xref: /openbmc/linux/drivers/mmc/host/dw_mmc.c (revision 3b23dc52)
1 /*
2  * Synopsys DesignWare Multimedia Card Interface driver
3  *  (Based on NXP driver for lpc 31xx)
4  *
5  * Copyright (C) 2009 NXP Semiconductors
6  * Copyright (C) 2009, 2010 Imagination Technologies Ltd.
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 
14 #include <linux/blkdev.h>
15 #include <linux/clk.h>
16 #include <linux/debugfs.h>
17 #include <linux/device.h>
18 #include <linux/dma-mapping.h>
19 #include <linux/err.h>
20 #include <linux/init.h>
21 #include <linux/interrupt.h>
22 #include <linux/iopoll.h>
23 #include <linux/ioport.h>
24 #include <linux/module.h>
25 #include <linux/platform_device.h>
26 #include <linux/pm_runtime.h>
27 #include <linux/seq_file.h>
28 #include <linux/slab.h>
29 #include <linux/stat.h>
30 #include <linux/delay.h>
31 #include <linux/irq.h>
32 #include <linux/mmc/card.h>
33 #include <linux/mmc/host.h>
34 #include <linux/mmc/mmc.h>
35 #include <linux/mmc/sd.h>
36 #include <linux/mmc/sdio.h>
37 #include <linux/bitops.h>
38 #include <linux/regulator/consumer.h>
39 #include <linux/of.h>
40 #include <linux/of_gpio.h>
41 #include <linux/mmc/slot-gpio.h>
42 
43 #include "dw_mmc.h"
44 
45 /* Common flag combinations */
46 #define DW_MCI_DATA_ERROR_FLAGS	(SDMMC_INT_DRTO | SDMMC_INT_DCRC | \
47 				 SDMMC_INT_HTO | SDMMC_INT_SBE  | \
48 				 SDMMC_INT_EBE | SDMMC_INT_HLE)
49 #define DW_MCI_CMD_ERROR_FLAGS	(SDMMC_INT_RTO | SDMMC_INT_RCRC | \
50 				 SDMMC_INT_RESP_ERR | SDMMC_INT_HLE)
51 #define DW_MCI_ERROR_FLAGS	(DW_MCI_DATA_ERROR_FLAGS | \
52 				 DW_MCI_CMD_ERROR_FLAGS)
53 #define DW_MCI_SEND_STATUS	1
54 #define DW_MCI_RECV_STATUS	2
55 #define DW_MCI_DMA_THRESHOLD	16
56 
57 #define DW_MCI_FREQ_MAX	200000000	/* unit: HZ */
58 #define DW_MCI_FREQ_MIN	100000		/* unit: HZ */
59 
60 #define IDMAC_INT_CLR		(SDMMC_IDMAC_INT_AI | SDMMC_IDMAC_INT_NI | \
61 				 SDMMC_IDMAC_INT_CES | SDMMC_IDMAC_INT_DU | \
62 				 SDMMC_IDMAC_INT_FBE | SDMMC_IDMAC_INT_RI | \
63 				 SDMMC_IDMAC_INT_TI)
64 
65 #define DESC_RING_BUF_SZ	PAGE_SIZE
66 
67 struct idmac_desc_64addr {
68 	u32		des0;	/* Control Descriptor */
69 #define IDMAC_OWN_CLR64(x) \
70 	!((x) & cpu_to_le32(IDMAC_DES0_OWN))
71 
72 	u32		des1;	/* Reserved */
73 
74 	u32		des2;	/*Buffer sizes */
75 #define IDMAC_64ADDR_SET_BUFFER1_SIZE(d, s) \
76 	((d)->des2 = ((d)->des2 & cpu_to_le32(0x03ffe000)) | \
77 	 ((cpu_to_le32(s)) & cpu_to_le32(0x1fff)))
78 
79 	u32		des3;	/* Reserved */
80 
81 	u32		des4;	/* Lower 32-bits of Buffer Address Pointer 1*/
82 	u32		des5;	/* Upper 32-bits of Buffer Address Pointer 1*/
83 
84 	u32		des6;	/* Lower 32-bits of Next Descriptor Address */
85 	u32		des7;	/* Upper 32-bits of Next Descriptor Address */
86 };
87 
88 struct idmac_desc {
89 	__le32		des0;	/* Control Descriptor */
90 #define IDMAC_DES0_DIC	BIT(1)
91 #define IDMAC_DES0_LD	BIT(2)
92 #define IDMAC_DES0_FD	BIT(3)
93 #define IDMAC_DES0_CH	BIT(4)
94 #define IDMAC_DES0_ER	BIT(5)
95 #define IDMAC_DES0_CES	BIT(30)
96 #define IDMAC_DES0_OWN	BIT(31)
97 
98 	__le32		des1;	/* Buffer sizes */
99 #define IDMAC_SET_BUFFER1_SIZE(d, s) \
100 	((d)->des1 = ((d)->des1 & cpu_to_le32(0x03ffe000)) | (cpu_to_le32((s) & 0x1fff)))
101 
102 	__le32		des2;	/* buffer 1 physical address */
103 
104 	__le32		des3;	/* buffer 2 physical address */
105 };
106 
107 /* Each descriptor can transfer up to 4KB of data in chained mode */
108 #define DW_MCI_DESC_DATA_LENGTH	0x1000
109 
110 #if defined(CONFIG_DEBUG_FS)
111 static int dw_mci_req_show(struct seq_file *s, void *v)
112 {
113 	struct dw_mci_slot *slot = s->private;
114 	struct mmc_request *mrq;
115 	struct mmc_command *cmd;
116 	struct mmc_command *stop;
117 	struct mmc_data	*data;
118 
119 	/* Make sure we get a consistent snapshot */
120 	spin_lock_bh(&slot->host->lock);
121 	mrq = slot->mrq;
122 
123 	if (mrq) {
124 		cmd = mrq->cmd;
125 		data = mrq->data;
126 		stop = mrq->stop;
127 
128 		if (cmd)
129 			seq_printf(s,
130 				   "CMD%u(0x%x) flg %x rsp %x %x %x %x err %d\n",
131 				   cmd->opcode, cmd->arg, cmd->flags,
132 				   cmd->resp[0], cmd->resp[1], cmd->resp[2],
133 				   cmd->resp[2], cmd->error);
134 		if (data)
135 			seq_printf(s, "DATA %u / %u * %u flg %x err %d\n",
136 				   data->bytes_xfered, data->blocks,
137 				   data->blksz, data->flags, data->error);
138 		if (stop)
139 			seq_printf(s,
140 				   "CMD%u(0x%x) flg %x rsp %x %x %x %x err %d\n",
141 				   stop->opcode, stop->arg, stop->flags,
142 				   stop->resp[0], stop->resp[1], stop->resp[2],
143 				   stop->resp[2], stop->error);
144 	}
145 
146 	spin_unlock_bh(&slot->host->lock);
147 
148 	return 0;
149 }
150 DEFINE_SHOW_ATTRIBUTE(dw_mci_req);
151 
152 static int dw_mci_regs_show(struct seq_file *s, void *v)
153 {
154 	struct dw_mci *host = s->private;
155 
156 	pm_runtime_get_sync(host->dev);
157 
158 	seq_printf(s, "STATUS:\t0x%08x\n", mci_readl(host, STATUS));
159 	seq_printf(s, "RINTSTS:\t0x%08x\n", mci_readl(host, RINTSTS));
160 	seq_printf(s, "CMD:\t0x%08x\n", mci_readl(host, CMD));
161 	seq_printf(s, "CTRL:\t0x%08x\n", mci_readl(host, CTRL));
162 	seq_printf(s, "INTMASK:\t0x%08x\n", mci_readl(host, INTMASK));
163 	seq_printf(s, "CLKENA:\t0x%08x\n", mci_readl(host, CLKENA));
164 
165 	pm_runtime_put_autosuspend(host->dev);
166 
167 	return 0;
168 }
169 DEFINE_SHOW_ATTRIBUTE(dw_mci_regs);
170 
171 static void dw_mci_init_debugfs(struct dw_mci_slot *slot)
172 {
173 	struct mmc_host	*mmc = slot->mmc;
174 	struct dw_mci *host = slot->host;
175 	struct dentry *root;
176 	struct dentry *node;
177 
178 	root = mmc->debugfs_root;
179 	if (!root)
180 		return;
181 
182 	node = debugfs_create_file("regs", S_IRUSR, root, host,
183 				   &dw_mci_regs_fops);
184 	if (!node)
185 		goto err;
186 
187 	node = debugfs_create_file("req", S_IRUSR, root, slot,
188 				   &dw_mci_req_fops);
189 	if (!node)
190 		goto err;
191 
192 	node = debugfs_create_u32("state", S_IRUSR, root, (u32 *)&host->state);
193 	if (!node)
194 		goto err;
195 
196 	node = debugfs_create_x32("pending_events", S_IRUSR, root,
197 				  (u32 *)&host->pending_events);
198 	if (!node)
199 		goto err;
200 
201 	node = debugfs_create_x32("completed_events", S_IRUSR, root,
202 				  (u32 *)&host->completed_events);
203 	if (!node)
204 		goto err;
205 
206 	return;
207 
208 err:
209 	dev_err(&mmc->class_dev, "failed to initialize debugfs for slot\n");
210 }
211 #endif /* defined(CONFIG_DEBUG_FS) */
212 
213 static bool dw_mci_ctrl_reset(struct dw_mci *host, u32 reset)
214 {
215 	u32 ctrl;
216 
217 	ctrl = mci_readl(host, CTRL);
218 	ctrl |= reset;
219 	mci_writel(host, CTRL, ctrl);
220 
221 	/* wait till resets clear */
222 	if (readl_poll_timeout_atomic(host->regs + SDMMC_CTRL, ctrl,
223 				      !(ctrl & reset),
224 				      1, 500 * USEC_PER_MSEC)) {
225 		dev_err(host->dev,
226 			"Timeout resetting block (ctrl reset %#x)\n",
227 			ctrl & reset);
228 		return false;
229 	}
230 
231 	return true;
232 }
233 
234 static void dw_mci_wait_while_busy(struct dw_mci *host, u32 cmd_flags)
235 {
236 	u32 status;
237 
238 	/*
239 	 * Databook says that before issuing a new data transfer command
240 	 * we need to check to see if the card is busy.  Data transfer commands
241 	 * all have SDMMC_CMD_PRV_DAT_WAIT set, so we'll key off that.
242 	 *
243 	 * ...also allow sending for SDMMC_CMD_VOLT_SWITCH where busy is
244 	 * expected.
245 	 */
246 	if ((cmd_flags & SDMMC_CMD_PRV_DAT_WAIT) &&
247 	    !(cmd_flags & SDMMC_CMD_VOLT_SWITCH)) {
248 		if (readl_poll_timeout_atomic(host->regs + SDMMC_STATUS,
249 					      status,
250 					      !(status & SDMMC_STATUS_BUSY),
251 					      10, 500 * USEC_PER_MSEC))
252 			dev_err(host->dev, "Busy; trying anyway\n");
253 	}
254 }
255 
256 static void mci_send_cmd(struct dw_mci_slot *slot, u32 cmd, u32 arg)
257 {
258 	struct dw_mci *host = slot->host;
259 	unsigned int cmd_status = 0;
260 
261 	mci_writel(host, CMDARG, arg);
262 	wmb(); /* drain writebuffer */
263 	dw_mci_wait_while_busy(host, cmd);
264 	mci_writel(host, CMD, SDMMC_CMD_START | cmd);
265 
266 	if (readl_poll_timeout_atomic(host->regs + SDMMC_CMD, cmd_status,
267 				      !(cmd_status & SDMMC_CMD_START),
268 				      1, 500 * USEC_PER_MSEC))
269 		dev_err(&slot->mmc->class_dev,
270 			"Timeout sending command (cmd %#x arg %#x status %#x)\n",
271 			cmd, arg, cmd_status);
272 }
273 
274 static u32 dw_mci_prepare_command(struct mmc_host *mmc, struct mmc_command *cmd)
275 {
276 	struct dw_mci_slot *slot = mmc_priv(mmc);
277 	struct dw_mci *host = slot->host;
278 	u32 cmdr;
279 
280 	cmd->error = -EINPROGRESS;
281 	cmdr = cmd->opcode;
282 
283 	if (cmd->opcode == MMC_STOP_TRANSMISSION ||
284 	    cmd->opcode == MMC_GO_IDLE_STATE ||
285 	    cmd->opcode == MMC_GO_INACTIVE_STATE ||
286 	    (cmd->opcode == SD_IO_RW_DIRECT &&
287 	     ((cmd->arg >> 9) & 0x1FFFF) == SDIO_CCCR_ABORT))
288 		cmdr |= SDMMC_CMD_STOP;
289 	else if (cmd->opcode != MMC_SEND_STATUS && cmd->data)
290 		cmdr |= SDMMC_CMD_PRV_DAT_WAIT;
291 
292 	if (cmd->opcode == SD_SWITCH_VOLTAGE) {
293 		u32 clk_en_a;
294 
295 		/* Special bit makes CMD11 not die */
296 		cmdr |= SDMMC_CMD_VOLT_SWITCH;
297 
298 		/* Change state to continue to handle CMD11 weirdness */
299 		WARN_ON(slot->host->state != STATE_SENDING_CMD);
300 		slot->host->state = STATE_SENDING_CMD11;
301 
302 		/*
303 		 * We need to disable low power mode (automatic clock stop)
304 		 * while doing voltage switch so we don't confuse the card,
305 		 * since stopping the clock is a specific part of the UHS
306 		 * voltage change dance.
307 		 *
308 		 * Note that low power mode (SDMMC_CLKEN_LOW_PWR) will be
309 		 * unconditionally turned back on in dw_mci_setup_bus() if it's
310 		 * ever called with a non-zero clock.  That shouldn't happen
311 		 * until the voltage change is all done.
312 		 */
313 		clk_en_a = mci_readl(host, CLKENA);
314 		clk_en_a &= ~(SDMMC_CLKEN_LOW_PWR << slot->id);
315 		mci_writel(host, CLKENA, clk_en_a);
316 		mci_send_cmd(slot, SDMMC_CMD_UPD_CLK |
317 			     SDMMC_CMD_PRV_DAT_WAIT, 0);
318 	}
319 
320 	if (cmd->flags & MMC_RSP_PRESENT) {
321 		/* We expect a response, so set this bit */
322 		cmdr |= SDMMC_CMD_RESP_EXP;
323 		if (cmd->flags & MMC_RSP_136)
324 			cmdr |= SDMMC_CMD_RESP_LONG;
325 	}
326 
327 	if (cmd->flags & MMC_RSP_CRC)
328 		cmdr |= SDMMC_CMD_RESP_CRC;
329 
330 	if (cmd->data) {
331 		cmdr |= SDMMC_CMD_DAT_EXP;
332 		if (cmd->data->flags & MMC_DATA_WRITE)
333 			cmdr |= SDMMC_CMD_DAT_WR;
334 	}
335 
336 	if (!test_bit(DW_MMC_CARD_NO_USE_HOLD, &slot->flags))
337 		cmdr |= SDMMC_CMD_USE_HOLD_REG;
338 
339 	return cmdr;
340 }
341 
342 static u32 dw_mci_prep_stop_abort(struct dw_mci *host, struct mmc_command *cmd)
343 {
344 	struct mmc_command *stop;
345 	u32 cmdr;
346 
347 	if (!cmd->data)
348 		return 0;
349 
350 	stop = &host->stop_abort;
351 	cmdr = cmd->opcode;
352 	memset(stop, 0, sizeof(struct mmc_command));
353 
354 	if (cmdr == MMC_READ_SINGLE_BLOCK ||
355 	    cmdr == MMC_READ_MULTIPLE_BLOCK ||
356 	    cmdr == MMC_WRITE_BLOCK ||
357 	    cmdr == MMC_WRITE_MULTIPLE_BLOCK ||
358 	    cmdr == MMC_SEND_TUNING_BLOCK ||
359 	    cmdr == MMC_SEND_TUNING_BLOCK_HS200) {
360 		stop->opcode = MMC_STOP_TRANSMISSION;
361 		stop->arg = 0;
362 		stop->flags = MMC_RSP_R1B | MMC_CMD_AC;
363 	} else if (cmdr == SD_IO_RW_EXTENDED) {
364 		stop->opcode = SD_IO_RW_DIRECT;
365 		stop->arg |= (1 << 31) | (0 << 28) | (SDIO_CCCR_ABORT << 9) |
366 			     ((cmd->arg >> 28) & 0x7);
367 		stop->flags = MMC_RSP_SPI_R5 | MMC_RSP_R5 | MMC_CMD_AC;
368 	} else {
369 		return 0;
370 	}
371 
372 	cmdr = stop->opcode | SDMMC_CMD_STOP |
373 		SDMMC_CMD_RESP_CRC | SDMMC_CMD_RESP_EXP;
374 
375 	if (!test_bit(DW_MMC_CARD_NO_USE_HOLD, &host->slot->flags))
376 		cmdr |= SDMMC_CMD_USE_HOLD_REG;
377 
378 	return cmdr;
379 }
380 
381 static inline void dw_mci_set_cto(struct dw_mci *host)
382 {
383 	unsigned int cto_clks;
384 	unsigned int cto_div;
385 	unsigned int cto_ms;
386 	unsigned long irqflags;
387 
388 	cto_clks = mci_readl(host, TMOUT) & 0xff;
389 	cto_div = (mci_readl(host, CLKDIV) & 0xff) * 2;
390 	if (cto_div == 0)
391 		cto_div = 1;
392 
393 	cto_ms = DIV_ROUND_UP_ULL((u64)MSEC_PER_SEC * cto_clks * cto_div,
394 				  host->bus_hz);
395 
396 	/* add a bit spare time */
397 	cto_ms += 10;
398 
399 	/*
400 	 * The durations we're working with are fairly short so we have to be
401 	 * extra careful about synchronization here.  Specifically in hardware a
402 	 * command timeout is _at most_ 5.1 ms, so that means we expect an
403 	 * interrupt (either command done or timeout) to come rather quickly
404 	 * after the mci_writel.  ...but just in case we have a long interrupt
405 	 * latency let's add a bit of paranoia.
