1// SPDX-License-Identifier: GPL-2.0-only
2/* Copyright (c) 2015, The Linux Foundation. All rights reserved.
3 */
4
5#include <linux/delay.h>
6#include <linux/highmem.h>
7#include <linux/io.h>
8#include <linux/iopoll.h>
9#include <linux/module.h>
10#include <linux/dma-mapping.h>
11#include <linux/slab.h>
12#include <linux/scatterlist.h>
13#include <linux/platform_device.h>
14#include <linux/ktime.h>
15
16#include <linux/mmc/mmc.h>
17#include <linux/mmc/host.h>
18#include <linux/mmc/card.h>
19
20#include "cqhci.h"
21#include "cqhci-crypto.h"
22
23#define DCMD_SLOT 31
24#define NUM_SLOTS 32
25
26struct cqhci_slot {
27	struct mmc_request *mrq;
28	unsigned int flags;
29#define CQHCI_EXTERNAL_TIMEOUT	BIT(0)
30#define CQHCI_COMPLETED		BIT(1)
31#define CQHCI_HOST_CRC		BIT(2)
32#define CQHCI_HOST_TIMEOUT	BIT(3)
33#define CQHCI_HOST_OTHER	BIT(4)
34};
35
36static inline u8 *get_desc(struct cqhci_host *cq_host, u8 tag)
37{
38	return cq_host->desc_base + (tag * cq_host->slot_sz);
39}
40
41static inline u8 *get_link_desc(struct cqhci_host *cq_host, u8 tag)
42{
43	u8 *desc = get_desc(cq_host, tag);
44
45	return desc + cq_host->task_desc_len;
46}
47
48static inline size_t get_trans_desc_offset(struct cqhci_host *cq_host, u8 tag)
49{
50	return cq_host->trans_desc_len * cq_host->mmc->max_segs * tag;
51}
52
53static inline dma_addr_t get_trans_desc_dma(struct cqhci_host *cq_host, u8 tag)
54{
55	size_t offset = get_trans_desc_offset(cq_host, tag);
56
57	return cq_host->trans_desc_dma_base + offset;
58}
59
60static inline u8 *get_trans_desc(struct cqhci_host *cq_host, u8 tag)
61{
62	size_t offset = get_trans_desc_offset(cq_host, tag);
63
64	return cq_host->trans_desc_base + offset;
65}
66
67static void setup_trans_desc(struct cqhci_host *cq_host, u8 tag)
68{
69	u8 *link_temp;
70	dma_addr_t trans_temp;
71
72	link_temp = get_link_desc(cq_host, tag);
73	trans_temp = get_trans_desc_dma(cq_host, tag);
74
75	memset(link_temp, 0, cq_host->link_desc_len);
76	if (cq_host->link_desc_len > 8)
77		*(link_temp + 8) = 0;
78
79	if (tag == DCMD_SLOT && (cq_host->mmc->caps2 & MMC_CAP2_CQE_DCMD)) {
80		*link_temp = CQHCI_VALID(0) | CQHCI_ACT(0) | CQHCI_END(1);
81		return;
82	}
83
84	*link_temp = CQHCI_VALID(1) | CQHCI_ACT(0x6) | CQHCI_END(0);
85
86	if (cq_host->dma64) {
87		__le64 *data_addr = (__le64 __force *)(link_temp + 4);
88
89		data_addr[0] = cpu_to_le64(trans_temp);
90	} else {
91		__le32 *data_addr = (__le32 __force *)(link_temp + 4);
92
93		data_addr[0] = cpu_to_le32(trans_temp);
94	}
95}
96
97static void cqhci_set_irqs(struct cqhci_host *cq_host, u32 set)
98{
99	cqhci_writel(cq_host, set, CQHCI_ISTE);
100	cqhci_writel(cq_host, set, CQHCI_ISGE);
101}
102
103#define DRV_NAME "cqhci"
104
105#define CQHCI_DUMP(f, x...) \
106	pr_err("%s: " DRV_NAME ": " f, mmc_hostname(mmc), ## x)
107
108static void cqhci_dumpregs(struct cqhci_host *cq_host)
109{
110	struct mmc_host *mmc = cq_host->mmc;
111
112	CQHCI_DUMP("============ CQHCI REGISTER DUMP ===========\n");
113
114	CQHCI_DUMP("Caps:      0x%08x | Version:  0x%08x\n",
115		   cqhci_readl(cq_host, CQHCI_CAP),
116		   cqhci_readl(cq_host, CQHCI_VER));
117	CQHCI_DUMP("Config:    0x%08x | Control:  0x%08x\n",
118		   cqhci_readl(cq_host, CQHCI_CFG),
119		   cqhci_readl(cq_host, CQHCI_CTL));
120	CQHCI_DUMP("Int stat:  0x%08x | Int enab: 0x%08x\n",
121		   cqhci_readl(cq_host, CQHCI_IS),
122		   cqhci_readl(cq_host, CQHCI_ISTE));
123	CQHCI_DUMP("Int sig:   0x%08x | Int Coal: 0x%08x\n",
124		   cqhci_readl(cq_host, CQHCI_ISGE),
125		   cqhci_readl(cq_host, CQHCI_IC));
126	CQHCI_DUMP("TDL base:  0x%08x | TDL up32: 0x%08x\n",
127		   cqhci_readl(cq_host, CQHCI_TDLBA),
128		   cqhci_readl(cq_host, CQHCI_TDLBAU));
129	CQHCI_DUMP("Doorbell:  0x%08x | TCN:      0x%08x\n",
130		   cqhci_readl(cq_host, CQHCI_TDBR),
131		   cqhci_readl(cq_host, CQHCI_TCN));
132	CQHCI_DUMP("Dev queue: 0x%08x | Dev Pend: 0x%08x\n",
133		   cqhci_readl(cq_host, CQHCI_DQS),
134		   cqhci_readl(cq_host, CQHCI_DPT));
135	CQHCI_DUMP("Task clr:  0x%08x | SSC1:     0x%08x\n",
136		   cqhci_readl(cq_host, CQHCI_TCLR),
137		   cqhci_readl(cq_host, CQHCI_SSC1));
138	CQHCI_DUMP("SSC2:      0x%08x | DCMD rsp: 0x%08x\n",
139		   cqhci_readl(cq_host, CQHCI_SSC2),
140		   cqhci_readl(cq_host, CQHCI_CRDCT));
141	CQHCI_DUMP("RED mask:  0x%08x | TERRI:    0x%08x\n",
142		   cqhci_readl(cq_host, CQHCI_RMEM),
143		   cqhci_readl(cq_host, CQHCI_TERRI));
144	CQHCI_DUMP("Resp idx:  0x%08x | Resp arg: 0x%08x\n",
145		   cqhci_readl(cq_host, CQHCI_CRI),
146		   cqhci_readl(cq_host, CQHCI_CRA));
147
148	if (cq_host->ops->dumpregs)
149		cq_host->ops->dumpregs(mmc);
150	else
151		CQHCI_DUMP(": ===========================================\n");
152}
153
154/*
155 * The allocated descriptor table for task, link & transfer descriptors
156 * looks like:
157 * |----------|
158 * |task desc |  |->|----------|
159 * |----------|  |  |trans desc|
160 * |link desc-|->|  |----------|
161 * |----------|          .
