1 /*
2 * Copyright 2022 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 #include "amdgpu.h"
24 #include "amdgpu_xcp.h"
25 #include "gfxhub_v1_2.h"
26 #include "gfxhub_v1_1.h"
27
28 #include "gc/gc_9_4_3_offset.h"
29 #include "gc/gc_9_4_3_sh_mask.h"
30 #include "vega10_enum.h"
31
32 #include "soc15_common.h"
33
34 #define regVM_L2_CNTL3_DEFAULT 0x80100007
35 #define regVM_L2_CNTL4_DEFAULT 0x000000c1
36
gfxhub_v1_2_get_mc_fb_offset(struct amdgpu_device * adev)37 static u64 gfxhub_v1_2_get_mc_fb_offset(struct amdgpu_device *adev)
38 {
39 return (u64)RREG32_SOC15(GC, GET_INST(GC, 0), regMC_VM_FB_OFFSET) << 24;
40 }
41
gfxhub_v1_2_xcc_setup_vm_pt_regs(struct amdgpu_device * adev,uint32_t vmid,uint64_t page_table_base,uint32_t xcc_mask)42 static void gfxhub_v1_2_xcc_setup_vm_pt_regs(struct amdgpu_device *adev,
43 uint32_t vmid,
44 uint64_t page_table_base,
45 uint32_t xcc_mask)
46 {
47 struct amdgpu_vmhub *hub;
48 int i;
49
50 for_each_inst(i, xcc_mask) {
51 hub = &adev->vmhub[AMDGPU_GFXHUB(i)];
52 WREG32_SOC15_OFFSET(GC, GET_INST(GC, i),
53 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32,
54 hub->ctx_addr_distance * vmid,
55 lower_32_bits(page_table_base));
56
57 WREG32_SOC15_OFFSET(GC, GET_INST(GC, i),
58 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32,
59 hub->ctx_addr_distance * vmid,
60 upper_32_bits(page_table_base));
61 }
62 }
63
gfxhub_v1_2_setup_vm_pt_regs(struct amdgpu_device * adev,uint32_t vmid,uint64_t page_table_base)64 static void gfxhub_v1_2_setup_vm_pt_regs(struct amdgpu_device *adev,
65 uint32_t vmid,
66 uint64_t page_table_base)
67 {
68 uint32_t xcc_mask;
69
70 xcc_mask = GENMASK(NUM_XCC(adev->gfx.xcc_mask) - 1, 0);
71 gfxhub_v1_2_xcc_setup_vm_pt_regs(adev, vmid, page_table_base, xcc_mask);
72 }
73
gfxhub_v1_2_xcc_init_gart_aperture_regs(struct amdgpu_device * adev,uint32_t xcc_mask)74 static void gfxhub_v1_2_xcc_init_gart_aperture_regs(struct amdgpu_device *adev,
75 uint32_t xcc_mask)
76 {
77 uint64_t pt_base;
78 int i;
79
80 if (adev->gmc.pdb0_bo)
81 pt_base = amdgpu_gmc_pd_addr(adev->gmc.pdb0_bo);
82 else
83 pt_base = amdgpu_gmc_pd_addr(adev->gart.bo);
84
85 gfxhub_v1_2_xcc_setup_vm_pt_regs(adev, 0, pt_base, xcc_mask);
86
87 /* If use GART for FB translation, vmid0 page table covers both
88 * vram and system memory (gart)
89 */
90 for_each_inst(i, xcc_mask) {
91 if (adev->gmc.pdb0_bo) {
92 WREG32_SOC15(GC, GET_INST(GC, i),
93 regVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
94 (u32)(adev->gmc.fb_start >> 12));
95 WREG32_SOC15(GC, GET_INST(GC, i),
96 regVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
97 (u32)(adev->gmc.fb_start >> 44));
98
99 WREG32_SOC15(GC, GET_INST(GC, i),
100 regVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
101 (u32)(adev->gmc.gart_end >> 12));
102 WREG32_SOC15(GC, GET_INST(GC, i),
103 regVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
104 (u32)(adev->gmc.gart_end >> 44));
105 } else {
106 WREG32_SOC15(GC, GET_INST(GC, i),
107 regVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
108 (u32)(adev->gmc.gart_start >> 12));
