2 * Copyright 2014 Advanced Micro Devices, Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
23 #include <linux/firmware.h>
27 #include "amdgpu_ucode.h"
29 #include "bif/bif_3_0_d.h"
30 #include "bif/bif_3_0_sh_mask.h"
31 #include "oss/oss_1_0_d.h"
32 #include "oss/oss_1_0_sh_mask.h"
33 #include "gmc/gmc_6_0_d.h"
34 #include "gmc/gmc_6_0_sh_mask.h"
35 #include "dce/dce_6_0_d.h"
36 #include "dce/dce_6_0_sh_mask.h"
39 static void gmc_v6_0_set_gart_funcs(struct amdgpu_device *adev);
40 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev);
41 static int gmc_v6_0_wait_for_idle(void *handle);
43 MODULE_FIRMWARE("radeon/tahiti_mc.bin");
44 MODULE_FIRMWARE("radeon/pitcairn_mc.bin");
45 MODULE_FIRMWARE("radeon/verde_mc.bin");
46 MODULE_FIRMWARE("radeon/oland_mc.bin");
47 MODULE_FIRMWARE("radeon/si58_mc.bin");
49 #define MC_SEQ_MISC0__MT__MASK 0xf0000000
50 #define MC_SEQ_MISC0__MT__GDDR1 0x10000000
51 #define MC_SEQ_MISC0__MT__DDR2 0x20000000
52 #define MC_SEQ_MISC0__MT__GDDR3 0x30000000
53 #define MC_SEQ_MISC0__MT__GDDR4 0x40000000
54 #define MC_SEQ_MISC0__MT__GDDR5 0x50000000
55 #define MC_SEQ_MISC0__MT__HBM 0x60000000
56 #define MC_SEQ_MISC0__MT__DDR3 0xB0000000
59 static const u32 crtc_offsets[6] =
61 SI_CRTC0_REGISTER_OFFSET,
62 SI_CRTC1_REGISTER_OFFSET,
63 SI_CRTC2_REGISTER_OFFSET,
64 SI_CRTC3_REGISTER_OFFSET,
65 SI_CRTC4_REGISTER_OFFSET,
66 SI_CRTC5_REGISTER_OFFSET
69 static void gmc_v6_0_mc_stop(struct amdgpu_device *adev,
70 struct amdgpu_mode_mc_save *save)
74 if (adev->mode_info.num_crtc)
75 amdgpu_display_stop_mc_access(adev, save);
77 gmc_v6_0_wait_for_idle((void *)adev);
79 blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
80 if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
81 /* Block CPU access */
82 WREG32(mmBIF_FB_EN, 0);
84 blackout = REG_SET_FIELD(blackout,
85 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
86 WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout | 1);
88 /* wait for the MC to settle */
93 static void gmc_v6_0_mc_resume(struct amdgpu_device *adev,
94 struct amdgpu_mode_mc_save *save)
98 /* unblackout the MC */
99 tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
100 tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
101 WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
102 /* allow CPU access */
103 tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
104 tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
105 WREG32(mmBIF_FB_EN, tmp);
107 if (adev->mode_info.num_crtc)
108 amdgpu_display_resume_mc_access(adev, save);
112 static int gmc_v6_0_init_microcode(struct amdgpu_device *adev)
114 const char *chip_name;
117 bool is_58_fw = false;
121 switch (adev->asic_type) {
123 chip_name = "tahiti";
126 chip_name = "pitcairn";
135 chip_name = "hainan";
140 /* this memory configuration requires special firmware */
141 if (((RREG32(mmMC_SEQ_MISC0) & 0xff000000) >> 24) == 0x58)
145 snprintf(fw_name, sizeof(fw_name), "radeon/si58_mc.bin");
147 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
148 err = request_firmware(&adev->mc.fw, fw_name, adev->dev);
152 err = amdgpu_ucode_validate(adev->mc.fw);
157 "si_mc: Failed to load firmware \"%s\"\n",
159 release_firmware(adev->mc.fw);
165 static int gmc_v6_0_mc_load_microcode(struct amdgpu_device *adev)
167 const __le32 *new_fw_data = NULL;
169 const __le32 *new_io_mc_regs = NULL;
170 int i, regs_size, ucode_size;
171 const struct mc_firmware_header_v1_0 *hdr;
176 hdr = (const struct mc_firmware_header_v1_0 *)adev->mc.fw->data;
178 amdgpu_ucode_print_mc_hdr(&hdr->header);
180 adev->mc.fw_version = le32_to_cpu(hdr->header.ucode_version);
181 regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
182 new_io_mc_regs = (const __le32 *)
183 (adev->mc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
184 ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
185 new_fw_data = (const __le32 *)
186 (adev->mc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
