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"
29 #include "amdgpu_gem.h"
31 #include "bif/bif_3_0_d.h"
32 #include "bif/bif_3_0_sh_mask.h"
33 #include "oss/oss_1_0_d.h"
34 #include "oss/oss_1_0_sh_mask.h"
35 #include "gmc/gmc_6_0_d.h"
36 #include "gmc/gmc_6_0_sh_mask.h"
37 #include "dce/dce_6_0_d.h"
38 #include "dce/dce_6_0_sh_mask.h"
41 static void gmc_v6_0_set_gmc_funcs(struct amdgpu_device *adev);
42 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev);
43 static int gmc_v6_0_wait_for_idle(void *handle);
45 MODULE_FIRMWARE("amdgpu/tahiti_mc.bin");
46 MODULE_FIRMWARE("amdgpu/pitcairn_mc.bin");
47 MODULE_FIRMWARE("amdgpu/verde_mc.bin");
48 MODULE_FIRMWARE("amdgpu/oland_mc.bin");
49 MODULE_FIRMWARE("amdgpu/hainan_mc.bin");
50 MODULE_FIRMWARE("amdgpu/si58_mc.bin");
52 #define MC_SEQ_MISC0__MT__MASK 0xf0000000
53 #define MC_SEQ_MISC0__MT__GDDR1 0x10000000
54 #define MC_SEQ_MISC0__MT__DDR2 0x20000000
55 #define MC_SEQ_MISC0__MT__GDDR3 0x30000000
56 #define MC_SEQ_MISC0__MT__GDDR4 0x40000000
57 #define MC_SEQ_MISC0__MT__GDDR5 0x50000000
58 #define MC_SEQ_MISC0__MT__HBM 0x60000000
59 #define MC_SEQ_MISC0__MT__DDR3 0xB0000000
62 static const u32 crtc_offsets[6] =
64 SI_CRTC0_REGISTER_OFFSET,
65 SI_CRTC1_REGISTER_OFFSET,
66 SI_CRTC2_REGISTER_OFFSET,
67 SI_CRTC3_REGISTER_OFFSET,
68 SI_CRTC4_REGISTER_OFFSET,
69 SI_CRTC5_REGISTER_OFFSET
72 static void gmc_v6_0_mc_stop(struct amdgpu_device *adev)
76 gmc_v6_0_wait_for_idle((void *)adev);
78 blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
79 if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
80 /* Block CPU access */
81 WREG32(mmBIF_FB_EN, 0);
83 blackout = REG_SET_FIELD(blackout,
84 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
85 WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout | 1);
87 /* wait for the MC to settle */
92 static void gmc_v6_0_mc_resume(struct amdgpu_device *adev)
96 /* unblackout the MC */
97 tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
98 tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
99 WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
100 /* allow CPU access */
101 tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
102 tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
103 WREG32(mmBIF_FB_EN, tmp);
106 static int gmc_v6_0_init_microcode(struct amdgpu_device *adev)
108 const char *chip_name;
111 bool is_58_fw = false;
115 switch (adev->asic_type) {
117 chip_name = "tahiti";
120 chip_name = "pitcairn";
129 chip_name = "hainan";
134 /* this memory configuration requires special firmware */
135 if (((RREG32(mmMC_SEQ_MISC0) & 0xff000000) >> 24) == 0x58)
139 snprintf(fw_name, sizeof(fw_name), "amdgpu/si58_mc.bin");
141 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mc.bin", chip_name);
142 err = request_firmware(&adev->gmc.fw, fw_name, adev->dev);
146 err = amdgpu_ucode_validate(adev->gmc.fw);
151 "si_mc: Failed to load firmware \"%s\"\n",
153 release_firmware(adev->gmc.fw);
159 static int gmc_v6_0_mc_load_microcode(struct amdgpu_device *adev)
161 const __le32 *new_fw_data = NULL;
163 const __le32 *new_io_mc_regs = NULL;
164 int i, regs_size, ucode_size;
165 const struct mc_firmware_header_v1_0 *hdr;
170 hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
172 amdgpu_ucode_print_mc_hdr(&hdr->header);
174 adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
175 regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
176 new_io_mc_regs = (const __le32 *)
177 (adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
178 ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
179 new_fw_data = (const __le32 *)
180 (adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
182 running = RREG32(mmMC_SEQ_SUP_CNTL) & MC_SEQ_SUP_CNTL__RUN_MASK;
186 /* reset the engine and set to writable */
187 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
188 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
190 /* load mc io regs */
191 for (i = 0; i < regs_size; i++) {
192 WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(new_io_mc_regs++));
193 WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(new_io_mc_regs++));
