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>
25 #include <drm/drm_cache.h>
28 #include "amdgpu_ucode.h"
30 #include "bif/bif_3_0_d.h"
31 #include "bif/bif_3_0_sh_mask.h"
32 #include "oss/oss_1_0_d.h"
33 #include "oss/oss_1_0_sh_mask.h"
34 #include "gmc/gmc_6_0_d.h"
35 #include "gmc/gmc_6_0_sh_mask.h"
36 #include "dce/dce_6_0_d.h"
37 #include "dce/dce_6_0_sh_mask.h"
40 static void gmc_v6_0_set_gmc_funcs(struct amdgpu_device *adev);
41 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev);
42 static int gmc_v6_0_wait_for_idle(void *handle);
44 MODULE_FIRMWARE("amdgpu/tahiti_mc.bin");
45 MODULE_FIRMWARE("amdgpu/pitcairn_mc.bin");
46 MODULE_FIRMWARE("amdgpu/verde_mc.bin");
47 MODULE_FIRMWARE("amdgpu/oland_mc.bin");
48 MODULE_FIRMWARE("amdgpu/si58_mc.bin");
50 #define MC_SEQ_MISC0__MT__MASK 0xf0000000
51 #define MC_SEQ_MISC0__MT__GDDR1 0x10000000
52 #define MC_SEQ_MISC0__MT__DDR2 0x20000000
53 #define MC_SEQ_MISC0__MT__GDDR3 0x30000000
54 #define MC_SEQ_MISC0__MT__GDDR4 0x40000000
55 #define MC_SEQ_MISC0__MT__GDDR5 0x50000000
56 #define MC_SEQ_MISC0__MT__HBM 0x60000000
57 #define MC_SEQ_MISC0__MT__DDR3 0xB0000000
60 static const u32 crtc_offsets[6] =
62 SI_CRTC0_REGISTER_OFFSET,
63 SI_CRTC1_REGISTER_OFFSET,
64 SI_CRTC2_REGISTER_OFFSET,
65 SI_CRTC3_REGISTER_OFFSET,
66 SI_CRTC4_REGISTER_OFFSET,
67 SI_CRTC5_REGISTER_OFFSET
70 static void gmc_v6_0_mc_stop(struct amdgpu_device *adev)
74 gmc_v6_0_wait_for_idle((void *)adev);
76 blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
77 if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
78 /* Block CPU access */
79 WREG32(mmBIF_FB_EN, 0);
81 blackout = REG_SET_FIELD(blackout,
82 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
83 WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout | 1);
85 /* wait for the MC to settle */
90 static void gmc_v6_0_mc_resume(struct amdgpu_device *adev)
94 /* unblackout the MC */
95 tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
96 tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
97 WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
98 /* allow CPU access */
99 tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
100 tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
101 WREG32(mmBIF_FB_EN, tmp);
104 static int gmc_v6_0_init_microcode(struct amdgpu_device *adev)
106 const char *chip_name;
109 bool is_58_fw = false;
113 switch (adev->asic_type) {
115 chip_name = "tahiti";
118 chip_name = "pitcairn";
127 chip_name = "hainan";
132 /* this memory configuration requires special firmware */
133 if (((RREG32(mmMC_SEQ_MISC0) & 0xff000000) >> 24) == 0x58)
137 snprintf(fw_name, sizeof(fw_name), "amdgpu/si58_mc.bin");
139 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mc.bin", chip_name);
140 err = request_firmware(&adev->gmc.fw, fw_name, adev->dev);
144 err = amdgpu_ucode_validate(adev->gmc.fw);
149 "si_mc: Failed to load firmware \"%s\"\n",
151 release_firmware(adev->gmc.fw);
157 static int gmc_v6_0_mc_load_microcode(struct amdgpu_device *adev)
159 const __le32 *new_fw_data = NULL;
161 const __le32 *new_io_mc_regs = NULL;
162 int i, regs_size, ucode_size;
163 const struct mc_firmware_header_v1_0 *hdr;
168 hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
170 amdgpu_ucode_print_mc_hdr(&hdr->header);
172 adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
173 regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
174 new_io_mc_regs = (const __le32 *)
175 (adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
176 ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
177 new_fw_data = (const __le32 *)
178 (adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
180 running = RREG32(mmMC_SEQ_SUP_CNTL) & MC_SEQ_SUP_CNTL__RUN_MASK;
184 /* reset the engine and set to writable */
185 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
186 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
188 /* load mc io regs */
189 for (i = 0; i < regs_size; i++) {
190 WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(new_io_mc_regs++));
191 WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(new_io_mc_regs++));
193 /* load the MC ucode */
194 for (i = 0; i < ucode_size; i++) {
195 WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(new_fw_data++));
