xref: /openbmc/linux/drivers/mmc/host/davinci_mmc.c (revision 4a9de8ad)
1 /*
2  * davinci_mmc.c - TI DaVinci MMC/SD/SDIO driver
3  *
4  * Copyright (C) 2006 Texas Instruments.
5  *       Original author: Purushotam Kumar
6  * Copyright (C) 2009 David Brownell
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12  *
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public License
19  * along with this program; if not, write to the Free Software
20  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21  */
22 
23 #include <linux/module.h>
24 #include <linux/ioport.h>
25 #include <linux/platform_device.h>
26 #include <linux/clk.h>
27 #include <linux/err.h>
28 #include <linux/cpufreq.h>
29 #include <linux/mmc/host.h>
30 #include <linux/io.h>
31 #include <linux/irq.h>
32 #include <linux/delay.h>
33 #include <linux/dma-mapping.h>
34 #include <linux/mmc/mmc.h>
35 
36 #include <mach/mmc.h>
37 #include <mach/edma.h>
38 
39 /*
40  * Register Definitions
41  */
42 #define DAVINCI_MMCCTL       0x00 /* Control Register                  */
43 #define DAVINCI_MMCCLK       0x04 /* Memory Clock Control Register     */
44 #define DAVINCI_MMCST0       0x08 /* Status Register 0                 */
45 #define DAVINCI_MMCST1       0x0C /* Status Register 1                 */
46 #define DAVINCI_MMCIM        0x10 /* Interrupt Mask Register           */
47 #define DAVINCI_MMCTOR       0x14 /* Response Time-Out Register        */
48 #define DAVINCI_MMCTOD       0x18 /* Data Read Time-Out Register       */
49 #define DAVINCI_MMCBLEN      0x1C /* Block Length Register             */
50 #define DAVINCI_MMCNBLK      0x20 /* Number of Blocks Register         */
51 #define DAVINCI_MMCNBLC      0x24 /* Number of Blocks Counter Register */
52 #define DAVINCI_MMCDRR       0x28 /* Data Receive Register             */
53 #define DAVINCI_MMCDXR       0x2C /* Data Transmit Register            */
54 #define DAVINCI_MMCCMD       0x30 /* Command Register                  */
55 #define DAVINCI_MMCARGHL     0x34 /* Argument Register                 */
56 #define DAVINCI_MMCRSP01     0x38 /* Response Register 0 and 1         */
57 #define DAVINCI_MMCRSP23     0x3C /* Response Register 0 and 1         */
58 #define DAVINCI_MMCRSP45     0x40 /* Response Register 0 and 1         */
59 #define DAVINCI_MMCRSP67     0x44 /* Response Register 0 and 1         */
60 #define DAVINCI_MMCDRSP      0x48 /* Data Response Register            */
61 #define DAVINCI_MMCETOK      0x4C
62 #define DAVINCI_MMCCIDX      0x50 /* Command Index Register            */
63 #define DAVINCI_MMCCKC       0x54
64 #define DAVINCI_MMCTORC      0x58
65 #define DAVINCI_MMCTODC      0x5C
66 #define DAVINCI_MMCBLNC      0x60
67 #define DAVINCI_SDIOCTL      0x64
68 #define DAVINCI_SDIOST0      0x68
69 #define DAVINCI_SDIOIEN      0x6C
70 #define DAVINCI_SDIOIST      0x70
71 #define DAVINCI_MMCFIFOCTL   0x74 /* FIFO Control Register             */
72 
73 /* DAVINCI_MMCCTL definitions */
74 #define MMCCTL_DATRST         (1 << 0)
75 #define MMCCTL_CMDRST         (1 << 1)
76 #define MMCCTL_WIDTH_8_BIT    (1 << 8)
77 #define MMCCTL_WIDTH_4_BIT    (1 << 2)
78 #define MMCCTL_DATEG_DISABLED (0 << 6)
79 #define MMCCTL_DATEG_RISING   (1 << 6)
80 #define MMCCTL_DATEG_FALLING  (2 << 6)
81 #define MMCCTL_DATEG_BOTH     (3 << 6)
82 #define MMCCTL_PERMDR_LE      (0 << 9)
83 #define MMCCTL_PERMDR_BE      (1 << 9)
84 #define MMCCTL_PERMDX_LE      (0 << 10)
85 #define MMCCTL_PERMDX_BE      (1 << 10)
86 
87 /* DAVINCI_MMCCLK definitions */
88 #define MMCCLK_CLKEN          (1 << 8)
89 #define MMCCLK_CLKRT_MASK     (0xFF << 0)
90 
91 /* IRQ bit definitions, for DAVINCI_MMCST0 and DAVINCI_MMCIM */
92 #define MMCST0_DATDNE         BIT(0)	/* data done */
93 #define MMCST0_BSYDNE         BIT(1)	/* busy done */
94 #define MMCST0_RSPDNE         BIT(2)	/* command done */
95 #define MMCST0_TOUTRD         BIT(3)	/* data read timeout */
96 #define MMCST0_TOUTRS         BIT(4)	/* command response timeout */
97 #define MMCST0_CRCWR          BIT(5)	/* data write CRC error */
98 #define MMCST0_CRCRD          BIT(6)	/* data read CRC error */
99 #define MMCST0_CRCRS          BIT(7)	/* command response CRC error */
100 #define MMCST0_DXRDY          BIT(9)	/* data transmit ready (fifo empty) */
101 #define MMCST0_DRRDY          BIT(10)	/* data receive ready (data in fifo)*/
102 #define MMCST0_DATED          BIT(11)	/* DAT3 edge detect */
103 #define MMCST0_TRNDNE         BIT(12)	/* transfer done */
104 
105 /* DAVINCI_MMCST1 definitions */
106 #define MMCST1_BUSY           (1 << 0)
107 
108 /* DAVINCI_MMCCMD definitions */
109 #define MMCCMD_CMD_MASK       (0x3F << 0)
110 #define MMCCMD_PPLEN          (1 << 7)
111 #define MMCCMD_BSYEXP         (1 << 8)
112 #define MMCCMD_RSPFMT_MASK    (3 << 9)
113 #define MMCCMD_RSPFMT_NONE    (0 << 9)
114 #define MMCCMD_RSPFMT_R1456   (1 << 9)
115 #define MMCCMD_RSPFMT_R2      (2 << 9)
116 #define MMCCMD_RSPFMT_R3      (3 << 9)
117 #define MMCCMD_DTRW           (1 << 11)
118 #define MMCCMD_STRMTP         (1 << 12)
119 #define MMCCMD_WDATX          (1 << 13)
120 #define MMCCMD_INITCK         (1 << 14)
121 #define MMCCMD_DCLR           (1 << 15)
122 #define MMCCMD_DMATRIG        (1 << 16)
123 
124 /* DAVINCI_MMCFIFOCTL definitions */
125 #define MMCFIFOCTL_FIFORST    (1 << 0)
126 #define MMCFIFOCTL_FIFODIR_WR (1 << 1)
127 #define MMCFIFOCTL_FIFODIR_RD (0 << 1)
128 #define MMCFIFOCTL_FIFOLEV    (1 << 2) /* 0 = 128 bits, 1 = 256 bits */
129 #define MMCFIFOCTL_ACCWD_4    (0 << 3) /* access width of 4 bytes    */
130 #define MMCFIFOCTL_ACCWD_3    (1 << 3) /* access width of 3 bytes    */
131 #define MMCFIFOCTL_ACCWD_2    (2 << 3) /* access width of 2 bytes    */
132 #define MMCFIFOCTL_ACCWD_1    (3 << 3) /* access width of 1 byte     */
133 
134 /* DAVINCI_SDIOST0 definitions */
135 #define SDIOST0_DAT1_HI       BIT(0)
136 
137 /* DAVINCI_SDIOIEN definitions */
138 #define SDIOIEN_IOINTEN       BIT(0)
139 
140 /* DAVINCI_SDIOIST definitions */
141 #define SDIOIST_IOINT         BIT(0)
142 
143 /* MMCSD Init clock in Hz in opendrain mode */
144 #define MMCSD_INIT_CLOCK		200000
145 
146 /*
147  * One scatterlist dma "segment" is at most MAX_CCNT rw_threshold units,
148  * and we handle up to MAX_NR_SG segments.  MMC_BLOCK_BOUNCE kicks in only
149  * for drivers with max_segs == 1, making the segments bigger (64KB)
150  * than the page or two that's otherwise typical. nr_sg (passed from
151  * platform data) == 16 gives at least the same throughput boost, using
152  * EDMA transfer linkage instead of spending CPU time copying pages.
