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#define mmSMNAID_AID0_MCA_SMU 0x03b30400
37
38static u64 mmhub_v1_8_get_fb_location(struct amdgpu_device *adev)
39{
40 u64 base = RREG32_SOC15(MMHUB, 0, regMC_VM_FB_LOCATION_BASE);
41 u64 top = RREG32_SOC15(MMHUB, 0, regMC_VM_FB_LOCATION_TOP);
42
43 base &= MC_VM_FB_LOCATION_BASE__FB_BASE_MASK;
44 base <<= 24;
45
46 top &= MC_VM_FB_LOCATION_TOP__FB_TOP_MASK;
47 top <<= 24;
48
49 adev->gmc.fb_start = base;
50 adev->gmc.fb_end = top;
51
52 return base;
53}
54
55static void mmhub_v1_8_setup_vm_pt_regs(struct amdgpu_device *adev, uint32_t vmid,
56 uint64_t page_table_base)
57{
58 struct amdgpu_vmhub *hub;
59 u32 inst_mask;
60 int i;
61
62 inst_mask = adev->aid_mask;
63 for_each_inst(i, inst_mask) {
64 hub = &adev->vmhub[AMDGPU_MMHUB0(i)];
65 WREG32_SOC15_OFFSET(MMHUB, i,
66 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32,
67 hub->ctx_addr_distance * vmid,
68 lower_32_bits(page_table_base));
69
70 WREG32_SOC15_OFFSET(MMHUB, i,
71 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32,
72 hub->ctx_addr_distance * vmid,
73 upper_32_bits(page_table_base));
74 }
75}
76
77static void mmhub_v1_8_init_gart_aperture_regs(struct amdgpu_device *adev)
78{
79 uint64_t pt_base;
80 u32 inst_mask;
81 int i;
82
83 if (adev->gmc.pdb0_bo)
84 pt_base = amdgpu_gmc_pd_addr(bo: adev->gmc.pdb0_bo);
85 else
86 pt_base = amdgpu_gmc_pd_addr(bo: adev->gart.bo);
87
88 mmhub_v1_8_setup_vm_pt_regs(adev, vmid: 0, page_table_base: pt_base);
89
90 /* If use GART for FB translation, vmid0 page table covers both
91 * vram and system memory (gart)
92 */
93 inst_mask = adev->aid_mask;
94 for_each_inst(i, inst_mask) {
95 if (adev->gmc.pdb0_bo) {
96 WREG32_SOC15(MMHUB, i,
97 regVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
98 (u32)(adev->gmc.fb_start >> 12));
99 WREG32_SOC15(MMHUB, i,
100 regVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
101 (u32)(adev->gmc.fb_start >> 44));
102
103 WREG32_SOC15(MMHUB, i,
104 regVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
105 (u32)(adev->gmc.gart_end >> 12));
106 WREG32_SOC15(MMHUB, i,
107 regVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
108 (u32)(adev->gmc.gart_end >> 44));
109
110 } else {
111 WREG32_SOC15(MMHUB, i,
112 regVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
113 (u32)(adev->gmc.gart_start >> 12));
114 WREG32_SOC15(MMHUB, i,
115 regVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
116 (u32)(adev->gmc.gart_start >> 44));
117
118 WREG32_SOC15(MMHUB, i,
119 regVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
120 (u32)(adev->gmc.gart_end >> 12));
121 WREG32_SOC15(MMHUB, i,
122 regVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
123 (u32)(adev->gmc.gart_end >> 44));
124 }
125 }
126}
127
128static void mmhub_v1_8_init_system_aperture_regs(struct amdgpu_device *adev)
129{
130 uint32_t tmp, inst_mask;
131 uint64_t value;
132 int i;
133
134 if (amdgpu_sriov_vf(adev))
135 return;
136
137 inst_mask = adev->aid_mask;
138 for_each_inst(i, inst_mask) {
139 /* Program the AGP BAR */
140 WREG32_SOC15(MMHUB, i, regMC_VM_AGP_BASE, 0);
141 WREG32_SOC15(MMHUB, i, regMC_VM_AGP_BOT,
142 adev->gmc.agp_start >> 24);
143 WREG32_SOC15(MMHUB, i, regMC_VM_AGP_TOP,
144 adev->gmc.agp_end >> 24);
145
146 /* Program the system aperture low logical page number. */
147 WREG32_SOC15(MMHUB, i, regMC_VM_SYSTEM_APERTURE_LOW_ADDR,
148 min(adev->gmc.fb_start, adev->gmc.agp_start) >> 18);
149
150 WREG32_SOC15(MMHUB, i, regMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
151 max(adev->gmc.fb_end, adev->gmc.agp_end) >> 18);
152
153 /* In the case squeezing vram into GART aperture, we don't use
154 * FB aperture and AGP aperture. Disable them.