198 /* put the engine back into the active state */
199 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
200 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
201 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
203 /* wait for training to complete */
204 for (i = 0; i < adev->usec_timeout; i++) {
205 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D0_MASK)
209 for (i = 0; i < adev->usec_timeout; i++) {
210 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D1_MASK)
220 static void gmc_v6_0_vram_gtt_location(struct amdgpu_device *adev,
221 struct amdgpu_gmc *mc)
223 u64 base = RREG32(mmMC_VM_FB_LOCATION) & 0xFFFF;
226 amdgpu_device_vram_location(adev, &adev->gmc, base);
227 amdgpu_device_gart_location(adev, mc);
230 static void gmc_v6_0_mc_program(struct amdgpu_device *adev)
235 for (i = 0, j = 0; i < 32; i++, j += 0x6) {
236 WREG32((0xb05 + j), 0x00000000);
237 WREG32((0xb06 + j), 0x00000000);
238 WREG32((0xb07 + j), 0x00000000);
239 WREG32((0xb08 + j), 0x00000000);
240 WREG32((0xb09 + j), 0x00000000);
242 WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
244 if (gmc_v6_0_wait_for_idle((void *)adev)) {
245 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
248 if (adev->mode_info.num_crtc) {
251 /* Lockout access through VGA aperture*/
252 tmp = RREG32(mmVGA_HDP_CONTROL);
253 tmp |= VGA_HDP_CONTROL__VGA_MEMORY_DISABLE_MASK;
254 WREG32(mmVGA_HDP_CONTROL, tmp);
256 /* disable VGA render */
257 tmp = RREG32(mmVGA_RENDER_CONTROL);
258 tmp &= ~VGA_VSTATUS_CNTL;
259 WREG32(mmVGA_RENDER_CONTROL, tmp);
261 /* Update configuration */
262 WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
263 adev->gmc.vram_start >> 12);
264 WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
265 adev->gmc.vram_end >> 12);
266 WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
267 adev->vram_scratch.gpu_addr >> 12);
268 WREG32(mmMC_VM_AGP_BASE, 0);
269 WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
270 WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
272 if (gmc_v6_0_wait_for_idle((void *)adev)) {
273 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
277 static int gmc_v6_0_mc_init(struct amdgpu_device *adev)
281 int chansize, numchan;
284 tmp = RREG32(mmMC_ARB_RAMCFG);
285 if (tmp & (1 << 11)) {
287 } else if (tmp & MC_ARB_RAMCFG__CHANSIZE_MASK) {
292 tmp = RREG32(mmMC_SHARED_CHMAP);
293 switch ((tmp & MC_SHARED_CHMAP__NOOFCHAN_MASK) >> MC_SHARED_CHMAP__NOOFCHAN__SHIFT) {
323 adev->gmc.vram_width = numchan * chansize;
324 /* size in MB on si */
325 adev->gmc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
326 adev->gmc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
328 if (!(adev->flags & AMD_IS_APU)) {
329 r = amdgpu_device_resize_fb_bar(adev);
333 adev->gmc.aper_base = pci_resource_start(adev->pdev, 0);
334 adev->gmc.aper_size = pci_resource_len(adev->pdev, 0);
335 adev->gmc.visible_vram_size = adev->gmc.aper_size;
337 /* set the gart size */
338 if (amdgpu_gart_size == -1) {
339 switch (adev->asic_type) {
340 case CHIP_HAINAN: /* no MM engines */
342 adev->gmc.gart_size = 256ULL << 20;
344 case CHIP_VERDE: /* UVD, VCE do not support GPUVM */
345 case CHIP_TAHITI: /* UVD, VCE do not support GPUVM */
346 case CHIP_PITCAIRN: /* UVD, VCE do not support GPUVM */
347 case CHIP_OLAND: /* UVD, VCE do not support GPUVM */
348 adev->gmc.gart_size = 1024ULL << 20;
352 adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
355 gmc_v6_0_vram_gtt_location(adev, &adev->gmc);
360 static void gmc_v6_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid)
362 WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
365 static uint64_t gmc_v6_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
366 unsigned vmid, uint64_t pd_addr)
370 /* write new base address */
372 reg = mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid;
374 reg = mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (vmid - 8);
375 amdgpu_ring_emit_wreg(ring, reg, pd_addr >> 12);
377 /* bits 0-15 are the VM contexts0-15 */
378 amdgpu_ring_emit_wreg(ring, mmVM_INVALIDATE_REQUEST, 1 << vmid);
383 static int gmc_v6_0_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr,
384 uint32_t gpu_page_idx, uint64_t addr,
387 void __iomem *ptr = (void *)cpu_pt_addr;