162 *      .                .
163 *  no. of slots      max-segs
164 *      .           |----------|
165 * |----------|
166 * The idea here is to create the [task+trans] table and mark & point the
167 * link desc to the transfer desc table on a per slot basis.
168 */
169static int cqhci_host_alloc_tdl(struct cqhci_host *cq_host)
170{
171	int i = 0;
172
173	/* task descriptor can be 64/128 bit irrespective of arch */
174	if (cq_host->caps & CQHCI_TASK_DESC_SZ_128) {
175		cqhci_writel(cq_host, cqhci_readl(cq_host, CQHCI_CFG) |
176			       CQHCI_TASK_DESC_SZ, CQHCI_CFG);
177		cq_host->task_desc_len = 16;
178	} else {
179		cq_host->task_desc_len = 8;
180	}
181
182	/*
183	 * 96 bits length of transfer desc instead of 128 bits which means
184	 * ADMA would expect next valid descriptor at the 96th bit
185	 * or 128th bit
186	 */
187	if (cq_host->dma64) {
188		if (cq_host->quirks & CQHCI_QUIRK_SHORT_TXFR_DESC_SZ)
189			cq_host->trans_desc_len = 12;
190		else
191			cq_host->trans_desc_len = 16;
192		cq_host->link_desc_len = 16;
193	} else {
194		cq_host->trans_desc_len = 8;
195		cq_host->link_desc_len = 8;
196	}
197
198	/* total size of a slot: 1 task & 1 transfer (link) */
199	cq_host->slot_sz = cq_host->task_desc_len + cq_host->link_desc_len;
200
201	cq_host->desc_size = cq_host->slot_sz * cq_host->num_slots;
202
203	cq_host->data_size = get_trans_desc_offset(cq_host, cq_host->mmc->cqe_qdepth);
204
205	pr_debug("%s: cqhci: desc_size: %zu data_sz: %zu slot-sz: %d\n",
206		 mmc_hostname(cq_host->mmc), cq_host->desc_size, cq_host->data_size,
207		 cq_host->slot_sz);
208
209	/*
210	 * allocate a dma-mapped chunk of memory for the descriptors
211	 * allocate a dma-mapped chunk of memory for link descriptors
212	 * setup each link-desc memory offset per slot-number to
213	 * the descriptor table.
214	 */
215	cq_host->desc_base = dmam_alloc_coherent(mmc_dev(cq_host->mmc),
216						 cq_host->desc_size,
217						 &cq_host->desc_dma_base,
218						 GFP_KERNEL);
219	if (!cq_host->desc_base)
220		return -ENOMEM;
221
222	cq_host->trans_desc_base = dmam_alloc_coherent(mmc_dev(cq_host->mmc),
223					      cq_host->data_size,
224					      &cq_host->trans_desc_dma_base,
225					      GFP_KERNEL);
226	if (!cq_host->trans_desc_base) {
227		dmam_free_coherent(mmc_dev(cq_host->mmc), cq_host->desc_size,
228				   cq_host->desc_base,
229				   cq_host->desc_dma_base);
230		cq_host->desc_base = NULL;
231		cq_host->desc_dma_base = 0;
232		return -ENOMEM;
233	}
234
235	pr_debug("%s: cqhci: desc-base: 0x%p trans-base: 0x%p\n desc_dma 0x%llx trans_dma: 0x%llx\n",
236		 mmc_hostname(cq_host->mmc), cq_host->desc_base, cq_host->trans_desc_base,
237		(unsigned long long)cq_host->desc_dma_base,
238		(unsigned long long)cq_host->trans_desc_dma_base);
239
240	for (; i < (cq_host->num_slots); i++)
241		setup_trans_desc(cq_host, i);
242
243	return 0;
244}
245
246static void __cqhci_enable(struct cqhci_host *cq_host)
247{
248	struct mmc_host *mmc = cq_host->mmc;
249	u32 cqcfg;
250
251	cqcfg = cqhci_readl(cq_host, CQHCI_CFG);
252
253	/* Configuration must not be changed while enabled */
254	if (cqcfg & CQHCI_ENABLE) {
255		cqcfg &= ~CQHCI_ENABLE;
256		cqhci_writel(cq_host, cqcfg, CQHCI_CFG);
257	}
258
259	cqcfg &= ~(CQHCI_DCMD | CQHCI_TASK_DESC_SZ);
260
261	if (mmc->caps2 & MMC_CAP2_CQE_DCMD)
262		cqcfg |= CQHCI_DCMD;
263
264	if (cq_host->caps & CQHCI_TASK_DESC_SZ_128)
265		cqcfg |= CQHCI_TASK_DESC_SZ;
266
267	if (mmc->caps2 & MMC_CAP2_CRYPTO)
268		cqcfg |= CQHCI_CRYPTO_GENERAL_ENABLE;
269
270	cqhci_writel(cq_host, cqcfg, CQHCI_CFG);
271
272	cqhci_writel(cq_host, lower_32_bits(cq_host->desc_dma_base),
273		     CQHCI_TDLBA);
274	cqhci_writel(cq_host, upper_32_bits(cq_host->desc_dma_base),
275		     CQHCI_TDLBAU);
276
277	cqhci_writel(cq_host, cq_host->rca, CQHCI_SSC2);
278
279	cqhci_set_irqs(cq_host, 0);
280
281	cqcfg |= CQHCI_ENABLE;
282
283	cqhci_writel(cq_host, cqcfg, CQHCI_CFG);
284
285	if (cqhci_readl(cq_host, CQHCI_CTL) & CQHCI_HALT)
286		cqhci_writel(cq_host, 0, CQHCI_CTL);
287
288	mmc->cqe_on = true;
289
290	if (cq_host->ops->enable)
291		cq_host->ops->enable(mmc);
292
293	/* Ensure all writes are done before interrupts are enabled */
294	wmb();
295
296	cqhci_set_irqs(cq_host, CQHCI_IS_MASK);
297
298	cq_host->activated = true;
299}
300