155 */
156 if (adev->gmc.pdb0_bo) {
157 WREG32_SOC15(MMHUB, i, regMC_VM_AGP_BOT, 0xFFFFFF);
158 WREG32_SOC15(MMHUB, i, regMC_VM_AGP_TOP, 0);
159 WREG32_SOC15(MMHUB, i, regMC_VM_FB_LOCATION_TOP, 0);
160 WREG32_SOC15(MMHUB, i, regMC_VM_FB_LOCATION_BASE,
161 0x00FFFFFF);
162 WREG32_SOC15(MMHUB, i,
163 regMC_VM_SYSTEM_APERTURE_LOW_ADDR,
164 0x3FFFFFFF);
165 WREG32_SOC15(MMHUB, i,
166 regMC_VM_SYSTEM_APERTURE_HIGH_ADDR, 0);
167 }
168
169 /* Set default page address. */
170 value = amdgpu_gmc_vram_mc2pa(adev, mc_addr: adev->mem_scratch.gpu_addr);
171 WREG32_SOC15(MMHUB, i, regMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_LSB,
172 (u32)(value >> 12));
173 WREG32_SOC15(MMHUB, i, regMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_MSB,
174 (u32)(value >> 44));
175
176 /* Program "protection fault". */
177 WREG32_SOC15(MMHUB, i,
178 regVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_LO32,
179 (u32)(adev->dummy_page_addr >> 12));
180 WREG32_SOC15(MMHUB, i,
181 regVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_HI32,
182 (u32)((u64)adev->dummy_page_addr >> 44));
183
184 tmp = RREG32_SOC15(MMHUB, i, regVM_L2_PROTECTION_FAULT_CNTL2);
185 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL2,
186 ACTIVE_PAGE_MIGRATION_PTE_READ_RETRY, 1);
187 WREG32_SOC15(MMHUB, i, regVM_L2_PROTECTION_FAULT_CNTL2, tmp);
188 }
189}
190
191static void mmhub_v1_8_init_tlb_regs(struct amdgpu_device *adev)
192{
193 uint32_t tmp, inst_mask;
194 int i;
195
196 /* Setup TLB control */
197 inst_mask = adev->aid_mask;
198 for_each_inst(i, inst_mask) {
199 tmp = RREG32_SOC15(MMHUB, i, regMC_VM_MX_L1_TLB_CNTL);
200
201 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB,
202 1);
203 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
204 SYSTEM_ACCESS_MODE, 3);
205 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
206 ENABLE_ADVANCED_DRIVER_MODEL, 1);
207 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
208 SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
209 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
210 MTYPE, MTYPE_UC);/* XXX for emulation. */
211 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ATC_EN, 1);
212
213 WREG32_SOC15(MMHUB, i, regMC_VM_MX_L1_TLB_CNTL, tmp);
214 }
215}
216
217static void mmhub_v1_8_init_cache_regs(struct amdgpu_device *adev)
218{
219 uint32_t tmp, inst_mask;
220 int i;
221
222 if (amdgpu_sriov_vf(adev))
223 return;
224
225 /* Setup L2 cache */
226 inst_mask = adev->aid_mask;
227 for_each_inst(i, inst_mask) {
228 tmp = RREG32_SOC15(MMHUB, 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,
231 ENABLE_L2_FRAGMENT_PROCESSING, 1);
232 /* XXX for emulation, Refer to closed source code.*/
233 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,
234 L2_PDE0_CACHE_TAG_GENERATION_MODE, 0);
235 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION,
236 0);
237 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,
238 CONTEXT1_IDENTITY_ACCESS_MODE, 1);
239 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,
240 IDENTITY_MODE_FRAGMENT_SIZE, 0);
241 WREG32_SOC15(MMHUB, i, regVM_L2_CNTL, tmp);
242
243 tmp = RREG32_SOC15(MMHUB, i, regVM_L2_CNTL2);
244 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS,
245 1);
246 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
247 WREG32_SOC15(MMHUB, i, regVM_L2_CNTL2, tmp);
248
249 tmp = regVM_L2_CNTL3_DEFAULT;
250 if (adev->gmc.translate_further) {
251 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 12);
252 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3,
253 L2_CACHE_BIGK_FRAGMENT_SIZE, 9);
254 } else {
255 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 9);
256 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3,
257 L2_CACHE_BIGK_FRAGMENT_SIZE, 6);
258 }
259 WREG32_SOC15(MMHUB, i, regVM_L2_CNTL3, tmp);
260
261 tmp = regVM_L2_CNTL4_DEFAULT;
262 /* For AMD APP APUs setup WC memory */
263 if (adev->gmc.xgmi.connected_to_cpu || adev->gmc.is_app_apu) {