109 WREG32_SOC15(GC, GET_INST(GC, i),
110 regVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
111 (u32)(adev->gmc.gart_start >> 44));
112
113 WREG32_SOC15(GC, GET_INST(GC, i),
114 regVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
115 (u32)(adev->gmc.gart_end >> 12));
116 WREG32_SOC15(GC, GET_INST(GC, i),
117 regVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
118 (u32)(adev->gmc.gart_end >> 44));
119 }
120 }
121 }
122
123 static void
gfxhub_v1_2_xcc_init_system_aperture_regs(struct amdgpu_device * adev,uint32_t xcc_mask)124 gfxhub_v1_2_xcc_init_system_aperture_regs(struct amdgpu_device *adev,
125 uint32_t xcc_mask)
126 {
127 uint64_t value;
128 uint32_t tmp;
129 int i;
130
131 for_each_inst(i, xcc_mask) {
132 /* Program the AGP BAR */
133 WREG32_SOC15_RLC(GC, GET_INST(GC, i), regMC_VM_AGP_BASE, 0);
134 WREG32_SOC15_RLC(GC, GET_INST(GC, i), regMC_VM_AGP_BOT, adev->gmc.agp_start >> 24);
135 WREG32_SOC15_RLC(GC, GET_INST(GC, i), regMC_VM_AGP_TOP, adev->gmc.agp_end >> 24);
136
137 if (!amdgpu_sriov_vf(adev) || adev->asic_type <= CHIP_VEGA10) {
138 /* Program the system aperture low logical page number. */
139 WREG32_SOC15_RLC(GC, GET_INST(GC, i), regMC_VM_SYSTEM_APERTURE_LOW_ADDR,
140 min(adev->gmc.fb_start, adev->gmc.agp_start) >> 18);
141
142 if (adev->apu_flags & (AMD_APU_IS_RAVEN2 |
143 AMD_APU_IS_RENOIR |
144 AMD_APU_IS_GREEN_SARDINE))
145 /*
146 * Raven2 has a HW issue that it is unable to use the
147 * vram which is out of MC_VM_SYSTEM_APERTURE_HIGH_ADDR.
148 * So here is the workaround that increase system
149 * aperture high address (add 1) to get rid of the VM
150 * fault and hardware hang.
151 */
152 WREG32_SOC15_RLC(GC, GET_INST(GC, i),
153 regMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
154 max((adev->gmc.fb_end >> 18) + 0x1,
155 adev->gmc.agp_end >> 18));
156 else
157 WREG32_SOC15_RLC(GC, GET_INST(GC, i),
158 regMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
159 max(adev->gmc.fb_end, adev->gmc.agp_end) >> 18);
160
161 /* Set default page address. */
162 value = amdgpu_gmc_vram_mc2pa(adev, adev->mem_scratch.gpu_addr);
163 WREG32_SOC15(GC, GET_INST(GC, i), regMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_LSB,
164 (u32)(value >> 12));
165 WREG32_SOC15(GC, GET_INST(GC, i), regMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_MSB,
166 (u32)(value >> 44));
167
168 /* Program "protection fault". */
169 WREG32_SOC15(GC, GET_INST(GC, i), regVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_LO32,
170 (u32)(adev->dummy_page_addr >> 12));
171 WREG32_SOC15(GC, GET_INST(GC, i), regVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_HI32,
172 (u32)((u64)adev->dummy_page_addr >> 44));
173
174 tmp = RREG32_SOC15(GC, GET_INST(GC, i), regVM_L2_PROTECTION_FAULT_CNTL2);
175 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL2,
176 ACTIVE_PAGE_MIGRATION_PTE_READ_RETRY, 1);
177 WREG32_SOC15(GC, GET_INST(GC, i), regVM_L2_PROTECTION_FAULT_CNTL2, tmp);
178 }
179
180 /* In the case squeezing vram into GART aperture, we don't use
181 * FB aperture and AGP aperture. Disable them.