153  */
154 #define MAX_CCNT	((1 << 16) - 1)
155 
156 #define MAX_NR_SG	16
157 
158 static unsigned rw_threshold = 32;
159 module_param(rw_threshold, uint, S_IRUGO);
160 MODULE_PARM_DESC(rw_threshold,
161 		"Read/Write threshold. Default = 32");
162 
163 static unsigned __initdata use_dma = 1;
164 module_param(use_dma, uint, 0);
165 MODULE_PARM_DESC(use_dma, "Whether to use DMA or not. Default = 1");
166 
167 struct mmc_davinci_host {
168 	struct mmc_command *cmd;
169 	struct mmc_data *data;
170 	struct mmc_host *mmc;
171 	struct clk *clk;
172 	unsigned int mmc_input_clk;
173 	void __iomem *base;
174 	struct resource *mem_res;
175 	int mmc_irq, sdio_irq;
176 	unsigned char bus_mode;
177 
178 #define DAVINCI_MMC_DATADIR_NONE	0
179 #define DAVINCI_MMC_DATADIR_READ	1
180 #define DAVINCI_MMC_DATADIR_WRITE	2
181 	unsigned char data_dir;
182 	unsigned char suspended;
183 
184 	/* buffer is used during PIO of one scatterlist segment, and
185 	 * is updated along with buffer_bytes_left.  bytes_left applies
186 	 * to all N blocks of the PIO transfer.
187 	 */
188 	u8 *buffer;
189 	u32 buffer_bytes_left;
190 	u32 bytes_left;
191 
192 	u32 rxdma, txdma;
193 	bool use_dma;
194 	bool do_dma;
195 	bool sdio_int;
196 
197 	/* Scatterlist DMA uses one or more parameter RAM entries:
198 	 * the main one (associated with rxdma or txdma) plus zero or
199 	 * more links.  The entries for a given transfer differ only
200 	 * by memory buffer (address, length) and link field.
201 	 */
202 	struct edmacc_param	tx_template;
203 	struct edmacc_param	rx_template;
204 	unsigned		n_link;
205 	u32			links[MAX_NR_SG - 1];
206 
207 	/* For PIO we walk scatterlists one segment at a time. */
208 	unsigned int		sg_len;
209 	struct scatterlist *sg;
210 
211 	/* Version of the MMC/SD controller */
212 	u8 version;
213 	/* for ns in one cycle calculation */
214 	unsigned ns_in_one_cycle;
215 	/* Number of sg segments */
216 	u8 nr_sg;
217 #ifdef CONFIG_CPU_FREQ
218 	struct notifier_block	freq_transition;
219 #endif
220 };
221 
222 
223 /* PIO only */
224 static void mmc_davinci_sg_to_buf(struct mmc_davinci_host *host)
225 {
226 	host->buffer_bytes_left = sg_dma_len(host->sg);
227 	host->buffer = sg_virt(host->sg);
228 	if (host->buffer_bytes_left > host->bytes_left)
229 		host->buffer_bytes_left = host->bytes_left;
230 }
231 
232 static void davinci_fifo_data_trans(struct mmc_davinci_host *host,
233 					unsigned int n)
234 {
235 	u8 *p;
236 	unsigned int i;
237 
238 	if (host->buffer_bytes_left == 0) {
239 		host->sg = sg_next(host->data->sg);
240 		mmc_davinci_sg_to_buf(host);
241 	}
242 
243 	p = host->buffer;
244 	if (n > host->buffer_bytes_left)
245 		n = host->buffer_bytes_left;
246 	host->buffer_bytes_left -= n;
247 	host->bytes_left -= n;
248 
249 	/* NOTE:  we never transfer more than rw_threshold bytes
250 	 * to/from the fifo here; there's no I/O overlap.
251 	 * This also assumes that access width( i.e. ACCWD) is 4 bytes
252 	 */
253 	if (host->data_dir == DAVINCI_MMC_DATADIR_WRITE) {
254 		for (i = 0; i < (n >> 2); i++) {
255 			writel(*((u32 *)p), host->base + DAVINCI_MMCDXR);
256 			p = p + 4;
257 		}
258 		if (n & 3) {
259 			iowrite8_rep(host->base + DAVINCI_MMCDXR, p, (n & 3));
260 			p = p + (n & 3);
261 		}
262 	} else {
263 		for (i = 0; i < (n >> 2); i++) {
264 			*((u32 *)p) = readl(host->base + DAVINCI_MMCDRR);
265 			p  = p + 4;
266 		}
267 		if (n & 3) {
268 			ioread8_rep(host->base + DAVINCI_MMCDRR, p, (n & 3));
269 			p = p + (n & 3);
270 		}
271 	}
272 	host->buffer = p;
273 }
274 
275 static void mmc_davinci_start_command(struct mmc_davinci_host *host,
276 		struct mmc_command *cmd)
277 {
278 	u32 cmd_reg = 0;
279 	u32 im_val;
280 
281 	dev_dbg(mmc_dev(host->mmc), "CMD%d, arg 0x%08x%s\n",
282 		cmd->opcode, cmd->arg,
283 		({ char *s;
284 		switch (mmc_resp_type(cmd)) {
285 		case MMC_RSP_R1:
286 			s = ", R1/R5/R6/R7 response";
287 			break;
288 		case MMC_RSP_R1B:
289 			s = ", R1b response";
290 			break;
291 		case MMC_RSP_R2:
292 			s = ", R2 response";
293 			break;
294 		case MMC_RSP_R3:
295 			s = ", R3/R4 response";
296 			break;
297 		default:
298 			s = ", (R? response)";
299 			break;
300 		}; s; }));
301 	host->cmd = cmd;
302 
303 	switch (mmc_resp_type(cmd)) {
304 	case MMC_RSP_R1B:
305 		/* There's some spec confusion about when R1B is
306 		 * allowed, but if the card doesn't issue a BUSY
307 		 * then it's harmless for us to allow it.
