xref: /linux/drivers/gpu/drm/omapdrm/omap_dmm_tiler.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 
19 #include <linux/completion.h>
20 #include <linux/delay.h>
21 #include <linux/dma-mapping.h>
22 #include <linux/errno.h>
23 #include <linux/init.h>
24 #include <linux/interrupt.h>
25 #include <linux/list.h>
26 #include <linux/mm.h>
27 #include <linux/module.h>
28 #include <linux/platform_device.h> /* platform_device() */
29 #include <linux/sched.h>
30 #include <linux/slab.h>
31 #include <linux/time.h>
32 #include <linux/vmalloc.h>
33 #include <linux/wait.h>
34 
35 #include "omap_dmm_tiler.h"
36 #include "omap_dmm_priv.h"
37 
38 #define DMM_DRIVER_NAME "dmm"
39 
40 /* mappings for associating views to luts */
41 static struct tcm *containers[TILFMT_NFORMATS];
42 static struct dmm *omap_dmm;
43 
44 #if defined(CONFIG_OF)
45 static const struct of_device_id dmm_of_match[];
46 #endif
47 
48 /* global spinlock for protecting lists */
49 static DEFINE_SPINLOCK(list_lock);
50 
51 /* Geometry table */
52 #define GEOM(xshift, yshift, bytes_per_pixel) { \
53 		.x_shft = (xshift), \
54 		.y_shft = (yshift), \
55 		.cpp    = (bytes_per_pixel), \
56 		.slot_w = 1 << (SLOT_WIDTH_BITS - (xshift)), \
57 		.slot_h = 1 << (SLOT_HEIGHT_BITS - (yshift)), \
58 	}
59 
60 static const struct {
61 	uint32_t x_shft;	/* unused X-bits (as part of bpp) */
62 	uint32_t y_shft;	/* unused Y-bits (as part of bpp) */
63 	uint32_t cpp;		/* bytes/chars per pixel */
64 	uint32_t slot_w;	/* width of each slot (in pixels) */
65 	uint32_t slot_h;	/* height of each slot (in pixels) */
66 } geom[TILFMT_NFORMATS] = {
67 	[TILFMT_8BIT]  = GEOM(0, 0, 1),
68 	[TILFMT_16BIT] = GEOM(0, 1, 2),
69 	[TILFMT_32BIT] = GEOM(1, 1, 4),
70 	[TILFMT_PAGE]  = GEOM(SLOT_WIDTH_BITS, SLOT_HEIGHT_BITS, 1),
71 };
72 
73 
74 /* lookup table for registers w/ per-engine instances */
75 static const uint32_t reg[][4] = {
76 	[PAT_STATUS] = {DMM_PAT_STATUS__0, DMM_PAT_STATUS__1,
77 			DMM_PAT_STATUS__2, DMM_PAT_STATUS__3},
78 	[PAT_DESCR]  = {DMM_PAT_DESCR__0, DMM_PAT_DESCR__1,
79 			DMM_PAT_DESCR__2, DMM_PAT_DESCR__3},
80 };
81 
82 /* simple allocator to grab next 16 byte aligned memory from txn */
83 static void *alloc_dma(struct dmm_txn *txn, size_t sz, dma_addr_t *pa)
84 {
85 	void *ptr;
86 	struct refill_engine *engine = txn->engine_handle;
87 
88 	/* dmm programming requires 16 byte aligned addresses */
89 	txn->current_pa = round_up(txn->current_pa, 16);
90 	txn->current_va = (void *)round_up((long)txn->current_va, 16);
91 
92 	ptr = txn->current_va;
93 	*pa = txn->current_pa;
94 
95 	txn->current_pa += sz;
96 	txn->current_va += sz;
97 
98 	BUG_ON((txn->current_va - engine->refill_va) > REFILL_BUFFER_SIZE);
99 
100 	return ptr;
101 }
102 
103 /* check status and spin until wait_mask comes true */
104 static int wait_status(struct refill_engine *engine, uint32_t wait_mask)
105 {
106 	struct dmm *dmm = engine->dmm;
107 	uint32_t r = 0, err, i;
108 
109 	i = DMM_FIXED_RETRY_COUNT;
110 	while (true) {
111 		r = readl(dmm->base + reg[PAT_STATUS][engine->id]);
112 		err = r & DMM_PATSTATUS_ERR;
113 		if (err)
114 			return -EFAULT;
115 
116 		if ((r & wait_mask) == wait_mask)
117 			break;
118 
119 		if (--i == 0)
120 			return -ETIMEDOUT;
121 
122 		udelay(1);
123 	}
124 
125 	return 0;
126 }
127 
