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#include <linux/seq_file.h>
21#include <linux/shmem_fs.h>
22#include <linux/spinlock.h>
23#include <linux/pfn_t.h>
24
25#include <drm/drm_vma_manager.h>
26
27#include "omap_drv.h"
28#include "omap_dmm_tiler.h"
29
30/*
31 * GEM buffer object implementation.
32 */
33
34/* note: we use upper 8 bits of flags for driver-internal flags: */
35#define OMAP_BO_MEM_DMA_API 0x01000000 /* memory allocated with the dma_alloc_* API */
36#define OMAP_BO_MEM_SHMEM 0x02000000 /* memory allocated through shmem backing */
37#define OMAP_BO_MEM_DMABUF 0x08000000 /* memory imported from a dmabuf */
38
39struct omap_gem_object {
40 struct drm_gem_object base;
41
42 struct list_head mm_list;
43
44 uint32_t flags;
45
46 /** width/height for tiled formats (rounded up to slot boundaries) */
47 uint16_t width, height;
48
49 /** roll applied when mapping to DMM */
50 uint32_t roll;
51
52 /**
53 * paddr contains the buffer DMA address. It is valid for
54 *
55 * - buffers allocated through the DMA mapping API (with the
56 * OMAP_BO_MEM_DMA_API flag set)
57 *
58 * - buffers imported from dmabuf (with the OMAP_BO_MEM_DMABUF flag set)
59 * if they are physically contiguous (when sgt->orig_nents == 1)
60 *
61 * - buffers mapped through the TILER when paddr_cnt is not zero, in
62 * which case the DMA address points to the TILER aperture
63 *
64 * Physically contiguous buffers have their DMA address equal to the
65 * physical address as we don't remap those buffers through the TILER.
66 *
67 * Buffers mapped to the TILER have their DMA address pointing to the
68 * TILER aperture. As TILER mappings are refcounted (through paddr_cnt)
69 * the DMA address must be accessed through omap_get_get_paddr() to
70 * ensure that the mapping won't disappear unexpectedly. References must
71 * be released with omap_gem_put_paddr().
72 */
73 dma_addr_t paddr;
74
75 /**
76 * # of users of paddr
77 */
78 uint32_t paddr_cnt;
79
80 /**
81 * If the buffer has been imported from a dmabuf the OMAP_DB_DMABUF flag
82 * is set and the sgt field is valid.
83 */
84 struct sg_table *sgt;
85
86 /**
87 * tiler block used when buffer is remapped in DMM/TILER.
88 */
89 struct tiler_block *block;
90
91 /**
92 * Array of backing pages, if allocated. Note that pages are never
93 * allocated for buffers originally allocated from contiguous memory
94 */
95 struct page **pages;
96
97 /** addresses corresponding to pages in above array */
98 dma_addr_t *addrs;
99
100 /**
101 * Virtual address, if mapped.
102 */
103 void *vaddr;
104
105 /**
106 * sync-object allocated on demand (if needed)
107 *
108 * Per-buffer sync-object for tracking pending and completed hw/dma
109 * read and write operations.
110 */
111 struct {
112 uint32_t write_pending;
113 uint32_t write_complete;
114 uint32_t read_pending;
115 uint32_t read_complete;
116 } *sync;
117};
118
119#define to_omap_bo(x) container_of(x, struct omap_gem_object, base)
120
121/* To deal with userspace mmap'ings of 2d tiled buffers, which (a) are
122 * not necessarily pinned in TILER all the time, and (b) when they are
123 * they are not necessarily page aligned, we reserve one or more small
124 * regions in each of the 2d containers to use as a user-GART where we
125 * can create a second page-aligned mapping of parts of the buffer
126 * being accessed from userspace.
127 *
128 * Note that we could optimize slightly when we know that multiple
129 * tiler containers are backed by the same PAT.. but I'll leave that
130 * for later..
131 */
132#define NUM_USERGART_ENTRIES 2
133struct omap_drm_usergart_entry {
134 struct tiler_block *block; /* the reserved tiler block */
135 dma_addr_t paddr;
136 struct drm_gem_object *obj; /* the current pinned obj */
137 pgoff_t obj_pgoff; /* page offset of obj currently
138 mapped in */
139};
140
141struct omap_drm_usergart {
142 struct omap_drm_usergart_entry entry[NUM_USERGART_ENTRIES];
143 int height; /* height in rows */
144 int height_shift; /* ilog2(height in rows) */
145 int slot_shift; /* ilog2(width per slot) */
146 int stride_pfn; /* stride in pages */
147 int last; /* index of last used entry */
148};
149
150/* -----------------------------------------------------------------------------
151 * Helpers
152 */
153
154/** get mmap offset */
155static uint64_t mmap_offset(struct drm_gem_object *obj)
156{
157 struct drm_device *dev = obj->dev;
158 int ret;
159 size_t size;
160
161 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
162
163 /* Make it mmapable */
164 size = omap_gem_mmap_size(obj);
165 ret = drm_gem_create_mmap_offset_size(obj, size);
166 if (ret) {
167 dev_err(dev->dev, "could not allocate mmap offset\n");
168 return 0;
169 }
170
171 return drm_vma_node_offset_addr(&obj->vma_node);
172}
173
174static bool is_contiguous(struct omap_gem_object *omap_obj)
175{
176 if (omap_obj->flags & OMAP_BO_MEM_DMA_API)
177 return true;
178
179 if ((omap_obj->flags & OMAP_BO_MEM_DMABUF) && omap_obj->sgt->nents == 1)
180 return true;
181
182 return false;
183}
184
185/* -----------------------------------------------------------------------------
186 * Eviction
187 */
188
189static void evict_entry(struct drm_gem_object *obj,
190 enum tiler_fmt fmt, struct omap_drm_usergart_entry *entry)
191{
192 struct omap_gem_object *omap_obj = to_omap_bo(obj);
193 struct omap_drm_private *priv = obj->dev->dev_private;
194 int n = priv->usergart[fmt].height;
195 size_t size = PAGE_SIZE * n;
196 loff_t off = mmap_offset(obj) +
197 (entry->obj_pgoff << PAGE_SHIFT);
198 const int m = 1 + ((omap_obj->width << fmt) / PAGE_SIZE);
199
200 if (m > 1) {
201 int i;
202 /* if stride > than PAGE_SIZE then sparse mapping: */
203 for (i = n; i > 0; i--) {
204 unmap_mapping_range(obj->dev->anon_inode->i_mapping,
205 off, PAGE_SIZE, 1);
206 off += PAGE_SIZE * m;
207 }
208 } else {
209 unmap_mapping_range(obj->dev->anon_inode->i_mapping,
210 off, size, 1);
211 }
212
213 entry->obj = NULL;
214}
215
216/* Evict a buffer from usergart, if it is mapped there */
217static void evict(struct drm_gem_object *obj)
218{
219 struct omap_gem_object *omap_obj = to_omap_bo(obj);