182 */
183 if (adev->gmc.pdb0_bo) {
184 WREG32_SOC15(GC, GET_INST(GC, i), regMC_VM_FB_LOCATION_TOP, 0);
185 WREG32_SOC15(GC, GET_INST(GC, i), regMC_VM_FB_LOCATION_BASE, 0x00FFFFFF);
186 WREG32_SOC15(GC, GET_INST(GC, i), regMC_VM_AGP_TOP, 0);
187 WREG32_SOC15(GC, GET_INST(GC, i), regMC_VM_AGP_BOT, 0xFFFFFF);
188 WREG32_SOC15(GC, GET_INST(GC, i), regMC_VM_SYSTEM_APERTURE_LOW_ADDR, 0x3FFFFFFF);
189 WREG32_SOC15(GC, GET_INST(GC, i), regMC_VM_SYSTEM_APERTURE_HIGH_ADDR, 0);
190 }
191 }
192 }
193
gfxhub_v1_2_xcc_init_tlb_regs(struct amdgpu_device * adev,uint32_t xcc_mask)194 static void gfxhub_v1_2_xcc_init_tlb_regs(struct amdgpu_device *adev,
195 uint32_t xcc_mask)
196 {
197 uint32_t tmp;
198 int i;
199
200 for_each_inst(i, xcc_mask) {
201 /* Setup TLB control */
202 tmp = RREG32_SOC15(GC, GET_INST(GC, i), regMC_VM_MX_L1_TLB_CNTL);
203
204 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
205 ENABLE_L1_TLB, 1);
206 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
207 SYSTEM_ACCESS_MODE, 3);
208 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
209 ENABLE_ADVANCED_DRIVER_MODEL, 1);
210 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
211 SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
212 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
213 MTYPE, MTYPE_UC);/* XXX for emulation. */
214 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ATC_EN, 1);
215
216 WREG32_SOC15_RLC(GC, GET_INST(GC, i), regMC_VM_MX_L1_TLB_CNTL, tmp);
217 }
218 }
219
gfxhub_v1_2_xcc_init_cache_regs(struct amdgpu_device * adev,uint32_t xcc_mask)220 static void gfxhub_v1_2_xcc_init_cache_regs(struct amdgpu_device *adev,
221 uint32_t xcc_mask)
222 {
223 uint32_t tmp;
224 int i;
225
226 for_each_inst(i, xcc_mask) {
227 /* Setup L2 cache */
228 tmp = RREG32_SOC15(GC, GET_INST(GC, i), regVM_L2_CNTL);
229 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
230 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
231 /* XXX for emulation, Refer to closed source code.*/
232 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
233 0);
234 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
235 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
236 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
237 WREG32_SOC15_RLC(GC, GET_INST(GC, i), regVM_L2_CNTL, tmp);
238
239 tmp = RREG32_SOC15(GC, GET_INST(GC, i), regVM_L2_CNTL2);
240 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
241 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
242 WREG32_SOC15_RLC(GC, GET_INST(GC, i), regVM_L2_CNTL2, tmp);
243
244 tmp = regVM_L2_CNTL3_DEFAULT;
245 if (adev->gmc.translate_further) {
246 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 12);
247 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3,
248 L2_CACHE_BIGK_FRAGMENT_SIZE, 9);
249 } else {
250 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 9);
251 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3,
252 L2_CACHE_BIGK_FRAGMENT_SIZE, 6);
253 }
254 WREG32_SOC15_RLC(GC, GET_INST(GC, i), regVM_L2_CNTL3, tmp);
255
256 tmp = regVM_L2_CNTL4_DEFAULT;
257 /* For AMD APP APUs setup WC memory */
258 if (adev->gmc.xgmi.connected_to_cpu || adev->gmc.is_app_apu) {
259 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PDE_REQUEST_PHYSICAL, 1);
260 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PTE_REQUEST_PHYSICAL, 1);
261 } else {
262 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PDE_REQUEST_PHYSICAL, 0);
263 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PTE_REQUEST_PHYSICAL, 0);
264 }
265 WREG32_SOC15_RLC(GC, GET_INST(GC, i), regVM_L2_CNTL4, tmp);
266 }
