1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #ifdef HAVE_CONFIG_H
25 #include "config.h"
26 #endif
27 
28 #include <stdlib.h>
29 #include <string.h>
30 #include <errno.h>
31 #include "amdgpu.h"
32 #include "amdgpu_drm.h"
33 #include "amdgpu_internal.h"
34 #include "util_math.h"
35 
amdgpu_va_range_query(amdgpu_device_handle dev,enum amdgpu_gpu_va_range type,uint64_t * start,uint64_t * end)36 int amdgpu_va_range_query(amdgpu_device_handle dev,
37 			  enum amdgpu_gpu_va_range type, uint64_t *start, uint64_t *end)
38 {
39 	if (type == amdgpu_gpu_va_range_general) {
40 		*start = dev->dev_info.virtual_address_offset;
41 		*end = dev->dev_info.virtual_address_max;
42 		return 0;
43 	}
44 	return -EINVAL;
45 }
46 
amdgpu_vamgr_init(struct amdgpu_bo_va_mgr * mgr,uint64_t start,uint64_t max,uint64_t alignment)47 drm_private void amdgpu_vamgr_init(struct amdgpu_bo_va_mgr *mgr, uint64_t start,
48 			      uint64_t max, uint64_t alignment)
49 {
50 	mgr->va_offset = start;
51 	mgr->va_max = max;
52 	mgr->va_alignment = alignment;
53 
54 	list_inithead(&mgr->va_holes);
55 	pthread_mutex_init(&mgr->bo_va_mutex, NULL);
56 }
57 
amdgpu_vamgr_deinit(struct amdgpu_bo_va_mgr * mgr)58 drm_private void amdgpu_vamgr_deinit(struct amdgpu_bo_va_mgr *mgr)
59 {
60 	struct amdgpu_bo_va_hole *hole, *tmp;
61 	LIST_FOR_EACH_ENTRY_SAFE(hole, tmp, &mgr->va_holes, list) {
62 		list_del(&hole->list);
63 		free(hole);
64 	}
65 	pthread_mutex_destroy(&mgr->bo_va_mutex);
66 }
67 
68 drm_private uint64_t
amdgpu_vamgr_find_va(struct amdgpu_bo_va_mgr * mgr,uint64_t size,uint64_t alignment,uint64_t base_required)69 amdgpu_vamgr_find_va(struct amdgpu_bo_va_mgr *mgr, uint64_t size,
70 		     uint64_t alignment, uint64_t base_required)
71 {
72 	struct amdgpu_bo_va_hole *hole, *n;
73 	uint64_t offset = 0, waste = 0;
74 
75 	alignment = MAX2(alignment, mgr->va_alignment);
76 	size = ALIGN(size, mgr->va_alignment);
77 
78 	if (base_required % alignment)
79 		return AMDGPU_INVALID_VA_ADDRESS;
80 
81 	pthread_mutex_lock(&mgr->bo_va_mutex);
82 	/* TODO: using more appropriate way to track the holes */
83 	/* first look for a hole */
84 	LIST_FOR_EACH_ENTRY_SAFE(hole, n, &mgr->va_holes, list) {
85 		if (base_required) {
86 			if(hole->offset > base_required ||
87 				(hole->offset + hole->size) < (base_required + size))
88 				continue;
89 			waste = base_required - hole->offset;
90 			offset = base_required;
91 		} else {
92 			offset = hole->offset;
93 			waste = offset % alignment;
94 			waste = waste ? alignment - waste : 0;
95 			offset += waste;
96 			if (offset >= (hole->offset + hole->size)) {
97 				continue;
98 			}
99 		}
100 		if (!waste && hole->size == size) {
101 			offset = hole->offset;
102 			list_del(&hole->list);
103 			free(hole);
104 			pthread_mutex_unlock(&mgr->bo_va_mutex);
105 			return offset;
106 		}
107 		if ((hole->size - waste) > size) {
108 			if (waste) {
109 				n = calloc(1, sizeof(struct amdgpu_bo_va_hole));
110 				n->size = waste;
111 				n->offset = hole->offset;
112 				list_add(&n->list, &hole->list);
113 			}
114 			hole->size -= (size + waste);
115 			hole->offset += size + waste;
116 			pthread_mutex_unlock(&mgr->bo_va_mutex);
117 			return offset;
118 		}
119 		if ((hole->size - waste) == size) {
120 			hole->size = waste;
121 			pthread_mutex_unlock(&mgr->bo_va_mutex);
122 			return offset;
123 		}
124 	}
125 
126 	if (base_required) {
127 		if (base_required < mgr->va_offset) {
128 			pthread_mutex_unlock(&mgr->bo_va_mutex);
129 			return AMDGPU_INVALID_VA_ADDRESS;
130 		}
131 		offset = mgr->va_offset;
132 		waste = base_required - mgr->va_offset;
133 	} else {
134 		offset = mgr->va_offset;
135 		waste = offset % alignment;
136 		waste = waste ? alignment - waste : 0;
137 	}
138 
139 	if (offset + waste + size > mgr->va_max) {
140 		pthread_mutex_unlock(&mgr->bo_va_mutex);
141 		return AMDGPU_INVALID_VA_ADDRESS;
142 	}
143 
144 	if (waste) {
145 		n = calloc(1, sizeof(struct amdgpu_bo_va_hole));
146 		n->size = waste;
147 		n->offset = offset;
148 		list_add(&n->list, &mgr->va_holes);
149 	}
150 
151 	offset += waste;
152 	mgr->va_offset += size + waste;
153 	pthread_mutex_unlock(&mgr->bo_va_mutex);
