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 "mmhub_v1_8.h"
25
26 #include "mmhub/mmhub_1_8_0_offset.h"
27 #include "mmhub/mmhub_1_8_0_sh_mask.h"
28 #include "vega10_enum.h"
29
30 #include "soc15_common.h"
31 #include "soc15.h"
32 #include "amdgpu_ras.h"
33
34 #define regVM_L2_CNTL3_DEFAULT 0x80100007
35 #define regVM_L2_CNTL4_DEFAULT 0x000000c1
36
mmhub_v1_8_get_fb_location(struct amdgpu_device * adev)37 static u64 mmhub_v1_8_get_fb_location(struct amdgpu_device *adev)
38 {
39 u64 base = RREG32_SOC15(MMHUB, 0, regMC_VM_FB_LOCATION_BASE);
40 u64 top = RREG32_SOC15(MMHUB, 0, regMC_VM_FB_LOCATION_TOP);
41
42 base &= MC_VM_FB_LOCATION_BASE__FB_BASE_MASK;
43 base <<= 24;
44
45 top &= MC_VM_FB_LOCATION_TOP__FB_TOP_MASK;
46 top <<= 24;
47
48 adev->gmc.fb_start = base;
49 adev->gmc.fb_end = top;
50
51 return base;
52 }
53
mmhub_v1_8_setup_vm_pt_regs(struct amdgpu_device * adev,uint32_t vmid,uint64_t page_table_base)54 static void mmhub_v1_8_setup_vm_pt_regs(struct amdgpu_device *adev, uint32_t vmid,
55 uint64_t page_table_base)
56 {
57 struct amdgpu_vmhub *hub;
58 u32 inst_mask;
59 int i;
60
61 inst_mask = adev->aid_mask;
62 for_each_inst(i, inst_mask) {
63 hub = &adev->vmhub[AMDGPU_MMHUB0(i)];
64 WREG32_SOC15_OFFSET(MMHUB, i,
65 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32,
66 hub->ctx_addr_distance * vmid,
67 lower_32_bits(page_table_base));
68
69 WREG32_SOC15_OFFSET(MMHUB, i,
70 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32,
71 hub->ctx_addr_distance * vmid,
72 upper_32_bits(page_table_base));
73 }
74 }
75
mmhub_v1_8_init_gart_aperture_regs(struct amdgpu_device * adev)76 static void mmhub_v1_8_init_gart_aperture_regs(struct amdgpu_device *adev)
77 {
78 uint64_t pt_base;
79 u32 inst_mask;
80 int i;
81
82 if (adev->gmc.pdb0_bo)
83 pt_base = amdgpu_gmc_pd_addr(adev->gmc.pdb0_bo);
84 else
85 pt_base = amdgpu_gmc_pd_addr(adev->gart.bo);
86
87 mmhub_v1_8_setup_vm_pt_regs(adev, 0, pt_base);
88
89 /* If use GART for FB translation, vmid0 page table covers both
90 * vram and system memory (gart)
91 */
92 inst_mask = adev->aid_mask;
93 for_each_inst(i, inst_mask) {
94 if (adev->gmc.pdb0_bo) {
95 WREG32_SOC15(MMHUB, i,
96 regVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
97 (u32)(adev->gmc.fb_start >> 12));
98 WREG32_SOC15(MMHUB, i,
99 regVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
100 (u32)(adev->gmc.fb_start >> 44));
101
102 WREG32_SOC15(MMHUB, i,
103 regVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
104 (u32)(adev->gmc.gart_end >> 12));
105 WREG32_SOC15(MMHUB, i,
106 regVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
107 (u32)(adev->gmc.gart_end >> 44));
108
109 } else {
110 WREG32_SOC15(MMHUB, i,
111 regVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
112 (u32)(adev->gmc.gart_start >> 12));
113 WREG32_SOC15(MMHUB, i,
114 regVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
115 (u32)(adev->gmc.gart_start >> 44));
116
117 WREG32_SOC15(MMHUB, i,
118 regVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
119 (u32)(adev->gmc.gart_end >> 12));
120 WREG32_SOC15(MMHUB, i,
121 regVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
122 (u32)(adev->gmc.gart_end >> 44));
123 }
124 }
125 }
126
mmhub_v1_8_init_system_aperture_regs(struct amdgpu_device * adev)127 static void mmhub_v1_8_init_system_aperture_regs(struct amdgpu_device *adev)
128 {
129 uint32_t tmp, inst_mask;
130 uint64_t value;
131 int i;
132
133 if (amdgpu_sriov_vf(adev))
134 return;
135
136 inst_mask = adev->aid_mask;
137 for_each_inst(i, inst_mask) {
138 /* Program the AGP BAR */
139 WREG32_SOC15(MMHUB, i, regMC_VM_AGP_BASE, 0);
140 WREG32_SOC15(MMHUB, i, regMC_VM_AGP_BOT,
141 adev->gmc.agp_start >> 24);
142 WREG32_SOC15(MMHUB, i, regMC_VM_AGP_TOP,
143 adev->gmc.agp_end >> 24);
144
145 /* Program the system aperture low logical page number. */
146 WREG32_SOC15(MMHUB, i, regMC_VM_SYSTEM_APERTURE_LOW_ADDR,
147 min(adev->gmc.fb_start, adev->gmc.agp_start) >> 18);
148
149 WREG32_SOC15(MMHUB, i, regMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
150 max(adev->gmc.fb_end, adev->gmc.agp_end) >> 18);
151
152 /* In the case squeezing vram into GART aperture, we don't use
153 * FB aperture and AGP aperture. Disable them.