220 struct omap_drm_private *priv = obj->dev->dev_private;
221
222 if (omap_obj->flags & OMAP_BO_TILED) {
223 enum tiler_fmt fmt = gem2fmt(omap_obj->flags);
224 int i;
225
226 for (i = 0; i < NUM_USERGART_ENTRIES; i++) {
227 struct omap_drm_usergart_entry *entry =
228 &priv->usergart[fmt].entry[i];
229
230 if (entry->obj == obj)
231 evict_entry(obj, fmt, entry);
232 }
233 }
234}
235
236/* -----------------------------------------------------------------------------
237 * Page Management
238 */
239
240/** ensure backing pages are allocated */
241static int omap_gem_attach_pages(struct drm_gem_object *obj)
242{
243 struct drm_device *dev = obj->dev;
244 struct omap_gem_object *omap_obj = to_omap_bo(obj);
245 struct page **pages;
246 int npages = obj->size >> PAGE_SHIFT;
247 int i, ret;
248 dma_addr_t *addrs;
249
250 WARN_ON(omap_obj->pages);
251
252 pages = drm_gem_get_pages(obj);
253 if (IS_ERR(pages)) {
254 dev_err(obj->dev->dev, "could not get pages: %ld\n", PTR_ERR(pages));
255 return PTR_ERR(pages);
256 }
257
258 /* for non-cached buffers, ensure the new pages are clean because
259 * DSS, GPU, etc. are not cache coherent:
260 */
261 if (omap_obj->flags & (OMAP_BO_WC|OMAP_BO_UNCACHED)) {
262 addrs = kmalloc(npages * sizeof(*addrs), GFP_KERNEL);
263 if (!addrs) {
264 ret = -ENOMEM;
265 goto free_pages;
266 }
267
268 for (i = 0; i < npages; i++) {
269 addrs[i] = dma_map_page(dev->dev, pages[i],
270 0, PAGE_SIZE, DMA_BIDIRECTIONAL);
271
272 if (dma_mapping_error(dev->dev, addrs[i])) {
273 dev_warn(dev->dev,
274 "%s: failed to map page\n", __func__);
275
276 for (i = i - 1; i >= 0; --i) {
277 dma_unmap_page(dev->dev, addrs[i],
278 PAGE_SIZE, DMA_BIDIRECTIONAL);
279 }
280
281 ret = -ENOMEM;
282 goto free_addrs;
283 }
284 }
285 } else {
286 addrs = kzalloc(npages * sizeof(*addrs), GFP_KERNEL);
287 if (!addrs) {
288 ret = -ENOMEM;
289 goto free_pages;
290 }
291 }
292
293 omap_obj->addrs = addrs;
294 omap_obj->pages = pages;
295
296 return 0;
297
298free_addrs:
299 kfree(addrs);
300free_pages:
301 drm_gem_put_pages(obj, pages, true, false);
302
303 return ret;
304}
305
306/* acquire pages when needed (for example, for DMA where physically
307 * contiguous buffer is not required
308 */
309static int get_pages(struct drm_gem_object *obj, struct page ***pages)
310{
311 struct omap_gem_object *omap_obj = to_omap_bo(obj);
312 int ret = 0;
313
314 if ((omap_obj->flags & OMAP_BO_MEM_SHMEM) && !omap_obj->pages) {
315 ret = omap_gem_attach_pages(obj);
316 if (ret) {
317 dev_err(obj->dev->dev, "could not attach pages\n");
318 return ret;
319 }
320 }
321
322 /* TODO: even phys-contig.. we should have a list of pages? */
323 *pages = omap_obj->pages;
324
325 return 0;
326}
327
328/** release backing pages */
329static void omap_gem_detach_pages(struct drm_gem_object *obj)
330{
331 struct omap_gem_object *omap_obj = to_omap_bo(obj);
332
333 /* for non-cached buffers, ensure the new pages are clean because
334 * DSS, GPU, etc. are not cache coherent:
335 */
336 if (omap_obj->flags & (OMAP_BO_WC|OMAP_BO_UNCACHED)) {
337 int i, npages = obj->size >> PAGE_SHIFT;
338 for (i = 0; i < npages; i++) {
339 if (omap_obj->addrs[i])
340 dma_unmap_page(obj->dev->dev,
341 omap_obj->addrs[i],
342 PAGE_SIZE, DMA_BIDIRECTIONAL);
343 }
344 }
345
346 kfree(omap_obj->addrs);
347 omap_obj->addrs = NULL;
348
349 drm_gem_put_pages(obj, omap_obj->pages, true, false);
350 omap_obj->pages = NULL;
351}
352
353/* get buffer flags */
354uint32_t omap_gem_flags(struct drm_gem_object *obj)
355{
356 return to_omap_bo(obj)->flags;
357}
358
359uint64_t omap_gem_mmap_offset(struct drm_gem_object *obj)
360{
361 uint64_t offset;
362 mutex_lock(&obj->dev->struct_mutex);
363 offset = mmap_offset(obj);
364 mutex_unlock(&obj->dev->struct_mutex);
365 return offset;
366}
367
368/** get mmap size */
369size_t omap_gem_mmap_size(struct drm_gem_object *obj)
370{
371 struct omap_gem_object *omap_obj = to_omap_bo(obj);
372 size_t size = obj->size;
373
374 if (omap_obj->flags & OMAP_BO_TILED) {
375 /* for tiled buffers, the virtual size has stride rounded up
376 * to 4kb.. (to hide the fact that row n+1 might start 16kb or
377 * 32kb later!). But we don't back the entire buffer with
378 * pages, only the valid picture part.. so need to adjust for
379 * this in the size used to mmap and generate mmap offset
380 */
381 size = tiler_vsize(gem2fmt(omap_obj->flags),
382 omap_obj->width, omap_obj->height);
383 }
384
385 return size;
386}
387
388/* -----------------------------------------------------------------------------
389 * Fault Handling
390 */
391
392/* Normal handling for the case of faulting in non-tiled buffers */
393static int fault_1d(struct drm_gem_object *obj,
394 struct vm_area_struct *vma, struct vm_fault *vmf)
395{
396 struct omap_gem_object *omap_obj = to_omap_bo(obj);
397 unsigned long pfn;
398 pgoff_t pgoff;
399
400 /* We don't use vmf->pgoff since that has the fake offset: */
401 pgoff = (vmf->address - vma->vm_start) >> PAGE_SHIFT;
402
403 if (omap_obj->pages) {
404 omap_gem_cpu_sync(obj, pgoff);
405 pfn = page_to_pfn(omap_obj->pages[pgoff]);
406 } else {
407 BUG_ON(!is_contiguous(omap_obj));
408 pfn = (omap_obj->paddr >> PAGE_SHIFT) + pgoff;
409 }
410
411 VERB("Inserting %p pfn %lx, pa %lx", (void *)vmf->address,
412 pfn, pfn << PAGE_SHIFT);
413
414 return vm_insert_mixed(vma, vmf->address, __pfn_to_pfn_t(pfn, PFN_DEV));
415}
416
417/* Special handling for the case of faulting in 2d tiled buffers */
418static int fault_2d(struct drm_gem_object *obj,
419 struct vm_area_struct *vma, struct vm_fault *vmf)
420{
421 struct omap_gem_object *omap_obj = to_omap_bo(obj);
422 struct omap_drm_private *priv = obj->dev->dev_private;
423 struct omap_drm_usergart_entry *entry;
424 enum tiler_fmt fmt = gem2fmt(omap_obj->flags);
425 struct page *pages[64]; /* XXX is this too much to have on stack? */
426 unsigned long pfn;
427 pgoff_t pgoff, base_pgoff;
428 unsigned long vaddr;
429 int i, ret, slots;
430
431 /*
432 * Note the height of the slot is also equal to the number of pages
433 * that need to be mapped in to fill 4kb wide CPU page. If the slot
434 * height is 64, then 64 pages fill a 4kb wide by 64 row region.
