xref: /linux/drivers/gpu/drm/i915/gem/i915_gem_internal.c (revision 6fdcba32711044c35c0e1b094cbd8f3f0b4472c9)
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 <drm/i915_drm.h>
12 
13 #include "i915_drv.h"
14 #include "i915_gem.h"
15 #include "i915_gem_object.h"
16 #include "i915_scatterlist.h"
17 #include "i915_utils.h"
18 
19 #define QUIET (__GFP_NORETRY | __GFP_NOWARN)
20 #define MAYFAIL (__GFP_RETRY_MAYFAIL | __GFP_NOWARN)
21 
22 static void internal_free_pages(struct sg_table *st)
23 {
24 	struct scatterlist *sg;
25 
26 	for (sg = st->sgl; sg; sg = __sg_next(sg)) {
27 		if (sg_page(sg))
28 			__free_pages(sg_page(sg), get_order(sg->length));
29 	}
30 
31 	sg_free_table(st);
32 	kfree(st);
33 }
34 
35 static int i915_gem_object_get_pages_internal(struct drm_i915_gem_object *obj)
36 {
37 	struct drm_i915_private *i915 = to_i915(obj->base.dev);
38 	struct sg_table *st;
39 	struct scatterlist *sg;
40 	unsigned int sg_page_sizes;
41 	unsigned int npages;
42 	int max_order;
43 	gfp_t gfp;
44 
45 	max_order = MAX_ORDER;
46 #ifdef CONFIG_SWIOTLB
47 	if (swiotlb_nr_tbl()) {
48 		unsigned int max_segment;
49 
50 		max_segment = swiotlb_max_segment();
51 		if (max_segment) {
52 			max_segment = max_t(unsigned int, max_segment,
53 					    PAGE_SIZE) >> PAGE_SHIFT;
54 			max_order = min(max_order, ilog2(max_segment));
55 		}
56 	}
57 #endif
58 
59 	gfp = GFP_KERNEL | __GFP_HIGHMEM | __GFP_RECLAIMABLE;
60 	if (IS_I965GM(i915) || IS_I965G(i915)) {
61 		/* 965gm cannot relocate objects above 4GiB. */
62 		gfp &= ~__GFP_HIGHMEM;
63 		gfp |= __GFP_DMA32;
64 	}
65 
66 create_st:
67 	st = kmalloc(sizeof(*st), GFP_KERNEL);
68 	if (!st)
69 		return -ENOMEM;
70 
71 	npages = obj->base.size / PAGE_SIZE;
72 	if (sg_alloc_table(st, npages, GFP_KERNEL)) {
73 		kfree(st);
74 		return -ENOMEM;
75 	}
76 
77 	sg = st->sgl;
78 	st->nents = 0;
79 	sg_page_sizes = 0;
80 
81 	do {
82 		int order = min(fls(npages) - 1, max_order);
83 		struct page *page;
84 
85 		do {
86 			page = alloc_pages(gfp | (order ? QUIET : MAYFAIL),
87 					   order);
88 			if (page)
89 				break;
90 			if (!order--)
91 				goto err;
92 
93 			/* Limit subsequent allocations as well */
94 			max_order = order;
95 		} while (1);
96 
97 		sg_set_page(sg, page, PAGE_SIZE << order, 0);
98 		sg_page_sizes |= PAGE_SIZE << order;
99 		st->nents++;
100 
101 		npages -= 1 << order;
102 		if (!npages) {
103 			sg_mark_end(sg);
104 			break;
105 		}
106 
107 		sg = __sg_next(sg);
108 	} while (1);
109 
110 	if (i915_gem_gtt_prepare_pages(obj, st)) {
111 		/* Failed to dma-map try again with single page sg segments */
112 		if (get_order(st->sgl->length)) {
113 			internal_free_pages(st);
114 			max_order = 0;
115 			goto create_st;
116 		}
117 		goto err;
118 	}
119 
120 	__i915_gem_object_set_pages(obj, st, sg_page_sizes);
121 
122 	return 0;
123 
124 err:
125 	sg_set_page(sg, NULL, 0, 0);
126 	sg_mark_end(sg);
127 	internal_free_pages(st);
128 
129 	return -ENOMEM;
130 }
131 
132 static void i915_gem_object_put_pages_internal(struct drm_i915_gem_object *obj,
133 					       struct sg_table *pages)
134 {
135 	i915_gem_gtt_finish_pages(obj, pages);
136 	internal_free_pages(pages);
137 
138 	obj->mm.dirty = false;
139 }
140 
141 static const struct drm_i915_gem_object_ops i915_gem_object_internal_ops = {
142 	.flags = I915_GEM_OBJECT_HAS_STRUCT_PAGE |
143 		 I915_GEM_OBJECT_IS_SHRINKABLE,
144 	.get_pages = i915_gem_object_get_pages_internal,
145 	.put_pages = i915_gem_object_put_pages_internal,
146 };
147 
148 /**
149  * i915_gem_object_create_internal: create an object with volatile pages
150  * @i915: the i915 device
151  * @size: the size in bytes of backing storage to allocate for the object
152  *
153  * Creates a new object that wraps some internal memory for private use.
154  * This object is not backed by swappable storage, and as such its contents
155  * are volatile and only valid whilst pinned. If the object is reaped by the
156  * shrinker, its pages and data will be discarded. Equally, it is not a full
157  * GEM object and so not valid for access from userspace. This makes it useful
158  * for hardware interfaces like ringbuffers (which are pinned from the time
159  * the request is written to the time the hardware stops accessing it), but
160  * not for contexts (which need to be preserved when not active for later
161  * reuse). Note that it is not cleared upon allocation.
162  */
163 struct drm_i915_gem_object *
164 i915_gem_object_create_internal(struct drm_i915_private *i915,
165 				phys_addr_t size)
166 {
167 	static struct lock_class_key lock_class;
168 	struct drm_i915_gem_object *obj;
169 	unsigned int cache_level;
170 
171 	GEM_BUG_ON(!size);
172 	GEM_BUG_ON(!IS_ALIGNED(size, PAGE_SIZE));
173 
174 	if (overflows_type(size, obj->base.size))
175 		return ERR_PTR(-E2BIG);
176 
177 	obj = i915_gem_object_alloc();
178 	if (!obj)
179 		return ERR_PTR(-ENOMEM);
180 
181 	drm_gem_private_object_init(&i915->drm, &obj->base, size);
182 	i915_gem_object_init(obj, &i915_gem_object_internal_ops, &lock_class);
183 
184 	/*
185 	 * Mark the object as volatile, such that the pages are marked as
186 	 * dontneed whilst they are still pinned. As soon as they are unpinned
187 	 * they are allowed to be reaped by the shrinker, and the caller is
188 	 * expected to repopulate - the contents of this object are only valid
189 	 * whilst active and pinned.
190 	 */
191 	i915_gem_object_set_volatile(obj);
192 
193 	obj->read_domains = I915_GEM_DOMAIN_CPU;
194 	obj->write_domain = I915_GEM_DOMAIN_CPU;
195 
196 	cache_level = HAS_LLC(i915) ? I915_CACHE_LLC : I915_CACHE_NONE;
197 	i915_gem_object_set_cache_coherency(obj, cache_level);
198 
199 	return obj;
200 }
201