301static void __cqhci_disable(struct cqhci_host *cq_host)
302{
303	u32 cqcfg;
304
305	cqcfg = cqhci_readl(cq_host, CQHCI_CFG);
306	cqcfg &= ~CQHCI_ENABLE;
307	cqhci_writel(cq_host, cqcfg, CQHCI_CFG);
308
309	cq_host->mmc->cqe_on = false;
310
311	cq_host->activated = false;
312}
313
314int cqhci_deactivate(struct mmc_host *mmc)
315{
316	struct cqhci_host *cq_host = mmc->cqe_private;
317
318	if (cq_host->enabled && cq_host->activated)
319		__cqhci_disable(cq_host);
320
321	return 0;
322}
323EXPORT_SYMBOL(cqhci_deactivate);
324
325int cqhci_resume(struct mmc_host *mmc)
326{
327	/* Re-enable is done upon first request */
328	return 0;
329}
330EXPORT_SYMBOL(cqhci_resume);
331
332static int cqhci_enable(struct mmc_host *mmc, struct mmc_card *card)
333{
334	struct cqhci_host *cq_host = mmc->cqe_private;
335	int err;
336
337	if (!card->ext_csd.cmdq_en)
338		return -EINVAL;
339
340	if (cq_host->enabled)
341		return 0;
342
343	cq_host->rca = card->rca;
344
345	err = cqhci_host_alloc_tdl(cq_host);
346	if (err) {
347		pr_err("%s: Failed to enable CQE, error %d\n",
348		       mmc_hostname(mmc), err);
349		return err;
350	}
351
352	__cqhci_enable(cq_host);
353
354	cq_host->enabled = true;
355
356#ifdef DEBUG
357	cqhci_dumpregs(cq_host);
358#endif
359	return 0;
360}
361
362/* CQHCI is idle and should halt immediately, so set a small timeout */
363#define CQHCI_OFF_TIMEOUT 100
364
365static u32 cqhci_read_ctl(struct cqhci_host *cq_host)
366{
367	return cqhci_readl(cq_host, CQHCI_CTL);
368}
369
370static void cqhci_off(struct mmc_host *mmc)
371{
372	struct cqhci_host *cq_host = mmc->cqe_private;
373	u32 reg;
374	int err;
375
376	if (!cq_host->enabled || !mmc->cqe_on || cq_host->recovery_halt)
377		return;
378
379	if (cq_host->ops->disable)
380		cq_host->ops->disable(mmc, false);
381
382	cqhci_writel(cq_host, CQHCI_HALT, CQHCI_CTL);
383
384	err = readx_poll_timeout(cqhci_read_ctl, cq_host, reg,
385				 reg & CQHCI_HALT, 0, CQHCI_OFF_TIMEOUT);
386	if (err < 0)
387		pr_err("%s: cqhci: CQE stuck on\n", mmc_hostname(mmc));
388	else
389		pr_debug("%s: cqhci: CQE off\n", mmc_hostname(mmc));
390
391	if (cq_host->ops->post_disable)
392		cq_host->ops->post_disable(mmc);
393
394	mmc->cqe_on = false;
395}
396
397static void cqhci_disable(struct mmc_host *mmc)
398{
399	struct cqhci_host *cq_host = mmc->cqe_private;
400
401	if (!cq_host->enabled)
402		return;
403
404	cqhci_off(mmc);
405
406	__cqhci_disable(cq_host);
407
408	dmam_free_coherent(mmc_dev(mmc), cq_host->data_size,
409			   cq_host->trans_desc_base,
410			   cq_host->trans_desc_dma_base);
411
412	dmam_free_coherent(mmc_dev(mmc), cq_host->desc_size,
413			   cq_host->desc_base,
414			   cq_host->desc_dma_base);
415
416	cq_host->trans_desc_base = NULL;
417	cq_host->desc_base = NULL;
418
419	cq_host->enabled = false;
420}
421
422static void cqhci_prep_task_desc(struct mmc_request *mrq,
423				 struct cqhci_host *cq_host, int tag)
424{
425	__le64 *task_desc = (__le64 __force *)get_desc(cq_host, tag);
426	u32 req_flags = mrq->data->flags;
427	u64 desc0;
428
429	desc0 = CQHCI_VALID(1) |
430		CQHCI_END(1) |
431		CQHCI_INT(1) |
432		CQHCI_ACT(0x5) |
433		CQHCI_FORCED_PROG(!!(req_flags & MMC_DATA_FORCED_PRG)) |
434		CQHCI_DATA_TAG(!!(req_flags & MMC_DATA_DAT_TAG)) |
435		CQHCI_DATA_DIR(!!(req_flags & MMC_DATA_READ)) |
436		CQHCI_PRIORITY(!!(req_flags & MMC_DATA_PRIO)) |
437		CQHCI_QBAR(!!(req_flags & MMC_DATA_QBR)) |
438		CQHCI_REL_WRITE(!!(req_flags & MMC_DATA_REL_WR)) |
439		CQHCI_BLK_COUNT(mrq->data->blocks) |
440		CQHCI_BLK_ADDR((u64)mrq->data->blk_addr);
441
442	task_desc[0] = cpu_to_le64(desc0);
443
444	if (cq_host->caps & CQHCI_TASK_DESC_SZ_128) {
445		u64 desc1 = cqhci_crypto_prep_task_desc(mrq);
446
447		task_desc[1] = cpu_to_le64(desc1);
448
449		pr_debug("%s: cqhci: tag %d task descriptor 0x%016llx%016llx\n",
450			 mmc_hostname(mrq->host), mrq->tag, desc1, desc0);
451	} else {
452		pr_debug("%s: cqhci: tag %d task descriptor 0x%016llx\n",
453			 mmc_hostname(mrq->host), mrq->tag, desc0);
454	}
455}
456
457static int cqhci_dma_map(struct mmc_host *host, struct mmc_request *mrq)
458{
459	int sg_count;
460	struct mmc_data *data = mrq->data;
461
462	if (!data)
463		return -EINVAL;
464
465	sg_count = dma_map_sg(mmc_dev(host), data->sg,
466			      data->sg_len,
467			      (data->flags & MMC_DATA_WRITE) ?