435 */
436 const int n = priv->usergart[fmt].height;
437 const int n_shift = priv->usergart[fmt].height_shift;
438
439 /*
440 * If buffer width in bytes > PAGE_SIZE then the virtual stride is
441 * rounded up to next multiple of PAGE_SIZE.. this need to be taken
442 * into account in some of the math, so figure out virtual stride
443 * in pages
444 */
445 const int m = 1 + ((omap_obj->width << fmt) / PAGE_SIZE);
446
447 /* We don't use vmf->pgoff since that has the fake offset: */
448 pgoff = (vmf->address - vma->vm_start) >> PAGE_SHIFT;
449
450 /*
451 * Actual address we start mapping at is rounded down to previous slot
452 * boundary in the y direction:
453 */
454 base_pgoff = round_down(pgoff, m << n_shift);
455
456 /* figure out buffer width in slots */
457 slots = omap_obj->width >> priv->usergart[fmt].slot_shift;
458
459 vaddr = vmf->address - ((pgoff - base_pgoff) << PAGE_SHIFT);
460
461 entry = &priv->usergart[fmt].entry[priv->usergart[fmt].last];
462
463 /* evict previous buffer using this usergart entry, if any: */
464 if (entry->obj)
465 evict_entry(entry->obj, fmt, entry);
466
467 entry->obj = obj;
468 entry->obj_pgoff = base_pgoff;
469
470 /* now convert base_pgoff to phys offset from virt offset: */
471 base_pgoff = (base_pgoff >> n_shift) * slots;
472
473 /* for wider-than 4k.. figure out which part of the slot-row we want: */
474 if (m > 1) {
475 int off = pgoff % m;
476 entry->obj_pgoff += off;
477 base_pgoff /= m;
478 slots = min(slots - (off << n_shift), n);
479 base_pgoff += off << n_shift;
480 vaddr += off << PAGE_SHIFT;
481 }
482
483 /*
484 * Map in pages. Beyond the valid pixel part of the buffer, we set
485 * pages[i] to NULL to get a dummy page mapped in.. if someone
486 * reads/writes it they will get random/undefined content, but at
487 * least it won't be corrupting whatever other random page used to
488 * be mapped in, or other undefined behavior.
489 */
490 memcpy(pages, &omap_obj->pages[base_pgoff],
491 sizeof(struct page *) * slots);
492 memset(pages + slots, 0,
493 sizeof(struct page *) * (n - slots));
494
495 ret = tiler_pin(entry->block, pages, ARRAY_SIZE(pages), 0, true);
496 if (ret) {
497 dev_err(obj->dev->dev, "failed to pin: %d\n", ret);
498 return ret;
499 }
500
501 pfn = entry->paddr >> PAGE_SHIFT;
502
503 VERB("Inserting %p pfn %lx, pa %lx", (void *)vmf->address,
504 pfn, pfn << PAGE_SHIFT);
505
506 for (i = n; i > 0; i--) {
507 vm_insert_mixed(vma, vaddr, __pfn_to_pfn_t(pfn, PFN_DEV));
508 pfn += priv->usergart[fmt].stride_pfn;
509 vaddr += PAGE_SIZE * m;
510 }
511
512 /* simple round-robin: */
513 priv->usergart[fmt].last = (priv->usergart[fmt].last + 1)
514 % NUM_USERGART_ENTRIES;
515
516 return 0;
517}
518
519/**
520 * omap_gem_fault - pagefault handler for GEM objects
521 * @vma: the VMA of the GEM object
522 * @vmf: fault detail
523 *
524 * Invoked when a fault occurs on an mmap of a GEM managed area. GEM
525 * does most of the work for us including the actual map/unmap calls
526 * but we need to do the actual page work.
527 *
528 * The VMA was set up by GEM. In doing so it also ensured that the
529 * vma->vm_private_data points to the GEM object that is backing this
530 * mapping.
531 */
532int omap_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
533{
534 struct drm_gem_object *obj = vma->vm_private_data;
535 struct omap_gem_object *omap_obj = to_omap_bo(obj);
536 struct drm_device *dev = obj->dev;
537 struct page **pages;
538 int ret;
539
540 /* Make sure we don't parallel update on a fault, nor move or remove
541 * something from beneath our feet
542 */
543 mutex_lock(&dev->struct_mutex);
544
545 /* if a shmem backed object, make sure we have pages attached now */
546 ret = get_pages(obj, &pages);
547 if (ret)
548 goto fail;
549
550 /* where should we do corresponding put_pages().. we are mapping
551 * the original page, rather than thru a GART, so we can't rely
552 * on eviction to trigger this. But munmap() or all mappings should
553 * probably trigger put_pages()?
554 */
555
556 if (omap_obj->flags & OMAP_BO_TILED)
557 ret = fault_2d(obj, vma, vmf);
558 else
559 ret = fault_1d(obj, vma, vmf);
560
561
562fail:
563 mutex_unlock(&dev->struct_mutex);
564 switch (ret) {
565 case 0:
566 case -ERESTARTSYS:
567 case -EINTR:
568 case -EBUSY:
569 /*
570 * EBUSY is ok: this just means that another thread
571 * already did the job.