308 		 */
309 		cmd_reg |= MMCCMD_BSYEXP;
310 		/* FALLTHROUGH */
311 	case MMC_RSP_R1:		/* 48 bits, CRC */
312 		cmd_reg |= MMCCMD_RSPFMT_R1456;
313 		break;
314 	case MMC_RSP_R2:		/* 136 bits, CRC */
315 		cmd_reg |= MMCCMD_RSPFMT_R2;
316 		break;
317 	case MMC_RSP_R3:		/* 48 bits, no CRC */
318 		cmd_reg |= MMCCMD_RSPFMT_R3;
319 		break;
320 	default:
321 		cmd_reg |= MMCCMD_RSPFMT_NONE;
322 		dev_dbg(mmc_dev(host->mmc), "unknown resp_type %04x\n",
323 			mmc_resp_type(cmd));
324 		break;
325 	}
326 
327 	/* Set command index */
328 	cmd_reg |= cmd->opcode;
329 
330 	/* Enable EDMA transfer triggers */
331 	if (host->do_dma)
332 		cmd_reg |= MMCCMD_DMATRIG;
333 
334 	if (host->version == MMC_CTLR_VERSION_2 && host->data != NULL &&
335 			host->data_dir == DAVINCI_MMC_DATADIR_READ)
336 		cmd_reg |= MMCCMD_DMATRIG;
337 
338 	/* Setting whether command involves data transfer or not */
339 	if (cmd->data)
340 		cmd_reg |= MMCCMD_WDATX;
341 
342 	/* Setting whether stream or block transfer */
343 	if (cmd->flags & MMC_DATA_STREAM)
344 		cmd_reg |= MMCCMD_STRMTP;
345 
346 	/* Setting whether data read or write */
347 	if (host->data_dir == DAVINCI_MMC_DATADIR_WRITE)
348 		cmd_reg |= MMCCMD_DTRW;
349 
350 	if (host->bus_mode == MMC_BUSMODE_PUSHPULL)
351 		cmd_reg |= MMCCMD_PPLEN;
352 
353 	/* set Command timeout */
354 	writel(0x1FFF, host->base + DAVINCI_MMCTOR);
355 
356 	/* Enable interrupt (calculate here, defer until FIFO is stuffed). */
357 	im_val =  MMCST0_RSPDNE | MMCST0_CRCRS | MMCST0_TOUTRS;
358 	if (host->data_dir == DAVINCI_MMC_DATADIR_WRITE) {
359 		im_val |= MMCST0_DATDNE | MMCST0_CRCWR;
360 
361 		if (!host->do_dma)
362 			im_val |= MMCST0_DXRDY;
363 	} else if (host->data_dir == DAVINCI_MMC_DATADIR_READ) {
364 		im_val |= MMCST0_DATDNE | MMCST0_CRCRD | MMCST0_TOUTRD;
365 
366 		if (!host->do_dma)
367 			im_val |= MMCST0_DRRDY;
368 	}
369 
370 	/*
371 	 * Before non-DMA WRITE commands the controller needs priming:
372 	 * FIFO should be populated with 32 bytes i.e. whatever is the FIFO size
373 	 */
374 	if (!host->do_dma && (host->data_dir == DAVINCI_MMC_DATADIR_WRITE))
375 		davinci_fifo_data_trans(host, rw_threshold);
376 
377 	writel(cmd->arg, host->base + DAVINCI_MMCARGHL);
378 	writel(cmd_reg,  host->base + DAVINCI_MMCCMD);
379 	writel(im_val, host->base + DAVINCI_MMCIM);
380 }
381 
382 /*----------------------------------------------------------------------*/
383 
384 /* DMA infrastructure */
385 
386 static void davinci_abort_dma(struct mmc_davinci_host *host)
387 {
388 	int sync_dev;
389 
390 	if (host->data_dir == DAVINCI_MMC_DATADIR_READ)
391 		sync_dev = host->rxdma;
392 	else
393 		sync_dev = host->txdma;
394 
395 	edma_stop(sync_dev);
396 	edma_clean_channel(sync_dev);
397 }
398 
399 static void
400 mmc_davinci_xfer_done(struct mmc_davinci_host *host, struct mmc_data *data);
401 
402 static void mmc_davinci_dma_cb(unsigned channel, u16 ch_status, void *data)
403 {
404 	if (DMA_COMPLETE != ch_status) {
405 		struct mmc_davinci_host *host = data;
406 
407 		/* Currently means:  DMA Event Missed, or "null" transfer
408 		 * request was seen.  In the future, TC errors (like bad
409 		 * addresses) might be presented too.
410 		 */
411 		dev_warn(mmc_dev(host->mmc), "DMA %s error\n",
412 			(host->data->flags & MMC_DATA_WRITE)
413 				? "write" : "read");
414 		host->data->error = -EIO;
415 		mmc_davinci_xfer_done(host, host->data);
416 	}
417 }
418 
419 /* Set up tx or rx template, to be modified and updated later */
420 static void __init mmc_davinci_dma_setup(struct mmc_davinci_host *host,
421 		bool tx, struct edmacc_param *template)
422 {
423 	unsigned	sync_dev;
424 	const u16	acnt = 4;
425 	const u16	bcnt = rw_threshold >> 2;
426 	const u16	ccnt = 0;
427 	u32		src_port = 0;
428 	u32		dst_port = 0;
429 	s16		src_bidx, dst_bidx;
430 	s16		src_cidx, dst_cidx;
431 
432 	/*
433 	 * A-B Sync transfer:  each DMA request is for one "frame" of
434 	 * rw_threshold bytes, broken into "acnt"-size chunks repeated
435 	 * "bcnt" times.  Each segment needs "ccnt" such frames; since
436 	 * we tell the block layer our mmc->max_seg_size limit, we can
437 	 * trust (later) that it's within bounds.
438 	 *
439 	 * The FIFOs are read/written in 4-byte chunks (acnt == 4) and
440 	 * EDMA will optimize memory operations to use larger bursts.
441 	 */
442 	if (tx) {
443 		sync_dev = host->txdma;
444 
445 		/* src_prt, ccnt, and link to be set up later */
446 		src_bidx = acnt;
447 		src_cidx = acnt * bcnt;
448 
449 		dst_port = host->mem_res->start + DAVINCI_MMCDXR;
450 		dst_bidx = 0;
451 		dst_cidx = 0;
452 	} else {
453 		sync_dev = host->rxdma;
454 
455 		src_port = host->mem_res->start + DAVINCI_MMCDRR;
456 		src_bidx = 0;
457 		src_cidx = 0;
458 
459 		/* dst_prt, ccnt, and link to be set up later */
460 		dst_bidx = acnt;
461 		dst_cidx = acnt * bcnt;
462 	}
463 
464 	/*
465 	 * We can't use FIFO mode for the FIFOs because MMC FIFO addresses
466 	 * are not 256-bit (32-byte) aligned.  So we use INCR, and the W8BIT
467 	 * parameter is ignored.
468 	 */
469 	edma_set_src(sync_dev, src_port, INCR, W8BIT);
470 	edma_set_dest(sync_dev, dst_port, INCR, W8BIT);
471 
472 	edma_set_src_index(sync_dev, src_bidx, src_cidx);
473 	edma_set_dest_index(sync_dev, dst_bidx, dst_cidx);
474 
475 	edma_set_transfer_params(sync_dev, acnt, bcnt, ccnt, 8, ABSYNC);
476 
477 	edma_read_slot(sync_dev, template);
478 
479 	/* don't bother with irqs or chaining */
480 	template->opt |= EDMA_CHAN_SLOT(sync_dev) << 12;
481 }
482 
483 static void mmc_davinci_send_dma_request(struct mmc_davinci_host *host,
484 		struct mmc_data *data)
485 {
486 	struct edmacc_param	*template;
487 	int			channel, slot;
488 	unsigned		link;
489 	struct scatterlist	*sg;
490 	unsigned		sg_len;
491 	unsigned		bytes_left = host->bytes_left;
492 	const unsigned		shift = ffs(rw_threshold) - 1;
493 
494 	if (host->data_dir == DAVINCI_MMC_DATADIR_WRITE) {
495 		template = &host->tx_template;
496 		channel = host->txdma;
497 	} else {
498 		template = &host->rx_template;
499 		channel = host->rxdma;
500 	}
501 
502 	/* We know sg_len and ccnt will never be out of range because
503 	 * we told the mmc layer which in turn tells the block layer
504 	 * to ensure that it only hands us one scatterlist segment
505 	 * per EDMA PARAM entry.  Update the PARAM
506 	 * entries needed for each segment of this scatterlist.