390 value = addr & 0xFFFFFFFFFFFFF000ULL;
392 writeq(value, ptr + (gpu_page_idx * 8));
397 static uint64_t gmc_v6_0_get_vm_pte_flags(struct amdgpu_device *adev,
400 uint64_t pte_flag = 0;
402 if (flags & AMDGPU_VM_PAGE_READABLE)
403 pte_flag |= AMDGPU_PTE_READABLE;
404 if (flags & AMDGPU_VM_PAGE_WRITEABLE)
405 pte_flag |= AMDGPU_PTE_WRITEABLE;
406 if (flags & AMDGPU_VM_PAGE_PRT)
407 pte_flag |= AMDGPU_PTE_PRT;
412 static void gmc_v6_0_get_vm_pde(struct amdgpu_device *adev, int level,
413 uint64_t *addr, uint64_t *flags)
415 BUG_ON(*addr & 0xFFFFFF0000000FFFULL);
418 static void gmc_v6_0_set_fault_enable_default(struct amdgpu_device *adev,
423 tmp = RREG32(mmVM_CONTEXT1_CNTL);
424 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
425 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
426 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
427 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
428 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
429 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
430 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
431 VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
432 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
433 READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
434 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
435 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
436 WREG32(mmVM_CONTEXT1_CNTL, tmp);
440 + * gmc_v8_0_set_prt - set PRT VM fault
442 + * @adev: amdgpu_device pointer
443 + * @enable: enable/disable VM fault handling for PRT
445 static void gmc_v6_0_set_prt(struct amdgpu_device *adev, bool enable)
449 if (enable && !adev->gmc.prt_warning) {
450 dev_warn(adev->dev, "Disabling VM faults because of PRT request!\n");
451 adev->gmc.prt_warning = true;
454 tmp = RREG32(mmVM_PRT_CNTL);
455 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
456 CB_DISABLE_FAULT_ON_UNMAPPED_ACCESS,
458 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
459 TC_DISABLE_FAULT_ON_UNMAPPED_ACCESS,
461 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
462 L2_CACHE_STORE_INVALID_ENTRIES,
464 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
465 L1_TLB_STORE_INVALID_ENTRIES,
467 WREG32(mmVM_PRT_CNTL, tmp);
470 uint32_t low = AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT;
471 uint32_t high = adev->vm_manager.max_pfn -
472 (AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT);
474 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, low);
475 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, low);
476 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, low);
477 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, low);
478 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, high);
479 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, high);
480 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, high);
481 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, high);
483 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, 0xfffffff);
484 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, 0xfffffff);
485 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, 0xfffffff);
486 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, 0xfffffff);
487 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, 0x0);
488 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, 0x0);
489 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, 0x0);
490 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, 0x0);
494 static int gmc_v6_0_gart_enable(struct amdgpu_device *adev)
499 if (adev->gart.robj == NULL) {
500 dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
503 r = amdgpu_gart_table_vram_pin(adev);
506 /* Setup TLB control */
507 WREG32(mmMC_VM_MX_L1_TLB_CNTL,
509 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_TLB_MASK |
510 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_FRAGMENT_PROCESSING_MASK |
511 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
512 MC_VM_MX_L1_TLB_CNTL__ENABLE_ADVANCED_DRIVER_MODEL_MASK |
513 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
516 VM_L2_CNTL__ENABLE_L2_CACHE_MASK |
517 VM_L2_CNTL__ENABLE_L2_FRAGMENT_PROCESSING_MASK |
518 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