264 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4,
265 VMC_TAP_PDE_REQUEST_PHYSICAL, 1);
266 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4,
267 VMC_TAP_PTE_REQUEST_PHYSICAL, 1);
268 } else {
269 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4,
270 VMC_TAP_PDE_REQUEST_PHYSICAL, 0);
271 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4,
272 VMC_TAP_PTE_REQUEST_PHYSICAL, 0);
273 }
274 WREG32_SOC15(MMHUB, i, regVM_L2_CNTL4, tmp);
275 }
276}
277
278static void mmhub_v1_8_enable_system_domain(struct amdgpu_device *adev)
279{
280 uint32_t tmp, inst_mask;
281 int i;
282
283 inst_mask = adev->aid_mask;
284 for_each_inst(i, inst_mask) {
285 tmp = RREG32_SOC15(MMHUB, i, regVM_CONTEXT0_CNTL);
286 tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
287 tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH,
288 adev->gmc.vmid0_page_table_depth);
289 tmp = REG_SET_FIELD(tmp,
290 VM_CONTEXT0_CNTL, PAGE_TABLE_BLOCK_SIZE,
291 adev->gmc.vmid0_page_table_block_size);
292 tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL,
293 RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 0);
294 WREG32_SOC15(MMHUB, i, regVM_CONTEXT0_CNTL, tmp);
295 }
296}
297
298static void mmhub_v1_8_disable_identity_aperture(struct amdgpu_device *adev)
299{
300 u32 inst_mask;
301 int i;
302
303 if (amdgpu_sriov_vf(adev))
304 return;
305
306 inst_mask = adev->aid_mask;
307 for_each_inst(i, inst_mask) {
308 WREG32_SOC15(MMHUB, i,
309 regVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_LO32,
310 0XFFFFFFFF);
311 WREG32_SOC15(MMHUB, i,
312 regVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_HI32,
313 0x0000000F);
314
315 WREG32_SOC15(MMHUB, i,
316 regVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_LO32,
317 0);
318 WREG32_SOC15(MMHUB, i,
319 regVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_HI32,
320 0);
321
322 WREG32_SOC15(MMHUB, i,
323 regVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_LO32, 0);
324 WREG32_SOC15(MMHUB, i,
325 regVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_HI32, 0);
326 }
327}
328
329static void mmhub_v1_8_setup_vmid_config(struct amdgpu_device *adev)
330{
331 struct amdgpu_vmhub *hub;
332 unsigned int num_level, block_size;
333 uint32_t tmp, inst_mask;
334 int i, j;
335
336 num_level = adev->vm_manager.num_level;
337 block_size = adev->vm_manager.block_size;
338 if (adev->gmc.translate_further)
339 num_level -= 1;
340 else
341 block_size -= 9;
342
343 inst_mask = adev->aid_mask;
344 for_each_inst(j, inst_mask) {
345 hub = &adev->vmhub[AMDGPU_MMHUB0(j)];
346 for (i = 0; i <= 14; i++) {
347 tmp = RREG32_SOC15_OFFSET(MMHUB, j, regVM_CONTEXT1_CNTL,
348 i * hub->ctx_distance);
349 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
350 ENABLE_CONTEXT, 1);
351 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
352 PAGE_TABLE_DEPTH, num_level);
353 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
354 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
355 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
356 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
357 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
358 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
359 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
360 VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
361 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
362 READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
363 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
364 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
365 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
366 EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
367 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
368 PAGE_TABLE_BLOCK_SIZE,
369 block_size);
370 /* On 9.4.3, XNACK can be enabled in the SQ
371 * per-process. Retry faults need to be enabled for
372 * that to work.
