_V0,	/* devacc_attr_version */
188 	DDI_STRUCTURE_LE_ACC,	/* devacc_attr_endian_flags */
189 	DDI_STRICTORDER_ACC,	/* devacc_attr_dataorder */
190 	DDI_DEFAULT_ACC,	/* devacc_attr_access */
191 };
192 static ddi_device_acc_attr_t sdhost_bufattr = {
193 	DDI_DEVICE_ATTR_V0,	/* devacc_attr_version */
194 	DDI_NEVERSWAP_ACC,	/* devacc_attr_endian_flags */
195 	DDI_STRICTORDER_ACC,	/* devacc_attr_dataorder */
196 	DDI_DEFAULT_ACC,	/* devacc_attr_access */
197 };
198 
199 #define	GET16(ss, reg)	\
200 	ddi_get16(ss->ss_acch, (void *)(ss->ss_regva + reg))
201 #define	PUT16(ss, reg, val)	\
202 	ddi_put16(ss->ss_acch, (void *)(ss->ss_regva + reg), val)
203 #define	GET32(ss, reg)	\
204 	ddi_get32(ss->ss_acch, (void *)(ss->ss_regva + reg))
205 #define	PUT32(ss, reg, val)	\
206 	ddi_put32(ss->ss_acch, (void *)(ss->ss_regva + reg), val)
207 #define	GET64(ss, reg)	\
208 	ddi_get64(ss->ss_acch, (void *)(ss->ss_regva + reg))
209 
210 #define	GET8(ss, reg)	\
211 	ddi_get8(ss->ss_acch, (void *)(ss->ss_regva + reg))
212 #define	PUT8(ss, reg, val)	\
213 	ddi_put8(ss->ss_acch, (void *)(ss->ss_regva + reg), val)
214 
215 #define	CLR8(ss, reg, mask)	PUT8(ss, reg, GET8(ss, reg) & ~(mask))
216 #define	SET8(ss, reg, mask)	PUT8(ss, reg, GET8(ss, reg) | (mask))
217 
218 /*
219  * If ever anyone uses PIO on SPARC, we have to endian-swap.  But we
220  * think that SD Host Controllers are likely to be uncommon on SPARC,
221  * and hopefully when they exist at all they will be able to use DMA.
222  */
223 #ifdef	_BIG_ENDIAN
224 #define	sw32(x)		ddi_swap32(x)
225 #define	sw16(x)		ddi_swap16(x)
226 #else
227 #define	sw32(x)		(x)
228 #define	sw16(x)		(x)
229 #endif
230 
231 #define	GETDATA32(ss)		sw32(GET32(ss, REG_DATA))
232 #define	GETDATA16(ss)		sw16(GET16(ss, REG_DATA))
233 #define	GETDATA8(ss)		GET8(ss, REG_DATA)
234 
235 #define	PUTDATA32(ss, val)	PUT32(ss, REG_DATA, sw32(val))
236 #define	PUTDATA16(ss, val)	PUT16(ss, REG_DATA, sw16(val))
237 #define	PUTDATA8(ss, val)	PUT8(ss, REG_DATA, val)
238 
239 #define	CHECK_STATE(ss, nm)	\
240 	((GET32(ss, REG_PRS) & PRS_ ## nm) != 0)
241 
242 int
243 _init(void)
244 {
245 	int	rv;
246 
247 	sda_host_init_ops(&sdhost_dev_ops);
248 
249 	if ((rv = mod_install(&modlinkage)) != 0) {
250 		sda_host_fini_ops(&sdhost_dev_ops);
251 	}
252 
253 	return (rv);
254 }
255 
256 int
257 _fini(void)
258 {
259 	int	rv;
260 
261 	if ((rv = mod_remove(&modlinkage)) == 0) {
262 		sda_host_fini_ops(&sdhost_dev_ops);
263 	}
264 	return (rv);
265 }
266 
267 int
268 _info(struct modinfo *modinfop)
269 {
270 	return (mod_info(&modlinkage, modinfop));
271 }
272 
273 int
274 sdhost_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
275 {
276 	sdhost_t		*shp;
277 	ddi_acc_handle_t	pcih;
278 	uint8_t			slotinfo;
279 	uint8_t			bar;
280 	int			i;
281 	int			rv;
282 
283 	switch (cmd) {
284 	case DDI_ATTACH:
285 		break;
286 
287 	case DDI_RESUME:
288 		return (sdhost_resume(dip));
289 
290 	default:
291 		return (DDI_FAILURE);
292 	}
293 
294 	/*
295 	 * Soft state allocation.
296 	 */
297 	shp = kmem_zalloc(sizeof (*shp), KM_SLEEP);
298 	ddi_set_driver_private(dip, shp);
299 
300 	/*
301 	 * Reset the "slot number", so uninit slot works properly.
302 	 */
303 	for (i = 0; i < SDHOST_MAXSLOTS; i++) {
304 		shp->sh_slots[i].ss_num = -1;
305 	}
306 
307 	/*
308 	 * Initialize DMA attributes.  For now we initialize as for
309 	 * SDMA.  If we add ADMA support we can improve this.
310 	 */
311 	shp->sh_dmaattr.dma_attr_version = DMA_ATTR_V0;
312 	shp->sh_dmaattr.dma_attr_addr_lo = 0;
313 	shp->sh_dmaattr.dma_attr_addr_hi = 0xffffffffU;
314 	shp->sh_dmaattr.dma_attr_count_max = 0xffffffffU;
315 	shp->sh_dmaattr.dma_attr_align = 4096;		/* Ricoh needs it */
316 	shp->sh_dmaattr.dma_attr_burstsizes = 0;	/* for now! */
317 	shp->sh_dmaattr.dma_attr_minxfer = 1;
318 	shp->sh_dmaattr.dma_attr_maxxfer = 0x7ffffU;
319 	shp->sh_dmaattr.dma_attr_sgllen = 1;		/* no scatter/gather */
320 	shp->sh_dmaattr.dma_attr_seg = 0x7ffffU;	/* not to cross 512K */
321 	shp->sh_dmaattr.dma_attr_granular = 1;
322 	shp->sh_dmaattr.dma_attr_flags = 0;
323 
324 	/*
325 	 * PCI configuration access to figure out number of slots present.
326 	 */
327 	if (pci_config_setup(dip, &pcih) != DDI_SUCCESS) {
328 		cmn_err(CE_WARN, "pci_config_setup failed");
329 		goto failed;
330 	}
331 
332 	slotinfo = pci_config_get8(pcih, SLOTINFO);
333 	shp->sh_numslots = SLOTINFO_NSLOT(slotinfo);
334 
335 	if (shp->sh_numslots > SDHOST_MAXSLOTS) {
336 		cmn_err(CE_WARN, "Host reports to have too many slots: %d",
337 		    shp->sh_numslots);
338 		pci_config_teardown(&pcih);
339 		goto failed;
340 	}
341 
342 	/*
343 	 * Enable master accesses and DMA.
344 	 */
345 	pci_config_put16(pcih, PCI_CONF_COMM,
346 	    pci_config_get16(pcih, PCI_CONF_COMM) |
347 	    PCI_COMM_MAE | PCI_COMM_ME);
348 
349 	/*
350 	 * Figure out which BAR to use.  Note that we number BARs from
351 	 * 1, although PCI and SD Host numbers from 0.  (We number
352 	 * from 1, because register number 0 means PCI configuration
353 	 * space in Solaris.)
354 	 */
355 	bar = SLOTINFO_BAR(slotinfo) + 1;
356 
357 	pci_config_teardown(&pcih);
358 
359 	/*
360 	 * Setup interrupts ... supports the new DDI interrupt API.  This
361 	 * will support MSI or MSI-X interrupts if a device is found to
362 	 * support it.
