1 /*
2 * SPDX-License-Identifier: MIT
3 *
4 * Copyright © 2014-2016 Intel Corporation
5 */
6
7 #include <linux/scatterlist.h>
8 #include <linux/slab.h>
9 #include <linux/swiotlb.h>
10
11 #include "i915_drv.h"
12 #include "i915_gem.h"
13 #include "i915_gem_object.h"
14 #include "i915_scatterlist.h"
15 #include "i915_utils.h"
16
17 #define QUIET (__GFP_NORETRY | __GFP_NOWARN)
18 #define MAYFAIL (__GFP_RETRY_MAYFAIL | __GFP_NOWARN)
19
internal_free_pages(struct sg_table * st)20 static void internal_free_pages(struct sg_table *st)
21 {
22 struct scatterlist *sg;
23
24 for (sg = st->sgl; sg; sg = __sg_next(sg)) {
25 if (sg_page(sg))
26 __free_pages(sg_page(sg), get_order(sg->length));
27 }
28
29 sg_free_table(st);
30 kfree(st);
31 }
32
i915_gem_object_get_pages_internal(struct drm_i915_gem_object * obj)33 static int i915_gem_object_get_pages_internal(struct drm_i915_gem_object *obj)
34 {
35 struct drm_i915_private *i915 = to_i915(obj->base.dev);
36 struct sg_table *st;
37 struct scatterlist *sg;
38 unsigned int sg_page_sizes;
39 unsigned int npages;
40 int max_order;
41 gfp_t gfp;
42
43 max_order = MAX_ORDER;
44 #ifdef CONFIG_SWIOTLB
45 if (is_swiotlb_active(obj->base.dev->dev)) {
46 unsigned int max_segment;
47
48 max_segment = swiotlb_max_segment();
49 if (max_segment) {
50 max_segment = max_t(unsigned int, max_segment,
51 PAGE_SIZE) >> PAGE_SHIFT;
52 max_order = min(max_order, ilog2(max_segment));
53 }
54 }
55 #endif
56
57 gfp = GFP_KERNEL | __GFP_HIGHMEM | __GFP_RECLAIMABLE;
58 if (IS_I965GM(i915) || IS_I965G(i915)) {
59 /* 965gm cannot relocate objects above 4GiB. */
60 gfp &= ~__GFP_HIGHMEM;
61 gfp |= __GFP_DMA32;
62 }
63
64 create_st:
65 st = kmalloc(sizeof(*st), GFP_KERNEL);
66 if (!st)
67 return -ENOMEM;
68
69 npages = obj->base.size / PAGE_SIZE;
70 if (sg_alloc_table(st, npages, GFP_KERNEL)) {
71 kfree(st);
72 return -ENOMEM;
73 }
74
75 sg = st->sgl;
76 st->nents = 0;
77 sg_page_sizes = 0;
78
79 do {
80 int order = min(fls(npages) - 1, max_order);
81 struct page *page;
82
83 do {
84 page = alloc_pages(gfp | (order ? QUIET : MAYFAIL),
85 order);
86 if (page)
87 break;
88 if (!order--)
89 goto err;
90
91 /* Limit subsequent allocations as well */
92 max_order = order;
93 } while (1);
94
95 sg_set_page(sg, page, PAGE_SIZE << order, 0);
96 sg_page_sizes |= PAGE_SIZE << order;
97 st->nents++;
98
99 npages -= 1 << order;
100 if (!npages) {
101 sg_mark_end(sg);
102 break;
103 }
104
105 sg = __sg_next(sg);
106 } while (1);
107
108 if (i915_gem_gtt_prepare_pages(obj, st)) {
109 /* Failed to dma-map try again with single page sg segments */
110 if (get_order(st->sgl->length)) {
111 internal_free_pages(st);
112 max_order = 0;
113 goto create_st;
114 }
115 goto err;
116 }
117
118 __i915_gem_object_set_pages(obj, st, sg_page_sizes);
119
120 return 0;
121
122 err:
123 sg_set_page(sg, NULL, 0, 0);
124 sg_mark_end(sg);
125 internal_free_pages(st);
126
127 return -ENOMEM;
128 }
129
i915_gem_object_put_pages_internal(struct drm_i915_gem_object * obj,struct sg_table * pages)130 static void i915_gem_object_put_pages_internal(struct drm_i915_gem_object *obj,