154 */
155 if (adev->gmc.pdb0_bo) {
156 WREG32_SOC15(MMHUB, i, regMC_VM_AGP_BOT, 0xFFFFFF);
157 WREG32_SOC15(MMHUB, i, regMC_VM_AGP_TOP, 0);
158 WREG32_SOC15(MMHUB, i, regMC_VM_FB_LOCATION_TOP, 0);
159 WREG32_SOC15(MMHUB, i, regMC_VM_FB_LOCATION_BASE,
160 0x00FFFFFF);
161 WREG32_SOC15(MMHUB, i,
162 regMC_VM_SYSTEM_APERTURE_LOW_ADDR,
163 0x3FFFFFFF);
164 WREG32_SOC15(MMHUB, i,
165 regMC_VM_SYSTEM_APERTURE_HIGH_ADDR, 0);
166 }
167
168 /* Set default page address. */
169 value = amdgpu_gmc_vram_mc2pa(adev, adev->mem_scratch.gpu_addr);
170 WREG32_SOC15(MMHUB, i, regMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_LSB,
171 (u32)(value >> 12));
172 WREG32_SOC15(MMHUB, i, regMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_MSB,
173 (u32)(value >> 44));
174
175 /* Program "protection fault". */
176 WREG32_SOC15(MMHUB, i,
177 regVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_LO32,
178 (u32)(adev->dummy_page_addr >> 12));
179 WREG32_SOC15(MMHUB, i,
180 regVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_HI32,
181 (u32)((u64)adev->dummy_page_addr >> 44));
182
183 tmp = RREG32_SOC15(MMHUB, i, regVM_L2_PROTECTION_FAULT_CNTL2);
184 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL2,
185 ACTIVE_PAGE_MIGRATION_PTE_READ_RETRY, 1);
186 WREG32_SOC15(MMHUB, i, regVM_L2_PROTECTION_FAULT_CNTL2, tmp);
187 }
188 }
189
mmhub_v1_8_init_tlb_regs(struct amdgpu_device * adev)190 static void mmhub_v1_8_init_tlb_regs(struct amdgpu_device *adev)
191 {
192 uint32_t tmp, inst_mask;
193 int i;
194
195 /* Setup TLB control */
196 inst_mask = adev->aid_mask;
197 for_each_inst(i, inst_mask) {
198 tmp = RREG32_SOC15(MMHUB, i, regMC_VM_MX_L1_TLB_CNTL);
199
200 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB,
201 1);
202 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
203 SYSTEM_ACCESS_MODE, 3);
204 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
205 ENABLE_ADVANCED_DRIVER_MODEL, 1);
206 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
207 SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
208 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
209 MTYPE, MTYPE_UC);/* XXX for emulation. */
210 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ATC_EN, 1);
211
212 WREG32_SOC15(MMHUB, i, regMC_VM_MX_L1_TLB_CNTL, tmp);
213 }
214 }
215
216 /* Set snoop bit for SDMA so that SDMA writes probe-invalidates RW lines */
mmhub_v1_8_init_snoop_override_regs(struct amdgpu_device * adev)217 static void mmhub_v1_8_init_snoop_override_regs(struct amdgpu_device *adev)
218 {
219 uint32_t tmp, inst_mask;
220 int i, j;
221 uint32_t distance = regDAGB1_WRCLI_GPU_SNOOP_OVERRIDE -
222 regDAGB0_WRCLI_GPU_SNOOP_OVERRIDE;
223
224 inst_mask = adev->aid_mask;
225 for_each_inst(i, inst_mask) {
226 for (j = 0; j < 5; j++) { /* DAGB instances */
227 tmp = RREG32_SOC15_OFFSET(MMHUB, i,
228 regDAGB0_WRCLI_GPU_SNOOP_OVERRIDE, j * distance);
229 tmp |= (1 << 15); /* SDMA client is BIT15 */
230 WREG32_SOC15_OFFSET(MMHUB, i,
231 regDAGB0_WRCLI_GPU_SNOOP_OVERRIDE, j * distance, tmp);
232
233 tmp = RREG32_SOC15_OFFSET(MMHUB, i,
234 regDAGB0_WRCLI_GPU_SNOOP_OVERRIDE_VALUE, j * distance);
235 tmp |= (1 << 15);
236 WREG32_SOC15_OFFSET(MMHUB, i,
237 regDAGB0_WRCLI_GPU_SNOOP_OVERRIDE_VALUE, j * distance, tmp);
238 }
239 }
240 }
241
mmhub_v1_8_init_cache_regs(struct amdgpu_device * adev)242 static void mmhub_v1_8_init_cache_regs(struct amdgpu_device *adev)
243 {
244 uint32_t tmp, inst_mask;
245 int i;
246
247 if (amdgpu_sriov_vf(adev))
248 return;
249
250 /* Setup L2 cache */
251 inst_mask = adev->aid_mask;
252 for_each_inst(i, inst_mask) {
253 tmp = RREG32_SOC15(MMHUB, i, regVM_L2_CNTL);
254 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