363 	 */
364 	if (sdhost_setup_intr(dip, shp) != DDI_SUCCESS) {
365 		cmn_err(CE_WARN, "Failed to setup interrupts");
366 		goto failed;
367 	}
368 
369 	shp->sh_host = sda_host_alloc(dip, shp->sh_numslots, &sdhost_ops,
370 	    NULL);
371 	if (shp->sh_host == NULL) {
372 		cmn_err(CE_WARN, "Failed allocating SD host structure");
373 		goto failed;
374 	}
375 
376 	/*
377 	 * Configure slots, this also maps registers, enables
378 	 * interrupts, etc.  Most of the hardware setup is done here.
379 	 */
380 	for (i = 0; i < shp->sh_numslots; i++) {
381 		if (sdhost_init_slot(dip, shp, i, bar + i) != DDI_SUCCESS) {
382 			cmn_err(CE_WARN, "Failed initializing slot %d", i);
383 			goto failed;
384 		}
385 	}
386 
387 	ddi_report_dev(dip);
388 
389 	/*
390 	 * Enable device interrupts at the DDI layer.
391 	 */
392 	if (shp->sh_icap & DDI_INTR_FLAG_BLOCK) {
393 		rv = ddi_intr_block_enable(&shp->sh_ihandle, 1);
394 	} else {
395 		rv = ddi_intr_enable(shp->sh_ihandle);
396 	}
397 	if (rv != DDI_SUCCESS) {
398 		cmn_err(CE_WARN, "Failed enabling interrupts");
399 		goto failed;
400 	}
401 
402 	/*
403 	 * Mark the slots online with the framework.  This will cause
404 	 * the framework to probe them for the presence of cards.
405 	 */
406 	if (sda_host_attach(shp->sh_host) != DDI_SUCCESS) {
407 		cmn_err(CE_WARN, "Failed attaching to SDA framework");
408 		if (shp->sh_icap & DDI_INTR_FLAG_BLOCK) {
409 			(void) ddi_intr_block_disable(&shp->sh_ihandle, 1);
410 		} else {
411 			(void) ddi_intr_disable(shp->sh_ihandle);
412 		}
413 		goto failed;
414 	}
415 
416 	return (DDI_SUCCESS);
417 
418 failed:
419 	if (shp->sh_ihandle != NULL) {
420 		(void) ddi_intr_remove_handler(shp->sh_ihandle);
421 		(void) ddi_intr_free(shp->sh_ihandle);
422 	}
423 	for (i = 0; i < shp->sh_numslots; i++)
424 		sdhost_uninit_slot(shp, i);
425 	if (shp->sh_host != NULL)
426 		sda_host_free(shp->sh_host);
427 	kmem_free(shp, sizeof (*shp));
428 
429 	return (DDI_FAILURE);
430 }
431 
432 int
433 sdhost_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
434 {
435 	sdhost_t	*shp;
436 	int		i;
437 
438 	switch (cmd) {
439 	case DDI_DETACH:
440 		break;
441 
442 	case DDI_SUSPEND:
443 		return (sdhost_suspend(dip));
444 
445 	default:
446 		return (DDI_FAILURE);
447 	}
448 
449 	shp = ddi_get_driver_private(dip);
450 
451 	/*
452 	 * Take host offline with the framework.
453 	 */
454 	sda_host_detach(shp->sh_host);
455 
456 	/*
457 	 * Tear down interrupts.
458 	 */
459 	if (shp->sh_ihandle != NULL) {
460 		if (shp->sh_icap & DDI_INTR_FLAG_BLOCK) {
461 			(void) ddi_intr_block_disable(&shp->sh_ihandle, 1);
462 		} else {
463 			(void) ddi_intr_disable(shp->sh_ihandle);
464 		}
465 		(void) ddi_intr_remove_handler(shp->sh_ihandle);
466 		(void) ddi_intr_free(shp->sh_ihandle);
467 	}
468 
469 	/*
470 	 * Tear down register mappings, etc.
471 	 */
472 	for (i = 0; i < shp->sh_numslots; i++)
473 		sdhost_uninit_slot(shp, i);
474 	sda_host_free(shp->sh_host);
475 	kmem_free(shp, sizeof (*shp));
476 
477 	return (DDI_SUCCESS);
478 }
479 
480 int
481 sdhost_quiesce(dev_info_t *dip)
482 {
483 	sdhost_t	*shp;
484 	sdslot_t	*ss;
485 
486 	shp = ddi_get_driver_private(dip);
487 
488 	/* reset each slot separately */
489 	for (int i = 0; i < shp->sh_numslots; i++) {
490 		ss = &shp->sh_slots[i];
491 		if (ss->ss_acch == NULL)
492 			continue;
493 
494 		(void) sdhost_soft_reset(ss, SOFT_RESET_ALL);
495 	}
496 	return (DDI_SUCCESS);
497 }
498 
499 int
500 sdhost_suspend(dev_info_t *dip)
501 {
502 	sdhost_t	*shp;
503 	sdslot_t	*ss;
504 	int		i;
505 
506 	shp = ddi_get_driver_private(dip);
507 
508 	sda_host_suspend(shp->sh_host);
509 
510 	for (i = 0; i < shp->sh_numslots; i++) {
511 		ss = &shp->sh_slots[i];
512 		mutex_enter(&ss->ss_lock);
513 		ss->ss_suspended = B_TRUE;
514 		sdhost_disable_interrupts(ss);
515 		(void) sdhost_soft_reset(ss, SOFT_RESET_ALL);
516 		mutex_exit(&ss->ss_lock);
517 	}
518 	return (DDI_SUCCESS);
519 }
520 
521 int
522 sdhost_resume(dev_info_t *dip)
523 {
524 	sdhost_t	*shp;
525 	sdslot_t	*ss;
526 	int		i;
527 
528 	shp = ddi_get_driver_private(dip);
529 
530 	for (i = 0; i < shp->sh_numslots; i++) {
531 		ss = &shp->sh_slots[i];
532 		mutex_enter(&ss->ss_lock);
533 		ss->ss_suspended = B_FALSE;
534 		(void) sdhost_soft_reset(ss, SOFT_RESET_ALL);
535 		sdhost_enable_interrupts(ss);
536 		mutex_exit(&ss->ss_lock);
537 	}
538 
539 	sda_host_resume(shp->sh_host);
540 
541 	return (DDI_SUCCESS);
542 }
543 
544 sda_err_t
545 sdhost_set_clock(sdslot_t *ss, uint32_t hz)
546 {
547 	uint16_t	div;
548 	uint32_t	val;
549 	uint32_t	clk;
550 	int		count;
551 
552 	/*
553 	 * Shut off the clock to begin.
554 	 */
555 	ss->ss_cardclk = 0;
556 	PUT16(ss, REG_CLOCK_CONTROL, 0);
557 	if (hz == 0) {
558 		return (SDA_EOK);
559 	}
560 
561 	if (ss->ss_baseclk == 0) {
562 		sda_host_log(ss->ss_host, ss->ss_num,
563 		    "Base clock frequency not established.");
564 		return (SDA_EINVAL);
565 	}
566 
567 	if ((hz > 25000000) && ((ss->ss_capab & CAPAB_HIGH_SPEED) != 0)) {
568 		/* this clock requires high speed timings! */
569 		SET8(ss, REG_HOST_CONTROL, HOST_CONTROL_HIGH_SPEED_EN);
570 	} else {
571 		/* don't allow clock to run faster than 25MHz */
572 		hz = min(hz, 25000000);
573 		CLR8(ss, REG_HOST_CONTROL, HOST_CONTROL_HIGH_SPEED_EN);
574 	}
575 
576 	/* figure out the divider */
577 	clk = ss->ss_baseclk;
578 	div  = 1;
579 	while (clk > hz) {
580 		if (div > 0x80)
581 			break;
582 		clk >>= 1;	/* divide clock by two */
583 		div <<= 1;	/* divider goes up by one */
584 	}
585 	div >>= 1;	/* 0 == divide by 1, 1 = divide by 2 */
586 
587 	/*
588 	 * Set the internal clock divider first, without enabling the
589 	 * card clock yet.
