Loading...
1/*
2 * Copyright (C) 2011 Texas Instruments Incorporated - http://www.ti.com/
3 * Author: Rob Clark <rob.clark@linaro.org>
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 as published by
7 * the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along with
15 * this program. If not, see <http://www.gnu.org/licenses/>.
16 */
17
18#include <linux/seq_file.h>
19#include <linux/shmem_fs.h>
20#include <linux/spinlock.h>
21#include <linux/pfn_t.h>
22
23#include <drm/drm_vma_manager.h>
24
25#include "omap_drv.h"
26#include "omap_dmm_tiler.h"
27
28/*
29 * GEM buffer object implementation.
30 */
31
32/* note: we use upper 8 bits of flags for driver-internal flags: */
33#define OMAP_BO_MEM_DMA_API 0x01000000 /* memory allocated with the dma_alloc_* API */
34#define OMAP_BO_MEM_SHMEM 0x02000000 /* memory allocated through shmem backing */
35#define OMAP_BO_MEM_DMABUF 0x08000000 /* memory imported from a dmabuf */
36
37struct omap_gem_object {
38 struct drm_gem_object base;
39
40 struct list_head mm_list;
41
42 u32 flags;
43
44 /** width/height for tiled formats (rounded up to slot boundaries) */
45 u16 width, height;
46
47 /** roll applied when mapping to DMM */
48 u32 roll;
49
50 /**
51 * dma_addr contains the buffer DMA address. It is valid for
52 *
53 * - buffers allocated through the DMA mapping API (with the
54 * OMAP_BO_MEM_DMA_API flag set)
55 *
56 * - buffers imported from dmabuf (with the OMAP_BO_MEM_DMABUF flag set)
57 * if they are physically contiguous (when sgt->orig_nents == 1)
58 *
59 * - buffers mapped through the TILER when dma_addr_cnt is not zero, in
60 * which case the DMA address points to the TILER aperture
61 *
62 * Physically contiguous buffers have their DMA address equal to the
63 * physical address as we don't remap those buffers through the TILER.
64 *
65 * Buffers mapped to the TILER have their DMA address pointing to the
66 * TILER aperture. As TILER mappings are refcounted (through
67 * dma_addr_cnt) the DMA address must be accessed through omap_gem_pin()
68 * to ensure that the mapping won't disappear unexpectedly. References
69 * must be released with omap_gem_unpin().
70 */
71 dma_addr_t dma_addr;
72
73 /**
74 * # of users of dma_addr
75 */
76 u32 dma_addr_cnt;
77
78 /**
79 * If the buffer has been imported from a dmabuf the OMAP_DB_DMABUF flag
80 * is set and the sgt field is valid.
81 */
82 struct sg_table *sgt;
83
84 /**
85 * tiler block used when buffer is remapped in DMM/TILER.
86 */
87 struct tiler_block *block;
88
89 /**
90 * Array of backing pages, if allocated. Note that pages are never
91 * allocated for buffers originally allocated from contiguous memory
92 */
93 struct page **pages;
94
95 /** addresses corresponding to pages in above array */
96 dma_addr_t *dma_addrs;
97
98 /**
99 * Virtual address, if mapped.
100 */
101 void *vaddr;
102};
103
104#define to_omap_bo(x) container_of(x, struct omap_gem_object, base)
105
106/* To deal with userspace mmap'ings of 2d tiled buffers, which (a) are
107 * not necessarily pinned in TILER all the time, and (b) when they are
108 * they are not necessarily page aligned, we reserve one or more small
109 * regions in each of the 2d containers to use as a user-GART where we
110 * can create a second page-aligned mapping of parts of the buffer
111 * being accessed from userspace.
112 *
113 * Note that we could optimize slightly when we know that multiple
114 * tiler containers are backed by the same PAT.. but I'll leave that
115 * for later..
116 */
117#define NUM_USERGART_ENTRIES 2
118struct omap_drm_usergart_entry {
119 struct tiler_block *block; /* the reserved tiler block */
120 dma_addr_t dma_addr;
121 struct drm_gem_object *obj; /* the current pinned obj */
122 pgoff_t obj_pgoff; /* page offset of obj currently
123 mapped in */
124};
125
126struct omap_drm_usergart {
127 struct omap_drm_usergart_entry entry[NUM_USERGART_ENTRIES];
128 int height; /* height in rows */
129 int height_shift; /* ilog2(height in rows) */
130 int slot_shift; /* ilog2(width per slot) */
131 int stride_pfn; /* stride in pages */
132 int last; /* index of last used entry */
133};
134
135/* -----------------------------------------------------------------------------
136 * Helpers
137 */
138
139/** get mmap offset */
140static u64 mmap_offset(struct drm_gem_object *obj)
141{
142 struct drm_device *dev = obj->dev;
143 int ret;
144 size_t size;
145
146 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
147
148 /* Make it mmapable */
149 size = omap_gem_mmap_size(obj);
150 ret = drm_gem_create_mmap_offset_size(obj, size);
151 if (ret) {
152 dev_err(dev->dev, "could not allocate mmap offset\n");
153 return 0;
154 }
155
156 return drm_vma_node_offset_addr(&obj->vma_node);
157}
158
159static bool is_contiguous(struct omap_gem_object *omap_obj)
160{
161 if (omap_obj->flags & OMAP_BO_MEM_DMA_API)
162 return true;
163
164 if ((omap_obj->flags & OMAP_BO_MEM_DMABUF) && omap_obj->sgt->nents == 1)
165 return true;
166
167 return false;
168}
169
170/* -----------------------------------------------------------------------------
171 * Eviction
172 */
173
174static void evict_entry(struct drm_gem_object *obj,
175 enum tiler_fmt fmt, struct omap_drm_usergart_entry *entry)
176{
177 struct omap_gem_object *omap_obj = to_omap_bo(obj);
178 struct omap_drm_private *priv = obj->dev->dev_private;
179 int n = priv->usergart[fmt].height;
180 size_t size = PAGE_SIZE * n;
181 loff_t off = mmap_offset(obj) +
182 (entry->obj_pgoff << PAGE_SHIFT);
183 const int m = DIV_ROUND_UP(omap_obj->width << fmt, PAGE_SIZE);
184
185 if (m > 1) {
186 int i;
187 /* if stride > than PAGE_SIZE then sparse mapping: */
188 for (i = n; i > 0; i--) {
189 unmap_mapping_range(obj->dev->anon_inode->i_mapping,
190 off, PAGE_SIZE, 1);
191 off += PAGE_SIZE * m;
192 }
193 } else {
194 unmap_mapping_range(obj->dev->anon_inode->i_mapping,
195 off, size, 1);
196 }
197
198 entry->obj = NULL;
199}
200
201/* Evict a buffer from usergart, if it is mapped there */
202static void evict(struct drm_gem_object *obj)
203{
204 struct omap_gem_object *omap_obj = to_omap_bo(obj);
205 struct omap_drm_private *priv = obj->dev->dev_private;
206
207 if (omap_obj->flags & OMAP_BO_TILED) {
208 enum tiler_fmt fmt = gem2fmt(omap_obj->flags);
209 int i;
210
211 for (i = 0; i < NUM_USERGART_ENTRIES; i++) {
212 struct omap_drm_usergart_entry *entry =
213 &priv->usergart[fmt].entry[i];
214
215 if (entry->obj == obj)
216 evict_entry(obj, fmt, entry);
217 }
218 }
219}
220
221/* -----------------------------------------------------------------------------
222 * Page Management
223 */
224
225/** ensure backing pages are allocated */
226static int omap_gem_attach_pages(struct drm_gem_object *obj)
227{
228 struct drm_device *dev = obj->dev;
229 struct omap_gem_object *omap_obj = to_omap_bo(obj);
230 struct page **pages;
231 int npages = obj->size >> PAGE_SHIFT;
232 int i, ret;
233 dma_addr_t *addrs;
234
235 WARN_ON(omap_obj->pages);
236
237 pages = drm_gem_get_pages(obj);
238 if (IS_ERR(pages)) {
239 dev_err(obj->dev->dev, "could not get pages: %ld\n", PTR_ERR(pages));
240 return PTR_ERR(pages);
241 }
242
243 /* for non-cached buffers, ensure the new pages are clean because
244 * DSS, GPU, etc. are not cache coherent:
245 */
246 if (omap_obj->flags & (OMAP_BO_WC|OMAP_BO_UNCACHED)) {
247 addrs = kmalloc(npages * sizeof(*addrs), GFP_KERNEL);
248 if (!addrs) {
249 ret = -ENOMEM;
250 goto free_pages;
251 }
252
253 for (i = 0; i < npages; i++) {
254 addrs[i] = dma_map_page(dev->dev, pages[i],
255 0, PAGE_SIZE, DMA_TO_DEVICE);
256
257 if (dma_mapping_error(dev->dev, addrs[i])) {
258 dev_warn(dev->dev,
259 "%s: failed to map page\n", __func__);
260
261 for (i = i - 1; i >= 0; --i) {
262 dma_unmap_page(dev->dev, addrs[i],
263 PAGE_SIZE, DMA_TO_DEVICE);
264 }
265
266 ret = -ENOMEM;
267 goto free_addrs;
268 }
269 }
270 } else {
271 addrs = kzalloc(npages * sizeof(*addrs), GFP_KERNEL);
272 if (!addrs) {
273 ret = -ENOMEM;
274 goto free_pages;
275 }
276 }
277
278 omap_obj->dma_addrs = addrs;
279 omap_obj->pages = pages;
280
281 return 0;
282
283free_addrs:
284 kfree(addrs);
285free_pages:
286 drm_gem_put_pages(obj, pages, true, false);
287
288 return ret;
289}
290
291/* acquire pages when needed (for example, for DMA where physically
292 * contiguous buffer is not required
293 */
294static int get_pages(struct drm_gem_object *obj, struct page ***pages)
295{
296 struct omap_gem_object *omap_obj = to_omap_bo(obj);
297 int ret = 0;
298
299 if ((omap_obj->flags & OMAP_BO_MEM_SHMEM) && !omap_obj->pages) {
300 ret = omap_gem_attach_pages(obj);
301 if (ret) {
302 dev_err(obj->dev->dev, "could not attach pages\n");
303 return ret;
304 }
305 }
306
307 /* TODO: even phys-contig.. we should have a list of pages? */
308 *pages = omap_obj->pages;
309
310 return 0;
311}
312
313/** release backing pages */
314static void omap_gem_detach_pages(struct drm_gem_object *obj)
315{
316 struct omap_gem_object *omap_obj = to_omap_bo(obj);
317 unsigned int npages = obj->size >> PAGE_SHIFT;
318 unsigned int i;
319
320 for (i = 0; i < npages; i++) {
321 if (omap_obj->dma_addrs[i])
322 dma_unmap_page(obj->dev->dev, omap_obj->dma_addrs[i],
323 PAGE_SIZE, DMA_TO_DEVICE);
324 }
325
326 kfree(omap_obj->dma_addrs);
327 omap_obj->dma_addrs = NULL;
328
329 drm_gem_put_pages(obj, omap_obj->pages, true, false);
330 omap_obj->pages = NULL;
331}
332
333/* get buffer flags */
334u32 omap_gem_flags(struct drm_gem_object *obj)
335{
336 return to_omap_bo(obj)->flags;
337}
338
339u64 omap_gem_mmap_offset(struct drm_gem_object *obj)
340{
341 u64 offset;
342
343 mutex_lock(&obj->dev->struct_mutex);
344 offset = mmap_offset(obj);
345 mutex_unlock(&obj->dev->struct_mutex);
346 return offset;
347}
348
349/** get mmap size */
350size_t omap_gem_mmap_size(struct drm_gem_object *obj)
351{
352 struct omap_gem_object *omap_obj = to_omap_bo(obj);
353 size_t size = obj->size;
354
355 if (omap_obj->flags & OMAP_BO_TILED) {
356 /* for tiled buffers, the virtual size has stride rounded up
357 * to 4kb.. (to hide the fact that row n+1 might start 16kb or
358 * 32kb later!). But we don't back the entire buffer with
359 * pages, only the valid picture part.. so need to adjust for
360 * this in the size used to mmap and generate mmap offset
361 */
362 size = tiler_vsize(gem2fmt(omap_obj->flags),
363 omap_obj->width, omap_obj->height);
364 }
365
366 return size;
367}
368
369/* -----------------------------------------------------------------------------
370 * Fault Handling
371 */
372
373/* Normal handling for the case of faulting in non-tiled buffers */
374static int fault_1d(struct drm_gem_object *obj,
375 struct vm_area_struct *vma, struct vm_fault *vmf)
376{
377 struct omap_gem_object *omap_obj = to_omap_bo(obj);
378 unsigned long pfn;
379 pgoff_t pgoff;
380
381 /* We don't use vmf->pgoff since that has the fake offset: */
382 pgoff = (vmf->address - vma->vm_start) >> PAGE_SHIFT;
383
384 if (omap_obj->pages) {
385 omap_gem_cpu_sync_page(obj, pgoff);
386 pfn = page_to_pfn(omap_obj->pages[pgoff]);
387 } else {
388 BUG_ON(!is_contiguous(omap_obj));
389 pfn = (omap_obj->dma_addr >> PAGE_SHIFT) + pgoff;
390 }
391
392 VERB("Inserting %p pfn %lx, pa %lx", (void *)vmf->address,
393 pfn, pfn << PAGE_SHIFT);
394
395 return vm_insert_mixed(vma, vmf->address, __pfn_to_pfn_t(pfn, PFN_DEV));
396}
397
398/* Special handling for the case of faulting in 2d tiled buffers */
399static int fault_2d(struct drm_gem_object *obj,
400 struct vm_area_struct *vma, struct vm_fault *vmf)
401{
402 struct omap_gem_object *omap_obj = to_omap_bo(obj);
403 struct omap_drm_private *priv = obj->dev->dev_private;
404 struct omap_drm_usergart_entry *entry;
405 enum tiler_fmt fmt = gem2fmt(omap_obj->flags);
406 struct page *pages[64]; /* XXX is this too much to have on stack? */
407 unsigned long pfn;
408 pgoff_t pgoff, base_pgoff;
409 unsigned long vaddr;
410 int i, ret, slots;
411
412 /*
413 * Note the height of the slot is also equal to the number of pages
414 * that need to be mapped in to fill 4kb wide CPU page. If the slot
415 * height is 64, then 64 pages fill a 4kb wide by 64 row region.