519 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
520 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
521 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
522 WREG32(mmVM_L2_CNTL2,
523 VM_L2_CNTL2__INVALIDATE_ALL_L1_TLBS_MASK |
524 VM_L2_CNTL2__INVALIDATE_L2_CACHE_MASK);
526 field = adev->vm_manager.fragment_size;
527 WREG32(mmVM_L2_CNTL3,
528 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
529 (field << VM_L2_CNTL3__BANK_SELECT__SHIFT) |
530 (field << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
532 WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->gmc.gart_start >> 12);
533 WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->gmc.gart_end >> 12);
534 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12);
535 WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
536 (u32)(adev->dummy_page_addr >> 12));
537 WREG32(mmVM_CONTEXT0_CNTL2, 0);
538 WREG32(mmVM_CONTEXT0_CNTL,
539 VM_CONTEXT0_CNTL__ENABLE_CONTEXT_MASK |
540 (0UL << VM_CONTEXT0_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
541 VM_CONTEXT0_CNTL__RANGE_PROTECTION_FAULT_ENABLE_DEFAULT_MASK);
547 /* empty context1-15 */
548 /* set vm size, must be a multiple of 4 */
549 WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
550 WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
551 /* Assign the pt base to something valid for now; the pts used for
552 * the VMs are determined by the application and setup and assigned
553 * on the fly in the vm part of radeon_gart.c
555 for (i = 1; i < 16; i++) {
557 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
558 adev->gart.table_addr >> 12);
560 WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
561 adev->gart.table_addr >> 12);
564 /* enable context1-15 */
565 WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
566 (u32)(adev->dummy_page_addr >> 12));
567 WREG32(mmVM_CONTEXT1_CNTL2, 4);
568 WREG32(mmVM_CONTEXT1_CNTL,
569 VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK |
570 (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
571 ((adev->vm_manager.block_size - 9)
572 << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT));
573 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
574 gmc_v6_0_set_fault_enable_default(adev, false);
576 gmc_v6_0_set_fault_enable_default(adev, true);
578 gmc_v6_0_flush_gpu_tlb(adev, 0);
579 dev_info(adev->dev, "PCIE GART of %uM enabled (table at 0x%016llX).\n",
580 (unsigned)(adev->gmc.gart_size >> 20),
581 (unsigned long long)adev->gart.table_addr);
582 adev->gart.ready = true;
586 static int gmc_v6_0_gart_init(struct amdgpu_device *adev)
590 if (adev->gart.robj) {
591 dev_warn(adev->dev, "gmc_v6_0 PCIE GART already initialized\n");
594 r = amdgpu_gart_init(adev);
597 adev->gart.table_size = adev->gart.num_gpu_pages * 8;
598 adev->gart.gart_pte_flags = 0;
599 return amdgpu_gart_table_vram_alloc(adev);
602 static void gmc_v6_0_gart_disable(struct amdgpu_device *adev)
606 for (i = 1; i < 16; ++i) {
609 reg = VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i ;
611 reg = VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (i - 8);
612 adev->vm_manager.saved_table_addr[i] = RREG32(reg);
615 /* Disable all tables */
616 WREG32(mmVM_CONTEXT0_CNTL, 0);
617 WREG32(mmVM_CONTEXT1_CNTL, 0);
618 /* Setup TLB control */
619 WREG32(mmMC_VM_MX_L1_TLB_CNTL,
620 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
621 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
624 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
625 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
626 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
627 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
628 WREG32(mmVM_L2_CNTL2, 0);
629 WREG32(mmVM_L2_CNTL3,
630 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
631 (0UL << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
632 amdgpu_gart_table_vram_unpin(adev);
635 static void gmc_v6_0_vm_decode_fault(struct amdgpu_device *adev,
636 u32 status, u32 addr, u32 mc_client)
639 u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
640 u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
642 char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
643 (mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
