xref: /linux/drivers/i3c/master/mipi-i3c-hci/pio.c (revision 24bce201d79807b668bf9d9e0aca801c5c0d5f78)
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 
8 #include <linux/bitfield.h>
9 #include <linux/device.h>
10 #include <linux/errno.h>
11 #include <linux/i3c/master.h>
12 #include <linux/io.h>
13 
14 #include "hci.h"
15 #include "cmd.h"
16 #include "ibi.h"
17 
18 
19 /*
20  * PIO Access Area
21  */
22 
23 #define pio_reg_read(r)		readl(hci->PIO_regs + (PIO_##r))
24 #define pio_reg_write(r, v)	writel(v, hci->PIO_regs + (PIO_##r))
25 
26 #define PIO_COMMAND_QUEUE_PORT		0x00
27 #define PIO_RESPONSE_QUEUE_PORT		0x04
28 #define PIO_XFER_DATA_PORT		0x08
29 #define PIO_IBI_PORT			0x0c
30 
31 #define PIO_QUEUE_THLD_CTRL		0x10
32 #define QUEUE_IBI_STATUS_THLD		GENMASK(31, 24)
33 #define QUEUE_IBI_DATA_THLD		GENMASK(23, 16)
34 #define QUEUE_RESP_BUF_THLD		GENMASK(15, 8)
35 #define QUEUE_CMD_EMPTY_BUF_THLD	GENMASK(7, 0)
36 
37 #define PIO_DATA_BUFFER_THLD_CTRL	0x14
38 #define DATA_RX_START_THLD		GENMASK(26, 24)
39 #define DATA_TX_START_THLD		GENMASK(18, 16)
40 #define DATA_RX_BUF_THLD		GENMASK(10, 8)
41 #define DATA_TX_BUF_THLD		GENMASK(2, 0)
42 
43 #define PIO_QUEUE_SIZE			0x18
44 #define TX_DATA_BUFFER_SIZE		GENMASK(31, 24)
45 #define RX_DATA_BUFFER_SIZE		GENMASK(23, 16)
46 #define IBI_STATUS_SIZE			GENMASK(15, 8)
47 #define CR_QUEUE_SIZE			GENMASK(7, 0)
48 
49 #define PIO_INTR_STATUS			0x20
50 #define PIO_INTR_STATUS_ENABLE		0x24
51 #define PIO_INTR_SIGNAL_ENABLE		0x28
52 #define PIO_INTR_FORCE			0x2c
53 #define STAT_TRANSFER_BLOCKED		BIT(25)
54 #define STAT_PERR_RESP_UFLOW		BIT(24)
55 #define STAT_PERR_CMD_OFLOW		BIT(23)
56 #define STAT_PERR_IBI_UFLOW		BIT(22)
57 #define STAT_PERR_RX_UFLOW		BIT(21)
58 #define STAT_PERR_TX_OFLOW		BIT(20)
59 #define STAT_ERR_RESP_QUEUE_FULL	BIT(19)
60 #define STAT_WARN_RESP_QUEUE_FULL	BIT(18)
61 #define STAT_ERR_IBI_QUEUE_FULL		BIT(17)
62 #define STAT_WARN_IBI_QUEUE_FULL	BIT(16)
63 #define STAT_ERR_RX_DATA_FULL		BIT(15)
64 #define STAT_WARN_RX_DATA_FULL		BIT(14)
65 #define STAT_ERR_TX_DATA_EMPTY		BIT(13)
66 #define STAT_WARN_TX_DATA_EMPTY		BIT(12)
67 #define STAT_TRANSFER_ERR		BIT(9)
68 #define STAT_WARN_INS_STOP_MODE		BIT(7)
69 #define STAT_TRANSFER_ABORT		BIT(5)
70 #define STAT_RESP_READY			BIT(4)
71 #define STAT_CMD_QUEUE_READY		BIT(3)
72 #define STAT_IBI_STATUS_THLD		BIT(2)
73 #define STAT_RX_THLD			BIT(1)
74 #define STAT_TX_THLD			BIT(0)
75 
76 #define PIO_QUEUE_CUR_STATUS		0x38
77 #define CUR_IBI_Q_LEVEL			GENMASK(28, 20)
78 #define CUR_RESP_Q_LEVEL		GENMASK(18, 10)
79 #define CUR_CMD_Q_EMPTY_LEVEL		GENMASK(8, 0)
80 
81 #define PIO_DATA_BUFFER_CUR_STATUS	0x3c
82 #define CUR_RX_BUF_LVL			GENMASK(26, 16)
83 #define CUR_TX_BUF_LVL			GENMASK(10, 0)
84 
85 /*
86  * Handy status bit combinations
87  */
88 
89 #define STAT_LATENCY_WARNINGS		(STAT_WARN_RESP_QUEUE_FULL | \
90 					 STAT_WARN_IBI_QUEUE_FULL | \
91 					 STAT_WARN_RX_DATA_FULL | \
92 					 STAT_WARN_TX_DATA_EMPTY | \
93 					 STAT_WARN_INS_STOP_MODE)
94 
95 #define STAT_LATENCY_ERRORS		(STAT_ERR_RESP_QUEUE_FULL | \
96 					 STAT_ERR_IBI_QUEUE_FULL | \
97 					 STAT_ERR_RX_DATA_FULL | \
98 					 STAT_ERR_TX_DATA_EMPTY)
99 
100 #define STAT_PROG_ERRORS		(STAT_TRANSFER_BLOCKED | \
101 					 STAT_PERR_RESP_UFLOW | \
102 					 STAT_PERR_CMD_OFLOW | \
103 					 STAT_PERR_IBI_UFLOW | \
104 					 STAT_PERR_RX_UFLOW | \
105 					 STAT_PERR_TX_OFLOW)
106 
107 #define STAT_ALL_ERRORS			(STAT_TRANSFER_ABORT | \
108 					 STAT_TRANSFER_ERR | \
109 					 STAT_LATENCY_ERRORS | \
110 					 STAT_PROG_ERRORS)
111 
112 struct hci_pio_dev_ibi_data {
113 	struct i3c_generic_ibi_pool *pool;
114 	unsigned int max_len;
115 };
116 
117 struct hci_pio_ibi_data {