267 }
268
gfxhub_v1_2_xcc_enable_system_domain(struct amdgpu_device * adev,uint32_t xcc_mask)269 static void gfxhub_v1_2_xcc_enable_system_domain(struct amdgpu_device *adev,
270 uint32_t xcc_mask)
271 {
272 uint32_t tmp;
273 int i;
274
275 for_each_inst(i, xcc_mask) {
276 tmp = RREG32_SOC15(GC, GET_INST(GC, i), regVM_CONTEXT0_CNTL);
277 tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
278 tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH,
279 adev->gmc.vmid0_page_table_depth);
280 tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_BLOCK_SIZE,
281 adev->gmc.vmid0_page_table_block_size);
282 tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL,
283 RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 0);
284 WREG32_SOC15(GC, GET_INST(GC, i), regVM_CONTEXT0_CNTL, tmp);
285 }
286 }
287
288 static void
gfxhub_v1_2_xcc_disable_identity_aperture(struct amdgpu_device * adev,uint32_t xcc_mask)289 gfxhub_v1_2_xcc_disable_identity_aperture(struct amdgpu_device *adev,
290 uint32_t xcc_mask)
291 {
292 int i;
293
294 for_each_inst(i, xcc_mask) {
295 WREG32_SOC15(GC, GET_INST(GC, i),
296 regVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_LO32,
297 0XFFFFFFFF);
298 WREG32_SOC15(GC, GET_INST(GC, i),
299 regVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_HI32,
300 0x0000000F);
301
302 WREG32_SOC15(GC, GET_INST(GC, i),
303 regVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_LO32,
304 0);
305 WREG32_SOC15(GC, GET_INST(GC, i),
306 regVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_HI32,
307 0);
308
309 WREG32_SOC15(GC, GET_INST(GC, i),
310 regVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_LO32, 0);
311 WREG32_SOC15(GC, GET_INST(GC, i),
312 regVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_HI32, 0);
313 }
314 }
315
gfxhub_v1_2_xcc_setup_vmid_config(struct amdgpu_device * adev,uint32_t xcc_mask)316 static void gfxhub_v1_2_xcc_setup_vmid_config(struct amdgpu_device *adev,
317 uint32_t xcc_mask)
318 {
319 struct amdgpu_vmhub *hub;
320 unsigned int num_level, block_size;
321 uint32_t tmp;
322 int i, j;
323
324 num_level = adev->vm_manager.num_level;
325 block_size = adev->vm_manager.block_size;
326 if (adev->gmc.translate_further)
327 num_level -= 1;
328 else
329 block_size -= 9;
330
331 for_each_inst(j, xcc_mask) {
332 hub = &adev->vmhub[AMDGPU_GFXHUB(j)];
333 for (i = 0; i <= 14; i++) {
334 tmp = RREG32_SOC15_OFFSET(GC, GET_INST(GC, j), regVM_CONTEXT1_CNTL,
335 i * hub->ctx_distance);
336 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
337 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH,
338 num_level);
339 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
340 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
341 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
342 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT,
343 1);
344 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
345 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
346 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
347 VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
348 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
349 READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
350 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
351 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
352 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
353 EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
354 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
355 PAGE_TABLE_BLOCK_SIZE,
356 block_size);
357 /* Send no-retry XNACK on fault to suppress VM fault storm.
