ttm_tt.c revision 1.22 
1/* SPDX-License-Identifier: GPL-2.0 OR MIT */
2/**************************************************************************
3 *
4 * Copyright (c) 2006-2009 VMware, Inc., Palo Alto, CA., USA
5 * All Rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
14 *
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
17 * of the Software.
18 *
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
22 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
23 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
24 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
25 * USE OR OTHER DEALINGS IN THE SOFTWARE.
26 *
27 **************************************************************************/
28/*
29 * Authors: Thomas Hellstrom <thellstrom-at-vmware-dot-com>
30 */
31
32#define pr_fmt(fmt) "[TTM] " fmt
33
34#include <linux/cc_platform.h>
35#include <linux/sched.h>
36#include <linux/shmem_fs.h>
37#include <linux/file.h>
38#include <linux/module.h>
39#include <drm/drm_cache.h>
40#include <drm/drm_device.h>
41#include <drm/drm_util.h>
42#include <drm/ttm/ttm_bo.h>
43#include <drm/ttm/ttm_tt.h>
44
45#include "ttm_module.h"
46
47static unsigned long ttm_pages_limit;
48
49MODULE_PARM_DESC(pages_limit, "Limit for the allocated pages");
50module_param_named(pages_limit, ttm_pages_limit, ulong, 0644);
51
52static unsigned long ttm_dma32_pages_limit;
53
54MODULE_PARM_DESC(dma32_pages_limit, "Limit for the allocated DMA32 pages");
55module_param_named(dma32_pages_limit, ttm_dma32_pages_limit, ulong, 0644);
56
57static atomic_long_t ttm_pages_allocated;
58static atomic_long_t ttm_dma32_pages_allocated;
59
60/*
61 * Allocates a ttm structure for the given BO.
62 */
63int ttm_tt_create(struct ttm_buffer_object *bo, bool zero_alloc)
64{
65	struct ttm_device *bdev = bo->bdev;
66	struct drm_device *ddev = bo->base.dev;
67	uint32_t page_flags = 0;
68
69	dma_resv_assert_held(bo->base.resv);
70
71	if (bo->ttm)
72		return 0;
73
74	switch (bo->type) {
75	case ttm_bo_type_device:
76		if (zero_alloc)
77			page_flags |= TTM_TT_FLAG_ZERO_ALLOC;
78		break;
79	case ttm_bo_type_kernel:
80		break;
81	case ttm_bo_type_sg:
82		page_flags |= TTM_TT_FLAG_EXTERNAL;
83		break;
84	default:
85		pr_err("Illegal buffer object type\n");
86		return -EINVAL;
87	}
88	/*
89	 * When using dma_alloc_coherent with memory encryption the
90	 * mapped TT pages need to be decrypted or otherwise the drivers
91	 * will end up sending encrypted mem to the gpu.
92	 */
93	if (bdev->pool.use_dma_alloc && cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) {
94		page_flags |= TTM_TT_FLAG_DECRYPTED;
95		drm_info_once(ddev, "TT memory decryption enabled.");
96	}
97
98	bo->ttm = bdev->funcs->ttm_tt_create(bo, page_flags);
99	if (unlikely(bo->ttm == NULL))
100		return -ENOMEM;
101
102	WARN_ON(bo->ttm->page_flags & TTM_TT_FLAG_EXTERNAL_MAPPABLE &&
103		!(bo->ttm->page_flags & TTM_TT_FLAG_EXTERNAL));
104
105	return 0;
106}
107
108/*
109 * Allocates storage for pointers to the pages that back the ttm.