188 running = RREG32(mmMC_SEQ_SUP_CNTL) & MC_SEQ_SUP_CNTL__RUN_MASK;
192 /* reset the engine and set to writable */
193 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
194 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
196 /* load mc io regs */
197 for (i = 0; i < regs_size; i++) {
198 WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(new_io_mc_regs++));
199 WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(new_io_mc_regs++));
201 /* load the MC ucode */
202 for (i = 0; i < ucode_size; i++) {
203 WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(new_fw_data++));
206 /* put the engine back into the active state */
207 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
208 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
209 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
211 /* wait for training to complete */
212 for (i = 0; i < adev->usec_timeout; i++) {
213 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D0_MASK)
217 for (i = 0; i < adev->usec_timeout; i++) {
218 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D1_MASK)
228 static void gmc_v6_0_vram_gtt_location(struct amdgpu_device *adev,
229 struct amdgpu_mc *mc)
231 if (mc->mc_vram_size > 0xFFC0000000ULL) {
232 dev_warn(adev->dev, "limiting VRAM\n");
233 mc->real_vram_size = 0xFFC0000000ULL;
234 mc->mc_vram_size = 0xFFC0000000ULL;
236 amdgpu_vram_location(adev, &adev->mc, 0);
237 adev->mc.gtt_base_align = 0;
238 amdgpu_gtt_location(adev, mc);
241 static void gmc_v6_0_mc_program(struct amdgpu_device *adev)
243 struct amdgpu_mode_mc_save save;
248 for (i = 0, j = 0; i < 32; i++, j += 0x6) {
249 WREG32((0xb05 + j), 0x00000000);
250 WREG32((0xb06 + j), 0x00000000);
251 WREG32((0xb07 + j), 0x00000000);
252 WREG32((0xb08 + j), 0x00000000);
253 WREG32((0xb09 + j), 0x00000000);
255 WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
257 if (adev->mode_info.num_crtc)
258 amdgpu_display_set_vga_render_state(adev, false);
260 gmc_v6_0_mc_stop(adev, &save);
262 if (gmc_v6_0_wait_for_idle((void *)adev)) {
263 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
266 WREG32(mmVGA_HDP_CONTROL, VGA_HDP_CONTROL__VGA_MEMORY_DISABLE_MASK);
267 /* Update configuration */
268 WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
269 adev->mc.vram_start >> 12);
270 WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
271 adev->mc.vram_end >> 12);
272 WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
273 adev->vram_scratch.gpu_addr >> 12);
274 tmp = ((adev->mc.vram_end >> 24) & 0xFFFF) << 16;
275 tmp |= ((adev->mc.vram_start >> 24) & 0xFFFF);
276 WREG32(mmMC_VM_FB_LOCATION, tmp);
277 /* XXX double check these! */
278 WREG32(mmHDP_NONSURFACE_BASE, (adev->mc.vram_start >> 8));
279 WREG32(mmHDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
280 WREG32(mmHDP_NONSURFACE_SIZE, 0x3FFFFFFF);
281 WREG32(mmMC_VM_AGP_BASE, 0);
282 WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
283 WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
285 if (gmc_v6_0_wait_for_idle((void *)adev)) {
286 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
288 gmc_v6_0_mc_resume(adev, &save);
291 static int gmc_v6_0_mc_init(struct amdgpu_device *adev)
295 int chansize, numchan;
297 tmp = RREG32(mmMC_ARB_RAMCFG);
298 if (tmp & (1 << 11)) {
300 } else if (tmp & MC_ARB_RAMCFG__CHANSIZE_MASK) {
305 tmp = RREG32(mmMC_SHARED_CHMAP);
306 switch ((tmp & MC_SHARED_CHMAP__NOOFCHAN_MASK) >> MC_SHARED_CHMAP__NOOFCHAN__SHIFT) {
336 adev->mc.vram_width = numchan * chansize;
337 /* Could aper size report 0 ? */
338 adev->mc.aper_base = pci_resource_start(adev->pdev, 0);
339 adev->mc.aper_size = pci_resource_len(adev->pdev, 0);
340 /* size in MB on si */
341 adev->mc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
342 adev->mc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
343 adev->mc.visible_vram_size = adev->mc.aper_size;
345 /* unless the user had overridden it, set the gart
346 * size equal to the 1024 or vram, whichever is larger.