416 */
417 const int n = priv->usergart[fmt].height;
418 const int n_shift = priv->usergart[fmt].height_shift;
419
420 /*
421 * If buffer width in bytes > PAGE_SIZE then the virtual stride is
422 * rounded up to next multiple of PAGE_SIZE.. this need to be taken
423 * into account in some of the math, so figure out virtual stride
424 * in pages
425 */
426 const int m = DIV_ROUND_UP(omap_obj->width << fmt, PAGE_SIZE);
427
428 /* We don't use vmf->pgoff since that has the fake offset: */
429 pgoff = (vmf->address - vma->vm_start) >> PAGE_SHIFT;
430
431 /*
432 * Actual address we start mapping at is rounded down to previous slot
433 * boundary in the y direction:
434 */
435 base_pgoff = round_down(pgoff, m << n_shift);
436
437 /* figure out buffer width in slots */
438 slots = omap_obj->width >> priv->usergart[fmt].slot_shift;
439
440 vaddr = vmf->address - ((pgoff - base_pgoff) << PAGE_SHIFT);
441
442 entry = &priv->usergart[fmt].entry[priv->usergart[fmt].last];
443
444 /* evict previous buffer using this usergart entry, if any: */
445 if (entry->obj)
446 evict_entry(entry->obj, fmt, entry);
447
448 entry->obj = obj;
449 entry->obj_pgoff = base_pgoff;
450
451 /* now convert base_pgoff to phys offset from virt offset: */
452 base_pgoff = (base_pgoff >> n_shift) * slots;
453
454 /* for wider-than 4k.. figure out which part of the slot-row we want: */
455 if (m > 1) {
456 int off = pgoff % m;
457 entry->obj_pgoff += off;
458 base_pgoff /= m;
459 slots = min(slots - (off << n_shift), n);
460 base_pgoff += off << n_shift;
461 vaddr += off << PAGE_SHIFT;
462 }
463
464 /*
465 * Map in pages. Beyond the valid pixel part of the buffer, we set
466 * pages[i] to NULL to get a dummy page mapped in.. if someone
467 * reads/writes it they will get random/undefined content, but at
468 * least it won't be corrupting whatever other random page used to
469 * be mapped in, or other undefined behavior.
470 */
471 memcpy(pages, &omap_obj->pages[base_pgoff],
472 sizeof(struct page *) * slots);
473 memset(pages + slots, 0,
474 sizeof(struct page *) * (n - slots));
475
476 ret = tiler_pin(entry->block, pages, ARRAY_SIZE(pages), 0, true);
477 if (ret) {
478 dev_err(obj->dev->dev, "failed to pin: %d\n", ret);
479 return ret;
480 }
481
482 pfn = entry->dma_addr >> PAGE_SHIFT;
483
484 VERB("Inserting %p pfn %lx, pa %lx", (void *)vmf->address,
485 pfn, pfn << PAGE_SHIFT);
486
487 for (i = n; i > 0; i--) {
488 vm_insert_mixed(vma, vaddr, __pfn_to_pfn_t(pfn, PFN_DEV));
489 pfn += priv->usergart[fmt].stride_pfn;
490 vaddr += PAGE_SIZE * m;
491 }
492
493 /* simple round-robin: */
494 priv->usergart[fmt].last = (priv->usergart[fmt].last + 1)
495 % NUM_USERGART_ENTRIES;
496
497 return 0;
498}
499
500/**
501 * omap_gem_fault - pagefault handler for GEM objects
502 * @vmf: fault detail
503 *
504 * Invoked when a fault occurs on an mmap of a GEM managed area. GEM
505 * does most of the work for us including the actual map/unmap calls
506 * but we need to do the actual page work.
507 *
508 * The VMA was set up by GEM. In doing so it also ensured that the
509 * vma->vm_private_data points to the GEM object that is backing this
510 * mapping.
511 */
512int omap_gem_fault(struct vm_fault *vmf)
513{
514 struct vm_area_struct *vma = vmf->vma;
515 struct drm_gem_object *obj = vma->vm_private_data;
516 struct omap_gem_object *omap_obj = to_omap_bo(obj);
517 struct drm_device *dev = obj->dev;
518 struct page **pages;
519 int ret;
520
521 /* Make sure we don't parallel update on a fault, nor move or remove
522 * something from beneath our feet
523 */
524 mutex_lock(&dev->struct_mutex);
525
526 /* if a shmem backed object, make sure we have pages attached now */
527 ret = get_pages(obj, &pages);
528 if (ret)
529 goto fail;
530
531 /* where should we do corresponding put_pages().. we are mapping
532 * the original page, rather than thru a GART, so we can't rely
533 * on eviction to trigger this. But munmap() or all mappings should
534 * probably trigger put_pages()?
535 */
536
537 if (omap_obj->flags & OMAP_BO_TILED)
538 ret = fault_2d(obj, vma, vmf);
539 else
540 ret = fault_1d(obj, vma, vmf);
541
542
543fail:
544 mutex_unlock(&dev->struct_mutex);
545 switch (ret) {
546 case 0:
547 case -ERESTARTSYS:
548 case -EINTR:
549 case -EBUSY:
550 /*
551 * EBUSY is ok: this just means that another thread
552 * already did the job.
553 */
554 return VM_FAULT_NOPAGE;
555 case -ENOMEM:
556 return VM_FAULT_OOM;
557 default:
558 return VM_FAULT_SIGBUS;
559 }
560}
561
562/** We override mainly to fix up some of the vm mapping flags.. */
563int omap_gem_mmap(struct file *filp, struct vm_area_struct *vma)
564{
565 int ret;
566
567 ret = drm_gem_mmap(filp, vma);
568 if (ret) {
569 DBG("mmap failed: %d", ret);
570 return ret;
571 }
572
573 return omap_gem_mmap_obj(vma->vm_private_data, vma);
574}
575
576int omap_gem_mmap_obj(struct drm_gem_object *obj,
577 struct vm_area_struct *vma)
578{
579 struct omap_gem_object *omap_obj = to_omap_bo(obj);
580
581 vma->vm_flags &= ~VM_PFNMAP;
582 vma->vm_flags |= VM_MIXEDMAP;
583
584 if (omap_obj->flags & OMAP_BO_WC) {
585 vma->vm_page_prot = pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
586 } else if (omap_obj->flags & OMAP_BO_UNCACHED) {
587 vma->vm_page_prot = pgprot_noncached(vm_get_page_prot(vma->vm_flags));
588 } else {
589 /*
590 * We do have some private objects, at least for scanout buffers
591 * on hardware without DMM/TILER. But these are allocated write-
592 * combine
593 */
594 if (WARN_ON(!obj->filp))
595 return -EINVAL;
596
597 /*
598 * Shunt off cached objs to shmem file so they have their own
599 * address_space (so unmap_mapping_range does what we want,
600 * in particular in the case of mmap'd dmabufs)
601 */
602 fput(vma->vm_file);
603 vma->vm_pgoff = 0;
604 vma->vm_file = get_file(obj->filp);
605
606 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
607 }
608
609 return 0;
610}
611
612/* -----------------------------------------------------------------------------
613 * Dumb Buffers
614 */
615
616/**
617 * omap_gem_dumb_create - create a dumb buffer
618 * @drm_file: our client file
619 * @dev: our device
620 * @args: the requested arguments copied from userspace
621 *
622 * Allocate a buffer suitable for use for a frame buffer of the
623 * form described by user space. Give userspace a handle by which
624 * to reference it.
625 */
626int omap_gem_dumb_create(struct drm_file *file, struct drm_device *dev,
627 struct drm_mode_create_dumb *args)
628{
629 union omap_gem_size gsize;
630
631 args->pitch = DIV_ROUND_UP(args->width * args->bpp, 8);
632
633 args->size = PAGE_ALIGN(args->pitch * args->height);
634
635 gsize = (union omap_gem_size){
636 .bytes = args->size,
637 };
638
639 return omap_gem_new_handle(dev, file, gsize,
640 OMAP_BO_SCANOUT | OMAP_BO_WC, &args->handle);
641}
642
643/**
644 * omap_gem_dumb_map - buffer mapping for dumb interface
645 * @file: our drm client file
646 * @dev: drm device
647 * @handle: GEM handle to the object (from dumb_create)
648 *
649 * Do the necessary setup to allow the mapping of the frame buffer
650 * into user memory. We don't have to do much here at the moment.
651 */
652int omap_gem_dumb_map_offset(struct drm_file *file, struct drm_device *dev,
653 u32 handle, u64 *offset)
654{
655 struct drm_gem_object *obj;
656 int ret = 0;
657
658 /* GEM does all our handle to object mapping */
659 obj = drm_gem_object_lookup(file, handle);
660 if (obj == NULL) {
661 ret = -ENOENT;
662 goto fail;
663 }
664
665 *offset = omap_gem_mmap_offset(obj);
666
667 drm_gem_object_unreference_unlocked(obj);
668
669fail:
670 return ret;
671}
672
673#ifdef CONFIG_DRM_FBDEV_EMULATION
674/* Set scrolling position. This allows us to implement fast scrolling
675 * for console.
676 *
677 * Call only from non-atomic contexts.
678 */
679int omap_gem_roll(struct drm_gem_object *obj, u32 roll)
680{
681 struct omap_gem_object *omap_obj = to_omap_bo(obj);
682 u32 npages = obj->size >> PAGE_SHIFT;
683 int ret = 0;
684
685 if (roll > npages) {
686 dev_err(obj->dev->dev, "invalid roll: %d\n", roll);
687 return -EINVAL;
688 }
689
690 omap_obj->roll = roll;
691
692 mutex_lock(&obj->dev->struct_mutex);
693
694 /* if we aren't mapped yet, we don't need to do anything */
695 if (omap_obj->block) {
696 struct page **pages;
697 ret = get_pages(obj, &pages);
698 if (ret)
699 goto fail;
700 ret = tiler_pin(omap_obj->block, pages, npages, roll, true);
701 if (ret)
702 dev_err(obj->dev->dev, "could not repin: %d\n", ret);
703 }
704
705fail:
706 mutex_unlock(&obj->dev->struct_mutex);
707
708 return ret;
709}
710#endif
711
712/* -----------------------------------------------------------------------------
713 * Memory Management & DMA Sync
714 */
715
716/*
717 * shmem buffers that are mapped cached are not coherent.
718 *
719 * We keep track of dirty pages using page faulting to perform cache management.
720 * When a page is mapped to the CPU in read/write mode the device can't access
721 * it and omap_obj->dma_addrs[i] is NULL. When a page is mapped to the device
722 * the omap_obj->dma_addrs[i] is set to the DMA address, and the page is
723 * unmapped from the CPU.