645 mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
648 dev_err(adev->dev, "VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
649 protections, vmid, addr,
650 REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
652 "write" : "read", block, mc_client, mc_id);
656 static const u32 mc_cg_registers[] = {
668 static const u32 mc_cg_ls_en[] = {
669 MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK,
670 MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK,
671 MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK,
672 MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK,
673 ATC_MISC_CG__MEM_LS_ENABLE_MASK,
674 MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK,
675 MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK,
676 MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK,
677 VM_L2_CG__MEM_LS_ENABLE_MASK,
680 static const u32 mc_cg_en[] = {
681 MC_HUB_MISC_HUB_CG__ENABLE_MASK,
682 MC_HUB_MISC_SIP_CG__ENABLE_MASK,
683 MC_HUB_MISC_VM_CG__ENABLE_MASK,
684 MC_XPB_CLK_GAT__ENABLE_MASK,
685 ATC_MISC_CG__ENABLE_MASK,
686 MC_CITF_MISC_WR_CG__ENABLE_MASK,
687 MC_CITF_MISC_RD_CG__ENABLE_MASK,
688 MC_CITF_MISC_VM_CG__ENABLE_MASK,
689 VM_L2_CG__ENABLE_MASK,
692 static void gmc_v6_0_enable_mc_ls(struct amdgpu_device *adev,
698 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
699 orig = data = RREG32(mc_cg_registers[i]);
700 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_LS))
701 data |= mc_cg_ls_en[i];
703 data &= ~mc_cg_ls_en[i];
705 WREG32(mc_cg_registers[i], data);
709 static void gmc_v6_0_enable_mc_mgcg(struct amdgpu_device *adev,
715 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
716 orig = data = RREG32(mc_cg_registers[i]);
717 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_MGCG))
720 data &= ~mc_cg_en[i];
722 WREG32(mc_cg_registers[i], data);
726 static void gmc_v6_0_enable_bif_mgls(struct amdgpu_device *adev,
731 orig = data = RREG32_PCIE(ixPCIE_CNTL2);
733 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_BIF_LS)) {
734 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 1);
735 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 1);
736 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 1);
737 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 1);
739 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 0);
740 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 0);
741 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 0);
742 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 0);
746 WREG32_PCIE(ixPCIE_CNTL2, data);
749 static void gmc_v6_0_enable_hdp_mgcg(struct amdgpu_device *adev,
754 orig = data = RREG32(mmHDP_HOST_PATH_CNTL);
756 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_MGCG))
757 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 0);
759 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 1);
762 WREG32(mmHDP_HOST_PATH_CNTL, data);
765 static void gmc_v6_0_enable_hdp_ls(struct amdgpu_device *adev,
770 orig = data = RREG32(mmHDP_MEM_POWER_LS);
772 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_LS))
773 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 1);
775 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 0);
778 WREG32(mmHDP_MEM_POWER_LS, data);
782 static int gmc_v6_0_convert_vram_type(int mc_seq_vram_type)
784 switch (mc_seq_vram_type) {
785 case MC_SEQ_MISC0__MT__GDDR1:
786 return AMDGPU_VRAM_TYPE_GDDR1;
787 case MC_SEQ_MISC0__MT__DDR2:
788 return AMDGPU_VRAM_TYPE_DDR2;
789 case MC_SEQ_MISC0__MT__GDDR3:
790 return AMDGPU_VRAM_TYPE_GDDR3;
791 case MC_SEQ_MISC0__MT__GDDR4:
792 return AMDGPU_VRAM_TYPE_GDDR4;
793 case MC_SEQ_MISC0__MT__GDDR5:
794 return AMDGPU_VRAM_TYPE_GDDR5;
795 case MC_SEQ_MISC0__MT__DDR3:
796 return AMDGPU_VRAM_TYPE_DDR3;
798 return AMDGPU_VRAM_TYPE_UNKNOWN;
802 static int gmc_v6_0_early_init(void *handle)
804 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
806 gmc_v6_0_set_gmc_funcs(adev);
807 gmc_v6_0_set_irq_funcs(adev);
812 static int gmc_v6_0_late_init(void *handle)
814 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
816 amdgpu_bo_late_init(adev);
818 if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
819 return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
824 static unsigned gmc_v6_0_get_vbios_fb_size(struct amdgpu_device *adev)
