xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h (revision 260f6f4fda93c8485c8037865c941b42b9cba5d2)
1 /*
2  * Copyright 2016 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 #ifndef __AMDGPU_TTM_H__
25 #define __AMDGPU_TTM_H__
26 
27 #include <linux/dma-direction.h>
28 #include <drm/gpu_scheduler.h>
29 #include <drm/ttm/ttm_placement.h>
30 #include "amdgpu_vram_mgr.h"
31 
32 #define AMDGPU_PL_GDS		(TTM_PL_PRIV + 0)
33 #define AMDGPU_PL_GWS		(TTM_PL_PRIV + 1)
34 #define AMDGPU_PL_OA		(TTM_PL_PRIV + 2)
35 #define AMDGPU_PL_PREEMPT	(TTM_PL_PRIV + 3)
36 #define AMDGPU_PL_DOORBELL	(TTM_PL_PRIV + 4)
37 #define __AMDGPU_PL_NUM	(TTM_PL_PRIV + 5)
38 
39 #define AMDGPU_GTT_MAX_TRANSFER_SIZE	512
40 #define AMDGPU_GTT_NUM_TRANSFER_WINDOWS	2
41 
42 extern const struct attribute_group amdgpu_vram_mgr_attr_group;
43 extern const struct attribute_group amdgpu_gtt_mgr_attr_group;
44 
45 struct hmm_range;
46 
47 struct amdgpu_gtt_mgr {
48 	struct ttm_resource_manager manager;
49 	struct drm_mm mm;
50 	spinlock_t lock;
51 };
52 
53 struct amdgpu_mman {
54 	struct ttm_device		bdev;
55 	struct ttm_pool			*ttm_pools;
56 	bool				initialized;
57 	void __iomem			*aper_base_kaddr;
58 
59 	/* buffer handling */
60 	const struct amdgpu_buffer_funcs	*buffer_funcs;
61 	struct amdgpu_ring			*buffer_funcs_ring;
62 	bool					buffer_funcs_enabled;
63 
64 	struct mutex				gtt_window_lock;
65 	/* High priority scheduler entity for buffer moves */
66 	struct drm_sched_entity			high_pr;
67 	/* Low priority scheduler entity for VRAM clearing */
68 	struct drm_sched_entity			low_pr;
69 
70 	struct amdgpu_vram_mgr vram_mgr;
71 	struct amdgpu_gtt_mgr gtt_mgr;
72 	struct ttm_resource_manager preempt_mgr;
73 
74 	uint64_t		stolen_vga_size;
75 	struct amdgpu_bo	*stolen_vga_memory;
76 	uint64_t		stolen_extended_size;
77 	struct amdgpu_bo	*stolen_extended_memory;
78 	bool			keep_stolen_vga_memory;
79 
80 	struct amdgpu_bo	*stolen_reserved_memory;
81 	uint64_t		stolen_reserved_offset;
82 	uint64_t		stolen_reserved_size;
83 
84 	/* discovery */
85 	uint8_t				*discovery_bin;
86 	uint32_t			discovery_tmr_size;
87 	/* fw reserved memory */
88 	struct amdgpu_bo		*fw_reserved_memory;
89 	struct amdgpu_bo		*fw_reserved_memory_extend;
90 
91 	/* firmware VRAM reservation */
92 	u64		fw_vram_usage_start_offset;
93 	u64		fw_vram_usage_size;
94 	struct amdgpu_bo	*fw_vram_usage_reserved_bo;
95 	void		*fw_vram_usage_va;
96 
97 	/* driver VRAM reservation */
98 	u64		drv_vram_usage_start_offset;
99 	u64		drv_vram_usage_size;
100 	struct amdgpu_bo	*drv_vram_usage_reserved_bo;
101 	void		*drv_vram_usage_va;
102 
103 	/* PAGE_SIZE'd BO for process memory r/w over SDMA. */
104 	struct amdgpu_bo	*sdma_access_bo;
105 	void			*sdma_access_ptr;
106 };
107 
108 struct amdgpu_copy_mem {
109 	struct ttm_buffer_object	*bo;
110 	struct ttm_resource		*mem;
111 	unsigned long			offset;
112 };
113 
114 #define AMDGPU_COPY_FLAGS_TMZ		(1 << 0)
115 #define AMDGPU_COPY_FLAGS_READ_DECOMPRESSED	(1 << 1)
116 #define AMDGPU_COPY_FLAGS_WRITE_COMPRESSED	(1 << 2)
117 #define AMDGPU_COPY_FLAGS_MAX_COMPRESSED_SHIFT		3
118 #define AMDGPU_COPY_FLAGS_MAX_COMPRESSED_MASK		0x03
119 #define AMDGPU_COPY_FLAGS_NUMBER_TYPE_SHIFT		5
120 #define AMDGPU_COPY_FLAGS_NUMBER_TYPE_MASK		0x07
121 #define AMDGPU_COPY_FLAGS_DATA_FORMAT_SHIFT		8
122 #define AMDGPU_COPY_FLAGS_DATA_FORMAT_MASK		0x3f
123 #define AMDGPU_COPY_FLAGS_WRITE_COMPRESS_DISABLE_SHIFT	14
124 #define AMDGPU_COPY_FLAGS_WRITE_COMPRESS_DISABLE_MASK	0x1
125 
126 #define AMDGPU_COPY_FLAGS_SET(field, value) \
127 	(((__u32)(value) & AMDGPU_COPY_FLAGS_##field##_MASK) << AMDGPU_COPY_FLAGS_##field##_SHIFT)
128 #define AMDGPU_COPY_FLAGS_GET(value, field) \
129 	(((__u32)(value) >> AMDGPU_COPY_FLAGS_##field##_SHIFT) & AMDGPU_COPY_FLAGS_##field##_MASK)
130 
131 int amdgpu_gtt_mgr_init(struct amdgpu_device *adev, uint64_t gtt_size);
132 void amdgpu_gtt_mgr_fini(struct amdgpu_device *adev);
133 int amdgpu_preempt_mgr_init(struct amdgpu_device *adev);
134 void amdgpu_preempt_mgr_fini(struct amdgpu_device *adev);
135 int amdgpu_vram_mgr_init(struct amdgpu_device *adev);
136 void amdgpu_vram_mgr_fini(struct amdgpu_device *adev);
137 
138 bool amdgpu_gtt_mgr_has_gart_addr(struct ttm_resource *mem);
139 void amdgpu_gtt_mgr_recover(struct amdgpu_gtt_mgr *mgr);
140 
141 uint64_t amdgpu_preempt_mgr_usage(struct ttm_resource_manager *man);