358 * On 9.4.2 and 9.4.3, XNACK can be enabled in
359 * the SQ per-process.
360 * Retry faults need to be enabled for that to work.
361 */
362 tmp = REG_SET_FIELD(
363 tmp, VM_CONTEXT1_CNTL,
364 RETRY_PERMISSION_OR_INVALID_PAGE_FAULT,
365 !adev->gmc.noretry ||
366 amdgpu_ip_version(adev, GC_HWIP, 0) ==
367 IP_VERSION(9, 4, 2) ||
368 amdgpu_ip_version(adev, GC_HWIP, 0) ==
369 IP_VERSION(9, 4, 3) ||
370 amdgpu_ip_version(adev, GC_HWIP, 0) ==
371 IP_VERSION(9, 4, 4) ||
372 amdgpu_ip_version(adev, GC_HWIP, 0) ==
373 IP_VERSION(9, 5, 0));
374 WREG32_SOC15_OFFSET(GC, GET_INST(GC, j), regVM_CONTEXT1_CNTL,
375 i * hub->ctx_distance, tmp);
376 WREG32_SOC15_OFFSET(GC, GET_INST(GC, j),
377 regVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32,
378 i * hub->ctx_addr_distance, 0);
379 WREG32_SOC15_OFFSET(GC, GET_INST(GC, j),
380 regVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32,
381 i * hub->ctx_addr_distance, 0);
382 WREG32_SOC15_OFFSET(GC, GET_INST(GC, j),
383 regVM_CONTEXT1_PAGE_TABLE_END_ADDR_LO32,
384 i * hub->ctx_addr_distance,
385 lower_32_bits(adev->vm_manager.max_pfn - 1));
386 WREG32_SOC15_OFFSET(GC, GET_INST(GC, j),
387 regVM_CONTEXT1_PAGE_TABLE_END_ADDR_HI32,
388 i * hub->ctx_addr_distance,
389 upper_32_bits(adev->vm_manager.max_pfn - 1));
390 }
391 }
392 }
393
gfxhub_v1_2_xcc_program_invalidation(struct amdgpu_device * adev,uint32_t xcc_mask)394 static void gfxhub_v1_2_xcc_program_invalidation(struct amdgpu_device *adev,
395 uint32_t xcc_mask)
396 {
397 struct amdgpu_vmhub *hub;
398 unsigned int i, j;
399
400 for_each_inst(j, xcc_mask) {
401 hub = &adev->vmhub[AMDGPU_GFXHUB(j)];
402
403 for (i = 0 ; i < 18; ++i) {
404 WREG32_SOC15_OFFSET(GC, GET_INST(GC, j), regVM_INVALIDATE_ENG0_ADDR_RANGE_LO32,
405 i * hub->eng_addr_distance, 0xffffffff);
406 WREG32_SOC15_OFFSET(GC, GET_INST(GC, j), regVM_INVALIDATE_ENG0_ADDR_RANGE_HI32,
407 i * hub->eng_addr_distance, 0x1f);
408 }
409 }
410 }
411
gfxhub_v1_2_xcc_gart_enable(struct amdgpu_device * adev,uint32_t xcc_mask)412 static int gfxhub_v1_2_xcc_gart_enable(struct amdgpu_device *adev,
413 uint32_t xcc_mask)
414 {
415 /* GART Enable. */
416 gfxhub_v1_2_xcc_init_gart_aperture_regs(adev, xcc_mask);
417 gfxhub_v1_2_xcc_init_system_aperture_regs(adev, xcc_mask);
418 gfxhub_v1_2_xcc_init_tlb_regs(adev, xcc_mask);
419 if (!amdgpu_sriov_vf(adev))
420 gfxhub_v1_2_xcc_init_cache_regs(adev, xcc_mask);
421
422 gfxhub_v1_2_xcc_enable_system_domain(adev, xcc_mask);
423 if (!amdgpu_sriov_vf(adev))
424 gfxhub_v1_2_xcc_disable_identity_aperture(adev, xcc_mask);
425 gfxhub_v1_2_xcc_setup_vmid_config(adev, xcc_mask);
426 gfxhub_v1_2_xcc_program_invalidation(adev, xcc_mask);
427
428 return 0;
429 }
430
gfxhub_v1_2_gart_enable(struct amdgpu_device * adev)431 static int gfxhub_v1_2_gart_enable(struct amdgpu_device *adev)
