1 /*
2  * DMM IOMMU driver support functions for TI OMAP processors.
3  *
4  * Author: Rob Clark <rob@ti.com>
5  *         Andy Gross <andy.gross@ti.com>
6  *
7  * Copyright (C) 2011 Texas Instruments Incorporated - http://www.ti.com/
8  *
9  * This program is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU General Public License as
11  * published by the Free Software Foundation version 2.
12  *
13  * This program is distributed "as is" WITHOUT ANY WARRANTY of any
14  * kind, whether express or implied; without even the implied warranty
15  * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17  */
18 #include <linux/init.h>
19 #include <linux/module.h>
20 #include <linux/platform_device.h> /* platform_device() */
21 #include <linux/errno.h>
22 #include <linux/sched.h>
23 #include <linux/wait.h>
24 #include <linux/interrupt.h>
25 #include <linux/dma-mapping.h>
26 #include <linux/slab.h>
27 #include <linux/vmalloc.h>
28 #include <linux/delay.h>
29 #include <linux/mm.h>
30 #include <linux/time.h>
31 #include <linux/list.h>
32 
33 #include "omap_dmm_tiler.h"
34 #include "omap_dmm_priv.h"
35 
36 #define DMM_DRIVER_NAME "dmm"
37 
38 /* mappings for associating views to luts */
39 static struct tcm *containers[TILFMT_NFORMATS];
40 static struct dmm *omap_dmm;
41 
42 /* global spinlock for protecting lists */
43 static DEFINE_SPINLOCK(list_lock);
44 
45 /* Geometry table */
46 #define GEOM(xshift, yshift, bytes_per_pixel) { \
47 		.x_shft = (xshift), \
48 		.y_shft = (yshift), \
49 		.cpp    = (bytes_per_pixel), \
50 		.slot_w = 1 << (SLOT_WIDTH_BITS - (xshift)), \
51 		.slot_h = 1 << (SLOT_HEIGHT_BITS - (yshift)), \
52 	}
53 
54 static const struct {
55 	uint32_t x_shft;	/* unused X-bits (as part of bpp) */
56 	uint32_t y_shft;	/* unused Y-bits (as part of bpp) */
57 	uint32_t cpp;		/* bytes/chars per pixel */
58 	uint32_t slot_w;	/* width of each slot (in pixels) */
59 	uint32_t slot_h;	/* height of each slot (in pixels) */
60 } geom[TILFMT_NFORMATS] = {
61 		[TILFMT_8BIT]  = GEOM(0, 0, 1),
62 		[TILFMT_16BIT] = GEOM(0, 1, 2),
63 		[TILFMT_32BIT] = GEOM(1, 1, 4),
64 		[TILFMT_PAGE]  = GEOM(SLOT_WIDTH_BITS, SLOT_HEIGHT_BITS, 1),
65 };
66 
67 
68 /* lookup table for registers w/ per-engine instances */
69 static const uint32_t reg[][4] = {
70 		[PAT_STATUS] = {DMM_PAT_STATUS__0, DMM_PAT_STATUS__1,
71 				DMM_PAT_STATUS__2, DMM_PAT_STATUS__3},
72 		[PAT_DESCR]  = {DMM_PAT_DESCR__0, DMM_PAT_DESCR__1,
73 				DMM_PAT_DESCR__2, DMM_PAT_DESCR__3},
74 };
75 
76 /* simple allocator to grab next 16 byte aligned memory from txn */
77 static void *alloc_dma(struct dmm_txn *txn, size_t sz, dma_addr_t *pa)
78 {
79 	void *ptr;
80 	struct refill_engine *engine = txn->engine_handle;
81 
82 	/* dmm programming requires 16 byte aligned addresses */
83 	txn->current_pa = round_up(txn->current_pa, 16);
84 	txn->current_va = (void *)round_up((long)txn->current_va, 16);
85 
86 	ptr = txn->current_va;
87 	*pa = txn->current_pa;
88 
89 	txn->current_pa += sz;
90 	txn->current_va += sz;
91 
92 	BUG_ON((txn->current_va - engine->refill_va) > REFILL_BUFFER_SIZE);
93 
94 	return ptr;
95 }
96 
97 /* check status and spin until wait_mask comes true */
98 static int wait_status(struct refill_engine *engine, uint32_t wait_mask)
99 {
100 	struct dmm *dmm = engine->dmm;
101 	uint32_t r = 0, err, i;
102 
103 	i = DMM_FIXED_RETRY_COUNT;
104 	while (true) {
105 		r = readl(dmm->base + reg[PAT_STATUS][engine->id]);
106 		err = r & DMM_PATSTATUS_ERR;
107 		if (err)
108 			return -EFAULT;
109 