507 	 */
508 	for (slot = channel, link = 0, sg = data->sg, sg_len = host->sg_len;
509 			sg_len-- != 0 && bytes_left;
510 			sg = sg_next(sg), slot = host->links[link++]) {
511 		u32		buf = sg_dma_address(sg);
512 		unsigned	count = sg_dma_len(sg);
513 
514 		template->link_bcntrld = sg_len
515 				? (EDMA_CHAN_SLOT(host->links[link]) << 5)
516 				: 0xffff;
517 
518 		if (count > bytes_left)
519 			count = bytes_left;
520 		bytes_left -= count;
521 
522 		if (host->data_dir == DAVINCI_MMC_DATADIR_WRITE)
523 			template->src = buf;
524 		else
525 			template->dst = buf;
526 		template->ccnt = count >> shift;
527 
528 		edma_write_slot(slot, template);
529 	}
530 
531 	if (host->version == MMC_CTLR_VERSION_2)
532 		edma_clear_event(channel);
533 
534 	edma_start(channel);
535 }
536 
537 static int mmc_davinci_start_dma_transfer(struct mmc_davinci_host *host,
538 		struct mmc_data *data)
539 {
540 	int i;
541 	int mask = rw_threshold - 1;
542 
543 	host->sg_len = dma_map_sg(mmc_dev(host->mmc), data->sg, data->sg_len,
544 				((data->flags & MMC_DATA_WRITE)
545 				? DMA_TO_DEVICE
546 				: DMA_FROM_DEVICE));
547 
548 	/* no individual DMA segment should need a partial FIFO */
549 	for (i = 0; i < host->sg_len; i++) {
550 		if (sg_dma_len(data->sg + i) & mask) {
551 			dma_unmap_sg(mmc_dev(host->mmc),
552 					data->sg, data->sg_len,
553 					(data->flags & MMC_DATA_WRITE)
554 					? DMA_TO_DEVICE
555 					: DMA_FROM_DEVICE);
556 			return -1;
557 		}
558 	}
559 
560 	host->do_dma = 1;
561 	mmc_davinci_send_dma_request(host, data);
562 
563 	return 0;
564 }
565 
566 static void __init_or_module
567 davinci_release_dma_channels(struct mmc_davinci_host *host)
568 {
569 	unsigned	i;
570 
571 	if (!host->use_dma)
572 		return;
573 
574 	for (i = 0; i < host->n_link; i++)
575 		edma_free_slot(host->links[i]);
576 
577 	edma_free_channel(host->txdma);
578 	edma_free_channel(host->rxdma);
579 }
580 
581 static int __init davinci_acquire_dma_channels(struct mmc_davinci_host *host)
582 {
583 	u32 link_size;
584 	int r, i;
585 
586 	/* Acquire master DMA write channel */
587 	r = edma_alloc_channel(host->txdma, mmc_davinci_dma_cb, host,
588 			EVENTQ_DEFAULT);
589 	if (r < 0) {
590 		dev_warn(mmc_dev(host->mmc), "alloc %s channel err %d\n",
591 				"tx", r);
592 		return r;
593 	}
594 	mmc_davinci_dma_setup(host, true, &host->tx_template);
595 
596 	/* Acquire master DMA read channel */
597 	r = edma_alloc_channel(host->rxdma, mmc_davinci_dma_cb, host,
598 			EVENTQ_DEFAULT);
599 	if (r < 0) {
600 		dev_warn(mmc_dev(host->mmc), "alloc %s channel err %d\n",
601 				"rx", r);
602 		goto free_master_write;
603 	}
604 	mmc_davinci_dma_setup(host, false, &host->rx_template);
605 
606 	/* Allocate parameter RAM slots, which will later be bound to a
607 	 * channel as needed to handle a scatterlist.
608 	 */
609 	link_size = min_t(unsigned, host->nr_sg, ARRAY_SIZE(host->links));
610 	for (i = 0; i < link_size; i++) {
611 		r = edma_alloc_slot(EDMA_CTLR(host->txdma), EDMA_SLOT_ANY);
612 		if (r < 0) {
613 			dev_dbg(mmc_dev(host->mmc), "dma PaRAM alloc --> %d\n",
614 				r);
615 			break;
616 		}
617 		host->links[i] = r;
618 	}
619 	host->n_link = i;
620 
621 	return 0;
622 
623 free_master_write:
624 	edma_free_channel(host->txdma);
625 
626 	return r;
627 }
628 
629 /*----------------------------------------------------------------------*/
630 
631 static void
632 mmc_davinci_prepare_data(struct mmc_davinci_host *host, struct mmc_request *req)
633 {
634 	int fifo_lev = (rw_threshold == 32) ? MMCFIFOCTL_FIFOLEV : 0;
635 	int timeout;
636 	struct mmc_data *data = req->data;
637 
638 	if (host->version == MMC_CTLR_VERSION_2)
639 		fifo_lev = (rw_threshold == 64) ? MMCFIFOCTL_FIFOLEV : 0;
640 
641 	host->data = data;
642 	if (data == NULL) {
643 		host->data_dir = DAVINCI_MMC_DATADIR_NONE;
644 		writel(0, host->base + DAVINCI_MMCBLEN);
645 		writel(0, host->base + DAVINCI_MMCNBLK);
646 		return;
647 	}
648 
649 	dev_dbg(mmc_dev(host->mmc), "%s %s, %d blocks of %d bytes\n",
650 		(data->flags & MMC_DATA_STREAM) ? "stream" : "block",
651 		(data->flags & MMC_DATA_WRITE) ? "write" : "read",
652 		data->blocks, data->blksz);
653 	dev_dbg(mmc_dev(host->mmc), "  DTO %d cycles + %d ns\n",
654 		data->timeout_clks, data->timeout_ns);
655 	timeout = data->timeout_clks +
656 		(data->timeout_ns / host->ns_in_one_cycle);
657 	if (timeout > 0xffff)
658 		timeout = 0xffff;
659 
660 	writel(timeout, host->base + DAVINCI_MMCTOD);
661 	writel(data->blocks, host->base + DAVINCI_MMCNBLK);
662 	writel(data->blksz, host->base + DAVINCI_MMCBLEN);
663 
664 	/* Configure the FIFO */
665 	switch (data->flags & MMC_DATA_WRITE) {
666 	case MMC_DATA_WRITE:
667 		host->data_dir = DAVINCI_MMC_DATADIR_WRITE;
668 		writel(fifo_lev | MMCFIFOCTL_FIFODIR_WR | MMCFIFOCTL_FIFORST,
669 			host->base + DAVINCI_MMCFIFOCTL);
670 		writel(fifo_lev | MMCFIFOCTL_FIFODIR_WR,
671 			host->base + DAVINCI_MMCFIFOCTL);
672 		break;
673 
674 	default:
675 		host->data_dir = DAVINCI_MMC_DATADIR_READ;
676 		writel(fifo_lev | MMCFIFOCTL_FIFODIR_RD | MMCFIFOCTL_FIFORST,
677 			host->base + DAVINCI_MMCFIFOCTL);
678 		writel(fifo_lev | MMCFIFOCTL_FIFODIR_RD,
679 			host->base + DAVINCI_MMCFIFOCTL);
680 		break;
681 	}
682 
683 	host->buffer = NULL;
684 	host->bytes_left = data->blocks * data->blksz;
685 
686 	/* For now we try to use DMA whenever we won't need partial FIFO
687 	 * reads or writes, either for the whole transfer (as tested here)
688 	 * or for any individual scatterlist segment (tested when we call
689 	 * start_dma_transfer).
690 	 *
691 	 * While we *could* change that, unusual block sizes are rarely
692 	 * used.  The occasional fallback to PIO should't hurt.
693 	 */
694 	if (host->use_dma && (host->bytes_left & (rw_threshold - 1)) == 0
695 			&& mmc_davinci_start_dma_transfer(host, data) == 0) {
696 		/* zero this to ensure we take no PIO paths */
697 		host->bytes_left = 0;
698 	} else {
699 		/* Revert to CPU Copy */
700 		host->sg_len = data->sg_len;
701 		host->sg = host->data->sg;
702 		mmc_davinci_sg_to_buf(host);
703 	}
704 }
705 
706 static void mmc_davinci_request(struct mmc_host *mmc, struct mmc_request *req)
707 {
708 	struct mmc_davinci_host *host = mmc_priv(mmc);
709 	unsigned long timeout = jiffies + msecs_to_jiffies(900);
710 	u32 mmcst1 = 0;
711 
712 	/* Card may still be sending BUSY after a previous operation,
713 	 * typically some kind of write.  If so, we can't proceed yet.