154 	return offset;
155 }
156 
157 drm_private void
amdgpu_vamgr_free_va(struct amdgpu_bo_va_mgr * mgr,uint64_t va,uint64_t size)158 amdgpu_vamgr_free_va(struct amdgpu_bo_va_mgr *mgr, uint64_t va, uint64_t size)
159 {
160 	struct amdgpu_bo_va_hole *hole;
161 
162 	if (va == AMDGPU_INVALID_VA_ADDRESS)
163 		return;
164 
165 	size = ALIGN(size, mgr->va_alignment);
166 
167 	pthread_mutex_lock(&mgr->bo_va_mutex);
168 	if ((va + size) == mgr->va_offset) {
169 		mgr->va_offset = va;
170 		/* Delete uppermost hole if it reaches the new top */
171 		if (!LIST_IS_EMPTY(&mgr->va_holes)) {
172 			hole = container_of(mgr->va_holes.next, hole, list);
173 			if ((hole->offset + hole->size) == va) {
174 				mgr->va_offset = hole->offset;
175 				list_del(&hole->list);
176 				free(hole);
177 			}
178 		}
179 	} else {
180 		struct amdgpu_bo_va_hole *next;
181 
182 		hole = container_of(&mgr->va_holes, hole, list);
183 		LIST_FOR_EACH_ENTRY(next, &mgr->va_holes, list) {
184 			if (next->offset < va)
185 				break;
186 			hole = next;
187 		}
188 
189 		if (&hole->list != &mgr->va_holes) {
190 			/* Grow upper hole if it's adjacent */
191 			if (hole->offset == (va + size)) {
192 				hole->offset = va;
193 				hole->size += size;
194 				/* Merge lower hole if it's adjacent */
195 				if (next != hole
196 						&& &next->list != &mgr->va_holes
197 						&& (next->offset + next->size) == va) {
198 					next->size += hole->size;
199 					list_del(&hole->list);
200 					free(hole);
201 				}
202 				goto out;
203 			}
204 		}
205 
206 		/* Grow lower hole if it's adjacent */
207 		if (next != hole && &next->list != &mgr->va_holes &&
208 				(next->offset + next->size) == va) {
209 			next->size += size;
210 			goto out;
211 		}
212 
213 		/* FIXME on allocation failure we just lose virtual address space
214 		 * maybe print a warning
215 		 */
216 		next = calloc(1, sizeof(struct amdgpu_bo_va_hole));
217 		if (next) {
218 			next->size = size;
219 			next->offset = va;
220 			list_add(&next->list, &hole->list);
221 		}
222 	}
223 out:
224 	pthread_mutex_unlock(&mgr->bo_va_mutex);
225 }
226 
amdgpu_va_range_alloc(amdgpu_device_handle dev,enum amdgpu_gpu_va_range va_range_type,uint64_t size,uint64_t va_base_alignment,uint64_t va_base_required,uint64_t * va_base_allocated,amdgpu_va_handle * va_range_handle,uint64_t flags)227 int amdgpu_va_range_alloc(amdgpu_device_handle dev,
228 			  enum amdgpu_gpu_va_range va_range_type,
229 			  uint64_t size,
230 			  uint64_t va_base_alignment,
231 			  uint64_t va_base_required,
232 			  uint64_t *va_base_allocated,
233 			  amdgpu_va_handle *va_range_handle,
234 			  uint64_t flags)
235 {
236 	struct amdgpu_bo_va_mgr *vamgr;
237 
238 	if (flags & AMDGPU_VA_RANGE_32_BIT)
239 		vamgr = dev->vamgr_32;
240 	else
241 		vamgr = dev->vamgr;
242 
243 	va_base_alignment = MAX2(va_base_alignment, vamgr->va_alignment);
244 	size = ALIGN(size, vamgr->va_alignment);
245 
246 	*va_base_allocated = amdgpu_vamgr_find_va(vamgr, size,
247 					va_base_alignment, va_base_required);
248 
249 	if (!(flags & AMDGPU_VA_RANGE_32_BIT) &&
250 	    (*va_base_allocated == AMDGPU_INVALID_VA_ADDRESS)) {
251 		/* fallback to 32bit address */
252 		vamgr = dev->vamgr_32;
253 		*va_base_allocated = amdgpu_vamgr_find_va(vamgr, size,
254 					va_base_alignment, va_base_required);
255 	}
256 
257 	if (*va_base_allocated != AMDGPU_INVALID_VA_ADDRESS) {
258 		struct amdgpu_va* va;
259 		va = calloc(1, sizeof(struct amdgpu_va));
260 		if(!va){
261 			amdgpu_vamgr_free_va(vamgr, *va_base_allocated, size);
262 			return -ENOMEM;
263 		}
264 		va->dev = dev;
265 		va->address = *va_base_allocated;
266 		va->size = size;
267 		va->range = va_range_type;
268 		va->vamgr = vamgr;
269 		*va_range_handle = va;
270 	} else {
271 		return -EINVAL;
272 	}
273 
274 	return 0;
275 }
276 
amdgpu_va_range_free(amdgpu_va_handle va_range_handle)277 int amdgpu_va_range_free(amdgpu_va_handle va_range_handle)
278 {
279 	if(!va_range_handle || !va_range_handle->address)
280 		return 0;
281 
282 	amdgpu_vamgr_free_va(va_range_handle->vamgr,
283 			va_range_handle->address,
284 			va_range_handle->size);
285 	free(va_range_handle);
286 	return 0;
287 }
288