572 */
573 return VM_FAULT_NOPAGE;
574 case -ENOMEM:
575 return VM_FAULT_OOM;
576 default:
577 return VM_FAULT_SIGBUS;
578 }
579}
580
581/** We override mainly to fix up some of the vm mapping flags.. */
582int omap_gem_mmap(struct file *filp, struct vm_area_struct *vma)
583{
584 int ret;
585
586 ret = drm_gem_mmap(filp, vma);
587 if (ret) {
588 DBG("mmap failed: %d", ret);
589 return ret;
590 }
591
592 return omap_gem_mmap_obj(vma->vm_private_data, vma);
593}
594
595int omap_gem_mmap_obj(struct drm_gem_object *obj,
596 struct vm_area_struct *vma)
597{
598 struct omap_gem_object *omap_obj = to_omap_bo(obj);
599
600 vma->vm_flags &= ~VM_PFNMAP;
601 vma->vm_flags |= VM_MIXEDMAP;
602
603 if (omap_obj->flags & OMAP_BO_WC) {
604 vma->vm_page_prot = pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
605 } else if (omap_obj->flags & OMAP_BO_UNCACHED) {
606 vma->vm_page_prot = pgprot_noncached(vm_get_page_prot(vma->vm_flags));
607 } else {
608 /*
609 * We do have some private objects, at least for scanout buffers
610 * on hardware without DMM/TILER. But these are allocated write-
611 * combine
612 */
613 if (WARN_ON(!obj->filp))
614 return -EINVAL;
615
616 /*
617 * Shunt off cached objs to shmem file so they have their own
618 * address_space (so unmap_mapping_range does what we want,
619 * in particular in the case of mmap'd dmabufs)
620 */
621 fput(vma->vm_file);
622 vma->vm_pgoff = 0;
623 vma->vm_file = get_file(obj->filp);
624
625 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
626 }
627
628 return 0;
629}
630
631/* -----------------------------------------------------------------------------
632 * Dumb Buffers
633 */
634
635/**
636 * omap_gem_dumb_create - create a dumb buffer
637 * @drm_file: our client file
638 * @dev: our device
639 * @args: the requested arguments copied from userspace
640 *
641 * Allocate a buffer suitable for use for a frame buffer of the
642 * form described by user space. Give userspace a handle by which
643 * to reference it.
644 */
645int omap_gem_dumb_create(struct drm_file *file, struct drm_device *dev,
646 struct drm_mode_create_dumb *args)
647{
648 union omap_gem_size gsize;
649
650 args->pitch = DIV_ROUND_UP(args->width * args->bpp, 8);
651
652 args->size = PAGE_ALIGN(args->pitch * args->height);
653
654 gsize = (union omap_gem_size){
655 .bytes = args->size,
656 };
657
658 return omap_gem_new_handle(dev, file, gsize,
659 OMAP_BO_SCANOUT | OMAP_BO_WC, &args->handle);
660}
661
662/**
663 * omap_gem_dumb_map - buffer mapping for dumb interface
664 * @file: our drm client file
665 * @dev: drm device
666 * @handle: GEM handle to the object (from dumb_create)
667 *
668 * Do the necessary setup to allow the mapping of the frame buffer
669 * into user memory. We don't have to do much here at the moment.
670 */
671int omap_gem_dumb_map_offset(struct drm_file *file, struct drm_device *dev,
672 uint32_t handle, uint64_t *offset)
673{
674 struct drm_gem_object *obj;
675 int ret = 0;
676
677 /* GEM does all our handle to object mapping */
678 obj = drm_gem_object_lookup(file, handle);
679 if (obj == NULL) {
680 ret = -ENOENT;
681 goto fail;
682 }
683
684 *offset = omap_gem_mmap_offset(obj);
685
686 drm_gem_object_unreference_unlocked(obj);
687
688fail:
689 return ret;
690}
691
692#ifdef CONFIG_DRM_FBDEV_EMULATION
693/* Set scrolling position. This allows us to implement fast scrolling
694 * for console.
695 *
696 * Call only from non-atomic contexts.
697 */
698int omap_gem_roll(struct drm_gem_object *obj, uint32_t roll)
699{
700 struct omap_gem_object *omap_obj = to_omap_bo(obj);
701 uint32_t npages = obj->size >> PAGE_SHIFT;
702 int ret = 0;
703
704 if (roll > npages) {
705 dev_err(obj->dev->dev, "invalid roll: %d\n", roll);
706 return -EINVAL;
707 }
708
709 omap_obj->roll = roll;
710
711 mutex_lock(&obj->dev->struct_mutex);
712
713 /* if we aren't mapped yet, we don't need to do anything */
714 if (omap_obj->block) {
715 struct page **pages;
716 ret = get_pages(obj, &pages);
717 if (ret)
718 goto fail;
719 ret = tiler_pin(omap_obj->block, pages, npages, roll, true);
720 if (ret)
721 dev_err(obj->dev->dev, "could not repin: %d\n", ret);
722 }
723
724fail:
725 mutex_unlock(&obj->dev->struct_mutex);
726
727 return ret;
728}
729#endif
730
731/* -----------------------------------------------------------------------------
732 * Memory Management & DMA Sync
733 */
734
735/**
736 * shmem buffers that are mapped cached can simulate coherency via using
737 * page faulting to keep track of dirty pages
738 */
739static inline bool is_cached_coherent(struct drm_gem_object *obj)
740{
741 struct omap_gem_object *omap_obj = to_omap_bo(obj);
742
743 return (omap_obj->flags & OMAP_BO_MEM_SHMEM) &&
744 ((omap_obj->flags & OMAP_BO_CACHE_MASK) == OMAP_BO_CACHED);
745}
746
747/* Sync the buffer for CPU access.. note pages should already be
748 * attached, ie. omap_gem_get_pages()
749 */
750void omap_gem_cpu_sync(struct drm_gem_object *obj, int pgoff)
751{
752 struct drm_device *dev = obj->dev;
753 struct omap_gem_object *omap_obj = to_omap_bo(obj);
754
755 if (is_cached_coherent(obj) && omap_obj->addrs[pgoff]) {
756 dma_unmap_page(dev->dev, omap_obj->addrs[pgoff],
757 PAGE_SIZE, DMA_BIDIRECTIONAL);
758 omap_obj->addrs[pgoff] = 0;
759 }
760}
761
762/* sync the buffer for DMA access */
763void omap_gem_dma_sync(struct drm_gem_object *obj,
764 enum dma_data_direction dir)
765{
766 struct drm_device *dev = obj->dev;
767 struct omap_gem_object *omap_obj = to_omap_bo(obj);
768
769 if (is_cached_coherent(obj)) {
770 int i, npages = obj->size >> PAGE_SHIFT;
771 struct page **pages = omap_obj->pages;
772 bool dirty = false;
773
774 for (i = 0; i < npages; i++) {
775 if (!omap_obj->addrs[i]) {
776 dma_addr_t addr;
777
778 addr = dma_map_page(dev->dev, pages[i], 0,
779 PAGE_SIZE, DMA_BIDIRECTIONAL);
780
781 if (dma_mapping_error(dev->dev, addr)) {
782 dev_warn(dev->dev,
783 "%s: failed to map page\n",
784 __func__);
785 break;
786 }
787
788 dirty = true;
789 omap_obj->addrs[i] = addr;
790 }
791 }
792
793 if (dirty) {
794 unmap_mapping_range(obj->filp->f_mapping, 0,
795 omap_gem_mmap_size(obj), 1);
796 }
797 }
798}
799
800/* Get physical address for DMA.. if 'remap' is true, and the buffer is not
801 * already contiguous, remap it to pin in physically contiguous memory.. (ie.