118 	struct i3c_ibi_slot *slot;
119 	void *data_ptr;
120 	unsigned int addr;
121 	unsigned int seg_len, seg_cnt;
122 	unsigned int max_len;
123 	bool last_seg;
124 };
125 
126 struct hci_pio_data {
127 	spinlock_t lock;
128 	struct hci_xfer *curr_xfer, *xfer_queue;
129 	struct hci_xfer *curr_rx, *rx_queue;
130 	struct hci_xfer *curr_tx, *tx_queue;
131 	struct hci_xfer *curr_resp, *resp_queue;
132 	struct hci_pio_ibi_data ibi;
133 	unsigned int rx_thresh_size, tx_thresh_size;
134 	unsigned int max_ibi_thresh;
135 	u32 reg_queue_thresh;
136 	u32 enabled_irqs;
137 };
138 
139 static int hci_pio_init(struct i3c_hci *hci)
140 {
141 	struct hci_pio_data *pio;
142 	u32 val, size_val, rx_thresh, tx_thresh, ibi_val;
143 
144 	pio = kzalloc(sizeof(*pio), GFP_KERNEL);
145 	if (!pio)
146 		return -ENOMEM;
147 
148 	hci->io_data = pio;
149 	spin_lock_init(&pio->lock);
150 
151 	size_val = pio_reg_read(QUEUE_SIZE);
152 	dev_info(&hci->master.dev, "CMD/RESP FIFO = %ld entries\n",
153 		 FIELD_GET(CR_QUEUE_SIZE, size_val));
154 	dev_info(&hci->master.dev, "IBI FIFO = %ld bytes\n",
155 		 4 * FIELD_GET(IBI_STATUS_SIZE, size_val));
156 	dev_info(&hci->master.dev, "RX data FIFO = %d bytes\n",
157 		 4 * (2 << FIELD_GET(RX_DATA_BUFFER_SIZE, size_val)));
158 	dev_info(&hci->master.dev, "TX data FIFO = %d bytes\n",
159 		 4 * (2 << FIELD_GET(TX_DATA_BUFFER_SIZE, size_val)));
160 
161 	/*
162 	 * Let's initialize data thresholds to half of the actual FIFO size.
163 	 * The start thresholds aren't used (set to 0) as the FIFO is always
164 	 * serviced before the corresponding command is queued.
165 	 */
166 	rx_thresh = FIELD_GET(RX_DATA_BUFFER_SIZE, size_val);
167 	tx_thresh = FIELD_GET(TX_DATA_BUFFER_SIZE, size_val);
168 	if (hci->version_major == 1) {
169 		/* those are expressed as 2^[n+1), so just sub 1 if not 0 */
170 		if (rx_thresh)
171 			rx_thresh -= 1;
172 		if (tx_thresh)
173 			tx_thresh -= 1;
174 		pio->rx_thresh_size = 2 << rx_thresh;
175 		pio->tx_thresh_size = 2 << tx_thresh;
176 	} else {
177 		/* size is 2^(n+1) and threshold is 2^n i.e. already halved */
178 		pio->rx_thresh_size = 1 << rx_thresh;
179 		pio->tx_thresh_size = 1 << tx_thresh;
180 	}
181 	val = FIELD_PREP(DATA_RX_BUF_THLD,   rx_thresh) |
182 	      FIELD_PREP(DATA_TX_BUF_THLD,   tx_thresh);
183 	pio_reg_write(DATA_BUFFER_THLD_CTRL, val);
184 
185 	/*
186 	 * Let's raise an interrupt as soon as there is one free cmd slot
187 	 * or one available response or IBI. For IBI data let's use half the
188 	 * IBI queue size within allowed bounds.
189 	 */
190 	ibi_val = FIELD_GET(IBI_STATUS_SIZE, size_val);
191 	pio->max_ibi_thresh = clamp_val(ibi_val/2, 1, 63);
192 	val = FIELD_PREP(QUEUE_IBI_STATUS_THLD, 1) |
193 	      FIELD_PREP(QUEUE_IBI_DATA_THLD, pio->max_ibi_thresh) |
194 	      FIELD_PREP(QUEUE_RESP_BUF_THLD, 1) |
195 	      FIELD_PREP(QUEUE_CMD_EMPTY_BUF_THLD, 1);
196 	pio_reg_write(QUEUE_THLD_CTRL, val);
197 	pio->reg_queue_thresh = val;
198 
199 	/* Disable all IRQs but allow all status bits */
200 	pio_reg_write(INTR_SIGNAL_ENABLE, 0x0);
201 	pio_reg_write(INTR_STATUS_ENABLE, 0xffffffff);
202 
203 	/* Always accept error interrupts (will be activated on first xfer) */
204 	pio->enabled_irqs = STAT_ALL_ERRORS;
205 
206 	return 0;
207 }
208 
209 static void hci_pio_cleanup(struct i3c_hci *hci)
210 {
211 	struct hci_pio_data *pio = hci->io_data;
212 
213 	pio_reg_write(INTR_SIGNAL_ENABLE, 0x0);
214 
215 	if (pio) {
216 		DBG("status = %#x/%#x",
217 		    pio_reg_read(INTR_STATUS), pio_reg_read(INTR_SIGNAL_ENABLE));
218 		BUG_ON(pio->curr_xfer);
219 		BUG_ON(pio->curr_rx);
220 		BUG_ON(pio->curr_tx);
221 		BUG_ON(pio->curr_resp);
222 		kfree(pio);