590 	 */
591 	PUT16(ss, REG_CLOCK_CONTROL,
592 	    (div << CLOCK_CONTROL_FREQ_SHIFT) | CLOCK_CONTROL_INT_CLOCK_EN);
593 
594 	/*
595 	 * Wait up to 100 msec for the internal clock to stabilize.
596 	 * (The spec does not seem to indicate a maximum timeout, but
597 	 * it also suggests that an infinite loop be used, which is
598 	 * not appropriate for hardened Solaris drivers.)
599 	 */
600 	for (count = 100000; count; count -= 10) {
601 
602 		val = GET16(ss, REG_CLOCK_CONTROL);
603 
604 		if (val & CLOCK_CONTROL_INT_CLOCK_STABLE) {
605 			/* if clock is stable, enable the SD clock pin */
606 			PUT16(ss, REG_CLOCK_CONTROL, val |
607 			    CLOCK_CONTROL_SD_CLOCK_EN);
608 
609 			ss->ss_cardclk = clk;
610 			return (SDA_EOK);
611 		}
612 
613 		drv_usecwait(10);
614 	}
615 
616 	return (SDA_ETIME);
617 }
618 
619 sda_err_t
620 sdhost_soft_reset(sdslot_t *ss, uint8_t bits)
621 {
622 	int	count;
623 
624 	/*
625 	 * There appears to be a bug where Ricoh hosts might have a
626 	 * problem if the host frequency is not set.  If the card
627 	 * isn't present, or we are doing a master reset, just enable
628 	 * the internal clock at its native speed.  (No dividers, and
629 	 * not exposed to card.).
630 	 */
631 	if ((bits == SOFT_RESET_ALL) || !(CHECK_STATE(ss, CARD_INSERTED))) {
632 		PUT16(ss, REG_CLOCK_CONTROL, CLOCK_CONTROL_INT_CLOCK_EN);
633 		/* simple 1msec wait, don't wait for clock to stabilize */
634 		drv_usecwait(1000);
635 		/*
636 		 * reset the card clock & width -- master reset also
637 		 * resets these
638 		 */
639 		ss->ss_cardclk = 0;
640 		ss->ss_width = 1;
641 	}
642 
643 
644 	PUT8(ss, REG_SOFT_RESET, bits);
645 	for (count = 100000; count != 0; count -= 10) {
646 		if ((GET8(ss, REG_SOFT_RESET) & bits) == 0) {
647 			return (SDA_EOK);
648 		}
649 		drv_usecwait(10);
650 	}
651 
652 	return (SDA_ETIME);
653 }
654 
655 void
656 sdhost_disable_interrupts(sdslot_t *ss)
657 {
658 	/* disable slot interrupts for card insert and remove */
659 	PUT16(ss, REG_INT_MASK, 0);
660 	PUT16(ss, REG_INT_EN, 0);
661 
662 	/* disable error interrupts */
663 	PUT16(ss, REG_ERR_MASK, 0);
664 	PUT16(ss, REG_ERR_EN, 0);
665 }
666 
667 void
668 sdhost_enable_interrupts(sdslot_t *ss)
669 {
670 	/*
671 	 * Note that we want to enable reading of the CMD related
672 	 * bits, but we do not want them to generate an interrupt.
673 	 * (The busy wait for typical CMD stuff will normally be less
674 	 * than 10usec, so its simpler/easier to just poll.  Even in
675 	 * the worst case of 100 kHz, the poll is at worst 2 msec.)
676 	 */
677 
678 	/* enable slot interrupts for card insert and remove */
679 	PUT16(ss, REG_INT_MASK, INT_MASK);
680 	PUT16(ss, REG_INT_EN, INT_ENAB);
681 
682 	/* enable error interrupts */
683 	PUT16(ss, REG_ERR_MASK, ERR_MASK);
684 	PUT16(ss, REG_ERR_EN, ERR_ENAB);
685 }
686 
687 int
688 sdhost_setup_intr(dev_info_t *dip, sdhost_t *shp)
689 {
690 	int		itypes;
691 	int		itype;
692 
693 	/*
694 	 * Set up interrupt handler.
695 	 */
696 	if (ddi_intr_get_supported_types(dip, &itypes) != DDI_SUCCESS) {
697 		cmn_err(CE_WARN, "ddi_intr_get_supported_types failed");
698 		return (DDI_FAILURE);
699 	}
700 
701 	/*
702 	 * It turns out that some controllers don't properly implement MSI,
703 	 * but advertise MSI capability in their  PCI config space.
704 	 *
705 	 * While this is really a chip-specific bug, the simplest solution
706 	 * is to just suppress MSI for now by default -- every device seen
707 	 * so far can use FIXED interrupts.
708 	 *
709 	 * We offer an override property, though, just in case someone really
710 	 * wants to force it.
711 	 *
712 	 * We don't do this if the FIXED type isn't supported though!
713 	 */
714 	if (itypes & DDI_INTR_TYPE_FIXED) {
715 		if (!PROPSET(SDHOST_PROP_ENABLE_MSI)) {
716 			itypes &= ~DDI_INTR_TYPE_MSI;
717 		}
718 		if (!PROPSET(SDHOST_PROP_ENABLE_MSIX)) {
719 			itypes &= ~DDI_INTR_TYPE_MSIX;
720 		}
721 	}
722 
723 	/*
724 	 * Interrupt types are bits in a mask.  We know about these ones:
725 	 * FIXED = 1
726 	 * MSI = 2
727 	 * MSIX = 4
728 	 */
729 	for (itype = DDI_INTR_TYPE_MSIX; itype != 0; itype >>= 1) {
730 
731 		int			count;
732 
733 		if ((itypes & itype) == 0) {
734 			/* this type is not supported on this device! */
735 			continue;
736 		}
737 
738 		if ((ddi_intr_get_nintrs(dip, itype, &count) != DDI_SUCCESS) ||
739 		    (count == 0)) {
740 			cmn_err(CE_WARN, "ddi_intr_get_nintrs failed");
741 			continue;
742 		}
743 
744 		/*
745 		 * We have not seen a host device with multiple
746 		 * interrupts (one per slot?), and the spec does not
747 		 * indicate that they exist.  But if one ever occurs,
748 		 * we spew a warning to help future debugging/support
749 		 * efforts.