348 if (amdgpu_gart_size == -1)
349 adev->mc.gtt_size = max((1024ULL << 20), adev->mc.mc_vram_size);
351 adev->mc.gtt_size = (uint64_t)amdgpu_gart_size << 20;
353 gmc_v6_0_vram_gtt_location(adev, &adev->mc);
358 static void gmc_v6_0_gart_flush_gpu_tlb(struct amdgpu_device *adev,
361 WREG32(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0);
363 WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
366 static int gmc_v6_0_gart_set_pte_pde(struct amdgpu_device *adev,
368 uint32_t gpu_page_idx,
372 void __iomem *ptr = (void *)cpu_pt_addr;
375 value = addr & 0xFFFFFFFFFFFFF000ULL;
377 writeq(value, ptr + (gpu_page_idx * 8));
382 static uint64_t gmc_v6_0_get_vm_pte_flags(struct amdgpu_device *adev,
385 uint64_t pte_flag = 0;
387 if (flags & AMDGPU_VM_PAGE_READABLE)
388 pte_flag |= AMDGPU_PTE_READABLE;
389 if (flags & AMDGPU_VM_PAGE_WRITEABLE)
390 pte_flag |= AMDGPU_PTE_WRITEABLE;
391 if (flags & AMDGPU_VM_PAGE_PRT)
392 pte_flag |= AMDGPU_PTE_PRT;
397 static void gmc_v6_0_set_fault_enable_default(struct amdgpu_device *adev,
402 tmp = RREG32(mmVM_CONTEXT1_CNTL);
403 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
404 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
405 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
406 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
407 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
408 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
409 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
410 VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
411 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
412 READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
413 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
414 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
415 WREG32(mmVM_CONTEXT1_CNTL, tmp);
419 + * gmc_v8_0_set_prt - set PRT VM fault
421 + * @adev: amdgpu_device pointer
422 + * @enable: enable/disable VM fault handling for PRT
424 static void gmc_v6_0_set_prt(struct amdgpu_device *adev, bool enable)
428 if (enable && !adev->mc.prt_warning) {
429 dev_warn(adev->dev, "Disabling VM faults because of PRT request!\n");
430 adev->mc.prt_warning = true;
433 tmp = RREG32(mmVM_PRT_CNTL);
434 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
435 CB_DISABLE_FAULT_ON_UNMAPPED_ACCESS,
437 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
438 TC_DISABLE_FAULT_ON_UNMAPPED_ACCESS,
440 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
441 L2_CACHE_STORE_INVALID_ENTRIES,
443 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
444 L1_TLB_STORE_INVALID_ENTRIES,
446 WREG32(mmVM_PRT_CNTL, tmp);
449 uint32_t low = AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT;
450 uint32_t high = adev->vm_manager.max_pfn;
452 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, low);
453 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, low);
454 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, low);
455 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, low);
456 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, high);
457 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, high);
458 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, high);
459 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, high);
461 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, 0xfffffff);
462 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, 0xfffffff);
463 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, 0xfffffff);
464 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, 0xfffffff);
465 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, 0x0);
466 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, 0x0);
467 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, 0x0);
468 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, 0x0);
472 static int gmc_v6_0_gart_enable(struct amdgpu_device *adev)
476 if (adev->gart.robj == NULL) {
477 dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
480 r = amdgpu_gart_table_vram_pin(adev);
483 /* Setup TLB control */
484 WREG32(mmMC_VM_MX_L1_TLB_CNTL,
486 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_TLB_MASK |
487 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_FRAGMENT_PROCESSING_MASK |
488 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
489 MC_VM_MX_L1_TLB_CNTL__ENABLE_ADVANCED_DRIVER_MODEL_MASK |
