1 /* SPDX-License-Identifier: GPL-2.0 OR MIT */
2 /*
3  * Copyright 2014-2022 Advanced Micro Devices, Inc.
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the "Software"),
7  * to deal in the Software without restriction, including without limitation
8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9  * and/or sell copies of the Software, and to permit persons to whom the
10  * Software is furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice shall be included in
13  * all copies or substantial portions of the Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
19  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
21  * OTHER DEALINGS IN THE SOFTWARE.
22  */
23 
24 #ifndef KFD_CRAT_H_INCLUDED
25 #define KFD_CRAT_H_INCLUDED
26 
27 #include <linux/types.h>
28 
29 #pragma pack(1)
30 
31 /*
32  * 4CC signature values for the CRAT and CDIT ACPI tables
33  */
34 
35 #define CRAT_SIGNATURE	"CRAT"
36 #define CDIT_SIGNATURE	"CDIT"
37 
38 /*
39  * Component Resource Association Table (CRAT)
40  */
41 
42 #define CRAT_OEMID_LENGTH	6
43 #define CRAT_OEMTABLEID_LENGTH	8
44 #define CRAT_RESERVED_LENGTH	6
45 
46 #define CRAT_OEMID_64BIT_MASK ((1ULL << (CRAT_OEMID_LENGTH * 8)) - 1)
47 
48 /* Compute Unit flags */
49 #define COMPUTE_UNIT_CPU	(1 << 0)  /* Create Virtual CRAT for CPU */
50 #define COMPUTE_UNIT_GPU	(1 << 1)  /* Create Virtual CRAT for GPU */
51 
52 struct crat_header {
53 	uint32_t	signature;
54 	uint32_t	length;
55 	uint8_t		revision;
56 	uint8_t		checksum;
57 	uint8_t		oem_id[CRAT_OEMID_LENGTH];
58 	uint8_t		oem_table_id[CRAT_OEMTABLEID_LENGTH];
59 	uint32_t	oem_revision;
60 	uint32_t	creator_id;
61 	uint32_t	creator_revision;
62 	uint32_t	total_entries;
63 	uint16_t	num_domains;
64 	uint8_t		reserved[CRAT_RESERVED_LENGTH];
65 };
66 
67 /*
68  * The header structure is immediately followed by total_entries of the
69  * data definitions
70  */
71 
72 /*
73  * The currently defined subtype entries in the CRAT
74  */
75 #define CRAT_SUBTYPE_COMPUTEUNIT_AFFINITY	0
76 #define CRAT_SUBTYPE_MEMORY_AFFINITY		1
77 #define CRAT_SUBTYPE_CACHE_AFFINITY		2
78 #define CRAT_SUBTYPE_TLB_AFFINITY		3
79 #define CRAT_SUBTYPE_CCOMPUTE_AFFINITY		4
80 #define CRAT_SUBTYPE_IOLINK_AFFINITY		5
81 #define CRAT_SUBTYPE_MAX			6
82 
83 #define CRAT_SIBLINGMAP_SIZE	32
84 
85 /*
86  * ComputeUnit Affinity structure and definitions
87  */
88 #define CRAT_CU_FLAGS_ENABLED		0x00000001
89 #define CRAT_CU_FLAGS_HOT_PLUGGABLE	0x00000002
90 #define CRAT_CU_FLAGS_CPU_PRESENT	0x00000004
91 #define CRAT_CU_FLAGS_GPU_PRESENT	0x00000008
92 #define CRAT_CU_FLAGS_IOMMU_PRESENT	0x00000010
93 #define CRAT_CU_FLAGS_RESERVED		0xffffffe0
94 
95 #define CRAT_COMPUTEUNIT_RESERVED_LENGTH 4
96 
97 struct crat_subtype_computeunit {
98 	uint8_t		type;
99 	uint8_t		length;
100 	uint16_t	reserved;
101 	uint32_t	flags;
102 	uint32_t	proximity_domain;
103 	uint32_t	processor_id_low;
104 	uint16_t	num_cpu_cores;
105 	uint16_t	num_simd_cores;
106 	uint16_t	max_waves_simd;
107 	uint16_t	io_count;
108 	uint16_t	hsa_capability;
109 	uint16_t	lds_size_in_kb;
110 	uint8_t		wave_front_size;
111 	uint8_t		num_banks;
112 	uint16_t	micro_engine_id;
113 	uint8_t		array_count;
114 	uint8_t		num_cu_per_array;
115 	uint8_t		num_simd_per_cu;