714 	 */
715 	while (time_before(jiffies, timeout)) {
716 		mmcst1  = readl(host->base + DAVINCI_MMCST1);
717 		if (!(mmcst1 & MMCST1_BUSY))
718 			break;
719 		cpu_relax();
720 	}
721 	if (mmcst1 & MMCST1_BUSY) {
722 		dev_err(mmc_dev(host->mmc), "still BUSY? bad ... \n");
723 		req->cmd->error = -ETIMEDOUT;
724 		mmc_request_done(mmc, req);
725 		return;
726 	}
727 
728 	host->do_dma = 0;
729 	mmc_davinci_prepare_data(host, req);
730 	mmc_davinci_start_command(host, req->cmd);
731 }
732 
733 static unsigned int calculate_freq_for_card(struct mmc_davinci_host *host,
734 	unsigned int mmc_req_freq)
735 {
736 	unsigned int mmc_freq = 0, mmc_pclk = 0, mmc_push_pull_divisor = 0;
737 
738 	mmc_pclk = host->mmc_input_clk;
739 	if (mmc_req_freq && mmc_pclk > (2 * mmc_req_freq))
740 		mmc_push_pull_divisor = ((unsigned int)mmc_pclk
741 				/ (2 * mmc_req_freq)) - 1;
742 	else
743 		mmc_push_pull_divisor = 0;
744 
745 	mmc_freq = (unsigned int)mmc_pclk
746 		/ (2 * (mmc_push_pull_divisor + 1));
747 
748 	if (mmc_freq > mmc_req_freq)
749 		mmc_push_pull_divisor = mmc_push_pull_divisor + 1;
750 	/* Convert ns to clock cycles */
751 	if (mmc_req_freq <= 400000)
752 		host->ns_in_one_cycle = (1000000) / (((mmc_pclk
753 				/ (2 * (mmc_push_pull_divisor + 1)))/1000));
754 	else
755 		host->ns_in_one_cycle = (1000000) / (((mmc_pclk
756 				/ (2 * (mmc_push_pull_divisor + 1)))/1000000));
757 
758 	return mmc_push_pull_divisor;
759 }
760 
761 static void calculate_clk_divider(struct mmc_host *mmc, struct mmc_ios *ios)
762 {
763 	unsigned int open_drain_freq = 0, mmc_pclk = 0;
764 	unsigned int mmc_push_pull_freq = 0;
765 	struct mmc_davinci_host *host = mmc_priv(mmc);
766 
767 	if (ios->bus_mode == MMC_BUSMODE_OPENDRAIN) {
768 		u32 temp;
769 
770 		/* Ignoring the init clock value passed for fixing the inter
771 		 * operability with different cards.
772 		 */
773 		open_drain_freq = ((unsigned int)mmc_pclk
774 				/ (2 * MMCSD_INIT_CLOCK)) - 1;
775 
776 		if (open_drain_freq > 0xFF)
777 			open_drain_freq = 0xFF;
778 
779 		temp = readl(host->base + DAVINCI_MMCCLK) & ~MMCCLK_CLKRT_MASK;
780 		temp |= open_drain_freq;
781 		writel(temp, host->base + DAVINCI_MMCCLK);
782 
783 		/* Convert ns to clock cycles */
784 		host->ns_in_one_cycle = (1000000) / (MMCSD_INIT_CLOCK/1000);
785 	} else {
786 		u32 temp;
787 		mmc_push_pull_freq = calculate_freq_for_card(host, ios->clock);
788 
789 		if (mmc_push_pull_freq > 0xFF)
790 			mmc_push_pull_freq = 0xFF;
791 
792 		temp = readl(host->base + DAVINCI_MMCCLK) & ~MMCCLK_CLKEN;
793 		writel(temp, host->base + DAVINCI_MMCCLK);
794 
795 		udelay(10);
796 
797 		temp = readl(host->base + DAVINCI_MMCCLK) & ~MMCCLK_CLKRT_MASK;
798 		temp |= mmc_push_pull_freq;
799 		writel(temp, host->base + DAVINCI_MMCCLK);
800 
801 		writel(temp | MMCCLK_CLKEN, host->base + DAVINCI_MMCCLK);
802 
803 		udelay(10);
804 	}
805 }
806 
807 static void mmc_davinci_set_ios(struct mmc_host *mmc, struct mmc_ios *ios)
808 {
809 	struct mmc_davinci_host *host = mmc_priv(mmc);
810 	struct platform_device *pdev = to_platform_device(mmc->parent);
811 	struct davinci_mmc_config *config = pdev->dev.platform_data;
812 
813 	dev_dbg(mmc_dev(host->mmc),
814 		"clock %dHz busmode %d powermode %d Vdd %04x\n",
815 		ios->clock, ios->bus_mode, ios->power_mode,
816 		ios->vdd);
817 
818 	switch (ios->power_mode) {
819 	case MMC_POWER_OFF:
820 		if (config && config->set_power)
821 			config->set_power(pdev->id, false);
822 		break;
823 	case MMC_POWER_UP:
824 		if (config && config->set_power)
825 			config->set_power(pdev->id, true);
826 		break;
827 	}
828 
829 	switch (ios->bus_width) {
830 	case MMC_BUS_WIDTH_8:
831 		dev_dbg(mmc_dev(host->mmc), "Enabling 8 bit mode\n");
832 		writel((readl(host->base + DAVINCI_MMCCTL) &
833 			~MMCCTL_WIDTH_4_BIT) | MMCCTL_WIDTH_8_BIT,
834 			host->base + DAVINCI_MMCCTL);
835 		break;
836 	case MMC_BUS_WIDTH_4:
837 		dev_dbg(mmc_dev(host->mmc), "Enabling 4 bit mode\n");
838 		if (host->version == MMC_CTLR_VERSION_2)
839 			writel((readl(host->base + DAVINCI_MMCCTL) &
840 				~MMCCTL_WIDTH_8_BIT) | MMCCTL_WIDTH_4_BIT,
841 				host->base + DAVINCI_MMCCTL);
842 		else
843 			writel(readl(host->base + DAVINCI_MMCCTL) |
844 				MMCCTL_WIDTH_4_BIT,
845 				host->base + DAVINCI_MMCCTL);
846 		break;
847 	case MMC_BUS_WIDTH_1:
848 		dev_dbg(mmc_dev(host->mmc), "Enabling 1 bit mode\n");
849 		if (host->version == MMC_CTLR_VERSION_2)
850 			writel(readl(host->base + DAVINCI_MMCCTL) &
851 				~(MMCCTL_WIDTH_8_BIT | MMCCTL_WIDTH_4_BIT),
852 				host->base + DAVINCI_MMCCTL);
853 		else
854 			writel(readl(host->base + DAVINCI_MMCCTL) &
855 				~MMCCTL_WIDTH_4_BIT,
856 				host->base + DAVINCI_MMCCTL);
857 		break;
858 	}
859 
860 	calculate_clk_divider(mmc, ios);
861 
862 	host->bus_mode = ios->bus_mode;
863 	if (ios->power_mode == MMC_POWER_UP) {
864 		unsigned long timeout = jiffies + msecs_to_jiffies(50);
865 		bool lose = true;
866 
867 		/* Send clock cycles, poll completion */
868 		writel(0, host->base + DAVINCI_MMCARGHL);
869 		writel(MMCCMD_INITCK, host->base + DAVINCI_MMCCMD);
870 		while (time_before(jiffies, timeout)) {
871 			u32 tmp = readl(host->base + DAVINCI_MMCST0);
872 
873 			if (tmp & MMCST0_RSPDNE) {
874 				lose = false;
875 				break;
876 			}
877 			cpu_relax();
878 		}
879 		if (lose)
880 			dev_warn(mmc_dev(host->mmc), "powerup timeout\n");
881 	}
882 
883 	/* FIXME on power OFF, reset things ... */
884 }
885 
886 static void
887 mmc_davinci_xfer_done(struct mmc_davinci_host *host, struct mmc_data *data)