406 	 *
407 	 * In general we'll assume that at least an interrupt will be asserted
408 	 * in hardware by the time the cto_timer runs.  ...and if it hasn't
409 	 * been asserted in hardware by that time then we'll assume it'll never
410 	 * come.
411 	 */
412 	spin_lock_irqsave(&host->irq_lock, irqflags);
413 	if (!test_bit(EVENT_CMD_COMPLETE, &host->pending_events))
414 		mod_timer(&host->cto_timer,
415 			jiffies + msecs_to_jiffies(cto_ms) + 1);
416 	spin_unlock_irqrestore(&host->irq_lock, irqflags);
417 }
418 
419 static void dw_mci_start_command(struct dw_mci *host,
420 				 struct mmc_command *cmd, u32 cmd_flags)
421 {
422 	host->cmd = cmd;
423 	dev_vdbg(host->dev,
424 		 "start command: ARGR=0x%08x CMDR=0x%08x\n",
425 		 cmd->arg, cmd_flags);
426 
427 	mci_writel(host, CMDARG, cmd->arg);
428 	wmb(); /* drain writebuffer */
429 	dw_mci_wait_while_busy(host, cmd_flags);
430 
431 	mci_writel(host, CMD, cmd_flags | SDMMC_CMD_START);
432 
433 	/* response expected command only */
434 	if (cmd_flags & SDMMC_CMD_RESP_EXP)
435 		dw_mci_set_cto(host);
436 }
437 
438 static inline void send_stop_abort(struct dw_mci *host, struct mmc_data *data)
439 {
440 	struct mmc_command *stop = &host->stop_abort;
441 
442 	dw_mci_start_command(host, stop, host->stop_cmdr);
443 }
444 
445 /* DMA interface functions */
446 static void dw_mci_stop_dma(struct dw_mci *host)
447 {
448 	if (host->using_dma) {
449 		host->dma_ops->stop(host);
450 		host->dma_ops->cleanup(host);
451 	}
452 
453 	/* Data transfer was stopped by the interrupt handler */
454 	set_bit(EVENT_XFER_COMPLETE, &host->pending_events);
455 }
456 
457 static void dw_mci_dma_cleanup(struct dw_mci *host)
458 {
459 	struct mmc_data *data = host->data;
460 
461 	if (data && data->host_cookie == COOKIE_MAPPED) {
462 		dma_unmap_sg(host->dev,
463 			     data->sg,
464 			     data->sg_len,
465 			     mmc_get_dma_dir(data));
466 		data->host_cookie = COOKIE_UNMAPPED;
467 	}
468 }
469 
470 static void dw_mci_idmac_reset(struct dw_mci *host)
471 {
472 	u32 bmod = mci_readl(host, BMOD);
473 	/* Software reset of DMA */
474 	bmod |= SDMMC_IDMAC_SWRESET;
475 	mci_writel(host, BMOD, bmod);
476 }
477 
478 static void dw_mci_idmac_stop_dma(struct dw_mci *host)
479 {
480 	u32 temp;
481 
482 	/* Disable and reset the IDMAC interface */
483 	temp = mci_readl(host, CTRL);
484 	temp &= ~SDMMC_CTRL_USE_IDMAC;
485 	temp |= SDMMC_CTRL_DMA_RESET;
486 	mci_writel(host, CTRL, temp);
487 
488 	/* Stop the IDMAC running */
489 	temp = mci_readl(host, BMOD);
490 	temp &= ~(SDMMC_IDMAC_ENABLE | SDMMC_IDMAC_FB);
491 	temp |= SDMMC_IDMAC_SWRESET;
492 	mci_writel(host, BMOD, temp);
493 }
494 
495 static void dw_mci_dmac_complete_dma(void *arg)
496 {
497 	struct dw_mci *host = arg;
498 	struct mmc_data *data = host->data;
499 
500 	dev_vdbg(host->dev, "DMA complete\n");
501 
502 	if ((host->use_dma == TRANS_MODE_EDMAC) &&
503 	    data && (data->flags & MMC_DATA_READ))
504 		/* Invalidate cache after read */
505 		dma_sync_sg_for_cpu(mmc_dev(host->slot->mmc),
506 				    data->sg,
507 				    data->sg_len,
508 				    DMA_FROM_DEVICE);
509 
510 	host->dma_ops->cleanup(host);
511 
512 	/*
513 	 * If the card was removed, data will be NULL. No point in trying to
514 	 * send the stop command or waiting for NBUSY in this case.
515 	 */
516 	if (data) {
517 		set_bit(EVENT_XFER_COMPLETE, &host->pending_events);
518 		tasklet_schedule(&host->tasklet);
519 	}
520 }
521 
522 static int dw_mci_idmac_init(struct dw_mci *host)
523 {
524 	int i;
525 
526 	if (host->dma_64bit_address == 1) {
527 		struct idmac_desc_64addr *p;
528 		/* Number of descriptors in the ring buffer */
529 		host->ring_size =
530 			DESC_RING_BUF_SZ / sizeof(struct idmac_desc_64addr);
531 
532 		/* Forward link the descriptor list */
533 		for (i = 0, p = host->sg_cpu; i < host->ring_size - 1;
534 								i++, p++) {
535 			p->des6 = (host->sg_dma +
536 					(sizeof(struct idmac_desc_64addr) *
537 							(i + 1))) & 0xffffffff;
538 
539 			p->des7 = (u64)(host->sg_dma +
540 					(sizeof(struct idmac_desc_64addr) *
541 							(i + 1))) >> 32;
542 			/* Initialize reserved and buffer size fields to "0" */
543 			p->des0 = 0;
544 			p->des1 = 0;
545 			p->des2 = 0;
546 			p->des3 = 0;
547 		}
548 
549 		/* Set the last descriptor as the end-of-ring descriptor */
550 		p->des6 = host->sg_dma & 0xffffffff;
551 		p->des7 = (u64)host->sg_dma >> 32;
552 		p->des0 = IDMAC_DES0_ER;
553 
554 	} else {
555 		struct idmac_desc *p;
556 		/* Number of descriptors in the ring buffer */
557 		host->ring_size =
558 			DESC_RING_BUF_SZ / sizeof(struct idmac_desc);
559 
560 		/* Forward link the descriptor list */
561 		for (i = 0, p = host->sg_cpu;
562 		     i < host->ring_size - 1;
563 		     i++, p++) {
564 			p->des3 = cpu_to_le32(host->sg_dma +
565 					(sizeof(struct idmac_desc) * (i + 1)));
566 			p->des0 = 0;
567 			p->des1 = 0;
568 		}
569 
570 		/* Set the last descriptor as the end-of-ring descriptor */
571 		p->des3 = cpu_to_le32(host->sg_dma);
572 		p->des0 = cpu_to_le32(IDMAC_DES0_ER);
573 	}
574 
575 	dw_mci_idmac_reset(host);
576 
577 	if (host->dma_64bit_address == 1) {
578 		/* Mask out interrupts - get Tx & Rx complete only */
579 		mci_writel(host, IDSTS64, IDMAC_INT_CLR);
580 		mci_writel(host, IDINTEN64, SDMMC_IDMAC_INT_NI |
581 				SDMMC_IDMAC_INT_RI | SDMMC_IDMAC_INT_TI);
582 
583 		/* Set the descriptor base address */
584 		mci_writel(host, DBADDRL, host->sg_dma & 0xffffffff);
585 		mci_writel(host, DBADDRU, (u64)host->sg_dma >> 32);
586 
587 	} else {
588 		/* Mask out interrupts - get Tx & Rx complete only */
589 		mci_writel(host, IDSTS, IDMAC_INT_CLR);
590 		mci_writel(host, IDINTEN, SDMMC_IDMAC_INT_NI |
591 				SDMMC_IDMAC_INT_RI | SDMMC_IDMAC_INT_TI);
592 
593 		/* Set the descriptor base address */
594 		mci_writel(host, DBADDR, host->sg_dma);
595 	}
596 
597 	return 0;
598 }
599 
600 static inline int dw_mci_prepare_desc64(struct dw_mci *host,
601 					 struct mmc_data *data,
602 					 unsigned int sg_len)
603 {
604 	unsigned int desc_len;
605 	struct idmac_desc_64addr *desc_first, *desc_last, *desc;
606 	u32 val;
607 	int i;
608 
609 	desc_first = desc_last = desc = host->sg_cpu;
610 
611 	for (i = 0; i < sg_len; i++) {
612 		unsigned int length = sg_dma_len(&data->sg[i]);
613 
614 		u64 mem_addr = sg_dma_address(&data->sg[i]);
615 
616 		for ( ; length ; desc++) {
617 			desc_len = (length <= DW_MCI_DESC_DATA_LENGTH) ?
618 				   length : DW_MCI_DESC_DATA_LENGTH;
619 
620 			length -= desc_len;
621 
622 			/*
623 			 * Wait for the former clear OWN bit operation
624 			 * of IDMAC to make sure that this descriptor
625 			 * isn't still owned by IDMAC as IDMAC's write
626 			 * ops and CPU's read ops are asynchronous.
627 			 */
628 			if (readl_poll_timeout_atomic(&desc->des0, val,
629 						!(val & IDMAC_DES0_OWN),
630 						10, 100 * USEC_PER_MSEC))
631 				goto err_own_bit;
632 
633 			/*
634 			 * Set the OWN bit and disable interrupts
635 			 * for this descriptor
636 			 */
637 			desc->des0 = IDMAC_DES0_OWN | IDMAC_DES0_DIC |
638 						IDMAC_DES0_CH;
639 
640 			/* Buffer length */
641 			IDMAC_64ADDR_SET_BUFFER1_SIZE(desc, desc_len);
642 
643 			/* Physical address to DMA to/from */
644 			desc->des4 = mem_addr & 0xffffffff;
645 			desc->des5 = mem_addr >> 32;
646 
647 			/* Update physical address for the next desc */
648 			mem_addr += desc_len;
649 
650 			/* Save pointer to the last descriptor */
651 			desc_last = desc;
652 		}
653 	}
654 
655 	/* Set first descriptor */
656 	desc_first->des0 |= IDMAC_DES0_FD;
657 
658 	/* Set last descriptor */
659 	desc_last->des0 &= ~(IDMAC_DES0_CH | IDMAC_DES0_DIC);
660 	desc_last->des0 |= IDMAC_DES0_LD;
661 
662 	return 0;
663 err_own_bit:
664 	/* restore the descriptor chain as it's polluted */
665 	dev_dbg(host->dev, "descriptor is still owned by IDMAC.\n");
666 	memset(host->sg_cpu, 0, DESC_RING_BUF_SZ);
667 	dw_mci_idmac_init(host);
668 	return -EINVAL;
669 }
670 
671 
672 static inline int dw_mci_prepare_desc32(struct dw_mci *host,
673 					 struct mmc_data *data,
674 					 unsigned int sg_len)
675 {
676 	unsigned int desc_len;
677 	struct idmac_desc *desc_first, *desc_last, *desc;
678 	u32 val;
679 	int i;
680 
681 	desc_first = desc_last = desc = host->sg_cpu;
682 
683 	for (i = 0; i < sg_len; i++) {
684 		unsigned int length = sg_dma_len(&data->sg[i]);
685 
686 		u32 mem_addr = sg_dma_address(&data->sg[i]);
687 
688 		for ( ; length ; desc++) {
689 			desc_len = (length <= DW_MCI_DESC_DATA_LENGTH) ?
690 				   length : DW_MCI_DESC_DATA_LENGTH;
691 
692 			length -= desc_len;
693 
694 			/*
695 			 * Wait for the former clear OWN bit operation
696 			 * of IDMAC to make sure that this descriptor
697 			 * isn't still owned by IDMAC as IDMAC's write
698 			 * ops and CPU's read ops are asynchronous.
699 			 */
700 			if (readl_poll_timeout_atomic(&desc->des0, val,
701 						      IDMAC_OWN_CLR64(val),
702 						      10,
703 						      100 * USEC_PER_MSEC))
704 				goto err_own_bit;
705 
706 			/*
707 			 * Set the OWN bit and disable interrupts
708 			 * for this descriptor
709 			 */
710 			desc->des0 = cpu_to_le32(IDMAC_DES0_OWN |
711 						 IDMAC_DES0_DIC |
712 						 IDMAC_DES0_CH);
713 
714 			/* Buffer length */
715 			IDMAC_SET_BUFFER1_SIZE(desc, desc_len);
716 
717 			/* Physical address to DMA to/from */
718 			desc->des2 = cpu_to_le32(mem_addr);
719 
720 			/* Update physical address for the next desc */
721 			mem_addr += desc_len;
722 
723 			/* Save pointer to the last descriptor */
724 			desc_last = desc;
725 		}
726 	}
727 
728 	/* Set first descriptor */
729 	desc_first->des0 |= cpu_to_le32(IDMAC_DES0_FD);
730 
731 	/* Set last descriptor */
732 	desc_last->des0 &= cpu_to_le32(~(IDMAC_DES0_CH |
733 				       IDMAC_DES0_DIC));
734 	desc_last->des0 |= cpu_to_le32(IDMAC_DES0_LD);
735 
736 	return 0;
737 err_own_bit:
738 	/* restore the descriptor chain as it's polluted */
739 	dev_dbg(host->dev, "descriptor is still owned by IDMAC.\n");
740 	memset(host->sg_cpu, 0, DESC_RING_BUF_SZ);
741 	dw_mci_idmac_init(host);
742 	return -EINVAL;
743 }
744 
745 static int dw_mci_idmac_start_dma(struct dw_mci *host, unsigned int sg_len)
746 {
747 	u32 temp;
748 	int ret;
749 
750 	if (host->dma_64bit_address == 1)
751 		ret = dw_mci_prepare_desc64(host, host->data, sg_len);
752 	else
753 		ret = dw_mci_prepare_desc32(host, host->data, sg_len);
754 
755 	if (ret)
756 		goto out;
757 
758 	/* drain writebuffer */
759 	wmb();
760 
761 	/* Make sure to reset DMA in case we did PIO before this */
762 	dw_mci_ctrl_reset(host, SDMMC_CTRL_DMA_RESET);
763 	dw_mci_idmac_reset(host);
764 
765 	/* Select IDMAC interface */
766 	temp = mci_readl(host, CTRL);
767 	temp |= SDMMC_CTRL_USE_IDMAC;
768 	mci_writel(host, CTRL, temp);
769 
770 	/* drain writebuffer */
771 	wmb();
772 
773 	/* Enable the IDMAC */
774 	temp = mci_readl(host, BMOD);
775 	temp |= SDMMC_IDMAC_ENABLE | SDMMC_IDMAC_FB;
776 	mci_writel(host, BMOD, temp);
777 
778 	/* Start it running */
779 	mci_writel(host, PLDMND, 1);
780 
781 out:
782 	return ret;
783 }
784 
785 static const struct dw_mci_dma_ops dw_mci_idmac_ops = {
786 	.init = dw_mci_idmac_init,
787 	.start = dw_mci_idmac_start_dma,
788 	.stop = dw_mci_idmac_stop_dma,
789 	.complete = dw_mci_dmac_complete_dma,
790 	.cleanup = dw_mci_dma_cleanup,
791 };
792 
793 static void dw_mci_edmac_stop_dma(struct dw_mci *host)
794 {
795 	dmaengine_terminate_async(host->dms->ch);
796 }
797 
798 static int dw_mci_edmac_start_dma(struct dw_mci *host,
799 					    unsigned int sg_len)
800 {
801 	struct dma_slave_config cfg;
802 	struct dma_async_tx_descriptor *desc = NULL;
803 	struct scatterlist *sgl = host->data->sg;
804 	static const u32 mszs[] = {1, 4, 8, 16, 32, 64, 128, 256};
805 	u32 sg_elems = host->data->sg_len;
806 	u32 fifoth_val;
807 	u32 fifo_offset = host->fifo_reg - host->regs;
808 	int ret = 0;
809 
810 	/* Set external dma config: burst size, burst width */
811 	cfg.dst_addr = host->phy_regs + fifo_offset;
812 	cfg.src_addr = cfg.dst_addr;
813 	cfg.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
814 	cfg.src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
815 
816 	/* Match burst msize with external dma config */
817 	fifoth_val = mci_readl(host, FIFOTH);
818 	cfg.dst_maxburst = mszs[(fifoth_val >> 28) & 0x7];
819 	cfg.src_maxburst = cfg.dst_maxburst;
820 
821 	if (host->data->flags & MMC_DATA_WRITE)
822 		cfg.direction = DMA_MEM_TO_DEV;
823 	else
824 		cfg.direction = DMA_DEV_TO_MEM;
825 
826 	ret = dmaengine_slave_config(host->dms->ch, &cfg);
827 	if (ret) {
828 		dev_err(host->dev, "Failed to config edmac.\n");
829 		return -EBUSY;
830 	}
831 
832 	desc = dmaengine_prep_slave_sg(host->dms->ch, sgl,
833 				       sg_len, cfg.direction,
834 				       DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
835 	if (!desc) {
836 		dev_err(host->dev, "Can't prepare slave sg.\n");
837 		return -EBUSY;
838 	}
839 
840 	/* Set dw_mci_dmac_complete_dma as callback */
841 	desc->callback = dw_mci_dmac_complete_dma;
842 	desc->callback_param = (void *)host;
843 	dmaengine_submit(desc);
844 
845 	/* Flush cache before write */
846 	if (host->data->flags & MMC_DATA_WRITE)
847 		dma_sync_sg_for_device(mmc_dev(host->slot->mmc), sgl,
848 				       sg_elems, DMA_TO_DEVICE);
849 
850 	dma_async_issue_pending(host->dms->ch);
851 
852 	return 0;
853 }
854 
855 static int dw_mci_edmac_init(struct dw_mci *host)
856 {
857 	/* Request external dma channel */
858 	host->dms = kzalloc(sizeof(struct dw_mci_dma_slave), GFP_KERNEL);
859 	if (!host->dms)
860 		return -ENOMEM;
861 
862 	host->dms->ch = dma_request_slave_channel(host->dev, "rx-tx");
863 	if (!host->dms->ch) {
864 		dev_err(host->dev, "Failed to get external DMA channel.\n");
865 		kfree(host->dms);
866 		host->dms = NULL;
867 		return -ENXIO;
868 	}
869 
870 	return 0;
871 }
872 
873 static void dw_mci_edmac_exit(struct dw_mci *host)
874 {
875 	if (host->dms) {
876 		if (host->dms->ch) {
877 			dma_release_channel(host->dms->ch);
878 			host->dms->ch = NULL;
879 		}
880 		kfree(host->dms);
881 		host->dms = NULL;
882 	}
883 }
884 
885 static const struct dw_mci_dma_ops dw_mci_edmac_ops = {
886 	.init = dw_mci_edmac_init,
887 	.exit = dw_mci_edmac_exit,
888 	.start = dw_mci_edmac_start_dma,
889 	.stop = dw_mci_edmac_stop_dma,
890 	.complete = dw_mci_dmac_complete_dma,
891 	.cleanup = dw_mci_dma_cleanup,
892 };
893 
894 static int dw_mci_pre_dma_transfer(struct dw_mci *host,
895 				   struct mmc_data *data,
896 				   int cookie)
897 {
898 	struct scatterlist *sg;
899 	unsigned int i, sg_len;
900 
901 	if (data->host_cookie == COOKIE_PRE_MAPPED)
902 		return data->sg_len;
903 
904 	/*
905 	 * We don't do DMA on "complex" transfers, i.e. with
906 	 * non-word-aligned buffers or lengths. Also, we don't bother
907 	 * with all the DMA setup overhead for short transfers.