373 */
374 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
375 RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 1);
376 WREG32_SOC15_OFFSET(MMHUB, j, regVM_CONTEXT1_CNTL,
377 i * hub->ctx_distance, tmp);
378 WREG32_SOC15_OFFSET(MMHUB, j,
379 regVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32,
380 i * hub->ctx_addr_distance, 0);
381 WREG32_SOC15_OFFSET(MMHUB, j,
382 regVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32,
383 i * hub->ctx_addr_distance, 0);
384 WREG32_SOC15_OFFSET(MMHUB, j,
385 regVM_CONTEXT1_PAGE_TABLE_END_ADDR_LO32,
386 i * hub->ctx_addr_distance,
387 lower_32_bits(adev->vm_manager.max_pfn - 1));
388 WREG32_SOC15_OFFSET(MMHUB, j,
389 regVM_CONTEXT1_PAGE_TABLE_END_ADDR_HI32,
390 i * hub->ctx_addr_distance,
391 upper_32_bits(adev->vm_manager.max_pfn - 1));
392 }
393 }
394}
395
396static void mmhub_v1_8_program_invalidation(struct amdgpu_device *adev)
397{
398 struct amdgpu_vmhub *hub;
399 u32 i, j, inst_mask;
400
401 inst_mask = adev->aid_mask;
402 for_each_inst(j, inst_mask) {
403 hub = &adev->vmhub[AMDGPU_MMHUB0(j)];
404 for (i = 0; i < 18; ++i) {
405 WREG32_SOC15_OFFSET(MMHUB, j,
406 regVM_INVALIDATE_ENG0_ADDR_RANGE_LO32,
407 i * hub->eng_addr_distance, 0xffffffff);
408 WREG32_SOC15_OFFSET(MMHUB, j,
409 regVM_INVALIDATE_ENG0_ADDR_RANGE_HI32,
410 i * hub->eng_addr_distance, 0x1f);
411 }
412 }
413}
414
415static int mmhub_v1_8_gart_enable(struct amdgpu_device *adev)
416{
417 /* GART Enable. */
418 mmhub_v1_8_init_gart_aperture_regs(adev);
419 mmhub_v1_8_init_system_aperture_regs(adev);
420 mmhub_v1_8_init_tlb_regs(adev);
421 mmhub_v1_8_init_cache_regs(adev);
422
423 mmhub_v1_8_enable_system_domain(adev);
424 mmhub_v1_8_disable_identity_aperture(adev);
425 mmhub_v1_8_setup_vmid_config(adev);
426 mmhub_v1_8_program_invalidation(adev);
427
428 return 0;
429}
430
431static void mmhub_v1_8_gart_disable(struct amdgpu_device *adev)
432{
433 struct amdgpu_vmhub *hub;
434 u32 tmp;
435 u32 i, j, inst_mask;
436
437 /* Disable all tables */
438 inst_mask = adev->aid_mask;
439 for_each_inst(j, inst_mask) {
440 hub = &adev->vmhub[AMDGPU_MMHUB0(j)];
441 for (i = 0; i < 16; i++)
442 WREG32_SOC15_OFFSET(MMHUB, j, regVM_CONTEXT0_CNTL,
443 i * hub->ctx_distance, 0);
444
445 /* Setup TLB control */
446 tmp = RREG32_SOC15(MMHUB, j, regMC_VM_MX_L1_TLB_CNTL);
447 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB,
448 0);
449 tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
450 ENABLE_ADVANCED_DRIVER_MODEL, 0);
451 WREG32_SOC15(MMHUB, j, regMC_VM_MX_L1_TLB_CNTL, tmp);
452
453 if (!amdgpu_sriov_vf(adev)) {
454 /* Setup L2 cache */
455 tmp = RREG32_SOC15(MMHUB, j, regVM_L2_CNTL);
456 tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE,
457 0);
458 WREG32_SOC15(MMHUB, j, regVM_L2_CNTL, tmp);
459 WREG32_SOC15(MMHUB, j, regVM_L2_CNTL3, 0);
460 }
461 }
462}
463
464/**
465 * mmhub_v1_8_set_fault_enable_default - update GART/VM fault handling
466 *
467 * @adev: amdgpu_device pointer
468 * @value: true redirects VM faults to the default page
469 */
470static void mmhub_v1_8_set_fault_enable_default(struct amdgpu_device *adev, bool value)
471{