468			      DMA_TO_DEVICE : DMA_FROM_DEVICE);
469	if (!sg_count) {
470		pr_err("%s: sg-len: %d\n", __func__, data->sg_len);
471		return -ENOMEM;
472	}
473
474	return sg_count;
475}
476
477static void cqhci_set_tran_desc(u8 *desc, dma_addr_t addr, int len, bool end,
478				bool dma64)
479{
480	__le32 *attr = (__le32 __force *)desc;
481
482	*attr = (CQHCI_VALID(1) |
483		 CQHCI_END(end ? 1 : 0) |
484		 CQHCI_INT(0) |
485		 CQHCI_ACT(0x4) |
486		 CQHCI_DAT_LENGTH(len));
487
488	if (dma64) {
489		__le64 *dataddr = (__le64 __force *)(desc + 4);
490
491		dataddr[0] = cpu_to_le64(addr);
492	} else {
493		__le32 *dataddr = (__le32 __force *)(desc + 4);
494
495		dataddr[0] = cpu_to_le32(addr);
496	}
497}
498
499static int cqhci_prep_tran_desc(struct mmc_request *mrq,
500			       struct cqhci_host *cq_host, int tag)
501{
502	struct mmc_data *data = mrq->data;
503	int i, sg_count, len;
504	bool end = false;
505	bool dma64 = cq_host->dma64;
506	dma_addr_t addr;
507	u8 *desc;
508	struct scatterlist *sg;
509
510	sg_count = cqhci_dma_map(mrq->host, mrq);
511	if (sg_count < 0) {
512		pr_err("%s: %s: unable to map sg lists, %d\n",
513				mmc_hostname(mrq->host), __func__, sg_count);
514		return sg_count;
515	}
516
517	desc = get_trans_desc(cq_host, tag);
518
519	for_each_sg(data->sg, sg, sg_count, i) {
520		addr = sg_dma_address(sg);
521		len = sg_dma_len(sg);
522
523		if ((i+1) == sg_count)
524			end = true;
525		cqhci_set_tran_desc(desc, addr, len, end, dma64);
526		desc += cq_host->trans_desc_len;
527	}
528
529	return 0;
530}
531
532static void cqhci_prep_dcmd_desc(struct mmc_host *mmc,
533				   struct mmc_request *mrq)
534{
535	u64 *task_desc = NULL;
536	u64 data = 0;
537	u8 resp_type;
538	u8 *desc;
539	__le64 *dataddr;
540	struct cqhci_host *cq_host = mmc->cqe_private;
541	u8 timing;
542
543	if (!(mrq->cmd->flags & MMC_RSP_PRESENT)) {
544		resp_type = 0x0;
545		timing = 0x1;
546	} else {
547		if (mrq->cmd->flags & MMC_RSP_R1B) {
548			resp_type = 0x3;
549			timing = 0x0;
550		} else {
551			resp_type = 0x2;
552			timing = 0x1;
553		}
554	}
555
556	task_desc = (__le64 __force *)get_desc(cq_host, cq_host->dcmd_slot);
557	memset(task_desc, 0, cq_host->task_desc_len);
558	data |= (CQHCI_VALID(1) |
559		 CQHCI_END(1) |
560		 CQHCI_INT(1) |
561		 CQHCI_QBAR(1) |
562		 CQHCI_ACT(0x5) |
563		 CQHCI_CMD_INDEX(mrq->cmd->opcode) |
564		 CQHCI_CMD_TIMING(timing) | CQHCI_RESP_TYPE(resp_type));
565	if (cq_host->ops->update_dcmd_desc)
566		cq_host->ops->update_dcmd_desc(mmc, mrq, &data);
567	*task_desc |= data;
568	desc = (u8 *)task_desc;
569	pr_debug("%s: cqhci: dcmd: cmd: %d timing: %d resp: %d\n",
570		 mmc_hostname(mmc), mrq->cmd->opcode, timing, resp_type);
571	dataddr = (__le64 __force *)(desc + 4);
572	dataddr[0] = cpu_to_le64((u64)mrq->cmd->arg);
573
574}
575
576static void cqhci_post_req(struct mmc_host *host, struct mmc_request *mrq)
577{
578	struct mmc_data *data = mrq->data;
579
580	if (data) {
581		dma_unmap_sg(mmc_dev(host), data->sg, data->sg_len,
582			     (data->flags & MMC_DATA_READ) ?