110 */
111static int ttm_tt_alloc_page_directory(struct ttm_tt *ttm)
112{
113	ttm->pages = kvcalloc(ttm->num_pages, sizeof(void*), GFP_KERNEL);
114	if (!ttm->pages)
115		return -ENOMEM;
116	ttm->orders = kvmalloc_array(ttm->num_pages,
117	    sizeof(unsigned long), GFP_KERNEL | __GFP_ZERO);
118	if (!ttm->orders)
119		return -ENOMEM;
120	return 0;
121}
122
123static int ttm_dma_tt_alloc_page_directory(struct ttm_tt *ttm)
124{
125	ttm->pages = kvcalloc(ttm->num_pages, sizeof(*ttm->pages) +
126			      sizeof(*ttm->dma_address), GFP_KERNEL);
127	if (!ttm->pages)
128		return -ENOMEM;
129
130	ttm->dma_address = (void *)(ttm->pages + ttm->num_pages);
131
132	ttm->orders = kvmalloc_array(ttm->num_pages,
133				      sizeof(unsigned long),
134				      GFP_KERNEL | __GFP_ZERO);
135	if (!ttm->orders)
136		return -ENOMEM;
137	return 0;
138}
139
140static int ttm_sg_tt_alloc_page_directory(struct ttm_tt *ttm)
141{
142	ttm->dma_address = kvcalloc(ttm->num_pages, sizeof(*ttm->dma_address),
143				    GFP_KERNEL);
144	if (!ttm->dma_address)
145		return -ENOMEM;
146
147	return 0;
148}
149
150void ttm_tt_destroy(struct ttm_device *bdev, struct ttm_tt *ttm)
151{
152	bdev->funcs->ttm_tt_destroy(bdev, ttm);
153}
154
155static void ttm_tt_init_fields(struct ttm_tt *ttm,
156			       struct ttm_buffer_object *bo,
157			       uint32_t page_flags,
158			       enum ttm_caching caching,
159			       unsigned long extra_pages)
160{
161	ttm->num_pages = (PAGE_ALIGN(bo->base.size) >> PAGE_SHIFT) + extra_pages;
162	ttm->page_flags = page_flags;
163	ttm->dma_address = NULL;
164	ttm->swap_storage = NULL;
165	ttm->sg = bo->sg;
166	ttm->caching = caching;
167	ttm->dmat = bo->bdev->dmat;
168	ttm->map = NULL;
169	ttm->segs = NULL;
170}
171
172int ttm_tt_init(struct ttm_tt *ttm, struct ttm_buffer_object *bo,
173		uint32_t page_flags, enum ttm_caching caching,
174		unsigned long extra_pages)
175{
176	ttm_tt_init_fields(ttm, bo, page_flags, caching, extra_pages);
177
178	if (ttm_tt_alloc_page_directory(ttm)) {
179		pr_err("Failed allocating page table\n");
180		return -ENOMEM;
181	}
182	return 0;
183}
184EXPORT_SYMBOL(ttm_tt_init);
185
186void ttm_tt_fini(struct ttm_tt *ttm)
187{
188	WARN_ON(ttm->page_flags & TTM_TT_FLAG_PRIV_POPULATED);
189
190	if (ttm->swap_storage)
191		uao_detach(ttm->swap_storage);
192	ttm->swap_storage = NULL;
193
194	if (ttm->pages)
195		kvfree(ttm->pages);
196	else
197		kvfree(ttm->dma_address);
198	kvfree(ttm->orders);
199	ttm->pages = NULL;
200	ttm->dma_address = NULL;
201	ttm->orders = NULL;
202
203	if (ttm->map)
204		bus_dmamap_destroy(ttm->dmat, ttm->map);
205	if (ttm->segs)
206		km_free(ttm->segs, round_page(ttm->num_pages *
207		    sizeof(bus_dma_segment_t)), &kv_any, &kp_zero);
208}
209EXPORT_SYMBOL(ttm_tt_fini);
210
211int ttm_sg_tt_init(struct ttm_tt *ttm, struct ttm_buffer_object *bo,
212		   uint32_t page_flags, enum ttm_caching caching)
213{
214	int ret;
215	int flags = BUS_DMA_WAITOK;
216
217	ttm_tt_init_fields(ttm, bo, page_flags, caching, 0);
218
219	if (page_flags & TTM_TT_FLAG_EXTERNAL)
220		ret = ttm_sg_tt_alloc_page_directory(ttm);
221	else
222		ret = ttm_dma_tt_alloc_page_directory(ttm);
223	if (ret) {
224		pr_err("Failed allocating page table\n");
225		return -ENOMEM;
226	}
227
228	ttm->segs = km_alloc(round_page(ttm->num_pages *
229	    sizeof(bus_dma_segment_t)), &kv_any, &kp_zero, &kd_waitok);
230
231	if (bo->bdev->pool.use_dma32 == false)
232		flags |= BUS_DMA_64BIT;
233	if (bus_dmamap_create(ttm->dmat, ttm->num_pages << PAGE_SHIFT,
234	    ttm->num_pages, ttm->num_pages << PAGE_SHIFT, 0, flags,
235	    &ttm->map)) {
236		km_free(ttm->segs, round_page(ttm->num_pages *
237		    sizeof(bus_dma_segment_t)), &kv_any, &kp_zero);
238		if (ttm->pages) {
239			kvfree(ttm->pages);
240			kvfree(ttm->orders);
241		} else
242			kvfree(ttm->dma_address);
243		ttm->pages = NULL;
244		ttm->orders = NULL;
245		ttm->dma_address = NULL;
246		pr_err("Failed allocating page table\n");
247		return -ENOMEM;
248	}
249
250	return 0;
251}
252EXPORT_SYMBOL(ttm_sg_tt_init);
253
254int ttm_tt_swapin(struct ttm_tt *ttm)
255{
256	struct uvm_object *swap_storage;
257	struct vm_page *from_page;
258	struct vm_page *to_page;
259	struct pglist plist;
260	int i, ret;
261
262	swap_storage = ttm->swap_storage;
263	BUG_ON(swap_storage == NULL);
264
265	TAILQ_INIT(&plist);
266	if (uvm_obj_wire(swap_storage, 0, ttm->num_pages << PAGE_SHIFT,
267	    &plist)) {
268		ret = -ENOMEM;
269		goto out_err;
270	}
271
272	from_page = TAILQ_FIRST(&plist);
273	for (i = 0; i < ttm->num_pages; ++i) {
274		to_page = ttm->pages[i];
275		if (unlikely(to_page == NULL)) {
276			ret = -ENOMEM;
277			goto out_err;
278		}
279
280		uvm_pagecopy(from_page, to_page);
281		from_page = TAILQ_NEXT(from_page, pageq);
282	}
283
284	uvm_obj_unwire(swap_storage, 0, ttm->num_pages << PAGE_SHIFT);
285
286	uao_detach(swap_storage);
287	ttm->swap_storage = NULL;
288	ttm->page_flags &= ~TTM_TT_FLAG_SWAPPED;
289
290	return 0;
291
292out_err:
293	return ret;
294}
295
296/**
297 * ttm_tt_swapout - swap out tt object
298 *
299 * @bdev: TTM device structure.