724 */
725static inline bool is_cached_coherent(struct drm_gem_object *obj)
726{
727 struct omap_gem_object *omap_obj = to_omap_bo(obj);
728
729 return !((omap_obj->flags & OMAP_BO_MEM_SHMEM) &&
730 ((omap_obj->flags & OMAP_BO_CACHE_MASK) == OMAP_BO_CACHED));
731}
732
733/* Sync the buffer for CPU access.. note pages should already be
734 * attached, ie. omap_gem_get_pages()
735 */
736void omap_gem_cpu_sync_page(struct drm_gem_object *obj, int pgoff)
737{
738 struct drm_device *dev = obj->dev;
739 struct omap_gem_object *omap_obj = to_omap_bo(obj);
740
741 if (is_cached_coherent(obj))
742 return;
743
744 if (omap_obj->dma_addrs[pgoff]) {
745 dma_unmap_page(dev->dev, omap_obj->dma_addrs[pgoff],
746 PAGE_SIZE, DMA_TO_DEVICE);
747 omap_obj->dma_addrs[pgoff] = 0;
748 }
749}
750
751/* sync the buffer for DMA access */
752void omap_gem_dma_sync_buffer(struct drm_gem_object *obj,
753 enum dma_data_direction dir)
754{
755 struct drm_device *dev = obj->dev;
756 struct omap_gem_object *omap_obj = to_omap_bo(obj);
757 int i, npages = obj->size >> PAGE_SHIFT;
758 struct page **pages = omap_obj->pages;
759 bool dirty = false;
760
761 if (is_cached_coherent(obj))
762 return;
763
764 for (i = 0; i < npages; i++) {
765 if (!omap_obj->dma_addrs[i]) {
766 dma_addr_t addr;
767
768 addr = dma_map_page(dev->dev, pages[i], 0,
769 PAGE_SIZE, dir);
770 if (dma_mapping_error(dev->dev, addr)) {
771 dev_warn(dev->dev, "%s: failed to map page\n",
772 __func__);
773 break;
774 }
775
776 dirty = true;
777 omap_obj->dma_addrs[i] = addr;
778 }
779 }
780
781 if (dirty) {
782 unmap_mapping_range(obj->filp->f_mapping, 0,
783 omap_gem_mmap_size(obj), 1);
784 }
785}
786
787/**
788 * omap_gem_pin() - Pin a GEM object in memory
789 * @obj: the GEM object
790 * @dma_addr: the DMA address
791 *
792 * Pin the given GEM object in memory and fill the dma_addr pointer with the
793 * object's DMA address. If the buffer is not physically contiguous it will be
794 * remapped through the TILER to provide a contiguous view.
795 *
796 * Pins are reference-counted, calling this function multiple times is allowed
797 * as long the corresponding omap_gem_unpin() calls are balanced.
798 *
799 * Return 0 on success or a negative error code otherwise.
800 */
801int omap_gem_pin(struct drm_gem_object *obj, dma_addr_t *dma_addr)
802{
803 struct omap_drm_private *priv = obj->dev->dev_private;
804 struct omap_gem_object *omap_obj = to_omap_bo(obj);
805 int ret = 0;
806
807 mutex_lock(&obj->dev->struct_mutex);
808
809 if (!is_contiguous(omap_obj) && priv->has_dmm) {
810 if (omap_obj->dma_addr_cnt == 0) {
811 struct page **pages;
812 u32 npages = obj->size >> PAGE_SHIFT;
813 enum tiler_fmt fmt = gem2fmt(omap_obj->flags);
814 struct tiler_block *block;
815
816 BUG_ON(omap_obj->block);
817
818 ret = get_pages(obj, &pages);
819 if (ret)
820 goto fail;
821
822 if (omap_obj->flags & OMAP_BO_TILED) {
823 block = tiler_reserve_2d(fmt,
824 omap_obj->width,
825 omap_obj->height, 0);
826 } else {
827 block = tiler_reserve_1d(obj->size);
828 }
829
830 if (IS_ERR(block)) {
831 ret = PTR_ERR(block);
832 dev_err(obj->dev->dev,
833 "could not remap: %d (%d)\n", ret, fmt);
834 goto fail;
835 }
836
837 /* TODO: enable async refill.. */
838 ret = tiler_pin(block, pages, npages,
839 omap_obj->roll, true);
840 if (ret) {
841 tiler_release(block);
842 dev_err(obj->dev->dev,
843 "could not pin: %d\n", ret);
844 goto fail;
845 }
846
847 omap_obj->dma_addr = tiler_ssptr(block);
848 omap_obj->block = block;
849
850 DBG("got dma address: %pad", &omap_obj->dma_addr);
851 }
852
853 omap_obj->dma_addr_cnt++;
854
855 *dma_addr = omap_obj->dma_addr;
856 } else if (is_contiguous(omap_obj)) {
857 *dma_addr = omap_obj->dma_addr;
858 } else {
859 ret = -EINVAL;
860 goto fail;
861 }
862
863fail:
864 mutex_unlock(&obj->dev->struct_mutex);
865
866 return ret;
867}
868
869/**
870 * omap_gem_unpin() - Unpin a GEM object from memory
871 * @obj: the GEM object
872 *
873 * Unpin the given GEM object previously pinned with omap_gem_pin(). Pins are
874 * reference-counted, the actualy unpin will only be performed when the number
875 * of calls to this function matches the number of calls to omap_gem_pin().
876 */
877void omap_gem_unpin(struct drm_gem_object *obj)
878{
879 struct omap_gem_object *omap_obj = to_omap_bo(obj);
880 int ret;
881
882 mutex_lock(&obj->dev->struct_mutex);
883 if (omap_obj->dma_addr_cnt > 0) {
884 omap_obj->dma_addr_cnt--;
885 if (omap_obj->dma_addr_cnt == 0) {
886 ret = tiler_unpin(omap_obj->block);
887 if (ret) {
888 dev_err(obj->dev->dev,
889 "could not unpin pages: %d\n", ret);
890 }
891 ret = tiler_release(omap_obj->block);
892 if (ret) {
893 dev_err(obj->dev->dev,
894 "could not release unmap: %d\n", ret);
895 }
896 omap_obj->dma_addr = 0;
897 omap_obj->block = NULL;
898 }
899 }
900
901 mutex_unlock(&obj->dev->struct_mutex);
902}
903
904/* Get rotated scanout address (only valid if already pinned), at the
905 * specified orientation and x,y offset from top-left corner of buffer
906 * (only valid for tiled 2d buffers)
907 */
908int omap_gem_rotated_dma_addr(struct drm_gem_object *obj, u32 orient,
909 int x, int y, dma_addr_t *dma_addr)
910{
911 struct omap_gem_object *omap_obj = to_omap_bo(obj);
912 int ret = -EINVAL;
913
914 mutex_lock(&obj->dev->struct_mutex);
915 if ((omap_obj->dma_addr_cnt > 0) && omap_obj->block &&
916 (omap_obj->flags & OMAP_BO_TILED)) {
917 *dma_addr = tiler_tsptr(omap_obj->block, orient, x, y);
918 ret = 0;
919 }
920 mutex_unlock(&obj->dev->struct_mutex);
921 return ret;
922}
923
924/* Get tiler stride for the buffer (only valid for 2d tiled buffers) */
925int omap_gem_tiled_stride(struct drm_gem_object *obj, u32 orient)
926{
927 struct omap_gem_object *omap_obj = to_omap_bo(obj);
928 int ret = -EINVAL;
929 if (omap_obj->flags & OMAP_BO_TILED)
930 ret = tiler_stride(gem2fmt(omap_obj->flags), orient);
931 return ret;
932}
933
934/* if !remap, and we don't have pages backing, then fail, rather than
935 * increasing the pin count (which we don't really do yet anyways,
936 * because we don't support swapping pages back out). And 'remap'
937 * might not be quite the right name, but I wanted to keep it working
938 * similarly to omap_gem_pin(). Note though that mutex is not
939 * aquired if !remap (because this can be called in atomic ctxt),
940 * but probably omap_gem_unpin() should be changed to work in the
941 * same way. If !remap, a matching omap_gem_put_pages() call is not
942 * required (and should not be made).
943 */
944int omap_gem_get_pages(struct drm_gem_object *obj, struct page ***pages,
945 bool remap)
946{
947 int ret;
948 if (!remap) {
949 struct omap_gem_object *omap_obj = to_omap_bo(obj);
950 if (!omap_obj->pages)
951 return -ENOMEM;
952 *pages = omap_obj->pages;
953 return 0;
954 }
955 mutex_lock(&obj->dev->struct_mutex);
956 ret = get_pages(obj, pages);
957 mutex_unlock(&obj->dev->struct_mutex);
958 return ret;
959}
960
961/* release pages when DMA no longer being performed */
962int omap_gem_put_pages(struct drm_gem_object *obj)
963{
964 /* do something here if we dynamically attach/detach pages.. at
965 * least they would no longer need to be pinned if everyone has
966 * released the pages..
967 */
968 return 0;
969}
970
971#ifdef CONFIG_DRM_FBDEV_EMULATION
972/* Get kernel virtual address for CPU access.. this more or less only
973 * exists for omap_fbdev. This should be called with struct_mutex
974 * held.
975 */
976void *omap_gem_vaddr(struct drm_gem_object *obj)
977{
978 struct omap_gem_object *omap_obj = to_omap_bo(obj);
979 WARN_ON(!mutex_is_locked(&obj->dev->struct_mutex));
980 if (!omap_obj->vaddr) {
981 struct page **pages;
982 int ret = get_pages(obj, &pages);
983 if (ret)
984 return ERR_PTR(ret);
985 omap_obj->vaddr = vmap(pages, obj->size >> PAGE_SHIFT,
986 VM_MAP, pgprot_writecombine(PAGE_KERNEL));
987 }
988 return omap_obj->vaddr;
989}
990#endif
991
992/* -----------------------------------------------------------------------------
993 * Power Management
994 */
995
996#ifdef CONFIG_PM
997/* re-pin objects in DMM in resume path: */
998int omap_gem_resume(struct drm_device *dev)
999{
1000 struct omap_drm_private *priv = dev->dev_private;
1001 struct omap_gem_object *omap_obj;
1002 int ret = 0;
1003
1004 list_for_each_entry(omap_obj, &priv->obj_list, mm_list) {
1005 if (omap_obj->block) {
1006 struct drm_gem_object *obj = &omap_obj->base;
1007 u32 npages = obj->size >> PAGE_SHIFT;
1008
1009 WARN_ON(!omap_obj->pages); /* this can't happen */
1010 ret = tiler_pin(omap_obj->block,
1011 omap_obj->pages, npages,
1012 omap_obj->roll, true);
1013 if (ret) {
1014 dev_err(dev->dev, "could not repin: %d\n", ret);
1015 return ret;
1016 }
1017 }
1018 }
1019
1020 return 0;
1021}
1022#endif
1023
1024/* -----------------------------------------------------------------------------
1025 * DebugFS
1026 */
1027
1028#ifdef CONFIG_DEBUG_FS
1029void omap_gem_describe(struct drm_gem_object *obj, struct seq_file *m)
1030{
1031 struct omap_gem_object *omap_obj = to_omap_bo(obj);
1032 u64 off;
1033
1034 off = drm_vma_node_start(&obj->vma_node);
1035
1036 seq_printf(m, "%08x: %2d (%2d) %08llx %pad (%2d) %p %4d",
1037 omap_obj->flags, obj->name, kref_read(&obj->refcount),
1038 off, &omap_obj->dma_addr, omap_obj->dma_addr_cnt,
1039 omap_obj->vaddr, omap_obj->roll);
1040
1041 if (omap_obj->flags & OMAP_BO_TILED) {
1042 seq_printf(m, " %dx%d", omap_obj->width, omap_obj->height);
1043 if (omap_obj->block) {
1044 struct tcm_area *area = &omap_obj->block->area;
1045 seq_printf(m, " (%dx%d, %dx%d)",
1046 area->p0.x, area->p0.y,
1047 area->p1.x, area->p1.y);
1048 }
1049 } else {
1050 seq_printf(m, " %zu", obj->size);
1051 }
1052
1053 seq_printf(m, "\n");
1054}
1055
1056void omap_gem_describe_objects(struct list_head *list, struct seq_file *m)
1057{
1058 struct omap_gem_object *omap_obj;
1059 int count = 0;
1060 size_t size = 0;
1061
1062 list_for_each_entry(omap_obj, list, mm_list) {
1063 struct drm_gem_object *obj = &omap_obj->base;
1064 seq_printf(m, " ");
1065 omap_gem_describe(obj, m);
1066 count++;
1067 size += obj->size;
1068 }
1069
1070 seq_printf(m, "Total %d objects, %zu bytes\n", count, size);
1071}
1072#endif
1073
1074/* -----------------------------------------------------------------------------
1075 * Constructor & Destructor
1076 */
1077
1078void omap_gem_free_object(struct drm_gem_object *obj)
1079{
1080 struct drm_device *dev = obj->dev;
1081 struct omap_drm_private *priv = dev->dev_private;
1082 struct omap_gem_object *omap_obj = to_omap_bo(obj);
1083
1084 evict(obj);
1085
1086 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
1087
1088 spin_lock(&priv->list_lock);
1089 list_del(&omap_obj->mm_list);
1090 spin_unlock(&priv->list_lock);
1091
1092 /* this means the object is still pinned.. which really should
1093 * not happen. I think..