490 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
493 VM_L2_CNTL__ENABLE_L2_CACHE_MASK |
494 VM_L2_CNTL__ENABLE_L2_FRAGMENT_PROCESSING_MASK |
495 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
496 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
497 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
498 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
499 WREG32(mmVM_L2_CNTL2,
500 VM_L2_CNTL2__INVALIDATE_ALL_L1_TLBS_MASK |
501 VM_L2_CNTL2__INVALIDATE_L2_CACHE_MASK);
502 WREG32(mmVM_L2_CNTL3,
503 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
504 (4UL << VM_L2_CNTL3__BANK_SELECT__SHIFT) |
505 (4UL << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
507 WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->mc.gtt_start >> 12);
508 WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->mc.gtt_end >> 12);
509 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12);
510 WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
511 (u32)(adev->dummy_page.addr >> 12));
512 WREG32(mmVM_CONTEXT0_CNTL2, 0);
513 WREG32(mmVM_CONTEXT0_CNTL,
514 VM_CONTEXT0_CNTL__ENABLE_CONTEXT_MASK |
515 (0UL << VM_CONTEXT0_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
516 VM_CONTEXT0_CNTL__RANGE_PROTECTION_FAULT_ENABLE_DEFAULT_MASK);
522 /* empty context1-15 */
523 /* set vm size, must be a multiple of 4 */
524 WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
525 WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
526 /* Assign the pt base to something valid for now; the pts used for
527 * the VMs are determined by the application and setup and assigned
528 * on the fly in the vm part of radeon_gart.c
530 for (i = 1; i < 16; i++) {
532 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
533 adev->gart.table_addr >> 12);
535 WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
536 adev->gart.table_addr >> 12);
539 /* enable context1-15 */
540 WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
541 (u32)(adev->dummy_page.addr >> 12));
542 WREG32(mmVM_CONTEXT1_CNTL2, 4);
543 WREG32(mmVM_CONTEXT1_CNTL,
544 VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK |
545 (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
546 ((adev->vm_manager.block_size - 9)
547 << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT));
548 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
549 gmc_v6_0_set_fault_enable_default(adev, false);
551 gmc_v6_0_set_fault_enable_default(adev, true);
553 gmc_v6_0_gart_flush_gpu_tlb(adev, 0);
554 dev_info(adev->dev, "PCIE GART of %uM enabled (table at 0x%016llX).\n",
555 (unsigned)(adev->mc.gtt_size >> 20),
556 (unsigned long long)adev->gart.table_addr);
557 adev->gart.ready = true;
561 static int gmc_v6_0_gart_init(struct amdgpu_device *adev)
565 if (adev->gart.robj) {
566 dev_warn(adev->dev, "gmc_v6_0 PCIE GART already initialized\n");
569 r = amdgpu_gart_init(adev);
572 adev->gart.table_size = adev->gart.num_gpu_pages * 8;
573 adev->gart.gart_pte_flags = 0;
574 return amdgpu_gart_table_vram_alloc(adev);
577 static void gmc_v6_0_gart_disable(struct amdgpu_device *adev)
581 for (i = 1; i < 16; ++i) {
584 reg = VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i ;
586 reg = VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (i - 8);
587 adev->vm_manager.saved_table_addr[i] = RREG32(reg);
590 /* Disable all tables */
591 WREG32(mmVM_CONTEXT0_CNTL, 0);
592 WREG32(mmVM_CONTEXT1_CNTL, 0);
593 /* Setup TLB control */
594 WREG32(mmMC_VM_MX_L1_TLB_CNTL,
595 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
596 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
599 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
600 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
601 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
602 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
603 WREG32(mmVM_L2_CNTL2, 0);
604 WREG32(mmVM_L2_CNTL3,
605 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
606 (0UL << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
607 amdgpu_gart_table_vram_unpin(adev);
610 static void gmc_v6_0_gart_fini(struct amdgpu_device *adev)
612 amdgpu_gart_table_vram_free(adev);
613 amdgpu_gart_fini(adev);
616 static int gmc_v6_0_vm_init(struct amdgpu_device *adev)
620 * VMID 0 is reserved for System
621 * amdgpu graphics/compute will use VMIDs 1-7