110 		if ((r & wait_mask) == wait_mask)
111 			break;
112 
113 		if (--i == 0)
114 			return -ETIMEDOUT;
115 
116 		udelay(1);
117 	}
118 
119 	return 0;
120 }
121 
122 static void release_engine(struct refill_engine *engine)
123 {
124 	unsigned long flags;
125 
126 	spin_lock_irqsave(&list_lock, flags);
127 	list_add(&engine->idle_node, &omap_dmm->idle_head);
128 	spin_unlock_irqrestore(&list_lock, flags);
129 
130 	atomic_inc(&omap_dmm->engine_counter);
131 	wake_up_interruptible(&omap_dmm->engine_queue);
132 }
133 
134 static irqreturn_t omap_dmm_irq_handler(int irq, void *arg)
135 {
136 	struct dmm *dmm = arg;
137 	uint32_t status = readl(dmm->base + DMM_PAT_IRQSTATUS);
138 	int i;
139 
140 	/* ack IRQ */
141 	writel(status, dmm->base + DMM_PAT_IRQSTATUS);
142 
143 	for (i = 0; i < dmm->num_engines; i++) {
144 		if (status & DMM_IRQSTAT_LST) {
145 			wake_up_interruptible(&dmm->engines[i].wait_for_refill);
146 
147 			if (dmm->engines[i].async)
148 				release_engine(&dmm->engines[i]);
149 		}
150 
151 		status >>= 8;
152 	}
153 
154 	return IRQ_HANDLED;
155 }
156 
157 /**
158  * Get a handle for a DMM transaction
159  */
160 static struct dmm_txn *dmm_txn_init(struct dmm *dmm, struct tcm *tcm)
161 {
162 	struct dmm_txn *txn = NULL;
163 	struct refill_engine *engine = NULL;
164 	int ret;
165 	unsigned long flags;
166 
167 
168 	/* wait until an engine is available */
169 	ret = wait_event_interruptible(omap_dmm->engine_queue,
170 		atomic_add_unless(&omap_dmm->engine_counter, -1, 0));
171 	if (ret)
172 		return ERR_PTR(ret);
173 
174 	/* grab an idle engine */
175 	spin_lock_irqsave(&list_lock, flags);
176 	if (!list_empty(&dmm->idle_head)) {
177 		engine = list_entry(dmm->idle_head.next, struct refill_engine,
178 					idle_node);
179 		list_del(&engine->idle_node);
180 	}
181 	spin_unlock_irqrestore(&list_lock, flags);
182 
183 	BUG_ON(!engine);
184 
185 	txn = &engine->txn;
186 	engine->tcm = tcm;
187 	txn->engine_handle = engine;
188 	txn->last_pat = NULL;
189 	txn->current_va = engine->refill_va;
190 	txn->current_pa = engine->refill_pa;
191 
192 	return txn;
193 }
194 
195 /**
196  * Add region to DMM transaction.  If pages or pages[i] is NULL, then the
197  * corresponding slot is cleared (ie. dummy_pa is programmed)
198  */
199 static void dmm_txn_append(struct dmm_txn *txn, struct pat_area *area,
200 		struct page **pages, uint32_t npages, uint32_t roll)
201 {
202 	dma_addr_t pat_pa = 0, data_pa = 0;
203 	uint32_t *data;
204 	struct pat *pat;
205 	struct refill_engine *engine = txn->engine_handle;
206 	int columns = (1 + area->x1 - area->x0);
207 	int rows = (1 + area->y1 - area->y0);
208 	int i = columns*rows;
209 
210 	pat = alloc_dma(txn, sizeof(struct pat), &pat_pa);
211 
212 	if (txn->last_pat)
213 		txn->last_pat->next_pa = (uint32_t)pat_pa;
214 
215 	pat->area = *area;
216 
217 	/* adjust Y coordinates based off of container parameters */
218 	pat->area.y0 += engine->tcm->y_offset;
219 	pat->area.y1 += engine->tcm->y_offset;
220 
221 	pat->ctrl = (struct pat_ctrl){
222 			.start = 1,
223 			.lut_id = engine->tcm->lut_id,
224 		};
225 
226 	data = alloc_dma(txn, 4*i, &data_pa);
227 	/* FIXME: what if data_pa is more than 32-bit ? */
228 	pat->data_pa = data_pa;
229 
230 	while (i--) {
231 		int n = i + roll;
232 		if (n >= npages)
233 			n -= npages;
234 		data[i] = (pages && pages[n]) ?
235 			page_to_phys(pages[n]) : engine->dmm->dummy_pa;
236 	}
237 
238 	txn->last_pat = pat;
239 
240 	return;
241 }
242 
243 /**
244  * Commit the DMM transaction.