255 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,
256 ENABLE_L2_FRAGMENT_PROCESSING, 1);
257 /* XXX for emulation, Refer to closed source code.*/
258 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,
259 L2_PDE0_CACHE_TAG_GENERATION_MODE, 0);
260 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION,
261 0);
262 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,
263 CONTEXT1_IDENTITY_ACCESS_MODE, 1);
264 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,
265 IDENTITY_MODE_FRAGMENT_SIZE, 0);
266 WREG32_SOC15(MMHUB, i, regVM_L2_CNTL, tmp);
267
268 tmp = RREG32_SOC15(MMHUB, i, regVM_L2_CNTL2);
269 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS,
270 1);
271 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
272 WREG32_SOC15(MMHUB, i, regVM_L2_CNTL2, tmp);
273
274 tmp = regVM_L2_CNTL3_DEFAULT;
275 if (adev->gmc.translate_further) {
276 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 12);
277 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3,
278 L2_CACHE_BIGK_FRAGMENT_SIZE, 9);
279 } else {
280 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 9);
281 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3,
282 L2_CACHE_BIGK_FRAGMENT_SIZE, 6);
283 }
284 WREG32_SOC15(MMHUB, i, regVM_L2_CNTL3, tmp);
285
286 tmp = regVM_L2_CNTL4_DEFAULT;
287 /* For AMD APP APUs setup WC memory */
288 if (adev->gmc.xgmi.connected_to_cpu || adev->gmc.is_app_apu) {
289 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4,
290 VMC_TAP_PDE_REQUEST_PHYSICAL, 1);
291 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4,
292 VMC_TAP_PTE_REQUEST_PHYSICAL, 1);
293 } else {
294 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4,
295 VMC_TAP_PDE_REQUEST_PHYSICAL, 0);
296 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4,
297 VMC_TAP_PTE_REQUEST_PHYSICAL, 0);
298 }
299 WREG32_SOC15(MMHUB, i, regVM_L2_CNTL4, tmp);
300 }
301 }
302
mmhub_v1_8_enable_system_domain(struct amdgpu_device * adev)303 static void mmhub_v1_8_enable_system_domain(struct amdgpu_device *adev)
304 {
305 uint32_t tmp, inst_mask;
306 int i;
307
308 inst_mask = adev->aid_mask;
309 for_each_inst(i, inst_mask) {
310 tmp = RREG32_SOC15(MMHUB, i, regVM_CONTEXT0_CNTL);
311 tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
312 tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH,
313 adev->gmc.vmid0_page_table_depth);
314 tmp = REG_SET_FIELD(tmp,
315 VM_CONTEXT0_CNTL, PAGE_TABLE_BLOCK_SIZE,
316 adev->gmc.vmid0_page_table_block_size);
317 tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL,
318 RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 0);
319 WREG32_SOC15(MMHUB, i, regVM_CONTEXT0_CNTL, tmp);
320 }
321 }
322
mmhub_v1_8_disable_identity_aperture(struct amdgpu_device * adev)323 static void mmhub_v1_8_disable_identity_aperture(struct amdgpu_device *adev)
324 {
325 u32 inst_mask;
326 int i;
327
328 if (amdgpu_sriov_vf(adev))
329 return;
330
331 inst_mask = adev->aid_mask;
332 for_each_inst(i, inst_mask) {
333 WREG32_SOC15(MMHUB, i,
334 regVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_LO32,
335 0XFFFFFFFF);
336 WREG32_SOC15(MMHUB, i,
337 regVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_HI32,
338 0x0000000F);
339
340 WREG32_SOC15(MMHUB, i,
341 regVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_LO32,
342 0);
343 WREG32_SOC15(MMHUB, i,
344 regVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_HI32,
345 0);
346
347 WREG32_SOC15(MMHUB, i,
348 regVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_LO32, 0);
349 WREG32_SOC15(MMHUB, i,
350 regVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_HI32, 0);
351 }
352 }
353