128 static void release_engine(struct refill_engine *engine)
129 {
130 	unsigned long flags;
131 
132 	spin_lock_irqsave(&list_lock, flags);
133 	list_add(&engine->idle_node, &omap_dmm->idle_head);
134 	spin_unlock_irqrestore(&list_lock, flags);
135 
136 	atomic_inc(&omap_dmm->engine_counter);
137 	wake_up_interruptible(&omap_dmm->engine_queue);
138 }
139 
140 static irqreturn_t omap_dmm_irq_handler(int irq, void *arg)
141 {
142 	struct dmm *dmm = arg;
143 	uint32_t status = readl(dmm->base + DMM_PAT_IRQSTATUS);
144 	int i;
145 
146 	/* ack IRQ */
147 	writel(status, dmm->base + DMM_PAT_IRQSTATUS);
148 
149 	for (i = 0; i < dmm->num_engines; i++) {
150 		if (status & DMM_IRQSTAT_LST) {
151 			if (dmm->engines[i].async)
152 				release_engine(&dmm->engines[i]);
153 
154 			complete(&dmm->engines[i].compl);
155 		}
156 
157 		status >>= 8;
158 	}
159 
160 	return IRQ_HANDLED;
161 }
162 
163 /**
164  * Get a handle for a DMM transaction
165  */
166 static struct dmm_txn *dmm_txn_init(struct dmm *dmm, struct tcm *tcm)
167 {
168 	struct dmm_txn *txn = NULL;
169 	struct refill_engine *engine = NULL;
170 	int ret;
171 	unsigned long flags;
172 
173 
174 	/* wait until an engine is available */
175 	ret = wait_event_interruptible(omap_dmm->engine_queue,
176 		atomic_add_unless(&omap_dmm->engine_counter, -1, 0));
177 	if (ret)
178 		return ERR_PTR(ret);
179 
180 	/* grab an idle engine */
181 	spin_lock_irqsave(&list_lock, flags);
182 	if (!list_empty(&dmm->idle_head)) {
183 		engine = list_entry(dmm->idle_head.next, struct refill_engine,
184 					idle_node);
185 		list_del(&engine->idle_node);
186 	}
187 	spin_unlock_irqrestore(&list_lock, flags);
188 
189 	BUG_ON(!engine);
190 
191 	txn = &engine->txn;
192 	engine->tcm = tcm;
193 	txn->engine_handle = engine;
194 	txn->last_pat = NULL;
195 	txn->current_va = engine->refill_va;
196 	txn->current_pa = engine->refill_pa;
197 
198 	return txn;
199 }
200 
201 /**
202  * Add region to DMM transaction.  If pages or pages[i] is NULL, then the
203  * corresponding slot is cleared (ie. dummy_pa is programmed)
204  */
205 static void dmm_txn_append(struct dmm_txn *txn, struct pat_area *area,
206 		struct page **pages, uint32_t npages, uint32_t roll)
207 {
208 	dma_addr_t pat_pa = 0, data_pa = 0;
209 	uint32_t *data;
210 	struct pat *pat;
211 	struct refill_engine *engine = txn->engine_handle;
212 	int columns = (1 + area->x1 - area->x0);
213 	int rows = (1 + area->y1 - area->y0);
214 	int i = columns*rows;
215 
216 	pat = alloc_dma(txn, sizeof(struct pat), &pat_pa);
217 
218 	if (txn->last_pat)
219 		txn->last_pat->next_pa = (uint32_t)pat_pa;
220 
221 	pat->area = *area;
222 
223 	/* adjust Y coordinates based off of container parameters */
224 	pat->area.y0 += engine->tcm->y_offset;
225 	pat->area.y1 += engine->tcm->y_offset;
226 
227 	pat->ctrl = (struct pat_ctrl){
228 			.start = 1,
229 			.lut_id = engine->tcm->lut_id,
230 		};
231 
232 	data = alloc_dma(txn, 4*i, &data_pa);
233 	/* FIXME: what if data_pa is more than 32-bit ? */
234 	pat->data_pa = data_pa;
235 
236 	while (i--) {
237 		int n = i + roll;
238 		if (n >= npages)
239 			n -= npages;
240 		data[i] = (pages && pages[n]) ?
241 			page_to_phys(pages[n]) : engine->dmm->dummy_pa;
242 	}
243 
244 	txn->last_pat = pat;
245 
246 	return;
247 }
248 
249 /**
250  * Commit the DMM transaction.