1094 */
1095 WARN_ON(omap_obj->dma_addr_cnt > 0);
1096
1097 if (omap_obj->pages) {
1098 if (omap_obj->flags & OMAP_BO_MEM_DMABUF)
1099 kfree(omap_obj->pages);
1100 else
1101 omap_gem_detach_pages(obj);
1102 }
1103
1104 if (omap_obj->flags & OMAP_BO_MEM_DMA_API) {
1105 dma_free_wc(dev->dev, obj->size, omap_obj->vaddr,
1106 omap_obj->dma_addr);
1107 } else if (omap_obj->vaddr) {
1108 vunmap(omap_obj->vaddr);
1109 } else if (obj->import_attach) {
1110 drm_prime_gem_destroy(obj, omap_obj->sgt);
1111 }
1112
1113 drm_gem_object_release(obj);
1114
1115 kfree(omap_obj);
1116}
1117
1118/* GEM buffer object constructor */
1119struct drm_gem_object *omap_gem_new(struct drm_device *dev,
1120 union omap_gem_size gsize, u32 flags)
1121{
1122 struct omap_drm_private *priv = dev->dev_private;
1123 struct omap_gem_object *omap_obj;
1124 struct drm_gem_object *obj;
1125 struct address_space *mapping;
1126 size_t size;
1127 int ret;
1128
1129 /* Validate the flags and compute the memory and cache flags. */
1130 if (flags & OMAP_BO_TILED) {
1131 if (!priv->usergart) {
1132 dev_err(dev->dev, "Tiled buffers require DMM\n");
1133 return NULL;
1134 }
1135
1136 /*
1137 * Tiled buffers are always shmem paged backed. When they are
1138 * scanned out, they are remapped into DMM/TILER.
1139 */
1140 flags &= ~OMAP_BO_SCANOUT;
1141 flags |= OMAP_BO_MEM_SHMEM;
1142
1143 /*
1144 * Currently don't allow cached buffers. There is some caching
1145 * stuff that needs to be handled better.
1146 */
1147 flags &= ~(OMAP_BO_CACHED|OMAP_BO_WC|OMAP_BO_UNCACHED);
1148 flags |= tiler_get_cpu_cache_flags();
1149 } else if ((flags & OMAP_BO_SCANOUT) && !priv->has_dmm) {
1150 /*
1151 * OMAP_BO_SCANOUT hints that the buffer doesn't need to be
1152 * tiled. However, to lower the pressure on memory allocation,
1153 * use contiguous memory only if no TILER is available.
1154 */
1155 flags |= OMAP_BO_MEM_DMA_API;
1156 } else if (!(flags & OMAP_BO_MEM_DMABUF)) {
1157 /*
1158 * All other buffers not backed by dma_buf are shmem-backed.
1159 */
1160 flags |= OMAP_BO_MEM_SHMEM;
1161 }
1162
1163 /* Allocate the initialize the OMAP GEM object. */
1164 omap_obj = kzalloc(sizeof(*omap_obj), GFP_KERNEL);
1165 if (!omap_obj)
1166 return NULL;
1167
1168 obj = &omap_obj->base;
1169 omap_obj->flags = flags;
1170
1171 if (flags & OMAP_BO_TILED) {
1172 /*
1173 * For tiled buffers align dimensions to slot boundaries and
1174 * calculate size based on aligned dimensions.
1175 */
1176 tiler_align(gem2fmt(flags), &gsize.tiled.width,
1177 &gsize.tiled.height);
1178
1179 size = tiler_size(gem2fmt(flags), gsize.tiled.width,
1180 gsize.tiled.height);
1181
1182 omap_obj->width = gsize.tiled.width;
1183 omap_obj->height = gsize.tiled.height;
1184 } else {
1185 size = PAGE_ALIGN(gsize.bytes);
1186 }
1187
1188 /* Initialize the GEM object. */
1189 if (!(flags & OMAP_BO_MEM_SHMEM)) {
1190 drm_gem_private_object_init(dev, obj, size);
1191 } else {
1192 ret = drm_gem_object_init(dev, obj, size);
1193 if (ret)
1194 goto err_free;
1195
1196 mapping = obj->filp->f_mapping;
1197 mapping_set_gfp_mask(mapping, GFP_USER | __GFP_DMA32);
1198 }
1199
1200 /* Allocate memory if needed. */
1201 if (flags & OMAP_BO_MEM_DMA_API) {
1202 omap_obj->vaddr = dma_alloc_wc(dev->dev, size,
1203 &omap_obj->dma_addr,
1204 GFP_KERNEL);
1205 if (!omap_obj->vaddr)
1206 goto err_release;
1207 }
1208
1209 spin_lock(&priv->list_lock);
1210 list_add(&omap_obj->mm_list, &priv->obj_list);
1211 spin_unlock(&priv->list_lock);
1212
1213 return obj;
1214
1215err_release:
1216 drm_gem_object_release(obj);
1217err_free:
1218 kfree(omap_obj);
1219 return NULL;
1220}
1221
1222struct drm_gem_object *omap_gem_new_dmabuf(struct drm_device *dev, size_t size,
1223 struct sg_table *sgt)
1224{
1225 struct omap_drm_private *priv = dev->dev_private;
1226 struct omap_gem_object *omap_obj;
1227 struct drm_gem_object *obj;
1228 union omap_gem_size gsize;
1229
1230 /* Without a DMM only physically contiguous buffers can be supported. */
1231 if (sgt->orig_nents != 1 && !priv->has_dmm)
1232 return ERR_PTR(-EINVAL);
1233
1234 mutex_lock(&dev->struct_mutex);
1235
1236 gsize.bytes = PAGE_ALIGN(size);
1237 obj = omap_gem_new(dev, gsize, OMAP_BO_MEM_DMABUF | OMAP_BO_WC);
1238 if (!obj) {
1239 obj = ERR_PTR(-ENOMEM);
1240 goto done;
1241 }
1242
1243 omap_obj = to_omap_bo(obj);
1244 omap_obj->sgt = sgt;
1245
1246 if (sgt->orig_nents == 1) {
1247 omap_obj->dma_addr = sg_dma_address(sgt->sgl);
1248 } else {
1249 /* Create pages list from sgt */
1250 struct sg_page_iter iter;
1251 struct page **pages;
1252 unsigned int npages;
1253 unsigned int i = 0;
1254
1255 npages = DIV_ROUND_UP(size, PAGE_SIZE);
1256 pages = kcalloc(npages, sizeof(*pages), GFP_KERNEL);
1257 if (!pages) {
1258 omap_gem_free_object(obj);
1259 obj = ERR_PTR(-ENOMEM);
1260 goto done;
1261 }
1262
1263 omap_obj->pages = pages;
1264
1265 for_each_sg_page(sgt->sgl, &iter, sgt->orig_nents, 0) {
1266 pages[i++] = sg_page_iter_page(&iter);
1267 if (i > npages)
1268 break;
1269 }
1270
1271 if (WARN_ON(i != npages)) {
1272 omap_gem_free_object(obj);
1273 obj = ERR_PTR(-ENOMEM);
1274 goto done;
1275 }
1276 }
1277
1278done:
1279 mutex_unlock(&dev->struct_mutex);
1280 return obj;
1281}
1282
1283/* convenience method to construct a GEM buffer object, and userspace handle */
1284int omap_gem_new_handle(struct drm_device *dev, struct drm_file *file,
1285 union omap_gem_size gsize, u32 flags, u32 *handle)
1286{
1287 struct drm_gem_object *obj;
1288 int ret;
1289
1290 obj = omap_gem_new(dev, gsize, flags);
1291 if (!obj)
1292 return -ENOMEM;
1293
1294 ret = drm_gem_handle_create(file, obj, handle);
1295 if (ret) {
1296 omap_gem_free_object(obj);
1297 return ret;
1298 }
1299
1300 /* drop reference from allocate - handle holds it now */
1301 drm_gem_object_unreference_unlocked(obj);
1302
1303 return 0;
1304}
1305
1306/* -----------------------------------------------------------------------------
1307 * Init & Cleanup
1308 */
1309
1310/* If DMM is used, we need to set some stuff up.. */
1311void omap_gem_init(struct drm_device *dev)
1312{
1313 struct omap_drm_private *priv = dev->dev_private;
1314 struct omap_drm_usergart *usergart;
1315 const enum tiler_fmt fmts[] = {
1316 TILFMT_8BIT, TILFMT_16BIT, TILFMT_32BIT
1317 };
1318 int i, j;
1319
1320 if (!dmm_is_available()) {
1321 /* DMM only supported on OMAP4 and later, so this isn't fatal */
1322 dev_warn(dev->dev, "DMM not available, disable DMM support\n");
1323 return;
1324 }
1325
1326 usergart = kcalloc(3, sizeof(*usergart), GFP_KERNEL);
1327 if (!usergart)
1328 return;
1329
1330 /* reserve 4k aligned/wide regions for userspace mappings: */
1331 for (i = 0; i < ARRAY_SIZE(fmts); i++) {
1332 u16 h = 1, w = PAGE_SIZE >> i;
1333
1334 tiler_align(fmts[i], &w, &h);
1335 /* note: since each region is 1 4kb page wide, and minimum
1336 * number of rows, the height ends up being the same as the
1337 * # of pages in the region
1338 */
1339 usergart[i].height = h;
1340 usergart[i].height_shift = ilog2(h);
1341 usergart[i].stride_pfn = tiler_stride(fmts[i], 0) >> PAGE_SHIFT;
1342 usergart[i].slot_shift = ilog2((PAGE_SIZE / h) >> i);
1343 for (j = 0; j < NUM_USERGART_ENTRIES; j++) {
1344 struct omap_drm_usergart_entry *entry;
1345 struct tiler_block *block;
1346
1347 entry = &usergart[i].entry[j];
1348 block = tiler_reserve_2d(fmts[i], w, h, PAGE_SIZE);
1349 if (IS_ERR(block)) {
1350 dev_err(dev->dev,
1351 "reserve failed: %d, %d, %ld\n",
1352 i, j, PTR_ERR(block));
1353 return;
1354 }
1355 entry->dma_addr = tiler_ssptr(block);
1356 entry->block = block;
1357
1358 DBG("%d:%d: %dx%d: dma_addr=%pad stride=%d", i, j, w, h,
1359 &entry->dma_addr,
1360 usergart[i].stride_pfn << PAGE_SHIFT);
1361 }
1362 }
1363
1364 priv->usergart = usergart;
1365 priv->has_dmm = true;
1366}
1367
1368void omap_gem_deinit(struct drm_device *dev)
1369{
1370 struct omap_drm_private *priv = dev->dev_private;
1371
1372 /* I believe we can rely on there being no more outstanding GEM
1373 * objects which could depend on usergart/dmm at this point.
1374 */
1375 kfree(priv->usergart);
1376}
1/*
2 * drivers/gpu/drm/omapdrm/omap_gem.c
3 *
4 * Copyright (C) 2011 Texas Instruments
5 * Author: Rob Clark <rob.clark@linaro.org>
6 *
7 * This program is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 as published by
9 * the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * more details.
15 *
16 * You should have received a copy of the GNU General Public License along with
17 * this program. If not, see <http://www.gnu.org/licenses/>.
18 */
19
20
21#include <linux/spinlock.h>
22#include <linux/shmem_fs.h>
23#include <drm/drm_vma_manager.h>
24
25#include "omap_drv.h"
26#include "omap_dmm_tiler.h"
27
28/* remove these once drm core helpers are merged */
29struct page **_drm_gem_get_pages(struct drm_gem_object *obj, gfp_t gfpmask);
30void _drm_gem_put_pages(struct drm_gem_object *obj, struct page **pages,
31 bool dirty, bool accessed);
32int _drm_gem_create_mmap_offset_size(struct drm_gem_object *obj, size_t size);
33
34/*
35 * GEM buffer object implementation.
36 */
37
38#define to_omap_bo(x) container_of(x, struct omap_gem_object, base)
39
40/* note: we use upper 8 bits of flags for driver-internal flags: */
41#define OMAP_BO_DMA 0x01000000 /* actually is physically contiguous */
42#define OMAP_BO_EXT_SYNC 0x02000000 /* externally allocated sync object */
43#define OMAP_BO_EXT_MEM 0x04000000 /* externally allocated memory */
44
45
46struct omap_gem_object {
47 struct drm_gem_object base;
48
49 struct list_head mm_list;
50
51 uint32_t flags;
52
53 /** width/height for tiled formats (rounded up to slot boundaries) */
54 uint16_t width, height;
55
56 /** roll applied when mapping to DMM */
57 uint32_t roll;
58
59 /**
60 * If buffer is allocated physically contiguous, the OMAP_BO_DMA flag
61 * is set and the paddr is valid. Also if the buffer is remapped in
62 * TILER and paddr_cnt > 0, then paddr is valid. But if you are using
63 * the physical address and OMAP_BO_DMA is not set, then you should
64 * be going thru omap_gem_{get,put}_paddr() to ensure the mapping is
65 * not removed from under your feet.