223 		hci->io_data = NULL;
224 	}
225 }
226 
227 static void hci_pio_write_cmd(struct i3c_hci *hci, struct hci_xfer *xfer)
228 {
229 	DBG("cmd_desc[%d] = 0x%08x", 0, xfer->cmd_desc[0]);
230 	DBG("cmd_desc[%d] = 0x%08x", 1, xfer->cmd_desc[1]);
231 	pio_reg_write(COMMAND_QUEUE_PORT, xfer->cmd_desc[0]);
232 	pio_reg_write(COMMAND_QUEUE_PORT, xfer->cmd_desc[1]);
233 	if (hci->cmd == &mipi_i3c_hci_cmd_v2) {
234 		DBG("cmd_desc[%d] = 0x%08x", 2, xfer->cmd_desc[2]);
235 		DBG("cmd_desc[%d] = 0x%08x", 3, xfer->cmd_desc[3]);
236 		pio_reg_write(COMMAND_QUEUE_PORT, xfer->cmd_desc[2]);
237 		pio_reg_write(COMMAND_QUEUE_PORT, xfer->cmd_desc[3]);
238 	}
239 }
240 
241 static bool hci_pio_do_rx(struct i3c_hci *hci, struct hci_pio_data *pio)
242 {
243 	struct hci_xfer *xfer = pio->curr_rx;
244 	unsigned int nr_words;
245 	u32 *p;
246 
247 	p = xfer->data;
248 	p += (xfer->data_len - xfer->data_left) / 4;
249 
250 	while (xfer->data_left >= 4) {
251 		/* bail out if FIFO hasn't reached the threshold value yet */
252 		if (!(pio_reg_read(INTR_STATUS) & STAT_RX_THLD))
253 			return false;
254 		nr_words = min(xfer->data_left / 4, pio->rx_thresh_size);
255 		/* extract data from FIFO */
256 		xfer->data_left -= nr_words * 4;
257 		DBG("now %d left %d", nr_words * 4, xfer->data_left);
258 		while (nr_words--)
259 			*p++ = pio_reg_read(XFER_DATA_PORT);
260 	}
261 
262 	/* trailing data is retrieved upon response reception */
263 	return !xfer->data_left;
264 }
265 
266 static void hci_pio_do_trailing_rx(struct i3c_hci *hci,
267 				   struct hci_pio_data *pio, unsigned int count)
268 {
269 	struct hci_xfer *xfer = pio->curr_rx;
270 	u32 *p;
271 
272 	DBG("%d remaining", count);
273 
274 	p = xfer->data;
275 	p += (xfer->data_len - xfer->data_left) / 4;
276 
277 	if (count >= 4) {
278 		unsigned int nr_words = count / 4;
279 		/* extract data from FIFO */
280 		xfer->data_left -= nr_words * 4;
281 		DBG("now %d left %d", nr_words * 4, xfer->data_left);
282 		while (nr_words--)
283 			*p++ = pio_reg_read(XFER_DATA_PORT);
284 	}
285 
286 	count &= 3;
287 	if (count) {
288 		/*
289 		 * There are trailing bytes in the last word.
290 		 * Fetch it and extract bytes in an endian independent way.
291 		 * Unlike the TX case, we must not write memory past the
292 		 * end of the destination buffer.
293 		 */
294 		u8 *p_byte = (u8 *)p;
295 		u32 data = pio_reg_read(XFER_DATA_PORT);
296 
297 		xfer->data_word_before_partial = data;
298 		xfer->data_left -= count;
299 		data = (__force u32) cpu_to_le32(data);
300 		while (count--) {
301 			*p_byte++ = data;
302 			data >>= 8;
303 		}
304 	}
305 }
306 
307 static bool hci_pio_do_tx(struct i3c_hci *hci, struct hci_pio_data *pio)
308 {
309 	struct hci_xfer *xfer = pio->curr_tx;
310 	unsigned int nr_words;
311 	u32 *p;
312 
313 	p = xfer->data;
314 	p += (xfer->data_len - xfer->data_left) / 4;
315 
316 	while (xfer->data_left >= 4) {
317 		/* bail out if FIFO free space is below set threshold */
318 		if (!(pio_reg_read(INTR_STATUS) & STAT_TX_THLD))
319 			return false;
320 		/* we can fill up to that TX threshold */
321 		nr_words = min(xfer->data_left / 4, pio->tx_thresh_size);
322 		/* push data into the FIFO */
323 		xfer->data_left -= nr_words * 4;
324 		DBG("now %d left %d", nr_words * 4, xfer->data_left);
325 		while (nr_words--)
326 			pio_reg_write(XFER_DATA_PORT, *p++);
327 	}
328 
329 	if (xfer->data_left) {
330 		/*
331 		 * There are trailing bytes to send. We can simply load
332 		 * them from memory as a word which will keep those bytes
333 		 * in their proper place even on a BE system. This will
334 		 * also get some bytes past the actual buffer but no one
335 		 * should care as they won't be sent out.