251  */
252 static int dmm_txn_commit(struct dmm_txn *txn, bool wait)
253 {
254 	int ret = 0;
255 	struct refill_engine *engine = txn->engine_handle;
256 	struct dmm *dmm = engine->dmm;
257 
258 	if (!txn->last_pat) {
259 		dev_err(engine->dmm->dev, "need at least one txn\n");
260 		ret = -EINVAL;
261 		goto cleanup;
262 	}
263 
264 	txn->last_pat->next_pa = 0;
265 
266 	/* write to PAT_DESCR to clear out any pending transaction */
267 	writel(0x0, dmm->base + reg[PAT_DESCR][engine->id]);
268 
269 	/* wait for engine ready: */
270 	ret = wait_status(engine, DMM_PATSTATUS_READY);
271 	if (ret) {
272 		ret = -EFAULT;
273 		goto cleanup;
274 	}
275 
276 	/* mark whether it is async to denote list management in IRQ handler */
277 	engine->async = wait ? false : true;
278 	reinit_completion(&engine->compl);
279 	/* verify that the irq handler sees the 'async' and completion value */
280 	smp_mb();
281 
282 	/* kick reload */
283 	writel(engine->refill_pa,
284 		dmm->base + reg[PAT_DESCR][engine->id]);
285 
286 	if (wait) {
287 		if (!wait_for_completion_timeout(&engine->compl,
288 				msecs_to_jiffies(100))) {
289 			dev_err(dmm->dev, "timed out waiting for done\n");
290 			ret = -ETIMEDOUT;
291 		}
292 	}
293 
294 cleanup:
295 	/* only place engine back on list if we are done with it */
296 	if (ret || wait)
297 		release_engine(engine);
298 
299 	return ret;
300 }
301 
302 /*
303  * DMM programming
304  */
305 static int fill(struct tcm_area *area, struct page **pages,
306 		uint32_t npages, uint32_t roll, bool wait)
307 {
308 	int ret = 0;
309 	struct tcm_area slice, area_s;
310 	struct dmm_txn *txn;
311 
312 	txn = dmm_txn_init(omap_dmm, area->tcm);
313 	if (IS_ERR_OR_NULL(txn))
314 		return -ENOMEM;
315 
316 	tcm_for_each_slice(slice, *area, area_s) {
317 		struct pat_area p_area = {
318 				.x0 = slice.p0.x,  .y0 = slice.p0.y,
319 				.x1 = slice.p1.x,  .y1 = slice.p1.y,
320 		};
321 
322 		dmm_txn_append(txn, &p_area, pages, npages, roll);
323 
324 		roll += tcm_sizeof(slice);
325 	}
326 
327 	ret = dmm_txn_commit(txn, wait);
328 
329 	return ret;
330 }
331 
332 /*
333  * Pin/unpin
334  */
335 
336 /* note: slots for which pages[i] == NULL are filled w/ dummy page
337  */
338 int tiler_pin(struct tiler_block *block, struct page **pages,
339 		uint32_t npages, uint32_t roll, bool wait)
340 {
341 	int ret;
342 
343 	ret = fill(&block->area, pages, npages, roll, wait);
344 
345 	if (ret)
346 		tiler_unpin(block);
347 
348 	return ret;
349 }
350 
351 int tiler_unpin(struct tiler_block *block)
352 {
353 	return fill(&block->area, NULL, 0, 0, false);
354 }
355 
356 /*
357  * Reserve/release
358  */
359 struct tiler_block *tiler_reserve_2d(enum tiler_fmt fmt, uint16_t w,
360 		uint16_t h, uint16_t align)
361 {
362 	struct tiler_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
363 	u32 min_align = 128;
364 	int ret;
365 	unsigned long flags;
366 
367 	BUG_ON(!validfmt(fmt));
368 
369 	/* convert width/height to slots */
370 	w = DIV_ROUND_UP(w, geom[fmt].slot_w);
371 	h = DIV_ROUND_UP(h, geom[fmt].slot_h);
372 
373 	/* convert alignment to slots */
374 	min_align = max(min_align, (geom[fmt].slot_w * geom[fmt].cpp));
375 	align = ALIGN(align, min_align);
376 	align /= geom[fmt].slot_w * geom[fmt].cpp;
377 
378 	block->fmt = fmt;
379 
380 	ret = tcm_reserve_2d(containers[fmt], w, h, align, &block->area);
381 	if (ret) {
382 		kfree(block);
383 		return ERR_PTR(-ENOMEM);
384 	}
385 
386 	/* add to allocation list */
387 	spin_lock_irqsave(&list_lock, flags);
388 	list_add(&block->alloc_node, &omap_dmm->alloc_head);
389 	spin_unlock_irqrestore(&list_lock, flags);
390 
391 	return block;
392 }
393 
394 struct tiler_block *tiler_reserve_1d(size_t size)
395 {
396 	struct tiler_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
397 	int num_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
398 	unsigned long flags;
399 
400 	if (!block)
401 		return ERR_PTR(-ENOMEM);
402 
403 	block->fmt = TILFMT_PAGE;
404 
405 	if (tcm_reserve_1d(containers[TILFMT_PAGE], num_pages,
406 				&block->area)) {
407 		kfree(block);
408 		return ERR_PTR(-ENOMEM);
409 	}
410 
411 	spin_lock_irqsave(&list_lock, flags);
412 	list_add(&block->alloc_node, &omap_dmm->alloc_head);
413 	spin_unlock_irqrestore(&list_lock, flags);
414 
415 	return block;
416 }
417 
418 /* note: if you have pin'd pages, you should have already unpin'd first! */
419 int tiler_release(struct tiler_block *block)
420 {
421 	int ret = tcm_free(&block->area);
422 	unsigned long flags;
423 
424 	if (block->area.tcm)
425 		dev_err(omap_dmm->dev, "failed to release block\n");
426 
427 	spin_lock_irqsave(&list_lock, flags);
428 	list_del(&block->alloc_node);
429 	spin_unlock_irqrestore(&list_lock, flags);
430 
431 	kfree(block);
432 	return ret;
433 }
434 
435 /*
436  * Utils
437  */
438 
439 /* calculate the tiler space address of a pixel in a view orientation...