66 *
67 * Note that OMAP_BO_SCANOUT is a hint from userspace that DMA capable
68 * buffer is requested, but doesn't mean that it is. Use the
69 * OMAP_BO_DMA flag to determine if the buffer has a DMA capable
70 * physical address.
71 */
72 dma_addr_t paddr;
73
74 /**
75 * # of users of paddr
76 */
77 uint32_t paddr_cnt;
78
79 /**
80 * tiler block used when buffer is remapped in DMM/TILER.
81 */
82 struct tiler_block *block;
83
84 /**
85 * Array of backing pages, if allocated. Note that pages are never
86 * allocated for buffers originally allocated from contiguous memory
87 */
88 struct page **pages;
89
90 /** addresses corresponding to pages in above array */
91 dma_addr_t *addrs;
92
93 /**
94 * Virtual address, if mapped.
95 */
96 void *vaddr;
97
98 /**
99 * sync-object allocated on demand (if needed)
100 *
101 * Per-buffer sync-object for tracking pending and completed hw/dma
102 * read and write operations. The layout in memory is dictated by
103 * the SGX firmware, which uses this information to stall the command
104 * stream if a surface is not ready yet.
105 *
106 * Note that when buffer is used by SGX, the sync-object needs to be
107 * allocated from a special heap of sync-objects. This way many sync
108 * objects can be packed in a page, and not waste GPU virtual address
109 * space. Because of this we have to have a omap_gem_set_sync_object()
110 * API to allow replacement of the syncobj after it has (potentially)
111 * already been allocated. A bit ugly but I haven't thought of a
112 * better alternative.
113 */
114 struct {
115 uint32_t write_pending;
116 uint32_t write_complete;
117 uint32_t read_pending;
118 uint32_t read_complete;
119 } *sync;
120};
121
122static int get_pages(struct drm_gem_object *obj, struct page ***pages);
123static uint64_t mmap_offset(struct drm_gem_object *obj);
124
125/* To deal with userspace mmap'ings of 2d tiled buffers, which (a) are
126 * not necessarily pinned in TILER all the time, and (b) when they are
127 * they are not necessarily page aligned, we reserve one or more small
128 * regions in each of the 2d containers to use as a user-GART where we
129 * can create a second page-aligned mapping of parts of the buffer
130 * being accessed from userspace.
131 *
132 * Note that we could optimize slightly when we know that multiple
133 * tiler containers are backed by the same PAT.. but I'll leave that
134 * for later..
135 */
136#define NUM_USERGART_ENTRIES 2
137struct usergart_entry {
138 struct tiler_block *block; /* the reserved tiler block */
139 dma_addr_t paddr;
140 struct drm_gem_object *obj; /* the current pinned obj */
141 pgoff_t obj_pgoff; /* page offset of obj currently
142 mapped in */
143};
144static struct {
145 struct usergart_entry entry[NUM_USERGART_ENTRIES];
146 int height; /* height in rows */
147 int height_shift; /* ilog2(height in rows) */
148 int slot_shift; /* ilog2(width per slot) */
149 int stride_pfn; /* stride in pages */
150 int last; /* index of last used entry */
151} *usergart;
152
153static void evict_entry(struct drm_gem_object *obj,
154 enum tiler_fmt fmt, struct usergart_entry *entry)
155{
156 struct omap_gem_object *omap_obj = to_omap_bo(obj);
157 int n = usergart[fmt].height;
158 size_t size = PAGE_SIZE * n;
159 loff_t off = mmap_offset(obj) +
160 (entry->obj_pgoff << PAGE_SHIFT);
161 const int m = 1 + ((omap_obj->width << fmt) / PAGE_SIZE);
162
163 if (m > 1) {
164 int i;
165 /* if stride > than PAGE_SIZE then sparse mapping: */
166 for (i = n; i > 0; i--) {
167 unmap_mapping_range(obj->dev->anon_inode->i_mapping,
168 off, PAGE_SIZE, 1);
169 off += PAGE_SIZE * m;
170 }
171 } else {
172 unmap_mapping_range(obj->dev->anon_inode->i_mapping,
173 off, size, 1);
174 }
175
176 entry->obj = NULL;
177}
178
179/* Evict a buffer from usergart, if it is mapped there */
180static void evict(struct drm_gem_object *obj)
181{
182 struct omap_gem_object *omap_obj = to_omap_bo(obj);
183
184 if (omap_obj->flags & OMAP_BO_TILED) {
185 enum tiler_fmt fmt = gem2fmt(omap_obj->flags);
186 int i;
187
188 if (!usergart)
189 return;
190
191 for (i = 0; i < NUM_USERGART_ENTRIES; i++) {
192 struct usergart_entry *entry = &usergart[fmt].entry[i];
193 if (entry->obj == obj)
194 evict_entry(obj, fmt, entry);
195 }
196 }
197}
198
199/* GEM objects can either be allocated from contiguous memory (in which
200 * case obj->filp==NULL), or w/ shmem backing (obj->filp!=NULL). But non
201 * contiguous buffers can be remapped in TILER/DMM if they need to be
202 * contiguous... but we don't do this all the time to reduce pressure
203 * on TILER/DMM space when we know at allocation time that the buffer
204 * will need to be scanned out.
205 */
206static inline bool is_shmem(struct drm_gem_object *obj)
207{
208 return obj->filp != NULL;
209}
210
211/**
212 * shmem buffers that are mapped cached can simulate coherency via using
213 * page faulting to keep track of dirty pages
214 */
215static inline bool is_cached_coherent(struct drm_gem_object *obj)
216{
217 struct omap_gem_object *omap_obj = to_omap_bo(obj);
218 return is_shmem(obj) &&
219 ((omap_obj->flags & OMAP_BO_CACHE_MASK) == OMAP_BO_CACHED);
220}
221
222static DEFINE_SPINLOCK(sync_lock);
223
224/** ensure backing pages are allocated */
225static int omap_gem_attach_pages(struct drm_gem_object *obj)
226{
227 struct drm_device *dev = obj->dev;
228 struct omap_gem_object *omap_obj = to_omap_bo(obj);
229 struct page **pages;
230 int npages = obj->size >> PAGE_SHIFT;
231 int i, ret;
232 dma_addr_t *addrs;
233
234 WARN_ON(omap_obj->pages);
235
236 /* TODO: __GFP_DMA32 .. but somehow GFP_HIGHMEM is coming from the
237 * mapping_gfp_mask(mapping) which conflicts w/ GFP_DMA32.. probably
238 * we actually want CMA memory for it all anyways..
239 */
240 pages = drm_gem_get_pages(obj, GFP_KERNEL);
241 if (IS_ERR(pages)) {
242 dev_err(obj->dev->dev, "could not get pages: %ld\n", PTR_ERR(pages));
243 return PTR_ERR(pages);
244 }
245
246 /* for non-cached buffers, ensure the new pages are clean because
247 * DSS, GPU, etc. are not cache coherent:
248 */
249 if (omap_obj->flags & (OMAP_BO_WC|OMAP_BO_UNCACHED)) {
250 addrs = kmalloc(npages * sizeof(*addrs), GFP_KERNEL);
251 if (!addrs) {
252 ret = -ENOMEM;
253 goto free_pages;
254 }
255
256 for (i = 0; i < npages; i++) {
257 addrs[i] = dma_map_page(dev->dev, pages[i],
258 0, PAGE_SIZE, DMA_BIDIRECTIONAL);
259 }
260 } else {
261 addrs = kzalloc(npages * sizeof(*addrs), GFP_KERNEL);
262 if (!addrs) {
263 ret = -ENOMEM;
264 goto free_pages;
265 }
266 }
267
268 omap_obj->addrs = addrs;
269 omap_obj->pages = pages;
270
271 return 0;
272
273free_pages:
274 drm_gem_put_pages(obj, pages, true, false);
275
276 return ret;
277}
278
279/** release backing pages */
280static void omap_gem_detach_pages(struct drm_gem_object *obj)
281{
282 struct omap_gem_object *omap_obj = to_omap_bo(obj);
283
284 /* for non-cached buffers, ensure the new pages are clean because
285 * DSS, GPU, etc. are not cache coherent:
286 */
287 if (omap_obj->flags & (OMAP_BO_WC|OMAP_BO_UNCACHED)) {
288 int i, npages = obj->size >> PAGE_SHIFT;
289 for (i = 0; i < npages; i++) {
290 dma_unmap_page(obj->dev->dev, omap_obj->addrs[i],
291 PAGE_SIZE, DMA_BIDIRECTIONAL);
292 }
293 }
294
295 kfree(omap_obj->addrs);
296 omap_obj->addrs = NULL;
297
298 drm_gem_put_pages(obj, omap_obj->pages, true, false);
299 omap_obj->pages = NULL;
300}
301
302/* get buffer flags */
303uint32_t omap_gem_flags(struct drm_gem_object *obj)
304{
305 return to_omap_bo(obj)->flags;
306}
307
308/** get mmap offset */
309static uint64_t mmap_offset(struct drm_gem_object *obj)
310{
311 struct drm_device *dev = obj->dev;
312 int ret;
313 size_t size;
314
315 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
316
317 /* Make it mmapable */
318 size = omap_gem_mmap_size(obj);
319 ret = drm_gem_create_mmap_offset_size(obj, size);
320 if (ret) {
321 dev_err(dev->dev, "could not allocate mmap offset\n");
322 return 0;
323 }
324
325 return drm_vma_node_offset_addr(&obj->vma_node);
326}
327
328uint64_t omap_gem_mmap_offset(struct drm_gem_object *obj)
329{
330 uint64_t offset;
331 mutex_lock(&obj->dev->struct_mutex);
332 offset = mmap_offset(obj);
333 mutex_unlock(&obj->dev->struct_mutex);
334 return offset;
335}
336
337/** get mmap size */
338size_t omap_gem_mmap_size(struct drm_gem_object *obj)
339{
340 struct omap_gem_object *omap_obj = to_omap_bo(obj);
341 size_t size = obj->size;
342
343 if (omap_obj->flags & OMAP_BO_TILED) {
344 /* for tiled buffers, the virtual size has stride rounded up
345 * to 4kb.. (to hide the fact that row n+1 might start 16kb or
346 * 32kb later!). But we don't back the entire buffer with
347 * pages, only the valid picture part.. so need to adjust for
348 * this in the size used to mmap and generate mmap offset
349 */
350 size = tiler_vsize(gem2fmt(omap_obj->flags),
351 omap_obj->width, omap_obj->height);
352 }
353
354 return size;
355}
356
357/* get tiled size, returns -EINVAL if not tiled buffer */
358int omap_gem_tiled_size(struct drm_gem_object *obj, uint16_t *w, uint16_t *h)
359{
360 struct omap_gem_object *omap_obj = to_omap_bo(obj);
361 if (omap_obj->flags & OMAP_BO_TILED) {
362 *w = omap_obj->width;
363 *h = omap_obj->height;
364 return 0;
365 }
366 return -EINVAL;
367}
368
369/* Normal handling for the case of faulting in non-tiled buffers */
370static int fault_1d(struct drm_gem_object *obj,
371 struct vm_area_struct *vma, struct vm_fault *vmf)
372{
373 struct omap_gem_object *omap_obj = to_omap_bo(obj);
374 unsigned long pfn;
375 pgoff_t pgoff;
376
377 /* We don't use vmf->pgoff since that has the fake offset: */
378 pgoff = ((unsigned long)vmf->virtual_address -
379 vma->vm_start) >> PAGE_SHIFT;
380
381 if (omap_obj->pages) {
382 omap_gem_cpu_sync(obj, pgoff);
383 pfn = page_to_pfn(omap_obj->pages[pgoff]);
384 } else {
385 BUG_ON(!(omap_obj->flags & OMAP_BO_DMA));
386 pfn = (omap_obj->paddr >> PAGE_SHIFT) + pgoff;
387 }
388
389 VERB("Inserting %p pfn %lx, pa %lx", vmf->virtual_address,
390 pfn, pfn << PAGE_SHIFT);
391
392 return vm_insert_mixed(vma, (unsigned long)vmf->virtual_address, pfn);
393}
394
395/* Special handling for the case of faulting in 2d tiled buffers */
396static int fault_2d(struct drm_gem_object *obj,
397 struct vm_area_struct *vma, struct vm_fault *vmf)
398{
399 struct omap_gem_object *omap_obj = to_omap_bo(obj);
400 struct usergart_entry *entry;
401 enum tiler_fmt fmt = gem2fmt(omap_obj->flags);
402 struct page *pages[64]; /* XXX is this too much to have on stack? */
403 unsigned long pfn;
404 pgoff_t pgoff, base_pgoff;
405 void __user *vaddr;
406 int i, ret, slots;
407
408 /*
409 * Note the height of the slot is also equal to the number of pages
410 * that need to be mapped in to fill 4kb wide CPU page. If the slot
411 * height is 64, then 64 pages fill a 4kb wide by 64 row region.