336 		 */
337 		if (!(pio_reg_read(INTR_STATUS) & STAT_TX_THLD))
338 			return false;
339 		DBG("trailing %d", xfer->data_left);
340 		pio_reg_write(XFER_DATA_PORT, *p);
341 		xfer->data_left = 0;
342 	}
343 
344 	return true;
345 }
346 
347 static bool hci_pio_process_rx(struct i3c_hci *hci, struct hci_pio_data *pio)
348 {
349 	while (pio->curr_rx && hci_pio_do_rx(hci, pio))
350 		pio->curr_rx = pio->curr_rx->next_data;
351 	return !pio->curr_rx;
352 }
353 
354 static bool hci_pio_process_tx(struct i3c_hci *hci, struct hci_pio_data *pio)
355 {
356 	while (pio->curr_tx && hci_pio_do_tx(hci, pio))
357 		pio->curr_tx = pio->curr_tx->next_data;
358 	return !pio->curr_tx;
359 }
360 
361 static void hci_pio_queue_data(struct i3c_hci *hci, struct hci_pio_data *pio)
362 {
363 	struct hci_xfer *xfer = pio->curr_xfer;
364 	struct hci_xfer *prev_queue_tail;
365 
366 	if (!xfer->data) {
367 		xfer->data_len = xfer->data_left = 0;
368 		return;
369 	}
370 
371 	if (xfer->rnw) {
372 		prev_queue_tail = pio->rx_queue;
373 		pio->rx_queue = xfer;
374 		if (pio->curr_rx) {
375 			prev_queue_tail->next_data = xfer;
376 		} else {
377 			pio->curr_rx = xfer;
378 			if (!hci_pio_process_rx(hci, pio))
379 				pio->enabled_irqs |= STAT_RX_THLD;
380 		}
381 	} else {
382 		prev_queue_tail = pio->tx_queue;
383 		pio->tx_queue = xfer;
384 		if (pio->curr_tx) {
385 			prev_queue_tail->next_data = xfer;
386 		} else {
387 			pio->curr_tx = xfer;
388 			if (!hci_pio_process_tx(hci, pio))
389 				pio->enabled_irqs |= STAT_TX_THLD;
390 		}
391 	}
392 }
393 
394 static void hci_pio_push_to_next_rx(struct i3c_hci *hci, struct hci_xfer *xfer,
395 				    unsigned int words_to_keep)
396 {
397 	u32 *from = xfer->data;
398 	u32 from_last;
399 	unsigned int received, count;
400 
401 	received = (xfer->data_len - xfer->data_left) / 4;
402 	if ((xfer->data_len - xfer->data_left) & 3) {
403 		from_last = xfer->data_word_before_partial;
404 		received += 1;
405 	} else {
406 		from_last = from[received];
407 	}
408 	from += words_to_keep;
409 	count = received - words_to_keep;
410 
411 	while (count) {
412 		unsigned int room, left, chunk, bytes_to_move;
413 		u32 last_word;
414 
415 		xfer = xfer->next_data;
416 		if (!xfer) {
417 			dev_err(&hci->master.dev, "pushing RX data to unexistent xfer\n");
418 			return;
419 		}
420 
421 		room = DIV_ROUND_UP(xfer->data_len, 4);
422 		left = DIV_ROUND_UP(xfer->data_left, 4);
423 		chunk = min(count, room);
424 		if (chunk > left) {
425 			hci_pio_push_to_next_rx(hci, xfer, chunk - left);
426 			left = chunk;
427 			xfer->data_left = left * 4;
428 		}
429 
430 		bytes_to_move = xfer->data_len - xfer->data_left;
431 		if (bytes_to_move & 3) {
432 			/* preserve word  to become partial */
433 			u32 *p = xfer->data;
434 
435 			xfer->data_word_before_partial = p[bytes_to_move / 4];
436 		}
437 		memmove(xfer->data + chunk, xfer->data, bytes_to_move);
438 
439 		/* treat last word specially because of partial word issues */
440 		chunk -= 1;
441 
442 		memcpy(xfer->data, from, chunk * 4);
443 		xfer->data_left -= chunk * 4;
444 		from += chunk;
445 		count -= chunk;
446 
447 		last_word = (count == 1) ? from_last : *from++;
448 		if (xfer->data_left < 4) {
449 			/*
450 			 * Like in hci_pio_do_trailing_rx(), preserve original
451 			 * word to be stored partially then store bytes it
452 			 * in an endian independent way.
453 			 */
454 			u8 *p_byte = xfer->data;
455 
456 			p_byte += chunk * 4;
457 			xfer->data_word_before_partial = last_word;
458 			last_word = (__force u32) cpu_to_le32(last_word);
459 			while (xfer->data_left--) {
460 				*p_byte++ = last_word;
461 				last_word >>= 8;
462 			}
463 		} else {
464 			u32 *p = xfer->data;
465 
466 			p[chunk] = last_word;
467 			xfer->data_left -= 4;
468 		}
469 		count--;
470 	}
471 }
472 
473 static void hci_pio_err(struct i3c_hci *hci, struct hci_pio_data *pio,
474 			u32 status);
475 
476 static bool hci_pio_process_resp(struct i3c_hci *hci, struct hci_pio_data *pio)
477 {
478 	while (pio->curr_resp &&
479 	       (pio_reg_read(INTR_STATUS) & STAT_RESP_READY)) {
480 		struct hci_xfer *xfer = pio->curr_resp;
481 		u32 resp = pio_reg_read(RESPONSE_QUEUE_PORT);
482 		unsigned int tid = RESP_TID(resp);
483 
484 		DBG("resp = 0x%08x", resp);
485 		if (tid != xfer->cmd_tid) {
486 			dev_err(&hci->master.dev,
487 				"response tid=%d when expecting %d\n",
488 				tid, xfer->cmd_tid);
489 			/* let's pretend it is a prog error... any of them  */
490 			hci_pio_err(hci, pio, STAT_PROG_ERRORS);
491 			return false;
492 		}
493 		xfer->response = resp;
494 
495 		if (pio->curr_rx == xfer) {
496 			/*
497 			 * Response availability implies RX completion.
498 			 * Retrieve trailing RX data if any.
499 			 * Note that short reads are possible.
500 			 */
501 			unsigned int received, expected, to_keep;
502 
503 			received = xfer->data_len - xfer->data_left;
504 			expected = RESP_DATA_LENGTH(xfer->response);
505 			if (expected > received) {
506 				hci_pio_do_trailing_rx(hci, pio,
507 						       expected - received);
508 			} else if (received > expected) {
509 				/* we consumed data meant for next xfer */
510 				to_keep = DIV_ROUND_UP(expected, 4);
511 				hci_pio_push_to_next_rx(hci, xfer, to_keep);
512 			}
513 
514 			/* then process the RX list pointer */
515 			if (hci_pio_process_rx(hci, pio))
516 				pio->enabled_irqs &= ~STAT_RX_THLD;
517 		}
518 
519 		/*
520 		 * We're about to give back ownership of the xfer structure
521 		 * to the waiting instance. Make sure no reference to it
522 		 * still exists.