245  */
246 static int dmm_txn_commit(struct dmm_txn *txn, bool wait)
247 {
248 	int ret = 0;
249 	struct refill_engine *engine = txn->engine_handle;
250 	struct dmm *dmm = engine->dmm;
251 
252 	if (!txn->last_pat) {
253 		dev_err(engine->dmm->dev, "need at least one txn\n");
254 		ret = -EINVAL;
255 		goto cleanup;
256 	}
257 
258 	txn->last_pat->next_pa = 0;
259 
260 	/* write to PAT_DESCR to clear out any pending transaction */
261 	writel(0x0, dmm->base + reg[PAT_DESCR][engine->id]);
262 
263 	/* wait for engine ready: */
264 	ret = wait_status(engine, DMM_PATSTATUS_READY);
265 	if (ret) {
266 		ret = -EFAULT;
267 		goto cleanup;
268 	}
269 
270 	/* mark whether it is async to denote list management in IRQ handler */
271 	engine->async = wait ? false : true;
272 
273 	/* kick reload */
274 	writel(engine->refill_pa,
275 		dmm->base + reg[PAT_DESCR][engine->id]);
276 
277 	if (wait) {
278 		if (wait_event_interruptible_timeout(engine->wait_for_refill,
279 				wait_status(engine, DMM_PATSTATUS_READY) == 0,
280 				msecs_to_jiffies(1)) <= 0) {
281 			dev_err(dmm->dev, "timed out waiting for done\n");
282 			ret = -ETIMEDOUT;
283 		}
284 	}
285 
286 cleanup:
287 	/* only place engine back on list if we are done with it */
288 	if (ret || wait)
289 		release_engine(engine);
290 
291 	return ret;
292 }
293 
294 /*
295  * DMM programming
296  */
297 static int fill(struct tcm_area *area, struct page **pages,
298 		uint32_t npages, uint32_t roll, bool wait)
299 {
300 	int ret = 0;
301 	struct tcm_area slice, area_s;
302 	struct dmm_txn *txn;
303 
304 	txn = dmm_txn_init(omap_dmm, area->tcm);
305 	if (IS_ERR_OR_NULL(txn))
306 		return -ENOMEM;
307 
308 	tcm_for_each_slice(slice, *area, area_s) {
309 		struct pat_area p_area = {
310 				.x0 = slice.p0.x,  .y0 = slice.p0.y,
311 				.x1 = slice.p1.x,  .y1 = slice.p1.y,
312 		};
313 
314 		dmm_txn_append(txn, &p_area, pages, npages, roll);
315 
316 		roll += tcm_sizeof(slice);
317 	}
318 
319 	ret = dmm_txn_commit(txn, wait);
320 
321 	return ret;
322 }
323 
324 /*
325  * Pin/unpin
326  */
327 
328 /* note: slots for which pages[i] == NULL are filled w/ dummy page
329  */
330 int tiler_pin(struct tiler_block *block, struct page **pages,
331 		uint32_t npages, uint32_t roll, bool wait)
332 {
333 	int ret;
334 
335 	ret = fill(&block->area, pages, npages, roll, wait);
336 
337 	if (ret)
338 		tiler_unpin(block);
339 
340 	return ret;
341 }
342 
343 int tiler_unpin(struct tiler_block *block)
344 {
345 	return fill(&block->area, NULL, 0, 0, false);
346 }
347 
348 /*
349  * Reserve/release
350  */
351 struct tiler_block *tiler_reserve_2d(enum tiler_fmt fmt, uint16_t w,
352 		uint16_t h, uint16_t align)
353 {
354 	struct tiler_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
355 	u32 min_align = 128;
356 	int ret;
357 	unsigned long flags;
358 
359 	BUG_ON(!validfmt(fmt));
360 
361 	/* convert width/height to slots */
362 	w = DIV_ROUND_UP(w, geom[fmt].slot_w);
363 	h = DIV_ROUND_UP(h, geom[fmt].slot_h);
364 
365 	/* convert alignment to slots */
366 	min_align = max(min_align, (geom[fmt].slot_w * geom[fmt].cpp));
367 	align = ALIGN(align, min_align);
368 	align /= geom[fmt].slot_w * geom[fmt].cpp;
369 
370 	block->fmt = fmt;
371 
372 	ret = tcm_reserve_2d(containers[fmt], w, h, align, &block->area);
373 	if (ret) {
374 		kfree(block);
375 		return ERR_PTR(-ENOMEM);
376 	}
377 
378 	/* add to allocation list */
379 	spin_lock_irqsave(&list_lock, flags);
380 	list_add(&block->alloc_node, &omap_dmm->alloc_head);
381 	spin_unlock_irqrestore(&list_lock, flags);
382 
383 	return block;
384 }
385 
386 struct tiler_block *tiler_reserve_1d(size_t size)
387 {
388 	struct tiler_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
389 	int num_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
390 	unsigned long flags;
391 
392 	if (!block)
393 		return ERR_PTR(-ENOMEM);
394 
395 	block->fmt = TILFMT_PAGE;
396 
397 	if (tcm_reserve_1d(containers[TILFMT_PAGE], num_pages,
398 				&block->area)) {
399 		kfree(block);
400 		return ERR_PTR(-ENOMEM);
401 	}
402 
403 	spin_lock_irqsave(&list_lock, flags);
404 	list_add(&block->alloc_node, &omap_dmm->alloc_head);
405 	spin_unlock_irqrestore(&list_lock, flags);
406 
407 	return block;
408 }
409 
410 /* note: if you have pin'd pages, you should have already unpin'd first! */
411 int tiler_release(struct tiler_block *block)
412 {
413 	int ret = tcm_free(&block->area);
414 	unsigned long flags;
415 
416 	if (block->area.tcm)
417 		dev_err(omap_dmm->dev, "failed to release block\n");
418 
419 	spin_lock_irqsave(&list_lock, flags);
420 	list_del(&block->alloc_node);
421 	spin_unlock_irqrestore(&list_lock, flags);
422 
423 	kfree(block);
424 	return ret;
425 }
426 
427 /*
428  * Utils
429  */
430 
431 /* calculate the tiler space address of a pixel in a view orientation...