472 u32 tmp, inst_mask;
473 int i;
474
475 if (amdgpu_sriov_vf(adev))
476 return;
477
478 inst_mask = adev->aid_mask;
479 for_each_inst(i, inst_mask) {
480 tmp = RREG32_SOC15(MMHUB, i, regVM_L2_PROTECTION_FAULT_CNTL);
481 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
482 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
483 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
484 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
485 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
486 PDE1_PROTECTION_FAULT_ENABLE_DEFAULT, value);
487 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
488 PDE2_PROTECTION_FAULT_ENABLE_DEFAULT, value);
489 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
490 TRANSLATE_FURTHER_PROTECTION_FAULT_ENABLE_DEFAULT,
491 value);
492 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
493 NACK_PROTECTION_FAULT_ENABLE_DEFAULT, value);
494 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
495 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
496 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
497 VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
498 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
499 READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
500 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
501 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
502 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
503 EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
504 if (!value) {
505 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
506 CRASH_ON_NO_RETRY_FAULT, 1);
507 tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
508 CRASH_ON_RETRY_FAULT, 1);
509 }
510
511 WREG32_SOC15(MMHUB, i, regVM_L2_PROTECTION_FAULT_CNTL, tmp);
512 }
513}
514
515static void mmhub_v1_8_init(struct amdgpu_device *adev)
516{
517 struct amdgpu_vmhub *hub;
518 u32 inst_mask;
519 int i;
520
521 inst_mask = adev->aid_mask;
522 for_each_inst(i, inst_mask) {
523 hub = &adev->vmhub[AMDGPU_MMHUB0(i)];
524
525 hub->ctx0_ptb_addr_lo32 = SOC15_REG_OFFSET(MMHUB, i,
526 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32);
527 hub->ctx0_ptb_addr_hi32 = SOC15_REG_OFFSET(MMHUB, i,
528 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32);
529 hub->vm_inv_eng0_req =
530 SOC15_REG_OFFSET(MMHUB, i, regVM_INVALIDATE_ENG0_REQ);
531 hub->vm_inv_eng0_ack =
532 SOC15_REG_OFFSET(MMHUB, i, regVM_INVALIDATE_ENG0_ACK);
533 hub->vm_context0_cntl =
534 SOC15_REG_OFFSET(MMHUB, i, regVM_CONTEXT0_CNTL);
535 hub->vm_l2_pro_fault_status = SOC15_REG_OFFSET(MMHUB, i,
536 regVM_L2_PROTECTION_FAULT_STATUS);
537 hub->vm_l2_pro_fault_cntl = SOC15_REG_OFFSET(MMHUB, i,
538 regVM_L2_PROTECTION_FAULT_CNTL);
539
540 hub->ctx_distance = regVM_CONTEXT1_CNTL - regVM_CONTEXT0_CNTL;
541 hub->ctx_addr_distance =
542 regVM_CONTEXT1_PAGE_TABLE_BASE_ADDR_LO32 -
543 regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32;
544 hub->eng_distance = regVM_INVALIDATE_ENG1_REQ -
545 regVM_INVALIDATE_ENG0_REQ;
546 hub->eng_addr_distance = regVM_INVALIDATE_ENG1_ADDR_RANGE_LO32 -
547 regVM_INVALIDATE_ENG0_ADDR_RANGE_LO32;
548 }
549}
550