116 	uint8_t		max_slots_scatch_cu;
117 	uint8_t		reserved2[CRAT_COMPUTEUNIT_RESERVED_LENGTH];
118 };
119 
120 /*
121  * HSA Memory Affinity structure and definitions
122  */
123 #define CRAT_MEM_FLAGS_ENABLED		0x00000001
124 #define CRAT_MEM_FLAGS_HOT_PLUGGABLE	0x00000002
125 #define CRAT_MEM_FLAGS_NON_VOLATILE	0x00000004
126 #define CRAT_MEM_FLAGS_RESERVED		0xfffffff8
127 
128 #define CRAT_MEMORY_RESERVED_LENGTH 8
129 
130 struct crat_subtype_memory {
131 	uint8_t		type;
132 	uint8_t		length;
133 	uint16_t	reserved;
134 	uint32_t	flags;
135 	uint32_t	proximity_domain;
136 	uint32_t	base_addr_low;
137 	uint32_t	base_addr_high;
138 	uint32_t	length_low;
139 	uint32_t	length_high;
140 	uint32_t	width;
141 	uint8_t		visibility_type; /* for virtual (dGPU) CRAT */
142 	uint8_t		reserved2[CRAT_MEMORY_RESERVED_LENGTH - 1];
143 };
144 
145 /*
146  * HSA Cache Affinity structure and definitions
147  */
148 #define CRAT_CACHE_FLAGS_ENABLED	0x00000001
149 #define CRAT_CACHE_FLAGS_DATA_CACHE	0x00000002
150 #define CRAT_CACHE_FLAGS_INST_CACHE	0x00000004
151 #define CRAT_CACHE_FLAGS_CPU_CACHE	0x00000008
152 #define CRAT_CACHE_FLAGS_SIMD_CACHE	0x00000010
153 #define CRAT_CACHE_FLAGS_RESERVED	0xffffffe0
154 
155 #define CRAT_CACHE_RESERVED_LENGTH 8
156 
157 struct crat_subtype_cache {
158 	uint8_t		type;
159 	uint8_t		length;
160 	uint16_t	reserved;
161 	uint32_t	flags;
162 	uint32_t	processor_id_low;
163 	uint8_t		sibling_map[CRAT_SIBLINGMAP_SIZE];
164 	uint32_t	cache_size;
165 	uint8_t		cache_level;
166 	uint8_t		lines_per_tag;
167 	uint16_t	cache_line_size;
168 	uint8_t		associativity;
169 	uint8_t		cache_properties;
170 	uint16_t	cache_latency;
171 	uint8_t		reserved2[CRAT_CACHE_RESERVED_LENGTH];
172 };
173 
174 /*
175  * HSA TLB Affinity structure and definitions
176  */
177 #define CRAT_TLB_FLAGS_ENABLED	0x00000001
178 #define CRAT_TLB_FLAGS_DATA_TLB	0x00000002
179 #define CRAT_TLB_FLAGS_INST_TLB	0x00000004
180 #define CRAT_TLB_FLAGS_CPU_TLB	0x00000008
181 #define CRAT_TLB_FLAGS_SIMD_TLB	0x00000010
182 #define CRAT_TLB_FLAGS_RESERVED	0xffffffe0
183 
184 #define CRAT_TLB_RESERVED_LENGTH 4
185 
186 struct crat_subtype_tlb {
187 	uint8_t		type;
188 	uint8_t		length;
189 	uint16_t	reserved;
190 	uint32_t	flags;
191 	uint32_t	processor_id_low;
192 	uint8_t		sibling_map[CRAT_SIBLINGMAP_SIZE];
193 	uint32_t	tlb_level;
194 	uint8_t		data_tlb_associativity_2mb;
195 	uint8_t		data_tlb_size_2mb;
196 	uint8_t		instruction_tlb_associativity_2mb;
197 	uint8_t		instruction_tlb_size_2mb;
198 	uint8_t		data_tlb_associativity_4k;
199 	uint8_t		data_tlb_size_4k;
200 	uint8_t		instruction_tlb_associativity_4k;
201 	uint8_t		instruction_tlb_size_4k;
202 	uint8_t		data_tlb_associativity_1gb;
203 	uint8_t		data_tlb_size_1gb;
204 	uint8_t		instruction_tlb_associativity_1gb;
205 	uint8_t		instruction_tlb_size_1gb;
206 	uint8_t		reserved2[CRAT_TLB_RESERVED_LENGTH];
207 };
208 
209 /*
210  * HSA CCompute/APU Affinity structure and definitions
211  */
212 #define CRAT_CCOMPUTE_FLAGS_ENABLED	0x00000001
213 #define CRAT_CCOMPUTE_FLAGS_RESERVED	0xfffffffe
214 
215 #define CRAT_CCOMPUTE_RESERVED_LENGTH 16
216 
217 struct crat_subtype_ccompute {
218 	uint8_t		type;
219 	uint8_t		length;
220 	uint16_t	reserved;
221 	uint32_t	flags;
222 	uint32_t	processor_id_low;