195 /* load the MC ucode */
196 for (i = 0; i < ucode_size; i++) {
197 WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(new_fw_data++));
200 /* put the engine back into the active state */
201 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
202 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
203 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
205 /* wait for training to complete */
206 for (i = 0; i < adev->usec_timeout; i++) {
207 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D0_MASK)
211 for (i = 0; i < adev->usec_timeout; i++) {
212 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D1_MASK)
222 static void gmc_v6_0_vram_gtt_location(struct amdgpu_device *adev,
223 struct amdgpu_gmc *mc)
225 u64 base = RREG32(mmMC_VM_FB_LOCATION) & 0xFFFF;
228 amdgpu_gmc_vram_location(adev, &adev->gmc, base);
229 amdgpu_gmc_gart_location(adev, mc);
232 static void gmc_v6_0_mc_program(struct amdgpu_device *adev)
237 for (i = 0, j = 0; i < 32; i++, j += 0x6) {
238 WREG32((0xb05 + j), 0x00000000);
239 WREG32((0xb06 + j), 0x00000000);
240 WREG32((0xb07 + j), 0x00000000);
241 WREG32((0xb08 + j), 0x00000000);
242 WREG32((0xb09 + j), 0x00000000);
244 WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
246 if (gmc_v6_0_wait_for_idle((void *)adev)) {
247 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
250 if (adev->mode_info.num_crtc) {
253 /* Lockout access through VGA aperture*/
254 tmp = RREG32(mmVGA_HDP_CONTROL);
255 tmp |= VGA_HDP_CONTROL__VGA_MEMORY_DISABLE_MASK;
256 WREG32(mmVGA_HDP_CONTROL, tmp);
258 /* disable VGA render */
259 tmp = RREG32(mmVGA_RENDER_CONTROL);
260 tmp &= ~VGA_VSTATUS_CNTL;
261 WREG32(mmVGA_RENDER_CONTROL, tmp);
263 /* Update configuration */
264 WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
265 adev->gmc.vram_start >> 12);
266 WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
267 adev->gmc.vram_end >> 12);
268 WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
269 adev->vram_scratch.gpu_addr >> 12);
270 WREG32(mmMC_VM_AGP_BASE, 0);
271 WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
272 WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
274 if (gmc_v6_0_wait_for_idle((void *)adev)) {
275 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
279 static int gmc_v6_0_mc_init(struct amdgpu_device *adev)
283 int chansize, numchan;
286 tmp = RREG32(mmMC_ARB_RAMCFG);
287 if (tmp & (1 << 11)) {
289 } else if (tmp & MC_ARB_RAMCFG__CHANSIZE_MASK) {
294 tmp = RREG32(mmMC_SHARED_CHMAP);
295 switch ((tmp & MC_SHARED_CHMAP__NOOFCHAN_MASK) >> MC_SHARED_CHMAP__NOOFCHAN__SHIFT) {
325 adev->gmc.vram_width = numchan * chansize;
326 /* size in MB on si */
327 adev->gmc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
328 adev->gmc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
330 if (!(adev->flags & AMD_IS_APU)) {
331 r = amdgpu_device_resize_fb_bar(adev);
335 adev->gmc.aper_base = pci_resource_start(adev->pdev, 0);
336 adev->gmc.aper_size = pci_resource_len(adev->pdev, 0);
337 adev->gmc.visible_vram_size = adev->gmc.aper_size;
339 /* set the gart size */
340 if (amdgpu_gart_size == -1) {
341 switch (adev->asic_type) {
342 case CHIP_HAINAN: /* no MM engines */
344 adev->gmc.gart_size = 256ULL << 20;
346 case CHIP_VERDE: /* UVD, VCE do not support GPUVM */
347 case CHIP_TAHITI: /* UVD, VCE do not support GPUVM */
348 case CHIP_PITCAIRN: /* UVD, VCE do not support GPUVM */
349 case CHIP_OLAND: /* UVD, VCE do not support GPUVM */
350 adev->gmc.gart_size = 1024ULL << 20;
354 adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
357 gmc_v6_0_vram_gtt_location(adev, &adev->gmc);
362 static void gmc_v6_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid)
364 WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
367 static uint64_t gmc_v6_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
368 unsigned vmid, uint64_t pd_addr)
372 /* write new base address */
374 reg = mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid;
376 reg = mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (vmid - 8);
377 amdgpu_ring_emit_wreg(ring, reg, pd_addr >> 12);
