xref: /linux/drivers/i3c/master/mipi-i3c-hci/dma.c (revision 55a42f78ffd386e01a5404419f8c5ded7db70a21)
1 // SPDX-License-Identifier: BSD-3-Clause
2 /*
3  * Copyright (c) 2020, MIPI Alliance, Inc.
4  *
5  * Author: Nicolas Pitre <npitre@baylibre.com>
6  *
7  * Note: The I3C HCI v2.0 spec is still in flux. The IBI support is based on
8  * v1.x of the spec and v2.0 will likely be split out.
9  */
10 
11 #include <linux/bitfield.h>
12 #include <linux/device.h>
13 #include <linux/dma-mapping.h>
14 #include <linux/errno.h>
15 #include <linux/i3c/master.h>
16 #include <linux/io.h>
17 #include <linux/pci.h>
18 
19 #include "hci.h"
20 #include "cmd.h"
21 #include "ibi.h"
22 
23 
24 /*
25  * Software Parameter Values (somewhat arb itrary for now).
26  * Some of them could be determined at run time eventually.
27  */
28 
29 #define XFER_RINGS			1	/* max: 8 */
30 #define XFER_RING_ENTRIES		16	/* max: 255 */
31 
32 #define IBI_RINGS			1	/* max: 8 */
33 #define IBI_STATUS_RING_ENTRIES		32	/* max: 255 */
34 #define IBI_CHUNK_CACHELINES		1	/* max: 256 bytes equivalent */
35 #define IBI_CHUNK_POOL_SIZE		128	/* max: 1023 */
36 
37 /*
38  * Ring Header Preamble
39  */
40 
41 #define rhs_reg_read(r)		readl(hci->RHS_regs + (RHS_##r))
42 #define rhs_reg_write(r, v)	writel(v, hci->RHS_regs + (RHS_##r))
43 
44 #define RHS_CONTROL			0x00
45 #define PREAMBLE_SIZE			GENMASK(31, 24)	/* Preamble Section Size */
46 #define HEADER_SIZE			GENMASK(23, 16)	/* Ring Header Size */
47 #define MAX_HEADER_COUNT_CAP		GENMASK(7, 4) /* HC Max Header Count */
48 #define MAX_HEADER_COUNT		GENMASK(3, 0) /* Driver Max Header Count */
49 
50 #define RHS_RHn_OFFSET(n)		(0x04 + (n)*4)
51 
52 /*
53  * Ring Header (Per-Ring Bundle)
54  */
55 
56 #define rh_reg_read(r)		readl(rh->regs + (RH_##r))
57 #define rh_reg_write(r, v)	writel(v, rh->regs + (RH_##r))
58 
59 #define RH_CR_SETUP			0x00	/* Command/Response Ring */
60 #define CR_XFER_STRUCT_SIZE		GENMASK(31, 24)
61 #define CR_RESP_STRUCT_SIZE		GENMASK(23, 16)
62 #define CR_RING_SIZE			GENMASK(8, 0)
63 
64 #define RH_IBI_SETUP			0x04
65 #define IBI_STATUS_STRUCT_SIZE		GENMASK(31, 24)
66 #define IBI_STATUS_RING_SIZE		GENMASK(23, 16)
67 #define IBI_DATA_CHUNK_SIZE		GENMASK(12, 10)
68 #define IBI_DATA_CHUNK_COUNT		GENMASK(9, 0)
69 
70 #define RH_CHUNK_CONTROL			0x08
71 
72 #define RH_INTR_STATUS			0x10
73 #define RH_INTR_STATUS_ENABLE		0x14
74 #define RH_INTR_SIGNAL_ENABLE		0x18
75 #define RH_INTR_FORCE			0x1c
76 #define INTR_IBI_READY			BIT(12)
77 #define INTR_TRANSFER_COMPLETION	BIT(11)
78 #define INTR_RING_OP			BIT(10)
79 #define INTR_TRANSFER_ERR		BIT(9)
80 #define INTR_IBI_RING_FULL		BIT(6)
81 #define INTR_TRANSFER_ABORT		BIT(5)
82 
83 #define RH_RING_STATUS			0x20
84 #define RING_STATUS_LOCKED		BIT(3)
85 #define RING_STATUS_ABORTED		BIT(2)
86 #define RING_STATUS_RUNNING		BIT(1)
87 #define RING_STATUS_ENABLED		BIT(0)
88 
89 #define RH_RING_CONTROL			0x24
90 #define RING_CTRL_ABORT			BIT(2)
91 #define RING_CTRL_RUN_STOP		BIT(1)
92 #define RING_CTRL_ENABLE		BIT(0)
93 
94 #define RH_RING_OPERATION1		0x28
95 #define RING_OP1_IBI_DEQ_PTR		GENMASK(23, 16)
96 #define RING_OP1_CR_SW_DEQ_PTR		GENMASK(15, 8)
97 #define RING_OP1_CR_ENQ_PTR		GENMASK(7, 0)
98 
99 #define RH_RING_OPERATION2		0x2c
100 #define RING_OP2_IBI_ENQ_PTR		GENMASK(23, 16)