802 * map in TILER)
803 */
804int omap_gem_get_paddr(struct drm_gem_object *obj,
805 dma_addr_t *paddr, bool remap)
806{
807 struct omap_drm_private *priv = obj->dev->dev_private;
808 struct omap_gem_object *omap_obj = to_omap_bo(obj);
809 int ret = 0;
810
811 mutex_lock(&obj->dev->struct_mutex);
812
813 if (!is_contiguous(omap_obj) && remap && priv->has_dmm) {
814 if (omap_obj->paddr_cnt == 0) {
815 struct page **pages;
816 uint32_t npages = obj->size >> PAGE_SHIFT;
817 enum tiler_fmt fmt = gem2fmt(omap_obj->flags);
818 struct tiler_block *block;
819
820 BUG_ON(omap_obj->block);
821
822 ret = get_pages(obj, &pages);
823 if (ret)
824 goto fail;
825
826 if (omap_obj->flags & OMAP_BO_TILED) {
827 block = tiler_reserve_2d(fmt,
828 omap_obj->width,
829 omap_obj->height, 0);
830 } else {
831 block = tiler_reserve_1d(obj->size);
832 }
833
834 if (IS_ERR(block)) {
835 ret = PTR_ERR(block);
836 dev_err(obj->dev->dev,
837 "could not remap: %d (%d)\n", ret, fmt);
838 goto fail;
839 }
840
841 /* TODO: enable async refill.. */
842 ret = tiler_pin(block, pages, npages,
843 omap_obj->roll, true);
844 if (ret) {
845 tiler_release(block);
846 dev_err(obj->dev->dev,
847 "could not pin: %d\n", ret);
848 goto fail;
849 }
850
851 omap_obj->paddr = tiler_ssptr(block);
852 omap_obj->block = block;
853
854 DBG("got paddr: %pad", &omap_obj->paddr);
855 }
856
857 omap_obj->paddr_cnt++;
858
859 *paddr = omap_obj->paddr;
860 } else if (is_contiguous(omap_obj)) {
861 *paddr = omap_obj->paddr;
862 } else {
863 ret = -EINVAL;
864 goto fail;
865 }
866
867fail:
868 mutex_unlock(&obj->dev->struct_mutex);
869
870 return ret;
871}
872
873/* Release physical address, when DMA is no longer being performed.. this
874 * could potentially unpin and unmap buffers from TILER
875 */
876void omap_gem_put_paddr(struct drm_gem_object *obj)
877{
878 struct omap_gem_object *omap_obj = to_omap_bo(obj);
879 int ret;
880
881 mutex_lock(&obj->dev->struct_mutex);
882 if (omap_obj->paddr_cnt > 0) {
883 omap_obj->paddr_cnt--;
884 if (omap_obj->paddr_cnt == 0) {
885 ret = tiler_unpin(omap_obj->block);
886 if (ret) {
887 dev_err(obj->dev->dev,
888 "could not unpin pages: %d\n", ret);
889 }
890 ret = tiler_release(omap_obj->block);
891 if (ret) {
892 dev_err(obj->dev->dev,
893 "could not release unmap: %d\n", ret);
894 }
895 omap_obj->paddr = 0;
896 omap_obj->block = NULL;
897 }
898 }
899
900 mutex_unlock(&obj->dev->struct_mutex);
901}
902
903/* Get rotated scanout address (only valid if already pinned), at the
904 * specified orientation and x,y offset from top-left corner of buffer
905 * (only valid for tiled 2d buffers)
906 */
907int omap_gem_rotated_paddr(struct drm_gem_object *obj, uint32_t orient,
908 int x, int y, dma_addr_t *paddr)
909{
910 struct omap_gem_object *omap_obj = to_omap_bo(obj);
911 int ret = -EINVAL;
912
913 mutex_lock(&obj->dev->struct_mutex);
914 if ((omap_obj->paddr_cnt > 0) && omap_obj->block &&
915 (omap_obj->flags & OMAP_BO_TILED)) {
916 *paddr = tiler_tsptr(omap_obj->block, orient, x, y);
917 ret = 0;
918 }
919 mutex_unlock(&obj->dev->struct_mutex);
920 return ret;
921}
922
923/* Get tiler stride for the buffer (only valid for 2d tiled buffers) */
924int omap_gem_tiled_stride(struct drm_gem_object *obj, uint32_t orient)
925{
926 struct omap_gem_object *omap_obj = to_omap_bo(obj);
927 int ret = -EINVAL;
928 if (omap_obj->flags & OMAP_BO_TILED)
929 ret = tiler_stride(gem2fmt(omap_obj->flags), orient);
930 return ret;
931}
932
933/* if !remap, and we don't have pages backing, then fail, rather than
934 * increasing the pin count (which we don't really do yet anyways,
935 * because we don't support swapping pages back out). And 'remap'
936 * might not be quite the right name, but I wanted to keep it working
937 * similarly to omap_gem_get_paddr(). Note though that mutex is not
938 * aquired if !remap (because this can be called in atomic ctxt),
939 * but probably omap_gem_get_paddr() should be changed to work in the
940 * same way. If !remap, a matching omap_gem_put_pages() call is not
941 * required (and should not be made).
942 */
943int omap_gem_get_pages(struct drm_gem_object *obj, struct page ***pages,
944 bool remap)
945{
946 int ret;
947 if (!remap) {
948 struct omap_gem_object *omap_obj = to_omap_bo(obj);
949 if (!omap_obj->pages)
950 return -ENOMEM;
951 *pages = omap_obj->pages;
952 return 0;
953 }
954 mutex_lock(&obj->dev->struct_mutex);
955 ret = get_pages(obj, pages);
956 mutex_unlock(&obj->dev->struct_mutex);
957 return ret;
958}
959
960/* release pages when DMA no longer being performed */
961int omap_gem_put_pages(struct drm_gem_object *obj)
962{
963 /* do something here if we dynamically attach/detach pages.. at
964 * least they would no longer need to be pinned if everyone has
965 * released the pages..
966 */
967 return 0;
968}
969
970#ifdef CONFIG_DRM_FBDEV_EMULATION
971/* Get kernel virtual address for CPU access.. this more or less only
972 * exists for omap_fbdev. This should be called with struct_mutex
973 * held.