750 		 */
751 		if (count > 1) {
752 			cmn_err(CE_WARN, "Controller offers %d interrupts, "
753 			    "but driver only supports one", count);
754 			continue;
755 		}
756 
757 		if ((ddi_intr_alloc(dip, &shp->sh_ihandle, itype, 0, 1,
758 		    &count, DDI_INTR_ALLOC_NORMAL) != DDI_SUCCESS) ||
759 		    (count != 1)) {
760 			cmn_err(CE_WARN, "ddi_intr_alloc failed");
761 			continue;
762 		}
763 
764 		if (ddi_intr_get_pri(shp->sh_ihandle, &shp->sh_ipri) !=
765 		    DDI_SUCCESS) {
766 			cmn_err(CE_WARN, "ddi_intr_get_pri failed");
767 			(void) ddi_intr_free(shp->sh_ihandle);
768 			shp->sh_ihandle = NULL;
769 			continue;
770 		}
771 
772 		if (shp->sh_ipri >= ddi_intr_get_hilevel_pri()) {
773 			cmn_err(CE_WARN, "Hi level interrupt not supported");
774 			(void) ddi_intr_free(shp->sh_ihandle);
775 			shp->sh_ihandle = NULL;
776 			continue;
777 		}
778 
779 		if (ddi_intr_get_cap(shp->sh_ihandle, &shp->sh_icap) !=
780 		    DDI_SUCCESS) {
781 			cmn_err(CE_WARN, "ddi_intr_get_cap failed");
782 			(void) ddi_intr_free(shp->sh_ihandle);
783 			shp->sh_ihandle = NULL;
784 			continue;
785 		}
786 
787 		if (ddi_intr_add_handler(shp->sh_ihandle, sdhost_intr,
788 		    shp, NULL) != DDI_SUCCESS) {
789 			cmn_err(CE_WARN, "ddi_intr_add_handler failed");
790 			(void) ddi_intr_free(shp->sh_ihandle);
791 			shp->sh_ihandle = NULL;
792 			continue;
793 		}
794 
795 		return (DDI_SUCCESS);
796 	}
797 
798 	return (DDI_FAILURE);
799 }
800 
801 void
802 sdhost_xfer_done(sdslot_t *ss, sda_err_t errno)
803 {
804 	if ((errno == SDA_EOK) && (ss->ss_resid != 0)) {
805 		/* an unexpected partial transfer was found */
806 		errno = SDA_ERESID;
807 	}
808 	ss->ss_blksz = 0;
809 	ss->ss_resid = 0;
810 
811 	if (errno != SDA_EOK) {
812 		(void) sdhost_soft_reset(ss, SOFT_RESET_CMD);
813 		(void) sdhost_soft_reset(ss, SOFT_RESET_DAT);
814 
815 		/* send a STOP command if necessary */
816 		if (ss->ss_mode & XFR_MODE_AUTO_CMD12) {
817 			PUT32(ss, REG_ARGUMENT, 0);
818 			PUT16(ss, REG_COMMAND,
819 			    (CMD_STOP_TRANSMIT << 8) |
820 			    COMMAND_TYPE_NORM | COMMAND_INDEX_CHECK_EN |
821 			    COMMAND_CRC_CHECK_EN | COMMAND_RESP_48_BUSY);
822 		}
823 	}
824 
825 	sda_host_transfer(ss->ss_host, ss->ss_num, errno);
826 }
827 
828 uint_t
829 sdhost_slot_intr(sdslot_t *ss)
830 {
831 	uint16_t	intr;
832 	uint16_t	errs;
833 	caddr_t		data;
834 	int		count;
835 
836 	mutex_enter(&ss->ss_lock);
837 
838 	if (ss->ss_suspended) {
839 		mutex_exit(&ss->ss_lock);
840 		return (DDI_INTR_UNCLAIMED);
841 	}
842 
843 	intr = GET16(ss, REG_INT_STAT);
844 	if (intr == 0) {
845 		mutex_exit(&ss->ss_lock);
846 		return (DDI_INTR_UNCLAIMED);
847 	}
848 	errs = GET16(ss, REG_ERR_STAT);
849 
850 	if (intr & (INT_REM | INT_INS)) {
851 
852 		PUT16(ss, REG_INT_STAT, intr);
853 		mutex_exit(&ss->ss_lock);
854 
855 		sda_host_detect(ss->ss_host, ss->ss_num);
856 		/* no further interrupt processing this cycle */
857 		return (DDI_INTR_CLAIMED);
858 	}
859 
860 	if (intr & INT_DMA) {
861 		/*
862 		 * We have crossed a DMA/page boundary.  Cope with it.
863 		 */
864 		/*
865 		 * Apparently some sdhost controllers issue a final
866 		 * DMA interrupt if the DMA completes on a boundary,
867 		 * even though there is no further data to transfer.
868 		 *
869 		 * There might be a risk here of the controller
870 		 * continuing to access the same data over and over
871 		 * again, but we accept the risk.
872 		 */
873 		PUT16(ss, REG_INT_STAT, INT_DMA);
874 	}
875 
876 	if (intr & INT_RD) {
877 		/*
878 		 * PIO read!  PIO is quite suboptimal, but we expect
879 		 * performance critical applications to use DMA
880 		 * whenever possible.  We have to stage this through
881 		 * the bounce buffer to meet alignment considerations.
882 		 */
883 
884 		PUT16(ss, REG_INT_STAT, INT_RD);
885 
886 		while ((ss->ss_resid > 0) && CHECK_STATE(ss, BUF_RD_EN)) {
887 
888 			data = ss->ss_bounce;
889 			count = ss->ss_blksz;
890 
891 			ASSERT(count > 0);
892 			ASSERT(ss->ss_kvaddr != NULL);
893 
894 			while (count >= sizeof (uint32_t)) {
895 				*(uint32_t *)(void *)data = GETDATA32(ss);
896 				data += sizeof (uint32_t);
897 				count -= sizeof (uint32_t);
898 			}
899 			while (count >= sizeof (uint16_t)) {
900 				*(uint16_t *)(void *)data = GETDATA16(ss);
901 				data += sizeof (uint16_t);
902 				count -= sizeof (uint16_t);
903 			}
904 			while (count >= sizeof (uint8_t)) {
905 				*(uint8_t *)data = GETDATA8(ss);
906 				data += sizeof (uint8_t);
907 				count -= sizeof (uint8_t);
908 			}
909 
910 			bcopy(ss->ss_bounce, ss->ss_kvaddr, ss->ss_blksz);
911 			ss->ss_kvaddr += ss->ss_blksz;
912 			ss->ss_resid--;
913 		}
914 	}
915 
916 	if (intr & INT_WR) {
917 		/*
918 		 * PIO write!  PIO is quite suboptimal, but we expect
919 		 * performance critical applications to use DMA
920 		 * whenever possible.  We have to stage this through
921 		 * the bounce buffer to meet alignment considerations.
922 		 */
923 
924 		PUT16(ss, REG_INT_STAT, INT_WR);
925 
926 		while ((ss->ss_resid > 0) && CHECK_STATE(ss, BUF_WR_EN)) {
927 
928 			data = ss->ss_bounce;
929 			count = ss->ss_blksz;
930 
931 			ASSERT(count > 0);
932 			ASSERT(ss->ss_kvaddr != NULL);
933 
934 			bcopy(ss->ss_kvaddr, data, count);
935 			while (count >= sizeof (uint32_t)) {
936 				PUTDATA32(ss, *(uint32_t *)(void *)data);
937 				data += sizeof (uint32_t);
938 				count -= sizeof (uint32_t);
939 			}
940 			while (count >= sizeof (uint16_t)) {
941 				PUTDATA16(ss, *(uint16_t *)(void *)data);
942 				data += sizeof (uint16_t);
943 				count -= sizeof (uint16_t);
944 			}
945 			while (count >= sizeof (uint8_t)) {
946 				PUTDATA8(ss, *(uint8_t *)data);
947 				data += sizeof (uint8_t);
948 				count -= sizeof (uint8_t);
949 			}
950 
951 			ss->ss_kvaddr += ss->ss_blksz;
952 			ss->ss_resid--;
953 		}
954 	}
955 
956 	if (intr & INT_XFR) {
957 		if ((ss->ss_mode & (XFR_MODE_READ | XFR_MODE_DMA_EN)) ==
958 		    (XFR_MODE_READ | XFR_MODE_DMA_EN)) {
959 			(void) ddi_dma_sync(ss->ss_bufdmah, 0, 0,
960 			    DDI_DMA_SYNC_FORKERNEL);
961 			bcopy(ss->ss_bounce, ss->ss_kvaddr, ss->ss_rcnt);
962 			ss->ss_rcnt = 0;
963 		}
964 		PUT16(ss, REG_INT_STAT, INT_XFR);
965 
966 		sdhost_xfer_done(ss, SDA_EOK);
967 	}
968 
969 	if (intr & INT_ERR) {
970 		PUT16(ss, REG_ERR_STAT, errs);
971 		PUT16(ss, REG_INT_STAT, INT_ERR);
972 
973 		if (errs & ERR_DAT) {
974 			if ((errs & ERR_DAT_END) == ERR_DAT_END) {
975 				sdhost_xfer_done(ss, SDA_EPROTO);
976 			} else if ((errs & ERR_DAT_CRC) == ERR_DAT_CRC) {
977 				sdhost_xfer_done(ss, SDA_ECRC7);
978 			} else {
979 				sdhost_xfer_done(ss, SDA_ETIME);
980 			}
981 
982 		} else if (errs & ERR_ACMD12) {
983 			/*
984 			 * Generally, this is bad news.  we need a full
985 			 * reset to recover properly.