908 	 */
909 	if (data->blocks * data->blksz < DW_MCI_DMA_THRESHOLD)
910 		return -EINVAL;
911 
912 	if (data->blksz & 3)
913 		return -EINVAL;
914 
915 	for_each_sg(data->sg, sg, data->sg_len, i) {
916 		if (sg->offset & 3 || sg->length & 3)
917 			return -EINVAL;
918 	}
919 
920 	sg_len = dma_map_sg(host->dev,
921 			    data->sg,
922 			    data->sg_len,
923 			    mmc_get_dma_dir(data));
924 	if (sg_len == 0)
925 		return -EINVAL;
926 
927 	data->host_cookie = cookie;
928 
929 	return sg_len;
930 }
931 
932 static void dw_mci_pre_req(struct mmc_host *mmc,
933 			   struct mmc_request *mrq)
934 {
935 	struct dw_mci_slot *slot = mmc_priv(mmc);
936 	struct mmc_data *data = mrq->data;
937 
938 	if (!slot->host->use_dma || !data)
939 		return;
940 
941 	/* This data might be unmapped at this time */
942 	data->host_cookie = COOKIE_UNMAPPED;
943 
944 	if (dw_mci_pre_dma_transfer(slot->host, mrq->data,
945 				COOKIE_PRE_MAPPED) < 0)
946 		data->host_cookie = COOKIE_UNMAPPED;
947 }
948 
949 static void dw_mci_post_req(struct mmc_host *mmc,
950 			    struct mmc_request *mrq,
951 			    int err)
952 {
953 	struct dw_mci_slot *slot = mmc_priv(mmc);
954 	struct mmc_data *data = mrq->data;
955 
956 	if (!slot->host->use_dma || !data)
957 		return;
958 
959 	if (data->host_cookie != COOKIE_UNMAPPED)
960 		dma_unmap_sg(slot->host->dev,
961 			     data->sg,
962 			     data->sg_len,
963 			     mmc_get_dma_dir(data));
964 	data->host_cookie = COOKIE_UNMAPPED;
965 }
966 
967 static int dw_mci_get_cd(struct mmc_host *mmc)
968 {
969 	int present;
970 	struct dw_mci_slot *slot = mmc_priv(mmc);
971 	struct dw_mci *host = slot->host;
972 	int gpio_cd = mmc_gpio_get_cd(mmc);
973 
974 	/* Use platform get_cd function, else try onboard card detect */
975 	if (((mmc->caps & MMC_CAP_NEEDS_POLL)
976 				|| !mmc_card_is_removable(mmc))) {
977 		present = 1;
978 
979 		if (!test_bit(DW_MMC_CARD_PRESENT, &slot->flags)) {
980 			if (mmc->caps & MMC_CAP_NEEDS_POLL) {
981 				dev_info(&mmc->class_dev,
982 					"card is polling.\n");
983 			} else {
984 				dev_info(&mmc->class_dev,
985 					"card is non-removable.\n");
986 			}
987 			set_bit(DW_MMC_CARD_PRESENT, &slot->flags);
988 		}
989 
990 		return present;
991 	} else if (gpio_cd >= 0)
992 		present = gpio_cd;
993 	else
994 		present = (mci_readl(slot->host, CDETECT) & (1 << slot->id))
995 			== 0 ? 1 : 0;
996 
997 	spin_lock_bh(&host->lock);
998 	if (present && !test_and_set_bit(DW_MMC_CARD_PRESENT, &slot->flags))
999 		dev_dbg(&mmc->class_dev, "card is present\n");
1000 	else if (!present &&
1001 			!test_and_clear_bit(DW_MMC_CARD_PRESENT, &slot->flags))
1002 		dev_dbg(&mmc->class_dev, "card is not present\n");
1003 	spin_unlock_bh(&host->lock);
1004 
1005 	return present;
1006 }
1007 
1008 static void dw_mci_adjust_fifoth(struct dw_mci *host, struct mmc_data *data)
1009 {
1010 	unsigned int blksz = data->blksz;
1011 	static const u32 mszs[] = {1, 4, 8, 16, 32, 64, 128, 256};
1012 	u32 fifo_width = 1 << host->data_shift;
1013 	u32 blksz_depth = blksz / fifo_width, fifoth_val;
1014 	u32 msize = 0, rx_wmark = 1, tx_wmark, tx_wmark_invers;
1015 	int idx = ARRAY_SIZE(mszs) - 1;
1016 
1017 	/* pio should ship this scenario */
1018 	if (!host->use_dma)
1019 		return;
1020 
1021 	tx_wmark = (host->fifo_depth) / 2;
1022 	tx_wmark_invers = host->fifo_depth - tx_wmark;
1023 
1024 	/*
1025 	 * MSIZE is '1',
1026 	 * if blksz is not a multiple of the FIFO width
1027 	 */
1028 	if (blksz % fifo_width)
1029 		goto done;
1030 
1031 	do {
1032 		if (!((blksz_depth % mszs[idx]) ||
1033 		     (tx_wmark_invers % mszs[idx]))) {
1034 			msize = idx;
1035 			rx_wmark = mszs[idx] - 1;
1036 			break;
1037 		}
1038 	} while (--idx > 0);
1039 	/*
1040 	 * If idx is '0', it won't be tried
1041 	 * Thus, initial values are uesed
1042 	 */
1043 done:
1044 	fifoth_val = SDMMC_SET_FIFOTH(msize, rx_wmark, tx_wmark);
1045 	mci_writel(host, FIFOTH, fifoth_val);
1046 }
1047 
1048 static void dw_mci_ctrl_thld(struct dw_mci *host, struct mmc_data *data)
1049 {
1050 	unsigned int blksz = data->blksz;
1051 	u32 blksz_depth, fifo_depth;
1052 	u16 thld_size;
1053 	u8 enable;
1054 
1055 	/*
1056 	 * CDTHRCTL doesn't exist prior to 240A (in fact that register offset is
1057 	 * in the FIFO region, so we really shouldn't access it).
1058 	 */
1059 	if (host->verid < DW_MMC_240A ||
1060 		(host->verid < DW_MMC_280A && data->flags & MMC_DATA_WRITE))
1061 		return;
1062 
1063 	/*
1064 	 * Card write Threshold is introduced since 2.80a
1065 	 * It's used when HS400 mode is enabled.
1066 	 */
1067 	if (data->flags & MMC_DATA_WRITE &&
1068 		host->timing != MMC_TIMING_MMC_HS400)
1069 		goto disable;
1070 
1071 	if (data->flags & MMC_DATA_WRITE)
1072 		enable = SDMMC_CARD_WR_THR_EN;
1073 	else
1074 		enable = SDMMC_CARD_RD_THR_EN;
1075 
1076 	if (host->timing != MMC_TIMING_MMC_HS200 &&
1077 	    host->timing != MMC_TIMING_UHS_SDR104 &&
1078 	    host->timing != MMC_TIMING_MMC_HS400)
1079 		goto disable;
1080 
1081 	blksz_depth = blksz / (1 << host->data_shift);
1082 	fifo_depth = host->fifo_depth;
1083 
1084 	if (blksz_depth > fifo_depth)
1085 		goto disable;
1086 
1087 	/*
1088 	 * If (blksz_depth) >= (fifo_depth >> 1), should be 'thld_size <= blksz'
1089 	 * If (blksz_depth) <  (fifo_depth >> 1), should be thld_size = blksz
1090 	 * Currently just choose blksz.
1091 	 */
1092 	thld_size = blksz;
1093 	mci_writel(host, CDTHRCTL, SDMMC_SET_THLD(thld_size, enable));
1094 	return;
1095 
1096 disable:
1097 	mci_writel(host, CDTHRCTL, 0);
1098 }
1099 
1100 static int dw_mci_submit_data_dma(struct dw_mci *host, struct mmc_data *data)
1101 {
1102 	unsigned long irqflags;
1103 	int sg_len;
1104 	u32 temp;
1105 
1106 	host->using_dma = 0;
1107 
1108 	/* If we don't have a channel, we can't do DMA */
1109 	if (!host->use_dma)
1110 		return -ENODEV;
1111 
1112 	sg_len = dw_mci_pre_dma_transfer(host, data, COOKIE_MAPPED);
1113 	if (sg_len < 0) {
1114 		host->dma_ops->stop(host);
1115 		return sg_len;
1116 	}
1117 
1118 	host->using_dma = 1;
1119 
1120 	if (host->use_dma == TRANS_MODE_IDMAC)
1121 		dev_vdbg(host->dev,
1122 			 "sd sg_cpu: %#lx sg_dma: %#lx sg_len: %d\n",
1123 			 (unsigned long)host->sg_cpu,
1124 			 (unsigned long)host->sg_dma,
1125 			 sg_len);
1126 
1127 	/*
1128 	 * Decide the MSIZE and RX/TX Watermark.
1129 	 * If current block size is same with previous size,
1130 	 * no need to update fifoth.
1131 	 */
1132 	if (host->prev_blksz != data->blksz)
1133 		dw_mci_adjust_fifoth(host, data);
1134 
1135 	/* Enable the DMA interface */
1136 	temp = mci_readl(host, CTRL);
1137 	temp |= SDMMC_CTRL_DMA_ENABLE;
1138 	mci_writel(host, CTRL, temp);
1139 
1140 	/* Disable RX/TX IRQs, let DMA handle it */
1141 	spin_lock_irqsave(&host->irq_lock, irqflags);
1142 	temp = mci_readl(host, INTMASK);
1143 	temp  &= ~(SDMMC_INT_RXDR | SDMMC_INT_TXDR);
1144 	mci_writel(host, INTMASK, temp);
1145 	spin_unlock_irqrestore(&host->irq_lock, irqflags);
1146 
1147 	if (host->dma_ops->start(host, sg_len)) {
1148 		host->dma_ops->stop(host);
1149 		/* We can't do DMA, try PIO for this one */
1150 		dev_dbg(host->dev,
1151 			"%s: fall back to PIO mode for current transfer\n",
1152 			__func__);
1153 		return -ENODEV;
1154 	}
1155 
1156 	return 0;
1157 }
1158 
1159 static void dw_mci_submit_data(struct dw_mci *host, struct mmc_data *data)
1160 {
1161 	unsigned long irqflags;
1162 	int flags = SG_MITER_ATOMIC;
1163 	u32 temp;
1164 
1165 	data->error = -EINPROGRESS;
1166 
1167 	WARN_ON(host->data);
1168 	host->sg = NULL;
1169 	host->data = data;
1170 
1171 	if (data->flags & MMC_DATA_READ)
1172 		host->dir_status = DW_MCI_RECV_STATUS;
1173 	else
1174 		host->dir_status = DW_MCI_SEND_STATUS;
1175 
1176 	dw_mci_ctrl_thld(host, data);
1177 
1178 	if (dw_mci_submit_data_dma(host, data)) {
1179 		if (host->data->flags & MMC_DATA_READ)
1180 			flags |= SG_MITER_TO_SG;
1181 		else
1182 			flags |= SG_MITER_FROM_SG;
1183 
1184 		sg_miter_start(&host->sg_miter, data->sg, data->sg_len, flags);
1185 		host->sg = data->sg;
1186 		host->part_buf_start = 0;
1187 		host->part_buf_count = 0;
1188 
1189 		mci_writel(host, RINTSTS, SDMMC_INT_TXDR | SDMMC_INT_RXDR);
1190 
1191 		spin_lock_irqsave(&host->irq_lock, irqflags);
1192 		temp = mci_readl(host, INTMASK);
1193 		temp |= SDMMC_INT_TXDR | SDMMC_INT_RXDR;
1194 		mci_writel(host, INTMASK, temp);
1195 		spin_unlock_irqrestore(&host->irq_lock, irqflags);
1196 
1197 		temp = mci_readl(host, CTRL);
1198 		temp &= ~SDMMC_CTRL_DMA_ENABLE;
1199 		mci_writel(host, CTRL, temp);
1200 
1201 		/*
1202 		 * Use the initial fifoth_val for PIO mode. If wm_algined
1203 		 * is set, we set watermark same as data size.
1204 		 * If next issued data may be transfered by DMA mode,
1205 		 * prev_blksz should be invalidated.
1206 		 */
1207 		if (host->wm_aligned)
1208 			dw_mci_adjust_fifoth(host, data);
1209 		else
1210 			mci_writel(host, FIFOTH, host->fifoth_val);
1211 		host->prev_blksz = 0;
1212 	} else {
1213 		/*
1214 		 * Keep the current block size.
1215 		 * It will be used to decide whether to update
1216 		 * fifoth register next time.
1217 		 */
1218 		host->prev_blksz = data->blksz;
1219 	}
1220 }
1221 
1222 static void dw_mci_setup_bus(struct dw_mci_slot *slot, bool force_clkinit)
1223 {
1224 	struct dw_mci *host = slot->host;
1225 	unsigned int clock = slot->clock;
1226 	u32 div;
1227 	u32 clk_en_a;
1228 	u32 sdmmc_cmd_bits = SDMMC_CMD_UPD_CLK | SDMMC_CMD_PRV_DAT_WAIT;
1229 
1230 	/* We must continue to set bit 28 in CMD until the change is complete */
1231 	if (host->state == STATE_WAITING_CMD11_DONE)
1232 		sdmmc_cmd_bits |= SDMMC_CMD_VOLT_SWITCH;
1233 
1234 	slot->mmc->actual_clock = 0;
1235 
1236 	if (!clock) {
1237 		mci_writel(host, CLKENA, 0);
1238 		mci_send_cmd(slot, sdmmc_cmd_bits, 0);
1239 	} else if (clock != host->current_speed || force_clkinit) {
1240 		div = host->bus_hz / clock;
1241 		if (host->bus_hz % clock && host->bus_hz > clock)
1242 			/*
1243 			 * move the + 1 after the divide to prevent
1244 			 * over-clocking the card.
1245 			 */
1246 			div += 1;
1247 
1248 		div = (host->bus_hz != clock) ? DIV_ROUND_UP(div, 2) : 0;
1249 
1250 		if ((clock != slot->__clk_old &&
1251 			!test_bit(DW_MMC_CARD_NEEDS_POLL, &slot->flags)) ||
1252 			force_clkinit) {
1253 			/* Silent the verbose log if calling from PM context */
1254 			if (!force_clkinit)
1255 				dev_info(&slot->mmc->class_dev,
1256 					 "Bus speed (slot %d) = %dHz (slot req %dHz, actual %dHZ div = %d)\n",
1257 					 slot->id, host->bus_hz, clock,
1258 					 div ? ((host->bus_hz / div) >> 1) :
1259 					 host->bus_hz, div);
1260 
1261 			/*
1262 			 * If card is polling, display the message only
1263 			 * one time at boot time.
1264 			 */
1265 			if (slot->mmc->caps & MMC_CAP_NEEDS_POLL &&
1266 					slot->mmc->f_min == clock)
1267 				set_bit(DW_MMC_CARD_NEEDS_POLL, &slot->flags);
1268 		}
1269 
1270 		/* disable clock */
1271 		mci_writel(host, CLKENA, 0);
1272 		mci_writel(host, CLKSRC, 0);
1273 
1274 		/* inform CIU */
1275 		mci_send_cmd(slot, sdmmc_cmd_bits, 0);
1276 
1277 		/* set clock to desired speed */
1278 		mci_writel(host, CLKDIV, div);
1279 
1280 		/* inform CIU */
1281 		mci_send_cmd(slot, sdmmc_cmd_bits, 0);
1282 
1283 		/* enable clock; only low power if no SDIO */
1284 		clk_en_a = SDMMC_CLKEN_ENABLE << slot->id;
1285 		if (!test_bit(DW_MMC_CARD_NO_LOW_PWR, &slot->flags))
1286 			clk_en_a |= SDMMC_CLKEN_LOW_PWR << slot->id;
1287 		mci_writel(host, CLKENA, clk_en_a);
1288 
1289 		/* inform CIU */
1290 		mci_send_cmd(slot, sdmmc_cmd_bits, 0);
1291 
1292 		/* keep the last clock value that was requested from core */
1293 		slot->__clk_old = clock;
1294 		slot->mmc->actual_clock = div ? ((host->bus_hz / div) >> 1) :
1295 					  host->bus_hz;
1296 	}
1297 
1298 	host->current_speed = clock;
1299 
1300 	/* Set the current slot bus width */
1301 	mci_writel(host, CTYPE, (slot->ctype << slot->id));
1302 }
1303 
1304 static void __dw_mci_start_request(struct dw_mci *host,
1305 				   struct dw_mci_slot *slot,
1306 				   struct mmc_command *cmd)
1307 {
1308 	struct mmc_request *mrq;
1309 	struct mmc_data	*data;
1310 	u32 cmdflags;
1311 
1312 	mrq = slot->mrq;
1313 
1314 	host->mrq = mrq;
1315 
1316 	host->pending_events = 0;
1317 	host->completed_events = 0;
1318 	host->cmd_status = 0;
1319 	host->data_status = 0;
1320 	host->dir_status = 0;
1321 
1322 	data = cmd->data;
1323 	if (data) {
1324 		mci_writel(host, TMOUT, 0xFFFFFFFF);
1325 		mci_writel(host, BYTCNT, data->blksz*data->blocks);
1326 		mci_writel(host, BLKSIZ, data->blksz);
1327 	}
1328 
1329 	cmdflags = dw_mci_prepare_command(slot->mmc, cmd);
1330 
1331 	/* this is the first command, send the initialization clock */
1332 	if (test_and_clear_bit(DW_MMC_CARD_NEED_INIT, &slot->flags))
1333 		cmdflags |= SDMMC_CMD_INIT;
1334 
1335 	if (data) {
1336 		dw_mci_submit_data(host, data);
1337 		wmb(); /* drain writebuffer */
1338 	}
1339 
1340 	dw_mci_start_command(host, cmd, cmdflags);
1341 
1342 	if (cmd->opcode == SD_SWITCH_VOLTAGE) {
1343 		unsigned long irqflags;
1344 
1345 		/*
1346 		 * Databook says to fail after 2ms w/ no response, but evidence
1347 		 * shows that sometimes the cmd11 interrupt takes over 130ms.