440  * below description copied from the display subsystem section of TRM:
441  *
442  * When the TILER is addressed, the bits:
443  *   [28:27] = 0x0 for 8-bit tiled
444  *             0x1 for 16-bit tiled
445  *             0x2 for 32-bit tiled
446  *             0x3 for page mode
447  *   [31:29] = 0x0 for 0-degree view
448  *             0x1 for 180-degree view + mirroring
449  *             0x2 for 0-degree view + mirroring
450  *             0x3 for 180-degree view
451  *             0x4 for 270-degree view + mirroring
452  *             0x5 for 270-degree view
453  *             0x6 for 90-degree view
454  *             0x7 for 90-degree view + mirroring
455  * Otherwise the bits indicated the corresponding bit address to access
456  * the SDRAM.
457  */
458 static u32 tiler_get_address(enum tiler_fmt fmt, u32 orient, u32 x, u32 y)
459 {
460 	u32 x_bits, y_bits, tmp, x_mask, y_mask, alignment;
461 
462 	x_bits = CONT_WIDTH_BITS - geom[fmt].x_shft;
463 	y_bits = CONT_HEIGHT_BITS - geom[fmt].y_shft;
464 	alignment = geom[fmt].x_shft + geom[fmt].y_shft;
465 
466 	/* validate coordinate */
467 	x_mask = MASK(x_bits);
468 	y_mask = MASK(y_bits);
469 
470 	if (x < 0 || x > x_mask || y < 0 || y > y_mask) {
471 		DBG("invalid coords: %u < 0 || %u > %u || %u < 0 || %u > %u",
472 				x, x, x_mask, y, y, y_mask);
473 		return 0;
474 	}
475 
476 	/* account for mirroring */
477 	if (orient & MASK_X_INVERT)
478 		x ^= x_mask;
479 	if (orient & MASK_Y_INVERT)
480 		y ^= y_mask;
481 
482 	/* get coordinate address */
483 	if (orient & MASK_XY_FLIP)
484 		tmp = ((x << y_bits) + y);
485 	else
486 		tmp = ((y << x_bits) + x);
487 
488 	return TIL_ADDR((tmp << alignment), orient, fmt);
489 }
490 
491 dma_addr_t tiler_ssptr(struct tiler_block *block)
492 {
493 	BUG_ON(!validfmt(block->fmt));
494 
495 	return TILVIEW_8BIT + tiler_get_address(block->fmt, 0,
496 			block->area.p0.x * geom[block->fmt].slot_w,
497 			block->area.p0.y * geom[block->fmt].slot_h);
498 }
499 
500 dma_addr_t tiler_tsptr(struct tiler_block *block, uint32_t orient,
501 		uint32_t x, uint32_t y)
502 {
503 	struct tcm_pt *p = &block->area.p0;
504 	BUG_ON(!validfmt(block->fmt));
505 
506 	return tiler_get_address(block->fmt, orient,
507 			(p->x * geom[block->fmt].slot_w) + x,
508 			(p->y * geom[block->fmt].slot_h) + y);
509 }
510 
511 void tiler_align(enum tiler_fmt fmt, uint16_t *w, uint16_t *h)
512 {
513 	BUG_ON(!validfmt(fmt));
514 	*w = round_up(*w, geom[fmt].slot_w);
515 	*h = round_up(*h, geom[fmt].slot_h);
516 }
517 
518 uint32_t tiler_stride(enum tiler_fmt fmt, uint32_t orient)
519 {
520 	BUG_ON(!validfmt(fmt));
521 
522 	if (orient & MASK_XY_FLIP)
523 		return 1 << (CONT_HEIGHT_BITS + geom[fmt].x_shft);
524 	else
525 		return 1 << (CONT_WIDTH_BITS + geom[fmt].y_shft);
526 }
527 
528 size_t tiler_size(enum tiler_fmt fmt, uint16_t w, uint16_t h)
529 {
530 	tiler_align(fmt, &w, &h);
531 	return geom[fmt].cpp * w * h;
532 }
533 
534 size_t tiler_vsize(enum tiler_fmt fmt, uint16_t w, uint16_t h)
535 {
536 	BUG_ON(!validfmt(fmt));
537 	return round_up(geom[fmt].cpp * w, PAGE_SIZE) * h;
538 }
539 
540 uint32_t tiler_get_cpu_cache_flags(void)
541 {
542 	return omap_dmm->plat_data->cpu_cache_flags;
543 }
544 
545 bool dmm_is_available(void)
546 {
547 	return omap_dmm ? true : false;
548 }
549 
550 static int omap_dmm_remove(struct platform_device *dev)
551 {
552 	struct tiler_block *block, *_block;
553 	int i;
554 	unsigned long flags;
555 
556 	if (omap_dmm) {
557 		/* free all area regions */
558 		spin_lock_irqsave(&list_lock, flags);
559 		list_for_each_entry_safe(block, _block, &omap_dmm->alloc_head,
560 					alloc_node) {
561 			list_del(&block->alloc_node);
562 			kfree(block);
563 		}
564 		spin_unlock_irqrestore(&list_lock, flags);
565 