986 			 */
987 			sdhost_xfer_done(ss, SDA_ECMD12);
988 		}
989 
990 		/*
991 		 * This asynchronous error leaves the slot more or less
992 		 * useless.  Report it to the framework.
993 		 */
994 		if (errs & ERR_CURRENT) {
995 			sda_host_fault(ss->ss_host, ss->ss_num,
996 			    SDA_FAULT_CURRENT);
997 		}
998 	}
999 
1000 	mutex_exit(&ss->ss_lock);
1001 
1002 	return (DDI_INTR_CLAIMED);
1003 }
1004 
1005 /*ARGSUSED1*/
1006 uint_t
1007 sdhost_intr(caddr_t arg1, caddr_t arg2)
1008 {
1009 	sdhost_t	*shp = (void *)arg1;
1010 	int		rv = DDI_INTR_UNCLAIMED;
1011 	int		num;
1012 
1013 	/* interrupt for each of the slots present in the system */
1014 	for (num = 0; num < shp->sh_numslots; num++) {
1015 		if (sdhost_slot_intr(&shp->sh_slots[num]) ==
1016 		    DDI_INTR_CLAIMED) {
1017 			rv = DDI_INTR_CLAIMED;
1018 		}
1019 	}
1020 	return (rv);
1021 }
1022 
1023 int
1024 sdhost_init_slot(dev_info_t *dip, sdhost_t *shp, int num, int bar)
1025 {
1026 	sdslot_t	*ss;
1027 	uint32_t	capab;
1028 	uint32_t	clk;
1029 	char		ksname[16];
1030 	size_t		blen;
1031 	unsigned	ndmac;
1032 	int		rv;
1033 
1034 	/*
1035 	 * Register the private state.
1036 	 */
1037 	ss = &shp->sh_slots[num];
1038 	ss->ss_host = shp->sh_host;
1039 	ss->ss_num = num;
1040 	sda_host_set_private(shp->sh_host, num, ss);
1041 	/*
1042 	 * Initialize core data structure, locks, etc.
1043 	 */
1044 	mutex_init(&ss->ss_lock, NULL, MUTEX_DRIVER,
1045 	    DDI_INTR_PRI(shp->sh_ipri));
1046 
1047 	/*
1048 	 * Set up DMA.
1049 	 */
1050 	rv = ddi_dma_alloc_handle(dip, &shp->sh_dmaattr,
1051 	    DDI_DMA_SLEEP, NULL, &ss->ss_bufdmah);
1052 	if (rv != DDI_SUCCESS) {
1053 		cmn_err(CE_WARN, "Failed to alloc dma handle (%d)!", rv);
1054 		return (DDI_FAILURE);
1055 	}
1056 
1057 	rv = ddi_dma_mem_alloc(ss->ss_bufdmah, SDHOST_BOUNCESZ,
1058 	    &sdhost_bufattr, DDI_DMA_CONSISTENT, DDI_DMA_SLEEP, NULL,
1059 	    &ss->ss_bounce, &blen, &ss->ss_bufacch);
1060 	if (rv != DDI_SUCCESS) {
1061 		cmn_err(CE_WARN, "Failed to alloc bounce buffer (%d)!", rv);
1062 		return (DDI_FAILURE);
1063 	}
1064 
1065 	rv = ddi_dma_addr_bind_handle(ss->ss_bufdmah, NULL, ss->ss_bounce,
1066 	    blen, DDI_DMA_RDWR | DDI_DMA_CONSISTENT, DDI_DMA_SLEEP, NULL,
1067 	    &ss->ss_bufdmac, &ndmac);
1068 	if ((rv != DDI_DMA_MAPPED) || (ndmac != 1)) {
1069 		cmn_err(CE_WARN, "Failed to bind DMA bounce buffer (%d, %u)!",
1070 		    rv, ndmac);
1071 		return (DDI_FAILURE);
1072 	}
1073 
1074 	/*
1075 	 * Set up virtual kstats.
1076 	 */
1077 	(void) snprintf(ksname, sizeof (ksname), "slot%d", num);
1078 	ss->ss_ksp = kstat_create(ddi_driver_name(dip), ddi_get_instance(dip),
1079 	    ksname, "misc", KSTAT_TYPE_NAMED,
1080 	    sizeof (sdstats_t) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL);
1081 	if (ss->ss_ksp != NULL) {
1082 		sdstats_t	*sp = &ss->ss_stats;
1083 		ss->ss_ksp->ks_data = sp;
1084 		ss->ss_ksp->ks_private = ss;
1085 		ss->ss_ksp->ks_lock = &ss->ss_lock;
1086 		/* counters are 64 bits wide */
1087 		kstat_named_init(&sp->ks_ncmd, "ncmd", KSTAT_DATA_UINT64);
1088 		kstat_named_init(&sp->ks_ixfr, "ixfr", KSTAT_DATA_UINT64);
1089 		kstat_named_init(&sp->ks_oxfr, "oxfr", KSTAT_DATA_UINT64);
1090 		kstat_named_init(&sp->ks_ibytes, "ibytes", KSTAT_DATA_UINT64);
1091 		kstat_named_init(&sp->ks_obytes, "obytes", KSTAT_DATA_UINT64);
1092 		kstat_named_init(&sp->ks_npio, "npio", KSTAT_DATA_UINT64);
1093 		kstat_named_init(&sp->ks_ndma, "ndma", KSTAT_DATA_UINT64);
1094 		kstat_named_init(&sp->ks_nmulti, "nmulti", KSTAT_DATA_UINT64);
1095 		/* these aren't counters -- leave them at 32 bits */
1096 		kstat_named_init(&sp->ks_baseclk, "baseclk", KSTAT_DATA_UINT32);
1097 		kstat_named_init(&sp->ks_cardclk, "cardclk", KSTAT_DATA_UINT32);
1098 		kstat_named_init(&sp->ks_tmusecs, "tmusecs", KSTAT_DATA_UINT32);
1099 		kstat_named_init(&sp->ks_width, "width", KSTAT_DATA_UINT32);
1100 		kstat_named_init(&sp->ks_flags, "flags", KSTAT_DATA_UINT32);
1101 		kstat_named_init(&sp->ks_capab, "capab", KSTAT_DATA_UINT32);
1102 		kstat_install(ss->ss_ksp);
1103 	}
1104 
1105 	if (PROPSET(SDHOST_PROP_FORCE_PIO)) {
1106 		ss->ss_flags |= SDFLAG_FORCE_PIO;
1107 	}
1108 	if (PROPSET(SDHOST_PROP_FORCE_DMA)) {
1109 		ss->ss_flags |= SDFLAG_FORCE_DMA;
1110 	}
1111 
1112 	if (ddi_regs_map_setup(dip, bar, &ss->ss_regva, 0, 0, &sdhost_regattr,
1113 	    &ss->ss_acch) != DDI_SUCCESS) {
1114 		cmn_err(CE_WARN, "Failed to map registers!");
1115 		return (DDI_FAILURE);
1116 	}
1117 
1118 	/* reset before reading capabilities */
1119 	if (sdhost_soft_reset(ss, SOFT_RESET_ALL) != SDA_EOK)
1120 		return (DDI_FAILURE);
1121 
1122 	capab = GET64(ss, REG_CAPAB) & 0xffffffffU; /* upper bits reserved */
1123 	ss->ss_capab = capab;
1124 
1125 	/* host voltages in OCR format */
1126 	ss->ss_ocr = 0;
1127 	if (capab & CAPAB_18V)
1128 		ss->ss_ocr |= OCR_18_19V;	/* 1.8V */
1129 	if (capab & CAPAB_30V)
1130 		ss->ss_ocr |= OCR_30_31V;
1131 	if (capab & CAPAB_33V)
1132 		ss->ss_ocr |= OCR_32_33V;
1133 
1134 	/* base clock */
1135 	ss->ss_baseclk =
1136 	    ((capab & CAPAB_BASE_FREQ_MASK) >> CAPAB_BASE_FREQ_SHIFT);
1137 	ss->ss_baseclk *= 1000000;
1138 
1139 	/*
1140 	 * Timeout clock.  We can calculate this using the following
1141 	 * formula:
1142 	 *
1143 	 * (1000000 usec/1sec) * (1sec/tmoutclk) * base factor = clock time
1144 	 *
1145 	 * Clock time is the length of the base clock in usecs.