101 #define RING_OP2_CR_DEQ_PTR		GENMASK(7, 0)
102 
103 #define RH_CMD_RING_BASE_LO		0x30
104 #define RH_CMD_RING_BASE_HI		0x34
105 #define RH_RESP_RING_BASE_LO		0x38
106 #define RH_RESP_RING_BASE_HI		0x3c
107 #define RH_IBI_STATUS_RING_BASE_LO	0x40
108 #define RH_IBI_STATUS_RING_BASE_HI	0x44
109 #define RH_IBI_DATA_RING_BASE_LO	0x48
110 #define RH_IBI_DATA_RING_BASE_HI	0x4c
111 
112 #define RH_CMD_RING_SG			0x50	/* Ring Scatter Gather Support */
113 #define RH_RESP_RING_SG			0x54
114 #define RH_IBI_STATUS_RING_SG		0x58
115 #define RH_IBI_DATA_RING_SG		0x5c
116 #define RING_SG_BLP			BIT(31)	/* Buffer Vs. List Pointer */
117 #define RING_SG_LIST_SIZE		GENMASK(15, 0)
118 
119 /*
120  * Data Buffer Descriptor (in memory)
121  */
122 
123 #define DATA_BUF_BLP			BIT(31)	/* Buffer Vs. List Pointer */
124 #define DATA_BUF_IOC			BIT(30)	/* Interrupt on Completion */
125 #define DATA_BUF_BLOCK_SIZE		GENMASK(15, 0)
126 
127 
128 struct hci_rh_data {
129 	void __iomem *regs;
130 	void *xfer, *resp, *ibi_status, *ibi_data;
131 	dma_addr_t xfer_dma, resp_dma, ibi_status_dma, ibi_data_dma;
132 	unsigned int xfer_entries, ibi_status_entries, ibi_chunks_total;
133 	unsigned int xfer_struct_sz, resp_struct_sz, ibi_status_sz, ibi_chunk_sz;
134 	unsigned int done_ptr, ibi_chunk_ptr;
135 	struct hci_xfer **src_xfers;
136 	spinlock_t lock;
137 	struct completion op_done;
138 };
139 
140 struct hci_rings_data {
141 	struct device *sysdev;
142 	unsigned int total;
143 	struct hci_rh_data headers[] __counted_by(total);
144 };
145 
146 struct hci_dma_dev_ibi_data {
147 	struct i3c_generic_ibi_pool *pool;
148 	unsigned int max_len;
149 };
150 
151 static void hci_dma_cleanup(struct i3c_hci *hci)
152 {
153 	struct hci_rings_data *rings = hci->io_data;
154 	struct hci_rh_data *rh;
155 	unsigned int i;
156 
157 	if (!rings)
158 		return;
159 
160 	for (i = 0; i < rings->total; i++) {
161 		rh = &rings->headers[i];
162 
163 		rh_reg_write(INTR_SIGNAL_ENABLE, 0);
164 		rh_reg_write(RING_CONTROL, 0);
165 		rh_reg_write(CR_SETUP, 0);
166 		rh_reg_write(IBI_SETUP, 0);
167 
168 		if (rh->xfer)
169 			dma_free_coherent(rings->sysdev,
170 					  rh->xfer_struct_sz * rh->xfer_entries,
171 					  rh->xfer, rh->xfer_dma);
172 		if (rh->resp)
173 			dma_free_coherent(rings->sysdev,
174 					  rh->resp_struct_sz * rh->xfer_entries,
175 					  rh->resp, rh->resp_dma);
176 		kfree(rh->src_xfers);
177 		if (rh->ibi_status)
178 			dma_free_coherent(rings->sysdev,
179 					  rh->ibi_status_sz * rh->ibi_status_entries,
180 					  rh->ibi_status, rh->ibi_status_dma);
181 		if (rh->ibi_data_dma)
182 			dma_unmap_single(rings->sysdev, rh->ibi_data_dma,
183 					 rh->ibi_chunk_sz * rh->ibi_chunks_total,
184 					 DMA_FROM_DEVICE);
185 		kfree(rh->ibi_data);
186 	}
187 
188 	rhs_reg_write(CONTROL, 0);
189 
190 	kfree(rings);
191 	hci->io_data = NULL;
192 }
193 
194 static int hci_dma_init(struct i3c_hci *hci)
195 {
196 	struct hci_rings_data *rings;
197 	struct hci_rh_data *rh;
198 	struct device *sysdev;
199 	u32 regval;
200 	unsigned int i, nr_rings, xfers_sz, resps_sz;
201 	unsigned int ibi_status_ring_sz, ibi_data_ring_sz;
202 	int ret;
203 
204 	/*
205 	 * Set pointer to a physical device that does DMA and has IOMMU setup
206 	 * done for it in case of enabled IOMMU and use it with the DMA API.