1348 		 * We'll set to 500ms, plus an extra jiffy just in case jiffies
1349 		 * is just about to roll over.
1350 		 *
1351 		 * We do this whole thing under spinlock and only if the
1352 		 * command hasn't already completed (indicating the the irq
1353 		 * already ran so we don't want the timeout).
1354 		 */
1355 		spin_lock_irqsave(&host->irq_lock, irqflags);
1356 		if (!test_bit(EVENT_CMD_COMPLETE, &host->pending_events))
1357 			mod_timer(&host->cmd11_timer,
1358 				jiffies + msecs_to_jiffies(500) + 1);
1359 		spin_unlock_irqrestore(&host->irq_lock, irqflags);
1360 	}
1361 
1362 	host->stop_cmdr = dw_mci_prep_stop_abort(host, cmd);
1363 }
1364 
1365 static void dw_mci_start_request(struct dw_mci *host,
1366 				 struct dw_mci_slot *slot)
1367 {
1368 	struct mmc_request *mrq = slot->mrq;
1369 	struct mmc_command *cmd;
1370 
1371 	cmd = mrq->sbc ? mrq->sbc : mrq->cmd;
1372 	__dw_mci_start_request(host, slot, cmd);
1373 }
1374 
1375 /* must be called with host->lock held */
1376 static void dw_mci_queue_request(struct dw_mci *host, struct dw_mci_slot *slot,
1377 				 struct mmc_request *mrq)
1378 {
1379 	dev_vdbg(&slot->mmc->class_dev, "queue request: state=%d\n",
1380 		 host->state);
1381 
1382 	slot->mrq = mrq;
1383 
1384 	if (host->state == STATE_WAITING_CMD11_DONE) {
1385 		dev_warn(&slot->mmc->class_dev,
1386 			 "Voltage change didn't complete\n");
1387 		/*
1388 		 * this case isn't expected to happen, so we can
1389 		 * either crash here or just try to continue on
1390 		 * in the closest possible state
1391 		 */
1392 		host->state = STATE_IDLE;
1393 	}
1394 
1395 	if (host->state == STATE_IDLE) {
1396 		host->state = STATE_SENDING_CMD;
1397 		dw_mci_start_request(host, slot);
1398 	} else {
1399 		list_add_tail(&slot->queue_node, &host->queue);
1400 	}
1401 }
1402 
1403 static void dw_mci_request(struct mmc_host *mmc, struct mmc_request *mrq)
1404 {
1405 	struct dw_mci_slot *slot = mmc_priv(mmc);
1406 	struct dw_mci *host = slot->host;
1407 
1408 	WARN_ON(slot->mrq);
1409 
1410 	/*
1411 	 * The check for card presence and queueing of the request must be
1412 	 * atomic, otherwise the card could be removed in between and the
1413 	 * request wouldn't fail until another card was inserted.
1414 	 */
1415 
1416 	if (!dw_mci_get_cd(mmc)) {
1417 		mrq->cmd->error = -ENOMEDIUM;
1418 		mmc_request_done(mmc, mrq);
1419 		return;
1420 	}
1421 
1422 	spin_lock_bh(&host->lock);
1423 
1424 	dw_mci_queue_request(host, slot, mrq);
1425 
1426 	spin_unlock_bh(&host->lock);
1427 }
1428 
1429 static void dw_mci_set_ios(struct mmc_host *mmc, struct mmc_ios *ios)
1430 {
1431 	struct dw_mci_slot *slot = mmc_priv(mmc);
1432 	const struct dw_mci_drv_data *drv_data = slot->host->drv_data;
1433 	u32 regs;
1434 	int ret;
1435 
1436 	switch (ios->bus_width) {
1437 	case MMC_BUS_WIDTH_4:
1438 		slot->ctype = SDMMC_CTYPE_4BIT;
1439 		break;
1440 	case MMC_BUS_WIDTH_8:
1441 		slot->ctype = SDMMC_CTYPE_8BIT;
1442 		break;
1443 	default:
1444 		/* set default 1 bit mode */
1445 		slot->ctype = SDMMC_CTYPE_1BIT;
1446 	}
1447 
1448 	regs = mci_readl(slot->host, UHS_REG);
1449 
1450 	/* DDR mode set */
1451 	if (ios->timing == MMC_TIMING_MMC_DDR52 ||
1452 	    ios->timing == MMC_TIMING_UHS_DDR50 ||
1453 	    ios->timing == MMC_TIMING_MMC_HS400)
1454 		regs |= ((0x1 << slot->id) << 16);
1455 	else
1456 		regs &= ~((0x1 << slot->id) << 16);
1457 
1458 	mci_writel(slot->host, UHS_REG, regs);
1459 	slot->host->timing = ios->timing;
1460 
1461 	/*
1462 	 * Use mirror of ios->clock to prevent race with mmc
1463 	 * core ios update when finding the minimum.
1464 	 */
1465 	slot->clock = ios->clock;
1466 
1467 	if (drv_data && drv_data->set_ios)
1468 		drv_data->set_ios(slot->host, ios);
1469 
1470 	switch (ios->power_mode) {
1471 	case MMC_POWER_UP:
1472 		if (!IS_ERR(mmc->supply.vmmc)) {
1473 			ret = mmc_regulator_set_ocr(mmc, mmc->supply.vmmc,
1474 					ios->vdd);
1475 			if (ret) {
1476 				dev_err(slot->host->dev,
1477 					"failed to enable vmmc regulator\n");
1478 				/*return, if failed turn on vmmc*/
1479 				return;
1480 			}
1481 		}
1482 		set_bit(DW_MMC_CARD_NEED_INIT, &slot->flags);
1483 		regs = mci_readl(slot->host, PWREN);
1484 		regs |= (1 << slot->id);
1485 		mci_writel(slot->host, PWREN, regs);
1486 		break;
1487 	case MMC_POWER_ON:
1488 		if (!slot->host->vqmmc_enabled) {
1489 			if (!IS_ERR(mmc->supply.vqmmc)) {
1490 				ret = regulator_enable(mmc->supply.vqmmc);
1491 				if (ret < 0)
1492 					dev_err(slot->host->dev,
1493 						"failed to enable vqmmc\n");
1494 				else
1495 					slot->host->vqmmc_enabled = true;
1496 
1497 			} else {
1498 				/* Keep track so we don't reset again */
1499 				slot->host->vqmmc_enabled = true;
1500 			}
1501 
1502 			/* Reset our state machine after powering on */
1503 			dw_mci_ctrl_reset(slot->host,
1504 					  SDMMC_CTRL_ALL_RESET_FLAGS);
1505 		}
1506 
1507 		/* Adjust clock / bus width after power is up */
1508 		dw_mci_setup_bus(slot, false);
1509 
1510 		break;
1511 	case MMC_POWER_OFF:
1512 		/* Turn clock off before power goes down */
1513 		dw_mci_setup_bus(slot, false);
1514 
1515 		if (!IS_ERR(mmc->supply.vmmc))
1516 			mmc_regulator_set_ocr(mmc, mmc->supply.vmmc, 0);
1517 
1518 		if (!IS_ERR(mmc->supply.vqmmc) && slot->host->vqmmc_enabled)
1519 			regulator_disable(mmc->supply.vqmmc);
1520 		slot->host->vqmmc_enabled = false;
1521 
1522 		regs = mci_readl(slot->host, PWREN);
1523 		regs &= ~(1 << slot->id);
1524 		mci_writel(slot->host, PWREN, regs);
1525 		break;
1526 	default:
1527 		break;
1528 	}
1529 
1530 	if (slot->host->state == STATE_WAITING_CMD11_DONE && ios->clock != 0)
1531 		slot->host->state = STATE_IDLE;
1532 }
1533 
1534 static int dw_mci_card_busy(struct mmc_host *mmc)
1535 {
1536 	struct dw_mci_slot *slot = mmc_priv(mmc);
1537 	u32 status;
1538 
1539 	/*
1540 	 * Check the busy bit which is low when DAT[3:0]
1541 	 * (the data lines) are 0000
1542 	 */
1543 	status = mci_readl(slot->host, STATUS);
1544 
1545 	return !!(status & SDMMC_STATUS_BUSY);
1546 }
1547 
1548 static int dw_mci_switch_voltage(struct mmc_host *mmc, struct mmc_ios *ios)
1549 {
1550 	struct dw_mci_slot *slot = mmc_priv(mmc);
1551 	struct dw_mci *host = slot->host;
1552 	const struct dw_mci_drv_data *drv_data = host->drv_data;
1553 	u32 uhs;
1554 	u32 v18 = SDMMC_UHS_18V << slot->id;
1555 	int ret;
1556 
1557 	if (drv_data && drv_data->switch_voltage)
1558 		return drv_data->switch_voltage(mmc, ios);
1559 
1560 	/*
1561 	 * Program the voltage.  Note that some instances of dw_mmc may use
1562 	 * the UHS_REG for this.  For other instances (like exynos) the UHS_REG
1563 	 * does no harm but you need to set the regulator directly.  Try both.
1564 	 */
1565 	uhs = mci_readl(host, UHS_REG);
1566 	if (ios->signal_voltage == MMC_SIGNAL_VOLTAGE_330)
1567 		uhs &= ~v18;
1568 	else
1569 		uhs |= v18;
1570 
1571 	if (!IS_ERR(mmc->supply.vqmmc)) {
1572 		ret = mmc_regulator_set_vqmmc(mmc, ios);
1573 
1574 		if (ret) {
1575 			dev_dbg(&mmc->class_dev,
1576 					 "Regulator set error %d - %s V\n",
1577 					 ret, uhs & v18 ? "1.8" : "3.3");
1578 			return ret;
1579 		}
1580 	}
1581 	mci_writel(host, UHS_REG, uhs);
1582 
1583 	return 0;
1584 }
1585 
1586 static int dw_mci_get_ro(struct mmc_host *mmc)
1587 {
1588 	int read_only;
1589 	struct dw_mci_slot *slot = mmc_priv(mmc);
1590 	int gpio_ro = mmc_gpio_get_ro(mmc);
1591 
1592 	/* Use platform get_ro function, else try on board write protect */
1593 	if (gpio_ro >= 0)
1594 		read_only = gpio_ro;
1595 	else
1596 		read_only =
1597 			mci_readl(slot->host, WRTPRT) & (1 << slot->id) ? 1 : 0;
1598 
1599 	dev_dbg(&mmc->class_dev, "card is %s\n",
1600 		read_only ? "read-only" : "read-write");
1601 
1602 	return read_only;
1603 }
1604 
1605 static void dw_mci_hw_reset(struct mmc_host *mmc)
1606 {
1607 	struct dw_mci_slot *slot = mmc_priv(mmc);
1608 	struct dw_mci *host = slot->host;
1609 	int reset;
1610 
1611 	if (host->use_dma == TRANS_MODE_IDMAC)
1612 		dw_mci_idmac_reset(host);
1613 
1614 	if (!dw_mci_ctrl_reset(host, SDMMC_CTRL_DMA_RESET |
1615 				     SDMMC_CTRL_FIFO_RESET))
1616 		return;
1617 
1618 	/*
1619 	 * According to eMMC spec, card reset procedure:
1620 	 * tRstW >= 1us:   RST_n pulse width
1621 	 * tRSCA >= 200us: RST_n to Command time
1622 	 * tRSTH >= 1us:   RST_n high period
1623 	 */
1624 	reset = mci_readl(host, RST_N);
1625 	reset &= ~(SDMMC_RST_HWACTIVE << slot->id);
1626 	mci_writel(host, RST_N, reset);
1627 	usleep_range(1, 2);
1628 	reset |= SDMMC_RST_HWACTIVE << slot->id;
1629 	mci_writel(host, RST_N, reset);
1630 	usleep_range(200, 300);
1631 }
1632 
1633 static void dw_mci_init_card(struct mmc_host *mmc, struct mmc_card *card)
1634 {
1635 	struct dw_mci_slot *slot = mmc_priv(mmc);
1636 	struct dw_mci *host = slot->host;
1637 
1638 	/*
1639 	 * Low power mode will stop the card clock when idle.  According to the
1640 	 * description of the CLKENA register we should disable low power mode
1641 	 * for SDIO cards if we need SDIO interrupts to work.
1642 	 */
1643 	if (mmc->caps & MMC_CAP_SDIO_IRQ) {
1644 		const u32 clken_low_pwr = SDMMC_CLKEN_LOW_PWR << slot->id;
1645 		u32 clk_en_a_old;
1646 		u32 clk_en_a;
1647 
1648 		clk_en_a_old = mci_readl(host, CLKENA);
1649 
1650 		if (card->type == MMC_TYPE_SDIO ||
1651 		    card->type == MMC_TYPE_SD_COMBO) {
1652 			set_bit(DW_MMC_CARD_NO_LOW_PWR, &slot->flags);
1653 			clk_en_a = clk_en_a_old & ~clken_low_pwr;
1654 		} else {
1655 			clear_bit(DW_MMC_CARD_NO_LOW_PWR, &slot->flags);
1656 			clk_en_a = clk_en_a_old | clken_low_pwr;
1657 		}
1658 
1659 		if (clk_en_a != clk_en_a_old) {
1660 			mci_writel(host, CLKENA, clk_en_a);
1661 			mci_send_cmd(slot, SDMMC_CMD_UPD_CLK |
1662 				     SDMMC_CMD_PRV_DAT_WAIT, 0);
1663 		}
1664 	}
1665 }
1666 
1667 static void __dw_mci_enable_sdio_irq(struct dw_mci_slot *slot, int enb)
1668 {
1669 	struct dw_mci *host = slot->host;
1670 	unsigned long irqflags;
1671 	u32 int_mask;
1672 
1673 	spin_lock_irqsave(&host->irq_lock, irqflags);
1674 
1675 	/* Enable/disable Slot Specific SDIO interrupt */
1676 	int_mask = mci_readl(host, INTMASK);
1677 	if (enb)
1678 		int_mask |= SDMMC_INT_SDIO(slot->sdio_id);
1679 	else
1680 		int_mask &= ~SDMMC_INT_SDIO(slot->sdio_id);
1681 	mci_writel(host, INTMASK, int_mask);
1682 
1683 	spin_unlock_irqrestore(&host->irq_lock, irqflags);
1684 }
1685 
1686 static void dw_mci_enable_sdio_irq(struct mmc_host *mmc, int enb)
1687 {
1688 	struct dw_mci_slot *slot = mmc_priv(mmc);
1689 	struct dw_mci *host = slot->host;
1690 
1691 	__dw_mci_enable_sdio_irq(slot, enb);
1692 
1693 	/* Avoid runtime suspending the device when SDIO IRQ is enabled */
1694 	if (enb)
1695 		pm_runtime_get_noresume(host->dev);
1696 	else
1697 		pm_runtime_put_noidle(host->dev);
1698 }
1699 
1700 static void dw_mci_ack_sdio_irq(struct mmc_host *mmc)
1701 {
1702 	struct dw_mci_slot *slot = mmc_priv(mmc);
1703 
1704 	__dw_mci_enable_sdio_irq(slot, 1);
1705 }
1706 
1707 static int dw_mci_execute_tuning(struct mmc_host *mmc, u32 opcode)
1708 {
1709 	struct dw_mci_slot *slot = mmc_priv(mmc);
1710 	struct dw_mci *host = slot->host;
1711 	const struct dw_mci_drv_data *drv_data = host->drv_data;
1712 	int err = -EINVAL;
1713 
1714 	if (drv_data && drv_data->execute_tuning)
1715 		err = drv_data->execute_tuning(slot, opcode);
1716 	return err;
1717 }
1718 
1719 static int dw_mci_prepare_hs400_tuning(struct mmc_host *mmc,
1720 				       struct mmc_ios *ios)
1721 {
1722 	struct dw_mci_slot *slot = mmc_priv(mmc);
1723 	struct dw_mci *host = slot->host;
1724 	const struct dw_mci_drv_data *drv_data = host->drv_data;
1725 
1726 	if (drv_data && drv_data->prepare_hs400_tuning)
1727 		return drv_data->prepare_hs400_tuning(host, ios);
1728 
1729 	return 0;
1730 }
1731 
1732 static bool dw_mci_reset(struct dw_mci *host)
1733 {
1734 	u32 flags = SDMMC_CTRL_RESET | SDMMC_CTRL_FIFO_RESET;
1735 	bool ret = false;
1736 	u32 status = 0;
1737 
1738 	/*
1739 	 * Resetting generates a block interrupt, hence setting
1740 	 * the scatter-gather pointer to NULL.
1741 	 */
1742 	if (host->sg) {
1743 		sg_miter_stop(&host->sg_miter);
1744 		host->sg = NULL;
1745 	}
1746 
1747 	if (host->use_dma)
1748 		flags |= SDMMC_CTRL_DMA_RESET;
1749 
1750 	if (dw_mci_ctrl_reset(host, flags)) {
1751 		/*
1752 		 * In all cases we clear the RAWINTS
1753 		 * register to clear any interrupts.