412 */
413 const int n = usergart[fmt].height;
414 const int n_shift = usergart[fmt].height_shift;
415
416 /*
417 * If buffer width in bytes > PAGE_SIZE then the virtual stride is
418 * rounded up to next multiple of PAGE_SIZE.. this need to be taken
419 * into account in some of the math, so figure out virtual stride
420 * in pages
421 */
422 const int m = 1 + ((omap_obj->width << fmt) / PAGE_SIZE);
423
424 /* We don't use vmf->pgoff since that has the fake offset: */
425 pgoff = ((unsigned long)vmf->virtual_address -
426 vma->vm_start) >> PAGE_SHIFT;
427
428 /*
429 * Actual address we start mapping at is rounded down to previous slot
430 * boundary in the y direction:
431 */
432 base_pgoff = round_down(pgoff, m << n_shift);
433
434 /* figure out buffer width in slots */
435 slots = omap_obj->width >> usergart[fmt].slot_shift;
436
437 vaddr = vmf->virtual_address - ((pgoff - base_pgoff) << PAGE_SHIFT);
438
439 entry = &usergart[fmt].entry[usergart[fmt].last];
440
441 /* evict previous buffer using this usergart entry, if any: */
442 if (entry->obj)
443 evict_entry(entry->obj, fmt, entry);
444
445 entry->obj = obj;
446 entry->obj_pgoff = base_pgoff;
447
448 /* now convert base_pgoff to phys offset from virt offset: */
449 base_pgoff = (base_pgoff >> n_shift) * slots;
450
451 /* for wider-than 4k.. figure out which part of the slot-row we want: */
452 if (m > 1) {
453 int off = pgoff % m;
454 entry->obj_pgoff += off;
455 base_pgoff /= m;
456 slots = min(slots - (off << n_shift), n);
457 base_pgoff += off << n_shift;
458 vaddr += off << PAGE_SHIFT;
459 }
460
461 /*
462 * Map in pages. Beyond the valid pixel part of the buffer, we set
463 * pages[i] to NULL to get a dummy page mapped in.. if someone
464 * reads/writes it they will get random/undefined content, but at
465 * least it won't be corrupting whatever other random page used to
466 * be mapped in, or other undefined behavior.
467 */
468 memcpy(pages, &omap_obj->pages[base_pgoff],
469 sizeof(struct page *) * slots);
470 memset(pages + slots, 0,
471 sizeof(struct page *) * (n - slots));
472
473 ret = tiler_pin(entry->block, pages, ARRAY_SIZE(pages), 0, true);
474 if (ret) {
475 dev_err(obj->dev->dev, "failed to pin: %d\n", ret);
476 return ret;
477 }
478
479 pfn = entry->paddr >> PAGE_SHIFT;
480
481 VERB("Inserting %p pfn %lx, pa %lx", vmf->virtual_address,
482 pfn, pfn << PAGE_SHIFT);
483
484 for (i = n; i > 0; i--) {
485 vm_insert_mixed(vma, (unsigned long)vaddr, pfn);
486 pfn += usergart[fmt].stride_pfn;
487 vaddr += PAGE_SIZE * m;
488 }
489
490 /* simple round-robin: */
491 usergart[fmt].last = (usergart[fmt].last + 1) % NUM_USERGART_ENTRIES;
492
493 return 0;
494}
495
496/**
497 * omap_gem_fault - pagefault handler for GEM objects
498 * @vma: the VMA of the GEM object
499 * @vmf: fault detail
500 *
501 * Invoked when a fault occurs on an mmap of a GEM managed area. GEM
502 * does most of the work for us including the actual map/unmap calls
503 * but we need to do the actual page work.
504 *
505 * The VMA was set up by GEM. In doing so it also ensured that the
506 * vma->vm_private_data points to the GEM object that is backing this
507 * mapping.
508 */
509int omap_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
510{
511 struct drm_gem_object *obj = vma->vm_private_data;
512 struct omap_gem_object *omap_obj = to_omap_bo(obj);
513 struct drm_device *dev = obj->dev;
514 struct page **pages;
515 int ret;
516
517 /* Make sure we don't parallel update on a fault, nor move or remove
518 * something from beneath our feet
519 */
520 mutex_lock(&dev->struct_mutex);
521
522 /* if a shmem backed object, make sure we have pages attached now */
523 ret = get_pages(obj, &pages);
524 if (ret)
525 goto fail;
526
527 /* where should we do corresponding put_pages().. we are mapping
528 * the original page, rather than thru a GART, so we can't rely
529 * on eviction to trigger this. But munmap() or all mappings should
530 * probably trigger put_pages()?
531 */
532
533 if (omap_obj->flags & OMAP_BO_TILED)
534 ret = fault_2d(obj, vma, vmf);
535 else
536 ret = fault_1d(obj, vma, vmf);
537
538
539fail:
540 mutex_unlock(&dev->struct_mutex);
541 switch (ret) {
542 case 0:
543 case -ERESTARTSYS:
544 case -EINTR:
545 return VM_FAULT_NOPAGE;
546 case -ENOMEM:
547 return VM_FAULT_OOM;
548 default:
549 return VM_FAULT_SIGBUS;
550 }
551}
552
553/** We override mainly to fix up some of the vm mapping flags.. */
554int omap_gem_mmap(struct file *filp, struct vm_area_struct *vma)
555{
556 int ret;
557
558 ret = drm_gem_mmap(filp, vma);
559 if (ret) {
560 DBG("mmap failed: %d", ret);
561 return ret;
562 }
563
564 return omap_gem_mmap_obj(vma->vm_private_data, vma);
565}
566
567int omap_gem_mmap_obj(struct drm_gem_object *obj,
568 struct vm_area_struct *vma)
569{
570 struct omap_gem_object *omap_obj = to_omap_bo(obj);
571
572 vma->vm_flags &= ~VM_PFNMAP;
573 vma->vm_flags |= VM_MIXEDMAP;
574
575 if (omap_obj->flags & OMAP_BO_WC) {
576 vma->vm_page_prot = pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
577 } else if (omap_obj->flags & OMAP_BO_UNCACHED) {
578 vma->vm_page_prot = pgprot_noncached(vm_get_page_prot(vma->vm_flags));
579 } else {
580 /*
581 * We do have some private objects, at least for scanout buffers
582 * on hardware without DMM/TILER. But these are allocated write-
583 * combine
584 */
585 if (WARN_ON(!obj->filp))
586 return -EINVAL;
587
588 /*
589 * Shunt off cached objs to shmem file so they have their own
590 * address_space (so unmap_mapping_range does what we want,
591 * in particular in the case of mmap'd dmabufs)
592 */
593 fput(vma->vm_file);
594 vma->vm_pgoff = 0;
595 vma->vm_file = get_file(obj->filp);
596
597 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
598 }
599
600 return 0;
601}
602
603
604/**
605 * omap_gem_dumb_create - create a dumb buffer
606 * @drm_file: our client file
607 * @dev: our device
608 * @args: the requested arguments copied from userspace
609 *
610 * Allocate a buffer suitable for use for a frame buffer of the
611 * form described by user space. Give userspace a handle by which
612 * to reference it.
613 */
614int omap_gem_dumb_create(struct drm_file *file, struct drm_device *dev,
615 struct drm_mode_create_dumb *args)
616{
617 union omap_gem_size gsize;
618
619 /* in case someone tries to feed us a completely bogus stride: */
620 args->pitch = align_pitch(args->pitch, args->width, args->bpp);
621 args->size = PAGE_ALIGN(args->pitch * args->height);
622
623 gsize = (union omap_gem_size){
624 .bytes = args->size,
625 };
626
627 return omap_gem_new_handle(dev, file, gsize,
628 OMAP_BO_SCANOUT | OMAP_BO_WC, &args->handle);
629}
630
631/**
632 * omap_gem_dumb_map - buffer mapping for dumb interface
633 * @file: our drm client file
634 * @dev: drm device
635 * @handle: GEM handle to the object (from dumb_create)
636 *
637 * Do the necessary setup to allow the mapping of the frame buffer
638 * into user memory. We don't have to do much here at the moment.
639 */
640int omap_gem_dumb_map_offset(struct drm_file *file, struct drm_device *dev,
641 uint32_t handle, uint64_t *offset)
642{
643 struct drm_gem_object *obj;
644 int ret = 0;
645
646 /* GEM does all our handle to object mapping */
647 obj = drm_gem_object_lookup(dev, file, handle);
648 if (obj == NULL) {
649 ret = -ENOENT;
650 goto fail;
651 }
652
653 *offset = omap_gem_mmap_offset(obj);
654
655 drm_gem_object_unreference_unlocked(obj);
656
657fail:
658 return ret;
659}
660
661/* Set scrolling position. This allows us to implement fast scrolling
662 * for console.
663 *
664 * Call only from non-atomic contexts.
665 */
666int omap_gem_roll(struct drm_gem_object *obj, uint32_t roll)
667{
668 struct omap_gem_object *omap_obj = to_omap_bo(obj);
669 uint32_t npages = obj->size >> PAGE_SHIFT;
670 int ret = 0;
671
672 if (roll > npages) {
673 dev_err(obj->dev->dev, "invalid roll: %d\n", roll);
674 return -EINVAL;
675 }
676
677 omap_obj->roll = roll;
678
679 mutex_lock(&obj->dev->struct_mutex);
680
681 /* if we aren't mapped yet, we don't need to do anything */
682 if (omap_obj->block) {
683 struct page **pages;
684 ret = get_pages(obj, &pages);
685 if (ret)
686 goto fail;
687 ret = tiler_pin(omap_obj->block, pages, npages, roll, true);
688 if (ret)
689 dev_err(obj->dev->dev, "could not repin: %d\n", ret);
690 }
691
692fail:
693 mutex_unlock(&obj->dev->struct_mutex);
694
695 return ret;
696}
697
698/* Sync the buffer for CPU access.. note pages should already be
699 * attached, ie. omap_gem_get_pages()
700 */
701void omap_gem_cpu_sync(struct drm_gem_object *obj, int pgoff)
702{
703 struct drm_device *dev = obj->dev;
704 struct omap_gem_object *omap_obj = to_omap_bo(obj);
705
706 if (is_cached_coherent(obj) && omap_obj->addrs[pgoff]) {
707 dma_unmap_page(dev->dev, omap_obj->addrs[pgoff],
708 PAGE_SIZE, DMA_BIDIRECTIONAL);
709 omap_obj->addrs[pgoff] = 0;
710 }
711}
712
713/* sync the buffer for DMA access */
714void omap_gem_dma_sync(struct drm_gem_object *obj,
715 enum dma_data_direction dir)
716{
717 struct drm_device *dev = obj->dev;
718 struct omap_gem_object *omap_obj = to_omap_bo(obj);
719
720 if (is_cached_coherent(obj)) {
721 int i, npages = obj->size >> PAGE_SHIFT;
722 struct page **pages = omap_obj->pages;
723 bool dirty = false;
724
725 for (i = 0; i < npages; i++) {
726 if (!omap_obj->addrs[i]) {
727 omap_obj->addrs[i] = dma_map_page(dev->dev, pages[i], 0,
728 PAGE_SIZE, DMA_BIDIRECTIONAL);
729 dirty = true;
730 }
731 }
732
733 if (dirty) {
734 unmap_mapping_range(obj->filp->f_mapping, 0,
735 omap_gem_mmap_size(obj), 1);
736 }
737 }
738}
739
740/* Get physical address for DMA.. if 'remap' is true, and the buffer is not
741 * already contiguous, remap it to pin in physically contiguous memory.. (ie.