207 	 * Here such device is either
208 	 * "mipi-i3c-hci" platform device (OF/ACPI enumeration) parent or
209 	 * grandparent (PCI enumeration).
210 	 */
211 	sysdev = hci->master.dev.parent;
212 	if (sysdev->parent && dev_is_pci(sysdev->parent))
213 		sysdev = sysdev->parent;
214 
215 	regval = rhs_reg_read(CONTROL);
216 	nr_rings = FIELD_GET(MAX_HEADER_COUNT_CAP, regval);
217 	dev_info(&hci->master.dev, "%d DMA rings available\n", nr_rings);
218 	if (unlikely(nr_rings > 8)) {
219 		dev_err(&hci->master.dev, "number of rings should be <= 8\n");
220 		nr_rings = 8;
221 	}
222 	if (nr_rings > XFER_RINGS)
223 		nr_rings = XFER_RINGS;
224 	rings = kzalloc(struct_size(rings, headers, nr_rings), GFP_KERNEL);
225 	if (!rings)
226 		return -ENOMEM;
227 	hci->io_data = rings;
228 	rings->total = nr_rings;
229 	rings->sysdev = sysdev;
230 
231 	regval = FIELD_PREP(MAX_HEADER_COUNT, rings->total);
232 	rhs_reg_write(CONTROL, regval);
233 
234 	for (i = 0; i < rings->total; i++) {
235 		u32 offset = rhs_reg_read(RHn_OFFSET(i));
236 
237 		dev_info(&hci->master.dev, "Ring %d at offset %#x\n", i, offset);
238 		ret = -EINVAL;
239 		if (!offset)
240 			goto err_out;
241 		rh = &rings->headers[i];
242 		rh->regs = hci->base_regs + offset;
243 		spin_lock_init(&rh->lock);
244 		init_completion(&rh->op_done);
245 
246 		rh->xfer_entries = XFER_RING_ENTRIES;
247 
248 		regval = rh_reg_read(CR_SETUP);
249 		rh->xfer_struct_sz = FIELD_GET(CR_XFER_STRUCT_SIZE, regval);
250 		rh->resp_struct_sz = FIELD_GET(CR_RESP_STRUCT_SIZE, regval);
251 		dev_dbg(&hci->master.dev,
252 			"xfer_struct_sz = %d, resp_struct_sz = %d",
253 			rh->xfer_struct_sz, rh->resp_struct_sz);
254 		xfers_sz = rh->xfer_struct_sz * rh->xfer_entries;
255 		resps_sz = rh->resp_struct_sz * rh->xfer_entries;
256 
257 		rh->xfer = dma_alloc_coherent(rings->sysdev, xfers_sz,
258 					      &rh->xfer_dma, GFP_KERNEL);
259 		rh->resp = dma_alloc_coherent(rings->sysdev, resps_sz,
260 					      &rh->resp_dma, GFP_KERNEL);
261 		rh->src_xfers =
262 			kmalloc_array(rh->xfer_entries, sizeof(*rh->src_xfers),
263 				      GFP_KERNEL);
264 		ret = -ENOMEM;
265 		if (!rh->xfer || !rh->resp || !rh->src_xfers)
266 			goto err_out;
267 
268 		rh_reg_write(CMD_RING_BASE_LO, lower_32_bits(rh->xfer_dma));
269 		rh_reg_write(CMD_RING_BASE_HI, upper_32_bits(rh->xfer_dma));
270 		rh_reg_write(RESP_RING_BASE_LO, lower_32_bits(rh->resp_dma));
271 		rh_reg_write(RESP_RING_BASE_HI, upper_32_bits(rh->resp_dma));
272 
273 		regval = FIELD_PREP(CR_RING_SIZE, rh->xfer_entries);
274 		rh_reg_write(CR_SETUP, regval);
275 
276 		rh_reg_write(INTR_STATUS_ENABLE, 0xffffffff);
277 		rh_reg_write(INTR_SIGNAL_ENABLE, INTR_IBI_READY |
278 						 INTR_TRANSFER_COMPLETION |
279 						 INTR_RING_OP |
280 						 INTR_TRANSFER_ERR |
281 						 INTR_IBI_RING_FULL |
282 						 INTR_TRANSFER_ABORT);
283 
284 		/* IBIs */
285 
286 		if (i >= IBI_RINGS)
287 			goto ring_ready;
288 
289 		regval = rh_reg_read(IBI_SETUP);
290 		rh->ibi_status_sz = FIELD_GET(IBI_STATUS_STRUCT_SIZE, regval);
291 		rh->ibi_status_entries = IBI_STATUS_RING_ENTRIES;
292 		rh->ibi_chunks_total = IBI_CHUNK_POOL_SIZE;
293 
294 		rh->ibi_chunk_sz = dma_get_cache_alignment();
295 		rh->ibi_chunk_sz *= IBI_CHUNK_CACHELINES;
296 		/*
297 		 * Round IBI data chunk size to number of bytes supported by
298 		 * the HW. Chunk size can be 2^n number of DWORDs which is the
299 		 * same as 2^(n+2) bytes, where n is 0..6.