1754 		 */
1755 		mci_writel(host, RINTSTS, 0xFFFFFFFF);
1756 
1757 		if (!host->use_dma) {
1758 			ret = true;
1759 			goto ciu_out;
1760 		}
1761 
1762 		/* Wait for dma_req to be cleared */
1763 		if (readl_poll_timeout_atomic(host->regs + SDMMC_STATUS,
1764 					      status,
1765 					      !(status & SDMMC_STATUS_DMA_REQ),
1766 					      1, 500 * USEC_PER_MSEC)) {
1767 			dev_err(host->dev,
1768 				"%s: Timeout waiting for dma_req to be cleared\n",
1769 				__func__);
1770 			goto ciu_out;
1771 		}
1772 
1773 		/* when using DMA next we reset the fifo again */
1774 		if (!dw_mci_ctrl_reset(host, SDMMC_CTRL_FIFO_RESET))
1775 			goto ciu_out;
1776 	} else {
1777 		/* if the controller reset bit did clear, then set clock regs */
1778 		if (!(mci_readl(host, CTRL) & SDMMC_CTRL_RESET)) {
1779 			dev_err(host->dev,
1780 				"%s: fifo/dma reset bits didn't clear but ciu was reset, doing clock update\n",
1781 				__func__);
1782 			goto ciu_out;
1783 		}
1784 	}
1785 
1786 	if (host->use_dma == TRANS_MODE_IDMAC)
1787 		/* It is also required that we reinit idmac */
1788 		dw_mci_idmac_init(host);
1789 
1790 	ret = true;
1791 
1792 ciu_out:
1793 	/* After a CTRL reset we need to have CIU set clock registers  */
1794 	mci_send_cmd(host->slot, SDMMC_CMD_UPD_CLK, 0);
1795 
1796 	return ret;
1797 }
1798 
1799 static const struct mmc_host_ops dw_mci_ops = {
1800 	.request		= dw_mci_request,
1801 	.pre_req		= dw_mci_pre_req,
1802 	.post_req		= dw_mci_post_req,
1803 	.set_ios		= dw_mci_set_ios,
1804 	.get_ro			= dw_mci_get_ro,
1805 	.get_cd			= dw_mci_get_cd,
1806 	.hw_reset               = dw_mci_hw_reset,
1807 	.enable_sdio_irq	= dw_mci_enable_sdio_irq,
1808 	.ack_sdio_irq		= dw_mci_ack_sdio_irq,
1809 	.execute_tuning		= dw_mci_execute_tuning,
1810 	.card_busy		= dw_mci_card_busy,
1811 	.start_signal_voltage_switch = dw_mci_switch_voltage,
1812 	.init_card		= dw_mci_init_card,
1813 	.prepare_hs400_tuning	= dw_mci_prepare_hs400_tuning,
1814 };
1815 
1816 static void dw_mci_request_end(struct dw_mci *host, struct mmc_request *mrq)
1817 	__releases(&host->lock)
1818 	__acquires(&host->lock)
1819 {
1820 	struct dw_mci_slot *slot;
1821 	struct mmc_host	*prev_mmc = host->slot->mmc;
1822 
1823 	WARN_ON(host->cmd || host->data);
1824 
1825 	host->slot->mrq = NULL;
1826 	host->mrq = NULL;
1827 	if (!list_empty(&host->queue)) {
1828 		slot = list_entry(host->queue.next,
1829 				  struct dw_mci_slot, queue_node);
1830 		list_del(&slot->queue_node);
1831 		dev_vdbg(host->dev, "list not empty: %s is next\n",
1832 			 mmc_hostname(slot->mmc));
1833 		host->state = STATE_SENDING_CMD;
1834 		dw_mci_start_request(host, slot);
1835 	} else {
1836 		dev_vdbg(host->dev, "list empty\n");
1837 
1838 		if (host->state == STATE_SENDING_CMD11)
1839 			host->state = STATE_WAITING_CMD11_DONE;
1840 		else
1841 			host->state = STATE_IDLE;
1842 	}
1843 
1844 	spin_unlock(&host->lock);
1845 	mmc_request_done(prev_mmc, mrq);
1846 	spin_lock(&host->lock);
1847 }
1848 
1849 static int dw_mci_command_complete(struct dw_mci *host, struct mmc_command *cmd)
1850 {
1851 	u32 status = host->cmd_status;
1852 
1853 	host->cmd_status = 0;
1854 
1855 	/* Read the response from the card (up to 16 bytes) */
1856 	if (cmd->flags & MMC_RSP_PRESENT) {
1857 		if (cmd->flags & MMC_RSP_136) {
1858 			cmd->resp[3] = mci_readl(host, RESP0);
1859 			cmd->resp[2] = mci_readl(host, RESP1);
1860 			cmd->resp[1] = mci_readl(host, RESP2);
1861 			cmd->resp[0] = mci_readl(host, RESP3);
1862 		} else {
1863 			cmd->resp[0] = mci_readl(host, RESP0);
1864 			cmd->resp[1] = 0;
1865 			cmd->resp[2] = 0;
1866 			cmd->resp[3] = 0;
1867 		}
1868 	}
1869 
1870 	if (status & SDMMC_INT_RTO)
1871 		cmd->error = -ETIMEDOUT;
1872 	else if ((cmd->flags & MMC_RSP_CRC) && (status & SDMMC_INT_RCRC))
1873 		cmd->error = -EILSEQ;
1874 	else if (status & SDMMC_INT_RESP_ERR)
1875 		cmd->error = -EIO;
1876 	else
1877 		cmd->error = 0;
1878 
1879 	return cmd->error;
1880 }
1881 
1882 static int dw_mci_data_complete(struct dw_mci *host, struct mmc_data *data)
1883 {
1884 	u32 status = host->data_status;
1885 
1886 	if (status & DW_MCI_DATA_ERROR_FLAGS) {
1887 		if (status & SDMMC_INT_DRTO) {
1888 			data->error = -ETIMEDOUT;
1889 		} else if (status & SDMMC_INT_DCRC) {
1890 			data->error = -EILSEQ;
1891 		} else if (status & SDMMC_INT_EBE) {
1892 			if (host->dir_status ==
1893 				DW_MCI_SEND_STATUS) {
1894 				/*
1895 				 * No data CRC status was returned.
1896 				 * The number of bytes transferred
1897 				 * will be exaggerated in PIO mode.
1898 				 */
1899 				data->bytes_xfered = 0;
1900 				data->error = -ETIMEDOUT;
1901 			} else if (host->dir_status ==
1902 					DW_MCI_RECV_STATUS) {
1903 				data->error = -EILSEQ;
1904 			}
1905 		} else {
1906 			/* SDMMC_INT_SBE is included */
1907 			data->error = -EILSEQ;
1908 		}
1909 
1910 		dev_dbg(host->dev, "data error, status 0x%08x\n", status);
1911 
1912 		/*
1913 		 * After an error, there may be data lingering
1914 		 * in the FIFO
1915 		 */
1916 		dw_mci_reset(host);
1917 	} else {
1918 		data->bytes_xfered = data->blocks * data->blksz;
1919 		data->error = 0;
1920 	}
1921 
1922 	return data->error;
1923 }
1924 
1925 static void dw_mci_set_drto(struct dw_mci *host)
1926 {
1927 	unsigned int drto_clks;
1928 	unsigned int drto_div;
1929 	unsigned int drto_ms;
1930 	unsigned long irqflags;
1931 
1932 	drto_clks = mci_readl(host, TMOUT) >> 8;
1933 	drto_div = (mci_readl(host, CLKDIV) & 0xff) * 2;
1934 	if (drto_div == 0)
1935 		drto_div = 1;
1936 
1937 	drto_ms = DIV_ROUND_UP_ULL((u64)MSEC_PER_SEC * drto_clks * drto_div,
1938 				   host->bus_hz);
1939 
1940 	/* add a bit spare time */
1941 	drto_ms += 10;
1942 
1943 	spin_lock_irqsave(&host->irq_lock, irqflags);
1944 	if (!test_bit(EVENT_DATA_COMPLETE, &host->pending_events))
1945 		mod_timer(&host->dto_timer,
1946 			  jiffies + msecs_to_jiffies(drto_ms));
1947 	spin_unlock_irqrestore(&host->irq_lock, irqflags);
1948 }
1949 
1950 static bool dw_mci_clear_pending_cmd_complete(struct dw_mci *host)
1951 {
1952 	if (!test_bit(EVENT_CMD_COMPLETE, &host->pending_events))
1953 		return false;
1954 
1955 	/*
1956 	 * Really be certain that the timer has stopped.  This is a bit of
1957 	 * paranoia and could only really happen if we had really bad
1958 	 * interrupt latency and the interrupt routine and timeout were
1959 	 * running concurrently so that the del_timer() in the interrupt
1960 	 * handler couldn't run.
1961 	 */
1962 	WARN_ON(del_timer_sync(&host->cto_timer));
1963 	clear_bit(EVENT_CMD_COMPLETE, &host->pending_events);
1964 
1965 	return true;
1966 }
1967 
1968 static bool dw_mci_clear_pending_data_complete(struct dw_mci *host)
1969 {
1970 	if (!test_bit(EVENT_DATA_COMPLETE, &host->pending_events))
1971 		return false;
1972 
1973 	/* Extra paranoia just like dw_mci_clear_pending_cmd_complete() */
1974 	WARN_ON(del_timer_sync(&host->dto_timer));
1975 	clear_bit(EVENT_DATA_COMPLETE, &host->pending_events);
1976 
1977 	return true;
1978 }
1979 
1980 static void dw_mci_tasklet_func(unsigned long priv)
1981 {
1982 	struct dw_mci *host = (struct dw_mci *)priv;
1983 	struct mmc_data	*data;
1984 	struct mmc_command *cmd;
1985 	struct mmc_request *mrq;
1986 	enum dw_mci_state state;
1987 	enum dw_mci_state prev_state;
1988 	unsigned int err;
1989 
1990 	spin_lock(&host->lock);
1991 
1992 	state = host->state;
1993 	data = host->data;
1994 	mrq = host->mrq;
1995 
1996 	do {
1997 		prev_state = state;
1998 
1999 		switch (state) {
2000 		case STATE_IDLE:
2001 		case STATE_WAITING_CMD11_DONE:
2002 			break;
2003 
2004 		case STATE_SENDING_CMD11:
2005 		case STATE_SENDING_CMD:
2006 			if (!dw_mci_clear_pending_cmd_complete(host))
2007 				break;
2008 
2009 			cmd = host->cmd;
2010 			host->cmd = NULL;
2011 			set_bit(EVENT_CMD_COMPLETE, &host->completed_events);
2012 			err = dw_mci_command_complete(host, cmd);
2013 			if (cmd == mrq->sbc && !err) {
2014 				__dw_mci_start_request(host, host->slot,
2015 						       mrq->cmd);
2016 				goto unlock;
2017 			}
2018 
2019 			if (cmd->data && err) {
2020 				/*
2021 				 * During UHS tuning sequence, sending the stop
2022 				 * command after the response CRC error would
2023 				 * throw the system into a confused state
2024 				 * causing all future tuning phases to report
2025 				 * failure.
2026 				 *
2027 				 * In such case controller will move into a data
2028 				 * transfer state after a response error or
2029 				 * response CRC error. Let's let that finish
2030 				 * before trying to send a stop, so we'll go to
2031 				 * STATE_SENDING_DATA.
2032 				 *
2033 				 * Although letting the data transfer take place
2034 				 * will waste a bit of time (we already know
2035 				 * the command was bad), it can't cause any
2036 				 * errors since it's possible it would have
2037 				 * taken place anyway if this tasklet got
2038 				 * delayed. Allowing the transfer to take place
2039 				 * avoids races and keeps things simple.
2040 				 */
2041 				if ((err != -ETIMEDOUT) &&
2042 				    (cmd->opcode == MMC_SEND_TUNING_BLOCK)) {
2043 					state = STATE_SENDING_DATA;
2044 					continue;
2045 				}
2046 
2047 				dw_mci_stop_dma(host);
2048 				send_stop_abort(host, data);
2049 				state = STATE_SENDING_STOP;
2050 				break;
2051 			}
2052 
2053 			if (!cmd->data || err) {
2054 				dw_mci_request_end(host, mrq);
2055 				goto unlock;
2056 			}
2057 
2058 			prev_state = state = STATE_SENDING_DATA;
2059 			/* fall through */
2060 
2061 		case STATE_SENDING_DATA:
2062 			/*
2063 			 * We could get a data error and never a transfer
2064 			 * complete so we'd better check for it here.
2065 			 *
2066 			 * Note that we don't really care if we also got a
2067 			 * transfer complete; stopping the DMA and sending an
2068 			 * abort won't hurt.
2069 			 */
2070 			if (test_and_clear_bit(EVENT_DATA_ERROR,
2071 					       &host->pending_events)) {
2072 				dw_mci_stop_dma(host);
2073 				if (!(host->data_status & (SDMMC_INT_DRTO |
2074 							   SDMMC_INT_EBE)))
2075 					send_stop_abort(host, data);
2076 				state = STATE_DATA_ERROR;
2077 				break;
2078 			}
2079 
2080 			if (!test_and_clear_bit(EVENT_XFER_COMPLETE,
2081 						&host->pending_events)) {
2082 				/*
2083 				 * If all data-related interrupts don't come
2084 				 * within the given time in reading data state.
2085 				 */
2086 				if (host->dir_status == DW_MCI_RECV_STATUS)
2087 					dw_mci_set_drto(host);
2088 				break;
2089 			}
2090 
2091 			set_bit(EVENT_XFER_COMPLETE, &host->completed_events);
2092 
2093 			/*
2094 			 * Handle an EVENT_DATA_ERROR that might have shown up
2095 			 * before the transfer completed.  This might not have
2096 			 * been caught by the check above because the interrupt
2097 			 * could have gone off between the previous check and
2098 			 * the check for transfer complete.
2099 			 *
2100 			 * Technically this ought not be needed assuming we
2101 			 * get a DATA_COMPLETE eventually (we'll notice the
2102 			 * error and end the request), but it shouldn't hurt.
2103 			 *
2104 			 * This has the advantage of sending the stop command.
2105 			 */
2106 			if (test_and_clear_bit(EVENT_DATA_ERROR,
2107 					       &host->pending_events)) {
2108 				dw_mci_stop_dma(host);
2109 				if (!(host->data_status & (SDMMC_INT_DRTO |
2110 							   SDMMC_INT_EBE)))
2111 					send_stop_abort(host, data);
2112 				state = STATE_DATA_ERROR;
2113 				break;
2114 			}
2115 			prev_state = state = STATE_DATA_BUSY;
2116 
2117 			/* fall through */
2118 
2119 		case STATE_DATA_BUSY:
2120 			if (!dw_mci_clear_pending_data_complete(host)) {
2121 				/*
2122 				 * If data error interrupt comes but data over
2123 				 * interrupt doesn't come within the given time.
2124 				 * in reading data state.
2125 				 */
2126 				if (host->dir_status == DW_MCI_RECV_STATUS)
2127 					dw_mci_set_drto(host);
2128 				break;
2129 			}
2130 
2131 			host->data = NULL;
2132 			set_bit(EVENT_DATA_COMPLETE, &host->completed_events);
2133 			err = dw_mci_data_complete(host, data);
2134 
2135 			if (!err) {
2136 				if (!data->stop || mrq->sbc) {
2137 					if (mrq->sbc && data->stop)
2138 						data->stop->error = 0;
2139 					dw_mci_request_end(host, mrq);
2140 					goto unlock;
2141 				}
2142 
2143 				/* stop command for open-ended transfer*/
2144 				if (data->stop)
2145 					send_stop_abort(host, data);
2146 			} else {
2147 				/*
2148 				 * If we don't have a command complete now we'll
2149 				 * never get one since we just reset everything;
2150 				 * better end the request.
2151 				 *
2152 				 * If we do have a command complete we'll fall
2153 				 * through to the SENDING_STOP command and
2154 				 * everything will be peachy keen.
2155 				 */
2156 				if (!test_bit(EVENT_CMD_COMPLETE,
2157 					      &host->pending_events)) {
2158 					host->cmd = NULL;
2159 					dw_mci_request_end(host, mrq);
2160 					goto unlock;
2161 				}
2162 			}
2163 
2164 			/*
2165 			 * If err has non-zero,
2166 			 * stop-abort command has been already issued.