523 		 */
524 		if (pio->curr_rx == xfer) {
525 			DBG("short RX ?");
526 			pio->curr_rx = pio->curr_rx->next_data;
527 		} else if (pio->curr_tx == xfer) {
528 			DBG("short TX ?");
529 			pio->curr_tx = pio->curr_tx->next_data;
530 		} else if (xfer->data_left) {
531 			DBG("PIO xfer count = %d after response",
532 			    xfer->data_left);
533 		}
534 
535 		pio->curr_resp = xfer->next_resp;
536 		if (xfer->completion)
537 			complete(xfer->completion);
538 	}
539 	return !pio->curr_resp;
540 }
541 
542 static void hci_pio_queue_resp(struct i3c_hci *hci, struct hci_pio_data *pio)
543 {
544 	struct hci_xfer *xfer = pio->curr_xfer;
545 	struct hci_xfer *prev_queue_tail;
546 
547 	if (!(xfer->cmd_desc[0] & CMD_0_ROC))
548 		return;
549 
550 	prev_queue_tail = pio->resp_queue;
551 	pio->resp_queue = xfer;
552 	if (pio->curr_resp) {
553 		prev_queue_tail->next_resp = xfer;
554 	} else {
555 		pio->curr_resp = xfer;
556 		if (!hci_pio_process_resp(hci, pio))
557 			pio->enabled_irqs |= STAT_RESP_READY;
558 	}
559 }
560 
561 static bool hci_pio_process_cmd(struct i3c_hci *hci, struct hci_pio_data *pio)
562 {
563 	while (pio->curr_xfer &&
564 	       (pio_reg_read(INTR_STATUS) & STAT_CMD_QUEUE_READY)) {
565 		/*
566 		 * Always process the data FIFO before sending the command
567 		 * so needed TX data or RX space is available upfront.
568 		 */
569 		hci_pio_queue_data(hci, pio);
570 		/*
571 		 * Then queue our response request. This will also process
572 		 * the response FIFO in case it got suddenly filled up
573 		 * with results from previous commands.
574 		 */
575 		hci_pio_queue_resp(hci, pio);
576 		/*
577 		 * Finally send the command.
578 		 */
579 		hci_pio_write_cmd(hci, pio->curr_xfer);
580 		/*
581 		 * And move on.
582 		 */
583 		pio->curr_xfer = pio->curr_xfer->next_xfer;
584 	}
585 	return !pio->curr_xfer;
586 }
587 
588 static int hci_pio_queue_xfer(struct i3c_hci *hci, struct hci_xfer *xfer, int n)
589 {
590 	struct hci_pio_data *pio = hci->io_data;
591 	struct hci_xfer *prev_queue_tail;
592 	int i;
593 
594 	DBG("n = %d", n);
595 
596 	/* link xfer instances together and initialize data count */
597 	for (i = 0; i < n; i++) {
598 		xfer[i].next_xfer = (i + 1 < n) ? &xfer[i + 1] : NULL;
599 		xfer[i].next_data = NULL;
600 		xfer[i].next_resp = NULL;
601 		xfer[i].data_left = xfer[i].data_len;
602 	}
603 
604 	spin_lock_irq(&pio->lock);
605 	prev_queue_tail = pio->xfer_queue;
606 	pio->xfer_queue = &xfer[n - 1];
607 	if (pio->curr_xfer) {
608 		prev_queue_tail->next_xfer = xfer;
609 	} else {
610 		pio->curr_xfer = xfer;
611 		if (!hci_pio_process_cmd(hci, pio))
612 			pio->enabled_irqs |= STAT_CMD_QUEUE_READY;
613 		pio_reg_write(INTR_SIGNAL_ENABLE, pio->enabled_irqs);
614 		DBG("status = %#x/%#x",
615 		    pio_reg_read(INTR_STATUS), pio_reg_read(INTR_SIGNAL_ENABLE));
616 	}
617 	spin_unlock_irq(&pio->lock);
618 	return 0;
619 }
620 
621 static bool hci_pio_dequeue_xfer_common(struct i3c_hci *hci,
622 					struct hci_pio_data *pio,
623 					struct hci_xfer *xfer, int n)
624 {
625 	struct hci_xfer *p, **p_prev_next;
626 	int i;
627 
628 	/*
629 	 * To safely dequeue a transfer request, it must be either entirely
630 	 * processed, or not yet processed at all. If our request tail is
631 	 * reachable from either the data or resp list that means the command
632 	 * was submitted and not yet completed.
633 	 */
634 	for (p = pio->curr_resp; p; p = p->next_resp)
635 		for (i = 0; i < n; i++)
636 			if (p == &xfer[i])
637 				goto pio_screwed;
638 	for (p = pio->curr_rx; p; p = p->next_data)
639 		for (i = 0; i < n; i++)
640 			if (p == &xfer[i])
641 				goto pio_screwed;
642 	for (p = pio->curr_tx; p; p = p->next_data)
643 		for (i = 0; i < n; i++)
644 			if (p == &xfer[i])
645 				goto pio_screwed;
646 
647 	/*
648 	 * The command was completed, or wasn't yet submitted.
649 	 * Unlink it from the que if the later.
650 	 */
651 	p_prev_next = &pio->curr_xfer;
652 	for (p = pio->curr_xfer; p; p = p->next_xfer) {
653 		if (p == &xfer[0]) {
654 			*p_prev_next = xfer[n - 1].next_xfer;
655 			break;
656 		}
657 		p_prev_next = &p->next_xfer;
658 	}
659 
660 	/* return true if we actually unqueued something */
661 	return !!p;
662 
663 pio_screwed:
664 	/*
665 	 * Life is tough. We must invalidate the hardware state and
666 	 * discard everything that is still queued.