888 {
889 	host->data = NULL;
890 
891 	if (host->mmc->caps & MMC_CAP_SDIO_IRQ) {
892 		/*
893 		 * SDIO Interrupt Detection work-around as suggested by
894 		 * Davinci Errata (TMS320DM355 Silicon Revision 1.1 Errata
895 		 * 2.1.6): Signal SDIO interrupt only if it is enabled by core
896 		 */
897 		if (host->sdio_int && !(readl(host->base + DAVINCI_SDIOST0) &
898 					SDIOST0_DAT1_HI)) {
899 			writel(SDIOIST_IOINT, host->base + DAVINCI_SDIOIST);
900 			mmc_signal_sdio_irq(host->mmc);
901 		}
902 	}
903 
904 	if (host->do_dma) {
905 		davinci_abort_dma(host);
906 
907 		dma_unmap_sg(mmc_dev(host->mmc), data->sg, data->sg_len,
908 			     (data->flags & MMC_DATA_WRITE)
909 			     ? DMA_TO_DEVICE
910 			     : DMA_FROM_DEVICE);
911 		host->do_dma = false;
912 	}
913 	host->data_dir = DAVINCI_MMC_DATADIR_NONE;
914 
915 	if (!data->stop || (host->cmd && host->cmd->error)) {
916 		mmc_request_done(host->mmc, data->mrq);
917 		writel(0, host->base + DAVINCI_MMCIM);
918 	} else
919 		mmc_davinci_start_command(host, data->stop);
920 }
921 
922 static void mmc_davinci_cmd_done(struct mmc_davinci_host *host,
923 				 struct mmc_command *cmd)
924 {
925 	host->cmd = NULL;
926 
927 	if (cmd->flags & MMC_RSP_PRESENT) {
928 		if (cmd->flags & MMC_RSP_136) {
929 			/* response type 2 */
930 			cmd->resp[3] = readl(host->base + DAVINCI_MMCRSP01);
931 			cmd->resp[2] = readl(host->base + DAVINCI_MMCRSP23);
932 			cmd->resp[1] = readl(host->base + DAVINCI_MMCRSP45);
933 			cmd->resp[0] = readl(host->base + DAVINCI_MMCRSP67);
934 		} else {
935 			/* response types 1, 1b, 3, 4, 5, 6 */
936 			cmd->resp[0] = readl(host->base + DAVINCI_MMCRSP67);
937 		}
938 	}
939 
940 	if (host->data == NULL || cmd->error) {
941 		if (cmd->error == -ETIMEDOUT)
942 			cmd->mrq->cmd->retries = 0;
943 		mmc_request_done(host->mmc, cmd->mrq);
944 		writel(0, host->base + DAVINCI_MMCIM);
945 	}
946 }
947 
948 static inline void mmc_davinci_reset_ctrl(struct mmc_davinci_host *host,
949 								int val)
950 {
951 	u32 temp;
952 
953 	temp = readl(host->base + DAVINCI_MMCCTL);
954 	if (val)	/* reset */
955 		temp |= MMCCTL_CMDRST | MMCCTL_DATRST;
956 	else		/* enable */
957 		temp &= ~(MMCCTL_CMDRST | MMCCTL_DATRST);
958 
959 	writel(temp, host->base + DAVINCI_MMCCTL);
960 	udelay(10);
961 }
962 
963 static void
964 davinci_abort_data(struct mmc_davinci_host *host, struct mmc_data *data)
965 {
966 	mmc_davinci_reset_ctrl(host, 1);
967 	mmc_davinci_reset_ctrl(host, 0);
968 }
969 
970 static irqreturn_t mmc_davinci_sdio_irq(int irq, void *dev_id)
971 {
972 	struct mmc_davinci_host *host = dev_id;
973 	unsigned int status;
974 
975 	status = readl(host->base + DAVINCI_SDIOIST);
976 	if (status & SDIOIST_IOINT) {
977 		dev_dbg(mmc_dev(host->mmc),
978 			"SDIO interrupt status %x\n", status);
979 		writel(status | SDIOIST_IOINT, host->base + DAVINCI_SDIOIST);
980 		mmc_signal_sdio_irq(host->mmc);
981 	}
982 	return IRQ_HANDLED;
983 }
984 
985 static irqreturn_t mmc_davinci_irq(int irq, void *dev_id)
986 {
987 	struct mmc_davinci_host *host = (struct mmc_davinci_host *)dev_id;
988 	unsigned int status, qstatus;
989 	int end_command = 0;
990 	int end_transfer = 0;
991 	struct mmc_data *data = host->data;
992 
993 	if (host->cmd == NULL && host->data == NULL) {
994 		status = readl(host->base + DAVINCI_MMCST0);
995 		dev_dbg(mmc_dev(host->mmc),
996 			"Spurious interrupt 0x%04x\n", status);
997 		/* Disable the interrupt from mmcsd */
998 		writel(0, host->base + DAVINCI_MMCIM);
999 		return IRQ_NONE;
1000 	}
1001 
1002 	status = readl(host->base + DAVINCI_MMCST0);
1003 	qstatus = status;
1004 
1005 	/* handle FIFO first when using PIO for data.
1006 	 * bytes_left will decrease to zero as I/O progress and status will
1007 	 * read zero over iteration because this controller status
1008 	 * register(MMCST0) reports any status only once and it is cleared
1009 	 * by read. So, it is not unbouned loop even in the case of
1010 	 * non-dma.
1011 	 */
1012 	while (host->bytes_left && (status & (MMCST0_DXRDY | MMCST0_DRRDY))) {
1013 		davinci_fifo_data_trans(host, rw_threshold);
1014 		status = readl(host->base + DAVINCI_MMCST0);
1015 		if (!status)
1016 			break;
1017 		qstatus |= status;
1018 	}
1019 
1020 	if (qstatus & MMCST0_DATDNE) {
1021 		/* All blocks sent/received, and CRC checks passed */
1022 		if (data != NULL) {
1023 			if ((host->do_dma == 0) && (host->bytes_left > 0)) {
1024 				/* if datasize < rw_threshold
1025 				 * no RX ints are generated
1026 				 */
1027 				davinci_fifo_data_trans(host, host->bytes_left);
1028 			}
1029 			end_transfer = 1;
1030 			data->bytes_xfered = data->blocks * data->blksz;
1031 		} else {
1032 			dev_err(mmc_dev(host->mmc),
1033 					"DATDNE with no host->data\n");
1034 		}
1035 	}
1036 
1037 	if (qstatus & MMCST0_TOUTRD) {
1038 		/* Read data timeout */
1039 		data->error = -ETIMEDOUT;
1040 		end_transfer = 1;
1041 
1042 		dev_dbg(mmc_dev(host->mmc),
1043 			"read data timeout, status %x\n",
1044 			qstatus);
1045 
1046 		davinci_abort_data(host, data);
1047 	}
1048 
1049 	if (qstatus & (MMCST0_CRCWR | MMCST0_CRCRD)) {
1050 		/* Data CRC error */
1051 		data->error = -EILSEQ;
1052 		end_transfer = 1;
1053 
1054 		/* NOTE:  this controller uses CRCWR to report both CRC
1055 		 * errors and timeouts (on writes).  MMCDRSP values are
1056 		 * only weakly documented, but 0x9f was clearly a timeout
1057 		 * case and the two three-bit patterns in various SD specs
1058 		 * (101, 010) aren't part of it ...