300 		 */
301 		rh->ibi_chunk_sz = umax(4, rh->ibi_chunk_sz);
302 		rh->ibi_chunk_sz = roundup_pow_of_two(rh->ibi_chunk_sz);
303 		if (rh->ibi_chunk_sz > 256) {
304 			ret = -EINVAL;
305 			goto err_out;
306 		}
307 
308 		ibi_status_ring_sz = rh->ibi_status_sz * rh->ibi_status_entries;
309 		ibi_data_ring_sz = rh->ibi_chunk_sz * rh->ibi_chunks_total;
310 
311 		rh->ibi_status =
312 			dma_alloc_coherent(rings->sysdev, ibi_status_ring_sz,
313 					   &rh->ibi_status_dma, GFP_KERNEL);
314 		rh->ibi_data = kmalloc(ibi_data_ring_sz, GFP_KERNEL);
315 		ret = -ENOMEM;
316 		if (!rh->ibi_status || !rh->ibi_data)
317 			goto err_out;
318 		rh->ibi_data_dma =
319 			dma_map_single(rings->sysdev, rh->ibi_data,
320 				       ibi_data_ring_sz, DMA_FROM_DEVICE);
321 		if (dma_mapping_error(rings->sysdev, rh->ibi_data_dma)) {
322 			rh->ibi_data_dma = 0;
323 			ret = -ENOMEM;
324 			goto err_out;
325 		}
326 
327 		rh_reg_write(IBI_STATUS_RING_BASE_LO, lower_32_bits(rh->ibi_status_dma));
328 		rh_reg_write(IBI_STATUS_RING_BASE_HI, upper_32_bits(rh->ibi_status_dma));
329 		rh_reg_write(IBI_DATA_RING_BASE_LO, lower_32_bits(rh->ibi_data_dma));
330 		rh_reg_write(IBI_DATA_RING_BASE_HI, upper_32_bits(rh->ibi_data_dma));
331 
332 		regval = FIELD_PREP(IBI_STATUS_RING_SIZE,
333 				    rh->ibi_status_entries) |
334 			 FIELD_PREP(IBI_DATA_CHUNK_SIZE,
335 				    ilog2(rh->ibi_chunk_sz) - 2) |
336 			 FIELD_PREP(IBI_DATA_CHUNK_COUNT,
337 				    rh->ibi_chunks_total);
338 		rh_reg_write(IBI_SETUP, regval);
339 
340 		regval = rh_reg_read(INTR_SIGNAL_ENABLE);
341 		regval |= INTR_IBI_READY;
342 		rh_reg_write(INTR_SIGNAL_ENABLE, regval);
343 
344 ring_ready:
345 		rh_reg_write(RING_CONTROL, RING_CTRL_ENABLE |
346 					   RING_CTRL_RUN_STOP);
347 	}
348 
349 	return 0;
350 
351 err_out:
352 	hci_dma_cleanup(hci);
353 	return ret;
354 }
355 
356 static void hci_dma_unmap_xfer(struct i3c_hci *hci,
357 			       struct hci_xfer *xfer_list, unsigned int n)
358 {
359 	struct hci_xfer *xfer;
360 	unsigned int i;
361 
362 	for (i = 0; i < n; i++) {
363 		xfer = xfer_list + i;
364 		if (!xfer->data)
365 			continue;
366 		i3c_master_dma_unmap_single(xfer->dma);
367 	}
368 }
369 
370 static int hci_dma_queue_xfer(struct i3c_hci *hci,
371 			      struct hci_xfer *xfer_list, int n)
372 {
373 	struct hci_rings_data *rings = hci->io_data;
374 	struct hci_rh_data *rh;
375 	unsigned int i, ring, enqueue_ptr;
376 	u32 op1_val, op2_val;
377 
378 	/* For now we only use ring 0 */
379 	ring = 0;
380 	rh = &rings->headers[ring];
381 
382 	op1_val = rh_reg_read(RING_OPERATION1);
383 	enqueue_ptr = FIELD_GET(RING_OP1_CR_ENQ_PTR, op1_val);
384 	for (i = 0; i < n; i++) {
385 		struct hci_xfer *xfer = xfer_list + i;
386 		u32 *ring_data = rh->xfer + rh->xfer_struct_sz * enqueue_ptr;
387 		enum dma_data_direction dir = xfer->rnw ? DMA_FROM_DEVICE :
388 							  DMA_TO_DEVICE;
389 		bool need_bounce;
390 
391 		/* store cmd descriptor */
392 		*ring_data++ = xfer->cmd_desc[0];
393 		*ring_data++ = xfer->cmd_desc[1];