742 * map in TILER)
743 */
744int omap_gem_get_paddr(struct drm_gem_object *obj,
745 dma_addr_t *paddr, bool remap)
746{
747 struct omap_drm_private *priv = obj->dev->dev_private;
748 struct omap_gem_object *omap_obj = to_omap_bo(obj);
749 int ret = 0;
750
751 mutex_lock(&obj->dev->struct_mutex);
752
753 if (remap && is_shmem(obj) && priv->has_dmm) {
754 if (omap_obj->paddr_cnt == 0) {
755 struct page **pages;
756 uint32_t npages = obj->size >> PAGE_SHIFT;
757 enum tiler_fmt fmt = gem2fmt(omap_obj->flags);
758 struct tiler_block *block;
759
760 BUG_ON(omap_obj->block);
761
762 ret = get_pages(obj, &pages);
763 if (ret)
764 goto fail;
765
766 if (omap_obj->flags & OMAP_BO_TILED) {
767 block = tiler_reserve_2d(fmt,
768 omap_obj->width,
769 omap_obj->height, 0);
770 } else {
771 block = tiler_reserve_1d(obj->size);
772 }
773
774 if (IS_ERR(block)) {
775 ret = PTR_ERR(block);
776 dev_err(obj->dev->dev,
777 "could not remap: %d (%d)\n", ret, fmt);
778 goto fail;
779 }
780
781 /* TODO: enable async refill.. */
782 ret = tiler_pin(block, pages, npages,
783 omap_obj->roll, true);
784 if (ret) {
785 tiler_release(block);
786 dev_err(obj->dev->dev,
787 "could not pin: %d\n", ret);
788 goto fail;
789 }
790
791 omap_obj->paddr = tiler_ssptr(block);
792 omap_obj->block = block;
793
794 DBG("got paddr: %08x", omap_obj->paddr);
795 }
796
797 omap_obj->paddr_cnt++;
798
799 *paddr = omap_obj->paddr;
800 } else if (omap_obj->flags & OMAP_BO_DMA) {
801 *paddr = omap_obj->paddr;
802 } else {
803 ret = -EINVAL;
804 goto fail;
805 }
806
807fail:
808 mutex_unlock(&obj->dev->struct_mutex);
809
810 return ret;
811}
812
813/* Release physical address, when DMA is no longer being performed.. this
814 * could potentially unpin and unmap buffers from TILER
815 */
816int omap_gem_put_paddr(struct drm_gem_object *obj)
817{
818 struct omap_gem_object *omap_obj = to_omap_bo(obj);
819 int ret = 0;
820
821 mutex_lock(&obj->dev->struct_mutex);
822 if (omap_obj->paddr_cnt > 0) {
823 omap_obj->paddr_cnt--;
824 if (omap_obj->paddr_cnt == 0) {
825 ret = tiler_unpin(omap_obj->block);
826 if (ret) {
827 dev_err(obj->dev->dev,
828 "could not unpin pages: %d\n", ret);
829 goto fail;
830 }
831 ret = tiler_release(omap_obj->block);
832 if (ret) {
833 dev_err(obj->dev->dev,
834 "could not release unmap: %d\n", ret);
835 }
836 omap_obj->block = NULL;
837 }
838 }
839fail:
840 mutex_unlock(&obj->dev->struct_mutex);
841 return ret;
842}
843
844/* Get rotated scanout address (only valid if already pinned), at the
845 * specified orientation and x,y offset from top-left corner of buffer
846 * (only valid for tiled 2d buffers)
847 */
848int omap_gem_rotated_paddr(struct drm_gem_object *obj, uint32_t orient,
849 int x, int y, dma_addr_t *paddr)
850{
851 struct omap_gem_object *omap_obj = to_omap_bo(obj);
852 int ret = -EINVAL;
853
854 mutex_lock(&obj->dev->struct_mutex);
855 if ((omap_obj->paddr_cnt > 0) && omap_obj->block &&
856 (omap_obj->flags & OMAP_BO_TILED)) {
857 *paddr = tiler_tsptr(omap_obj->block, orient, x, y);
858 ret = 0;
859 }
860 mutex_unlock(&obj->dev->struct_mutex);
861 return ret;
862}
863
864/* Get tiler stride for the buffer (only valid for 2d tiled buffers) */
865int omap_gem_tiled_stride(struct drm_gem_object *obj, uint32_t orient)
866{
867 struct omap_gem_object *omap_obj = to_omap_bo(obj);
868 int ret = -EINVAL;
869 if (omap_obj->flags & OMAP_BO_TILED)
870 ret = tiler_stride(gem2fmt(omap_obj->flags), orient);
871 return ret;
872}
873
874/* acquire pages when needed (for example, for DMA where physically
875 * contiguous buffer is not required
876 */
877static int get_pages(struct drm_gem_object *obj, struct page ***pages)
878{
879 struct omap_gem_object *omap_obj = to_omap_bo(obj);
880 int ret = 0;
881
882 if (is_shmem(obj) && !omap_obj->pages) {
883 ret = omap_gem_attach_pages(obj);
884 if (ret) {
885 dev_err(obj->dev->dev, "could not attach pages\n");
886 return ret;
887 }
888 }
889
890 /* TODO: even phys-contig.. we should have a list of pages? */
891 *pages = omap_obj->pages;
892
893 return 0;
894}
895
896/* if !remap, and we don't have pages backing, then fail, rather than
897 * increasing the pin count (which we don't really do yet anyways,
898 * because we don't support swapping pages back out). And 'remap'
899 * might not be quite the right name, but I wanted to keep it working
900 * similarly to omap_gem_get_paddr(). Note though that mutex is not
901 * aquired if !remap (because this can be called in atomic ctxt),
902 * but probably omap_gem_get_paddr() should be changed to work in the
903 * same way. If !remap, a matching omap_gem_put_pages() call is not
904 * required (and should not be made).
905 */
906int omap_gem_get_pages(struct drm_gem_object *obj, struct page ***pages,
907 bool remap)
908{
909 int ret;
910 if (!remap) {
911 struct omap_gem_object *omap_obj = to_omap_bo(obj);
912 if (!omap_obj->pages)
913 return -ENOMEM;
914 *pages = omap_obj->pages;
915 return 0;
916 }
917 mutex_lock(&obj->dev->struct_mutex);
918 ret = get_pages(obj, pages);
919 mutex_unlock(&obj->dev->struct_mutex);
920 return ret;
921}
922
923/* release pages when DMA no longer being performed */
924int omap_gem_put_pages(struct drm_gem_object *obj)
925{
926 /* do something here if we dynamically attach/detach pages.. at
927 * least they would no longer need to be pinned if everyone has
928 * released the pages..
929 */
930 return 0;
931}
932
933/* Get kernel virtual address for CPU access.. this more or less only
934 * exists for omap_fbdev. This should be called with struct_mutex
935 * held.
936 */
937void *omap_gem_vaddr(struct drm_gem_object *obj)
938{
939 struct omap_gem_object *omap_obj = to_omap_bo(obj);
940 WARN_ON(!mutex_is_locked(&obj->dev->struct_mutex));
941 if (!omap_obj->vaddr) {
942 struct page **pages;
943 int ret = get_pages(obj, &pages);
944 if (ret)
945 return ERR_PTR(ret);
946 omap_obj->vaddr = vmap(pages, obj->size >> PAGE_SHIFT,
947 VM_MAP, pgprot_writecombine(PAGE_KERNEL));
948 }
949 return omap_obj->vaddr;
950}
951
952#ifdef CONFIG_PM
953/* re-pin objects in DMM in resume path: */
954int omap_gem_resume(struct device *dev)
955{
956 struct drm_device *drm_dev = dev_get_drvdata(dev);
957 struct omap_drm_private *priv = drm_dev->dev_private;
958 struct omap_gem_object *omap_obj;
959 int ret = 0;
960
961 list_for_each_entry(omap_obj, &priv->obj_list, mm_list) {
962 if (omap_obj->block) {
963 struct drm_gem_object *obj = &omap_obj->base;
964 uint32_t npages = obj->size >> PAGE_SHIFT;
965 WARN_ON(!omap_obj->pages); /* this can't happen */
966 ret = tiler_pin(omap_obj->block,
967 omap_obj->pages, npages,
968 omap_obj->roll, true);
969 if (ret) {
970 dev_err(dev, "could not repin: %d\n", ret);
971 return ret;
972 }
973 }
974 }
975
976 return 0;
977}
978#endif
979
980#ifdef CONFIG_DEBUG_FS
981void omap_gem_describe(struct drm_gem_object *obj, struct seq_file *m)
982{
983 struct omap_gem_object *omap_obj = to_omap_bo(obj);
984 uint64_t off;
985
986 off = drm_vma_node_start(&obj->vma_node);
987
988 seq_printf(m, "%08x: %2d (%2d) %08llx %08Zx (%2d) %p %4d",
989 omap_obj->flags, obj->name, obj->refcount.refcount.counter,
990 off, omap_obj->paddr, omap_obj->paddr_cnt,
991 omap_obj->vaddr, omap_obj->roll);
992
993 if (omap_obj->flags & OMAP_BO_TILED) {
994 seq_printf(m, " %dx%d", omap_obj->width, omap_obj->height);
995 if (omap_obj->block) {
996 struct tcm_area *area = &omap_obj->block->area;
997 seq_printf(m, " (%dx%d, %dx%d)",
998 area->p0.x, area->p0.y,
999 area->p1.x, area->p1.y);
1000 }
1001 } else {
1002 seq_printf(m, " %d", obj->size);
1003 }
1004
1005 seq_printf(m, "\n");
1006}
1007
1008void omap_gem_describe_objects(struct list_head *list, struct seq_file *m)
1009{
1010 struct omap_gem_object *omap_obj;
1011 int count = 0;
1012 size_t size = 0;
1013
1014 list_for_each_entry(omap_obj, list, mm_list) {
1015 struct drm_gem_object *obj = &omap_obj->base;
1016 seq_printf(m, " ");
1017 omap_gem_describe(obj, m);
1018 count++;
1019 size += obj->size;
1020 }
1021
1022 seq_printf(m, "Total %d objects, %zu bytes\n", count, size);
1023}
1024#endif
1025
1026/* Buffer Synchronization:
1027 */
1028
1029struct omap_gem_sync_waiter {
1030 struct list_head list;
1031 struct omap_gem_object *omap_obj;
1032 enum omap_gem_op op;
1033 uint32_t read_target, write_target;
1034 /* notify called w/ sync_lock held */
1035 void (*notify)(void *arg);
1036 void *arg;
1037};
1038
1039/* list of omap_gem_sync_waiter.. the notify fxn gets called back when
1040 * the read and/or write target count is achieved which can call a user
1041 * callback (ex. to kick 3d and/or 2d), wakeup blocked task (prep for
1042 * cpu access), etc.
1043 */
1044static LIST_HEAD(waiters);
1045
1046static inline bool is_waiting(struct omap_gem_sync_waiter *waiter)
1047{
1048 struct omap_gem_object *omap_obj = waiter->omap_obj;
1049 if ((waiter->op & OMAP_GEM_READ) &&
1050 (omap_obj->sync->write_complete < waiter->write_target))
1051 return true;
1052 if ((waiter->op & OMAP_GEM_WRITE) &&
1053 (omap_obj->sync->read_complete < waiter->read_target))
1054 return true;
1055 return false;
1056}
1057
1058/* macro for sync debug.. */
1059#define SYNCDBG 0
1060#define SYNC(fmt, ...) do { if (SYNCDBG) \
1061 printk(KERN_ERR "%s:%d: "fmt"\n", \
1062 __func__, __LINE__, ##__VA_ARGS__); \
1063 } while (0)
1064
1065
1066static void sync_op_update(void)
1067{
1068 struct omap_gem_sync_waiter *waiter, *n;
1069 list_for_each_entry_safe(waiter, n, &waiters, list) {
1070 if (!is_waiting(waiter)) {
1071 list_del(&waiter->list);
1072 SYNC("notify: %p", waiter);
1073 waiter->notify(waiter->arg);
1074 kfree(waiter);
1075 }
1076 }
1077}
1078
1079static inline int sync_op(struct drm_gem_object *obj,
1080 enum omap_gem_op op, bool start)
1081{
1082 struct omap_gem_object *omap_obj = to_omap_bo(obj);
1083 int ret = 0;
1084
1085 spin_lock(&sync_lock);
1086
1087 if (!omap_obj->sync) {
1088 omap_obj->sync = kzalloc(sizeof(*omap_obj->sync), GFP_ATOMIC);
1089 if (!omap_obj->sync) {
1090 ret = -ENOMEM;
1091 goto unlock;
1092 }
1093 }
1094
1095 if (start) {
1096 if (op & OMAP_GEM_READ)
1097 omap_obj->sync->read_pending++;
1098 if (op & OMAP_GEM_WRITE)
1099 omap_obj->sync->write_pending++;
1100 } else {
1101 if (op & OMAP_GEM_READ)
1102 omap_obj->sync->read_complete++;
1103 if (op & OMAP_GEM_WRITE)
1104 omap_obj->sync->write_complete++;
1105 sync_op_update();
1106 }
1107
1108unlock:
1109 spin_unlock(&sync_lock);
1110
1111 return ret;
1112}
1113
1114/* it is a bit lame to handle updates in this sort of polling way, but
1115 * in case of PVR, the GPU can directly update read/write complete
1116 * values, and not really tell us which ones it updated.. this also
1117 * means that sync_lock is not quite sufficient. So we'll need to
1118 * do something a bit better when it comes time to add support for
1119 * separate 2d hw..