566 		for (i = 0; i < omap_dmm->num_lut; i++)
567 			if (omap_dmm->tcm && omap_dmm->tcm[i])
568 				omap_dmm->tcm[i]->deinit(omap_dmm->tcm[i]);
569 		kfree(omap_dmm->tcm);
570 
571 		kfree(omap_dmm->engines);
572 		if (omap_dmm->refill_va)
573 			dma_free_writecombine(omap_dmm->dev,
574 				REFILL_BUFFER_SIZE * omap_dmm->num_engines,
575 				omap_dmm->refill_va,
576 				omap_dmm->refill_pa);
577 		if (omap_dmm->dummy_page)
578 			__free_page(omap_dmm->dummy_page);
579 
580 		if (omap_dmm->irq > 0)
581 			free_irq(omap_dmm->irq, omap_dmm);
582 
583 		iounmap(omap_dmm->base);
584 		kfree(omap_dmm);
585 		omap_dmm = NULL;
586 	}
587 
588 	return 0;
589 }
590 
591 static int omap_dmm_probe(struct platform_device *dev)
592 {
593 	int ret = -EFAULT, i;
594 	struct tcm_area area = {0};
595 	u32 hwinfo, pat_geom;
596 	struct resource *mem;
597 
598 	omap_dmm = kzalloc(sizeof(*omap_dmm), GFP_KERNEL);
599 	if (!omap_dmm)
600 		goto fail;
601 
602 	/* initialize lists */
603 	INIT_LIST_HEAD(&omap_dmm->alloc_head);
604 	INIT_LIST_HEAD(&omap_dmm->idle_head);
605 
606 	init_waitqueue_head(&omap_dmm->engine_queue);
607 
608 	if (dev->dev.of_node) {
609 		const struct of_device_id *match;
610 
611 		match = of_match_node(dmm_of_match, dev->dev.of_node);
612 		if (!match) {
613 			dev_err(&dev->dev, "failed to find matching device node\n");
614 			return -ENODEV;
615 		}
616 
617 		omap_dmm->plat_data = match->data;
618 	}
619 
620 	/* lookup hwmod data - base address and irq */
621 	mem = platform_get_resource(dev, IORESOURCE_MEM, 0);
622 	if (!mem) {
623 		dev_err(&dev->dev, "failed to get base address resource\n");
624 		goto fail;
625 	}
626 
627 	omap_dmm->base = ioremap(mem->start, SZ_2K);
628 
629 	if (!omap_dmm->base) {
630 		dev_err(&dev->dev, "failed to get dmm base address\n");
631 		goto fail;
632 	}
633 
634 	omap_dmm->irq = platform_get_irq(dev, 0);
635 	if (omap_dmm->irq < 0) {
636 		dev_err(&dev->dev, "failed to get IRQ resource\n");
637 		goto fail;
638 	}
639 
640 	omap_dmm->dev = &dev->dev;
641 
642 	hwinfo = readl(omap_dmm->base + DMM_PAT_HWINFO);
643 	omap_dmm->num_engines = (hwinfo >> 24) & 0x1F;
644 	omap_dmm->num_lut = (hwinfo >> 16) & 0x1F;
645 	omap_dmm->container_width = 256;
646 	omap_dmm->container_height = 128;
647 
648 	atomic_set(&omap_dmm->engine_counter, omap_dmm->num_engines);
649 
650 	/* read out actual LUT width and height */
651 	pat_geom = readl(omap_dmm->base + DMM_PAT_GEOMETRY);
652 	omap_dmm->lut_width = ((pat_geom >> 16) & 0xF) << 5;
653 	omap_dmm->lut_height = ((pat_geom >> 24) & 0xF) << 5;
654 
655 	/* increment LUT by one if on OMAP5 */
656 	/* LUT has twice the height, and is split into a separate container */
657 	if (omap_dmm->lut_height != omap_dmm->container_height)
658 		omap_dmm->num_lut++;
659 
660 	/* initialize DMM registers */
661 	writel(0x88888888, omap_dmm->base + DMM_PAT_VIEW__0);
662 	writel(0x88888888, omap_dmm->base + DMM_PAT_VIEW__1);
663 	writel(0x80808080, omap_dmm->base + DMM_PAT_VIEW_MAP__0);
664 	writel(0x80000000, omap_dmm->base + DMM_PAT_VIEW_MAP_BASE);
665 	writel(0x88888888, omap_dmm->base + DMM_TILER_OR__0);
666 	writel(0x88888888, omap_dmm->base + DMM_TILER_OR__1);
667 
668 	ret = request_irq(omap_dmm->irq, omap_dmm_irq_handler, IRQF_SHARED,
669 				"omap_dmm_irq_handler", omap_dmm);
670 
671 	if (ret) {
672 		dev_err(&dev->dev, "couldn't register IRQ %d, error %d\n",
673 			omap_dmm->irq, ret);
674 		omap_dmm->irq = -1;
675 		goto fail;
676 	}
677 
678 	/* Enable all interrupts for each refill engine except
679 	 * ERR_LUT_MISS<n> (which is just advisory, and we don't care
680 	 * about because we want to be able to refill live scanout
681 	 * buffers for accelerated pan/scroll) and FILL_DSC<n> which
682 	 * we just generally don't care about.