300 * @ttm: The struct ttm_tt.
301 * @gfp_flags: Flags to use for memory allocation.
302 *
303 * Swapout a TT object to a shmem_file, return number of pages swapped out or
304 * negative error code.
305 */
306int ttm_tt_swapout(struct ttm_device *bdev, struct ttm_tt *ttm,
307		   gfp_t gfp_flags)
308{
309	STUB();
310	return -ENOSYS;
311#ifdef notyet
312	loff_t size = (loff_t)ttm->num_pages << PAGE_SHIFT;
313	struct uvm_object *swap_storage;
314	struct vm_page *from_page;
315	struct vm_page *to_page;
316	int i, ret;
317
318	swap_storage = uao_create(size, 0);
319#ifdef notyet
320	if (IS_ERR(swap_storage)) {
321		pr_err("Failed allocating swap storage\n");
322		return PTR_ERR(swap_storage);
323	}
324#endif
325
326	for (i = 0; i < ttm->num_pages; ++i) {
327		from_page = ttm->pages[i];
328		if (unlikely(from_page == NULL))
329			continue;
330
331		to_page = shmem_read_mapping_page_gfp(swap_space, i, gfp_flags);
332		if (IS_ERR(to_page)) {
333			ret = PTR_ERR(to_page);
334			goto out_err;
335		}
336		copy_highpage(to_page, from_page);
337		set_page_dirty(to_page);
338		mark_page_accessed(to_page);
339		put_page(to_page);
340	}
341
342	ttm_tt_unpopulate(bdev, ttm);
343	ttm->swap_storage = swap_storage;
344	ttm->page_flags |= TTM_TT_FLAG_SWAPPED;
345
346	return ttm->num_pages;
347
348out_err:
349	uao_detach(swap_storage);
350
351	return ret;
352#endif
353}
354
355int ttm_tt_populate(struct ttm_device *bdev,
356		    struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
357{
358	int ret;
359
360	if (!ttm)
361		return -EINVAL;
362
363	if (ttm_tt_is_populated(ttm))
364		return 0;
365
366	if (!(ttm->page_flags & TTM_TT_FLAG_EXTERNAL)) {
367		atomic_long_add(ttm->num_pages, &ttm_pages_allocated);
368		if (bdev->pool.use_dma32)
369			atomic_long_add(ttm->num_pages,
370					&ttm_dma32_pages_allocated);
371	}
372
373	while (atomic_long_read(&ttm_pages_allocated) > ttm_pages_limit ||
374	       atomic_long_read(&ttm_dma32_pages_allocated) >
375	       ttm_dma32_pages_limit) {
376
377		ret = ttm_global_swapout(ctx, GFP_KERNEL);
378		if (ret == 0)
379			break;
380		if (ret < 0)
381			goto error;
382	}
383
384	if (bdev->funcs->ttm_tt_populate)
385		ret = bdev->funcs->ttm_tt_populate(bdev, ttm, ctx);
386	else
387		ret = ttm_pool_alloc(&bdev->pool, ttm, ctx);
388	if (ret)
389		goto error;
390
391	ttm->page_flags |= TTM_TT_FLAG_PRIV_POPULATED;
392	if (unlikely(ttm->page_flags & TTM_TT_FLAG_SWAPPED)) {
393		ret = ttm_tt_swapin(ttm);
394		if (unlikely(ret != 0)) {
395			ttm_tt_unpopulate(bdev, ttm);
396			return ret;
397		}
398	}
399
400	return 0;
401
402error:
403	if (!(ttm->page_flags & TTM_TT_FLAG_EXTERNAL)) {
404		atomic_long_sub(ttm->num_pages, &ttm_pages_allocated);
405		if (bdev->pool.use_dma32)
406			atomic_long_sub(ttm->num_pages,
407					&ttm_dma32_pages_allocated);
408	}
409	return ret;
410}
411EXPORT_SYMBOL(ttm_tt_populate);
412
413void ttm_tt_unpopulate(struct ttm_device *bdev, struct ttm_tt *ttm)
414{
415	if (!ttm_tt_is_populated(ttm))
416		return;
417
418	if (bdev->funcs->ttm_tt_unpopulate)