mmhub_v1_8_setup_vmid_config(struct amdgpu_device * adev)354 static void mmhub_v1_8_setup_vmid_config(struct amdgpu_device *adev)
355 {
356 struct amdgpu_vmhub *hub;
357 unsigned int num_level, block_size;
358 uint32_t tmp, inst_mask;
359 int i, j;
360
361 num_level = adev->vm_manager.num_level;
362 block_size = adev->vm_manager.block_size;
363 if (adev->gmc.translate_further)
364 num_level -= 1;
365 else
366 block_size -= 9;
367
368 inst_mask = adev->aid_mask;
369 for_each_inst(j, inst_mask) {
370 hub = &adev->vmhub[AMDGPU_MMHUB0(j)];
371 for (i = 0; i <= 14; i++) {
372 tmp = RREG32_SOC15_OFFSET(MMHUB, j, regVM_CONTEXT1_CNTL,
373 i * hub->ctx_distance);
374 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
375 ENABLE_CONTEXT, 1);
376 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
377 PAGE_TABLE_DEPTH, num_level);
378 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
379 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
380 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
381 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
382 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
383 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
384 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
385 VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
386 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
387 READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
388 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
389 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
390 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
391 EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
392 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
393 PAGE_TABLE_BLOCK_SIZE,
394 block_size);
395 /* On 9.4.3, XNACK can be enabled in the SQ
396 * per-process. Retry faults need to be enabled for
397 * that to work.
398 */
399 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
400 RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 1);
401 WREG32_SOC15_OFFSET(MMHUB, j, regVM_CONTEXT1_CNTL,
402 i * hub->ctx_distance, tmp);
403 WREG32_SOC15_OFFSET(MMHUB, j,
404 regVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32,
405 i * hub->ctx_addr_distance, 0);
406 WREG32_SOC15_OFFSET(MMHUB, j,
407 regVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32,
408 i * hub->ctx_addr_distance, 0);
409 WREG32_SOC15_OFFSET(MMHUB, j,
410 regVM_CONTEXT1_PAGE_TABLE_END_ADDR_LO32,
411 i * hub->ctx_addr_distance,
412 lower_32_bits(adev->vm_manager.max_pfn - 1));
413 WREG32_SOC15_OFFSET(MMHUB, j,
414 regVM_CONTEXT1_PAGE_TABLE_END_ADDR_HI32,
415 i * hub->ctx_addr_distance,
416 upper_32_bits(adev->vm_manager.max_pfn - 1));
417 }
418 }
419 }
420
mmhub_v1_8_program_invalidation(struct amdgpu_device * adev)421 static void mmhub_v1_8_program_invalidation(struct amdgpu_device *adev)
422 {
423 struct amdgpu_vmhub *hub;
424 u32 i, j, inst_mask;
425
426 inst_mask = adev->aid_mask;
427 for_each_inst(j, inst_mask) {
428 hub = &adev->vmhub[AMDGPU_MMHUB0(j)];
429 for (i = 0; i < 18; ++i) {
430 WREG32_SOC15_OFFSET(MMHUB, j,
431 regVM_INVALIDATE_ENG0_ADDR_RANGE_LO32,
432 i * hub->eng_addr_distance, 0xffffffff);
433 WREG32_SOC15_OFFSET(MMHUB, j,
434 regVM_INVALIDATE_ENG0_ADDR_RANGE_HI32,
435 i * hub->eng_addr_distance, 0x1f);
436 }
437 }
438 }
439
mmhub_v1_8_gart_enable(struct amdgpu_device * adev)440 static int mmhub_v1_8_gart_enable(struct amdgpu_device *adev)
441 {
442 /* GART Enable. */
443 mmhub_v1_8_init_gart_aperture_regs(adev);
444 mmhub_v1_8_init_system_aperture_regs(adev);
445 mmhub_v1_8_init_tlb_regs(adev);
446 mmhub_v1_8_init_cache_regs(adev);
447 mmhub_v1_8_init_snoop_override_regs(adev);
448
449 mmhub_v1_8_enable_system_domain(adev);