622 * amdkfd will use VMIDs 8-15
624 adev->vm_manager.num_ids = AMDGPU_NUM_OF_VMIDS;
625 adev->vm_manager.num_level = 1;
626 amdgpu_vm_manager_init(adev);
628 /* base offset of vram pages */
629 if (adev->flags & AMD_IS_APU) {
630 u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
632 adev->vm_manager.vram_base_offset = tmp;
634 adev->vm_manager.vram_base_offset = 0;
639 static void gmc_v6_0_vm_fini(struct amdgpu_device *adev)
643 static void gmc_v6_0_vm_decode_fault(struct amdgpu_device *adev,
644 u32 status, u32 addr, u32 mc_client)
647 u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
648 u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
650 char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
651 (mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
653 mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
656 dev_err(adev->dev, "VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
657 protections, vmid, addr,
658 REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
660 "write" : "read", block, mc_client, mc_id);
664 static const u32 mc_cg_registers[] = {
676 static const u32 mc_cg_ls_en[] = {
677 MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK,
678 MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK,
679 MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK,
680 MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK,
681 ATC_MISC_CG__MEM_LS_ENABLE_MASK,
682 MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK,
683 MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK,
684 MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK,
685 VM_L2_CG__MEM_LS_ENABLE_MASK,
688 static const u32 mc_cg_en[] = {
689 MC_HUB_MISC_HUB_CG__ENABLE_MASK,
690 MC_HUB_MISC_SIP_CG__ENABLE_MASK,
691 MC_HUB_MISC_VM_CG__ENABLE_MASK,
692 MC_XPB_CLK_GAT__ENABLE_MASK,
693 ATC_MISC_CG__ENABLE_MASK,
694 MC_CITF_MISC_WR_CG__ENABLE_MASK,
695 MC_CITF_MISC_RD_CG__ENABLE_MASK,
696 MC_CITF_MISC_VM_CG__ENABLE_MASK,
697 VM_L2_CG__ENABLE_MASK,
700 static void gmc_v6_0_enable_mc_ls(struct amdgpu_device *adev,
706 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
707 orig = data = RREG32(mc_cg_registers[i]);
708 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_LS))
709 data |= mc_cg_ls_en[i];
711 data &= ~mc_cg_ls_en[i];
713 WREG32(mc_cg_registers[i], data);
717 static void gmc_v6_0_enable_mc_mgcg(struct amdgpu_device *adev,
723 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
724 orig = data = RREG32(mc_cg_registers[i]);
725 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_MGCG))
728 data &= ~mc_cg_en[i];
730 WREG32(mc_cg_registers[i], data);
734 static void gmc_v6_0_enable_bif_mgls(struct amdgpu_device *adev,
739 orig = data = RREG32_PCIE(ixPCIE_CNTL2);
741 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_BIF_LS)) {
742 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 1);
743 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 1);
744 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 1);
745 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 1);
747 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 0);
748 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 0);
749 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 0);
750 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 0);
754 WREG32_PCIE(ixPCIE_CNTL2, data);
757 static void gmc_v6_0_enable_hdp_mgcg(struct amdgpu_device *adev,
762 orig = data = RREG32(mmHDP_HOST_PATH_CNTL);
764 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_MGCG))
765 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 0);
767 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 1);
770 WREG32(mmHDP_HOST_PATH_CNTL, data);
773 static void gmc_v6_0_enable_hdp_ls(struct amdgpu_device *adev,
778 orig = data = RREG32(mmHDP_MEM_POWER_LS);
780 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_LS))
781 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 1);
783 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 0);
786 WREG32(mmHDP_MEM_POWER_LS, data);
790 static int gmc_v6_0_convert_vram_type(int mc_seq_vram_type)
792 switch (mc_seq_vram_type) {
793 case MC_SEQ_MISC0__MT__GDDR1:
794 return AMDGPU_VRAM_TYPE_GDDR1;
795 case MC_SEQ_MISC0__MT__DDR2:
796 return AMDGPU_VRAM_TYPE_DDR2;
797 case MC_SEQ_MISC0__MT__GDDR3:
798 return AMDGPU_VRAM_TYPE_GDDR3;