432  * below description copied from the display subsystem section of TRM:
433  *
434  * When the TILER is addressed, the bits:
435  *   [28:27] = 0x0 for 8-bit tiled
436  *             0x1 for 16-bit tiled
437  *             0x2 for 32-bit tiled
438  *             0x3 for page mode
439  *   [31:29] = 0x0 for 0-degree view
440  *             0x1 for 180-degree view + mirroring
441  *             0x2 for 0-degree view + mirroring
442  *             0x3 for 180-degree view
443  *             0x4 for 270-degree view + mirroring
444  *             0x5 for 270-degree view
445  *             0x6 for 90-degree view
446  *             0x7 for 90-degree view + mirroring
447  * Otherwise the bits indicated the corresponding bit address to access
448  * the SDRAM.
449  */
450 static u32 tiler_get_address(enum tiler_fmt fmt, u32 orient, u32 x, u32 y)
451 {
452 	u32 x_bits, y_bits, tmp, x_mask, y_mask, alignment;
453 
454 	x_bits = CONT_WIDTH_BITS - geom[fmt].x_shft;
455 	y_bits = CONT_HEIGHT_BITS - geom[fmt].y_shft;
456 	alignment = geom[fmt].x_shft + geom[fmt].y_shft;
457 
458 	/* validate coordinate */
459 	x_mask = MASK(x_bits);
460 	y_mask = MASK(y_bits);
461 
462 	if (x < 0 || x > x_mask || y < 0 || y > y_mask) {
463 		DBG("invalid coords: %u < 0 || %u > %u || %u < 0 || %u > %u",
464 				x, x, x_mask, y, y, y_mask);
465 		return 0;
466 	}
467 
468 	/* account for mirroring */
469 	if (orient & MASK_X_INVERT)
470 		x ^= x_mask;
471 	if (orient & MASK_Y_INVERT)
472 		y ^= y_mask;
473 
474 	/* get coordinate address */
475 	if (orient & MASK_XY_FLIP)
476 		tmp = ((x << y_bits) + y);
477 	else
478 		tmp = ((y << x_bits) + x);
479 
480 	return TIL_ADDR((tmp << alignment), orient, fmt);
481 }
482 
483 dma_addr_t tiler_ssptr(struct tiler_block *block)
484 {
485 	BUG_ON(!validfmt(block->fmt));
486 
487 	return TILVIEW_8BIT + tiler_get_address(block->fmt, 0,
488 			block->area.p0.x * geom[block->fmt].slot_w,
489 			block->area.p0.y * geom[block->fmt].slot_h);
490 }
491 
492 dma_addr_t tiler_tsptr(struct tiler_block *block, uint32_t orient,
493 		uint32_t x, uint32_t y)
494 {
495 	struct tcm_pt *p = &block->area.p0;
496 	BUG_ON(!validfmt(block->fmt));
497 
498 	return tiler_get_address(block->fmt, orient,
499 			(p->x * geom[block->fmt].slot_w) + x,
500 			(p->y * geom[block->fmt].slot_h) + y);
501 }
502 
503 void tiler_align(enum tiler_fmt fmt, uint16_t *w, uint16_t *h)
504 {
505 	BUG_ON(!validfmt(fmt));
506 	*w = round_up(*w, geom[fmt].slot_w);
507 	*h = round_up(*h, geom[fmt].slot_h);
508 }
509 
510 uint32_t tiler_stride(enum tiler_fmt fmt, uint32_t orient)
511 {
512 	BUG_ON(!validfmt(fmt));
513 
514 	if (orient & MASK_XY_FLIP)
515 		return 1 << (CONT_HEIGHT_BITS + geom[fmt].x_shft);
516 	else
517 		return 1 << (CONT_WIDTH_BITS + geom[fmt].y_shft);
518 }
519 
520 size_t tiler_size(enum tiler_fmt fmt, uint16_t w, uint16_t h)
521 {
522 	tiler_align(fmt, &w, &h);
523 	return geom[fmt].cpp * w * h;
524 }
525 
526 size_t tiler_vsize(enum tiler_fmt fmt, uint16_t w, uint16_t h)
527 {
528 	BUG_ON(!validfmt(fmt));
529 	return round_up(geom[fmt].cpp * w, PAGE_SIZE) * h;
530 }
531 
532 bool dmm_is_available(void)
533 {
534 	return omap_dmm ? true : false;
535 }
536 
537 static int omap_dmm_remove(struct platform_device *dev)
538 {
539 	struct tiler_block *block, *_block;
540 	int i;
541 	unsigned long flags;
542 
543 	if (omap_dmm) {
544 		/* free all area regions */
545 		spin_lock_irqsave(&list_lock, flags);
546 		list_for_each_entry_safe(block, _block, &omap_dmm->alloc_head,
547 					alloc_node) {
548 			list_del(&block->alloc_node);
