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;
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, &pat->data_pa);
227 
228 	while (i--) {
229 		int n = i + roll;
230 		if (n >= npages)
231 			n -= npages;
232 		data[i] = (pages && pages[n]) ?
233 			page_to_phys(pages[n]) : engine->dmm->dummy_pa;
234 	}
235 
236 	txn->last_pat = pat;
237 
238 	return;
239 }
240 
241 /**
242  * Commit the DMM transaction.
243  */
244 static int dmm_txn_commit(struct dmm_txn *txn, bool wait)
245 {
246 	int ret = 0;
247 	struct refill_engine *engine = txn->engine_handle;
248 	struct dmm *dmm = engine->dmm;
249 
250 	if (!txn->last_pat) {
251 		dev_err(engine->dmm->dev, "need at least one txn\n");
252 		ret = -EINVAL;
253 		goto cleanup;
254 	}
255 
256 	txn->last_pat->next_pa = 0;
257 
258 	/* write to PAT_DESCR to clear out any pending transaction */
259 	writel(0x0, dmm->base + reg[PAT_DESCR][engine->id]);
260 
261 	/* wait for engine ready: */
262 	ret = wait_status(engine, DMM_PATSTATUS_READY);
263 	if (ret) {
264 		ret = -EFAULT;
265 		goto cleanup;
266 	}
267 
268 	/* mark whether it is async to denote list management in IRQ handler */
269 	engine->async = wait ? false : true;
270 
271 	/* kick reload */
272 	writel(engine->refill_pa,
273 		dmm->base + reg[PAT_DESCR][engine->id]);
274 
275 	if (wait) {
276 		if (wait_event_interruptible_timeout(engine->wait_for_refill,
277 				wait_status(engine, DMM_PATSTATUS_READY) == 0,
278 				msecs_to_jiffies(1)) <= 0) {
279 			dev_err(dmm->dev, "timed out waiting for done\n");
280 			ret = -ETIMEDOUT;
281 		}
282 	}
283 
284 cleanup:
285 	/* only place engine back on list if we are done with it */
286 	if (ret || wait)
287 		release_engine(engine);
288 
289 	return ret;
290 }
291 
292 /*
293  * DMM programming
294  */
295 static int fill(struct tcm_area *area, struct page **pages,
296 		uint32_t npages, uint32_t roll, bool wait)
297 {
298 	int ret = 0;
299 	struct tcm_area slice, area_s;
300 	struct dmm_txn *txn;
301 
302 	txn = dmm_txn_init(omap_dmm, area->tcm);
303 	if (IS_ERR_OR_NULL(txn))
304 		return -ENOMEM;
305 
306 	tcm_for_each_slice(slice, *area, area_s) {
307 		struct pat_area p_area = {
308 				.x0 = slice.p0.x,  .y0 = slice.p0.y,
309 				.x1 = slice.p1.x,  .y1 = slice.p1.y,
310 		};
311 
312 		dmm_txn_append(txn, &p_area, pages, npages, roll);
313 
314 		roll += tcm_sizeof(slice);
315 	}
316 
317 	ret = dmm_txn_commit(txn, wait);
318 
319 	return ret;
320 }
321 
322 /*
323  * Pin/unpin
324  */
325 
326 /* note: slots for which pages[i] == NULL are filled w/ dummy page
327  */
328 int tiler_pin(struct tiler_block *block, struct page **pages,
329 		uint32_t npages, uint32_t roll, bool wait)
330 {
331 	int ret;
332 
333 	ret = fill(&block->area, pages, npages, roll, wait);
334 
335 	if (ret)
336 		tiler_unpin(block);
337 
338 	return ret;
339 }
340 
341 int tiler_unpin(struct tiler_block *block)
342 {
343 	return fill(&block->area, NULL, 0, 0, false);
344 }
345 
346 /*
347  * Reserve/release
348  */