394 		if (hci->cmd == &mipi_i3c_hci_cmd_v2) {
395 			*ring_data++ = xfer->cmd_desc[2];
396 			*ring_data++ = xfer->cmd_desc[3];
397 		}
398 
399 		/* first word of Data Buffer Descriptor Structure */
400 		if (!xfer->data)
401 			xfer->data_len = 0;
402 		*ring_data++ =
403 			FIELD_PREP(DATA_BUF_BLOCK_SIZE, xfer->data_len) |
404 			((i == n - 1) ? DATA_BUF_IOC : 0);
405 
406 		/* 2nd and 3rd words of Data Buffer Descriptor Structure */
407 		if (xfer->data) {
408 			need_bounce = device_iommu_mapped(rings->sysdev) &&
409 				      xfer->rnw &&
410 				      xfer->data_len != ALIGN(xfer->data_len, 4);
411 			xfer->dma = i3c_master_dma_map_single(rings->sysdev,
412 							      xfer->data,
413 							      xfer->data_len,
414 							      need_bounce,
415 							      dir);
416 			if (!xfer->dma) {
417 				hci_dma_unmap_xfer(hci, xfer_list, i);
418 				return -ENOMEM;
419 			}
420 			*ring_data++ = lower_32_bits(xfer->dma->addr);
421 			*ring_data++ = upper_32_bits(xfer->dma->addr);
422 		} else {
423 			*ring_data++ = 0;
424 			*ring_data++ = 0;
425 		}
426 
427 		/* remember corresponding xfer struct */
428 		rh->src_xfers[enqueue_ptr] = xfer;
429 		/* remember corresponding ring/entry for this xfer structure */
430 		xfer->ring_number = ring;
431 		xfer->ring_entry = enqueue_ptr;
432 
433 		enqueue_ptr = (enqueue_ptr + 1) % rh->xfer_entries;
434 
435 		/*
436 		 * We may update the hardware view of the enqueue pointer
437 		 * only if we didn't reach its dequeue pointer.
438 		 */
439 		op2_val = rh_reg_read(RING_OPERATION2);
440 		if (enqueue_ptr == FIELD_GET(RING_OP2_CR_DEQ_PTR, op2_val)) {
441 			/* the ring is full */
442 			hci_dma_unmap_xfer(hci, xfer_list, i + 1);
443 			return -EBUSY;
444 		}
445 	}
446 
447 	/* take care to update the hardware enqueue pointer atomically */
448 	spin_lock_irq(&rh->lock);
449 	op1_val = rh_reg_read(RING_OPERATION1);
450 	op1_val &= ~RING_OP1_CR_ENQ_PTR;
451 	op1_val |= FIELD_PREP(RING_OP1_CR_ENQ_PTR, enqueue_ptr);
452 	rh_reg_write(RING_OPERATION1, op1_val);
453 	spin_unlock_irq(&rh->lock);
454 
455 	return 0;
456 }
457 
458 static bool hci_dma_dequeue_xfer(struct i3c_hci *hci,
459 				 struct hci_xfer *xfer_list, int n)
460 {
461 	struct hci_rings_data *rings = hci->io_data;
462 	struct hci_rh_data *rh = &rings->headers[xfer_list[0].ring_number];
463 	unsigned int i;
464 	bool did_unqueue = false;
465 
466 	/* stop the ring */
467 	rh_reg_write(RING_CONTROL, RING_CTRL_ABORT);
468 	if (wait_for_completion_timeout(&rh->op_done, HZ) == 0) {
469 		/*
470 		 * We're deep in it if ever this condition is ever met.
471 		 * Hardware might still be writing to memory, etc.
472 		 */
473 		dev_crit(&hci->master.dev, "unable to abort the ring\n");
474 		WARN_ON(1);
475 	}
476 
477 	for (i = 0; i < n; i++) {
478 		struct hci_xfer *xfer = xfer_list + i;
479 		int idx = xfer->ring_entry;
480 
481 		/*
482 		 * At the time the abort happened, the xfer might have
483 		 * completed already. If not then replace corresponding
484 		 * descriptor entries with a no-op.