549 			kfree(block);
550 		}
551 		spin_unlock_irqrestore(&list_lock, flags);
552 
553 		for (i = 0; i < omap_dmm->num_lut; i++)
554 			if (omap_dmm->tcm && omap_dmm->tcm[i])
555 				omap_dmm->tcm[i]->deinit(omap_dmm->tcm[i]);
556 		kfree(omap_dmm->tcm);
557 
558 		kfree(omap_dmm->engines);
559 		if (omap_dmm->refill_va)
560 			dma_free_writecombine(omap_dmm->dev,
561 				REFILL_BUFFER_SIZE * omap_dmm->num_engines,
562 				omap_dmm->refill_va,
563 				omap_dmm->refill_pa);
564 		if (omap_dmm->dummy_page)
565 			__free_page(omap_dmm->dummy_page);
566 
567 		if (omap_dmm->irq > 0)
568 			free_irq(omap_dmm->irq, omap_dmm);
569 
570 		iounmap(omap_dmm->base);
571 		kfree(omap_dmm);
572 		omap_dmm = NULL;
573 	}
574 
575 	return 0;
576 }
577 
578 static int omap_dmm_probe(struct platform_device *dev)
579 {
580 	int ret = -EFAULT, i;
581 	struct tcm_area area = {0};
582 	u32 hwinfo, pat_geom;
583 	struct resource *mem;
584 
585 	omap_dmm = kzalloc(sizeof(*omap_dmm), GFP_KERNEL);
586 	if (!omap_dmm)
587 		goto fail;
588 
589 	/* initialize lists */
590 	INIT_LIST_HEAD(&omap_dmm->alloc_head);
591 	INIT_LIST_HEAD(&omap_dmm->idle_head);
592 
593 	init_waitqueue_head(&omap_dmm->engine_queue);
594 
595 	/* lookup hwmod data - base address and irq */
596 	mem = platform_get_resource(dev, IORESOURCE_MEM, 0);
597 	if (!mem) {
598 		dev_err(&dev->dev, "failed to get base address resource\n");
599 		goto fail;
600 	}
601 
602 	omap_dmm->base = ioremap(mem->start, SZ_2K);
603 
604 	if (!omap_dmm->base) {
605 		dev_err(&dev->dev, "failed to get dmm base address\n");
606 		goto fail;
607 	}
608 
609 	omap_dmm->irq = platform_get_irq(dev, 0);
610 	if (omap_dmm->irq < 0) {
611 		dev_err(&dev->dev, "failed to get IRQ resource\n");
612 		goto fail;
613 	}
614 
615 	omap_dmm->dev = &dev->dev;
616 
617 	hwinfo = readl(omap_dmm->base + DMM_PAT_HWINFO);
618 	omap_dmm->num_engines = (hwinfo >> 24) & 0x1F;
619 	omap_dmm->num_lut = (hwinfo >> 16) & 0x1F;
620 	omap_dmm->container_width = 256;
621 	omap_dmm->container_height = 128;
622 
623 	atomic_set(&omap_dmm->engine_counter, omap_dmm->num_engines);
624 
625 	/* read out actual LUT width and height */
626 	pat_geom = readl(omap_dmm->base + DMM_PAT_GEOMETRY);
627 	omap_dmm->lut_width = ((pat_geom >> 16) & 0xF) << 5;
628 	omap_dmm->lut_height = ((pat_geom >> 24) & 0xF) << 5;
629 
630 	/* increment LUT by one if on OMAP5 */
631 	/* LUT has twice the height, and is split into a separate container */
632 	if (omap_dmm->lut_height != omap_dmm->container_height)
633 		omap_dmm->num_lut++;
634 
635 	/* initialize DMM registers */
636 	writel(0x88888888, omap_dmm->base + DMM_PAT_VIEW__0);
637 	writel(0x88888888, omap_dmm->base + DMM_PAT_VIEW__1);
638 	writel(0x80808080, omap_dmm->base + DMM_PAT_VIEW_MAP__0);
639 	writel(0x80000000, omap_dmm->base + DMM_PAT_VIEW_MAP_BASE);
640 	writel(0x88888888, omap_dmm->base + DMM_TILER_OR__0);
641 	writel(0x88888888, omap_dmm->base + DMM_TILER_OR__1);
642 
643 	ret = request_irq(omap_dmm->irq, omap_dmm_irq_handler, IRQF_SHARED,
644 				"omap_dmm_irq_handler", omap_dmm);
645 
646 	if (ret) {
647 		dev_err(&dev->dev, "couldn't register IRQ %d, error %d\n",
648 			omap_dmm->irq, ret);
649 		omap_dmm->irq = -1;
650 		goto fail;
651 	}
652 
653 	/* Enable all interrupts for each refill engine except
654 	 * ERR_LUT_MISS<n> (which is just advisory, and we don't care
655 	 * about because we want to be able to refill live scanout
656 	 * buffers for accelerated pan/scroll) and FILL_DSC<n> which
657 	 * we just generally don't care about.