349 struct tiler_block *tiler_reserve_2d(enum tiler_fmt fmt, uint16_t w,
350 		uint16_t h, uint16_t align)
351 {
352 	struct tiler_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
353 	u32 min_align = 128;
354 	int ret;
355 	unsigned long flags;
356 
357 	BUG_ON(!validfmt(fmt));
358 
359 	/* convert width/height to slots */
360 	w = DIV_ROUND_UP(w, geom[fmt].slot_w);
361 	h = DIV_ROUND_UP(h, geom[fmt].slot_h);
362 
363 	/* convert alignment to slots */
364 	min_align = max(min_align, (geom[fmt].slot_w * geom[fmt].cpp));
365 	align = ALIGN(align, min_align);
366 	align /= geom[fmt].slot_w * geom[fmt].cpp;
367 
368 	block->fmt = fmt;
369 
370 	ret = tcm_reserve_2d(containers[fmt], w, h, align, &block->area);
371 	if (ret) {
372 		kfree(block);
373 		return ERR_PTR(-ENOMEM);
374 	}
375 
376 	/* add to allocation list */
377 	spin_lock_irqsave(&list_lock, flags);
378 	list_add(&block->alloc_node, &omap_dmm->alloc_head);
379 	spin_unlock_irqrestore(&list_lock, flags);
380 
381 	return block;
382 }
383 
384 struct tiler_block *tiler_reserve_1d(size_t size)
385 {
386 	struct tiler_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
387 	int num_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
388 	unsigned long flags;
389 
390 	if (!block)
391 		return ERR_PTR(-ENOMEM);
392 
393 	block->fmt = TILFMT_PAGE;
394 
395 	if (tcm_reserve_1d(containers[TILFMT_PAGE], num_pages,
396 				&block->area)) {
397 		kfree(block);
398 		return ERR_PTR(-ENOMEM);
399 	}
400 
401 	spin_lock_irqsave(&list_lock, flags);
402 	list_add(&block->alloc_node, &omap_dmm->alloc_head);
403 	spin_unlock_irqrestore(&list_lock, flags);
404 
405 	return block;
406 }
407 
408 /* note: if you have pin'd pages, you should have already unpin'd first! */
409 int tiler_release(struct tiler_block *block)
410 {
411 	int ret = tcm_free(&block->area);
412 	unsigned long flags;
413 
414 	if (block->area.tcm)
415 		dev_err(omap_dmm->dev, "failed to release block\n");
416 
417 	spin_lock_irqsave(&list_lock, flags);
418 	list_del(&block->alloc_node);
419 	spin_unlock_irqrestore(&list_lock, flags);
420 
421 	kfree(block);
422 	return ret;
423 }
424 
425 /*
426  * Utils
427  */
428 
429 /* calculate the tiler space address of a pixel in a view orientation...
430  * below description copied from the display subsystem section of TRM:
431  *
432  * When the TILER is addressed, the bits:
433  *   [28:27] = 0x0 for 8-bit tiled
434  *             0x1 for 16-bit tiled
435  *             0x2 for 32-bit tiled
436  *             0x3 for page mode
437  *   [31:29] = 0x0 for 0-degree view
438  *             0x1 for 180-degree view + mirroring
439  *             0x2 for 0-degree view + mirroring
440  *             0x3 for 180-degree view
441  *             0x4 for 270-degree view + mirroring
442  *             0x5 for 270-degree view
443  *             0x6 for 90-degree view
444  *             0x7 for 90-degree view + mirroring
445  * Otherwise the bits indicated the corresponding bit address to access
446  * the SDRAM.