1059 		 */
1060 		if (qstatus & MMCST0_CRCWR) {
1061 			u32 temp = readb(host->base + DAVINCI_MMCDRSP);
1062 
1063 			if (temp == 0x9f)
1064 				data->error = -ETIMEDOUT;
1065 		}
1066 		dev_dbg(mmc_dev(host->mmc), "data %s %s error\n",
1067 			(qstatus & MMCST0_CRCWR) ? "write" : "read",
1068 			(data->error == -ETIMEDOUT) ? "timeout" : "CRC");
1069 
1070 		davinci_abort_data(host, data);
1071 	}
1072 
1073 	if (qstatus & MMCST0_TOUTRS) {
1074 		/* Command timeout */
1075 		if (host->cmd) {
1076 			dev_dbg(mmc_dev(host->mmc),
1077 				"CMD%d timeout, status %x\n",
1078 				host->cmd->opcode, qstatus);
1079 			host->cmd->error = -ETIMEDOUT;
1080 			if (data) {
1081 				end_transfer = 1;
1082 				davinci_abort_data(host, data);
1083 			} else
1084 				end_command = 1;
1085 		}
1086 	}
1087 
1088 	if (qstatus & MMCST0_CRCRS) {
1089 		/* Command CRC error */
1090 		dev_dbg(mmc_dev(host->mmc), "Command CRC error\n");
1091 		if (host->cmd) {
1092 			host->cmd->error = -EILSEQ;
1093 			end_command = 1;
1094 		}
1095 	}
1096 
1097 	if (qstatus & MMCST0_RSPDNE) {
1098 		/* End of command phase */
1099 		end_command = (int) host->cmd;
1100 	}
1101 
1102 	if (end_command)
1103 		mmc_davinci_cmd_done(host, host->cmd);
1104 	if (end_transfer)
1105 		mmc_davinci_xfer_done(host, data);
1106 	return IRQ_HANDLED;
1107 }
1108 
1109 static int mmc_davinci_get_cd(struct mmc_host *mmc)
1110 {
1111 	struct platform_device *pdev = to_platform_device(mmc->parent);
1112 	struct davinci_mmc_config *config = pdev->dev.platform_data;
1113 
1114 	if (!config || !config->get_cd)
1115 		return -ENOSYS;
1116 	return config->get_cd(pdev->id);
1117 }
1118 
1119 static int mmc_davinci_get_ro(struct mmc_host *mmc)
1120 {
1121 	struct platform_device *pdev = to_platform_device(mmc->parent);
1122 	struct davinci_mmc_config *config = pdev->dev.platform_data;
1123 
1124 	if (!config || !config->get_ro)
1125 		return -ENOSYS;
1126 	return config->get_ro(pdev->id);
1127 }
1128 
1129 static void mmc_davinci_enable_sdio_irq(struct mmc_host *mmc, int enable)
1130 {
1131 	struct mmc_davinci_host *host = mmc_priv(mmc);
1132 
1133 	if (enable) {
1134 		if (!(readl(host->base + DAVINCI_SDIOST0) & SDIOST0_DAT1_HI)) {
1135 			writel(SDIOIST_IOINT, host->base + DAVINCI_SDIOIST);
1136 			mmc_signal_sdio_irq(host->mmc);
1137 		} else {
1138 			host->sdio_int = true;
1139 			writel(readl(host->base + DAVINCI_SDIOIEN) |
1140 			       SDIOIEN_IOINTEN, host->base + DAVINCI_SDIOIEN);
1141 		}
1142 	} else {
1143 		host->sdio_int = false;
1144 		writel(readl(host->base + DAVINCI_SDIOIEN) & ~SDIOIEN_IOINTEN,
1145 		       host->base + DAVINCI_SDIOIEN);
1146 	}
1147 }
1148 
1149 static struct mmc_host_ops mmc_davinci_ops = {
1150 	.request	= mmc_davinci_request,
1151 	.set_ios	= mmc_davinci_set_ios,
1152 	.get_cd		= mmc_davinci_get_cd,
1153 	.get_ro		= mmc_davinci_get_ro,
1154 	.enable_sdio_irq = mmc_davinci_enable_sdio_irq,
1155 };
1156 
1157 /*----------------------------------------------------------------------*/
1158 
1159 #ifdef CONFIG_CPU_FREQ
1160 static int mmc_davinci_cpufreq_transition(struct notifier_block *nb,
1161 				     unsigned long val, void *data)
1162 {
1163 	struct mmc_davinci_host *host;
1164 	unsigned int mmc_pclk;
1165 	struct mmc_host *mmc;
1166 	unsigned long flags;
1167 
1168 	host = container_of(nb, struct mmc_davinci_host, freq_transition);
1169 	mmc = host->mmc;
1170 	mmc_pclk = clk_get_rate(host->clk);
1171 
1172 	if (val == CPUFREQ_POSTCHANGE) {
1173 		spin_lock_irqsave(&mmc->lock, flags);
1174 		host->mmc_input_clk = mmc_pclk;
1175 		calculate_clk_divider(mmc, &mmc->ios);
1176 		spin_unlock_irqrestore(&mmc->lock, flags);
1177 	}
1178 
1179 	return 0;
1180 }
1181 
1182 static inline int mmc_davinci_cpufreq_register(struct mmc_davinci_host *host)
1183 {
1184 	host->freq_transition.notifier_call = mmc_davinci_cpufreq_transition;
1185 
1186 	return cpufreq_register_notifier(&host->freq_transition,
1187 					 CPUFREQ_TRANSITION_NOTIFIER);
1188 }
1189 
1190 static inline void mmc_davinci_cpufreq_deregister(struct mmc_davinci_host *host)
1191 {
1192 	cpufreq_unregister_notifier(&host->freq_transition,
1193 				    CPUFREQ_TRANSITION_NOTIFIER);
1194 }
1195 #else
1196 static inline int mmc_davinci_cpufreq_register(struct mmc_davinci_host *host)
1197 {
1198 	return 0;
1199 }
1200 
1201 static inline void mmc_davinci_cpufreq_deregister(struct mmc_davinci_host *host)
1202 {
1203 }
1204 #endif
1205 static void __init init_mmcsd_host(struct mmc_davinci_host *host)
1206 {
1207 
1208 	mmc_davinci_reset_ctrl(host, 1);
1209 
1210 	writel(0, host->base + DAVINCI_MMCCLK);
1211 	writel(MMCCLK_CLKEN, host->base + DAVINCI_MMCCLK);
1212 
1213 	writel(0x1FFF, host->base + DAVINCI_MMCTOR);
1214 	writel(0xFFFF, host->base + DAVINCI_MMCTOD);
1215 
1216 	mmc_davinci_reset_ctrl(host, 0);
1217 }
1218 
1219 static int __init davinci_mmcsd_probe(struct platform_device *pdev)
1220 {
1221 	struct davinci_mmc_config *pdata = pdev->dev.platform_data;
1222 	struct mmc_davinci_host *host = NULL;
1223 	struct mmc_host *mmc = NULL;
1224 	struct resource *r, *mem = NULL;
1225 	int ret = 0, irq = 0;
1226 	size_t mem_size;
1227 
1228 	/* REVISIT:  when we're fully converted, fail if pdata is NULL */
1229 
1230 	ret = -ENODEV;
1231 	r = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1232 	irq = platform_get_irq(pdev, 0);
1233 	if (!r || irq == NO_IRQ)
1234 		goto out;
1235 
1236 	ret = -EBUSY;
1237 	mem_size = resource_size(r);
1238 	mem = request_mem_region(r->start, mem_size, pdev->name);
1239 	if (!mem)
1240 		goto out;
1241 
1242 	ret = -ENOMEM;
1243 	mmc = mmc_alloc_host(sizeof(struct mmc_davinci_host), &pdev->dev);
1244 	if (!mmc)
1245 		goto out;
1246 
1247 	host = mmc_priv(mmc);
1248 	host->mmc = mmc;	/* Important */
1249 
1250 	r = platform_get_resource(pdev, IORESOURCE_DMA, 0);
1251 	if (!r)
1252 		goto out;
1253 	host->rxdma = r->start;
1254 
1255 	r = platform_get_resource(pdev, IORESOURCE_DMA, 1);
1256 	if (!r)
1257 		goto out;
1258 	host->txdma = r->start;
1259 
1260 	host->mem_res = mem;
1261 	host->base = ioremap(mem->start, mem_size);
1262 	if (!host->base)
1263 		goto out;
1264 
1265 	ret = -ENXIO;
1266 	host->clk = clk_get(&pdev->dev, "MMCSDCLK");
1267 	if (IS_ERR(host->clk)) {
1268 		ret = PTR_ERR(host->clk);
1269 		goto out;
1270 	}
1271 	clk_enable(host->clk);
1272 	host->mmc_input_clk = clk_get_rate(host->clk);
1273 
1274 	init_mmcsd_host(host);
1275 
1276 	if (pdata->nr_sg)
1277 		host->nr_sg = pdata->nr_sg - 1;
1278 
1279 	if (host->nr_sg > MAX_NR_SG || !host->nr_sg)
1280 		host->nr_sg = MAX_NR_SG;
1281 
1282 	host->use_dma = use_dma;
1283 	host->mmc_irq = irq;
1284 	host->sdio_irq = platform_get_irq(pdev, 1);
1285 
1286 	if (host->use_dma && davinci_acquire_dma_channels(host) != 0)
1287 		host->use_dma = 0;
1288 
1289 	/* REVISIT:  someday, support IRQ-driven card detection.  */
1290 	mmc->caps |= MMC_CAP_NEEDS_POLL;
1291 	mmc->caps |= MMC_CAP_WAIT_WHILE_BUSY;
1292 
1293 	if (pdata && (pdata->wires == 4 || pdata->wires == 0))
1294 		mmc->caps |= MMC_CAP_4_BIT_DATA;
1295 
1296 	if (pdata && (pdata->wires == 8))
1297 		mmc->caps |= (MMC_CAP_4_BIT_DATA | MMC_CAP_8_BIT_DATA);
1298 
1299 	host->version = pdata->version;
1300 
1301 	mmc->ops = &mmc_davinci_ops;
1302 	mmc->f_min = 312500;
1303 	mmc->f_max = 25000000;
1304 	if (pdata && pdata->max_freq)
1305 		mmc->f_max = pdata->max_freq;
1306 	if (pdata && pdata->caps)
1307 		mmc->caps |= pdata->caps;
1308 	mmc->ocr_avail = MMC_VDD_32_33 | MMC_VDD_33_34;
1309 
1310 	/* With no iommu coalescing pages, each phys_seg is a hw_seg.