658 	 */
659 	writel(0x7e7e7e7e, omap_dmm->base + DMM_PAT_IRQENABLE_SET);
660 
661 	omap_dmm->dummy_page = alloc_page(GFP_KERNEL | __GFP_DMA32);
662 	if (!omap_dmm->dummy_page) {
663 		dev_err(&dev->dev, "could not allocate dummy page\n");
664 		ret = -ENOMEM;
665 		goto fail;
666 	}
667 
668 	/* set dma mask for device */
669 	ret = dma_set_coherent_mask(&dev->dev, DMA_BIT_MASK(32));
670 	if (ret)
671 		goto fail;
672 
673 	omap_dmm->dummy_pa = page_to_phys(omap_dmm->dummy_page);
674 
675 	/* alloc refill memory */
676 	omap_dmm->refill_va = dma_alloc_writecombine(&dev->dev,
677 				REFILL_BUFFER_SIZE * omap_dmm->num_engines,
678 				&omap_dmm->refill_pa, GFP_KERNEL);
679 	if (!omap_dmm->refill_va) {
680 		dev_err(&dev->dev, "could not allocate refill memory\n");
681 		goto fail;
682 	}
683 
684 	/* alloc engines */
685 	omap_dmm->engines = kcalloc(omap_dmm->num_engines,
686 				    sizeof(struct refill_engine), GFP_KERNEL);
687 	if (!omap_dmm->engines) {
688 		ret = -ENOMEM;
689 		goto fail;
690 	}
691 
692 	for (i = 0; i < omap_dmm->num_engines; i++) {
693 		omap_dmm->engines[i].id = i;
694 		omap_dmm->engines[i].dmm = omap_dmm;
695 		omap_dmm->engines[i].refill_va = omap_dmm->refill_va +
696 						(REFILL_BUFFER_SIZE * i);
697 		omap_dmm->engines[i].refill_pa = omap_dmm->refill_pa +
698 						(REFILL_BUFFER_SIZE * i);
699 		init_waitqueue_head(&omap_dmm->engines[i].wait_for_refill);
700 
701 		list_add(&omap_dmm->engines[i].idle_node, &omap_dmm->idle_head);
702 	}
703 
704 	omap_dmm->tcm = kcalloc(omap_dmm->num_lut, sizeof(*omap_dmm->tcm),
705 				GFP_KERNEL);
706 	if (!omap_dmm->tcm) {
707 		ret = -ENOMEM;
708 		goto fail;
709 	}
710 
711 	/* init containers */
712 	/* Each LUT is associated with a TCM (container manager).  We use the
713 	   lut_id to denote the lut_id used to identify the correct LUT for
714 	   programming during reill operations */
715 	for (i = 0; i < omap_dmm->num_lut; i++) {
716 		omap_dmm->tcm[i] = sita_init(omap_dmm->container_width,
717 						omap_dmm->container_height,
718 						NULL);
719 
720 		if (!omap_dmm->tcm[i]) {
721 			dev_err(&dev->dev, "failed to allocate container\n");
722 			ret = -ENOMEM;
723 			goto fail;
724 		}
725 
726 		omap_dmm->tcm[i]->lut_id = i;
727 	}
728 
729 	/* assign access mode containers to applicable tcm container */
730 	/* OMAP 4 has 1 container for all 4 views */
731 	/* OMAP 5 has 2 containers, 1 for 2D and 1 for 1D */
732 	containers[TILFMT_8BIT] = omap_dmm->tcm[0];
733 	containers[TILFMT_16BIT] = omap_dmm->tcm[0];
734 	containers[TILFMT_32BIT] = omap_dmm->tcm[0];
735 
736 	if (omap_dmm->container_height != omap_dmm->lut_height) {
737 		/* second LUT is used for PAGE mode.  Programming must use
738 		   y offset that is added to all y coordinates.  LUT id is still
739 		   0, because it is the same LUT, just the upper 128 lines */
740 		containers[TILFMT_PAGE] = omap_dmm->tcm[1];
741 		omap_dmm->tcm[1]->y_offset = OMAP5_LUT_OFFSET;
742 		omap_dmm->tcm[1]->lut_id = 0;
743 	} else {
744 		containers[TILFMT_PAGE] = omap_dmm->tcm[0];
745 	}
746 
747 	area = (struct tcm_area) {
748 		.tcm = NULL,
749 		.p1.x = omap_dmm->container_width - 1,
750 		.p1.y = omap_dmm->container_height - 1,
751 	};
752 
753 	/* initialize all LUTs to dummy page entries */
754 	for (i = 0; i < omap_dmm->num_lut; i++) {
755 		area.tcm = omap_dmm->tcm[i];
756 		if (fill(&area, NULL, 0, 0, true))