419		bdev->funcs->ttm_tt_unpopulate(bdev, ttm);
420	else
421		ttm_pool_free(&bdev->pool, ttm);
422
423	if (!(ttm->page_flags & TTM_TT_FLAG_EXTERNAL)) {
424		atomic_long_sub(ttm->num_pages, &ttm_pages_allocated);
425		if (bdev->pool.use_dma32)
426			atomic_long_sub(ttm->num_pages,
427					&ttm_dma32_pages_allocated);
428	}
429
430	ttm->page_flags &= ~TTM_TT_FLAG_PRIV_POPULATED;
431}
432
433#ifdef CONFIG_DEBUG_FS
434
435/* Test the shrinker functions and dump the result */
436static int ttm_tt_debugfs_shrink_show(struct seq_file *m, void *data)
437{
438	struct ttm_operation_ctx ctx = { false, false };
439
440	seq_printf(m, "%d\n", ttm_global_swapout(&ctx, GFP_KERNEL));
441	return 0;
442}
443DEFINE_SHOW_ATTRIBUTE(ttm_tt_debugfs_shrink);
444
445#endif
446
447
448/*
449 * ttm_tt_mgr_init - register with the MM shrinker
450 *
451 * Register with the MM shrinker for swapping out BOs.
452 */
453void ttm_tt_mgr_init(unsigned long num_pages, unsigned long num_dma32_pages)
454{
455#ifdef CONFIG_DEBUG_FS
456	debugfs_create_file("tt_shrink", 0400, ttm_debugfs_root, NULL,
457			    &ttm_tt_debugfs_shrink_fops);
458#endif
459
460	if (!ttm_pages_limit)
461		ttm_pages_limit = num_pages;
462
463	if (!ttm_dma32_pages_limit)
464		ttm_dma32_pages_limit = num_dma32_pages;
465}
466
467static void ttm_kmap_iter_tt_map_local(struct ttm_kmap_iter *iter,
468				       struct iosys_map *dmap,
469				       pgoff_t i, bus_space_tag_t bst)
470{
471	struct ttm_kmap_iter_tt *iter_tt =
472		container_of(iter, typeof(*iter_tt), base);
473
474#ifdef __linux__
475	iosys_map_set_vaddr(dmap, kmap_local_page_prot(iter_tt->tt->pages[i],
476						       iter_tt->prot));
477#else
478	iosys_map_set_vaddr(dmap, kmap_atomic_prot(iter_tt->tt->pages[i],
479						       iter_tt->prot));
480#endif
481}
482
483static void ttm_kmap_iter_tt_unmap_local(struct ttm_kmap_iter *iter,
484					 struct iosys_map *map, bus_space_tag_t bst)
485{
486#ifdef __linux__
487	kunmap_local(map->vaddr);
488#else
489	kunmap_atomic(map->vaddr);
490#endif
491}
492
493static const struct ttm_kmap_iter_ops ttm_kmap_iter_tt_ops = {
494	.map_local = ttm_kmap_iter_tt_map_local,
495	.unmap_local = ttm_kmap_iter_tt_unmap_local,
496	.maps_tt = true,
497};
498
499/**
500 * ttm_kmap_iter_tt_init - Initialize a struct ttm_kmap_iter_tt
501 * @iter_tt: The struct ttm_kmap_iter_tt to initialize.
502 * @tt: Struct ttm_tt holding page pointers of the struct ttm_resource.
503 *
504 * Return: Pointer to the embedded struct ttm_kmap_iter.
505 */
506struct ttm_kmap_iter *
507ttm_kmap_iter_tt_init(struct ttm_kmap_iter_tt *iter_tt,
508		      struct ttm_tt *tt)
509{
510	iter_tt->base.ops = &ttm_kmap_iter_tt_ops;
511	iter_tt->tt = tt;
512	if (tt)
513		iter_tt->prot = ttm_prot_from_caching(tt->caching, PAGE_KERNEL);
514	else
515		iter_tt->prot = PAGE_KERNEL;
516
517	return &iter_tt->base;
518}
519EXPORT_SYMBOL(ttm_kmap_iter_tt_init);
520
521unsigned long ttm_tt_pages_limit(void)
522{
523	return ttm_pages_limit;
524}
525EXPORT_SYMBOL(ttm_tt_pages_limit);
526