683 	 */
684 	writel(0x7e7e7e7e, omap_dmm->base + DMM_PAT_IRQENABLE_SET);
685 
686 	omap_dmm->dummy_page = alloc_page(GFP_KERNEL | __GFP_DMA32);
687 	if (!omap_dmm->dummy_page) {
688 		dev_err(&dev->dev, "could not allocate dummy page\n");
689 		ret = -ENOMEM;
690 		goto fail;
691 	}
692 
693 	/* set dma mask for device */
694 	ret = dma_set_coherent_mask(&dev->dev, DMA_BIT_MASK(32));
695 	if (ret)
696 		goto fail;
697 
698 	omap_dmm->dummy_pa = page_to_phys(omap_dmm->dummy_page);
699 
700 	/* alloc refill memory */
701 	omap_dmm->refill_va = dma_alloc_writecombine(&dev->dev,
702 				REFILL_BUFFER_SIZE * omap_dmm->num_engines,
703 				&omap_dmm->refill_pa, GFP_KERNEL);
704 	if (!omap_dmm->refill_va) {
705 		dev_err(&dev->dev, "could not allocate refill memory\n");
706 		goto fail;
707 	}
708 
709 	/* alloc engines */
710 	omap_dmm->engines = kcalloc(omap_dmm->num_engines,
711 				    sizeof(struct refill_engine), GFP_KERNEL);
712 	if (!omap_dmm->engines) {
713 		ret = -ENOMEM;
714 		goto fail;
715 	}
716 
717 	for (i = 0; i < omap_dmm->num_engines; i++) {
718 		omap_dmm->engines[i].id = i;
719 		omap_dmm->engines[i].dmm = omap_dmm;
720 		omap_dmm->engines[i].refill_va = omap_dmm->refill_va +
721 						(REFILL_BUFFER_SIZE * i);
722 		omap_dmm->engines[i].refill_pa = omap_dmm->refill_pa +
723 						(REFILL_BUFFER_SIZE * i);
724 		init_completion(&omap_dmm->engines[i].compl);
725 
726 		list_add(&omap_dmm->engines[i].idle_node, &omap_dmm->idle_head);
727 	}
728 
729 	omap_dmm->tcm = kcalloc(omap_dmm->num_lut, sizeof(*omap_dmm->tcm),
730 				GFP_KERNEL);
731 	if (!omap_dmm->tcm) {
732 		ret = -ENOMEM;
733 		goto fail;
734 	}
735 
736 	/* init containers */
737 	/* Each LUT is associated with a TCM (container manager).  We use the
738 	   lut_id to denote the lut_id used to identify the correct LUT for
739 	   programming during reill operations */
740 	for (i = 0; i < omap_dmm->num_lut; i++) {
741 		omap_dmm->tcm[i] = sita_init(omap_dmm->container_width,
742 						omap_dmm->container_height,
743 						NULL);
744 
745 		if (!omap_dmm->tcm[i]) {
746 			dev_err(&dev->dev, "failed to allocate container\n");
747 			ret = -ENOMEM;
748 			goto fail;
749 		}
750 
751 		omap_dmm->tcm[i]->lut_id = i;
752 	}
753 
754 	/* assign access mode containers to applicable tcm container */
755 	/* OMAP 4 has 1 container for all 4 views */
756 	/* OMAP 5 has 2 containers, 1 for 2D and 1 for 1D */
757 	containers[TILFMT_8BIT] = omap_dmm->tcm[0];
758 	containers[TILFMT_16BIT] = omap_dmm->tcm[0];
759 	containers[TILFMT_32BIT] = omap_dmm->tcm[0];
760 
761 	if (omap_dmm->container_height != omap_dmm->lut_height) {
762 		/* second LUT is used for PAGE mode.  Programming must use
763 		   y offset that is added to all y coordinates.  LUT id is still
764 		   0, because it is the same LUT, just the upper 128 lines */
765 		containers[TILFMT_PAGE] = omap_dmm->tcm[1];
766 		omap_dmm->tcm[1]->y_offset = OMAP5_LUT_OFFSET;
767 		omap_dmm->tcm[1]->lut_id = 0;
768 	} else {
769 		containers[TILFMT_PAGE] = omap_dmm->tcm[0];
770 	}
771 
772 	area = (struct tcm_area) {
773 		.tcm = NULL,
774 		.p1.x = omap_dmm->container_width - 1,
775 		.p1.y = omap_dmm->container_height - 1,
776 	};
777 
778 	/* initialize all LUTs to dummy page entries */
779 	for (i = 0; i < omap_dmm->num_lut; i++) {
780 		area.tcm = omap_dmm->tcm[i];
781 		if (fill(&area, NULL, 0, 0, true))
782 			dev_err(omap_dmm->dev, "refill failed");
783 	}
784 
785 	dev_info(omap_dmm->dev, "initialized all PAT entries\n");