2167 			 */
2168 			prev_state = state = STATE_SENDING_STOP;
2169 
2170 			/* fall through */
2171 
2172 		case STATE_SENDING_STOP:
2173 			if (!dw_mci_clear_pending_cmd_complete(host))
2174 				break;
2175 
2176 			/* CMD error in data command */
2177 			if (mrq->cmd->error && mrq->data)
2178 				dw_mci_reset(host);
2179 
2180 			host->cmd = NULL;
2181 			host->data = NULL;
2182 
2183 			if (!mrq->sbc && mrq->stop)
2184 				dw_mci_command_complete(host, mrq->stop);
2185 			else
2186 				host->cmd_status = 0;
2187 
2188 			dw_mci_request_end(host, mrq);
2189 			goto unlock;
2190 
2191 		case STATE_DATA_ERROR:
2192 			if (!test_and_clear_bit(EVENT_XFER_COMPLETE,
2193 						&host->pending_events))
2194 				break;
2195 
2196 			state = STATE_DATA_BUSY;
2197 			break;
2198 		}
2199 	} while (state != prev_state);
2200 
2201 	host->state = state;
2202 unlock:
2203 	spin_unlock(&host->lock);
2204 
2205 }
2206 
2207 /* push final bytes to part_buf, only use during push */
2208 static void dw_mci_set_part_bytes(struct dw_mci *host, void *buf, int cnt)
2209 {
2210 	memcpy((void *)&host->part_buf, buf, cnt);
2211 	host->part_buf_count = cnt;
2212 }
2213 
2214 /* append bytes to part_buf, only use during push */
2215 static int dw_mci_push_part_bytes(struct dw_mci *host, void *buf, int cnt)
2216 {
2217 	cnt = min(cnt, (1 << host->data_shift) - host->part_buf_count);
2218 	memcpy((void *)&host->part_buf + host->part_buf_count, buf, cnt);
2219 	host->part_buf_count += cnt;
2220 	return cnt;
2221 }
2222 
2223 /* pull first bytes from part_buf, only use during pull */
2224 static int dw_mci_pull_part_bytes(struct dw_mci *host, void *buf, int cnt)
2225 {
2226 	cnt = min_t(int, cnt, host->part_buf_count);
2227 	if (cnt) {
2228 		memcpy(buf, (void *)&host->part_buf + host->part_buf_start,
2229 		       cnt);
2230 		host->part_buf_count -= cnt;
2231 		host->part_buf_start += cnt;
2232 	}
2233 	return cnt;
2234 }
2235 
2236 /* pull final bytes from the part_buf, assuming it's just been filled */
2237 static void dw_mci_pull_final_bytes(struct dw_mci *host, void *buf, int cnt)
2238 {
2239 	memcpy(buf, &host->part_buf, cnt);
2240 	host->part_buf_start = cnt;
2241 	host->part_buf_count = (1 << host->data_shift) - cnt;
2242 }
2243 
2244 static void dw_mci_push_data16(struct dw_mci *host, void *buf, int cnt)
2245 {
2246 	struct mmc_data *data = host->data;
2247 	int init_cnt = cnt;
2248 
2249 	/* try and push anything in the part_buf */
2250 	if (unlikely(host->part_buf_count)) {
2251 		int len = dw_mci_push_part_bytes(host, buf, cnt);
2252 
2253 		buf += len;
2254 		cnt -= len;
2255 		if (host->part_buf_count == 2) {
2256 			mci_fifo_writew(host->fifo_reg, host->part_buf16);
2257 			host->part_buf_count = 0;
2258 		}
2259 	}
2260 #ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
2261 	if (unlikely((unsigned long)buf & 0x1)) {
2262 		while (cnt >= 2) {
2263 			u16 aligned_buf[64];
2264 			int len = min(cnt & -2, (int)sizeof(aligned_buf));
2265 			int items = len >> 1;
2266 			int i;
2267 			/* memcpy from input buffer into aligned buffer */
2268 			memcpy(aligned_buf, buf, len);
2269 			buf += len;
2270 			cnt -= len;
2271 			/* push data from aligned buffer into fifo */
2272 			for (i = 0; i < items; ++i)
2273 				mci_fifo_writew(host->fifo_reg, aligned_buf[i]);
2274 		}
2275 	} else
2276 #endif
2277 	{
2278 		u16 *pdata = buf;
2279 
2280 		for (; cnt >= 2; cnt -= 2)
2281 			mci_fifo_writew(host->fifo_reg, *pdata++);
2282 		buf = pdata;
2283 	}
2284 	/* put anything remaining in the part_buf */
2285 	if (cnt) {
2286 		dw_mci_set_part_bytes(host, buf, cnt);
2287 		 /* Push data if we have reached the expected data length */
2288 		if ((data->bytes_xfered + init_cnt) ==
2289 		    (data->blksz * data->blocks))
2290 			mci_fifo_writew(host->fifo_reg, host->part_buf16);
2291 	}
2292 }
2293 
2294 static void dw_mci_pull_data16(struct dw_mci *host, void *buf, int cnt)
2295 {
2296 #ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
2297 	if (unlikely((unsigned long)buf & 0x1)) {
2298 		while (cnt >= 2) {
2299 			/* pull data from fifo into aligned buffer */
2300 			u16 aligned_buf[64];
2301 			int len = min(cnt & -2, (int)sizeof(aligned_buf));
2302 			int items = len >> 1;
2303 			int i;
2304 
2305 			for (i = 0; i < items; ++i)
2306 				aligned_buf[i] = mci_fifo_readw(host->fifo_reg);
2307 			/* memcpy from aligned buffer into output buffer */
2308 			memcpy(buf, aligned_buf, len);
2309 			buf += len;
2310 			cnt -= len;
2311 		}
2312 	} else
2313 #endif
2314 	{
2315 		u16 *pdata = buf;
2316 
2317 		for (; cnt >= 2; cnt -= 2)
2318 			*pdata++ = mci_fifo_readw(host->fifo_reg);
2319 		buf = pdata;
2320 	}
2321 	if (cnt) {
2322 		host->part_buf16 = mci_fifo_readw(host->fifo_reg);
2323 		dw_mci_pull_final_bytes(host, buf, cnt);
2324 	}
2325 }
2326 
2327 static void dw_mci_push_data32(struct dw_mci *host, void *buf, int cnt)
2328 {
2329 	struct mmc_data *data = host->data;
2330 	int init_cnt = cnt;
2331 
2332 	/* try and push anything in the part_buf */
2333 	if (unlikely(host->part_buf_count)) {
2334 		int len = dw_mci_push_part_bytes(host, buf, cnt);
2335 
2336 		buf += len;
2337 		cnt -= len;
2338 		if (host->part_buf_count == 4) {
2339 			mci_fifo_writel(host->fifo_reg,	host->part_buf32);
2340 			host->part_buf_count = 0;
2341 		}
2342 	}
2343 #ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
2344 	if (unlikely((unsigned long)buf & 0x3)) {
2345 		while (cnt >= 4) {
2346 			u32 aligned_buf[32];
2347 			int len = min(cnt & -4, (int)sizeof(aligned_buf));
2348 			int items = len >> 2;
2349 			int i;
2350 			/* memcpy from input buffer into aligned buffer */
2351 			memcpy(aligned_buf, buf, len);
2352 			buf += len;
2353 			cnt -= len;
2354 			/* push data from aligned buffer into fifo */
2355 			for (i = 0; i < items; ++i)
2356 				mci_fifo_writel(host->fifo_reg,	aligned_buf[i]);
2357 		}
2358 	} else
2359 #endif
2360 	{
2361 		u32 *pdata = buf;
2362 
2363 		for (; cnt >= 4; cnt -= 4)
2364 			mci_fifo_writel(host->fifo_reg, *pdata++);
2365 		buf = pdata;
2366 	}
2367 	/* put anything remaining in the part_buf */
2368 	if (cnt) {
2369 		dw_mci_set_part_bytes(host, buf, cnt);
2370 		 /* Push data if we have reached the expected data length */
2371 		if ((data->bytes_xfered + init_cnt) ==
2372 		    (data->blksz * data->blocks))
2373 			mci_fifo_writel(host->fifo_reg, host->part_buf32);
2374 	}
2375 }
2376 
2377 static void dw_mci_pull_data32(struct dw_mci *host, void *buf, int cnt)
2378 {
2379 #ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
2380 	if (unlikely((unsigned long)buf & 0x3)) {
2381 		while (cnt >= 4) {
2382 			/* pull data from fifo into aligned buffer */
2383 			u32 aligned_buf[32];
2384 			int len = min(cnt & -4, (int)sizeof(aligned_buf));
2385 			int items = len >> 2;
2386 			int i;
2387 
2388 			for (i = 0; i < items; ++i)
2389 				aligned_buf[i] = mci_fifo_readl(host->fifo_reg);
2390 			/* memcpy from aligned buffer into output buffer */
2391 			memcpy(buf, aligned_buf, len);
2392 			buf += len;
2393 			cnt -= len;
2394 		}
2395 	} else
2396 #endif
2397 	{
2398 		u32 *pdata = buf;
2399 
2400 		for (; cnt >= 4; cnt -= 4)
2401 			*pdata++ = mci_fifo_readl(host->fifo_reg);
2402 		buf = pdata;
2403 	}
2404 	if (cnt) {
2405 		host->part_buf32 = mci_fifo_readl(host->fifo_reg);
2406 		dw_mci_pull_final_bytes(host, buf, cnt);
2407 	}
2408 }
2409 
2410 static void dw_mci_push_data64(struct dw_mci *host, void *buf, int cnt)
2411 {
2412 	struct mmc_data *data = host->data;
2413 	int init_cnt = cnt;
2414 
2415 	/* try and push anything in the part_buf */
2416 	if (unlikely(host->part_buf_count)) {
2417 		int len = dw_mci_push_part_bytes(host, buf, cnt);
2418 
2419 		buf += len;
2420 		cnt -= len;
2421 
2422 		if (host->part_buf_count == 8) {
2423 			mci_fifo_writeq(host->fifo_reg,	host->part_buf);
2424 			host->part_buf_count = 0;
2425 		}
2426 	}
2427 #ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
2428 	if (unlikely((unsigned long)buf & 0x7)) {
2429 		while (cnt >= 8) {
2430 			u64 aligned_buf[16];
2431 			int len = min(cnt & -8, (int)sizeof(aligned_buf));
2432 			int items = len >> 3;
2433 			int i;
2434 			/* memcpy from input buffer into aligned buffer */
2435 			memcpy(aligned_buf, buf, len);
2436 			buf += len;
2437 			cnt -= len;
2438 			/* push data from aligned buffer into fifo */
2439 			for (i = 0; i < items; ++i)
2440 				mci_fifo_writeq(host->fifo_reg,	aligned_buf[i]);
2441 		}
2442 	} else
2443 #endif
2444 	{
2445 		u64 *pdata = buf;
2446 
2447 		for (; cnt >= 8; cnt -= 8)
2448 			mci_fifo_writeq(host->fifo_reg, *pdata++);
2449 		buf = pdata;
2450 	}
2451 	/* put anything remaining in the part_buf */
2452 	if (cnt) {
2453 		dw_mci_set_part_bytes(host, buf, cnt);
2454 		/* Push data if we have reached the expected data length */
2455 		if ((data->bytes_xfered + init_cnt) ==
2456 		    (data->blksz * data->blocks))
2457 			mci_fifo_writeq(host->fifo_reg, host->part_buf);
2458 	}
2459 }
2460 
2461 static void dw_mci_pull_data64(struct dw_mci *host, void *buf, int cnt)
2462 {
2463 #ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
2464 	if (unlikely((unsigned long)buf & 0x7)) {
2465 		while (cnt >= 8) {
2466 			/* pull data from fifo into aligned buffer */
2467 			u64 aligned_buf[16];
2468 			int len = min(cnt & -8, (int)sizeof(aligned_buf));
2469 			int items = len >> 3;
2470 			int i;
2471 
2472 			for (i = 0; i < items; ++i)
2473 				aligned_buf[i] = mci_fifo_readq(host->fifo_reg);
2474 
2475 			/* memcpy from aligned buffer into output buffer */
2476 			memcpy(buf, aligned_buf, len);
2477 			buf += len;
2478 			cnt -= len;
2479 		}
2480 	} else
2481 #endif
2482 	{
2483 		u64 *pdata = buf;
2484 
2485 		for (; cnt >= 8; cnt -= 8)
2486 			*pdata++ = mci_fifo_readq(host->fifo_reg);
2487 		buf = pdata;
2488 	}
2489 	if (cnt) {
2490 		host->part_buf = mci_fifo_readq(host->fifo_reg);
2491 		dw_mci_pull_final_bytes(host, buf, cnt);
2492 	}
2493 }
2494 
2495 static void dw_mci_pull_data(struct dw_mci *host, void *buf, int cnt)
2496 {
2497 	int len;
2498 
2499 	/* get remaining partial bytes */
2500 	len = dw_mci_pull_part_bytes(host, buf, cnt);
2501 	if (unlikely(len == cnt))
2502 		return;
2503 	buf += len;
2504 	cnt -= len;
2505 
2506 	/* get the rest of the data */
2507 	host->pull_data(host, buf, cnt);
2508 }
2509 
2510 static void dw_mci_read_data_pio(struct dw_mci *host, bool dto)
2511 {
2512 	struct sg_mapping_iter *sg_miter = &host->sg_miter;
2513 	void *buf;
2514 	unsigned int offset;
2515 	struct mmc_data	*data = host->data;
2516 	int shift = host->data_shift;
2517 	u32 status;
2518 	unsigned int len;
2519 	unsigned int remain, fcnt;
2520 
2521 	do {
2522 		if (!sg_miter_next(sg_miter))
2523 			goto done;
2524 
2525 		host->sg = sg_miter->piter.sg;
2526 		buf = sg_miter->addr;
2527 		remain = sg_miter->length;
2528 		offset = 0;
2529 
2530 		do {
2531 			fcnt = (SDMMC_GET_FCNT(mci_readl(host, STATUS))
2532 					<< shift) + host->part_buf_count;
2533 			len = min(remain, fcnt);
2534 			if (!len)
2535 				break;
2536 			dw_mci_pull_data(host, (void *)(buf + offset), len);
2537 			data->bytes_xfered += len;
2538 			offset += len;
2539 			remain -= len;
2540 		} while (remain);
2541 
2542 		sg_miter->consumed = offset;
2543 		status = mci_readl(host, MINTSTS);
2544 		mci_writel(host, RINTSTS, SDMMC_INT_RXDR);
2545 	/* if the RXDR is ready read again */
2546 	} while ((status & SDMMC_INT_RXDR) ||
2547 		 (dto && SDMMC_GET_FCNT(mci_readl(host, STATUS))));
2548 
2549 	if (!remain) {
2550 		if (!sg_miter_next(sg_miter))
2551 			goto done;
2552 		sg_miter->consumed = 0;
2553 	}
2554 	sg_miter_stop(sg_miter);
2555 	return;
2556 
2557 done:
2558 	sg_miter_stop(sg_miter);
2559 	host->sg = NULL;
2560 	smp_wmb(); /* drain writebuffer */
2561 	set_bit(EVENT_XFER_COMPLETE, &host->pending_events);
2562 }
2563 
2564 static void dw_mci_write_data_pio(struct dw_mci *host)
2565 {
2566 	struct sg_mapping_iter *sg_miter = &host->sg_miter;
2567 	void *buf;
2568 	unsigned int offset;
2569 	struct mmc_data	*data = host->data;
2570 	int shift = host->data_shift;
2571 	u32 status;
2572 	unsigned int len;
2573 	unsigned int fifo_depth = host->fifo_depth;
2574 	unsigned int remain, fcnt;
2575 
2576 	do {
2577 		if (!sg_miter_next(sg_miter))
2578 			goto done;
2579 
2580 		host->sg = sg_miter->piter.sg;
2581 		buf = sg_miter->addr;
2582 		remain = sg_miter->length;
2583 		offset = 0;
2584 
2585 		do {
2586 			fcnt = ((fifo_depth -
2587 				 SDMMC_GET_FCNT(mci_readl(host, STATUS)))
2588 					<< shift) - host->part_buf_count;
2589 			len = min(remain, fcnt);
2590 			if (!len)
2591 				break;
2592 			host->push_data(host, (void *)(buf + offset), len);
2593 			data->bytes_xfered += len;
2594 			offset += len;
2595 			remain -= len;
2596 		} while (remain);
2597 
2598 		sg_miter->consumed = offset;
2599 		status = mci_readl(host, MINTSTS);
2600 		mci_writel(host, RINTSTS, SDMMC_INT_TXDR);
2601 	} while (status & SDMMC_INT_TXDR); /* if TXDR write again */
2602 
2603 	if (!remain) {
2604 		if (!sg_miter_next(sg_miter))
2605 			goto done;
2606 		sg_miter->consumed = 0;
2607 	}
2608 	sg_miter_stop(sg_miter);
2609 	return;
2610 
2611 done:
2612 	sg_miter_stop(sg_miter);
2613 	host->sg = NULL;
2614 	smp_wmb(); /* drain writebuffer */
2615 	set_bit(EVENT_XFER_COMPLETE, &host->pending_events);
2616 }
2617 
2618 static void dw_mci_cmd_interrupt(struct dw_mci *host, u32 status)
2619 {
2620 	del_timer(&host->cto_timer);
2621 
2622 	if (!host->cmd_status)
2623 		host->cmd_status = status;
2624 
2625 	smp_wmb(); /* drain writebuffer */
2626 
2627 	set_bit(EVENT_CMD_COMPLETE, &host->pending_events);
2628 	tasklet_schedule(&host->tasklet);
2629 }
2630 
2631 static void dw_mci_handle_cd(struct dw_mci *host)
2632 {
2633 	struct dw_mci_slot *slot = host->slot;
2634 
2635 	if (slot->mmc->ops->card_event)
2636 		slot->mmc->ops->card_event(slot->mmc);
2637 	mmc_detect_change(slot->mmc,
2638 		msecs_to_jiffies(host->pdata->detect_delay_ms));
2639 }
2640 
2641 static irqreturn_t dw_mci_interrupt(int irq, void *dev_id)
2642 {
2643 	struct dw_mci *host = dev_id;
2644 	u32 pending;
2645 	struct dw_mci_slot *slot = host->slot;
2646 	unsigned long irqflags;
2647 
2648 	pending = mci_readl(host, MINTSTS); /* read-only mask reg */
2649 
2650 	if (pending) {
2651 		/* Check volt switch first, since it can look like an error */
2652 		if ((host->state == STATE_SENDING_CMD11) &&
2653 		    (pending & SDMMC_INT_VOLT_SWITCH)) {
2654 			mci_writel(host, RINTSTS, SDMMC_INT_VOLT_SWITCH);
2655 			pending &= ~SDMMC_INT_VOLT_SWITCH;
2656 
2657 			/*
2658 			 * Hold the lock; we know cmd11_timer can't be kicked
2659 			 * off after the lock is released, so safe to delete.