447  */
448 static u32 tiler_get_address(enum tiler_fmt fmt, u32 orient, u32 x, u32 y)
449 {
450 	u32 x_bits, y_bits, tmp, x_mask, y_mask, alignment;
451 
452 	x_bits = CONT_WIDTH_BITS - geom[fmt].x_shft;
453 	y_bits = CONT_HEIGHT_BITS - geom[fmt].y_shft;
454 	alignment = geom[fmt].x_shft + geom[fmt].y_shft;
455 
456 	/* validate coordinate */
457 	x_mask = MASK(x_bits);
458 	y_mask = MASK(y_bits);
459 
460 	if (x < 0 || x > x_mask || y < 0 || y > y_mask) {
461 		DBG("invalid coords: %u < 0 || %u > %u || %u < 0 || %u > %u",
462 				x, x, x_mask, y, y, y_mask);
463 		return 0;
464 	}
465 
466 	/* account for mirroring */
467 	if (orient & MASK_X_INVERT)
468 		x ^= x_mask;
469 	if (orient & MASK_Y_INVERT)
470 		y ^= y_mask;
471 
472 	/* get coordinate address */
473 	if (orient & MASK_XY_FLIP)
474 		tmp = ((x << y_bits) + y);
475 	else
476 		tmp = ((y << x_bits) + x);
477 
478 	return TIL_ADDR((tmp << alignment), orient, fmt);
479 }
480 
481 dma_addr_t tiler_ssptr(struct tiler_block *block)
482 {
483 	BUG_ON(!validfmt(block->fmt));
484 
485 	return TILVIEW_8BIT + tiler_get_address(block->fmt, 0,
486 			block->area.p0.x * geom[block->fmt].slot_w,
487 			block->area.p0.y * geom[block->fmt].slot_h);
488 }
489 
490 dma_addr_t tiler_tsptr(struct tiler_block *block, uint32_t orient,
491 		uint32_t x, uint32_t y)
492 {
493 	struct tcm_pt *p = &block->area.p0;
494 	BUG_ON(!validfmt(block->fmt));
495 
496 	return tiler_get_address(block->fmt, orient,
497 			(p->x * geom[block->fmt].slot_w) + x,
498 			(p->y * geom[block->fmt].slot_h) + y);
499 }
500 
501 void tiler_align(enum tiler_fmt fmt, uint16_t *w, uint16_t *h)
502 {
503 	BUG_ON(!validfmt(fmt));
504 	*w = round_up(*w, geom[fmt].slot_w);
505 	*h = round_up(*h, geom[fmt].slot_h);
506 }
507 
508 uint32_t tiler_stride(enum tiler_fmt fmt, uint32_t orient)
509 {
510 	BUG_ON(!validfmt(fmt));
511 
512 	if (orient & MASK_XY_FLIP)
513 		return 1 << (CONT_HEIGHT_BITS + geom[fmt].x_shft);
514 	else
515 		return 1 << (CONT_WIDTH_BITS + geom[fmt].y_shft);
516 }
517 
518 size_t tiler_size(enum tiler_fmt fmt, uint16_t w, uint16_t h)
519 {
520 	tiler_align(fmt, &w, &h);
521 	return geom[fmt].cpp * w * h;
522 }
523 
524 size_t tiler_vsize(enum tiler_fmt fmt, uint16_t w, uint16_t h)
525 {
526 	BUG_ON(!validfmt(fmt));
527 	return round_up(geom[fmt].cpp * w, PAGE_SIZE) * h;
528 }
529 
530 bool dmm_is_available(void)
531 {
532 	return omap_dmm ? true : false;
533 }
534 
535 static int omap_dmm_remove(struct platform_device *dev)
536 {
537 	struct tiler_block *block, *_block;
538 	int i;
539 	unsigned long flags;
540 
541 	if (omap_dmm) {
542 		/* free all area regions */
543 		spin_lock_irqsave(&list_lock, flags);
544 		list_for_each_entry_safe(block, _block, &omap_dmm->alloc_head,
545 					alloc_node) {
546 			list_del(&block->alloc_node);
547 			kfree(block);
548 		}
549 		spin_unlock_irqrestore(&list_lock, flags);
550 
551 		for (i = 0; i < omap_dmm->num_lut; i++)
552 			if (omap_dmm->tcm && omap_dmm->tcm[i])
553 				omap_dmm->tcm[i]->deinit(omap_dmm->tcm[i]);
554 		kfree(omap_dmm->tcm);
555 
556 		kfree(omap_dmm->engines);
557 		if (omap_dmm->refill_va)
558 			dma_free_writecombine(omap_dmm->dev,