432 {
433 uint32_t xcc_mask;
434
435 xcc_mask = GENMASK(NUM_XCC(adev->gfx.xcc_mask) - 1, 0);
436 return gfxhub_v1_2_xcc_gart_enable(adev, xcc_mask);
437 }
438
gfxhub_v1_2_xcc_gart_disable(struct amdgpu_device * adev,uint32_t xcc_mask)439 static void gfxhub_v1_2_xcc_gart_disable(struct amdgpu_device *adev,
440 uint32_t xcc_mask)
441 {
442 struct amdgpu_vmhub *hub;
443 u32 tmp;
444 u32 i, j;
445
446 for_each_inst(j, xcc_mask) {
447 hub = &adev->vmhub[AMDGPU_GFXHUB(j)];
448 /* Disable all tables */
449 for (i = 0; i < 16; i++)
450 WREG32_SOC15_OFFSET(GC, GET_INST(GC, j), regVM_CONTEXT0_CNTL,
451 i * hub->ctx_distance, 0);
452
453 /* Setup TLB control */
454 tmp = RREG32_SOC15(GC, GET_INST(GC, j), regMC_VM_MX_L1_TLB_CNTL);
455 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
456 tmp = REG_SET_FIELD(tmp,
457 MC_VM_MX_L1_TLB_CNTL,
458 ENABLE_ADVANCED_DRIVER_MODEL,
459 0);
460 WREG32_SOC15_RLC(GC, GET_INST(GC, j), regMC_VM_MX_L1_TLB_CNTL, tmp);
461
462 /* Setup L2 cache */
463 if (!amdgpu_sriov_vf(adev)) {
464 tmp = RREG32_SOC15(GC, GET_INST(GC, j), regVM_L2_CNTL);
465 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
466 WREG32_SOC15(GC, GET_INST(GC, j), regVM_L2_CNTL, tmp);
467 WREG32_SOC15(GC, GET_INST(GC, j), regVM_L2_CNTL3, 0);
468 }
469 }
470 }
471
gfxhub_v1_2_gart_disable(struct amdgpu_device * adev)472 static void gfxhub_v1_2_gart_disable(struct amdgpu_device *adev)
473 {
474 uint32_t xcc_mask;
475
476 xcc_mask = GENMASK(NUM_XCC(adev->gfx.xcc_mask) - 1, 0);
477 gfxhub_v1_2_xcc_gart_disable(adev, xcc_mask);
478 }
479
gfxhub_v1_2_xcc_set_fault_enable_default(struct amdgpu_device * adev,bool value,uint32_t xcc_mask)480 static void gfxhub_v1_2_xcc_set_fault_enable_default(struct amdgpu_device *adev,
481 bool value,
482 uint32_t xcc_mask)
483 {
484 u32 tmp;
485 int i;
486
487 for_each_inst(i, xcc_mask) {
488 tmp = RREG32_SOC15(GC, GET_INST(GC, i), regVM_L2_PROTECTION_FAULT_CNTL);
489 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
490 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
491 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
492 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
493 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
494 PDE1_PROTECTION_FAULT_ENABLE_DEFAULT, value);
495 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
496 PDE2_PROTECTION_FAULT_ENABLE_DEFAULT, value);
497 tmp = REG_SET_FIELD(tmp,
498 VM_L2_PROTECTION_FAULT_CNTL,
499 TRANSLATE_FURTHER_PROTECTION_FAULT_ENABLE_DEFAULT,
500 value);
501 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
502 NACK_PROTECTION_FAULT_ENABLE_DEFAULT, value);
503 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
504 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
505 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