450 mmhub_v1_8_disable_identity_aperture(adev);
451 mmhub_v1_8_setup_vmid_config(adev);
452 mmhub_v1_8_program_invalidation(adev);
453
454 return 0;
455 }
456
mmhub_v1_8_gart_disable(struct amdgpu_device * adev)457 static void mmhub_v1_8_gart_disable(struct amdgpu_device *adev)
458 {
459 struct amdgpu_vmhub *hub;
460 u32 tmp;
461 u32 i, j, inst_mask;
462
463 /* Disable all tables */
464 inst_mask = adev->aid_mask;
465 for_each_inst(j, inst_mask) {
466 hub = &adev->vmhub[AMDGPU_MMHUB0(j)];
467 for (i = 0; i < 16; i++)
468 WREG32_SOC15_OFFSET(MMHUB, j, regVM_CONTEXT0_CNTL,
469 i * hub->ctx_distance, 0);
470
471 /* Setup TLB control */
472 tmp = RREG32_SOC15(MMHUB, j, regMC_VM_MX_L1_TLB_CNTL);
473 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB,
474 0);
475 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
476 ENABLE_ADVANCED_DRIVER_MODEL, 0);
477 WREG32_SOC15(MMHUB, j, regMC_VM_MX_L1_TLB_CNTL, tmp);
478
479 if (!amdgpu_sriov_vf(adev)) {
480 /* Setup L2 cache */
481 tmp = RREG32_SOC15(MMHUB, j, regVM_L2_CNTL);
482 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE,
483 0);
484 WREG32_SOC15(MMHUB, j, regVM_L2_CNTL, tmp);
485 WREG32_SOC15(MMHUB, j, regVM_L2_CNTL3, 0);
486 }
487 }
488 }
489
490 /**
491 * mmhub_v1_8_set_fault_enable_default - update GART/VM fault handling
492 *
493 * @adev: amdgpu_device pointer
494 * @value: true redirects VM faults to the default page
495 */
mmhub_v1_8_set_fault_enable_default(struct amdgpu_device * adev,bool value)496 static void mmhub_v1_8_set_fault_enable_default(struct amdgpu_device *adev, bool value)
497 {
498 u32 tmp, inst_mask;
499 int i;
500
501 if (amdgpu_sriov_vf(adev))
502 return;
503
504 inst_mask = adev->aid_mask;
505 for_each_inst(i, inst_mask) {
506 tmp = RREG32_SOC15(MMHUB, i, regVM_L2_PROTECTION_FAULT_CNTL);
507 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
508 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
509 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
510 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
511 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
512 PDE1_PROTECTION_FAULT_ENABLE_DEFAULT, value);
513 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
514 PDE2_PROTECTION_FAULT_ENABLE_DEFAULT, value);
515 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
516 TRANSLATE_FURTHER_PROTECTION_FAULT_ENABLE_DEFAULT,
517 value);
518 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
519 NACK_PROTECTION_FAULT_ENABLE_DEFAULT, value);
520 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
521 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
522 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
523 VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
524 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
525 READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
526 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
527 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
528 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
529 EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
530 if (!value) {
531 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
532 CRASH_ON_NO_RETRY_FAULT, 1);
533 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
534 CRASH_ON_RETRY_FAULT, 1);
535 }
536
537 WREG32_SOC15(MMHUB, i, regVM_L2_PROTECTION_FAULT_CNTL, tmp);
538 }
539 }
540
mmhub_v1_8_init(struct amdgpu_device * adev)541 static void mmhub_v1_8_init(struct amdgpu_device *adev)
542 {
543 struct amdgpu_vmhub *hub;
544 u32 inst_mask;
545 int i;
546
547 inst_mask = adev->aid_mask;
548 for_each_inst(i, inst_mask) {
549 hub = &adev->vmhub[AMDGPU_MMHUB0(i)];
550
551 hub->ctx0_ptb_addr_lo32 = SOC15_REG_OFFSET(MMHUB, i,