583			     DMA_FROM_DEVICE : DMA_TO_DEVICE);
584	}
585}
586
587static inline int cqhci_tag(struct mmc_request *mrq)
588{
589	return mrq->cmd ? DCMD_SLOT : mrq->tag;
590}
591
592static int cqhci_request(struct mmc_host *mmc, struct mmc_request *mrq)
593{
594	int err = 0;
595	int tag = cqhci_tag(mrq);
596	struct cqhci_host *cq_host = mmc->cqe_private;
597	unsigned long flags;
598
599	if (!cq_host->enabled) {
600		pr_err("%s: cqhci: not enabled\n", mmc_hostname(mmc));
601		return -EINVAL;
602	}
603
604	/* First request after resume has to re-enable */
605	if (!cq_host->activated)
606		__cqhci_enable(cq_host);
607
608	if (!mmc->cqe_on) {
609		if (cq_host->ops->pre_enable)
610			cq_host->ops->pre_enable(mmc);
611
612		cqhci_writel(cq_host, 0, CQHCI_CTL);
613		mmc->cqe_on = true;
614		pr_debug("%s: cqhci: CQE on\n", mmc_hostname(mmc));
615		if (cqhci_readl(cq_host, CQHCI_CTL) && CQHCI_HALT) {
616			pr_err("%s: cqhci: CQE failed to exit halt state\n",
617			       mmc_hostname(mmc));
618		}
619		if (cq_host->ops->enable)
620			cq_host->ops->enable(mmc);
621	}
622
623	if (mrq->data) {
624		cqhci_prep_task_desc(mrq, cq_host, tag);
625
626		err = cqhci_prep_tran_desc(mrq, cq_host, tag);
627		if (err) {
628			pr_err("%s: cqhci: failed to setup tx desc: %d\n",
629			       mmc_hostname(mmc), err);
630			return err;
631		}
632	} else {
633		cqhci_prep_dcmd_desc(mmc, mrq);
634	}
635
636	spin_lock_irqsave(&cq_host->lock, flags);
637
638	if (cq_host->recovery_halt) {
639		err = -EBUSY;
640		goto out_unlock;
641	}
642
643	cq_host->slot[tag].mrq = mrq;
644	cq_host->slot[tag].flags = 0;
645
646	cq_host->qcnt += 1;
647	/* Make sure descriptors are ready before ringing the doorbell */
648	wmb();
649	cqhci_writel(cq_host, 1 << tag, CQHCI_TDBR);
650	if (!(cqhci_readl(cq_host, CQHCI_TDBR) & (1 << tag)))
651		pr_debug("%s: cqhci: doorbell not set for tag %d\n",
652			 mmc_hostname(mmc), tag);
653out_unlock:
654	spin_unlock_irqrestore(&cq_host->lock, flags);
655
656	if (err)
657		cqhci_post_req(mmc, mrq);
658
659	return err;
660}
661
662static void cqhci_recovery_needed(struct mmc_host *mmc, struct mmc_request *mrq,
663				  bool notify)
664{
665	struct cqhci_host *cq_host = mmc->cqe_private;
666
667	if (!cq_host->recovery_halt) {
668		cq_host->recovery_halt = true;
669		pr_debug("%s: cqhci: recovery needed\n", mmc_hostname(mmc));
670		wake_up(&cq_host->wait_queue);
671		if (notify && mrq->recovery_notifier)
672			mrq->recovery_notifier(mrq);
673	}
674}
675
676static unsigned int cqhci_error_flags(int error1, int error2)
677{
678	int error = error1 ? error1 : error2;
679
680	switch (error) {
681	case -EILSEQ:
682		return CQHCI_HOST_CRC;
683	case -ETIMEDOUT:
684		return CQHCI_HOST_TIMEOUT;
685	default:
686		return CQHCI_HOST_OTHER;
687	}
688}
689
690static void cqhci_error_irq(struct mmc_host *mmc, u32 status, int cmd_error,
691			    int data_error)
692{
693	struct cqhci_host *cq_host = mmc->cqe_private;
694	struct cqhci_slot *slot;
695	u32 terri;
696	u32 tdpe;
697	int tag;
698
699	spin_lock(&cq_host->lock);
700
701	terri = cqhci_readl(cq_host, CQHCI_TERRI);
702
703	pr_debug("%s: cqhci: error IRQ status: 0x%08x cmd error %d data error %d TERRI: 0x%08x\n",
704		 mmc_hostname(mmc), status, cmd_error, data_error, terri);
705
706	/* Forget about errors when recovery has already been triggered */
707	if (cq_host->recovery_halt)
708		goto out_unlock;
709
710	if (!cq_host->qcnt) {
711		WARN_ONCE(1, "%s: cqhci: error when idle. IRQ status: 0x%08x cmd error %d data error %d TERRI: 0x%08x\n",
712			  mmc_hostname(mmc), status, cmd_error, data_error,
713			  terri);
714		goto out_unlock;
715	}
716
717	if (CQHCI_TERRI_C_VALID(terri)) {
718		tag = CQHCI_TERRI_C_TASK(terri);
719		slot = &cq_host->slot[tag];
720		if (slot->mrq) {
721			slot->flags = cqhci_error_flags(cmd_error, data_error);
722			cqhci_recovery_needed(mmc, slot->mrq, true);
723		}
724	}
725
726	if (CQHCI_TERRI_D_VALID(terri)) {
727		tag = CQHCI_TERRI_D_TASK(terri);
728		slot = &cq_host->slot[tag];
729		if (slot->mrq) {
730			slot->flags = cqhci_error_flags(data_error, cmd_error);
731			cqhci_recovery_needed(mmc, slot->mrq, true);
732		}
733	}
734
735	/*
736	 * Handle ICCE ("Invalid Crypto Configuration Error").  This should
737	 * never happen, since the block layer ensures that all crypto-enabled
738	 * I/O requests have a valid keyslot before they reach the driver.
739	 *
740	 * Note that GCE ("General Crypto Error") is different; it already got
741	 * handled above by checking TERRI.
742	 */
743	if (status & CQHCI_IS_ICCE) {
744		tdpe = cqhci_readl(cq_host, CQHCI_TDPE);
745		WARN_ONCE(1,
746			  "%s: cqhci: invalid crypto configuration error. IRQ status: 0x%08x TDPE: 0x%08x\n",
747			  mmc_hostname(mmc), status, tdpe);
748		while (tdpe != 0) {
749			tag = __ffs(tdpe);
750			tdpe &= ~(1 << tag);
751			slot = &cq_host->slot[tag];
752			if (!slot->mrq)
753				continue;
754			slot->flags = cqhci_error_flags(data_error, cmd_error);
755			cqhci_recovery_needed(mmc, slot->mrq, true);
756		}
757	}
758
759	if (!cq_host->recovery_halt) {
760		/*
761		 * The only way to guarantee forward progress is to mark at
762		 * least one task in error, so if none is indicated, pick one.