506 VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
507 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
508 READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
509 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
510 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
511 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
512 EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
513 if (!value) {
514 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
515 CRASH_ON_NO_RETRY_FAULT, 1);
516 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
517 CRASH_ON_RETRY_FAULT, 1);
518 }
519 WREG32_SOC15(GC, GET_INST(GC, i), regVM_L2_PROTECTION_FAULT_CNTL, tmp);
520 }
521 }
522
523 /**
524 * gfxhub_v1_2_set_fault_enable_default - update GART/VM fault handling
525 *
526 * @adev: amdgpu_device pointer
527 * @value: true redirects VM faults to the default page
528 */
gfxhub_v1_2_set_fault_enable_default(struct amdgpu_device * adev,bool value)529 static void gfxhub_v1_2_set_fault_enable_default(struct amdgpu_device *adev,
530 bool value)
531 {
532 uint32_t xcc_mask;
533
534 xcc_mask = GENMASK(NUM_XCC(adev->gfx.xcc_mask) - 1, 0);
535 gfxhub_v1_2_xcc_set_fault_enable_default(adev, value, xcc_mask);
536 }
537
gfxhub_v1_2_xcc_init(struct amdgpu_device * adev,uint32_t xcc_mask)538 static void gfxhub_v1_2_xcc_init(struct amdgpu_device *adev, uint32_t xcc_mask)
539 {
540 struct amdgpu_vmhub *hub;
541 int i;
542
543 for_each_inst(i, xcc_mask) {
544 hub = &adev->vmhub[AMDGPU_GFXHUB(i)];
545
546 hub->ctx0_ptb_addr_lo32 =
547 SOC15_REG_OFFSET(GC, GET_INST(GC, i),
548 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32);
549 hub->ctx0_ptb_addr_hi32 =
550 SOC15_REG_OFFSET(GC, GET_INST(GC, i),
551 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32);
552 hub->vm_inv_eng0_sem =
553 SOC15_REG_OFFSET(GC, GET_INST(GC, i), regVM_INVALIDATE_ENG0_SEM);
554 hub->vm_inv_eng0_req =
555 SOC15_REG_OFFSET(GC, GET_INST(GC, i), regVM_INVALIDATE_ENG0_REQ);
556 hub->vm_inv_eng0_ack =
557 SOC15_REG_OFFSET(GC, GET_INST(GC, i), regVM_INVALIDATE_ENG0_ACK);
558 hub->vm_context0_cntl =
559 SOC15_REG_OFFSET(GC, GET_INST(GC, i), regVM_CONTEXT0_CNTL);
560 hub->vm_l2_pro_fault_status =
561 SOC15_REG_OFFSET(GC, GET_INST(GC, i),
562 regVM_L2_PROTECTION_FAULT_STATUS);
563 hub->vm_l2_pro_fault_cntl =
564 SOC15_REG_OFFSET(GC, GET_INST(GC, i), regVM_L2_PROTECTION_FAULT_CNTL);
565
566 hub->ctx_distance = regVM_CONTEXT1_CNTL -
567 regVM_CONTEXT0_CNTL;
568 hub->ctx_addr_distance =
569 regVM_CONTEXT1_PAGE_TABLE_BASE_ADDR_LO32 -
570 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32;
571 hub->eng_distance = regVM_INVALIDATE_ENG1_REQ -
572 regVM_INVALIDATE_ENG0_REQ;
573 hub->eng_addr_distance =
574 regVM_INVALIDATE_ENG1_ADDR_RANGE_LO32 -
575 regVM_INVALIDATE_ENG0_ADDR_RANGE_LO32;
576 }
577 }
578