552 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32);
553 hub->ctx0_ptb_addr_hi32 = SOC15_REG_OFFSET(MMHUB, i,
554 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32);
555 hub->vm_inv_eng0_req =
556 SOC15_REG_OFFSET(MMHUB, i, regVM_INVALIDATE_ENG0_REQ);
557 hub->vm_inv_eng0_ack =
558 SOC15_REG_OFFSET(MMHUB, i, regVM_INVALIDATE_ENG0_ACK);
559 hub->vm_context0_cntl =
560 SOC15_REG_OFFSET(MMHUB, i, regVM_CONTEXT0_CNTL);
561 hub->vm_l2_pro_fault_status = SOC15_REG_OFFSET(MMHUB, i,
562 regVM_L2_PROTECTION_FAULT_STATUS);
563 hub->vm_l2_pro_fault_cntl = SOC15_REG_OFFSET(MMHUB, i,
564 regVM_L2_PROTECTION_FAULT_CNTL);
565
566 hub->ctx_distance = regVM_CONTEXT1_CNTL - regVM_CONTEXT0_CNTL;
567 hub->ctx_addr_distance =
568 regVM_CONTEXT1_PAGE_TABLE_BASE_ADDR_LO32 -
569 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32;
570 hub->eng_distance = regVM_INVALIDATE_ENG1_REQ -
571 regVM_INVALIDATE_ENG0_REQ;
572 hub->eng_addr_distance = regVM_INVALIDATE_ENG1_ADDR_RANGE_LO32 -
573 regVM_INVALIDATE_ENG0_ADDR_RANGE_LO32;
574 }
575 }
576
mmhub_v1_8_set_clockgating(struct amdgpu_device * adev,enum amd_clockgating_state state)577 static int mmhub_v1_8_set_clockgating(struct amdgpu_device *adev,
578 enum amd_clockgating_state state)
579 {
580 return 0;
581 }
582
mmhub_v1_8_get_clockgating(struct amdgpu_device * adev,u64 * flags)583 static void mmhub_v1_8_get_clockgating(struct amdgpu_device *adev, u64 *flags)
584 {
585
586 }
587
588 const struct amdgpu_mmhub_funcs mmhub_v1_8_funcs = {
589 .get_fb_location = mmhub_v1_8_get_fb_location,
590 .init = mmhub_v1_8_init,
591 .gart_enable = mmhub_v1_8_gart_enable,
592 .set_fault_enable_default = mmhub_v1_8_set_fault_enable_default,
593 .gart_disable = mmhub_v1_8_gart_disable,
594 .setup_vm_pt_regs = mmhub_v1_8_setup_vm_pt_regs,
595 .set_clockgating = mmhub_v1_8_set_clockgating,
596 .get_clockgating = mmhub_v1_8_get_clockgating,
597 };
598
599 static const struct amdgpu_ras_err_status_reg_entry mmhub_v1_8_ce_reg_list[] = {
600 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA0_CE_ERR_STATUS_LO, regMMEA0_CE_ERR_STATUS_HI),
601 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA0"},
602 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA1_CE_ERR_STATUS_LO, regMMEA1_CE_ERR_STATUS_HI),
603 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA1"},
604 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA2_CE_ERR_STATUS_LO, regMMEA2_CE_ERR_STATUS_HI),
605 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA2"},
606 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA3_CE_ERR_STATUS_LO, regMMEA3_CE_ERR_STATUS_HI),
607 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA3"},
608 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA4_CE_ERR_STATUS_LO, regMMEA4_CE_ERR_STATUS_HI),
609 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA4"},
610 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMM_CANE_CE_ERR_STATUS_LO, regMM_CANE_CE_ERR_STATUS_HI),
611 1, 0, "MM_CANE"},
612 };
613
614 static const struct amdgpu_ras_err_status_reg_entry mmhub_v1_8_ue_reg_list[] = {
615 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA0_UE_ERR_STATUS_LO, regMMEA0_UE_ERR_STATUS_HI),
616 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA0"},
617 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA1_UE_ERR_STATUS_LO, regMMEA1_UE_ERR_STATUS_HI),
618 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA1"},
619 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA2_UE_ERR_STATUS_LO, regMMEA2_UE_ERR_STATUS_HI),
620 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA2"},
621 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA3_UE_ERR_STATUS_LO, regMMEA3_UE_ERR_STATUS_HI),