786 
787 	return 0;
788 
789 fail:
790 	if (omap_dmm_remove(dev))
791 		dev_err(&dev->dev, "cleanup failed\n");
792 	return ret;
793 }
794 
795 /*
796  * debugfs support
797  */
798 
799 #ifdef CONFIG_DEBUG_FS
800 
801 static const char *alphabet = "abcdefghijklmnopqrstuvwxyz"
802 				"ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789";
803 static const char *special = ".,:;'\"`~!^-+";
804 
805 static void fill_map(char **map, int xdiv, int ydiv, struct tcm_area *a,
806 							char c, bool ovw)
807 {
808 	int x, y;
809 	for (y = a->p0.y / ydiv; y <= a->p1.y / ydiv; y++)
810 		for (x = a->p0.x / xdiv; x <= a->p1.x / xdiv; x++)
811 			if (map[y][x] == ' ' || ovw)
812 				map[y][x] = c;
813 }
814 
815 static void fill_map_pt(char **map, int xdiv, int ydiv, struct tcm_pt *p,
816 									char c)
817 {
818 	map[p->y / ydiv][p->x / xdiv] = c;
819 }
820 
821 static char read_map_pt(char **map, int xdiv, int ydiv, struct tcm_pt *p)
822 {
823 	return map[p->y / ydiv][p->x / xdiv];
824 }
825 
826 static int map_width(int xdiv, int x0, int x1)
827 {
828 	return (x1 / xdiv) - (x0 / xdiv) + 1;
829 }
830 
831 static void text_map(char **map, int xdiv, char *nice, int yd, int x0, int x1)
832 {
833 	char *p = map[yd] + (x0 / xdiv);
834 	int w = (map_width(xdiv, x0, x1) - strlen(nice)) / 2;
835 	if (w >= 0) {
836 		p += w;
837 		while (*nice)
838 			*p++ = *nice++;
839 	}
840 }
841 
842 static void map_1d_info(char **map, int xdiv, int ydiv, char *nice,
843 							struct tcm_area *a)
844 {
845 	sprintf(nice, "%dK", tcm_sizeof(*a) * 4);
846 	if (a->p0.y + 1 < a->p1.y) {
847 		text_map(map, xdiv, nice, (a->p0.y + a->p1.y) / 2 / ydiv, 0,
848 							256 - 1);
849 	} else if (a->p0.y < a->p1.y) {
850 		if (strlen(nice) < map_width(xdiv, a->p0.x, 256 - 1))
851 			text_map(map, xdiv, nice, a->p0.y / ydiv,
852 					a->p0.x + xdiv,	256 - 1);
853 		else if (strlen(nice) < map_width(xdiv, 0, a->p1.x))
854 			text_map(map, xdiv, nice, a->p1.y / ydiv,
855 					0, a->p1.y - xdiv);
856 	} else if (strlen(nice) + 1 < map_width(xdiv, a->p0.x, a->p1.x)) {
857 		text_map(map, xdiv, nice, a->p0.y / ydiv, a->p0.x, a->p1.x);
858 	}
859 }
860 
861 static void map_2d_info(char **map, int xdiv, int ydiv, char *nice,
862 							struct tcm_area *a)
863 {
864 	sprintf(nice, "(%d*%d)", tcm_awidth(*a), tcm_aheight(*a));
865 	if (strlen(nice) + 1 < map_width(xdiv, a->p0.x, a->p1.x))
866 		text_map(map, xdiv, nice, (a->p0.y + a->p1.y) / 2 / ydiv,
867 							a->p0.x, a->p1.x);
868 }
869 
870 int tiler_map_show(struct seq_file *s, void *arg)
871 {
872 	int xdiv = 2, ydiv = 1;
873 	char **map = NULL, *global_map;
874 	struct tiler_block *block;
875 	struct tcm_area a, p;
876 	int i;
877 	const char *m2d = alphabet;
878 	const char *a2d = special;
879 	const char *m2dp = m2d, *a2dp = a2d;
880 	char nice[128];
881 	int h_adj;
882 	int w_adj;
883 	unsigned long flags;
884 	int lut_idx;
885 
886 
887 	if (!omap_dmm) {
888 		/* early return if dmm/tiler device is not initialized */
889 		return 0;
890 	}
891 
892 	h_adj = omap_dmm->container_height / ydiv;
893 	w_adj = omap_dmm->container_width / xdiv;
894 
895 	map = kmalloc(h_adj * sizeof(*map), GFP_KERNEL);
896 	global_map = kmalloc((w_adj + 1) * h_adj, GFP_KERNEL);
897 
898 	if (!map || !global_map)
899 		goto error;
900 
901 	for (lut_idx = 0; lut_idx < omap_dmm->num_lut; lut_idx++) {
902 		memset(map, 0, h_adj * sizeof(*map));
903 		memset(global_map, ' ', (w_adj + 1) * h_adj);
904 
905 		for (i = 0; i < omap_dmm->container_height; i++) {