gfxhub_v1_2_init(struct amdgpu_device * adev)579 static void gfxhub_v1_2_init(struct amdgpu_device *adev)
580 {
581 uint32_t xcc_mask;
582
583 xcc_mask = GENMASK(NUM_XCC(adev->gfx.xcc_mask) - 1, 0);
584 gfxhub_v1_2_xcc_init(adev, xcc_mask);
585 }
586
gfxhub_v1_2_get_xgmi_info(struct amdgpu_device * adev)587 static int gfxhub_v1_2_get_xgmi_info(struct amdgpu_device *adev)
588 {
589 u32 max_num_physical_nodes;
590 u32 max_physical_node_id;
591 u32 xgmi_lfb_cntl;
592 u32 max_region;
593 u64 seg_size;
594
595 xgmi_lfb_cntl = RREG32_SOC15(GC, GET_INST(GC, 0), regMC_VM_XGMI_LFB_CNTL);
596 seg_size = REG_GET_FIELD(
597 RREG32_SOC15(GC, GET_INST(GC, 0), regMC_VM_XGMI_LFB_SIZE),
598 MC_VM_XGMI_LFB_SIZE, PF_LFB_SIZE) << 24;
599 max_region =
600 REG_GET_FIELD(xgmi_lfb_cntl, MC_VM_XGMI_LFB_CNTL, PF_MAX_REGION);
601
602
603
604 max_num_physical_nodes = 8;
605 max_physical_node_id = 7;
606
607 /* PF_MAX_REGION=0 means xgmi is disabled */
608 if (max_region || adev->gmc.xgmi.connected_to_cpu) {
609 adev->gmc.xgmi.num_physical_nodes = max_region + 1;
610
611 if (adev->gmc.xgmi.num_physical_nodes > max_num_physical_nodes)
612 return -EINVAL;
613
614 adev->gmc.xgmi.physical_node_id =
615 REG_GET_FIELD(xgmi_lfb_cntl, MC_VM_XGMI_LFB_CNTL,
616 PF_LFB_REGION);
617
618 if (adev->gmc.xgmi.physical_node_id > max_physical_node_id)
619 return -EINVAL;
620
621 adev->gmc.xgmi.node_segment_size = seg_size;
622 }
623
624 return 0;
625 }
626
627 const struct amdgpu_gfxhub_funcs gfxhub_v1_2_funcs = {
628 .get_mc_fb_offset = gfxhub_v1_2_get_mc_fb_offset,
629 .setup_vm_pt_regs = gfxhub_v1_2_setup_vm_pt_regs,
630 .gart_enable = gfxhub_v1_2_gart_enable,
631 .gart_disable = gfxhub_v1_2_gart_disable,
632 .set_fault_enable_default = gfxhub_v1_2_set_fault_enable_default,
633 .init = gfxhub_v1_2_init,
634 .get_xgmi_info = gfxhub_v1_2_get_xgmi_info,
635 };
636
gfxhub_v1_2_xcp_resume(void * handle,uint32_t inst_mask)637 static int gfxhub_v1_2_xcp_resume(void *handle, uint32_t inst_mask)
638 {
639 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
640 bool value;
641
642 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
643 value = false;
644 else
645 value = true;
646
647 gfxhub_v1_2_xcc_set_fault_enable_default(adev, value, inst_mask);
648
649 if (!amdgpu_sriov_vf(adev))
650 return gfxhub_v1_2_xcc_gart_enable(adev, inst_mask);
651
652 return 0;
653 }
654
gfxhub_v1_2_xcp_suspend(void * handle,uint32_t inst_mask)655 static int gfxhub_v1_2_xcp_suspend(void *handle, uint32_t inst_mask)
656 {
657 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
658
659 if (!amdgpu_sriov_vf(adev))
660 gfxhub_v1_2_xcc_gart_disable(adev, inst_mask);
661
662 return 0;
663 }
664
665 struct amdgpu_xcp_ip_funcs gfxhub_v1_2_xcp_funcs = {
666 .suspend = &gfxhub_v1_2_xcp_suspend,
667 .resume = &gfxhub_v1_2_xcp_resume
668 };
669