622 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA3"},
623 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA4_UE_ERR_STATUS_LO, regMMEA4_UE_ERR_STATUS_HI),
624 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA4"},
625 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMM_CANE_UE_ERR_STATUS_LO, regMM_CANE_UE_ERR_STATUS_HI),
626 1, 0, "MM_CANE"},
627 };
628
629 static const struct amdgpu_ras_memory_id_entry mmhub_v1_8_ras_memory_list[] = {
630 {AMDGPU_MMHUB_WGMI_PAGEMEM, "MMEA_WGMI_PAGEMEM"},
631 {AMDGPU_MMHUB_RGMI_PAGEMEM, "MMEA_RGMI_PAGEMEM"},
632 {AMDGPU_MMHUB_WDRAM_PAGEMEM, "MMEA_WDRAM_PAGEMEM"},
633 {AMDGPU_MMHUB_RDRAM_PAGEMEM, "MMEA_RDRAM_PAGEMEM"},
634 {AMDGPU_MMHUB_WIO_CMDMEM, "MMEA_WIO_CMDMEM"},
635 {AMDGPU_MMHUB_RIO_CMDMEM, "MMEA_RIO_CMDMEM"},
636 {AMDGPU_MMHUB_WGMI_CMDMEM, "MMEA_WGMI_CMDMEM"},
637 {AMDGPU_MMHUB_RGMI_CMDMEM, "MMEA_RGMI_CMDMEM"},
638 {AMDGPU_MMHUB_WDRAM_CMDMEM, "MMEA_WDRAM_CMDMEM"},
639 {AMDGPU_MMHUB_RDRAM_CMDMEM, "MMEA_RDRAM_CMDMEM"},
640 {AMDGPU_MMHUB_MAM_DMEM0, "MMEA_MAM_DMEM0"},
641 {AMDGPU_MMHUB_MAM_DMEM1, "MMEA_MAM_DMEM1"},
642 {AMDGPU_MMHUB_MAM_DMEM2, "MMEA_MAM_DMEM2"},
643 {AMDGPU_MMHUB_MAM_DMEM3, "MMEA_MAM_DMEM3"},
644 {AMDGPU_MMHUB_WRET_TAGMEM, "MMEA_WRET_TAGMEM"},
645 {AMDGPU_MMHUB_RRET_TAGMEM, "MMEA_RRET_TAGMEM"},
646 {AMDGPU_MMHUB_WIO_DATAMEM, "MMEA_WIO_DATAMEM"},
647 {AMDGPU_MMHUB_WGMI_DATAMEM, "MMEA_WGMI_DATAMEM"},
648 {AMDGPU_MMHUB_WDRAM_DATAMEM, "MMEA_WDRAM_DATAMEM"},
649 };
650
mmhub_v1_8_inst_query_ras_error_count(struct amdgpu_device * adev,uint32_t mmhub_inst,void * ras_err_status)651 static void mmhub_v1_8_inst_query_ras_error_count(struct amdgpu_device *adev,
652 uint32_t mmhub_inst,
653 void *ras_err_status)
654 {
655 struct ras_err_data *err_data = (struct ras_err_data *)ras_err_status;
656 unsigned long ue_count = 0, ce_count = 0;
657
658 /* NOTE: mmhub is converted by aid_mask and the range is 0-3,
659 * which can be used as die ID directly */
660 struct amdgpu_smuio_mcm_config_info mcm_info = {
661 .socket_id = adev->smuio.funcs->get_socket_id(adev),
662 .die_id = mmhub_inst,
663 };
664
665 amdgpu_ras_inst_query_ras_error_count(adev,
666 mmhub_v1_8_ce_reg_list,
667 ARRAY_SIZE(mmhub_v1_8_ce_reg_list),
668 mmhub_v1_8_ras_memory_list,
669 ARRAY_SIZE(mmhub_v1_8_ras_memory_list),
670 mmhub_inst,
671 AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE,
672 &ce_count);
673 amdgpu_ras_inst_query_ras_error_count(adev,
674 mmhub_v1_8_ue_reg_list,
675 ARRAY_SIZE(mmhub_v1_8_ue_reg_list),
676 mmhub_v1_8_ras_memory_list,
677 ARRAY_SIZE(mmhub_v1_8_ras_memory_list),
678 mmhub_inst,
679 AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
680 &ue_count);
681
682 amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, ce_count);
683 amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, ue_count);
684 }
685
mmhub_v1_8_query_ras_error_count(struct amdgpu_device * adev,void * ras_err_status)686 static void mmhub_v1_8_query_ras_error_count(struct amdgpu_device *adev,
687 void *ras_err_status)
688 {
689 uint32_t inst_mask;
690 uint32_t i;
691
692 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__MMHUB)) {
693 dev_warn(adev->dev, "MMHUB RAS is not supported\n");
694 return;
695 }
696
697 inst_mask = adev->aid_mask;
698 for_each_inst(i, inst_mask)
699 mmhub_v1_8_inst_query_ras_error_count(adev, i, ras_err_status);
700 }
701
mmhub_v1_8_inst_reset_ras_error_count(struct amdgpu_device * adev,uint32_t mmhub_inst)702 static void mmhub_v1_8_inst_reset_ras_error_count(struct amdgpu_device *adev,
703 uint32_t mmhub_inst)
704 {
705 amdgpu_ras_inst_reset_ras_error_count(adev,
706 mmhub_v1_8_ce_reg_list,
707 ARRAY_SIZE(mmhub_v1_8_ce_reg_list),