142 
143 u64 amdgpu_vram_mgr_bo_visible_size(struct amdgpu_bo *bo);
144 int amdgpu_vram_mgr_alloc_sgt(struct amdgpu_device *adev,
145 			      struct ttm_resource *mem,
146 			      u64 offset, u64 size,
147 			      struct device *dev,
148 			      enum dma_data_direction dir,
149 			      struct sg_table **sgt);
150 void amdgpu_vram_mgr_free_sgt(struct device *dev,
151 			      enum dma_data_direction dir,
152 			      struct sg_table *sgt);
153 uint64_t amdgpu_vram_mgr_vis_usage(struct amdgpu_vram_mgr *mgr);
154 int amdgpu_vram_mgr_reserve_range(struct amdgpu_vram_mgr *mgr,
155 				  uint64_t start, uint64_t size);
156 int amdgpu_vram_mgr_query_page_status(struct amdgpu_vram_mgr *mgr,
157 				      uint64_t start);
158 void amdgpu_vram_mgr_clear_reset_blocks(struct amdgpu_device *adev);
159 
160 bool amdgpu_res_cpu_visible(struct amdgpu_device *adev,
161 			    struct ttm_resource *res);
162 
163 int amdgpu_ttm_init(struct amdgpu_device *adev);
164 void amdgpu_ttm_fini(struct amdgpu_device *adev);
165 void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev,
166 					bool enable);
167 int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,
168 		       uint64_t dst_offset, uint32_t byte_count,
169 		       struct dma_resv *resv,
170 		       struct dma_fence **fence, bool direct_submit,
171 		       bool vm_needs_flush, uint32_t copy_flags);
172 int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev,
173 			       const struct amdgpu_copy_mem *src,
174 			       const struct amdgpu_copy_mem *dst,
175 			       uint64_t size, bool tmz,
176 			       struct dma_resv *resv,
177 			       struct dma_fence **f);
178 int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo,
179 			    struct dma_resv *resv,
180 			    struct dma_fence **fence);
181 int amdgpu_fill_buffer(struct amdgpu_bo *bo,
182 			uint32_t src_data,
183 			struct dma_resv *resv,
184 			struct dma_fence **fence,
185 			bool delayed);
186 
187 int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo);
188 void amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo);
189 uint64_t amdgpu_ttm_domain_start(struct amdgpu_device *adev, uint32_t type);
190 
191 #if IS_ENABLED(CONFIG_DRM_AMDGPU_USERPTR)
192 int amdgpu_ttm_tt_get_user_pages(struct amdgpu_bo *bo, struct page **pages,
193 				 struct hmm_range **range);
194 void amdgpu_ttm_tt_discard_user_pages(struct ttm_tt *ttm,
195 				      struct hmm_range *range);
196 bool amdgpu_ttm_tt_get_user_pages_done(struct ttm_tt *ttm,
197 				       struct hmm_range *range);
198 #else
amdgpu_ttm_tt_get_user_pages(struct amdgpu_bo * bo,struct page ** pages,struct hmm_range ** range)199 static inline int amdgpu_ttm_tt_get_user_pages(struct amdgpu_bo *bo,
200 					       struct page **pages,
201 					       struct hmm_range **range)
202 {
203 	return -EPERM;
204 }
amdgpu_ttm_tt_discard_user_pages(struct ttm_tt * ttm,struct hmm_range * range)205 static inline void amdgpu_ttm_tt_discard_user_pages(struct ttm_tt *ttm,
206 						    struct hmm_range *range)
207 {
208 }
amdgpu_ttm_tt_get_user_pages_done(struct ttm_tt * ttm,struct hmm_range * range)209 static inline bool amdgpu_ttm_tt_get_user_pages_done(struct ttm_tt *ttm,
210 						     struct hmm_range *range)
211 {
212 	return false;
213 }
214 #endif
215 
216 void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct page **pages);
217 int amdgpu_ttm_tt_get_userptr(const struct ttm_buffer_object *tbo,
218 			      uint64_t *user_addr);
219 int amdgpu_ttm_tt_set_userptr(struct ttm_buffer_object *bo,
220 			      uint64_t addr, uint32_t flags);
221 bool amdgpu_ttm_tt_has_userptr(struct ttm_tt *ttm);
222 struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm);
223 bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
224 				  unsigned long end, unsigned long *userptr);
225 bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm,
226 				       int *last_invalidated);
227 bool amdgpu_ttm_tt_is_userptr(struct ttm_tt *ttm);
228 bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm);
229 uint64_t amdgpu_ttm_tt_pde_flags(struct ttm_tt *ttm, struct ttm_resource *mem);
230 uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
231 				 struct ttm_resource *mem);
232 int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type);
233 
234 void amdgpu_ttm_debugfs_init(struct amdgpu_device *adev);
235 
236 #endif
237