551static int mmhub_v1_8_set_clockgating(struct amdgpu_device *adev,
552 enum amd_clockgating_state state)
553{
554 return 0;
555}
556
557static void mmhub_v1_8_get_clockgating(struct amdgpu_device *adev, u64 *flags)
558{
559
560}
561
562const struct amdgpu_mmhub_funcs mmhub_v1_8_funcs = {
563 .get_fb_location = mmhub_v1_8_get_fb_location,
564 .init = mmhub_v1_8_init,
565 .gart_enable = mmhub_v1_8_gart_enable,
566 .set_fault_enable_default = mmhub_v1_8_set_fault_enable_default,
567 .gart_disable = mmhub_v1_8_gart_disable,
568 .setup_vm_pt_regs = mmhub_v1_8_setup_vm_pt_regs,
569 .set_clockgating = mmhub_v1_8_set_clockgating,
570 .get_clockgating = mmhub_v1_8_get_clockgating,
571};
572
573static const struct amdgpu_ras_err_status_reg_entry mmhub_v1_8_ce_reg_list[] = {
574 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA0_CE_ERR_STATUS_LO, regMMEA0_CE_ERR_STATUS_HI),
575 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA0"},
576 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA1_CE_ERR_STATUS_LO, regMMEA1_CE_ERR_STATUS_HI),
577 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA1"},
578 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA2_CE_ERR_STATUS_LO, regMMEA2_CE_ERR_STATUS_HI),
579 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA2"},
580 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA3_CE_ERR_STATUS_LO, regMMEA3_CE_ERR_STATUS_HI),
581 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA3"},
582 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA4_CE_ERR_STATUS_LO, regMMEA4_CE_ERR_STATUS_HI),
583 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA4"},
584 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMM_CANE_CE_ERR_STATUS_LO, regMM_CANE_CE_ERR_STATUS_HI),
585 1, 0, "MM_CANE"},
586};
587
588static const struct amdgpu_ras_err_status_reg_entry mmhub_v1_8_ue_reg_list[] = {
589 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA0_UE_ERR_STATUS_LO, regMMEA0_UE_ERR_STATUS_HI),
590 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA0"},
591 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA1_UE_ERR_STATUS_LO, regMMEA1_UE_ERR_STATUS_HI),
592 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA1"},
593 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA2_UE_ERR_STATUS_LO, regMMEA2_UE_ERR_STATUS_HI),
594 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA2"},
595 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA3_UE_ERR_STATUS_LO, regMMEA3_UE_ERR_STATUS_HI),
596 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA3"},
597 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMMEA4_UE_ERR_STATUS_LO, regMMEA4_UE_ERR_STATUS_HI),
598 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "MMEA4"},
599 {AMDGPU_RAS_REG_ENTRY(MMHUB, 0, regMM_CANE_UE_ERR_STATUS_LO, regMM_CANE_UE_ERR_STATUS_HI),
600 1, 0, "MM_CANE"},
601};
602
603static const struct amdgpu_ras_memory_id_entry mmhub_v1_8_ras_memory_list[] = {
604 {AMDGPU_MMHUB_WGMI_PAGEMEM, "MMEA_WGMI_PAGEMEM"},
605 {AMDGPU_MMHUB_RGMI_PAGEMEM, "MMEA_RGMI_PAGEMEM"},
606 {AMDGPU_MMHUB_WDRAM_PAGEMEM, "MMEA_WDRAM_PAGEMEM"},
607 {AMDGPU_MMHUB_RDRAM_PAGEMEM, "MMEA_RDRAM_PAGEMEM"},