708 mmhub_inst);
709 amdgpu_ras_inst_reset_ras_error_count(adev,
710 mmhub_v1_8_ue_reg_list,
711 ARRAY_SIZE(mmhub_v1_8_ue_reg_list),
712 mmhub_inst);
713 }
714
mmhub_v1_8_reset_ras_error_count(struct amdgpu_device * adev)715 static void mmhub_v1_8_reset_ras_error_count(struct amdgpu_device *adev)
716 {
717 uint32_t inst_mask;
718 uint32_t i;
719
720 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__MMHUB)) {
721 dev_warn(adev->dev, "MMHUB RAS is not supported\n");
722 return;
723 }
724
725 inst_mask = adev->aid_mask;
726 for_each_inst(i, inst_mask)
727 mmhub_v1_8_inst_reset_ras_error_count(adev, i);
728 }
729
730 static const struct amdgpu_ras_block_hw_ops mmhub_v1_8_ras_hw_ops = {
731 .query_ras_error_count = mmhub_v1_8_query_ras_error_count,
732 .reset_ras_error_count = mmhub_v1_8_reset_ras_error_count,
733 };
734
mmhub_v1_8_aca_bank_parser(struct aca_handle * handle,struct aca_bank * bank,enum aca_smu_type type,void * data)735 static int mmhub_v1_8_aca_bank_parser(struct aca_handle *handle, struct aca_bank *bank,
736 enum aca_smu_type type, void *data)
737 {
738 struct aca_bank_info info;
739 u64 misc0;
740 int ret;
741
742 ret = aca_bank_info_decode(bank, &info);
743 if (ret)
744 return ret;
745
746 misc0 = bank->regs[ACA_REG_IDX_MISC0];
747 switch (type) {
748 case ACA_SMU_TYPE_UE:
749 bank->aca_err_type = ACA_ERROR_TYPE_UE;
750 ret = aca_error_cache_log_bank_error(handle, &info, ACA_ERROR_TYPE_UE,
751 1ULL);
752 break;
753 case ACA_SMU_TYPE_CE:
754 bank->aca_err_type = ACA_ERROR_TYPE_CE;
755 ret = aca_error_cache_log_bank_error(handle, &info, bank->aca_err_type,
756 ACA_REG__MISC0__ERRCNT(misc0));
757 break;
758 default:
759 return -EINVAL;
760 }
761
762 return ret;
763 }
764
765 /* reference to smu driver if header file */
766 static int mmhub_v1_8_err_codes[] = {
767 0, 1, 2, 3, 4, /* CODE_DAGB0 - 4 */
768 5, 6, 7, 8, 9, /* CODE_EA0 - 4 */
769 10, /* CODE_UTCL2_ROUTER */
770 11, /* CODE_VML2 */
771 12, /* CODE_VML2_WALKER */
772 13, /* CODE_MMCANE */
773 };
774
mmhub_v1_8_aca_bank_is_valid(struct aca_handle * handle,struct aca_bank * bank,enum aca_smu_type type,void * data)775 static bool mmhub_v1_8_aca_bank_is_valid(struct aca_handle *handle, struct aca_bank *bank,
776 enum aca_smu_type type, void *data)
777 {
778 u32 instlo;
779
780 instlo = ACA_REG__IPID__INSTANCEIDLO(bank->regs[ACA_REG_IDX_IPID]);
781 instlo &= GENMASK(31, 1);
782
783 if (instlo != mmSMNAID_AID0_MCA_SMU)
784 return false;
785
786 if (aca_bank_check_error_codes(handle->adev, bank,
787 mmhub_v1_8_err_codes,
788 ARRAY_SIZE(mmhub_v1_8_err_codes)))
789 return false;
790
791 return true;
792 }
793
794 static const struct aca_bank_ops mmhub_v1_8_aca_bank_ops = {
795 .aca_bank_parser = mmhub_v1_8_aca_bank_parser,
796 .aca_bank_is_valid = mmhub_v1_8_aca_bank_is_valid,
797 };
798
799 static const struct aca_info mmhub_v1_8_aca_info = {
800 .hwip = ACA_HWIP_TYPE_SMU,
801 .mask = ACA_ERROR_UE_MASK,
802 .bank_ops = &mmhub_v1_8_aca_bank_ops,
803 };
804
mmhub_v1_8_ras_late_init(struct amdgpu_device * adev,struct ras_common_if * ras_block)805 static int mmhub_v1_8_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block)
806 {
807 int r;
808
809 r = amdgpu_ras_block_late_init(adev, ras_block);
810 if (r)
811 return r;
812
813 r = amdgpu_ras_bind_aca(adev, AMDGPU_RAS_BLOCK__MMHUB,
814 &mmhub_v1_8_aca_info, NULL);
815 if (r)
816 goto late_fini;
817
818 return 0;
819
820 late_fini:
821 amdgpu_ras_block_late_fini(adev, ras_block);
822
823 return r;
824 }
825
826 struct amdgpu_mmhub_ras mmhub_v1_8_ras = {
827 .ras_block = {
828 .hw_ops = &mmhub_v1_8_ras_hw_ops,
829 .ras_late_init = mmhub_v1_8_ras_late_init,
830 },
831 };
832