485 		 */
486 		if (idx >= 0) {
487 			u32 *ring_data = rh->xfer + rh->xfer_struct_sz * idx;
488 
489 			/* store no-op cmd descriptor */
490 			*ring_data++ = FIELD_PREP(CMD_0_ATTR, 0x7);
491 			*ring_data++ = 0;
492 			if (hci->cmd == &mipi_i3c_hci_cmd_v2) {
493 				*ring_data++ = 0;
494 				*ring_data++ = 0;
495 			}
496 
497 			/* disassociate this xfer struct */
498 			rh->src_xfers[idx] = NULL;
499 
500 			/* and unmap it */
501 			hci_dma_unmap_xfer(hci, xfer, 1);
502 
503 			did_unqueue = true;
504 		}
505 	}
506 
507 	/* restart the ring */
508 	rh_reg_write(RING_CONTROL, RING_CTRL_ENABLE);
509 
510 	return did_unqueue;
511 }
512 
513 static void hci_dma_xfer_done(struct i3c_hci *hci, struct hci_rh_data *rh)
514 {
515 	u32 op1_val, op2_val, resp, *ring_resp;
516 	unsigned int tid, done_ptr = rh->done_ptr;
517 	struct hci_xfer *xfer;
518 
519 	for (;;) {
520 		op2_val = rh_reg_read(RING_OPERATION2);
521 		if (done_ptr == FIELD_GET(RING_OP2_CR_DEQ_PTR, op2_val))
522 			break;
523 
524 		ring_resp = rh->resp + rh->resp_struct_sz * done_ptr;
525 		resp = *ring_resp;
526 		tid = RESP_TID(resp);
527 		dev_dbg(&hci->master.dev, "resp = 0x%08x", resp);
528 
529 		xfer = rh->src_xfers[done_ptr];
530 		if (!xfer) {
531 			dev_dbg(&hci->master.dev, "orphaned ring entry");
532 		} else {
533 			hci_dma_unmap_xfer(hci, xfer, 1);
534 			xfer->ring_entry = -1;
535 			xfer->response = resp;
536 			if (tid != xfer->cmd_tid) {
537 				dev_err(&hci->master.dev,
538 					"response tid=%d when expecting %d\n",
539 					tid, xfer->cmd_tid);
540 				/* TODO: do something about it? */
541 			}
542 			if (xfer->completion)
543 				complete(xfer->completion);
544 		}
545 
546 		done_ptr = (done_ptr + 1) % rh->xfer_entries;
547 		rh->done_ptr = done_ptr;
548 	}
549 
550 	/* take care to update the software dequeue pointer atomically */
551 	spin_lock(&rh->lock);
552 	op1_val = rh_reg_read(RING_OPERATION1);
553 	op1_val &= ~RING_OP1_CR_SW_DEQ_PTR;
554 	op1_val |= FIELD_PREP(RING_OP1_CR_SW_DEQ_PTR, done_ptr);
555 	rh_reg_write(RING_OPERATION1, op1_val);
556 	spin_unlock(&rh->lock);
557 }
558 
559 static int hci_dma_request_ibi(struct i3c_hci *hci, struct i3c_dev_desc *dev,
560 			       const struct i3c_ibi_setup *req)
561 {
562 	struct i3c_hci_dev_data *dev_data = i3c_dev_get_master_data(dev);
563 	struct i3c_generic_ibi_pool *pool;
564 	struct hci_dma_dev_ibi_data *dev_ibi;
565 
566 	dev_ibi = kmalloc(sizeof(*dev_ibi), GFP_KERNEL);
567 	if (!dev_ibi)
568 		return -ENOMEM;
569 	pool = i3c_generic_ibi_alloc_pool(dev, req);
570 	if (IS_ERR(pool)) {
571 		kfree(dev_ibi);
572 		return PTR_ERR(pool);
573 	}
574 	dev_ibi->pool = pool;
575 	dev_ibi->max_len = req->max_payload_len;
576 	dev_data->ibi_data = dev_ibi;
577 	return 0;
578 }
579 
580 static void hci_dma_free_ibi(struct i3c_hci *hci, struct i3c_dev_desc *dev)
581 {
582 	struct i3c_hci_dev_data *dev_data = i3c_dev_get_master_data(dev);
583 	struct hci_dma_dev_ibi_data *dev_ibi = dev_data->ibi_data;
584 
585 	dev_data->ibi_data = NULL;
586 	i3c_generic_ibi_free_pool(dev_ibi->pool);
587 	kfree(dev_ibi);
588 }
589 
590 static void hci_dma_recycle_ibi_slot(struct i3c_hci *hci,
591 				     struct i3c_dev_desc *dev,
592 				     struct i3c_ibi_slot *slot)
593 {
594 	struct i3c_hci_dev_data *dev_data = i3c_dev_get_master_data(dev);
595 	struct hci_dma_dev_ibi_data *dev_ibi = dev_data->ibi_data;
596 
597 	i3c_generic_ibi_recycle_slot(dev_ibi->pool, slot);