906 			map[i] = global_map + i * (w_adj + 1);
907 			map[i][w_adj] = 0;
908 		}
909 
910 		spin_lock_irqsave(&list_lock, flags);
911 
912 		list_for_each_entry(block, &omap_dmm->alloc_head, alloc_node) {
913 			if (block->area.tcm == omap_dmm->tcm[lut_idx]) {
914 				if (block->fmt != TILFMT_PAGE) {
915 					fill_map(map, xdiv, ydiv, &block->area,
916 						*m2dp, true);
917 					if (!*++a2dp)
918 						a2dp = a2d;
919 					if (!*++m2dp)
920 						m2dp = m2d;
921 					map_2d_info(map, xdiv, ydiv, nice,
922 							&block->area);
923 				} else {
924 					bool start = read_map_pt(map, xdiv,
925 						ydiv, &block->area.p0) == ' ';
926 					bool end = read_map_pt(map, xdiv, ydiv,
927 							&block->area.p1) == ' ';
928 
929 					tcm_for_each_slice(a, block->area, p)
930 						fill_map(map, xdiv, ydiv, &a,
931 							'=', true);
932 					fill_map_pt(map, xdiv, ydiv,
933 							&block->area.p0,
934 							start ? '<' : 'X');
935 					fill_map_pt(map, xdiv, ydiv,
936 							&block->area.p1,
937 							end ? '>' : 'X');
938 					map_1d_info(map, xdiv, ydiv, nice,
939 							&block->area);
940 				}
941 			}
942 		}
943 
944 		spin_unlock_irqrestore(&list_lock, flags);
945 
946 		if (s) {
947 			seq_printf(s, "CONTAINER %d DUMP BEGIN\n", lut_idx);
948 			for (i = 0; i < 128; i++)
949 				seq_printf(s, "%03d:%s\n", i, map[i]);
950 			seq_printf(s, "CONTAINER %d DUMP END\n", lut_idx);
951 		} else {
952 			dev_dbg(omap_dmm->dev, "CONTAINER %d DUMP BEGIN\n",
953 				lut_idx);
954 			for (i = 0; i < 128; i++)
955 				dev_dbg(omap_dmm->dev, "%03d:%s\n", i, map[i]);
956 			dev_dbg(omap_dmm->dev, "CONTAINER %d DUMP END\n",
957 				lut_idx);
958 		}
959 	}
960 
961 error:
962 	kfree(map);
963 	kfree(global_map);
964 
965 	return 0;
966 }
967 #endif
968 
969 #ifdef CONFIG_PM_SLEEP
970 static int omap_dmm_resume(struct device *dev)
971 {
972 	struct tcm_area area;
973 	int i;
974 
975 	if (!omap_dmm)
976 		return -ENODEV;
977 
978 	area = (struct tcm_area) {
979 		.tcm = NULL,
980 		.p1.x = omap_dmm->container_width - 1,
981 		.p1.y = omap_dmm->container_height - 1,
982 	};
983 
984 	/* initialize all LUTs to dummy page entries */
985 	for (i = 0; i < omap_dmm->num_lut; i++) {
986 		area.tcm = omap_dmm->tcm[i];
987 		if (fill(&area, NULL, 0, 0, true))
988 			dev_err(dev, "refill failed");
989 	}
990 
991 	return 0;
992 }
993 #endif
994 
995 static SIMPLE_DEV_PM_OPS(omap_dmm_pm_ops, NULL, omap_dmm_resume);
996 
997 #if defined(CONFIG_OF)
998 static const struct dmm_platform_data dmm_omap4_platform_data = {
999 	.cpu_cache_flags = OMAP_BO_WC,
1000 };
1001 
1002 static const struct dmm_platform_data dmm_omap5_platform_data = {
1003 	.cpu_cache_flags = OMAP_BO_UNCACHED,
1004 };
1005 
1006 static const struct of_device_id dmm_of_match[] = {
1007 	{
1008 		.compatible = "ti,omap4-dmm",
1009 		.data = &dmm_omap4_platform_data,
1010 	},
1011 	{
1012 		.compatible = "ti,omap5-dmm",
1013 		.data = &dmm_omap5_platform_data,
1014 	},
1015 	{},
1016 };
1017 #endif
1018 
1019 struct platform_driver omap_dmm_driver = {
1020 	.probe = omap_dmm_probe,
1021 	.remove = omap_dmm_remove,
1022 	.driver = {
1023 		.owner = THIS_MODULE,
1024 		.name = DMM_DRIVER_NAME,
1025 		.of_match_table = of_match_ptr(dmm_of_match),
1026 		.pm = &omap_dmm_pm_ops,
1027 	},
1028 };
1029 
1030 MODULE_LICENSE("GPL v2");
1031 MODULE_AUTHOR("Andy Gross <andy.gross@ti.com>");
1032 MODULE_DESCRIPTION("OMAP DMM/Tiler Driver");
1033 MODULE_ALIAS("platform:" DMM_DRIVER_NAME);
1034