667 	 */
668 	for (p = pio->curr_resp; p; p = p->next_resp) {
669 		p->response = FIELD_PREP(RESP_ERR_FIELD, RESP_ERR_HC_TERMINATED);
670 		if (p->completion)
671 			complete(p->completion);
672 	}
673 	for (p = pio->curr_xfer; p; p = p->next_xfer) {
674 		p->response = FIELD_PREP(RESP_ERR_FIELD, RESP_ERR_HC_TERMINATED);
675 		if (p->completion)
676 			complete(p->completion);
677 	}
678 	pio->curr_xfer = pio->curr_rx = pio->curr_tx = pio->curr_resp = NULL;
679 
680 	return true;
681 }
682 
683 static bool hci_pio_dequeue_xfer(struct i3c_hci *hci, struct hci_xfer *xfer, int n)
684 {
685 	struct hci_pio_data *pio = hci->io_data;
686 	int ret;
687 
688 	spin_lock_irq(&pio->lock);
689 	DBG("n=%d status=%#x/%#x", n,
690 	    pio_reg_read(INTR_STATUS), pio_reg_read(INTR_SIGNAL_ENABLE));
691 	DBG("main_status = %#x/%#x",
692 	    readl(hci->base_regs + 0x20), readl(hci->base_regs + 0x28));
693 
694 	ret = hci_pio_dequeue_xfer_common(hci, pio, xfer, n);
695 	spin_unlock_irq(&pio->lock);
696 	return ret;
697 }
698 
699 static void hci_pio_err(struct i3c_hci *hci, struct hci_pio_data *pio,
700 			u32 status)
701 {
702 	/* TODO: this ought to be more sophisticated eventually */
703 
704 	if (pio_reg_read(INTR_STATUS) & STAT_RESP_READY) {
705 		/* this may happen when an error is signaled with ROC unset */
706 		u32 resp = pio_reg_read(RESPONSE_QUEUE_PORT);
707 
708 		dev_err(&hci->master.dev,
709 			"orphan response (%#x) on error\n", resp);
710 	}
711 
712 	/* dump states on programming errors */
713 	if (status & STAT_PROG_ERRORS) {
714 		u32 queue = pio_reg_read(QUEUE_CUR_STATUS);
715 		u32 data = pio_reg_read(DATA_BUFFER_CUR_STATUS);
716 
717 		dev_err(&hci->master.dev,
718 			"prog error %#lx (C/R/I = %ld/%ld/%ld, TX/RX = %ld/%ld)\n",
719 			status & STAT_PROG_ERRORS,
720 			FIELD_GET(CUR_CMD_Q_EMPTY_LEVEL, queue),
721 			FIELD_GET(CUR_RESP_Q_LEVEL, queue),
722 			FIELD_GET(CUR_IBI_Q_LEVEL, queue),
723 			FIELD_GET(CUR_TX_BUF_LVL, data),
724 			FIELD_GET(CUR_RX_BUF_LVL, data));
725 	}
726 
727 	/* just bust out everything with pending responses for now */
728 	hci_pio_dequeue_xfer_common(hci, pio, pio->curr_resp, 1);
729 	/* ... and half-way TX transfers if any */
730 	if (pio->curr_tx && pio->curr_tx->data_left != pio->curr_tx->data_len)
731 		hci_pio_dequeue_xfer_common(hci, pio, pio->curr_tx, 1);
732 	/* then reset the hardware */
733 	mipi_i3c_hci_pio_reset(hci);
734 	mipi_i3c_hci_resume(hci);
735 
736 	DBG("status=%#x/%#x",
737 	    pio_reg_read(INTR_STATUS), pio_reg_read(INTR_SIGNAL_ENABLE));
738 }
739 
740 static void hci_pio_set_ibi_thresh(struct i3c_hci *hci,
741 				   struct hci_pio_data *pio,
742 				   unsigned int thresh_val)
743 {
744 	u32 regval = pio->reg_queue_thresh;
745 
746 	regval &= ~QUEUE_IBI_STATUS_THLD;
747 	regval |= FIELD_PREP(QUEUE_IBI_STATUS_THLD, thresh_val);
748 	/* write the threshold reg only if it changes */
749 	if (regval != pio->reg_queue_thresh) {
750 		pio_reg_write(QUEUE_THLD_CTRL, regval);
751 		pio->reg_queue_thresh = regval;
752 		DBG("%d", thresh_val);
753 	}
754 }
755 
756 static bool hci_pio_get_ibi_segment(struct i3c_hci *hci,
757 				    struct hci_pio_data *pio)
758 {
759 	struct hci_pio_ibi_data *ibi = &pio->ibi;
760 	unsigned int nr_words, thresh_val;
761 	u32 *p;
762 
763 	p = ibi->data_ptr;
764 	p += (ibi->seg_len - ibi->seg_cnt) / 4;
765 
766 	while ((nr_words = ibi->seg_cnt/4)) {
767 		/* determine our IBI queue threshold value */
768 		thresh_val = min(nr_words, pio->max_ibi_thresh);
769 		hci_pio_set_ibi_thresh(hci, pio, thresh_val);
770 		/* bail out if we don't have that amount of data ready */
771 		if (!(pio_reg_read(INTR_STATUS) & STAT_IBI_STATUS_THLD))
772 			return false;
773 		/* extract the data from the IBI port */
774 		nr_words = thresh_val;
775 		ibi->seg_cnt -= nr_words * 4;
776 		DBG("now %d left %d", nr_words * 4, ibi->seg_cnt);
777 		while (nr_words--)
778 			*p++ = pio_reg_read(IBI_PORT);
779 	}
780 
781 	if (ibi->seg_cnt) {
782 		/*
783 		 * There are trailing bytes in the last word.
784 		 * Fetch it and extract bytes in an endian independent way.
785 		 * Unlike the TX case, we must not write past the end of
786 		 * the destination buffer.