379 /* bits 0-15 are the VM contexts0-15 */
380 amdgpu_ring_emit_wreg(ring, mmVM_INVALIDATE_REQUEST, 1 << vmid);
385 static int gmc_v6_0_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr,
386 uint32_t gpu_page_idx, uint64_t addr,
389 void __iomem *ptr = (void *)cpu_pt_addr;
392 value = addr & 0xFFFFFFFFFFFFF000ULL;
394 writeq(value, ptr + (gpu_page_idx * 8));
399 static uint64_t gmc_v6_0_get_vm_pte_flags(struct amdgpu_device *adev,
402 uint64_t pte_flag = 0;
404 if (flags & AMDGPU_VM_PAGE_READABLE)
405 pte_flag |= AMDGPU_PTE_READABLE;
406 if (flags & AMDGPU_VM_PAGE_WRITEABLE)
407 pte_flag |= AMDGPU_PTE_WRITEABLE;
408 if (flags & AMDGPU_VM_PAGE_PRT)
409 pte_flag |= AMDGPU_PTE_PRT;
414 static void gmc_v6_0_get_vm_pde(struct amdgpu_device *adev, int level,
415 uint64_t *addr, uint64_t *flags)
417 BUG_ON(*addr & 0xFFFFFF0000000FFFULL);
420 static void gmc_v6_0_set_fault_enable_default(struct amdgpu_device *adev,
425 tmp = RREG32(mmVM_CONTEXT1_CNTL);
426 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
427 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
428 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
429 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
430 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
431 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
432 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
433 VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
434 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
435 READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
436 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
437 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
438 WREG32(mmVM_CONTEXT1_CNTL, tmp);
442 + * gmc_v8_0_set_prt - set PRT VM fault
444 + * @adev: amdgpu_device pointer
445 + * @enable: enable/disable VM fault handling for PRT
447 static void gmc_v6_0_set_prt(struct amdgpu_device *adev, bool enable)
451 if (enable && !adev->gmc.prt_warning) {
452 dev_warn(adev->dev, "Disabling VM faults because of PRT request!\n");
453 adev->gmc.prt_warning = true;
456 tmp = RREG32(mmVM_PRT_CNTL);
457 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
458 CB_DISABLE_FAULT_ON_UNMAPPED_ACCESS,
460 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
461 TC_DISABLE_FAULT_ON_UNMAPPED_ACCESS,
463 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
464 L2_CACHE_STORE_INVALID_ENTRIES,
466 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
467 L1_TLB_STORE_INVALID_ENTRIES,
469 WREG32(mmVM_PRT_CNTL, tmp);
472 uint32_t low = AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT;
473 uint32_t high = adev->vm_manager.max_pfn -
474 (AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT);
476 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, low);
477 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, low);
478 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, low);
479 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, low);
480 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, high);
481 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, high);
482 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, high);
483 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, high);
485 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, 0xfffffff);
486 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, 0xfffffff);
487 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, 0xfffffff);
488 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, 0xfffffff);
489 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, 0x0);
490 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, 0x0);
491 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, 0x0);
492 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, 0x0);
496 static int gmc_v6_0_gart_enable(struct amdgpu_device *adev)
502 if (adev->gart.bo == NULL) {
503 dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
506 r = amdgpu_gart_table_vram_pin(adev);
510 table_addr = amdgpu_bo_gpu_offset(adev->gart.bo);
512 /* Setup TLB control */
513 WREG32(mmMC_VM_MX_L1_TLB_CNTL,