799 case MC_SEQ_MISC0__MT__GDDR4:
800 return AMDGPU_VRAM_TYPE_GDDR4;
801 case MC_SEQ_MISC0__MT__GDDR5:
802 return AMDGPU_VRAM_TYPE_GDDR5;
803 case MC_SEQ_MISC0__MT__DDR3:
804 return AMDGPU_VRAM_TYPE_DDR3;
806 return AMDGPU_VRAM_TYPE_UNKNOWN;
810 static int gmc_v6_0_early_init(void *handle)
812 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
814 gmc_v6_0_set_gart_funcs(adev);
815 gmc_v6_0_set_irq_funcs(adev);
817 if (adev->flags & AMD_IS_APU) {
818 adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
820 u32 tmp = RREG32(mmMC_SEQ_MISC0);
821 tmp &= MC_SEQ_MISC0__MT__MASK;
822 adev->mc.vram_type = gmc_v6_0_convert_vram_type(tmp);
828 static int gmc_v6_0_late_init(void *handle)
830 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
832 if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
833 return amdgpu_irq_get(adev, &adev->mc.vm_fault, 0);
838 static int gmc_v6_0_sw_init(void *handle)
842 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
844 r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_LEGACY, 146, &adev->mc.vm_fault);
848 r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_LEGACY, 147, &adev->mc.vm_fault);
852 amdgpu_vm_adjust_size(adev, 64);
853 adev->vm_manager.max_pfn = adev->vm_manager.vm_size << 18;
855 adev->mc.mc_mask = 0xffffffffffULL;
857 adev->need_dma32 = false;
858 dma_bits = adev->need_dma32 ? 32 : 40;
859 r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
861 adev->need_dma32 = true;
863 dev_warn(adev->dev, "amdgpu: No suitable DMA available.\n");
865 r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
867 pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
868 dev_warn(adev->dev, "amdgpu: No coherent DMA available.\n");
871 r = gmc_v6_0_init_microcode(adev);
873 dev_err(adev->dev, "Failed to load mc firmware!\n");
877 r = gmc_v6_0_mc_init(adev);
881 r = amdgpu_bo_init(adev);
885 r = gmc_v6_0_gart_init(adev);
889 if (!adev->vm_manager.enabled) {
890 r = gmc_v6_0_vm_init(adev);
892 dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
895 adev->vm_manager.enabled = true;
901 static int gmc_v6_0_sw_fini(void *handle)
903 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
905 if (adev->vm_manager.enabled) {
906 gmc_v6_0_vm_fini(adev);
907 adev->vm_manager.enabled = false;
909 gmc_v6_0_gart_fini(adev);
910 amdgpu_gem_force_release(adev);
911 amdgpu_bo_fini(adev);
916 static int gmc_v6_0_hw_init(void *handle)
919 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
921 gmc_v6_0_mc_program(adev);
923 if (!(adev->flags & AMD_IS_APU)) {
924 r = gmc_v6_0_mc_load_microcode(adev);
926 dev_err(adev->dev, "Failed to load MC firmware!\n");
931 r = gmc_v6_0_gart_enable(adev);
938 static int gmc_v6_0_hw_fini(void *handle)
940 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
942 amdgpu_irq_put(adev, &adev->mc.vm_fault, 0);
943 gmc_v6_0_gart_disable(adev);
948 static int gmc_v6_0_suspend(void *handle)
950 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
952 if (adev->vm_manager.enabled) {
953 gmc_v6_0_vm_fini(adev);
954 adev->vm_manager.enabled = false;
956 gmc_v6_0_hw_fini(adev);
961 static int gmc_v6_0_resume(void *handle)
964 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
966 r = gmc_v6_0_hw_init(adev);
970 if (!adev->vm_manager.enabled) {
971 r = gmc_v6_0_vm_init(adev);
973 dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
976 adev->vm_manager.enabled = true;
982 static bool gmc_v6_0_is_idle(void *handle)
984 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
985 u32 tmp = RREG32(mmSRBM_STATUS);
987 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
988 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
994 static int gmc_v6_0_wait_for_idle(void *handle)
998 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1000 for (i = 0; i < adev->usec_timeout; i++) {
1001 tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
1002 SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1003 SRBM_STATUS__MCC_BUSY_MASK |
1004 SRBM_STATUS__MCD_BUSY_MASK |
1005 SRBM_STATUS__VMC_BUSY_MASK);
1014 static int gmc_v6_0_soft_reset(void *handle)
1016 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1017 struct amdgpu_mode_mc_save save;
1018 u32 srbm_soft_reset = 0;
1019 u32 tmp = RREG32(mmSRBM_STATUS);