559 				REFILL_BUFFER_SIZE * omap_dmm->num_engines,
560 				omap_dmm->refill_va,
561 				omap_dmm->refill_pa);
562 		if (omap_dmm->dummy_page)
563 			__free_page(omap_dmm->dummy_page);
564 
565 		if (omap_dmm->irq > 0)
566 			free_irq(omap_dmm->irq, omap_dmm);
567 
568 		iounmap(omap_dmm->base);
569 		kfree(omap_dmm);
570 		omap_dmm = NULL;
571 	}
572 
573 	return 0;
574 }
575 
576 static int omap_dmm_probe(struct platform_device *dev)
577 {
578 	int ret = -EFAULT, i;
579 	struct tcm_area area = {0};
580 	u32 hwinfo, pat_geom;
581 	struct resource *mem;
582 
583 	omap_dmm = kzalloc(sizeof(*omap_dmm), GFP_KERNEL);
584 	if (!omap_dmm)
585 		goto fail;
586 
587 	/* initialize lists */
588 	INIT_LIST_HEAD(&omap_dmm->alloc_head);
589 	INIT_LIST_HEAD(&omap_dmm->idle_head);
590 
591 	init_waitqueue_head(&omap_dmm->engine_queue);
592 
593 	/* lookup hwmod data - base address and irq */
594 	mem = platform_get_resource(dev, IORESOURCE_MEM, 0);
595 	if (!mem) {
596 		dev_err(&dev->dev, "failed to get base address resource\n");
597 		goto fail;
598 	}
599 
600 	omap_dmm->base = ioremap(mem->start, SZ_2K);
601 
602 	if (!omap_dmm->base) {
603 		dev_err(&dev->dev, "failed to get dmm base address\n");
604 		goto fail;
605 	}
606 
607 	omap_dmm->irq = platform_get_irq(dev, 0);
608 	if (omap_dmm->irq < 0) {
609 		dev_err(&dev->dev, "failed to get IRQ resource\n");
610 		goto fail;
611 	}
612 
613 	omap_dmm->dev = &dev->dev;
614 
615 	hwinfo = readl(omap_dmm->base + DMM_PAT_HWINFO);
616 	omap_dmm->num_engines = (hwinfo >> 24) & 0x1F;
617 	omap_dmm->num_lut = (hwinfo >> 16) & 0x1F;
618 	omap_dmm->container_width = 256;
619 	omap_dmm->container_height = 128;
620 
621 	atomic_set(&omap_dmm->engine_counter, omap_dmm->num_engines);
622 
623 	/* read out actual LUT width and height */
624 	pat_geom = readl(omap_dmm->base + DMM_PAT_GEOMETRY);
625 	omap_dmm->lut_width = ((pat_geom >> 16) & 0xF) << 5;
626 	omap_dmm->lut_height = ((pat_geom >> 24) & 0xF) << 5;
627 
628 	/* increment LUT by one if on OMAP5 */
629 	/* LUT has twice the height, and is split into a separate container */
630 	if (omap_dmm->lut_height != omap_dmm->container_height)
631 		omap_dmm->num_lut++;
632 
633 	/* initialize DMM registers */
634 	writel(0x88888888, omap_dmm->base + DMM_PAT_VIEW__0);
635 	writel(0x88888888, omap_dmm->base + DMM_PAT_VIEW__1);
636 	writel(0x80808080, omap_dmm->base + DMM_PAT_VIEW_MAP__0);
637 	writel(0x80000000, omap_dmm->base + DMM_PAT_VIEW_MAP_BASE);
638 	writel(0x88888888, omap_dmm->base + DMM_TILER_OR__0);
639 	writel(0x88888888, omap_dmm->base + DMM_TILER_OR__1);
640 
641 	ret = request_irq(omap_dmm->irq, omap_dmm_irq_handler, IRQF_SHARED,
642 				"omap_dmm_irq_handler", omap_dmm);
643 
644 	if (ret) {
645 		dev_err(&dev->dev, "couldn't register IRQ %d, error %d\n",
646 			omap_dmm->irq, ret);
647 		omap_dmm->irq = -1;
648 		goto fail;
649 	}
650 
651 	/* Enable all interrupts for each refill engine except
652 	 * ERR_LUT_MISS<n> (which is just advisory, and we don't care
653 	 * about because we want to be able to refill live scanout
654 	 * buffers for accelerated pan/scroll) and FILL_DSC<n> which
655 	 * we just generally don't care about.