826 u32 d1vga_control = RREG32(mmD1VGA_CONTROL);
829 if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
830 size = 9 * 1024 * 1024; /* reserve 8MB for vga emulator and 1 MB for FB */
832 u32 viewport = RREG32(mmVIEWPORT_SIZE);
833 size = (REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
834 REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_WIDTH) *
837 /* return 0 if the pre-OS buffer uses up most of vram */
838 if ((adev->gmc.real_vram_size - size) < (8 * 1024 * 1024))
843 static int gmc_v6_0_sw_init(void *handle)
847 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
849 if (adev->flags & AMD_IS_APU) {
850 adev->gmc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
852 u32 tmp = RREG32(mmMC_SEQ_MISC0);
853 tmp &= MC_SEQ_MISC0__MT__MASK;
854 adev->gmc.vram_type = gmc_v6_0_convert_vram_type(tmp);
857 r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_LEGACY, 146, &adev->gmc.vm_fault);
861 r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_LEGACY, 147, &adev->gmc.vm_fault);
865 amdgpu_vm_adjust_size(adev, 64, 9, 1, 40);
867 adev->gmc.mc_mask = 0xffffffffffULL;
869 adev->need_dma32 = false;
870 dma_bits = adev->need_dma32 ? 32 : 40;
871 r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
873 adev->need_dma32 = true;
875 dev_warn(adev->dev, "amdgpu: No suitable DMA available.\n");
877 r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
879 pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
880 dev_warn(adev->dev, "amdgpu: No coherent DMA available.\n");
882 adev->need_swiotlb = drm_get_max_iomem() > ((u64)1 << dma_bits);
884 r = gmc_v6_0_init_microcode(adev);
886 dev_err(adev->dev, "Failed to load mc firmware!\n");
890 r = gmc_v6_0_mc_init(adev);
894 adev->gmc.stolen_size = gmc_v6_0_get_vbios_fb_size(adev);
896 r = amdgpu_bo_init(adev);
900 r = gmc_v6_0_gart_init(adev);
906 * VMID 0 is reserved for System
907 * amdgpu graphics/compute will use VMIDs 1-7
908 * amdkfd will use VMIDs 8-15
910 adev->vm_manager.id_mgr[0].num_ids = AMDGPU_NUM_OF_VMIDS;
911 amdgpu_vm_manager_init(adev);
913 /* base offset of vram pages */
914 if (adev->flags & AMD_IS_APU) {
915 u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
918 adev->vm_manager.vram_base_offset = tmp;
920 adev->vm_manager.vram_base_offset = 0;
926 static int gmc_v6_0_sw_fini(void *handle)
928 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
930 amdgpu_gem_force_release(adev);
931 amdgpu_vm_manager_fini(adev);
932 amdgpu_gart_table_vram_free(adev);
933 amdgpu_bo_fini(adev);
934 amdgpu_gart_fini(adev);
935 release_firmware(adev->gmc.fw);
941 static int gmc_v6_0_hw_init(void *handle)
944 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
946 gmc_v6_0_mc_program(adev);
948 if (!(adev->flags & AMD_IS_APU)) {
949 r = gmc_v6_0_mc_load_microcode(adev);
951 dev_err(adev->dev, "Failed to load MC firmware!\n");
956 r = gmc_v6_0_gart_enable(adev);
963 static int gmc_v6_0_hw_fini(void *handle)
965 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
967 amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
968 gmc_v6_0_gart_disable(adev);
973 static int gmc_v6_0_suspend(void *handle)
975 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
977 gmc_v6_0_hw_fini(adev);
982 static int gmc_v6_0_resume(void *handle)
985 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
987 r = gmc_v6_0_hw_init(adev);
991 amdgpu_vmid_reset_all(adev);
996 static bool gmc_v6_0_is_idle(void *handle)
998 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
999 u32 tmp = RREG32(mmSRBM_STATUS);
1001 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1002 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
1008 static int gmc_v6_0_wait_for_idle(void *handle)
1011 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1013 for (i = 0; i < adev->usec_timeout; i++) {
1014 if (gmc_v6_0_is_idle(handle))
1022 static int gmc_v6_0_soft_reset(void *handle)
1024 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1025 u32 srbm_soft_reset = 0;
1026 u32 tmp = RREG32(mmSRBM_STATUS);
1028 if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1029 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1030 SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1032 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1033 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1034 if (!(adev->flags & AMD_IS_APU))