974 */
975void *omap_gem_vaddr(struct drm_gem_object *obj)
976{
977 struct omap_gem_object *omap_obj = to_omap_bo(obj);
978 WARN_ON(!mutex_is_locked(&obj->dev->struct_mutex));
979 if (!omap_obj->vaddr) {
980 struct page **pages;
981 int ret = get_pages(obj, &pages);
982 if (ret)
983 return ERR_PTR(ret);
984 omap_obj->vaddr = vmap(pages, obj->size >> PAGE_SHIFT,
985 VM_MAP, pgprot_writecombine(PAGE_KERNEL));
986 }
987 return omap_obj->vaddr;
988}
989#endif
990
991/* -----------------------------------------------------------------------------
992 * Power Management
993 */
994
995#ifdef CONFIG_PM
996/* re-pin objects in DMM in resume path: */
997int omap_gem_resume(struct device *dev)
998{
999 struct drm_device *drm_dev = dev_get_drvdata(dev);
1000 struct omap_drm_private *priv = drm_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 uint32_t npages = obj->size >> PAGE_SHIFT;
1008 WARN_ON(!omap_obj->pages); /* this can't happen */
1009 ret = tiler_pin(omap_obj->block,
1010 omap_obj->pages, npages,
1011 omap_obj->roll, true);
1012 if (ret) {
1013 dev_err(dev, "could not repin: %d\n", ret);
1014 return ret;
1015 }
1016 }
1017 }
1018
1019 return 0;
1020}
1021#endif
1022
1023/* -----------------------------------------------------------------------------
1024 * DebugFS
1025 */
1026
1027#ifdef CONFIG_DEBUG_FS
1028void omap_gem_describe(struct drm_gem_object *obj, struct seq_file *m)
1029{
1030 struct omap_gem_object *omap_obj = to_omap_bo(obj);
1031 uint64_t off;
1032
1033 off = drm_vma_node_start(&obj->vma_node);
1034
1035 seq_printf(m, "%08x: %2d (%2d) %08llx %pad (%2d) %p %4d",
1036 omap_obj->flags, obj->name, obj->refcount.refcount.counter,
1037 off, &omap_obj->paddr, omap_obj->paddr_cnt,
1038 omap_obj->vaddr, omap_obj->roll);
1039
1040 if (omap_obj->flags & OMAP_BO_TILED) {
1041 seq_printf(m, " %dx%d", omap_obj->width, omap_obj->height);
1042 if (omap_obj->block) {
1043 struct tcm_area *area = &omap_obj->block->area;
1044 seq_printf(m, " (%dx%d, %dx%d)",
1045 area->p0.x, area->p0.y,
1046 area->p1.x, area->p1.y);
1047 }
1048 } else {
1049 seq_printf(m, " %d", obj->size);
1050 }
1051
1052 seq_printf(m, "\n");
1053}
1054
1055void omap_gem_describe_objects(struct list_head *list, struct seq_file *m)
1056{
1057 struct omap_gem_object *omap_obj;
1058 int count = 0;
1059 size_t size = 0;
1060
1061 list_for_each_entry(omap_obj, list, mm_list) {
1062 struct drm_gem_object *obj = &omap_obj->base;
1063 seq_printf(m, " ");
1064 omap_gem_describe(obj, m);
1065 count++;
1066 size += obj->size;
1067 }
1068
1069 seq_printf(m, "Total %d objects, %zu bytes\n", count, size);
1070}
1071#endif
1072
1073/* -----------------------------------------------------------------------------
1074 * Buffer Synchronization
1075 */
1076
1077static DEFINE_SPINLOCK(sync_lock);
1078
1079struct omap_gem_sync_waiter {
1080 struct list_head list;
1081 struct omap_gem_object *omap_obj;
1082 enum omap_gem_op op;
1083 uint32_t read_target, write_target;
1084 /* notify called w/ sync_lock held */
1085 void (*notify)(void *arg);
1086 void *arg;
1087};
1088
1089/* list of omap_gem_sync_waiter.. the notify fxn gets called back when
1090 * the read and/or write target count is achieved which can call a user
1091 * callback (ex. to kick 3d and/or 2d), wakeup blocked task (prep for
1092 * cpu access), etc.
1093 */
1094static LIST_HEAD(waiters);
1095
1096static inline bool is_waiting(struct omap_gem_sync_waiter *waiter)
1097{
1098 struct omap_gem_object *omap_obj = waiter->omap_obj;
1099 if ((waiter->op & OMAP_GEM_READ) &&
1100 (omap_obj->sync->write_complete < waiter->write_target))
1101 return true;
1102 if ((waiter->op & OMAP_GEM_WRITE) &&
1103 (omap_obj->sync->read_complete < waiter->read_target))
1104 return true;
1105 return false;
1106}
1107
1108/* macro for sync debug.. */
1109#define SYNCDBG 0
1110#define SYNC(fmt, ...) do { if (SYNCDBG) \
1111 printk(KERN_ERR "%s:%d: "fmt"\n", \
1112 __func__, __LINE__, ##__VA_ARGS__); \
1113 } while (0)
1114
1115
1116static void sync_op_update(void)
1117{
1118 struct omap_gem_sync_waiter *waiter, *n;
1119 list_for_each_entry_safe(waiter, n, &waiters, list) {
1120 if (!is_waiting(waiter)) {
1121 list_del(&waiter->list);
1122 SYNC("notify: %p", waiter);
1123 waiter->notify(waiter->arg);
1124 kfree(waiter);
1125 }
1126 }
1127}
1128
1129static inline int sync_op(struct drm_gem_object *obj,
1130 enum omap_gem_op op, bool start)
1131{
1132 struct omap_gem_object *omap_obj = to_omap_bo(obj);
1133 int ret = 0;
1134
1135 spin_lock(&sync_lock);
1136
1137 if (!omap_obj->sync) {
1138 omap_obj->sync = kzalloc(sizeof(*omap_obj->sync), GFP_ATOMIC);
1139 if (!omap_obj->sync) {
1140 ret = -ENOMEM;
1141 goto unlock;
1142 }
1143 }
1144
1145 if (start) {
1146 if (op & OMAP_GEM_READ)
1147 omap_obj->sync->read_pending++;
1148 if (op & OMAP_GEM_WRITE)
1149 omap_obj->sync->write_pending++;
1150 } else {
1151 if (op & OMAP_GEM_READ)
1152 omap_obj->sync->read_complete++;
1153 if (op & OMAP_GEM_WRITE)
1154 omap_obj->sync->write_complete++;
1155 sync_op_update();
1156 }
1157
1158unlock:
1159 spin_unlock(&sync_lock);
1160
1161 return ret;
1162}
1163
1164/* mark the start of read and/or write operation */
1165int omap_gem_op_start(struct drm_gem_object *obj, enum omap_gem_op op)
1166{
1167 return sync_op(obj, op, true);
1168}
1169
1170int omap_gem_op_finish(struct drm_gem_object *obj, enum omap_gem_op op)
1171{
1172 return sync_op(obj, op, false);
1173}
1174
1175static DECLARE_WAIT_QUEUE_HEAD(sync_event);
1176
1177static void sync_notify(void *arg)
1178{
1179 struct task_struct **waiter_task = arg;
1180 *waiter_task = NULL;
1181 wake_up_all(&sync_event);
1182}
1183
1184int omap_gem_op_sync(struct drm_gem_object *obj, enum omap_gem_op op)
1185{
1186 struct omap_gem_object *omap_obj = to_omap_bo(obj);
1187 int ret = 0;
1188 if (omap_obj->sync) {
1189 struct task_struct *waiter_task = current;
1190 struct omap_gem_sync_waiter *waiter =
1191 kzalloc(sizeof(*waiter), GFP_KERNEL);
1192
1193 if (!waiter)
1194 return -ENOMEM;
1195
1196 waiter->omap_obj = omap_obj;
1197 waiter->op = op;
1198 waiter->read_target = omap_obj->sync->read_pending;
1199 waiter->write_target = omap_obj->sync->write_pending;
1200 waiter->notify = sync_notify;
1201 waiter->arg = &waiter_task;
1202
1203 spin_lock(&sync_lock);
1204 if (is_waiting(waiter)) {
1205 SYNC("waited: %p", waiter);
1206 list_add_tail(&waiter->list, &waiters);
1207 spin_unlock(&sync_lock);
1208 ret = wait_event_interruptible(sync_event,
1209 (waiter_task == NULL));
1210 spin_lock(&sync_lock);
1211 if (waiter_task) {
1212 SYNC("interrupted: %p", waiter);
1213 /* we were interrupted */
1214 list_del(&waiter->list);
1215 waiter_task = NULL;
1216 } else {
1217 /* freed in sync_op_update() */
1218 waiter = NULL;
1219 }
1220 }
1221 spin_unlock(&sync_lock);
1222 kfree(waiter);
1223 }
1224 return ret;
1225}
1226
1227/* call fxn(arg), either synchronously or asynchronously if the op
1228 * is currently blocked.. fxn() can be called from any context
1229 *
1230 * (TODO for now fxn is called back from whichever context calls
1231 * omap_gem_op_finish().. but this could be better defined later
1232 * if needed)
1233 *
1234 * TODO more code in common w/ _sync()..