1311 	 * Each hw_seg uses one EDMA parameter RAM slot, always one
1312 	 * channel and then usually some linked slots.
1313 	 */
1314 	mmc->max_segs		= 1 + host->n_link;
1315 
1316 	/* EDMA limit per hw segment (one or two MBytes) */
1317 	mmc->max_seg_size	= MAX_CCNT * rw_threshold;
1318 
1319 	/* MMC/SD controller limits for multiblock requests */
1320 	mmc->max_blk_size	= 4095;  /* BLEN is 12 bits */
1321 	mmc->max_blk_count	= 65535; /* NBLK is 16 bits */
1322 	mmc->max_req_size	= mmc->max_blk_size * mmc->max_blk_count;
1323 
1324 	dev_dbg(mmc_dev(host->mmc), "max_segs=%d\n", mmc->max_segs);
1325 	dev_dbg(mmc_dev(host->mmc), "max_blk_size=%d\n", mmc->max_blk_size);
1326 	dev_dbg(mmc_dev(host->mmc), "max_req_size=%d\n", mmc->max_req_size);
1327 	dev_dbg(mmc_dev(host->mmc), "max_seg_size=%d\n", mmc->max_seg_size);
1328 
1329 	platform_set_drvdata(pdev, host);
1330 
1331 	ret = mmc_davinci_cpufreq_register(host);
1332 	if (ret) {
1333 		dev_err(&pdev->dev, "failed to register cpufreq\n");
1334 		goto cpu_freq_fail;
1335 	}
1336 
1337 	ret = mmc_add_host(mmc);
1338 	if (ret < 0)
1339 		goto out;
1340 
1341 	ret = request_irq(irq, mmc_davinci_irq, 0, mmc_hostname(mmc), host);
1342 	if (ret)
1343 		goto out;
1344 
1345 	if (host->sdio_irq >= 0) {
1346 		ret = request_irq(host->sdio_irq, mmc_davinci_sdio_irq, 0,
1347 				  mmc_hostname(mmc), host);
1348 		if (!ret)
1349 			mmc->caps |= MMC_CAP_SDIO_IRQ;
1350 	}
1351 
1352 	rename_region(mem, mmc_hostname(mmc));
1353 
1354 	dev_info(mmc_dev(host->mmc), "Using %s, %d-bit mode\n",
1355 		host->use_dma ? "DMA" : "PIO",
1356 		(mmc->caps & MMC_CAP_4_BIT_DATA) ? 4 : 1);
1357 
1358 	return 0;
1359 
1360 out:
1361 	mmc_davinci_cpufreq_deregister(host);
1362 cpu_freq_fail:
1363 	if (host) {
1364 		davinci_release_dma_channels(host);
1365 
1366 		if (host->clk) {
1367 			clk_disable(host->clk);
1368 			clk_put(host->clk);
1369 		}
1370 
1371 		if (host->base)
1372 			iounmap(host->base);
1373 	}
1374 
1375 	if (mmc)
1376 		mmc_free_host(mmc);
1377 
1378 	if (mem)
1379 		release_resource(mem);
1380 
1381 	dev_dbg(&pdev->dev, "probe err %d\n", ret);
1382 
1383 	return ret;
1384 }
1385 
1386 static int __exit davinci_mmcsd_remove(struct platform_device *pdev)
1387 {
1388 	struct mmc_davinci_host *host = platform_get_drvdata(pdev);
1389 
1390 	platform_set_drvdata(pdev, NULL);
1391 	if (host) {
1392 		mmc_davinci_cpufreq_deregister(host);
1393 
1394 		mmc_remove_host(host->mmc);
1395 		free_irq(host->mmc_irq, host);
1396 		if (host->mmc->caps & MMC_CAP_SDIO_IRQ)
1397 			free_irq(host->sdio_irq, host);
1398 
1399 		davinci_release_dma_channels(host);
1400 
1401 		clk_disable(host->clk);
1402 		clk_put(host->clk);
1403 
1404 		iounmap(host->base);
1405 
1406 		release_resource(host->mem_res);
1407 
1408 		mmc_free_host(host->mmc);
1409 	}
1410 
1411 	return 0;
1412 }
1413 
1414 #ifdef CONFIG_PM
1415 static int davinci_mmcsd_suspend(struct device *dev)
1416 {
1417 	struct platform_device *pdev = to_platform_device(dev);
1418 	struct mmc_davinci_host *host = platform_get_drvdata(pdev);
1419 	int ret;
1420 
1421 	mmc_host_enable(host->mmc);
1422 	ret = mmc_suspend_host(host->mmc);
1423 	if (!ret) {
1424 		writel(0, host->base + DAVINCI_MMCIM);
1425 		mmc_davinci_reset_ctrl(host, 1);
1426 		mmc_host_disable(host->mmc);
1427 		clk_disable(host->clk);
1428 		host->suspended = 1;
1429 	} else {
1430 		host->suspended = 0;
1431 		mmc_host_disable(host->mmc);
1432 	}
1433 
1434 	return ret;
1435 }
1436 
1437 static int davinci_mmcsd_resume(struct device *dev)
1438 {
1439 	struct platform_device *pdev = to_platform_device(dev);
1440 	struct mmc_davinci_host *host = platform_get_drvdata(pdev);
1441 	int ret;
1442 
1443 	if (!host->suspended)
1444 		return 0;
1445 
1446 	clk_enable(host->clk);
1447 	mmc_host_enable(host->mmc);
1448 
1449 	mmc_davinci_reset_ctrl(host, 0);
1450 	ret = mmc_resume_host(host->mmc);
1451 	if (!ret)
1452 		host->suspended = 0;
1453 
1454 	return ret;
1455 }
1456 
1457 static const struct dev_pm_ops davinci_mmcsd_pm = {
1458 	.suspend        = davinci_mmcsd_suspend,
1459 	.resume         = davinci_mmcsd_resume,
1460 };
1461 
1462 #define davinci_mmcsd_pm_ops (&davinci_mmcsd_pm)
1463 #else
1464 #define davinci_mmcsd_pm_ops NULL
1465 #endif
1466 
1467 static struct platform_driver davinci_mmcsd_driver = {
1468 	.driver		= {
1469 		.name	= "davinci_mmc",
1470 		.owner	= THIS_MODULE,
1471 		.pm	= davinci_mmcsd_pm_ops,
1472 	},
1473 	.remove		= __exit_p(davinci_mmcsd_remove),
1474 };
1475 
1476 static int __init davinci_mmcsd_init(void)
1477 {
1478 	return platform_driver_probe(&davinci_mmcsd_driver,
1479 				     davinci_mmcsd_probe);
1480 }
1481 module_init(davinci_mmcsd_init);
1482 
1483 static void __exit davinci_mmcsd_exit(void)
1484 {
1485 	platform_driver_unregister(&davinci_mmcsd_driver);
1486 }
1487 module_exit(davinci_mmcsd_exit);
1488 
1489 MODULE_AUTHOR("Texas Instruments India");
1490 MODULE_LICENSE("GPL");
1491 MODULE_DESCRIPTION("MMC/SD driver for Davinci MMC controller");
1492 
1493