787 		 */
788 		u32 data;
789 		u8 *p_byte = (u8 *)p;
790 
791 		hci_pio_set_ibi_thresh(hci, pio, 1);
792 		if (!(pio_reg_read(INTR_STATUS) & STAT_IBI_STATUS_THLD))
793 			return false;
794 		DBG("trailing %d", ibi->seg_cnt);
795 		data = pio_reg_read(IBI_PORT);
796 		data = (__force u32) cpu_to_le32(data);
797 		while (ibi->seg_cnt--) {
798 			*p_byte++ = data;
799 			data >>= 8;
800 		}
801 	}
802 
803 	return true;
804 }
805 
806 static bool hci_pio_prep_new_ibi(struct i3c_hci *hci, struct hci_pio_data *pio)
807 {
808 	struct hci_pio_ibi_data *ibi = &pio->ibi;
809 	struct i3c_dev_desc *dev;
810 	struct i3c_hci_dev_data *dev_data;
811 	struct hci_pio_dev_ibi_data *dev_ibi;
812 	u32 ibi_status;
813 
814 	/*
815 	 * We have a new IBI. Try to set up its payload retrieval.
816 	 * When returning true, the IBI data has to be consumed whether
817 	 * or not we are set up to capture it. If we return true with
818 	 * ibi->slot == NULL that means the data payload has to be
819 	 * drained out of the IBI port and dropped.
820 	 */
821 
822 	ibi_status = pio_reg_read(IBI_PORT);
823 	DBG("status = %#x", ibi_status);
824 	ibi->addr = FIELD_GET(IBI_TARGET_ADDR, ibi_status);
825 	if (ibi_status & IBI_ERROR) {
826 		dev_err(&hci->master.dev, "IBI error from %#x\n", ibi->addr);
827 		return false;
828 	}
829 
830 	ibi->last_seg = ibi_status & IBI_LAST_STATUS;
831 	ibi->seg_len = FIELD_GET(IBI_DATA_LENGTH, ibi_status);
832 	ibi->seg_cnt = ibi->seg_len;
833 
834 	dev = i3c_hci_addr_to_dev(hci, ibi->addr);
835 	if (!dev) {
836 		dev_err(&hci->master.dev,
837 			"IBI for unknown device %#x\n", ibi->addr);
838 		return true;
839 	}
840 
841 	dev_data = i3c_dev_get_master_data(dev);
842 	dev_ibi = dev_data->ibi_data;
843 	ibi->max_len = dev_ibi->max_len;
844 
845 	if (ibi->seg_len > ibi->max_len) {
846 		dev_err(&hci->master.dev, "IBI payload too big (%d > %d)\n",
847 			ibi->seg_len, ibi->max_len);
848 		return true;
849 	}
850 
851 	ibi->slot = i3c_generic_ibi_get_free_slot(dev_ibi->pool);
852 	if (!ibi->slot) {
853 		dev_err(&hci->master.dev, "no free slot for IBI\n");
854 	} else {
855 		ibi->slot->len = 0;
856 		ibi->data_ptr = ibi->slot->data;
857 	}
858 	return true;
859 }
860 
861 static void hci_pio_free_ibi_slot(struct i3c_hci *hci, struct hci_pio_data *pio)
862 {
863 	struct hci_pio_ibi_data *ibi = &pio->ibi;
864 	struct hci_pio_dev_ibi_data *dev_ibi;
865 
866 	if (ibi->slot) {
867 		dev_ibi = ibi->slot->dev->common.master_priv;
868 		i3c_generic_ibi_recycle_slot(dev_ibi->pool, ibi->slot);
869 		ibi->slot = NULL;
870 	}
871 }
872 
873 static bool hci_pio_process_ibi(struct i3c_hci *hci, struct hci_pio_data *pio)
874 {
875 	struct hci_pio_ibi_data *ibi = &pio->ibi;
876 
877 	if (!ibi->slot && !ibi->seg_cnt && ibi->last_seg)
878 		if (!hci_pio_prep_new_ibi(hci, pio))
879 			return false;
880 
881 	for (;;) {
882 		u32 ibi_status;
883 		unsigned int ibi_addr;
884 
885 		if (ibi->slot) {
886 			if (!hci_pio_get_ibi_segment(hci, pio))
887 				return false;
888 			ibi->slot->len += ibi->seg_len;
889 			ibi->data_ptr += ibi->seg_len;
890 			if (ibi->last_seg) {
891 				/* was the last segment: submit it and leave */
892 				i3c_master_queue_ibi(ibi->slot->dev, ibi->slot);
893 				ibi->slot = NULL;
894 				hci_pio_set_ibi_thresh(hci, pio, 1);
895 				return true;
896 			}
897 		} else if (ibi->seg_cnt) {
898 			/*
899 			 * No slot but a non-zero count. This is the result
900 			 * of some error and the payload must be drained.
901 			 * This normally does not happen therefore no need
902 			 * to be extra optimized here.