757 			dev_err(omap_dmm->dev, "refill failed");
758 	}
759 
760 	dev_info(omap_dmm->dev, "initialized all PAT entries\n");
761 
762 	return 0;
763 
764 fail:
765 	if (omap_dmm_remove(dev))
766 		dev_err(&dev->dev, "cleanup failed\n");
767 	return ret;
768 }
769 
770 /*
771  * debugfs support
772  */
773 
774 #ifdef CONFIG_DEBUG_FS
775 
776 static const char *alphabet = "abcdefghijklmnopqrstuvwxyz"
777 				"ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789";
778 static const char *special = ".,:;'\"`~!^-+";
779 
780 static void fill_map(char **map, int xdiv, int ydiv, struct tcm_area *a,
781 							char c, bool ovw)
782 {
783 	int x, y;
784 	for (y = a->p0.y / ydiv; y <= a->p1.y / ydiv; y++)
785 		for (x = a->p0.x / xdiv; x <= a->p1.x / xdiv; x++)
786 			if (map[y][x] == ' ' || ovw)
787 				map[y][x] = c;
788 }
789 
790 static void fill_map_pt(char **map, int xdiv, int ydiv, struct tcm_pt *p,
791 									char c)
792 {
793 	map[p->y / ydiv][p->x / xdiv] = c;
794 }
795 
796 static char read_map_pt(char **map, int xdiv, int ydiv, struct tcm_pt *p)
797 {
798 	return map[p->y / ydiv][p->x / xdiv];
799 }
800 
801 static int map_width(int xdiv, int x0, int x1)
802 {
803 	return (x1 / xdiv) - (x0 / xdiv) + 1;
804 }
805 
806 static void text_map(char **map, int xdiv, char *nice, int yd, int x0, int x1)
807 {
808 	char *p = map[yd] + (x0 / xdiv);
809 	int w = (map_width(xdiv, x0, x1) - strlen(nice)) / 2;
810 	if (w >= 0) {
811 		p += w;
812 		while (*nice)
813 			*p++ = *nice++;
814 	}
815 }
816 
817 static void map_1d_info(char **map, int xdiv, int ydiv, char *nice,
818 							struct tcm_area *a)
819 {
820 	sprintf(nice, "%dK", tcm_sizeof(*a) * 4);
821 	if (a->p0.y + 1 < a->p1.y) {
822 		text_map(map, xdiv, nice, (a->p0.y + a->p1.y) / 2 / ydiv, 0,
823 							256 - 1);
824 	} else if (a->p0.y < a->p1.y) {
825 		if (strlen(nice) < map_width(xdiv, a->p0.x, 256 - 1))
826 			text_map(map, xdiv, nice, a->p0.y / ydiv,
827 					a->p0.x + xdiv,	256 - 1);
828 		else if (strlen(nice) < map_width(xdiv, 0, a->p1.x))
829 			text_map(map, xdiv, nice, a->p1.y / ydiv,
830 					0, a->p1.y - xdiv);
831 	} else if (strlen(nice) + 1 < map_width(xdiv, a->p0.x, a->p1.x)) {
832 		text_map(map, xdiv, nice, a->p0.y / ydiv, a->p0.x, a->p1.x);
833 	}
834 }
835 
836 static void map_2d_info(char **map, int xdiv, int ydiv, char *nice,
837 							struct tcm_area *a)
838 {
839 	sprintf(nice, "(%d*%d)", tcm_awidth(*a), tcm_aheight(*a));
840 	if (strlen(nice) + 1 < map_width(xdiv, a->p0.x, a->p1.x))
841 		text_map(map, xdiv, nice, (a->p0.y + a->p1.y) / 2 / ydiv,
842 							a->p0.x, a->p1.x);
843 }
844 
845 int tiler_map_show(struct seq_file *s, void *arg)
846 {
847 	int xdiv = 2, ydiv = 1;
848 	char **map = NULL, *global_map;
849 	struct tiler_block *block;
850 	struct tcm_area a, p;
851 	int i;
852 	const char *m2d = alphabet;
853 	const char *a2d = special;
854 	const char *m2dp = m2d, *a2dp = a2d;
855 	char nice[128];
856 	int h_adj;
857 	int w_adj;
858 	unsigned long flags;
859 	int lut_idx;
860 
861 
862 	if (!omap_dmm) {
863 		/* early return if dmm/tiler device is not initialized */
864 		return 0;
865 	}
866 
867 	h_adj = omap_dmm->container_height / ydiv;
868 	w_adj = omap_dmm->container_width / xdiv;
869 
870 	map = kmalloc(h_adj * sizeof(*map), GFP_KERNEL);
871 	global_map = kmalloc((w_adj + 1) * h_adj, GFP_KERNEL);
872 
873 	if (!map || !global_map)