1035 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1036 SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1039 if (srbm_soft_reset) {
1040 gmc_v6_0_mc_stop(adev);
1041 if (gmc_v6_0_wait_for_idle(adev)) {
1042 dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1046 tmp = RREG32(mmSRBM_SOFT_RESET);
1047 tmp |= srbm_soft_reset;
1048 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1049 WREG32(mmSRBM_SOFT_RESET, tmp);
1050 tmp = RREG32(mmSRBM_SOFT_RESET);
1054 tmp &= ~srbm_soft_reset;
1055 WREG32(mmSRBM_SOFT_RESET, tmp);
1056 tmp = RREG32(mmSRBM_SOFT_RESET);
1060 gmc_v6_0_mc_resume(adev);
1067 static int gmc_v6_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1068 struct amdgpu_irq_src *src,
1070 enum amdgpu_interrupt_state state)
1073 u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1074 VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1075 VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1076 VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1077 VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1078 VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1081 case AMDGPU_IRQ_STATE_DISABLE:
1082 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1084 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1085 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1087 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1089 case AMDGPU_IRQ_STATE_ENABLE:
1090 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1092 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1093 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1095 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1104 static int gmc_v6_0_process_interrupt(struct amdgpu_device *adev,
1105 struct amdgpu_irq_src *source,
1106 struct amdgpu_iv_entry *entry)
1110 addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1111 status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1112 WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1114 if (!addr && !status)
1117 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1118 gmc_v6_0_set_fault_enable_default(adev, false);
1120 if (printk_ratelimit()) {
1121 dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1122 entry->src_id, entry->src_data[0]);
1123 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
1125 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1127 gmc_v6_0_vm_decode_fault(adev, status, addr, 0);
1133 static int gmc_v6_0_set_clockgating_state(void *handle,
1134 enum amd_clockgating_state state)
1139 static int gmc_v6_0_set_powergating_state(void *handle,
1140 enum amd_powergating_state state)
1145 static const struct amd_ip_funcs gmc_v6_0_ip_funcs = {
1147 .early_init = gmc_v6_0_early_init,
1148 .late_init = gmc_v6_0_late_init,
1149 .sw_init = gmc_v6_0_sw_init,
1150 .sw_fini = gmc_v6_0_sw_fini,
1151 .hw_init = gmc_v6_0_hw_init,
1152 .hw_fini = gmc_v6_0_hw_fini,
1153 .suspend = gmc_v6_0_suspend,
1154 .resume = gmc_v6_0_resume,
1155 .is_idle = gmc_v6_0_is_idle,
1156 .wait_for_idle = gmc_v6_0_wait_for_idle,
1157 .soft_reset = gmc_v6_0_soft_reset,
1158 .set_clockgating_state = gmc_v6_0_set_clockgating_state,
1159 .set_powergating_state = gmc_v6_0_set_powergating_state,
1162 static const struct amdgpu_gmc_funcs gmc_v6_0_gmc_funcs = {
1163 .flush_gpu_tlb = gmc_v6_0_flush_gpu_tlb,
1164 .emit_flush_gpu_tlb = gmc_v6_0_emit_flush_gpu_tlb,
1165 .set_pte_pde = gmc_v6_0_set_pte_pde,
1166 .set_prt = gmc_v6_0_set_prt,
1167 .get_vm_pde = gmc_v6_0_get_vm_pde,
1168 .get_vm_pte_flags = gmc_v6_0_get_vm_pte_flags
1171 static const struct amdgpu_irq_src_funcs gmc_v6_0_irq_funcs = {
1172 .set = gmc_v6_0_vm_fault_interrupt_state,
1173 .process = gmc_v6_0_process_interrupt,
1176 static void gmc_v6_0_set_gmc_funcs(struct amdgpu_device *adev)
1178 if (adev->gmc.gmc_funcs == NULL)
1179 adev->gmc.gmc_funcs = &gmc_v6_0_gmc_funcs;
1182 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev)
1184 adev->gmc.vm_fault.num_types = 1;
1185 adev->gmc.vm_fault.funcs = &gmc_v6_0_irq_funcs;
1188 const struct amdgpu_ip_block_version gmc_v6_0_ip_block =
1190 .type = AMD_IP_BLOCK_TYPE_GMC,
1194 .funcs = &gmc_v6_0_ip_funcs,