223 	uint8_t		sibling_map[CRAT_SIBLINGMAP_SIZE];
224 	uint32_t	apu_size;
225 	uint8_t		reserved2[CRAT_CCOMPUTE_RESERVED_LENGTH];
226 };
227 
228 /*
229  * HSA IO Link Affinity structure and definitions
230  */
231 #define CRAT_IOLINK_FLAGS_ENABLED		(1 << 0)
232 #define CRAT_IOLINK_FLAGS_NON_COHERENT		(1 << 1)
233 #define CRAT_IOLINK_FLAGS_NO_ATOMICS_32_BIT	(1 << 2)
234 #define CRAT_IOLINK_FLAGS_NO_ATOMICS_64_BIT	(1 << 3)
235 #define CRAT_IOLINK_FLAGS_NO_PEER_TO_PEER_DMA	(1 << 4)
236 #define CRAT_IOLINK_FLAGS_BI_DIRECTIONAL	(1 << 31)
237 #define CRAT_IOLINK_FLAGS_RESERVED_MASK		0x7fffffe0
238 
239 /*
240  * IO interface types
241  */
242 #define CRAT_IOLINK_TYPE_UNDEFINED	0
243 #define CRAT_IOLINK_TYPE_HYPERTRANSPORT	1
244 #define CRAT_IOLINK_TYPE_PCIEXPRESS	2
245 #define CRAT_IOLINK_TYPE_AMBA		3
246 #define CRAT_IOLINK_TYPE_MIPI		4
247 #define CRAT_IOLINK_TYPE_QPI_1_1	5
248 #define CRAT_IOLINK_TYPE_RESERVED1	6
249 #define CRAT_IOLINK_TYPE_RESERVED2	7
250 #define CRAT_IOLINK_TYPE_RAPID_IO	8
251 #define CRAT_IOLINK_TYPE_INFINIBAND	9
252 #define CRAT_IOLINK_TYPE_RESERVED3	10
253 #define CRAT_IOLINK_TYPE_XGMI		11
254 #define CRAT_IOLINK_TYPE_XGOP		12
255 #define CRAT_IOLINK_TYPE_GZ		13
256 #define CRAT_IOLINK_TYPE_ETHERNET_RDMA	14
257 #define CRAT_IOLINK_TYPE_RDMA_OTHER	15
258 #define CRAT_IOLINK_TYPE_OTHER		16
259 #define CRAT_IOLINK_TYPE_MAX		255
260 
261 #define CRAT_IOLINK_RESERVED_LENGTH	24
262 
263 struct crat_subtype_iolink {
264 	uint8_t		type;
265 	uint8_t		length;
266 	uint16_t	reserved;
267 	uint32_t	flags;
268 	uint32_t	proximity_domain_from;
269 	uint32_t	proximity_domain_to;
270 	uint8_t		io_interface_type;
271 	uint8_t		version_major;
272 	uint16_t	version_minor;
273 	uint32_t	minimum_latency;
274 	uint32_t	maximum_latency;
275 	uint32_t	minimum_bandwidth_mbs;
276 	uint32_t	maximum_bandwidth_mbs;
277 	uint32_t	recommended_transfer_size;
278 	uint8_t		reserved2[CRAT_IOLINK_RESERVED_LENGTH - 1];
279 	uint8_t		num_hops_xgmi;
280 };
281 
282 /*
283  * HSA generic sub-type header
284  */
285 
286 #define CRAT_SUBTYPE_FLAGS_ENABLED 0x00000001
287 
288 struct crat_subtype_generic {
289 	uint8_t		type;
290 	uint8_t		length;
291 	uint16_t	reserved;
292 	uint32_t	flags;
293 };
294 
295 /*
296  * Component Locality Distance Information Table (CDIT)
297  */
298 #define CDIT_OEMID_LENGTH	6
299 #define CDIT_OEMTABLEID_LENGTH	8
300 
301 struct cdit_header {
302 	uint32_t	signature;
303 	uint32_t	length;
304 	uint8_t		revision;
305 	uint8_t		checksum;
306 	uint8_t		oem_id[CDIT_OEMID_LENGTH];
307 	uint8_t		oem_table_id[CDIT_OEMTABLEID_LENGTH];
308 	uint32_t	oem_revision;
309 	uint32_t	creator_id;
310 	uint32_t	creator_revision;
311 	uint32_t	total_entries;
312 	uint16_t	num_domains;
313 	uint8_t		entry[1];
314 };
315 
316 #pragma pack()
317 
318 struct kfd_dev;
319 
320 int kfd_create_crat_image_acpi(void **crat_image, size_t *size);
321 void kfd_destroy_crat_image(void *crat_image);
322 int kfd_parse_crat_table(void *crat_image, struct list_head *device_list,
323 			 uint32_t proximity_domain);
324 int kfd_create_crat_image_virtual(void **crat_image, size_t *size,
325 				  int flags, struct kfd_dev *kdev,
326 				  uint32_t proximity_domain);
327 
328 #endif /* KFD_CRAT_H_INCLUDED */
329