1146 	 *
1147 	 * Our base factor is 2^13, which is the shortest clock we
1148 	 * can count.
1149 	 *
1150 	 * To simplify the math and avoid overflow, we cancel out the
1151 	 * zeros for kHz or MHz.  Since we want to wait more clocks, not
1152 	 * less, on error, we truncate the result rather than rounding
1153 	 * up.
1154 	 */
1155 	clk = ((capab & CAPAB_TIMEOUT_FREQ_MASK) >> CAPAB_TIMEOUT_FREQ_SHIFT);
1156 	if ((ss->ss_baseclk == 0) || (clk == 0)) {
1157 		cmn_err(CE_WARN, "Unable to determine clock frequencies");
1158 		return (DDI_FAILURE);
1159 	}
1160 
1161 	if (capab & CAPAB_TIMEOUT_UNITS) {
1162 		/* MHz */
1163 		ss->ss_tmusecs = (1 << 13) / clk;
1164 		clk *= 1000000;
1165 	} else {
1166 		/* kHz */
1167 		ss->ss_tmusecs = (1000 * (1 << 13)) / clk;
1168 		clk *= 1000;
1169 	}
1170 
1171 	/*
1172 	 * Calculation of the timeout.
1173 	 *
1174 	 * SDIO cards use a 1sec timeout, and SDHC cards use fixed
1175 	 * 100msec for read and 250 msec for write.
1176 	 *
1177 	 * Legacy cards running at 375kHz have a worst case of about
1178 	 * 15 seconds.  Running at 25MHz (the standard speed) it is
1179 	 * about 100msec for read, and about 3.2 sec for write.
1180 	 * Typical values are 1/100th that, or about 1msec for read,
1181 	 * and 32 msec for write.
1182 	 *
1183 	 * No transaction at full speed should ever take more than 4
1184 	 * seconds.  (Some slow legacy cards might have trouble, but
1185 	 * we'll worry about them if they ever are seen.  Nobody wants
1186 	 * to wait 4 seconds to access a single block anyway!)
1187 	 *
1188 	 * To get to 4 seconds, we continuously double usec until we
1189 	 * get to the maximum value, or a timeout greater than 4
1190 	 * seconds.
1191 	 *
1192 	 * Note that for high-speed timeout clocks, we might not be
1193 	 * able to get to the full 4 seconds.  E.g. with a 48MHz
1194 	 * timeout clock, we can only get to about 2.8 seconds.  Its
1195 	 * possible that there could be some slow MMC cards that will
1196 	 * timeout at this clock rate, but it seems unlikely.  (The
1197 	 * device would have to be pressing the very worst times,
1198 	 * against the 100-fold "permissive" window allowed, and
1199 	 * running at only 12.5MHz.)
1200 	 *
1201 	 * XXX: this could easily be a tunable.  Someone dealing with only
1202 	 * reasonable cards could set this to just 1 second.
1203 	 */
1204 	for (ss->ss_tmoutclk = 0; ss->ss_tmoutclk < 14; ss->ss_tmoutclk++) {
1205 		if ((ss->ss_tmusecs * (1 << ss->ss_tmoutclk)) >= 4000000) {
1206 			break;
1207 		}
1208 	}
1209 
1210 	/*
1211 	 * Enable slot interrupts.
1212 	 */
1213 	sdhost_enable_interrupts(ss);
1214 
1215 	return (DDI_SUCCESS);
1216 }
1217 
1218 void
1219 sdhost_uninit_slot(sdhost_t *shp, int num)
1220 {
1221 	sdslot_t	*ss;
1222 
1223 	ss = &shp->sh_slots[num];
1224 
1225 	if (ss->ss_acch != NULL)
1226 		(void) sdhost_soft_reset(ss, SOFT_RESET_ALL);
1227 
1228 	if (ss->ss_bufdmac.dmac_address)
1229 		(void) ddi_dma_unbind_handle(ss->ss_bufdmah);
1230 
1231 	if (ss->ss_bufacch != NULL)
1232 		ddi_dma_mem_free(&ss->ss_bufacch);
1233 
1234 	if (ss->ss_bufdmah != NULL)
1235 		ddi_dma_free_handle(&ss->ss_bufdmah);
1236 
1237 	if (ss->ss_ksp != NULL)
1238 		kstat_delete(ss->ss_ksp);
1239 
1240 	if (ss->ss_acch != NULL)
1241 		ddi_regs_map_free(&ss->ss_acch);
1242 
1243 	if (ss->ss_num != -1)
1244 		mutex_destroy(&ss->ss_lock);
1245 }
1246 
1247 void
1248 sdhost_get_response(sdslot_t *ss, sda_cmd_t *cmdp)
1249 {
1250 	uint32_t	*resp = cmdp->sc_response;
1251 	int		i;
1252 
1253 	resp[0] = GET32(ss, REG_RESP1);
1254 	resp[1] = GET32(ss, REG_RESP2);
1255 	resp[2] = GET32(ss, REG_RESP3);
1256 	resp[3] = GET32(ss, REG_RESP4);
1257 
1258 	/*
1259 	 * Response 2 is goofy because the host drops the low
1260 	 * order CRC bits.  This makes it a bit awkward, so we
1261 	 * have to shift the bits to make it work out right.
1262 	 *
1263 	 * Note that the framework expects the 32 bit
1264 	 * words to be ordered in LE fashion.  (The
1265 	 * bits within the words are in native order).
1266 	 */
1267 	if (cmdp->sc_rtype == R2) {
1268 		for (i = 3; i > 0; i--) {
1269 			resp[i] <<= 8;
1270 			resp[i] |= (resp[i - 1] >> 24);
1271 		}
1272 		resp[0] <<= 8;
1273 	}
1274 }
1275 
1276 sda_err_t
1277 sdhost_wait_cmd(sdslot_t *ss, sda_cmd_t *cmdp)
1278 {
1279 	int		i;
1280 	uint16_t	errs;
1281 	sda_err_t	rv;
1282 
1283 	/*
1284 	 * Worst case for 100kHz timeout is 2msec (200 clocks), we add
1285 	 * a tiny bit for safety.  (Generally timeout will be far, far
1286 	 * less than that.)
1287 	 *
1288 	 * Note that at more typical 12MHz (and normally it will be
1289 	 * even faster than that!) that the device timeout is only
1290 	 * 16.67 usec.  We could be smarter and reduce the delay time,
1291 	 * but that would require putting more intelligence into the
1292 	 * code, and we don't expect CMD timeout to normally occur
1293 	 * except during initialization.  (At which time we need the
1294 	 * full timeout anyway.)
1295 	 *
1296 	 * Checking the ERR_STAT will normally cause the timeout to
1297 	 * terminate to finish early if the device is healthy, anyway.