608 {AMDGPU_MMHUB_WIO_CMDMEM, "MMEA_WIO_CMDMEM"},
609 {AMDGPU_MMHUB_RIO_CMDMEM, "MMEA_RIO_CMDMEM"},
610 {AMDGPU_MMHUB_WGMI_CMDMEM, "MMEA_WGMI_CMDMEM"},
611 {AMDGPU_MMHUB_RGMI_CMDMEM, "MMEA_RGMI_CMDMEM"},
612 {AMDGPU_MMHUB_WDRAM_CMDMEM, "MMEA_WDRAM_CMDMEM"},
613 {AMDGPU_MMHUB_RDRAM_CMDMEM, "MMEA_RDRAM_CMDMEM"},
614 {AMDGPU_MMHUB_MAM_DMEM0, "MMEA_MAM_DMEM0"},
615 {AMDGPU_MMHUB_MAM_DMEM1, "MMEA_MAM_DMEM1"},
616 {AMDGPU_MMHUB_MAM_DMEM2, "MMEA_MAM_DMEM2"},
617 {AMDGPU_MMHUB_MAM_DMEM3, "MMEA_MAM_DMEM3"},
618 {AMDGPU_MMHUB_WRET_TAGMEM, "MMEA_WRET_TAGMEM"},
619 {AMDGPU_MMHUB_RRET_TAGMEM, "MMEA_RRET_TAGMEM"},
620 {AMDGPU_MMHUB_WIO_DATAMEM, "MMEA_WIO_DATAMEM"},
621 {AMDGPU_MMHUB_WGMI_DATAMEM, "MMEA_WGMI_DATAMEM"},
622 {AMDGPU_MMHUB_WDRAM_DATAMEM, "MMEA_WDRAM_DATAMEM"},
623};
624
625static void mmhub_v1_8_inst_query_ras_error_count(struct amdgpu_device *adev,
626 uint32_t mmhub_inst,
627 void *ras_err_status)
628{
629 struct ras_err_data *err_data = (struct ras_err_data *)ras_err_status;
630 unsigned long ue_count = 0, ce_count = 0;
631
632 /* NOTE: mmhub is converted by aid_mask and the range is 0-3,
633 * which can be used as die ID directly */
634 struct amdgpu_smuio_mcm_config_info mcm_info = {
635 .socket_id = adev->smuio.funcs->get_socket_id(adev),
636 .die_id = mmhub_inst,
637 };
638
639 amdgpu_ras_inst_query_ras_error_count(adev,
640 reg_list: mmhub_v1_8_ce_reg_list,
641 ARRAY_SIZE(mmhub_v1_8_ce_reg_list),
642 mem_list: mmhub_v1_8_ras_memory_list,
643 ARRAY_SIZE(mmhub_v1_8_ras_memory_list),
644 instance: mmhub_inst,
645 err_type: AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE,
646 err_count: &ce_count);
647 amdgpu_ras_inst_query_ras_error_count(adev,
648 reg_list: mmhub_v1_8_ue_reg_list,
649 ARRAY_SIZE(mmhub_v1_8_ue_reg_list),
650 mem_list: mmhub_v1_8_ras_memory_list,
651 ARRAY_SIZE(mmhub_v1_8_ras_memory_list),
652 instance: mmhub_inst,
653 err_type: AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
654 err_count: &ue_count);
655
656 amdgpu_ras_error_statistic_ce_count(err_data, mcm_info: &mcm_info, NULL, count: ce_count);
657 amdgpu_ras_error_statistic_ue_count(err_data, mcm_info: &mcm_info, NULL, count: ue_count);
658}
659
660static void mmhub_v1_8_query_ras_error_count(struct amdgpu_device *adev,
661 void *ras_err_status)
662{
663 uint32_t inst_mask;
664 uint32_t i;
665
666 if (!amdgpu_ras_is_supported(adev, block: AMDGPU_RAS_BLOCK__MMHUB)) {
667 dev_warn(adev->dev, "MMHUB RAS is not supported\n");
668 return;
669 }
670
671 inst_mask = adev->aid_mask;
672 for_each_inst(i, inst_mask)
673 mmhub_v1_8_inst_query_ras_error_count(adev, mmhub_inst: i, ras_err_status);
674}
675
676static void mmhub_v1_8_inst_reset_ras_error_count(struct amdgpu_device *adev,
677 uint32_t mmhub_inst)
678{
679 amdgpu_ras_inst_reset_ras_error_count(adev,
680 reg_list: mmhub_v1_8_ce_reg_list,
681 ARRAY_SIZE(mmhub_v1_8_ce_reg_list),
682 instance: mmhub_inst);
683 amdgpu_ras_inst_reset_ras_error_count(adev,
684 reg_list: mmhub_v1_8_ue_reg_list,
685 ARRAY_SIZE(mmhub_v1_8_ue_reg_list),
686 instance: mmhub_inst);
687}
688
689static void mmhub_v1_8_reset_ras_error_count(struct amdgpu_device *adev)