598 }
599 
600 static void hci_dma_process_ibi(struct i3c_hci *hci, struct hci_rh_data *rh)
601 {
602 	struct hci_rings_data *rings = hci->io_data;
603 	struct i3c_dev_desc *dev;
604 	struct i3c_hci_dev_data *dev_data;
605 	struct hci_dma_dev_ibi_data *dev_ibi;
606 	struct i3c_ibi_slot *slot;
607 	u32 op1_val, op2_val, ibi_status_error;
608 	unsigned int ptr, enq_ptr, deq_ptr;
609 	unsigned int ibi_size, ibi_chunks, ibi_data_offset, first_part;
610 	int ibi_addr, last_ptr;
611 	void *ring_ibi_data;
612 	dma_addr_t ring_ibi_data_dma;
613 
614 	op1_val = rh_reg_read(RING_OPERATION1);
615 	deq_ptr = FIELD_GET(RING_OP1_IBI_DEQ_PTR, op1_val);
616 
617 	op2_val = rh_reg_read(RING_OPERATION2);
618 	enq_ptr = FIELD_GET(RING_OP2_IBI_ENQ_PTR, op2_val);
619 
620 	ibi_status_error = 0;
621 	ibi_addr = -1;
622 	ibi_chunks = 0;
623 	ibi_size = 0;
624 	last_ptr = -1;
625 
626 	/* let's find all we can about this IBI */
627 	for (ptr = deq_ptr; ptr != enq_ptr;
628 	     ptr = (ptr + 1) % rh->ibi_status_entries) {
629 		u32 ibi_status, *ring_ibi_status;
630 		unsigned int chunks;
631 
632 		ring_ibi_status = rh->ibi_status + rh->ibi_status_sz * ptr;
633 		ibi_status = *ring_ibi_status;
634 		dev_dbg(&hci->master.dev, "status = %#x", ibi_status);
635 
636 		if (ibi_status_error) {
637 			/* we no longer care */
638 		} else if (ibi_status & IBI_ERROR) {
639 			ibi_status_error = ibi_status;
640 		} else if (ibi_addr ==  -1) {
641 			ibi_addr = FIELD_GET(IBI_TARGET_ADDR, ibi_status);
642 		} else if (ibi_addr != FIELD_GET(IBI_TARGET_ADDR, ibi_status)) {
643 			/* the address changed unexpectedly */
644 			ibi_status_error = ibi_status;
645 		}
646 
647 		chunks = FIELD_GET(IBI_CHUNKS, ibi_status);
648 		ibi_chunks += chunks;
649 		if (!(ibi_status & IBI_LAST_STATUS)) {
650 			ibi_size += chunks * rh->ibi_chunk_sz;
651 		} else {
652 			ibi_size += FIELD_GET(IBI_DATA_LENGTH, ibi_status);
653 			last_ptr = ptr;
654 			break;
655 		}
656 	}
657 
658 	/* validate what we've got */
659 
660 	if (last_ptr == -1) {
661 		/* this IBI sequence is not yet complete */
662 		dev_dbg(&hci->master.dev,
663 			"no LAST_STATUS available (e=%d d=%d)",
664 			enq_ptr, deq_ptr);
665 		return;
666 	}
667 	deq_ptr = last_ptr + 1;
668 	deq_ptr %= rh->ibi_status_entries;
669 
670 	if (ibi_status_error) {
671 		dev_err(&hci->master.dev, "IBI error from %#x\n", ibi_addr);
672 		goto done;
673 	}
674 
675 	/* determine who this is for */
676 	dev = i3c_hci_addr_to_dev(hci, ibi_addr);
677 	if (!dev) {
678 		dev_err(&hci->master.dev,
679 			"IBI for unknown device %#x\n", ibi_addr);
680 		goto done;
681 	}
682 
683 	dev_data = i3c_dev_get_master_data(dev);
684 	dev_ibi = dev_data->ibi_data;
685 	if (ibi_size > dev_ibi->max_len) {
686 		dev_err(&hci->master.dev, "IBI payload too big (%d > %d)\n",
687 			ibi_size, dev_ibi->max_len);
688 		goto done;
689 	}
690 
691 	/*
692 	 * This ring model is not suitable for zero-copy processing of IBIs.
693 	 * We have the data chunk ring wrap-around to deal with, meaning
694 	 * that the payload might span multiple chunks beginning at the
695 	 * end of the ring and wrap to the start of the ring. Furthermore
696 	 * there is no guarantee that those chunks will be released in order
697 	 * and in a timely manner by the upper driver. So let's just copy
698 	 * them to a discrete buffer. In practice they're supposed to be
699 	 * small anyway.