1120 */
1121void omap_gem_op_update(void)
1122{
1123 spin_lock(&sync_lock);
1124 sync_op_update();
1125 spin_unlock(&sync_lock);
1126}
1127
1128/* mark the start of read and/or write operation */
1129int omap_gem_op_start(struct drm_gem_object *obj, enum omap_gem_op op)
1130{
1131 return sync_op(obj, op, true);
1132}
1133
1134int omap_gem_op_finish(struct drm_gem_object *obj, enum omap_gem_op op)
1135{
1136 return sync_op(obj, op, false);
1137}
1138
1139static DECLARE_WAIT_QUEUE_HEAD(sync_event);
1140
1141static void sync_notify(void *arg)
1142{
1143 struct task_struct **waiter_task = arg;
1144 *waiter_task = NULL;
1145 wake_up_all(&sync_event);
1146}
1147
1148int omap_gem_op_sync(struct drm_gem_object *obj, enum omap_gem_op op)
1149{
1150 struct omap_gem_object *omap_obj = to_omap_bo(obj);
1151 int ret = 0;
1152 if (omap_obj->sync) {
1153 struct task_struct *waiter_task = current;
1154 struct omap_gem_sync_waiter *waiter =
1155 kzalloc(sizeof(*waiter), GFP_KERNEL);
1156
1157 if (!waiter)
1158 return -ENOMEM;
1159
1160 waiter->omap_obj = omap_obj;
1161 waiter->op = op;
1162 waiter->read_target = omap_obj->sync->read_pending;
1163 waiter->write_target = omap_obj->sync->write_pending;
1164 waiter->notify = sync_notify;
1165 waiter->arg = &waiter_task;
1166
1167 spin_lock(&sync_lock);
1168 if (is_waiting(waiter)) {
1169 SYNC("waited: %p", waiter);
1170 list_add_tail(&waiter->list, &waiters);
1171 spin_unlock(&sync_lock);
1172 ret = wait_event_interruptible(sync_event,
1173 (waiter_task == NULL));
1174 spin_lock(&sync_lock);
1175 if (waiter_task) {
1176 SYNC("interrupted: %p", waiter);
1177 /* we were interrupted */
1178 list_del(&waiter->list);
1179 waiter_task = NULL;
1180 } else {
1181 /* freed in sync_op_update() */
1182 waiter = NULL;
1183 }
1184 }
1185 spin_unlock(&sync_lock);
1186
1187 if (waiter)
1188 kfree(waiter);
1189 }
1190 return ret;
1191}
1192
1193/* call fxn(arg), either synchronously or asynchronously if the op
1194 * is currently blocked.. fxn() can be called from any context
1195 *
1196 * (TODO for now fxn is called back from whichever context calls
1197 * omap_gem_op_update().. but this could be better defined later
1198 * if needed)
1199 *
1200 * TODO more code in common w/ _sync()..
1201 */
1202int omap_gem_op_async(struct drm_gem_object *obj, enum omap_gem_op op,
1203 void (*fxn)(void *arg), void *arg)
1204{
1205 struct omap_gem_object *omap_obj = to_omap_bo(obj);
1206 if (omap_obj->sync) {
1207 struct omap_gem_sync_waiter *waiter =
1208 kzalloc(sizeof(*waiter), GFP_ATOMIC);
1209
1210 if (!waiter)
1211 return -ENOMEM;
1212
1213 waiter->omap_obj = omap_obj;
1214 waiter->op = op;
1215 waiter->read_target = omap_obj->sync->read_pending;
1216 waiter->write_target = omap_obj->sync->write_pending;
1217 waiter->notify = fxn;
1218 waiter->arg = arg;
1219
1220 spin_lock(&sync_lock);
1221 if (is_waiting(waiter)) {
1222 SYNC("waited: %p", waiter);
1223 list_add_tail(&waiter->list, &waiters);
1224 spin_unlock(&sync_lock);
1225 return 0;
1226 }
1227
1228 spin_unlock(&sync_lock);
1229
1230 kfree(waiter);
1231 }
1232
1233 /* no waiting.. */
1234 fxn(arg);
1235
1236 return 0;
1237}
1238
1239/* special API so PVR can update the buffer to use a sync-object allocated
1240 * from it's sync-obj heap. Only used for a newly allocated (from PVR's
1241 * perspective) sync-object, so we overwrite the new syncobj w/ values
1242 * from the already allocated syncobj (if there is one)
1243 */
1244int omap_gem_set_sync_object(struct drm_gem_object *obj, void *syncobj)
1245{
1246 struct omap_gem_object *omap_obj = to_omap_bo(obj);
1247 int ret = 0;
1248
1249 spin_lock(&sync_lock);
1250
1251 if ((omap_obj->flags & OMAP_BO_EXT_SYNC) && !syncobj) {
1252 /* clearing a previously set syncobj */
1253 syncobj = kmemdup(omap_obj->sync, sizeof(*omap_obj->sync),
1254 GFP_ATOMIC);
1255 if (!syncobj) {
1256 ret = -ENOMEM;
1257 goto unlock;
1258 }
1259 omap_obj->flags &= ~OMAP_BO_EXT_SYNC;
1260 omap_obj->sync = syncobj;
1261 } else if (syncobj && !(omap_obj->flags & OMAP_BO_EXT_SYNC)) {
1262 /* replacing an existing syncobj */
1263 if (omap_obj->sync) {
1264 memcpy(syncobj, omap_obj->sync, sizeof(*omap_obj->sync));
1265 kfree(omap_obj->sync);
1266 }
1267 omap_obj->flags |= OMAP_BO_EXT_SYNC;
1268 omap_obj->sync = syncobj;
1269 }
1270
1271unlock:
1272 spin_unlock(&sync_lock);
1273 return ret;
1274}
1275
1276/* don't call directly.. called from GEM core when it is time to actually
1277 * free the object..
1278 */
1279void omap_gem_free_object(struct drm_gem_object *obj)
1280{
1281 struct drm_device *dev = obj->dev;
1282 struct omap_gem_object *omap_obj = to_omap_bo(obj);
1283
1284 evict(obj);
1285
1286 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
1287
1288 list_del(&omap_obj->mm_list);
1289
1290 drm_gem_free_mmap_offset(obj);
1291
1292 /* this means the object is still pinned.. which really should
1293 * not happen. I think..
1294 */
1295 WARN_ON(omap_obj->paddr_cnt > 0);
1296
1297 /* don't free externally allocated backing memory */
1298 if (!(omap_obj->flags & OMAP_BO_EXT_MEM)) {
1299 if (omap_obj->pages)
1300 omap_gem_detach_pages(obj);
1301
1302 if (!is_shmem(obj)) {
1303 dma_free_writecombine(dev->dev, obj->size,
1304 omap_obj->vaddr, omap_obj->paddr);
1305 } else if (omap_obj->vaddr) {
1306 vunmap(omap_obj->vaddr);
1307 }
1308 }
1309
1310 /* don't free externally allocated syncobj */
1311 if (!(omap_obj->flags & OMAP_BO_EXT_SYNC))
1312 kfree(omap_obj->sync);
1313
1314 drm_gem_object_release(obj);
1315
1316 kfree(obj);
1317}
1318
1319/* convenience method to construct a GEM buffer object, and userspace handle */
1320int omap_gem_new_handle(struct drm_device *dev, struct drm_file *file,
1321 union omap_gem_size gsize, uint32_t flags, uint32_t *handle)
1322{
1323 struct drm_gem_object *obj;
1324 int ret;
1325
1326 obj = omap_gem_new(dev, gsize, flags);
1327 if (!obj)
1328 return -ENOMEM;
1329
1330 ret = drm_gem_handle_create(file, obj, handle);
1331 if (ret) {
1332 drm_gem_object_release(obj);
1333 kfree(obj); /* TODO isn't there a dtor to call? just copying i915 */
1334 return ret;
1335 }
1336
1337 /* drop reference from allocate - handle holds it now */
1338 drm_gem_object_unreference_unlocked(obj);
1339
1340 return 0;
1341}
1342
1343/* GEM buffer object constructor */
1344struct drm_gem_object *omap_gem_new(struct drm_device *dev,
1345 union omap_gem_size gsize, uint32_t flags)
1346{
1347 struct omap_drm_private *priv = dev->dev_private;
1348 struct omap_gem_object *omap_obj;
1349 struct drm_gem_object *obj = NULL;
1350 size_t size;
1351 int ret;
1352
1353 if (flags & OMAP_BO_TILED) {
1354 if (!usergart) {
1355 dev_err(dev->dev, "Tiled buffers require DMM\n");
1356 goto fail;
1357 }
1358
1359 /* tiled buffers are always shmem paged backed.. when they are
1360 * scanned out, they are remapped into DMM/TILER
1361 */
1362 flags &= ~OMAP_BO_SCANOUT;
1363
1364 /* currently don't allow cached buffers.. there is some caching
1365 * stuff that needs to be handled better
1366 */
1367 flags &= ~(OMAP_BO_CACHED|OMAP_BO_UNCACHED);
1368 flags |= OMAP_BO_WC;
1369
1370 /* align dimensions to slot boundaries... */
1371 tiler_align(gem2fmt(flags),
1372 &gsize.tiled.width, &gsize.tiled.height);
1373
1374 /* ...and calculate size based on aligned dimensions */
1375 size = tiler_size(gem2fmt(flags),
1376 gsize.tiled.width, gsize.tiled.height);
1377 } else {
1378 size = PAGE_ALIGN(gsize.bytes);
1379 }
1380
1381 omap_obj = kzalloc(sizeof(*omap_obj), GFP_KERNEL);
1382 if (!omap_obj)
1383 goto fail;
1384
1385 list_add(&omap_obj->mm_list, &priv->obj_list);
1386
1387 obj = &omap_obj->base;
1388
1389 if ((flags & OMAP_BO_SCANOUT) && !priv->has_dmm) {
1390 /* attempt to allocate contiguous memory if we don't
1391 * have DMM for remappign discontiguous buffers
1392 */
1393 omap_obj->vaddr = dma_alloc_writecombine(dev->dev, size,
1394 &omap_obj->paddr, GFP_KERNEL);
1395 if (omap_obj->vaddr)
1396 flags |= OMAP_BO_DMA;
1397
1398 }
1399
1400 omap_obj->flags = flags;
1401
1402 if (flags & OMAP_BO_TILED) {
1403 omap_obj->width = gsize.tiled.width;
1404 omap_obj->height = gsize.tiled.height;
1405 }
1406
1407 ret = 0;
1408 if (flags & (OMAP_BO_DMA|OMAP_BO_EXT_MEM))
1409 drm_gem_private_object_init(dev, obj, size);
1410 else
1411 ret = drm_gem_object_init(dev, obj, size);
1412
1413 if (ret)
1414 goto fail;
1415
1416 return obj;
1417
1418fail:
1419 if (obj)
1420 omap_gem_free_object(obj);
1421
1422 return NULL;
1423}
1424
1425/* init/cleanup.. if DMM is used, we need to set some stuff up.. */
1426void omap_gem_init(struct drm_device *dev)
1427{
1428 struct omap_drm_private *priv = dev->dev_private;
1429 const enum tiler_fmt fmts[] = {
1430 TILFMT_8BIT, TILFMT_16BIT, TILFMT_32BIT
1431 };
1432 int i, j;
1433
1434 if (!dmm_is_available()) {
1435 /* DMM only supported on OMAP4 and later, so this isn't fatal */
1436 dev_warn(dev->dev, "DMM not available, disable DMM support\n");
1437 return;
1438 }
1439
1440 usergart = kcalloc(3, sizeof(*usergart), GFP_KERNEL);
1441 if (!usergart)
1442 return;
1443
1444 /* reserve 4k aligned/wide regions for userspace mappings: */
1445 for (i = 0; i < ARRAY_SIZE(fmts); i++) {
1446 uint16_t h = 1, w = PAGE_SIZE >> i;
1447 tiler_align(fmts[i], &w, &h);
1448 /* note: since each region is 1 4kb page wide, and minimum
1449 * number of rows, the height ends up being the same as the
1450 * # of pages in the region
1451 */
1452 usergart[i].height = h;
1453 usergart[i].height_shift = ilog2(h);
1454 usergart[i].stride_pfn = tiler_stride(fmts[i], 0) >> PAGE_SHIFT;
1455 usergart[i].slot_shift = ilog2((PAGE_SIZE / h) >> i);
1456 for (j = 0; j < NUM_USERGART_ENTRIES; j++) {
1457 struct usergart_entry *entry = &usergart[i].entry[j];
1458 struct tiler_block *block =
1459 tiler_reserve_2d(fmts[i], w, h,
1460 PAGE_SIZE);
1461 if (IS_ERR(block)) {
1462 dev_err(dev->dev,
1463 "reserve failed: %d, %d, %ld\n",
1464 i, j, PTR_ERR(block));
1465 return;
1466 }
1467 entry->paddr = tiler_ssptr(block);
1468 entry->block = block;
1469
1470 DBG("%d:%d: %dx%d: paddr=%08x stride=%d", i, j, w, h,
1471 entry->paddr,
1472 usergart[i].stride_pfn << PAGE_SHIFT);
1473 }
1474 }
1475
1476 priv->has_dmm = true;
1477}
1478
1479void omap_gem_deinit(struct drm_device *dev)
1480{
1481 /* I believe we can rely on there being no more outstanding GEM
1482 * objects which could depend on usergart/dmm at this point.
1483 */
1484 kfree(usergart);
1485}