763		 */
764		for (tag = 0; tag < NUM_SLOTS; tag++) {
765			slot = &cq_host->slot[tag];
766			if (!slot->mrq)
767				continue;
768			slot->flags = cqhci_error_flags(data_error, cmd_error);
769			cqhci_recovery_needed(mmc, slot->mrq, true);
770			break;
771		}
772	}
773
774out_unlock:
775	spin_unlock(&cq_host->lock);
776}
777
778static void cqhci_finish_mrq(struct mmc_host *mmc, unsigned int tag)
779{
780	struct cqhci_host *cq_host = mmc->cqe_private;
781	struct cqhci_slot *slot = &cq_host->slot[tag];
782	struct mmc_request *mrq = slot->mrq;
783	struct mmc_data *data;
784
785	if (!mrq) {
786		WARN_ONCE(1, "%s: cqhci: spurious TCN for tag %d\n",
787			  mmc_hostname(mmc), tag);
788		return;
789	}
790
791	/* No completions allowed during recovery */
792	if (cq_host->recovery_halt) {
793		slot->flags |= CQHCI_COMPLETED;
794		return;
795	}
796
797	slot->mrq = NULL;
798
799	cq_host->qcnt -= 1;
800
801	data = mrq->data;
802	if (data) {
803		if (data->error)
804			data->bytes_xfered = 0;
805		else
806			data->bytes_xfered = data->blksz * data->blocks;
807	}
808
809	mmc_cqe_request_done(mmc, mrq);
810}
811
812irqreturn_t cqhci_irq(struct mmc_host *mmc, u32 intmask, int cmd_error,
813		      int data_error)
814{
815	u32 status;
816	unsigned long tag = 0, comp_status;
817	struct cqhci_host *cq_host = mmc->cqe_private;
818
819	status = cqhci_readl(cq_host, CQHCI_IS);
820	cqhci_writel(cq_host, status, CQHCI_IS);
821
822	pr_debug("%s: cqhci: IRQ status: 0x%08x\n", mmc_hostname(mmc), status);
823
824	if ((status & (CQHCI_IS_RED | CQHCI_IS_GCE | CQHCI_IS_ICCE)) ||
825	    cmd_error || data_error) {
826		if (status & CQHCI_IS_RED)
827			mmc_debugfs_err_stats_inc(mmc, MMC_ERR_CMDQ_RED);
828		if (status & CQHCI_IS_GCE)
829			mmc_debugfs_err_stats_inc(mmc, MMC_ERR_CMDQ_GCE);
830		if (status & CQHCI_IS_ICCE)
831			mmc_debugfs_err_stats_inc(mmc, MMC_ERR_CMDQ_ICCE);
832		cqhci_error_irq(mmc, status, cmd_error, data_error);
833	}
834
835	if (status & CQHCI_IS_TCC) {
836		/* read TCN and complete the request */
837		comp_status = cqhci_readl(cq_host, CQHCI_TCN);
838		cqhci_writel(cq_host, comp_status, CQHCI_TCN);
839		pr_debug("%s: cqhci: TCN: 0x%08lx\n",
840			 mmc_hostname(mmc), comp_status);
841
842		spin_lock(&cq_host->lock);
843
844		for_each_set_bit(tag, &comp_status, cq_host->num_slots) {
845			/* complete the corresponding mrq */
846			pr_debug("%s: cqhci: completing tag %lu\n",
847				 mmc_hostname(mmc), tag);
848			cqhci_finish_mrq(mmc, tag);
849		}
850
851		if (cq_host->waiting_for_idle && !cq_host->qcnt) {
852			cq_host->waiting_for_idle = false;
853			wake_up(&cq_host->wait_queue);
854		}
855
856		spin_unlock(&cq_host->lock);
857	}
858
859	if (status & CQHCI_IS_TCL)
860		wake_up(&cq_host->wait_queue);
861
862	if (status & CQHCI_IS_HAC)
863		wake_up(&cq_host->wait_queue);
864
865	return IRQ_HANDLED;
866}
867EXPORT_SYMBOL(cqhci_irq);
868
869static bool cqhci_is_idle(struct cqhci_host *cq_host, int *ret)
870{
871	unsigned long flags;
872	bool is_idle;
873
874	spin_lock_irqsave(&cq_host->lock, flags);
875	is_idle = !cq_host->qcnt || cq_host->recovery_halt;
876	*ret = cq_host->recovery_halt ? -EBUSY : 0;
877	cq_host->waiting_for_idle = !is_idle;
878	spin_unlock_irqrestore(&cq_host->lock, flags);
879
880	return is_idle;
881}
882
883static int cqhci_wait_for_idle(struct mmc_host *mmc)
884{
885	struct cqhci_host *cq_host = mmc->cqe_private;
886	int ret;
887
888	wait_event(cq_host->wait_queue, cqhci_is_idle(cq_host, &ret));
889
890	return ret;
891}
892
893static bool cqhci_timeout(struct mmc_host *mmc, struct mmc_request *mrq,
894			  bool *recovery_needed)
895{
896	struct cqhci_host *cq_host = mmc->cqe_private;
897	int tag = cqhci_tag(mrq);
898	struct cqhci_slot *slot = &cq_host->slot[tag];
899	unsigned long flags;
900	bool timed_out;
901
902	spin_lock_irqsave(&cq_host->lock, flags);
903	timed_out = slot->mrq == mrq;
904	if (timed_out) {
905		slot->flags |= CQHCI_EXTERNAL_TIMEOUT;
906		cqhci_recovery_needed(mmc, mrq, false);
907		*recovery_needed = cq_host->recovery_halt;
908	}
909	spin_unlock_irqrestore(&cq_host->lock, flags);
910
911	if (timed_out) {
912		pr_err("%s: cqhci: timeout for tag %d, qcnt %d\n",
913		       mmc_hostname(mmc), tag, cq_host->qcnt);
914		cqhci_dumpregs(cq_host);
915	}
916
917	return timed_out;
918}
919
920static bool cqhci_tasks_cleared(struct cqhci_host *cq_host)
921{