1235 */
1236int omap_gem_op_async(struct drm_gem_object *obj, enum omap_gem_op op,
1237 void (*fxn)(void *arg), void *arg)
1238{
1239 struct omap_gem_object *omap_obj = to_omap_bo(obj);
1240 if (omap_obj->sync) {
1241 struct omap_gem_sync_waiter *waiter =
1242 kzalloc(sizeof(*waiter), GFP_ATOMIC);
1243
1244 if (!waiter)
1245 return -ENOMEM;
1246
1247 waiter->omap_obj = omap_obj;
1248 waiter->op = op;
1249 waiter->read_target = omap_obj->sync->read_pending;
1250 waiter->write_target = omap_obj->sync->write_pending;
1251 waiter->notify = fxn;
1252 waiter->arg = arg;
1253
1254 spin_lock(&sync_lock);
1255 if (is_waiting(waiter)) {
1256 SYNC("waited: %p", waiter);
1257 list_add_tail(&waiter->list, &waiters);
1258 spin_unlock(&sync_lock);
1259 return 0;
1260 }
1261
1262 spin_unlock(&sync_lock);
1263
1264 kfree(waiter);
1265 }
1266
1267 /* no waiting.. */
1268 fxn(arg);
1269
1270 return 0;
1271}
1272
1273/* -----------------------------------------------------------------------------
1274 * Constructor & Destructor
1275 */
1276
1277void omap_gem_free_object(struct drm_gem_object *obj)
1278{
1279 struct drm_device *dev = obj->dev;
1280 struct omap_drm_private *priv = dev->dev_private;
1281 struct omap_gem_object *omap_obj = to_omap_bo(obj);
1282
1283 evict(obj);
1284
1285 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
1286
1287 spin_lock(&priv->list_lock);
1288 list_del(&omap_obj->mm_list);
1289 spin_unlock(&priv->list_lock);
1290
1291 /* this means the object is still pinned.. which really should
1292 * not happen. I think..
1293 */
1294 WARN_ON(omap_obj->paddr_cnt > 0);
1295
1296 if (omap_obj->pages) {
1297 if (omap_obj->flags & OMAP_BO_MEM_DMABUF)
1298 kfree(omap_obj->pages);
1299 else
1300 omap_gem_detach_pages(obj);
1301 }
1302
1303 if (omap_obj->flags & OMAP_BO_MEM_DMA_API) {
1304 dma_free_wc(dev->dev, obj->size, omap_obj->vaddr,
1305 omap_obj->paddr);
1306 } else if (omap_obj->vaddr) {
1307 vunmap(omap_obj->vaddr);
1308 } else if (obj->import_attach) {
1309 drm_prime_gem_destroy(obj, omap_obj->sgt);
1310 }
1311
1312 kfree(omap_obj->sync);
1313
1314 drm_gem_object_release(obj);
1315
1316 kfree(omap_obj);
1317}
1318
1319/* GEM buffer object constructor */
1320struct drm_gem_object *omap_gem_new(struct drm_device *dev,
1321 union omap_gem_size gsize, uint32_t flags)
1322{
1323 struct omap_drm_private *priv = dev->dev_private;
1324 struct omap_gem_object *omap_obj;
1325 struct drm_gem_object *obj;
1326 struct address_space *mapping;
1327 size_t size;
1328 int ret;
1329
1330 /* Validate the flags and compute the memory and cache flags. */
1331 if (flags & OMAP_BO_TILED) {
1332 if (!priv->usergart) {
1333 dev_err(dev->dev, "Tiled buffers require DMM\n");
1334 return NULL;
1335 }
1336
1337 /*
1338 * Tiled buffers are always shmem paged backed. When they are
1339 * scanned out, they are remapped into DMM/TILER.
1340 */
1341 flags &= ~OMAP_BO_SCANOUT;
1342 flags |= OMAP_BO_MEM_SHMEM;
1343
1344 /*
1345 * Currently don't allow cached buffers. There is some caching
1346 * stuff that needs to be handled better.
1347 */
1348 flags &= ~(OMAP_BO_CACHED|OMAP_BO_WC|OMAP_BO_UNCACHED);
1349 flags |= tiler_get_cpu_cache_flags();
1350 } else if ((flags & OMAP_BO_SCANOUT) && !priv->has_dmm) {
1351 /*
1352 * OMAP_BO_SCANOUT hints that the buffer doesn't need to be
1353 * tiled. However, to lower the pressure on memory allocation,
1354 * use contiguous memory only if no TILER is available.
1355 */
1356 flags |= OMAP_BO_MEM_DMA_API;
1357 } else if (!(flags & OMAP_BO_MEM_DMABUF)) {
1358 /*
1359 * All other buffers not backed by dma_buf are shmem-backed.
1360 */
1361 flags |= OMAP_BO_MEM_SHMEM;
1362 }
1363
1364 /* Allocate the initialize the OMAP GEM object. */
1365 omap_obj = kzalloc(sizeof(*omap_obj), GFP_KERNEL);
1366 if (!omap_obj)
1367 return NULL;
1368
1369 obj = &omap_obj->base;
1370 omap_obj->flags = flags;
1371
1372 if (flags & OMAP_BO_TILED) {
1373 /*
1374 * For tiled buffers align dimensions to slot boundaries and
1375 * calculate size based on aligned dimensions.