656 	 */
657 	writel(0x7e7e7e7e, omap_dmm->base + DMM_PAT_IRQENABLE_SET);
658 
659 	omap_dmm->dummy_page = alloc_page(GFP_KERNEL | __GFP_DMA32);
660 	if (!omap_dmm->dummy_page) {
661 		dev_err(&dev->dev, "could not allocate dummy page\n");
662 		ret = -ENOMEM;
663 		goto fail;
664 	}
665 
666 	/* set dma mask for device */
667 	ret = dma_set_coherent_mask(&dev->dev, DMA_BIT_MASK(32));
668 	if (ret)
669 		goto fail;
670 
671 	omap_dmm->dummy_pa = page_to_phys(omap_dmm->dummy_page);
672 
673 	/* alloc refill memory */
674 	omap_dmm->refill_va = dma_alloc_writecombine(&dev->dev,
675 				REFILL_BUFFER_SIZE * omap_dmm->num_engines,
676 				&omap_dmm->refill_pa, GFP_KERNEL);
677 	if (!omap_dmm->refill_va) {
678 		dev_err(&dev->dev, "could not allocate refill memory\n");
679 		goto fail;
680 	}
681 
682 	/* alloc engines */
683 	omap_dmm->engines = kcalloc(omap_dmm->num_engines,
684 				    sizeof(struct refill_engine), GFP_KERNEL);
685 	if (!omap_dmm->engines) {
686 		ret = -ENOMEM;
687 		goto fail;
688 	}
689 
690 	for (i = 0; i < omap_dmm->num_engines; i++) {
691 		omap_dmm->engines[i].id = i;
692 		omap_dmm->engines[i].dmm = omap_dmm;
693 		omap_dmm->engines[i].refill_va = omap_dmm->refill_va +
694 						(REFILL_BUFFER_SIZE * i);
695 		omap_dmm->engines[i].refill_pa = omap_dmm->refill_pa +
696 						(REFILL_BUFFER_SIZE * i);
697 		init_waitqueue_head(&omap_dmm->engines[i].wait_for_refill);
698 
699 		list_add(&omap_dmm->engines[i].idle_node, &omap_dmm->idle_head);
700 	}
701 
702 	omap_dmm->tcm = kcalloc(omap_dmm->num_lut, sizeof(*omap_dmm->tcm),
703 				GFP_KERNEL);
704 	if (!omap_dmm->tcm) {
705 		ret = -ENOMEM;
706 		goto fail;
707 	}
708 
709 	/* init containers */
710 	/* Each LUT is associated with a TCM (container manager).  We use the
711 	   lut_id to denote the lut_id used to identify the correct LUT for
712 	   programming during reill operations */
713 	for (i = 0; i < omap_dmm->num_lut; i++) {
714 		omap_dmm->tcm[i] = sita_init(omap_dmm->container_width,
715 						omap_dmm->container_height,
716 						NULL);
717 
718 		if (!omap_dmm->tcm[i]) {
719 			dev_err(&dev->dev, "failed to allocate container\n");
720 			ret = -ENOMEM;
721 			goto fail;
722 		}
723 
724 		omap_dmm->tcm[i]->lut_id = i;
725 	}
726 
727 	/* assign access mode containers to applicable tcm container */
728 	/* OMAP 4 has 1 container for all 4 views */
729 	/* OMAP 5 has 2 containers, 1 for 2D and 1 for 1D */
730 	containers[TILFMT_8BIT] = omap_dmm->tcm[0];
731 	containers[TILFMT_16BIT] = omap_dmm->tcm[0];
732 	containers[TILFMT_32BIT] = omap_dmm->tcm[0];
733 
734 	if (omap_dmm->container_height != omap_dmm->lut_height) {
735 		/* second LUT is used for PAGE mode.  Programming must use
736 		   y offset that is added to all y coordinates.  LUT id is still
737 		   0, because it is the same LUT, just the upper 128 lines */
738 		containers[TILFMT_PAGE] = omap_dmm->tcm[1];
739 		omap_dmm->tcm[1]->y_offset = OMAP5_LUT_OFFSET;
740 		omap_dmm->tcm[1]->lut_id = 0;
741 	} else {
742 		containers[TILFMT_PAGE] = omap_dmm->tcm[0];
743 	}
744 
745 	area = (struct tcm_area) {
746 		.tcm = NULL,
747 		.p1.x = omap_dmm->container_width - 1,
748 		.p1.y = omap_dmm->container_height - 1,
749 	};
750 
751 	/* initialize all LUTs to dummy page entries */
752 	for (i = 0; i < omap_dmm->num_lut; i++) {