1298 	 */
1299 
1300 	for (i = 3000; i > 0; i -= 5) {
1301 		if (GET16(ss, REG_INT_STAT) & INT_CMD) {
1302 
1303 			PUT16(ss, REG_INT_STAT, INT_CMD);
1304 
1305 			/* command completed */
1306 			sdhost_get_response(ss, cmdp);
1307 			return (SDA_EOK);
1308 		}
1309 
1310 		if ((errs = (GET16(ss, REG_ERR_STAT) & ERR_CMD)) != 0) {
1311 			PUT16(ss, REG_ERR_STAT, errs);
1312 
1313 			/* command timeout isn't a host failure */
1314 			if ((errs & ERR_CMD_TMO) == ERR_CMD_TMO) {
1315 				rv = SDA_ETIME;
1316 			} else if ((errs & ERR_CMD_CRC) == ERR_CMD_CRC) {
1317 				rv = SDA_ECRC7;
1318 			} else {
1319 				rv = SDA_EPROTO;
1320 			}
1321 			goto error;
1322 		}
1323 
1324 		drv_usecwait(5);
1325 	}
1326 
1327 	rv = SDA_ETIME;
1328 
1329 error:
1330 	/*
1331 	 * NB: We need to soft reset the CMD and DAT
1332 	 * lines after a failure of this sort.
1333 	 */
1334 	(void) sdhost_soft_reset(ss, SOFT_RESET_CMD);
1335 	(void) sdhost_soft_reset(ss, SOFT_RESET_DAT);
1336 
1337 	return (rv);
1338 }
1339 
1340 sda_err_t
1341 sdhost_poll(void *arg)
1342 {
1343 	sdslot_t	*ss = arg;
1344 
1345 	(void) sdhost_slot_intr(ss);
1346 	return (SDA_EOK);
1347 }
1348 
1349 sda_err_t
1350 sdhost_cmd(void *arg, sda_cmd_t *cmdp)
1351 {
1352 	sdslot_t	*ss = arg;
1353 	uint16_t	command;
1354 	uint16_t	mode;
1355 	sda_err_t	rv;
1356 
1357 	/*
1358 	 * Command register:
1359 	 * bit 13-8	= command index
1360 	 * bit 7-6	= command type (always zero for us!)
1361 	 * bit 5	= data present select
1362 	 * bit 4	= command index check (always on!)
1363 	 * bit 3	= command CRC check enable
1364 	 * bit 2	= reserved
1365 	 * bit 1-0	= response type
1366 	 */
1367 
1368 	command = ((uint16_t)cmdp->sc_index << 8);
1369 	command |= COMMAND_TYPE_NORM |
1370 	    COMMAND_INDEX_CHECK_EN | COMMAND_CRC_CHECK_EN;
1371 
1372 	switch (cmdp->sc_rtype) {
1373 	case R0:
1374 		command |= COMMAND_RESP_NONE;
1375 		break;
1376 	case R1:
1377 	case R5:
1378 	case R6:
1379 	case R7:
1380 		command |= COMMAND_RESP_48;
1381 		break;
1382 	case R1b:
1383 	case R5b:
1384 		command |= COMMAND_RESP_48_BUSY;
1385 		break;
1386 	case R2:
1387 		command |= COMMAND_RESP_136;
1388 		command &= ~(COMMAND_INDEX_CHECK_EN | COMMAND_CRC_CHECK_EN);
1389 		break;
1390 	case R3:
1391 	case R4:
1392 		command |= COMMAND_RESP_48;
1393 		command &= ~COMMAND_CRC_CHECK_EN;
1394 		command &= ~COMMAND_INDEX_CHECK_EN;
1395 		break;
1396 	default:
1397 		return (SDA_EINVAL);
1398 	}
1399 
1400 	mutex_enter(&ss->ss_lock);
1401 	if (ss->ss_suspended) {
1402 		mutex_exit(&ss->ss_lock);
1403 		return (SDA_ESUSPENDED);
1404 	}
1405 
1406 	if (cmdp->sc_nblks != 0) {
1407 		uint16_t	blksz;
1408 		uint16_t	nblks;
1409 
1410 		blksz = cmdp->sc_blksz;
1411 		nblks = cmdp->sc_nblks;
1412 
1413 		/*
1414 		 * Ensure that we have good data.
1415 		 */
1416 		if ((blksz < 1) || (blksz > 2048)) {
1417 			mutex_exit(&ss->ss_lock);
1418 			return (SDA_EINVAL);
1419 		}
1420 		command |= COMMAND_DATA_PRESENT;
1421 
1422 		ss->ss_blksz = blksz;
1423 
1424 		ss->ss_kvaddr = (void *)cmdp->sc_kvaddr;
1425 		ss->ss_rcnt = 0;
1426 		ss->ss_resid = 0;
1427 
1428 		/*
1429 		 * Only SDMA for now.  We can investigate ADMA2 later.
1430 		 * (Right now we don't have ADMA2 capable hardware.)
1431 		 * We always use a bounce buffer, which solves weird
1432 		 * problems with certain controllers.  Doing this with
1433 		 * a large contiguous buffer may be faster than
1434 		 * servicing all the little per-page interrupts
1435 		 * anyway. (Bcopy of 64 K vs. 16 interrupts.)
1436 		 */
1437 		if (((ss->ss_capab & CAPAB_SDMA) != 0) &&
1438 		    ((ss->ss_flags & SDFLAG_FORCE_PIO) == 0) &&
1439 		    ((blksz * nblks) <= SDHOST_BOUNCESZ)) {
1440 
1441 			if (cmdp->sc_flags & SDA_CMDF_WRITE) {
1442 				/*
1443 				 * if we're writing, prepare initial round
1444 				 * of data
1445 				 */
1446 				bcopy(cmdp->sc_kvaddr, ss->ss_bounce,
1447 				    nblks * blksz);
1448 				(void) ddi_dma_sync(ss->ss_bufdmah, 0, 0,
1449 				    DDI_DMA_SYNC_FORDEV);
1450 			} else {
1451 				ss->ss_rcnt = nblks * blksz;
1452 			}
1453 			PUT32(ss, REG_SDMA_ADDR, ss->ss_bufdmac.dmac_address);
1454 			mode = XFR_MODE_DMA_EN;
1455 			PUT16(ss, REG_BLKSZ, BLKSZ_BOUNDARY_512K | blksz);
1456 			ss->ss_ndma++;
1457 
1458 		} else {
1459 			mode = 0;
1460 			ss->ss_npio++;
1461 			ss->ss_resid = nblks;
1462 			PUT16(ss, REG_BLKSZ, blksz);
1463 		}
1464 
1465 		if (nblks > 1) {
1466 			mode |= XFR_MODE_MULTI | XFR_MODE_COUNT;
1467 			if (cmdp->sc_flags & SDA_CMDF_AUTO_CMD12)
1468 				mode |= XFR_MODE_AUTO_CMD12;
1469 			ss->ss_nmulti++;
1470 		}
1471 		if ((cmdp->sc_flags & SDA_CMDF_READ) != 0) {
1472 			mode |= XFR_MODE_READ;
1473 			ss->ss_ixfr++;
1474 			ss->ss_ibytes += nblks * blksz;
1475 		} else {
1476 			ss->ss_oxfr++;
1477 			ss->ss_obytes += nblks * blksz;
1478 		}
1479 
1480 		ss->ss_mode = mode;
1481 
1482 		PUT8(ss, REG_TIMEOUT_CONTROL, ss->ss_tmoutclk);
1483 		PUT16(ss, REG_BLOCK_COUNT, nblks);
1484 		PUT16(ss, REG_XFR_MODE, mode);
1485 	}
1486 
1487 	PUT32(ss, REG_ARGUMENT, cmdp->sc_argument);
1488 	PUT16(ss, REG_COMMAND, command);
1489 
1490 	ss->ss_ncmd++;
1491 	rv = sdhost_wait_cmd(ss, cmdp);
1492 
1493 	mutex_exit(&ss->ss_lock);
1494 
1495 	return (rv);
1496 }
1497 
1498 sda_err_t
1499 sdhost_getprop(void *arg, sda_prop_t prop, uint32_t *val)
1500 {
1501 	sdslot_t	*ss = arg;
1502 	sda_err_t	rv = 0;
1503 
1504 	mutex_enter(&ss->ss_lock);
1505 
1506 	if (ss->ss_suspended) {
1507 		mutex_exit(&ss->ss_lock);
1508 		return (SDA_ESUSPENDED);
1509 	}
1510 	switch (prop) {
1511 	case SDA_PROP_INSERTED:
1512 		if (CHECK_STATE(ss, CARD_INSERTED)) {
1513 			*val = B_TRUE;
1514 		} else {
1515 			*val = B_FALSE;
1516 		}
1517 		break;
1518 
1519 	case SDA_PROP_WPROTECT:
1520 		if (CHECK_STATE(ss, WRITE_ENABLE)) {
1521 			*val = B_FALSE;
1522 		} else {
1523 			*val = B_TRUE;
1524 		}
1525 		break;
1526 
1527 	case SDA_PROP_OCR:
1528 		*val = ss->ss_ocr;
1529 		break;
1530 
1531 	case SDA_PROP_CLOCK:
1532 		*val = ss->ss_cardclk;
1533 		break;
1534 
1535 	case SDA_PROP_CAP_HISPEED:
1536 		if ((ss->ss_capab & CAPAB_HIGH_SPEED) != 0) {
1537 			*val = B_TRUE;
1538 		} else {
1539 			*val = B_FALSE;
1540 		}
1541 		break;
1542 
1543 	case SDA_PROP_CAP_4BITS:
1544 		*val = B_TRUE;
1545 		break;
1546 
1547 	case SDA_PROP_CAP_NOPIO:
1548 		/*
1549 		 * We might have to use PIO for buffers that don't
1550 		 * have reasonable alignments.  A few controllers seem
1551 		 * not to deal with granularity or alignments of
1552 		 * something other 32-bits.