903 			 */
904 			hci_pio_set_ibi_thresh(hci, pio, 1);
905 			do {
906 				if (!(pio_reg_read(INTR_STATUS) & STAT_IBI_STATUS_THLD))
907 					return false;
908 				pio_reg_read(IBI_PORT);
909 			} while (--ibi->seg_cnt);
910 			if (ibi->last_seg)
911 				return true;
912 		}
913 
914 		/* try to move to the next segment right away */
915 		hci_pio_set_ibi_thresh(hci, pio, 1);
916 		if (!(pio_reg_read(INTR_STATUS) & STAT_IBI_STATUS_THLD))
917 			return false;
918 		ibi_status = pio_reg_read(IBI_PORT);
919 		ibi_addr = FIELD_GET(IBI_TARGET_ADDR, ibi_status);
920 		if (ibi->addr != ibi_addr) {
921 			/* target address changed before last segment */
922 			dev_err(&hci->master.dev,
923 				"unexp IBI address changed from %d to %d\n",
924 				ibi->addr, ibi_addr);
925 			hci_pio_free_ibi_slot(hci, pio);
926 		}
927 		ibi->last_seg = ibi_status & IBI_LAST_STATUS;
928 		ibi->seg_len = FIELD_GET(IBI_DATA_LENGTH, ibi_status);
929 		ibi->seg_cnt = ibi->seg_len;
930 		if (ibi->slot && ibi->slot->len + ibi->seg_len > ibi->max_len) {
931 			dev_err(&hci->master.dev,
932 				"IBI payload too big (%d > %d)\n",
933 				ibi->slot->len + ibi->seg_len, ibi->max_len);
934 			hci_pio_free_ibi_slot(hci, pio);
935 		}
936 	}
937 
938 	return false;
939 }
940 
941 static int hci_pio_request_ibi(struct i3c_hci *hci, struct i3c_dev_desc *dev,
942 			       const struct i3c_ibi_setup *req)
943 {
944 	struct i3c_hci_dev_data *dev_data = i3c_dev_get_master_data(dev);
945 	struct i3c_generic_ibi_pool *pool;
946 	struct hci_pio_dev_ibi_data *dev_ibi;
947 
948 	dev_ibi = kmalloc(sizeof(*dev_ibi), GFP_KERNEL);
949 	if (!dev_ibi)
950 		return -ENOMEM;
951 	pool = i3c_generic_ibi_alloc_pool(dev, req);
952 	if (IS_ERR(pool)) {
953 		kfree(dev_ibi);
954 		return PTR_ERR(pool);
955 	}
956 	dev_ibi->pool = pool;
957 	dev_ibi->max_len = req->max_payload_len;
958 	dev_data->ibi_data = dev_ibi;
959 	return 0;
960 }
961 
962 static void hci_pio_free_ibi(struct i3c_hci *hci, struct i3c_dev_desc *dev)
963 {
964 	struct i3c_hci_dev_data *dev_data = i3c_dev_get_master_data(dev);
965 	struct hci_pio_dev_ibi_data *dev_ibi = dev_data->ibi_data;
966 
967 	dev_data->ibi_data = NULL;
968 	i3c_generic_ibi_free_pool(dev_ibi->pool);
969 	kfree(dev_ibi);
970 }
971 
972 static void hci_pio_recycle_ibi_slot(struct i3c_hci *hci,
973 				    struct i3c_dev_desc *dev,
974 				    struct i3c_ibi_slot *slot)
975 {
976 	struct i3c_hci_dev_data *dev_data = i3c_dev_get_master_data(dev);
977 	struct hci_pio_dev_ibi_data *dev_ibi = dev_data->ibi_data;
978 
979 	i3c_generic_ibi_recycle_slot(dev_ibi->pool, slot);
980 }
981 
982 static bool hci_pio_irq_handler(struct i3c_hci *hci, unsigned int unused)
983 {
984 	struct hci_pio_data *pio = hci->io_data;
985 	u32 status;
986 
987 	spin_lock(&pio->lock);
988 	status = pio_reg_read(INTR_STATUS);
989 	DBG("(in) status: %#x/%#x", status, pio->enabled_irqs);
990 	status &= pio->enabled_irqs | STAT_LATENCY_WARNINGS;
991 	if (!status) {
992 		spin_unlock(&pio->lock);
993 		return false;
994 	}
995 
996 	if (status & STAT_IBI_STATUS_THLD)
997 		hci_pio_process_ibi(hci, pio);
998 
999 	if (status & STAT_RX_THLD)
1000 		if (hci_pio_process_rx(hci, pio))
1001 			pio->enabled_irqs &= ~STAT_RX_THLD;
1002 	if (status & STAT_TX_THLD)
1003 		if (hci_pio_process_tx(hci, pio))
1004 			pio->enabled_irqs &= ~STAT_TX_THLD;
1005 	if (status & STAT_RESP_READY)
1006 		if (hci_pio_process_resp(hci, pio))
1007 			pio->enabled_irqs &= ~STAT_RESP_READY;
1008 
1009 	if (unlikely(status & STAT_LATENCY_WARNINGS)) {
1010 		pio_reg_write(INTR_STATUS, status & STAT_LATENCY_WARNINGS);
1011 		dev_warn_ratelimited(&hci->master.dev,
1012 				     "encountered warning condition %#lx\n",
1013 				     status & STAT_LATENCY_WARNINGS);
1014 	}
1015 
1016 	if (unlikely(status & STAT_ALL_ERRORS)) {
1017 		pio_reg_write(INTR_STATUS, status & STAT_ALL_ERRORS);
1018 		hci_pio_err(hci, pio, status & STAT_ALL_ERRORS);
1019 	}
1020 
1021 	if (status & STAT_CMD_QUEUE_READY)
1022 		if (hci_pio_process_cmd(hci, pio))
1023 			pio->enabled_irqs &= ~STAT_CMD_QUEUE_READY;
1024 
1025 	pio_reg_write(INTR_SIGNAL_ENABLE, pio->enabled_irqs);
1026 	DBG("(out) status: %#x/%#x",
1027 	    pio_reg_read(INTR_STATUS), pio_reg_read(INTR_SIGNAL_ENABLE));
1028 	spin_unlock(&pio->lock);
1029 	return true;
1030 }
1031 
1032 const struct hci_io_ops mipi_i3c_hci_pio = {
1033 	.init			= hci_pio_init,
1034 	.cleanup		= hci_pio_cleanup,
1035 	.queue_xfer		= hci_pio_queue_xfer,
1036 	.dequeue_xfer		= hci_pio_dequeue_xfer,
1037 	.irq_handler		= hci_pio_irq_handler,
1038 	.request_ibi		= hci_pio_request_ibi,
1039 	.free_ibi		= hci_pio_free_ibi,
1040 	.recycle_ibi_slot	= hci_pio_recycle_ibi_slot,
1041 };
1042