753 		area.tcm = omap_dmm->tcm[i];
754 		if (fill(&area, NULL, 0, 0, true))
755 			dev_err(omap_dmm->dev, "refill failed");
756 	}
757 
758 	dev_info(omap_dmm->dev, "initialized all PAT entries\n");
759 
760 	return 0;
761 
762 fail:
763 	if (omap_dmm_remove(dev))
764 		dev_err(&dev->dev, "cleanup failed\n");
765 	return ret;
766 }
767 
768 /*
769  * debugfs support
770  */
771 
772 #ifdef CONFIG_DEBUG_FS
773 
774 static const char *alphabet = "abcdefghijklmnopqrstuvwxyz"
775 				"ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789";
776 static const char *special = ".,:;'\"`~!^-+";
777 
778 static void fill_map(char **map, int xdiv, int ydiv, struct tcm_area *a,
779 							char c, bool ovw)
780 {
781 	int x, y;
782 	for (y = a->p0.y / ydiv; y <= a->p1.y / ydiv; y++)
783 		for (x = a->p0.x / xdiv; x <= a->p1.x / xdiv; x++)
784 			if (map[y][x] == ' ' || ovw)
785 				map[y][x] = c;
786 }
787 
788 static void fill_map_pt(char **map, int xdiv, int ydiv, struct tcm_pt *p,
789 									char c)
790 {
791 	map[p->y / ydiv][p->x / xdiv] = c;
792 }
793 
794 static char read_map_pt(char **map, int xdiv, int ydiv, struct tcm_pt *p)
795 {
796 	return map[p->y / ydiv][p->x / xdiv];
797 }
798 
799 static int map_width(int xdiv, int x0, int x1)
800 {
801 	return (x1 / xdiv) - (x0 / xdiv) + 1;
802 }
803 
804 static void text_map(char **map, int xdiv, char *nice, int yd, int x0, int x1)
805 {
806 	char *p = map[yd] + (x0 / xdiv);
807 	int w = (map_width(xdiv, x0, x1) - strlen(nice)) / 2;
808 	if (w >= 0) {
809 		p += w;
810 		while (*nice)
811 			*p++ = *nice++;
812 	}
813 }
814 
815 static void map_1d_info(char **map, int xdiv, int ydiv, char *nice,
816 							struct tcm_area *a)
817 {
818 	sprintf(nice, "%dK", tcm_sizeof(*a) * 4);
819 	if (a->p0.y + 1 < a->p1.y) {
820 		text_map(map, xdiv, nice, (a->p0.y + a->p1.y) / 2 / ydiv, 0,
821 							256 - 1);
822 	} else if (a->p0.y < a->p1.y) {
823 		if (strlen(nice) < map_width(xdiv, a->p0.x, 256 - 1))
824 			text_map(map, xdiv, nice, a->p0.y / ydiv,
825 					a->p0.x + xdiv,	256 - 1);
826 		else if (strlen(nice) < map_width(xdiv, 0, a->p1.x))
827 			text_map(map, xdiv, nice, a->p1.y / ydiv,
828 					0, a->p1.y - xdiv);
829 	} else if (strlen(nice) + 1 < map_width(xdiv, a->p0.x, a->p1.x)) {
830 		text_map(map, xdiv, nice, a->p0.y / ydiv, a->p0.x, a->p1.x);
831 	}
832 }
833 
834 static void map_2d_info(char **map, int xdiv, int ydiv, char *nice,
835 							struct tcm_area *a)
836 {
837 	sprintf(nice, "(%d*%d)", tcm_awidth(*a), tcm_aheight(*a));
838 	if (strlen(nice) + 1 < map_width(xdiv, a->p0.x, a->p1.x))
839 		text_map(map, xdiv, nice, (a->p0.y + a->p1.y) / 2 / ydiv,
840 							a->p0.x, a->p1.x);
841 }
842 
843 int tiler_map_show(struct seq_file *s, void *arg)
844 {
845 	int xdiv = 2, ydiv = 1;
846 	char **map = NULL, *global_map;
847 	struct tiler_block *block;
848 	struct tcm_area a, p;
849 	int i;
850 	const char *m2d = alphabet;
851 	const char *a2d = special;
852 	const char *m2dp = m2d, *a2dp = a2d;
853 	char nice[128];
854 	int h_adj;
855 	int w_adj;
856 	unsigned long flags;
857 	int lut_idx;
858 
859 
860 	if (!omap_dmm) {
861 		/* early return if dmm/tiler device is not initialized */
862 		return 0;
863 	}
864 
865 	h_adj = omap_dmm->container_height / ydiv;
866 	w_adj = omap_dmm->container_width / xdiv;