2660 			 */
2661 			spin_lock_irqsave(&host->irq_lock, irqflags);
2662 			dw_mci_cmd_interrupt(host, pending);
2663 			spin_unlock_irqrestore(&host->irq_lock, irqflags);
2664 
2665 			del_timer(&host->cmd11_timer);
2666 		}
2667 
2668 		if (pending & DW_MCI_CMD_ERROR_FLAGS) {
2669 			spin_lock_irqsave(&host->irq_lock, irqflags);
2670 
2671 			del_timer(&host->cto_timer);
2672 			mci_writel(host, RINTSTS, DW_MCI_CMD_ERROR_FLAGS);
2673 			host->cmd_status = pending;
2674 			smp_wmb(); /* drain writebuffer */
2675 			set_bit(EVENT_CMD_COMPLETE, &host->pending_events);
2676 
2677 			spin_unlock_irqrestore(&host->irq_lock, irqflags);
2678 		}
2679 
2680 		if (pending & DW_MCI_DATA_ERROR_FLAGS) {
2681 			/* if there is an error report DATA_ERROR */
2682 			mci_writel(host, RINTSTS, DW_MCI_DATA_ERROR_FLAGS);
2683 			host->data_status = pending;
2684 			smp_wmb(); /* drain writebuffer */
2685 			set_bit(EVENT_DATA_ERROR, &host->pending_events);
2686 			tasklet_schedule(&host->tasklet);
2687 		}
2688 
2689 		if (pending & SDMMC_INT_DATA_OVER) {
2690 			spin_lock_irqsave(&host->irq_lock, irqflags);
2691 
2692 			del_timer(&host->dto_timer);
2693 
2694 			mci_writel(host, RINTSTS, SDMMC_INT_DATA_OVER);
2695 			if (!host->data_status)
2696 				host->data_status = pending;
2697 			smp_wmb(); /* drain writebuffer */
2698 			if (host->dir_status == DW_MCI_RECV_STATUS) {
2699 				if (host->sg != NULL)
2700 					dw_mci_read_data_pio(host, true);
2701 			}
2702 			set_bit(EVENT_DATA_COMPLETE, &host->pending_events);
2703 			tasklet_schedule(&host->tasklet);
2704 
2705 			spin_unlock_irqrestore(&host->irq_lock, irqflags);
2706 		}
2707 
2708 		if (pending & SDMMC_INT_RXDR) {
2709 			mci_writel(host, RINTSTS, SDMMC_INT_RXDR);
2710 			if (host->dir_status == DW_MCI_RECV_STATUS && host->sg)
2711 				dw_mci_read_data_pio(host, false);
2712 		}
2713 
2714 		if (pending & SDMMC_INT_TXDR) {
2715 			mci_writel(host, RINTSTS, SDMMC_INT_TXDR);
2716 			if (host->dir_status == DW_MCI_SEND_STATUS && host->sg)
2717 				dw_mci_write_data_pio(host);
2718 		}
2719 
2720 		if (pending & SDMMC_INT_CMD_DONE) {
2721 			spin_lock_irqsave(&host->irq_lock, irqflags);
2722 
2723 			mci_writel(host, RINTSTS, SDMMC_INT_CMD_DONE);
2724 			dw_mci_cmd_interrupt(host, pending);
2725 
2726 			spin_unlock_irqrestore(&host->irq_lock, irqflags);
2727 		}
2728 
2729 		if (pending & SDMMC_INT_CD) {
2730 			mci_writel(host, RINTSTS, SDMMC_INT_CD);
2731 			dw_mci_handle_cd(host);
2732 		}
2733 
2734 		if (pending & SDMMC_INT_SDIO(slot->sdio_id)) {
2735 			mci_writel(host, RINTSTS,
2736 				   SDMMC_INT_SDIO(slot->sdio_id));
2737 			__dw_mci_enable_sdio_irq(slot, 0);
2738 			sdio_signal_irq(slot->mmc);
2739 		}
2740 
2741 	}
2742 
2743 	if (host->use_dma != TRANS_MODE_IDMAC)
2744 		return IRQ_HANDLED;
2745 
2746 	/* Handle IDMA interrupts */
2747 	if (host->dma_64bit_address == 1) {
2748 		pending = mci_readl(host, IDSTS64);
2749 		if (pending & (SDMMC_IDMAC_INT_TI | SDMMC_IDMAC_INT_RI)) {
2750 			mci_writel(host, IDSTS64, SDMMC_IDMAC_INT_TI |
2751 							SDMMC_IDMAC_INT_RI);
2752 			mci_writel(host, IDSTS64, SDMMC_IDMAC_INT_NI);
2753 			if (!test_bit(EVENT_DATA_ERROR, &host->pending_events))
2754 				host->dma_ops->complete((void *)host);
2755 		}
2756 	} else {
2757 		pending = mci_readl(host, IDSTS);
2758 		if (pending & (SDMMC_IDMAC_INT_TI | SDMMC_IDMAC_INT_RI)) {
2759 			mci_writel(host, IDSTS, SDMMC_IDMAC_INT_TI |
2760 							SDMMC_IDMAC_INT_RI);
2761 			mci_writel(host, IDSTS, SDMMC_IDMAC_INT_NI);
2762 			if (!test_bit(EVENT_DATA_ERROR, &host->pending_events))
2763 				host->dma_ops->complete((void *)host);
2764 		}
2765 	}
2766 
2767 	return IRQ_HANDLED;
2768 }
2769 
2770 static int dw_mci_init_slot_caps(struct dw_mci_slot *slot)
2771 {
2772 	struct dw_mci *host = slot->host;
2773 	const struct dw_mci_drv_data *drv_data = host->drv_data;
2774 	struct mmc_host *mmc = slot->mmc;
2775 	int ctrl_id;
2776 
2777 	if (host->pdata->caps)
2778 		mmc->caps = host->pdata->caps;
2779 
2780 	/*
2781 	 * Support MMC_CAP_ERASE by default.
2782 	 * It needs to use trim/discard/erase commands.
2783 	 */
2784 	mmc->caps |= MMC_CAP_ERASE;
2785 
2786 	if (host->pdata->pm_caps)
2787 		mmc->pm_caps = host->pdata->pm_caps;
2788 
2789 	if (host->dev->of_node) {
2790 		ctrl_id = of_alias_get_id(host->dev->of_node, "mshc");
2791 		if (ctrl_id < 0)
2792 			ctrl_id = 0;
2793 	} else {
2794 		ctrl_id = to_platform_device(host->dev)->id;
2795 	}
2796 
2797 	if (drv_data && drv_data->caps) {
2798 		if (ctrl_id >= drv_data->num_caps) {
2799 			dev_err(host->dev, "invalid controller id %d\n",
2800 				ctrl_id);
2801 			return -EINVAL;
2802 		}
2803 		mmc->caps |= drv_data->caps[ctrl_id];
2804 	}
2805 
2806 	if (host->pdata->caps2)
2807 		mmc->caps2 = host->pdata->caps2;
2808 
2809 	mmc->f_min = DW_MCI_FREQ_MIN;
2810 	if (!mmc->f_max)
2811 		mmc->f_max = DW_MCI_FREQ_MAX;
2812 
2813 	/* Process SDIO IRQs through the sdio_irq_work. */
2814 	if (mmc->caps & MMC_CAP_SDIO_IRQ)
2815 		mmc->caps2 |= MMC_CAP2_SDIO_IRQ_NOTHREAD;
2816 
2817 	return 0;
2818 }
2819 
2820 static int dw_mci_init_slot(struct dw_mci *host)
2821 {
2822 	struct mmc_host *mmc;
2823 	struct dw_mci_slot *slot;
2824 	int ret;
2825 
2826 	mmc = mmc_alloc_host(sizeof(struct dw_mci_slot), host->dev);
2827 	if (!mmc)
2828 		return -ENOMEM;
2829 
2830 	slot = mmc_priv(mmc);
2831 	slot->id = 0;
2832 	slot->sdio_id = host->sdio_id0 + slot->id;
2833 	slot->mmc = mmc;
2834 	slot->host = host;
2835 	host->slot = slot;
2836 
2837 	mmc->ops = &dw_mci_ops;
2838 
2839 	/*if there are external regulators, get them*/
2840 	ret = mmc_regulator_get_supply(mmc);
2841 	if (ret)
2842 		goto err_host_allocated;
2843 
2844 	if (!mmc->ocr_avail)
2845 		mmc->ocr_avail = MMC_VDD_32_33 | MMC_VDD_33_34;
2846 
2847 	ret = mmc_of_parse(mmc);
2848 	if (ret)
2849 		goto err_host_allocated;
2850 
2851 	ret = dw_mci_init_slot_caps(slot);
2852 	if (ret)
2853 		goto err_host_allocated;
2854 
2855 	/* Useful defaults if platform data is unset. */
2856 	if (host->use_dma == TRANS_MODE_IDMAC) {
2857 		mmc->max_segs = host->ring_size;
2858 		mmc->max_blk_size = 65535;
2859 		mmc->max_seg_size = 0x1000;
2860 		mmc->max_req_size = mmc->max_seg_size * host->ring_size;
2861 		mmc->max_blk_count = mmc->max_req_size / 512;
2862 	} else if (host->use_dma == TRANS_MODE_EDMAC) {
2863 		mmc->max_segs = 64;
2864 		mmc->max_blk_size = 65535;
2865 		mmc->max_blk_count = 65535;
2866 		mmc->max_req_size =
2867 				mmc->max_blk_size * mmc->max_blk_count;
2868 		mmc->max_seg_size = mmc->max_req_size;
2869 	} else {
2870 		/* TRANS_MODE_PIO */
2871 		mmc->max_segs = 64;
2872 		mmc->max_blk_size = 65535; /* BLKSIZ is 16 bits */
2873 		mmc->max_blk_count = 512;
2874 		mmc->max_req_size = mmc->max_blk_size *
2875 				    mmc->max_blk_count;
2876 		mmc->max_seg_size = mmc->max_req_size;
2877 	}
2878 
2879 	dw_mci_get_cd(mmc);
2880 
2881 	ret = mmc_add_host(mmc);
2882 	if (ret)
2883 		goto err_host_allocated;
2884 
2885 #if defined(CONFIG_DEBUG_FS)
2886 	dw_mci_init_debugfs(slot);
2887 #endif
2888 
2889 	return 0;
2890 
2891 err_host_allocated:
2892 	mmc_free_host(mmc);
2893 	return ret;
2894 }
2895 
2896 static void dw_mci_cleanup_slot(struct dw_mci_slot *slot)
2897 {
2898 	/* Debugfs stuff is cleaned up by mmc core */
2899 	mmc_remove_host(slot->mmc);
2900 	slot->host->slot = NULL;
2901 	mmc_free_host(slot->mmc);
2902 }
2903 
2904 static void dw_mci_init_dma(struct dw_mci *host)
2905 {
2906 	int addr_config;
2907 	struct device *dev = host->dev;
2908 
2909 	/*
2910 	* Check tansfer mode from HCON[17:16]
2911 	* Clear the ambiguous description of dw_mmc databook:
2912 	* 2b'00: No DMA Interface -> Actually means using Internal DMA block
2913 	* 2b'01: DesignWare DMA Interface -> Synopsys DW-DMA block
2914 	* 2b'10: Generic DMA Interface -> non-Synopsys generic DMA block
2915 	* 2b'11: Non DW DMA Interface -> pio only
2916 	* Compared to DesignWare DMA Interface, Generic DMA Interface has a
2917 	* simpler request/acknowledge handshake mechanism and both of them
2918 	* are regarded as external dma master for dw_mmc.
2919 	*/
2920 	host->use_dma = SDMMC_GET_TRANS_MODE(mci_readl(host, HCON));
2921 	if (host->use_dma == DMA_INTERFACE_IDMA) {
2922 		host->use_dma = TRANS_MODE_IDMAC;
2923 	} else if (host->use_dma == DMA_INTERFACE_DWDMA ||
2924 		   host->use_dma == DMA_INTERFACE_GDMA) {
2925 		host->use_dma = TRANS_MODE_EDMAC;
2926 	} else {
2927 		goto no_dma;
2928 	}
2929 
2930 	/* Determine which DMA interface to use */
2931 	if (host->use_dma == TRANS_MODE_IDMAC) {
2932 		/*
2933 		* Check ADDR_CONFIG bit in HCON to find
2934 		* IDMAC address bus width
2935 		*/
2936 		addr_config = SDMMC_GET_ADDR_CONFIG(mci_readl(host, HCON));
2937 
2938 		if (addr_config == 1) {
2939 			/* host supports IDMAC in 64-bit address mode */
2940 			host->dma_64bit_address = 1;
2941 			dev_info(host->dev,
2942 				 "IDMAC supports 64-bit address mode.\n");
2943 			if (!dma_set_mask(host->dev, DMA_BIT_MASK(64)))
2944 				dma_set_coherent_mask(host->dev,
2945 						      DMA_BIT_MASK(64));
2946 		} else {
2947 			/* host supports IDMAC in 32-bit address mode */
2948 			host->dma_64bit_address = 0;
2949 			dev_info(host->dev,
2950 				 "IDMAC supports 32-bit address mode.\n");
2951 		}
2952 
2953 		/* Alloc memory for sg translation */
2954 		host->sg_cpu = dmam_alloc_coherent(host->dev,
2955 						   DESC_RING_BUF_SZ,
2956 						   &host->sg_dma, GFP_KERNEL);
2957 		if (!host->sg_cpu) {
2958 			dev_err(host->dev,
2959 				"%s: could not alloc DMA memory\n",
2960 				__func__);
2961 			goto no_dma;
2962 		}
2963 
2964 		host->dma_ops = &dw_mci_idmac_ops;
2965 		dev_info(host->dev, "Using internal DMA controller.\n");
2966 	} else {
2967 		/* TRANS_MODE_EDMAC: check dma bindings again */
2968 		if ((device_property_read_string_array(dev, "dma-names",
2969 						       NULL, 0) < 0) ||
2970 		    !device_property_present(dev, "dmas")) {
2971 			goto no_dma;
2972 		}
2973 		host->dma_ops = &dw_mci_edmac_ops;
2974 		dev_info(host->dev, "Using external DMA controller.\n");
2975 	}
2976 
2977 	if (host->dma_ops->init && host->dma_ops->start &&
2978 	    host->dma_ops->stop && host->dma_ops->cleanup) {
2979 		if (host->dma_ops->init(host)) {
2980 			dev_err(host->dev, "%s: Unable to initialize DMA Controller.\n",
2981 				__func__);
2982 			goto no_dma;
2983 		}
2984 	} else {
2985 		dev_err(host->dev, "DMA initialization not found.\n");
2986 		goto no_dma;
2987 	}
2988 
2989 	return;
2990 
2991 no_dma:
2992 	dev_info(host->dev, "Using PIO mode.\n");
2993 	host->use_dma = TRANS_MODE_PIO;
2994 }
2995 
2996 static void dw_mci_cmd11_timer(struct timer_list *t)
2997 {
2998 	struct dw_mci *host = from_timer(host, t, cmd11_timer);
2999 
3000 	if (host->state != STATE_SENDING_CMD11) {
3001 		dev_warn(host->dev, "Unexpected CMD11 timeout\n");
3002 		return;
3003 	}
3004 
3005 	host->cmd_status = SDMMC_INT_RTO;
3006 	set_bit(EVENT_CMD_COMPLETE, &host->pending_events);
3007 	tasklet_schedule(&host->tasklet);
3008 }
3009 
3010 static void dw_mci_cto_timer(struct timer_list *t)
3011 {
3012 	struct dw_mci *host = from_timer(host, t, cto_timer);
3013 	unsigned long irqflags;
3014 	u32 pending;
3015 
3016 	spin_lock_irqsave(&host->irq_lock, irqflags);
3017 
3018 	/*
3019 	 * If somehow we have very bad interrupt latency it's remotely possible
3020 	 * that the timer could fire while the interrupt is still pending or
3021 	 * while the interrupt is midway through running.  Let's be paranoid
3022 	 * and detect those two cases.  Note that this is paranoia is somewhat
3023 	 * justified because in this function we don't actually cancel the
3024 	 * pending command in the controller--we just assume it will never come.
3025 	 */
3026 	pending = mci_readl(host, MINTSTS); /* read-only mask reg */
3027 	if (pending & (DW_MCI_CMD_ERROR_FLAGS | SDMMC_INT_CMD_DONE)) {
3028 		/* The interrupt should fire; no need to act but we can warn */
3029 		dev_warn(host->dev, "Unexpected interrupt latency\n");
3030 		goto exit;
3031 	}
3032 	if (test_bit(EVENT_CMD_COMPLETE, &host->pending_events)) {
3033 		/* Presumably interrupt handler couldn't delete the timer */
3034 		dev_warn(host->dev, "CTO timeout when already completed\n");
3035 		goto exit;
3036 	}
3037 
3038 	/*
3039 	 * Continued paranoia to make sure we're in the state we expect.
3040 	 * This paranoia isn't really justified but it seems good to be safe.
3041 	 */
3042 	switch (host->state) {
3043 	case STATE_SENDING_CMD11:
3044 	case STATE_SENDING_CMD:
3045 	case STATE_SENDING_STOP:
3046 		/*
3047 		 * If CMD_DONE interrupt does NOT come in sending command
3048 		 * state, we should notify the driver to terminate current
3049 		 * transfer and report a command timeout to the core.
3050 		 */
3051 		host->cmd_status = SDMMC_INT_RTO;
3052 		set_bit(EVENT_CMD_COMPLETE, &host->pending_events);
3053 		tasklet_schedule(&host->tasklet);
3054 		break;
3055 	default:
3056 		dev_warn(host->dev, "Unexpected command timeout, state %d\n",
3057 			 host->state);
3058 		break;
3059 	}
3060 
3061 exit:
3062 	spin_unlock_irqrestore(&host->irq_lock, irqflags);
3063 }
3064 
3065 static void dw_mci_dto_timer(struct timer_list *t)
3066 {
3067 	struct dw_mci *host = from_timer(host, t, dto_timer);
3068 	unsigned long irqflags;
3069 	u32 pending;
3070 
3071 	spin_lock_irqsave(&host->irq_lock, irqflags);
3072 
3073 	/*
3074 	 * The DTO timer is much longer than the CTO timer, so it's even less
3075 	 * likely that we'll these cases, but it pays to be paranoid.
3076 	 */
3077 	pending = mci_readl(host, MINTSTS); /* read-only mask reg */
3078 	if (pending & SDMMC_INT_DATA_OVER) {
3079 		/* The interrupt should fire; no need to act but we can warn */
3080 		dev_warn(host->dev, "Unexpected data interrupt latency\n");
3081 		goto exit;
3082 	}
3083 	if (test_bit(EVENT_DATA_COMPLETE, &host->pending_events)) {
3084 		/* Presumably interrupt handler couldn't delete the timer */
3085 		dev_warn(host->dev, "DTO timeout when already completed\n");
3086 		goto exit;
3087 	}
3088 
3089 	/*
3090 	 * Continued paranoia to make sure we're in the state we expect.