1021 if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1022 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1023 SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1025 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1026 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1027 if (!(adev->flags & AMD_IS_APU))
1028 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1029 SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1032 if (srbm_soft_reset) {
1033 gmc_v6_0_mc_stop(adev, &save);
1034 if (gmc_v6_0_wait_for_idle(adev)) {
1035 dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1039 tmp = RREG32(mmSRBM_SOFT_RESET);
1040 tmp |= srbm_soft_reset;
1041 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1042 WREG32(mmSRBM_SOFT_RESET, tmp);
1043 tmp = RREG32(mmSRBM_SOFT_RESET);
1047 tmp &= ~srbm_soft_reset;
1048 WREG32(mmSRBM_SOFT_RESET, tmp);
1049 tmp = RREG32(mmSRBM_SOFT_RESET);
1053 gmc_v6_0_mc_resume(adev, &save);
1060 static int gmc_v6_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1061 struct amdgpu_irq_src *src,
1063 enum amdgpu_interrupt_state state)
1066 u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1067 VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1068 VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1069 VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1070 VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1071 VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1074 case AMDGPU_IRQ_STATE_DISABLE:
1075 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1077 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1078 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1080 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1082 case AMDGPU_IRQ_STATE_ENABLE:
1083 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1085 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1086 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1088 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1097 static int gmc_v6_0_process_interrupt(struct amdgpu_device *adev,
1098 struct amdgpu_irq_src *source,
1099 struct amdgpu_iv_entry *entry)
1103 addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1104 status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1105 WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1107 if (!addr && !status)
1110 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1111 gmc_v6_0_set_fault_enable_default(adev, false);
1113 if (printk_ratelimit()) {
1114 dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1115 entry->src_id, entry->src_data[0]);
1116 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
1118 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1120 gmc_v6_0_vm_decode_fault(adev, status, addr, 0);
1126 static int gmc_v6_0_set_clockgating_state(void *handle,
1127 enum amd_clockgating_state state)
1132 static int gmc_v6_0_set_powergating_state(void *handle,
1133 enum amd_powergating_state state)
1138 static const struct amd_ip_funcs gmc_v6_0_ip_funcs = {
1140 .early_init = gmc_v6_0_early_init,
1141 .late_init = gmc_v6_0_late_init,
1142 .sw_init = gmc_v6_0_sw_init,
1143 .sw_fini = gmc_v6_0_sw_fini,
1144 .hw_init = gmc_v6_0_hw_init,
1145 .hw_fini = gmc_v6_0_hw_fini,
1146 .suspend = gmc_v6_0_suspend,
1147 .resume = gmc_v6_0_resume,
1148 .is_idle = gmc_v6_0_is_idle,
1149 .wait_for_idle = gmc_v6_0_wait_for_idle,
1150 .soft_reset = gmc_v6_0_soft_reset,
1151 .set_clockgating_state = gmc_v6_0_set_clockgating_state,
1152 .set_powergating_state = gmc_v6_0_set_powergating_state,
1155 static const struct amdgpu_gart_funcs gmc_v6_0_gart_funcs = {
1156 .flush_gpu_tlb = gmc_v6_0_gart_flush_gpu_tlb,
1157 .set_pte_pde = gmc_v6_0_gart_set_pte_pde,
1158 .set_prt = gmc_v6_0_set_prt,
1159 .get_vm_pte_flags = gmc_v6_0_get_vm_pte_flags
1162 static const struct amdgpu_irq_src_funcs gmc_v6_0_irq_funcs = {
1163 .set = gmc_v6_0_vm_fault_interrupt_state,
1164 .process = gmc_v6_0_process_interrupt,
1167 static void gmc_v6_0_set_gart_funcs(struct amdgpu_device *adev)
1169 if (adev->gart.gart_funcs == NULL)
1170 adev->gart.gart_funcs = &gmc_v6_0_gart_funcs;
1173 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev)
1175 adev->mc.vm_fault.num_types = 1;
1176 adev->mc.vm_fault.funcs = &gmc_v6_0_irq_funcs;
1179 const struct amdgpu_ip_block_version gmc_v6_0_ip_block =
1181 .type = AMD_IP_BLOCK_TYPE_GMC,
1185 .funcs = &gmc_v6_0_ip_funcs,