1553 		 */
1554 		*val = B_FALSE;
1555 		break;
1556 
1557 	case SDA_PROP_CAP_INTR:
1558 	case SDA_PROP_CAP_8BITS:
1559 		*val = B_FALSE;
1560 		break;
1561 
1562 	default:
1563 		rv = SDA_ENOTSUP;
1564 		break;
1565 	}
1566 	mutex_exit(&ss->ss_lock);
1567 
1568 	return (rv);
1569 }
1570 
1571 sda_err_t
1572 sdhost_setprop(void *arg, sda_prop_t prop, uint32_t val)
1573 {
1574 	sdslot_t	*ss = arg;
1575 	sda_err_t	rv = SDA_EOK;
1576 
1577 	mutex_enter(&ss->ss_lock);
1578 
1579 	if (ss->ss_suspended) {
1580 		mutex_exit(&ss->ss_lock);
1581 		return (SDA_ESUSPENDED);
1582 	}
1583 
1584 	switch (prop) {
1585 	case SDA_PROP_LED:
1586 		if (val) {
1587 			SET8(ss, REG_HOST_CONTROL, HOST_CONTROL_LED_ON);
1588 		} else {
1589 			CLR8(ss, REG_HOST_CONTROL, HOST_CONTROL_LED_ON);
1590 		}
1591 		break;
1592 
1593 	case SDA_PROP_CLOCK:
1594 		rv = sdhost_set_clock(arg, val);
1595 		break;
1596 
1597 	case SDA_PROP_BUSWIDTH:
1598 		switch (val) {
1599 		case 1:
1600 			ss->ss_width = val;
1601 			CLR8(ss, REG_HOST_CONTROL, HOST_CONTROL_DATA_WIDTH);
1602 			break;
1603 		case 4:
1604 			ss->ss_width = val;
1605 			SET8(ss, REG_HOST_CONTROL, HOST_CONTROL_DATA_WIDTH);
1606 			break;
1607 		default:
1608 			rv = SDA_EINVAL;
1609 		}
1610 		break;
1611 
1612 	case SDA_PROP_OCR:
1613 		val &= ss->ss_ocr;
1614 
1615 		if (val & OCR_17_18V) {
1616 			PUT8(ss, REG_POWER_CONTROL, POWER_CONTROL_18V);
1617 			PUT8(ss, REG_POWER_CONTROL, POWER_CONTROL_18V |
1618 			    POWER_CONTROL_BUS_POWER);
1619 		} else if (val & OCR_29_30V) {
1620 			PUT8(ss, REG_POWER_CONTROL, POWER_CONTROL_30V);
1621 			PUT8(ss, REG_POWER_CONTROL, POWER_CONTROL_30V |
1622 			    POWER_CONTROL_BUS_POWER);
1623 		} else if (val & OCR_32_33V) {
1624 			PUT8(ss, REG_POWER_CONTROL, POWER_CONTROL_33V);
1625 			PUT8(ss, REG_POWER_CONTROL, POWER_CONTROL_33V |
1626 			    POWER_CONTROL_BUS_POWER);
1627 		} else if (val == 0) {
1628 			/* turn off power */
1629 			PUT8(ss, REG_POWER_CONTROL, 0);
1630 		} else {
1631 			rv = SDA_EINVAL;
1632 		}
1633 		break;
1634 
1635 	case SDA_PROP_HISPEED:
1636 		if (val) {
1637 			SET8(ss, REG_HOST_CONTROL, HOST_CONTROL_HIGH_SPEED_EN);
1638 		} else {
1639 			CLR8(ss, REG_HOST_CONTROL, HOST_CONTROL_HIGH_SPEED_EN);
1640 		}
1641 		/* give clocks time to settle */
1642 		drv_usecwait(10);
1643 		break;
1644 
1645 	default:
1646 		rv = SDA_ENOTSUP;
1647 		break;
1648 	}
1649 
1650 	/*
1651 	 * Apparently some controllers (ENE) have issues with changing
1652 	 * certain parameters (bus width seems to be one), requiring
1653 	 * a reset of the DAT and CMD lines.
1654 	 */
1655 	if (rv == SDA_EOK) {
1656 		(void) sdhost_soft_reset(ss, SOFT_RESET_CMD);
1657 		(void) sdhost_soft_reset(ss, SOFT_RESET_DAT);
1658 	}
1659 	mutex_exit(&ss->ss_lock);
1660 	return (rv);
1661 }
1662 
1663 sda_err_t
1664 sdhost_reset(void *arg)
1665 {
1666 	sdslot_t	*ss = arg;
1667 
1668 	mutex_enter(&ss->ss_lock);
1669 	if (!ss->ss_suspended) {
1670 		if (sdhost_soft_reset(ss, SOFT_RESET_ALL) != SDA_EOK) {
1671 			mutex_exit(&ss->ss_lock);
1672 			return (SDA_ETIME);
1673 		}
1674 		sdhost_enable_interrupts(ss);
1675 	}
1676 	mutex_exit(&ss->ss_lock);
1677 	return (SDA_EOK);
1678 }
1679 
1680 sda_err_t
1681 sdhost_halt(void *arg)
1682 {
1683 	sdslot_t	*ss = arg;
1684 
1685 	mutex_enter(&ss->ss_lock);
1686 	if (!ss->ss_suspended) {
1687 		sdhost_disable_interrupts(ss);
1688 		/* this has the side effect of removing power from the card */
1689 		if (sdhost_soft_reset(ss, SOFT_RESET_ALL) != SDA_EOK) {
1690 			mutex_exit(&ss->ss_lock);
1691 			return (SDA_ETIME);
1692 		}
1693 	}
1694 	mutex_exit(&ss->ss_lock);
1695 	return (SDA_EOK);
1696 }
1697