874 		goto error;
875 
876 	for (lut_idx = 0; lut_idx < omap_dmm->num_lut; lut_idx++) {
877 		memset(map, 0, h_adj * sizeof(*map));
878 		memset(global_map, ' ', (w_adj + 1) * h_adj);
879 
880 		for (i = 0; i < omap_dmm->container_height; i++) {
881 			map[i] = global_map + i * (w_adj + 1);
882 			map[i][w_adj] = 0;
883 		}
884 
885 		spin_lock_irqsave(&list_lock, flags);
886 
887 		list_for_each_entry(block, &omap_dmm->alloc_head, alloc_node) {
888 			if (block->area.tcm == omap_dmm->tcm[lut_idx]) {
889 				if (block->fmt != TILFMT_PAGE) {
890 					fill_map(map, xdiv, ydiv, &block->area,
891 						*m2dp, true);
892 					if (!*++a2dp)
893 						a2dp = a2d;
894 					if (!*++m2dp)
895 						m2dp = m2d;
896 					map_2d_info(map, xdiv, ydiv, nice,
897 							&block->area);
898 				} else {
899 					bool start = read_map_pt(map, xdiv,
900 						ydiv, &block->area.p0) == ' ';
901 					bool end = read_map_pt(map, xdiv, ydiv,
902 							&block->area.p1) == ' ';
903 
904 					tcm_for_each_slice(a, block->area, p)
905 						fill_map(map, xdiv, ydiv, &a,
906 							'=', true);
907 					fill_map_pt(map, xdiv, ydiv,
908 							&block->area.p0,
909 							start ? '<' : 'X');
910 					fill_map_pt(map, xdiv, ydiv,
911 							&block->area.p1,
912 							end ? '>' : 'X');
913 					map_1d_info(map, xdiv, ydiv, nice,
914 							&block->area);
915 				}
916 			}
917 		}
918 
919 		spin_unlock_irqrestore(&list_lock, flags);
920 
921 		if (s) {
922 			seq_printf(s, "CONTAINER %d DUMP BEGIN\n", lut_idx);
923 			for (i = 0; i < 128; i++)
924 				seq_printf(s, "%03d:%s\n", i, map[i]);
925 			seq_printf(s, "CONTAINER %d DUMP END\n", lut_idx);
926 		} else {
927 			dev_dbg(omap_dmm->dev, "CONTAINER %d DUMP BEGIN\n",
928 				lut_idx);
929 			for (i = 0; i < 128; i++)
930 				dev_dbg(omap_dmm->dev, "%03d:%s\n", i, map[i]);
931 			dev_dbg(omap_dmm->dev, "CONTAINER %d DUMP END\n",
932 				lut_idx);
933 		}
934 	}
935 
936 error:
937 	kfree(map);
938 	kfree(global_map);
939 
940 	return 0;
941 }
942 #endif
943 
944 #ifdef CONFIG_PM
945 static int omap_dmm_resume(struct device *dev)
946 {
947 	struct tcm_area area;
948 	int i;
949 
950 	if (!omap_dmm)
951 		return -ENODEV;
952 
953 	area = (struct tcm_area) {
954 		.tcm = NULL,
955 		.p1.x = omap_dmm->container_width - 1,
956 		.p1.y = omap_dmm->container_height - 1,
957 	};
958 
959 	/* initialize all LUTs to dummy page entries */
960 	for (i = 0; i < omap_dmm->num_lut; i++) {
961 		area.tcm = omap_dmm->tcm[i];
962 		if (fill(&area, NULL, 0, 0, true))
963 			dev_err(dev, "refill failed");
964 	}
965 
966 	return 0;
967 }
968 
969 static const struct dev_pm_ops omap_dmm_pm_ops = {
970 	.resume = omap_dmm_resume,
971 };
972 #endif
973 
974 #if defined(CONFIG_OF)
975 static const struct of_device_id dmm_of_match[] = {
976 	{ .compatible = "ti,omap4-dmm", },
977 	{ .compatible = "ti,omap5-dmm", },
978 	{},
979 };
980 #endif
981 
982 struct platform_driver omap_dmm_driver = {
983 	.probe = omap_dmm_probe,
984 	.remove = omap_dmm_remove,
985 	.driver = {
986 		.owner = THIS_MODULE,
987 		.name = DMM_DRIVER_NAME,
988 		.of_match_table = of_match_ptr(dmm_of_match),
989 #ifdef CONFIG_PM
990 		.pm = &omap_dmm_pm_ops,
991 #endif
992 	},
993 };
994 
995 MODULE_LICENSE("GPL v2");
996 MODULE_AUTHOR("Andy Gross <andy.gross@ti.com>");
997 MODULE_DESCRIPTION("OMAP DMM/Tiler Driver");
998 MODULE_ALIAS("platform:" DMM_DRIVER_NAME);
999