1376 */
1377 tiler_align(gem2fmt(flags), &gsize.tiled.width,
1378 &gsize.tiled.height);
1379
1380 size = tiler_size(gem2fmt(flags), gsize.tiled.width,
1381 gsize.tiled.height);
1382
1383 omap_obj->width = gsize.tiled.width;
1384 omap_obj->height = gsize.tiled.height;
1385 } else {
1386 size = PAGE_ALIGN(gsize.bytes);
1387 }
1388
1389 /* Initialize the GEM object. */
1390 if (!(flags & OMAP_BO_MEM_SHMEM)) {
1391 drm_gem_private_object_init(dev, obj, size);
1392 } else {
1393 ret = drm_gem_object_init(dev, obj, size);
1394 if (ret)
1395 goto err_free;
1396
1397 mapping = obj->filp->f_mapping;
1398 mapping_set_gfp_mask(mapping, GFP_USER | __GFP_DMA32);
1399 }
1400
1401 /* Allocate memory if needed. */
1402 if (flags & OMAP_BO_MEM_DMA_API) {
1403 omap_obj->vaddr = dma_alloc_wc(dev->dev, size,
1404 &omap_obj->paddr,
1405 GFP_KERNEL);
1406 if (!omap_obj->vaddr)
1407 goto err_release;
1408 }
1409
1410 spin_lock(&priv->list_lock);
1411 list_add(&omap_obj->mm_list, &priv->obj_list);
1412 spin_unlock(&priv->list_lock);
1413
1414 return obj;
1415
1416err_release:
1417 drm_gem_object_release(obj);
1418err_free:
1419 kfree(omap_obj);
1420 return NULL;
1421}
1422
1423struct drm_gem_object *omap_gem_new_dmabuf(struct drm_device *dev, size_t size,
1424 struct sg_table *sgt)
1425{
1426 struct omap_drm_private *priv = dev->dev_private;
1427 struct omap_gem_object *omap_obj;
1428 struct drm_gem_object *obj;
1429 union omap_gem_size gsize;
1430
1431 /* Without a DMM only physically contiguous buffers can be supported. */
1432 if (sgt->orig_nents != 1 && !priv->has_dmm)
1433 return ERR_PTR(-EINVAL);
1434
1435 mutex_lock(&dev->struct_mutex);
1436
1437 gsize.bytes = PAGE_ALIGN(size);
1438 obj = omap_gem_new(dev, gsize, OMAP_BO_MEM_DMABUF | OMAP_BO_WC);
1439 if (!obj) {
1440 obj = ERR_PTR(-ENOMEM);
1441 goto done;
1442 }
1443
1444 omap_obj = to_omap_bo(obj);
1445 omap_obj->sgt = sgt;
1446
1447 if (sgt->orig_nents == 1) {
1448 omap_obj->paddr = sg_dma_address(sgt->sgl);
1449 } else {
1450 /* Create pages list from sgt */
1451 struct sg_page_iter iter;
1452 struct page **pages;
1453 unsigned int npages;
1454 unsigned int i = 0;
1455
1456 npages = DIV_ROUND_UP(size, PAGE_SIZE);
1457 pages = kcalloc(npages, sizeof(*pages), GFP_KERNEL);
1458 if (!pages) {
1459 omap_gem_free_object(obj);
1460 obj = ERR_PTR(-ENOMEM);
1461 goto done;
1462 }
1463
1464 omap_obj->pages = pages;
1465
1466 for_each_sg_page(sgt->sgl, &iter, sgt->orig_nents, 0) {
1467 pages[i++] = sg_page_iter_page(&iter);
1468 if (i > npages)
1469 break;
1470 }
1471
1472 if (WARN_ON(i != npages)) {
1473 omap_gem_free_object(obj);
1474 obj = ERR_PTR(-ENOMEM);
1475 goto done;
1476 }
1477 }
1478
1479done:
1480 mutex_unlock(&dev->struct_mutex);
1481 return obj;
1482}
1483
1484/* convenience method to construct a GEM buffer object, and userspace handle */
1485int omap_gem_new_handle(struct drm_device *dev, struct drm_file *file,
1486 union omap_gem_size gsize, uint32_t flags, uint32_t *handle)
1487{
1488 struct drm_gem_object *obj;
1489 int ret;
1490
1491 obj = omap_gem_new(dev, gsize, flags);
1492 if (!obj)
1493 return -ENOMEM;
1494
1495 ret = drm_gem_handle_create(file, obj, handle);
1496 if (ret) {
1497 omap_gem_free_object(obj);
1498 return ret;
1499 }
1500
1501 /* drop reference from allocate - handle holds it now */
1502 drm_gem_object_unreference_unlocked(obj);
1503
1504 return 0;
1505}
1506
1507/* -----------------------------------------------------------------------------
1508 * Init & Cleanup
1509 */
1510
1511/* If DMM is used, we need to set some stuff up.. */
1512void omap_gem_init(struct drm_device *dev)
1513{
1514 struct omap_drm_private *priv = dev->dev_private;
1515 struct omap_drm_usergart *usergart;
1516 const enum tiler_fmt fmts[] = {
1517 TILFMT_8BIT, TILFMT_16BIT, TILFMT_32BIT
1518 };
1519 int i, j;
1520
1521 if (!dmm_is_available()) {
1522 /* DMM only supported on OMAP4 and later, so this isn't fatal */
1523 dev_warn(dev->dev, "DMM not available, disable DMM support\n");
1524 return;
1525 }
1526
1527 usergart = kcalloc(3, sizeof(*usergart), GFP_KERNEL);
1528 if (!usergart)
1529 return;
1530
1531 /* reserve 4k aligned/wide regions for userspace mappings: */
1532 for (i = 0; i < ARRAY_SIZE(fmts); i++) {
1533 uint16_t h = 1, w = PAGE_SIZE >> i;
1534 tiler_align(fmts[i], &w, &h);
1535 /* note: since each region is 1 4kb page wide, and minimum
1536 * number of rows, the height ends up being the same as the
1537 * # of pages in the region
1538 */
1539 usergart[i].height = h;
1540 usergart[i].height_shift = ilog2(h);
1541 usergart[i].stride_pfn = tiler_stride(fmts[i], 0) >> PAGE_SHIFT;
1542 usergart[i].slot_shift = ilog2((PAGE_SIZE / h) >> i);
1543 for (j = 0; j < NUM_USERGART_ENTRIES; j++) {
1544 struct omap_drm_usergart_entry *entry;
1545 struct tiler_block *block;
1546
1547 entry = &usergart[i].entry[j];
1548 block = tiler_reserve_2d(fmts[i], w, h, PAGE_SIZE);
1549 if (IS_ERR(block)) {
1550 dev_err(dev->dev,
1551 "reserve failed: %d, %d, %ld\n",
1552 i, j, PTR_ERR(block));
1553 return;
1554 }
1555 entry->paddr = tiler_ssptr(block);
1556 entry->block = block;
1557
1558 DBG("%d:%d: %dx%d: paddr=%pad stride=%d", i, j, w, h,
1559 &entry->paddr,
1560 usergart[i].stride_pfn << PAGE_SHIFT);
1561 }
1562 }
1563
1564 priv->usergart = usergart;
1565 priv->has_dmm = true;
1566}
1567
1568void omap_gem_deinit(struct drm_device *dev)
1569{
1570 struct omap_drm_private *priv = dev->dev_private;
1571
1572 /* I believe we can rely on there being no more outstanding GEM
1573 * objects which could depend on usergart/dmm at this point.
1574 */
1575 kfree(priv->usergart);
1576}