131 struct sg_table *pages)
132 {
133 i915_gem_gtt_finish_pages(obj, pages);
134 internal_free_pages(pages);
135
136 obj->mm.dirty = false;
137 }
138
139 static const struct drm_i915_gem_object_ops i915_gem_object_internal_ops = {
140 .name = "i915_gem_object_internal",
141 .flags = I915_GEM_OBJECT_IS_SHRINKABLE,
142 .get_pages = i915_gem_object_get_pages_internal,
143 .put_pages = i915_gem_object_put_pages_internal,
144 };
145
146 struct drm_i915_gem_object *
__i915_gem_object_create_internal(struct drm_i915_private * i915,const struct drm_i915_gem_object_ops * ops,phys_addr_t size)147 __i915_gem_object_create_internal(struct drm_i915_private *i915,
148 const struct drm_i915_gem_object_ops *ops,
149 phys_addr_t size)
150 {
151 static struct lock_class_key lock_class;
152 struct drm_i915_gem_object *obj;
153 unsigned int cache_level;
154
155 GEM_BUG_ON(!size);
156 GEM_BUG_ON(!IS_ALIGNED(size, PAGE_SIZE));
157
158 if (overflows_type(size, obj->base.size))
159 return ERR_PTR(-E2BIG);
160
161 obj = i915_gem_object_alloc();
162 if (!obj)
163 return ERR_PTR(-ENOMEM);
164
165 drm_gem_private_object_init(&i915->drm, &obj->base, size);
166 i915_gem_object_init(obj, ops, &lock_class, 0);
167 obj->mem_flags |= I915_BO_FLAG_STRUCT_PAGE;
168
169 /*
170 * Mark the object as volatile, such that the pages are marked as
171 * dontneed whilst they are still pinned. As soon as they are unpinned
172 * they are allowed to be reaped by the shrinker, and the caller is
173 * expected to repopulate - the contents of this object are only valid
174 * whilst active and pinned.
175 */
176 i915_gem_object_set_volatile(obj);
177
178 obj->read_domains = I915_GEM_DOMAIN_CPU;
179 obj->write_domain = I915_GEM_DOMAIN_CPU;
180
181 cache_level = HAS_LLC(i915) ? I915_CACHE_LLC : I915_CACHE_NONE;
182 i915_gem_object_set_cache_coherency(obj, cache_level);
183
184 return obj;
185 }
186
187 /**
188 * i915_gem_object_create_internal: create an object with volatile pages
189 * @i915: the i915 device
190 * @size: the size in bytes of backing storage to allocate for the object
191 *
192 * Creates a new object that wraps some internal memory for private use.
193 * This object is not backed by swappable storage, and as such its contents
194 * are volatile and only valid whilst pinned. If the object is reaped by the
195 * shrinker, its pages and data will be discarded. Equally, it is not a full
196 * GEM object and so not valid for access from userspace. This makes it useful
197 * for hardware interfaces like ringbuffers (which are pinned from the time
198 * the request is written to the time the hardware stops accessing it), but
199 * not for contexts (which need to be preserved when not active for later
200 * reuse). Note that it is not cleared upon allocation.
201 */
202 struct drm_i915_gem_object *
i915_gem_object_create_internal(struct drm_i915_private * i915,phys_addr_t size)203 i915_gem_object_create_internal(struct drm_i915_private *i915,
204 phys_addr_t size)
205 {
206 return __i915_gem_object_create_internal(i915, &i915_gem_object_internal_ops, size);
207 }
208