690{
691 uint32_t inst_mask;
692 uint32_t i;
693
694 if (!amdgpu_ras_is_supported(adev, block: AMDGPU_RAS_BLOCK__MMHUB)) {
695 dev_warn(adev->dev, "MMHUB RAS is not supported\n");
696 return;
697 }
698
699 inst_mask = adev->aid_mask;
700 for_each_inst(i, inst_mask)
701 mmhub_v1_8_inst_reset_ras_error_count(adev, mmhub_inst: i);
702}
703
704static const struct amdgpu_ras_block_hw_ops mmhub_v1_8_ras_hw_ops = {
705 .query_ras_error_count = mmhub_v1_8_query_ras_error_count,
706 .reset_ras_error_count = mmhub_v1_8_reset_ras_error_count,
707};
708
709static int mmhub_v1_8_aca_bank_generate_report(struct aca_handle *handle,
710 struct aca_bank *bank, enum aca_error_type type,
711 struct aca_bank_report *report, void *data)
712{
713 u64 status, misc0;
714 int ret;
715
716 status = bank->regs[ACA_REG_IDX_STATUS];
717 if ((type == ACA_ERROR_TYPE_UE &&
718 ACA_REG__STATUS__ERRORCODEEXT(status) == ACA_EXTERROR_CODE_FAULT) ||
719 (type == ACA_ERROR_TYPE_CE &&
720 ACA_REG__STATUS__ERRORCODEEXT(status) == ACA_EXTERROR_CODE_CE)) {
721
722 ret = aca_bank_info_decode(bank, info: &report->info);
723 if (ret)
724 return ret;
725
726 misc0 = bank->regs[ACA_REG_IDX_MISC0];
727 report->count[type] = ACA_REG__MISC0__ERRCNT(misc0);
728 }
729
730 return 0;
731}
732
733/* reference to smu driver if header file */
734static int mmhub_v1_8_err_codes[] = {
735 0, 1, 2, 3, 4, /* CODE_DAGB0 - 4 */
736 5, 6, 7, 8, 9, /* CODE_EA0 - 4 */
737 10, /* CODE_UTCL2_ROUTER */
738 11, /* CODE_VML2 */
739 12, /* CODE_VML2_WALKER */
740 13, /* CODE_MMCANE */
741};
742
743static bool mmhub_v1_8_aca_bank_is_valid(struct aca_handle *handle, struct aca_bank *bank,
744 enum aca_error_type type, void *data)
745{
746 u32 instlo;
747
748 instlo = ACA_REG__IPID__INSTANCEIDLO(bank->regs[ACA_REG_IDX_IPID]);
749 instlo &= GENMASK(31, 1);
750
751 if (instlo != mmSMNAID_AID0_MCA_SMU)
752 return false;
753
754 if (aca_bank_check_error_codes(adev: handle->adev, bank,
755 err_codes: mmhub_v1_8_err_codes,
756 ARRAY_SIZE(mmhub_v1_8_err_codes)))
757 return false;
758
759 return true;
760}
761
762static const struct aca_bank_ops mmhub_v1_8_aca_bank_ops = {
763 .aca_bank_generate_report = mmhub_v1_8_aca_bank_generate_report,
764 .aca_bank_is_valid = mmhub_v1_8_aca_bank_is_valid,
765};
766
767static const struct aca_info mmhub_v1_8_aca_info = {
768 .hwip = ACA_HWIP_TYPE_SMU,
769 .mask = ACA_ERROR_UE_MASK,
770 .bank_ops = &mmhub_v1_8_aca_bank_ops,
771};
772
773static int mmhub_v1_8_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block)
774{
775 int r;
776
777 r = amdgpu_ras_block_late_init(adev, ras_block);
778 if (r)
779 return r;
780
781 r = amdgpu_ras_bind_aca(adev, blk: AMDGPU_RAS_BLOCK__MMHUB,
782 aca_info: &mmhub_v1_8_aca_info, NULL);
783 if (r)
784 goto late_fini;
785
786 return 0;
787
788late_fini:
789 amdgpu_ras_block_late_fini(adev, ras_block);
790
791 return r;
792}
793
794struct amdgpu_mmhub_ras mmhub_v1_8_ras = {
795 .ras_block = {
796 .hw_ops = &mmhub_v1_8_ras_hw_ops,
797 .ras_late_init = mmhub_v1_8_ras_late_init,
798 },
799};
800

source code of linux/drivers/gpu/drm/amd/amdgpu/mmhub_v1_8.c