922	return !(cqhci_readl(cq_host, CQHCI_CTL) & CQHCI_CLEAR_ALL_TASKS);
923}
924
925static bool cqhci_clear_all_tasks(struct mmc_host *mmc, unsigned int timeout)
926{
927	struct cqhci_host *cq_host = mmc->cqe_private;
928	bool ret;
929	u32 ctl;
930
931	cqhci_set_irqs(cq_host, CQHCI_IS_TCL);
932
933	ctl = cqhci_readl(cq_host, CQHCI_CTL);
934	ctl |= CQHCI_CLEAR_ALL_TASKS;
935	cqhci_writel(cq_host, ctl, CQHCI_CTL);
936
937	wait_event_timeout(cq_host->wait_queue, cqhci_tasks_cleared(cq_host),
938			   msecs_to_jiffies(timeout) + 1);
939
940	cqhci_set_irqs(cq_host, 0);
941
942	ret = cqhci_tasks_cleared(cq_host);
943
944	if (!ret)
945		pr_warn("%s: cqhci: Failed to clear tasks\n",
946			mmc_hostname(mmc));
947
948	return ret;
949}
950
951static bool cqhci_halted(struct cqhci_host *cq_host)
952{
953	return cqhci_readl(cq_host, CQHCI_CTL) & CQHCI_HALT;
954}
955
956static bool cqhci_halt(struct mmc_host *mmc, unsigned int timeout)
957{
958	struct cqhci_host *cq_host = mmc->cqe_private;
959	bool ret;
960	u32 ctl;
961
962	if (cqhci_halted(cq_host))
963		return true;
964
965	cqhci_set_irqs(cq_host, CQHCI_IS_HAC);
966
967	ctl = cqhci_readl(cq_host, CQHCI_CTL);
968	ctl |= CQHCI_HALT;
969	cqhci_writel(cq_host, ctl, CQHCI_CTL);
970
971	wait_event_timeout(cq_host->wait_queue, cqhci_halted(cq_host),
972			   msecs_to_jiffies(timeout) + 1);
973
974	cqhci_set_irqs(cq_host, 0);
975
976	ret = cqhci_halted(cq_host);
977
978	if (!ret)
979		pr_warn("%s: cqhci: Failed to halt\n", mmc_hostname(mmc));
980
981	return ret;
982}
983
984/*
985 * After halting we expect to be able to use the command line. We interpret the
986 * failure to halt to mean the data lines might still be in use (and the upper
987 * layers will need to send a STOP command), however failing to halt complicates
988 * the recovery, so set a timeout that would reasonably allow I/O to complete.
989 */
990#define CQHCI_START_HALT_TIMEOUT	500
991
992static void cqhci_recovery_start(struct mmc_host *mmc)
993{
994	struct cqhci_host *cq_host = mmc->cqe_private;
995
996	pr_debug("%s: cqhci: %s\n", mmc_hostname(mmc), __func__);
997
998	WARN_ON(!cq_host->recovery_halt);
999
1000	cqhci_halt(mmc, CQHCI_START_HALT_TIMEOUT);
1001
1002	if (cq_host->ops->disable)
1003		cq_host->ops->disable(mmc, true);
1004
1005	mmc->cqe_on = false;
1006}
1007
1008static int cqhci_error_from_flags(unsigned int flags)
1009{
1010	if (!flags)
1011		return 0;
1012
1013	/* CRC errors might indicate re-tuning so prefer to report that */
1014	if (flags & CQHCI_HOST_CRC)
1015		return -EILSEQ;
1016
1017	if (flags & (CQHCI_EXTERNAL_TIMEOUT | CQHCI_HOST_TIMEOUT))
1018		return -ETIMEDOUT;
1019
1020	return -EIO;
1021}
1022
1023static void cqhci_recover_mrq(struct cqhci_host *cq_host, unsigned int tag)
1024{
1025	struct cqhci_slot *slot = &cq_host->slot[tag];
1026	struct mmc_request *mrq = slot->mrq;
1027	struct mmc_data *data;
1028
1029	if (!mrq)
1030		return;
1031
1032	slot->mrq = NULL;
1033
1034	cq_host->qcnt -= 1;
1035
1036	data = mrq->data;
1037	if (data) {
1038		data->bytes_xfered = 0;
1039		data->error = cqhci_error_from_flags(slot->flags);
1040	} else {
1041		mrq->cmd->error = cqhci_error_from_flags(slot->flags);
1042	}
1043
1044	mmc_cqe_request_done(cq_host->mmc, mrq);
1045}
1046
1047static void cqhci_recover_mrqs(struct cqhci_host *cq_host)
1048{
1049	int i;
1050
1051	for (i = 0; i < cq_host->num_slots; i++)
1052		cqhci_recover_mrq(cq_host, i);
1053}
1054
1055/*
1056 * By now the command and data lines should be unused so there is no reason for
1057 * CQHCI to take a long time to halt, but if it doesn't halt there could be
1058 * problems clearing tasks, so be generous.
1059 */
1060#define CQHCI_FINISH_HALT_TIMEOUT	20
1061
1062/* CQHCI could be expected to clear it's internal state pretty quickly */
1063#define CQHCI_CLEAR_TIMEOUT		20
1064
1065static void cqhci_recovery_finish(struct mmc_host *mmc)
1066{
1067	struct cqhci_host *cq_host = mmc->cqe_private;
1068	unsigned long flags;
1069	u32 cqcfg;
1070	bool ok;
1071
1072	pr_debug("%s: cqhci: %s\n", mmc_hostname(mmc), __func__);
1073
1074	WARN_ON(!cq_host->recovery_halt);
1075
1076	ok = cqhci_halt(mmc, CQHCI_FINISH_HALT_TIMEOUT);
1077
1078	/*
1079	 * The specification contradicts itself, by saying that tasks cannot be
1080	 * cleared if CQHCI does not halt, but if CQHCI does not halt, it should
1081	 * be disabled/re-enabled, but not to disable before clearing tasks.
1082	 * Have a go anyway.