867 
868 	map = kmalloc(h_adj * sizeof(*map), GFP_KERNEL);
869 	global_map = kmalloc((w_adj + 1) * h_adj, GFP_KERNEL);
870 
871 	if (!map || !global_map)
872 		goto error;
873 
874 	for (lut_idx = 0; lut_idx < omap_dmm->num_lut; lut_idx++) {
875 		memset(map, 0, h_adj * sizeof(*map));
876 		memset(global_map, ' ', (w_adj + 1) * h_adj);
877 
878 		for (i = 0; i < omap_dmm->container_height; i++) {
879 			map[i] = global_map + i * (w_adj + 1);
880 			map[i][w_adj] = 0;
881 		}
882 
883 		spin_lock_irqsave(&list_lock, flags);
884 
885 		list_for_each_entry(block, &omap_dmm->alloc_head, alloc_node) {
886 			if (block->area.tcm == omap_dmm->tcm[lut_idx]) {
887 				if (block->fmt != TILFMT_PAGE) {
888 					fill_map(map, xdiv, ydiv, &block->area,
889 						*m2dp, true);
890 					if (!*++a2dp)
891 						a2dp = a2d;
892 					if (!*++m2dp)
893 						m2dp = m2d;
894 					map_2d_info(map, xdiv, ydiv, nice,
895 							&block->area);
896 				} else {
897 					bool start = read_map_pt(map, xdiv,
898 						ydiv, &block->area.p0) == ' ';
899 					bool end = read_map_pt(map, xdiv, ydiv,
900 							&block->area.p1) == ' ';
901 
902 					tcm_for_each_slice(a, block->area, p)
903 						fill_map(map, xdiv, ydiv, &a,
904 							'=', true);
905 					fill_map_pt(map, xdiv, ydiv,
906 							&block->area.p0,
907 							start ? '<' : 'X');
908 					fill_map_pt(map, xdiv, ydiv,
909 							&block->area.p1,
910 							end ? '>' : 'X');
911 					map_1d_info(map, xdiv, ydiv, nice,
912 							&block->area);
913 				}
914 			}
915 		}
916 
917 		spin_unlock_irqrestore(&list_lock, flags);
918 
919 		if (s) {
920 			seq_printf(s, "CONTAINER %d DUMP BEGIN\n", lut_idx);
921 			for (i = 0; i < 128; i++)
922 				seq_printf(s, "%03d:%s\n", i, map[i]);
923 			seq_printf(s, "CONTAINER %d DUMP END\n", lut_idx);
924 		} else {
925 			dev_dbg(omap_dmm->dev, "CONTAINER %d DUMP BEGIN\n",
926 				lut_idx);
927 			for (i = 0; i < 128; i++)
928 				dev_dbg(omap_dmm->dev, "%03d:%s\n", i, map[i]);
929 			dev_dbg(omap_dmm->dev, "CONTAINER %d DUMP END\n",
930 				lut_idx);
931 		}
932 	}
933 
934 error:
935 	kfree(map);
936 	kfree(global_map);
937 
938 	return 0;
939 }
940 #endif
941 
942 #ifdef CONFIG_PM
943 static int omap_dmm_resume(struct device *dev)
944 {
945 	struct tcm_area area;
946 	int i;
947 
948 	if (!omap_dmm)
949 		return -ENODEV;
950 
951 	area = (struct tcm_area) {
952 		.tcm = NULL,
953 		.p1.x = omap_dmm->container_width - 1,
954 		.p1.y = omap_dmm->container_height - 1,
955 	};
956 
957 	/* initialize all LUTs to dummy page entries */
958 	for (i = 0; i < omap_dmm->num_lut; i++) {
959 		area.tcm = omap_dmm->tcm[i];
960 		if (fill(&area, NULL, 0, 0, true))
961 			dev_err(dev, "refill failed");
962 	}
963 
964 	return 0;
965 }
966 
967 static const struct dev_pm_ops omap_dmm_pm_ops = {
968 	.resume = omap_dmm_resume,
969 };
970 #endif
971 
972 struct platform_driver omap_dmm_driver = {
973 	.probe = omap_dmm_probe,
974 	.remove = omap_dmm_remove,
975 	.driver = {
976 		.owner = THIS_MODULE,
977 		.name = DMM_DRIVER_NAME,
978 #ifdef CONFIG_PM
979 		.pm = &omap_dmm_pm_ops,
980 #endif
981 	},
982 };
983 
984 MODULE_LICENSE("GPL v2");
985 MODULE_AUTHOR("Andy Gross <andy.gross@ti.com>");
986 MODULE_DESCRIPTION("OMAP DMM/Tiler Driver");
987 MODULE_ALIAS("platform:" DMM_DRIVER_NAME);
988