3091 	 * This paranoia isn't really justified but it seems good to be safe.
3092 	 */
3093 	switch (host->state) {
3094 	case STATE_SENDING_DATA:
3095 	case STATE_DATA_BUSY:
3096 		/*
3097 		 * If DTO interrupt does NOT come in sending data state,
3098 		 * we should notify the driver to terminate current transfer
3099 		 * and report a data timeout to the core.
3100 		 */
3101 		host->data_status = SDMMC_INT_DRTO;
3102 		set_bit(EVENT_DATA_ERROR, &host->pending_events);
3103 		set_bit(EVENT_DATA_COMPLETE, &host->pending_events);
3104 		tasklet_schedule(&host->tasklet);
3105 		break;
3106 	default:
3107 		dev_warn(host->dev, "Unexpected data timeout, state %d\n",
3108 			 host->state);
3109 		break;
3110 	}
3111 
3112 exit:
3113 	spin_unlock_irqrestore(&host->irq_lock, irqflags);
3114 }
3115 
3116 #ifdef CONFIG_OF
3117 static struct dw_mci_board *dw_mci_parse_dt(struct dw_mci *host)
3118 {
3119 	struct dw_mci_board *pdata;
3120 	struct device *dev = host->dev;
3121 	const struct dw_mci_drv_data *drv_data = host->drv_data;
3122 	int ret;
3123 	u32 clock_frequency;
3124 
3125 	pdata = devm_kzalloc(dev, sizeof(*pdata), GFP_KERNEL);
3126 	if (!pdata)
3127 		return ERR_PTR(-ENOMEM);
3128 
3129 	/* find reset controller when exist */
3130 	pdata->rstc = devm_reset_control_get_optional_exclusive(dev, "reset");
3131 	if (IS_ERR(pdata->rstc)) {
3132 		if (PTR_ERR(pdata->rstc) == -EPROBE_DEFER)
3133 			return ERR_PTR(-EPROBE_DEFER);
3134 	}
3135 
3136 	if (device_property_read_u32(dev, "fifo-depth", &pdata->fifo_depth))
3137 		dev_info(dev,
3138 			 "fifo-depth property not found, using value of FIFOTH register as default\n");
3139 
3140 	device_property_read_u32(dev, "card-detect-delay",
3141 				 &pdata->detect_delay_ms);
3142 
3143 	device_property_read_u32(dev, "data-addr", &host->data_addr_override);
3144 
3145 	if (device_property_present(dev, "fifo-watermark-aligned"))
3146 		host->wm_aligned = true;
3147 
3148 	if (!device_property_read_u32(dev, "clock-frequency", &clock_frequency))
3149 		pdata->bus_hz = clock_frequency;
3150 
3151 	if (drv_data && drv_data->parse_dt) {
3152 		ret = drv_data->parse_dt(host);
3153 		if (ret)
3154 			return ERR_PTR(ret);
3155 	}
3156 
3157 	return pdata;
3158 }
3159 
3160 #else /* CONFIG_OF */
3161 static struct dw_mci_board *dw_mci_parse_dt(struct dw_mci *host)
3162 {
3163 	return ERR_PTR(-EINVAL);
3164 }
3165 #endif /* CONFIG_OF */
3166 
3167 static void dw_mci_enable_cd(struct dw_mci *host)
3168 {
3169 	unsigned long irqflags;
3170 	u32 temp;
3171 
3172 	/*
3173 	 * No need for CD if all slots have a non-error GPIO
3174 	 * as well as broken card detection is found.
3175 	 */
3176 	if (host->slot->mmc->caps & MMC_CAP_NEEDS_POLL)
3177 		return;
3178 
3179 	if (mmc_gpio_get_cd(host->slot->mmc) < 0) {
3180 		spin_lock_irqsave(&host->irq_lock, irqflags);
3181 		temp = mci_readl(host, INTMASK);
3182 		temp  |= SDMMC_INT_CD;
3183 		mci_writel(host, INTMASK, temp);
3184 		spin_unlock_irqrestore(&host->irq_lock, irqflags);
3185 	}
3186 }
3187 
3188 int dw_mci_probe(struct dw_mci *host)
3189 {
3190 	const struct dw_mci_drv_data *drv_data = host->drv_data;
3191 	int width, i, ret = 0;
3192 	u32 fifo_size;
3193 
3194 	if (!host->pdata) {
3195 		host->pdata = dw_mci_parse_dt(host);
3196 		if (PTR_ERR(host->pdata) == -EPROBE_DEFER) {
3197 			return -EPROBE_DEFER;
3198 		} else if (IS_ERR(host->pdata)) {
3199 			dev_err(host->dev, "platform data not available\n");
3200 			return -EINVAL;
3201 		}
3202 	}
3203 
3204 	host->biu_clk = devm_clk_get(host->dev, "biu");
3205 	if (IS_ERR(host->biu_clk)) {
3206 		dev_dbg(host->dev, "biu clock not available\n");
3207 	} else {
3208 		ret = clk_prepare_enable(host->biu_clk);
3209 		if (ret) {
3210 			dev_err(host->dev, "failed to enable biu clock\n");
3211 			return ret;
3212 		}
3213 	}
3214 
3215 	host->ciu_clk = devm_clk_get(host->dev, "ciu");
3216 	if (IS_ERR(host->ciu_clk)) {
3217 		dev_dbg(host->dev, "ciu clock not available\n");
3218 		host->bus_hz = host->pdata->bus_hz;
3219 	} else {
3220 		ret = clk_prepare_enable(host->ciu_clk);
3221 		if (ret) {
3222 			dev_err(host->dev, "failed to enable ciu clock\n");
3223 			goto err_clk_biu;
3224 		}
3225 
3226 		if (host->pdata->bus_hz) {
3227 			ret = clk_set_rate(host->ciu_clk, host->pdata->bus_hz);
3228 			if (ret)
3229 				dev_warn(host->dev,
3230 					 "Unable to set bus rate to %uHz\n",
3231 					 host->pdata->bus_hz);
3232 		}
3233 		host->bus_hz = clk_get_rate(host->ciu_clk);
3234 	}
3235 
3236 	if (!host->bus_hz) {
3237 		dev_err(host->dev,
3238 			"Platform data must supply bus speed\n");
3239 		ret = -ENODEV;
3240 		goto err_clk_ciu;
3241 	}
3242 
3243 	if (!IS_ERR(host->pdata->rstc)) {
3244 		reset_control_assert(host->pdata->rstc);
3245 		usleep_range(10, 50);
3246 		reset_control_deassert(host->pdata->rstc);
3247 	}
3248 
3249 	if (drv_data && drv_data->init) {
3250 		ret = drv_data->init(host);
3251 		if (ret) {
3252 			dev_err(host->dev,
3253 				"implementation specific init failed\n");
3254 			goto err_clk_ciu;
3255 		}
3256 	}
3257 
3258 	timer_setup(&host->cmd11_timer, dw_mci_cmd11_timer, 0);
3259 	timer_setup(&host->cto_timer, dw_mci_cto_timer, 0);
3260 	timer_setup(&host->dto_timer, dw_mci_dto_timer, 0);
3261 
3262 	spin_lock_init(&host->lock);
3263 	spin_lock_init(&host->irq_lock);
3264 	INIT_LIST_HEAD(&host->queue);
3265 
3266 	/*
3267 	 * Get the host data width - this assumes that HCON has been set with
3268 	 * the correct values.
3269 	 */
3270 	i = SDMMC_GET_HDATA_WIDTH(mci_readl(host, HCON));
3271 	if (!i) {
3272 		host->push_data = dw_mci_push_data16;
3273 		host->pull_data = dw_mci_pull_data16;
3274 		width = 16;
3275 		host->data_shift = 1;
3276 	} else if (i == 2) {
3277 		host->push_data = dw_mci_push_data64;
3278 		host->pull_data = dw_mci_pull_data64;
3279 		width = 64;
3280 		host->data_shift = 3;
3281 	} else {
3282 		/* Check for a reserved value, and warn if it is */
3283 		WARN((i != 1),
3284 		     "HCON reports a reserved host data width!\n"
3285 		     "Defaulting to 32-bit access.\n");
3286 		host->push_data = dw_mci_push_data32;
3287 		host->pull_data = dw_mci_pull_data32;
3288 		width = 32;
3289 		host->data_shift = 2;
3290 	}
3291 
3292 	/* Reset all blocks */
3293 	if (!dw_mci_ctrl_reset(host, SDMMC_CTRL_ALL_RESET_FLAGS)) {
3294 		ret = -ENODEV;
3295 		goto err_clk_ciu;
3296 	}
3297 
3298 	host->dma_ops = host->pdata->dma_ops;
3299 	dw_mci_init_dma(host);
3300 
3301 	/* Clear the interrupts for the host controller */
3302 	mci_writel(host, RINTSTS, 0xFFFFFFFF);
3303 	mci_writel(host, INTMASK, 0); /* disable all mmc interrupt first */
3304 
3305 	/* Put in max timeout */
3306 	mci_writel(host, TMOUT, 0xFFFFFFFF);
3307 
3308 	/*
3309 	 * FIFO threshold settings  RxMark  = fifo_size / 2 - 1,
3310 	 *                          Tx Mark = fifo_size / 2 DMA Size = 8
3311 	 */
3312 	if (!host->pdata->fifo_depth) {
3313 		/*
3314 		 * Power-on value of RX_WMark is FIFO_DEPTH-1, but this may
3315 		 * have been overwritten by the bootloader, just like we're
3316 		 * about to do, so if you know the value for your hardware, you
3317 		 * should put it in the platform data.
3318 		 */
3319 		fifo_size = mci_readl(host, FIFOTH);
3320 		fifo_size = 1 + ((fifo_size >> 16) & 0xfff);
3321 	} else {
3322 		fifo_size = host->pdata->fifo_depth;
3323 	}
3324 	host->fifo_depth = fifo_size;
3325 	host->fifoth_val =
3326 		SDMMC_SET_FIFOTH(0x2, fifo_size / 2 - 1, fifo_size / 2);
3327 	mci_writel(host, FIFOTH, host->fifoth_val);
3328 
3329 	/* disable clock to CIU */
3330 	mci_writel(host, CLKENA, 0);
3331 	mci_writel(host, CLKSRC, 0);
3332 
3333 	/*
3334 	 * In 2.40a spec, Data offset is changed.
3335 	 * Need to check the version-id and set data-offset for DATA register.
3336 	 */
3337 	host->verid = SDMMC_GET_VERID(mci_readl(host, VERID));
3338 	dev_info(host->dev, "Version ID is %04x\n", host->verid);
3339 
3340 	if (host->data_addr_override)
3341 		host->fifo_reg = host->regs + host->data_addr_override;
3342 	else if (host->verid < DW_MMC_240A)
3343 		host->fifo_reg = host->regs + DATA_OFFSET;
3344 	else
3345 		host->fifo_reg = host->regs + DATA_240A_OFFSET;
3346 
3347 	tasklet_init(&host->tasklet, dw_mci_tasklet_func, (unsigned long)host);
3348 	ret = devm_request_irq(host->dev, host->irq, dw_mci_interrupt,
3349 			       host->irq_flags, "dw-mci", host);
3350 	if (ret)
3351 		goto err_dmaunmap;
3352 
3353 	/*
3354 	 * Enable interrupts for command done, data over, data empty,
3355 	 * receive ready and error such as transmit, receive timeout, crc error
3356 	 */
3357 	mci_writel(host, INTMASK, SDMMC_INT_CMD_DONE | SDMMC_INT_DATA_OVER |
3358 		   SDMMC_INT_TXDR | SDMMC_INT_RXDR |
3359 		   DW_MCI_ERROR_FLAGS);
3360 	/* Enable mci interrupt */
3361 	mci_writel(host, CTRL, SDMMC_CTRL_INT_ENABLE);
3362 
3363 	dev_info(host->dev,
3364 		 "DW MMC controller at irq %d,%d bit host data width,%u deep fifo\n",
3365 		 host->irq, width, fifo_size);
3366 
3367 	/* We need at least one slot to succeed */
3368 	ret = dw_mci_init_slot(host);
3369 	if (ret) {
3370 		dev_dbg(host->dev, "slot %d init failed\n", i);
3371 		goto err_dmaunmap;
3372 	}
3373 
3374 	/* Now that slots are all setup, we can enable card detect */
3375 	dw_mci_enable_cd(host);
3376 
3377 	return 0;
3378 
3379 err_dmaunmap:
3380 	if (host->use_dma && host->dma_ops->exit)
3381 		host->dma_ops->exit(host);
3382 
3383 	if (!IS_ERR(host->pdata->rstc))
3384 		reset_control_assert(host->pdata->rstc);
3385 
3386 err_clk_ciu:
3387 	clk_disable_unprepare(host->ciu_clk);
3388 
3389 err_clk_biu:
3390 	clk_disable_unprepare(host->biu_clk);
3391 
3392 	return ret;
3393 }
3394 EXPORT_SYMBOL(dw_mci_probe);
3395 
3396 void dw_mci_remove(struct dw_mci *host)
3397 {
3398 	dev_dbg(host->dev, "remove slot\n");
3399 	if (host->slot)
3400 		dw_mci_cleanup_slot(host->slot);
3401 
3402 	mci_writel(host, RINTSTS, 0xFFFFFFFF);
3403 	mci_writel(host, INTMASK, 0); /* disable all mmc interrupt first */
3404 
3405 	/* disable clock to CIU */
3406 	mci_writel(host, CLKENA, 0);
3407 	mci_writel(host, CLKSRC, 0);
3408 
3409 	if (host->use_dma && host->dma_ops->exit)
3410 		host->dma_ops->exit(host);
3411 
3412 	if (!IS_ERR(host->pdata->rstc))
3413 		reset_control_assert(host->pdata->rstc);
3414 
3415 	clk_disable_unprepare(host->ciu_clk);
3416 	clk_disable_unprepare(host->biu_clk);
3417 }
3418 EXPORT_SYMBOL(dw_mci_remove);
3419 
3420 
3421 
3422 #ifdef CONFIG_PM
3423 int dw_mci_runtime_suspend(struct device *dev)
3424 {
3425 	struct dw_mci *host = dev_get_drvdata(dev);
3426 
3427 	if (host->use_dma && host->dma_ops->exit)
3428 		host->dma_ops->exit(host);
3429 
3430 	clk_disable_unprepare(host->ciu_clk);
3431 
3432 	if (host->slot &&
3433 	    (mmc_can_gpio_cd(host->slot->mmc) ||
3434 	     !mmc_card_is_removable(host->slot->mmc)))
3435 		clk_disable_unprepare(host->biu_clk);
3436 
3437 	return 0;
3438 }
3439 EXPORT_SYMBOL(dw_mci_runtime_suspend);
3440 
3441 int dw_mci_runtime_resume(struct device *dev)
3442 {
3443 	int ret = 0;
3444 	struct dw_mci *host = dev_get_drvdata(dev);
3445 
3446 	if (host->slot &&
3447 	    (mmc_can_gpio_cd(host->slot->mmc) ||
3448 	     !mmc_card_is_removable(host->slot->mmc))) {
3449 		ret = clk_prepare_enable(host->biu_clk);
3450 		if (ret)
3451 			return ret;
3452 	}
3453 
3454 	ret = clk_prepare_enable(host->ciu_clk);
3455 	if (ret)
3456 		goto err;
3457 
3458 	if (!dw_mci_ctrl_reset(host, SDMMC_CTRL_ALL_RESET_FLAGS)) {
3459 		clk_disable_unprepare(host->ciu_clk);
3460 		ret = -ENODEV;
3461 		goto err;
3462 	}
3463 
3464 	if (host->use_dma && host->dma_ops->init)
3465 		host->dma_ops->init(host);
3466 
3467 	/*
3468 	 * Restore the initial value at FIFOTH register
3469 	 * And Invalidate the prev_blksz with zero
3470 	 */
3471 	 mci_writel(host, FIFOTH, host->fifoth_val);
3472 	 host->prev_blksz = 0;
3473 
3474 	/* Put in max timeout */
3475 	mci_writel(host, TMOUT, 0xFFFFFFFF);
3476 
3477 	mci_writel(host, RINTSTS, 0xFFFFFFFF);
3478 	mci_writel(host, INTMASK, SDMMC_INT_CMD_DONE | SDMMC_INT_DATA_OVER |
3479 		   SDMMC_INT_TXDR | SDMMC_INT_RXDR |
3480 		   DW_MCI_ERROR_FLAGS);
3481 	mci_writel(host, CTRL, SDMMC_CTRL_INT_ENABLE);
3482 
3483 
3484 	if (host->slot->mmc->pm_flags & MMC_PM_KEEP_POWER)
3485 		dw_mci_set_ios(host->slot->mmc, &host->slot->mmc->ios);
3486 
3487 	/* Force setup bus to guarantee available clock output */
3488 	dw_mci_setup_bus(host->slot, true);
3489 
3490 	/* Now that slots are all setup, we can enable card detect */
3491 	dw_mci_enable_cd(host);
3492 
3493 	return 0;
3494 
3495 err:
3496 	if (host->slot &&
3497 	    (mmc_can_gpio_cd(host->slot->mmc) ||
3498 	     !mmc_card_is_removable(host->slot->mmc)))
3499 		clk_disable_unprepare(host->biu_clk);
3500 
3501 	return ret;
3502 }
3503 EXPORT_SYMBOL(dw_mci_runtime_resume);
3504 #endif /* CONFIG_PM */
3505 
3506 static int __init dw_mci_init(void)
3507 {
3508 	pr_info("Synopsys Designware Multimedia Card Interface Driver\n");
3509 	return 0;
3510 }
3511 
3512 static void __exit dw_mci_exit(void)
3513 {
3514 }
3515 
3516 module_init(dw_mci_init);
3517 module_exit(dw_mci_exit);
3518 
3519 MODULE_DESCRIPTION("DW Multimedia Card Interface driver");
3520 MODULE_AUTHOR("NXP Semiconductor VietNam");
3521 MODULE_AUTHOR("Imagination Technologies Ltd");
3522 MODULE_LICENSE("GPL v2");
3523