1083	 */
1084	if (!cqhci_clear_all_tasks(mmc, CQHCI_CLEAR_TIMEOUT))
1085		ok = false;
1086
1087	/* Disable to make sure tasks really are cleared */
1088	cqcfg = cqhci_readl(cq_host, CQHCI_CFG);
1089	cqcfg &= ~CQHCI_ENABLE;
1090	cqhci_writel(cq_host, cqcfg, CQHCI_CFG);
1091
1092	cqcfg = cqhci_readl(cq_host, CQHCI_CFG);
1093	cqcfg |= CQHCI_ENABLE;
1094	cqhci_writel(cq_host, cqcfg, CQHCI_CFG);
1095
1096	cqhci_halt(mmc, CQHCI_FINISH_HALT_TIMEOUT);
1097
1098	if (!ok)
1099		cqhci_clear_all_tasks(mmc, CQHCI_CLEAR_TIMEOUT);
1100
1101	cqhci_recover_mrqs(cq_host);
1102
1103	WARN_ON(cq_host->qcnt);
1104
1105	spin_lock_irqsave(&cq_host->lock, flags);
1106	cq_host->qcnt = 0;
1107	cq_host->recovery_halt = false;
1108	mmc->cqe_on = false;
1109	spin_unlock_irqrestore(&cq_host->lock, flags);
1110
1111	/* Ensure all writes are done before interrupts are re-enabled */
1112	wmb();
1113
1114	cqhci_writel(cq_host, CQHCI_IS_HAC | CQHCI_IS_TCL, CQHCI_IS);
1115
1116	cqhci_set_irqs(cq_host, CQHCI_IS_MASK);
1117
1118	pr_debug("%s: cqhci: recovery done\n", mmc_hostname(mmc));
1119}
1120
1121static const struct mmc_cqe_ops cqhci_cqe_ops = {
1122	.cqe_enable = cqhci_enable,
1123	.cqe_disable = cqhci_disable,
1124	.cqe_request = cqhci_request,
1125	.cqe_post_req = cqhci_post_req,
1126	.cqe_off = cqhci_off,
1127	.cqe_wait_for_idle = cqhci_wait_for_idle,
1128	.cqe_timeout = cqhci_timeout,
1129	.cqe_recovery_start = cqhci_recovery_start,
1130	.cqe_recovery_finish = cqhci_recovery_finish,
1131};
1132
1133struct cqhci_host *cqhci_pltfm_init(struct platform_device *pdev)
1134{
1135	struct cqhci_host *cq_host;
1136	struct resource *cqhci_memres = NULL;
1137
1138	/* check and setup CMDQ interface */
1139	cqhci_memres = platform_get_resource_byname(pdev, IORESOURCE_MEM,
1140						   "cqhci");
1141	if (!cqhci_memres) {
1142		dev_dbg(&pdev->dev, "CMDQ not supported\n");
1143		return ERR_PTR(-EINVAL);
1144	}
1145
1146	cq_host = devm_kzalloc(&pdev->dev, sizeof(*cq_host), GFP_KERNEL);
1147	if (!cq_host)
1148		return ERR_PTR(-ENOMEM);
1149	cq_host->mmio = devm_ioremap(&pdev->dev,
1150				     cqhci_memres->start,
1151				     resource_size(cqhci_memres));
1152	if (!cq_host->mmio) {
1153		dev_err(&pdev->dev, "failed to remap cqhci regs\n");
1154		return ERR_PTR(-EBUSY);
1155	}
1156	dev_dbg(&pdev->dev, "CMDQ ioremap: done\n");
1157
1158	return cq_host;
1159}
1160EXPORT_SYMBOL(cqhci_pltfm_init);
1161
1162static unsigned int cqhci_ver_major(struct cqhci_host *cq_host)
1163{
1164	return CQHCI_VER_MAJOR(cqhci_readl(cq_host, CQHCI_VER));
1165}
1166
1167static unsigned int cqhci_ver_minor(struct cqhci_host *cq_host)
1168{
1169	u32 ver = cqhci_readl(cq_host, CQHCI_VER);
1170
1171	return CQHCI_VER_MINOR1(ver) * 10 + CQHCI_VER_MINOR2(ver);
1172}
1173
1174int cqhci_init(struct cqhci_host *cq_host, struct mmc_host *mmc,
1175	      bool dma64)
1176{
1177	int err;
1178
1179	cq_host->dma64 = dma64;
1180	cq_host->mmc = mmc;
1181	cq_host->mmc->cqe_private = cq_host;
1182
1183	cq_host->num_slots = NUM_SLOTS;
1184	cq_host->dcmd_slot = DCMD_SLOT;
1185
1186	mmc->cqe_ops = &cqhci_cqe_ops;
1187
1188	mmc->cqe_qdepth = NUM_SLOTS;
1189	if (mmc->caps2 & MMC_CAP2_CQE_DCMD)
1190		mmc->cqe_qdepth -= 1;
1191
1192	cq_host->slot = devm_kcalloc(mmc_dev(mmc), cq_host->num_slots,
1193				     sizeof(*cq_host->slot), GFP_KERNEL);
1194	if (!cq_host->slot) {
1195		err = -ENOMEM;
1196		goto out_err;
1197	}
1198
1199	err = cqhci_crypto_init(cq_host);
1200	if (err) {
1201		pr_err("%s: CQHCI crypto initialization failed\n",
1202		       mmc_hostname(mmc));
1203		goto out_err;
1204	}
1205
1206	spin_lock_init(&cq_host->lock);
1207
1208	init_completion(&cq_host->halt_comp);
1209	init_waitqueue_head(&cq_host->wait_queue);
1210
1211	pr_info("%s: CQHCI version %u.%02u\n",
1212		mmc_hostname(mmc), cqhci_ver_major(cq_host),
1213		cqhci_ver_minor(cq_host));
1214
1215	return 0;
1216
1217out_err:
1218	pr_err("%s: CQHCI version %u.%02u failed to initialize, error %d\n",
1219	       mmc_hostname(mmc), cqhci_ver_major(cq_host),
1220	       cqhci_ver_minor(cq_host), err);
1221	return err;
1222}
1223EXPORT_SYMBOL(cqhci_init);
1224
1225MODULE_AUTHOR("Venkat Gopalakrishnan <venkatg@codeaurora.org>");
1226MODULE_DESCRIPTION("Command Queue Host Controller Interface driver");
1227MODULE_LICENSE("GPL v2");
1228