700 	 */
701 	slot = i3c_generic_ibi_get_free_slot(dev_ibi->pool);
702 	if (!slot) {
703 		dev_err(&hci->master.dev, "no free slot for IBI\n");
704 		goto done;
705 	}
706 
707 	/* copy first part of the payload */
708 	ibi_data_offset = rh->ibi_chunk_sz * rh->ibi_chunk_ptr;
709 	ring_ibi_data = rh->ibi_data + ibi_data_offset;
710 	ring_ibi_data_dma = rh->ibi_data_dma + ibi_data_offset;
711 	first_part = (rh->ibi_chunks_total - rh->ibi_chunk_ptr)
712 			* rh->ibi_chunk_sz;
713 	if (first_part > ibi_size)
714 		first_part = ibi_size;
715 	dma_sync_single_for_cpu(rings->sysdev, ring_ibi_data_dma,
716 				first_part, DMA_FROM_DEVICE);
717 	memcpy(slot->data, ring_ibi_data, first_part);
718 
719 	/* copy second part if any */
720 	if (ibi_size > first_part) {
721 		/* we wrap back to the start and copy remaining data */
722 		ring_ibi_data = rh->ibi_data;
723 		ring_ibi_data_dma = rh->ibi_data_dma;
724 		dma_sync_single_for_cpu(rings->sysdev, ring_ibi_data_dma,
725 					ibi_size - first_part, DMA_FROM_DEVICE);
726 		memcpy(slot->data + first_part, ring_ibi_data,
727 		       ibi_size - first_part);
728 	}
729 
730 	/* submit it */
731 	slot->dev = dev;
732 	slot->len = ibi_size;
733 	i3c_master_queue_ibi(dev, slot);
734 
735 done:
736 	/* take care to update the ibi dequeue pointer atomically */
737 	spin_lock(&rh->lock);
738 	op1_val = rh_reg_read(RING_OPERATION1);
739 	op1_val &= ~RING_OP1_IBI_DEQ_PTR;
740 	op1_val |= FIELD_PREP(RING_OP1_IBI_DEQ_PTR, deq_ptr);
741 	rh_reg_write(RING_OPERATION1, op1_val);
742 	spin_unlock(&rh->lock);
743 
744 	/* update the chunk pointer */
745 	rh->ibi_chunk_ptr += ibi_chunks;
746 	rh->ibi_chunk_ptr %= rh->ibi_chunks_total;
747 
748 	/* and tell the hardware about freed chunks */
749 	rh_reg_write(CHUNK_CONTROL, rh_reg_read(CHUNK_CONTROL) + ibi_chunks);
750 }
751 
752 static bool hci_dma_irq_handler(struct i3c_hci *hci)
753 {
754 	struct hci_rings_data *rings = hci->io_data;
755 	unsigned int i;
756 	bool handled = false;
757 
758 	for (i = 0; i < rings->total; i++) {
759 		struct hci_rh_data *rh;
760 		u32 status;
761 
762 		rh = &rings->headers[i];
763 		status = rh_reg_read(INTR_STATUS);
764 		dev_dbg(&hci->master.dev, "Ring %d: RH_INTR_STATUS %#x",
765 			i, status);
766 		if (!status)
767 			continue;
768 		rh_reg_write(INTR_STATUS, status);
769 
770 		if (status & INTR_IBI_READY)
771 			hci_dma_process_ibi(hci, rh);
772 		if (status & (INTR_TRANSFER_COMPLETION | INTR_TRANSFER_ERR))
773 			hci_dma_xfer_done(hci, rh);
774 		if (status & INTR_RING_OP)
775 			complete(&rh->op_done);
776 
777 		if (status & INTR_TRANSFER_ABORT) {
778 			u32 ring_status;
779 
780 			dev_notice_ratelimited(&hci->master.dev,
781 				"Ring %d: Transfer Aborted\n", i);
782 			mipi_i3c_hci_resume(hci);
783 			ring_status = rh_reg_read(RING_STATUS);
784 			if (!(ring_status & RING_STATUS_RUNNING) &&
785 			    status & INTR_TRANSFER_COMPLETION &&
786 			    status & INTR_TRANSFER_ERR) {
787 				/*
788 				 * Ring stop followed by run is an Intel
789 				 * specific required quirk after resuming the
790 				 * halted controller. Do it only when the ring
791 				 * is not in running state after a transfer
792 				 * error.
793 				 */
794 				rh_reg_write(RING_CONTROL, RING_CTRL_ENABLE);
795 				rh_reg_write(RING_CONTROL, RING_CTRL_ENABLE |
796 							   RING_CTRL_RUN_STOP);
797 			}
798 		}
799 		if (status & INTR_IBI_RING_FULL)
800 			dev_err_ratelimited(&hci->master.dev,
801 				"Ring %d: IBI Ring Full Condition\n", i);
802 
803 		handled = true;
804 	}
805 
806 	return handled;
807 }
808 
809 const struct hci_io_ops mipi_i3c_hci_dma = {
810 	.init			= hci_dma_init,
811 	.cleanup		= hci_dma_cleanup,
812 	.queue_xfer		= hci_dma_queue_xfer,
813 	.dequeue_xfer		= hci_dma_dequeue_xfer,
814 	.irq_handler		= hci_dma_irq_handler,
815 	.request_ibi		= hci_dma_request_ibi,
816 	.free_ibi		= hci_dma_free_ibi,
817 	.recycle_ibi_slot	= hci_dma_recycle_ibi_slot,
818 };
819