515 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_TLB_MASK |
516 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_FRAGMENT_PROCESSING_MASK |
517 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
518 MC_VM_MX_L1_TLB_CNTL__ENABLE_ADVANCED_DRIVER_MODEL_MASK |
519 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
522 VM_L2_CNTL__ENABLE_L2_CACHE_MASK |
523 VM_L2_CNTL__ENABLE_L2_FRAGMENT_PROCESSING_MASK |
524 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
525 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
526 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
527 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
528 WREG32(mmVM_L2_CNTL2,
529 VM_L2_CNTL2__INVALIDATE_ALL_L1_TLBS_MASK |
530 VM_L2_CNTL2__INVALIDATE_L2_CACHE_MASK);
532 field = adev->vm_manager.fragment_size;
533 WREG32(mmVM_L2_CNTL3,
534 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
535 (field << VM_L2_CNTL3__BANK_SELECT__SHIFT) |
536 (field << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
538 WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->gmc.gart_start >> 12);
539 WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->gmc.gart_end >> 12);
540 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, table_addr >> 12);
541 WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
542 (u32)(adev->dummy_page_addr >> 12));
543 WREG32(mmVM_CONTEXT0_CNTL2, 0);
544 WREG32(mmVM_CONTEXT0_CNTL,
545 VM_CONTEXT0_CNTL__ENABLE_CONTEXT_MASK |
546 (0UL << VM_CONTEXT0_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
547 VM_CONTEXT0_CNTL__RANGE_PROTECTION_FAULT_ENABLE_DEFAULT_MASK);
553 /* empty context1-15 */
554 /* set vm size, must be a multiple of 4 */
555 WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
556 WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
557 /* Assign the pt base to something valid for now; the pts used for
558 * the VMs are determined by the application and setup and assigned
559 * on the fly in the vm part of radeon_gart.c
561 for (i = 1; i < 16; i++) {
563 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
566 WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
570 /* enable context1-15 */
571 WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
572 (u32)(adev->dummy_page_addr >> 12));
573 WREG32(mmVM_CONTEXT1_CNTL2, 4);
574 WREG32(mmVM_CONTEXT1_CNTL,
575 VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK |
576 (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
577 ((adev->vm_manager.block_size - 9)
578 << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT));
579 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
580 gmc_v6_0_set_fault_enable_default(adev, false);
582 gmc_v6_0_set_fault_enable_default(adev, true);
584 gmc_v6_0_flush_gpu_tlb(adev, 0);
585 dev_info(adev->dev, "PCIE GART of %uM enabled (table at 0x%016llX).\n",
586 (unsigned)(adev->gmc.gart_size >> 20),
587 (unsigned long long)table_addr);
588 adev->gart.ready = true;
592 static int gmc_v6_0_gart_init(struct amdgpu_device *adev)
597 dev_warn(adev->dev, "gmc_v6_0 PCIE GART already initialized\n");
600 r = amdgpu_gart_init(adev);
603 adev->gart.table_size = adev->gart.num_gpu_pages * 8;
604 adev->gart.gart_pte_flags = 0;
605 return amdgpu_gart_table_vram_alloc(adev);
608 static void gmc_v6_0_gart_disable(struct amdgpu_device *adev)
612 for (i = 1; i < 16; ++i) {
615 reg = VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i ;
617 reg = VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (i - 8);
618 adev->vm_manager.saved_table_addr[i] = RREG32(reg);
621 /* Disable all tables */
622 WREG32(mmVM_CONTEXT0_CNTL, 0);
623 WREG32(mmVM_CONTEXT1_CNTL, 0);
624 /* Setup TLB control */
625 WREG32(mmMC_VM_MX_L1_TLB_CNTL,
626 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
627 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
630 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
631 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
632 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
633 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
634 WREG32(mmVM_L2_CNTL2, 0);
635 WREG32(mmVM_L2_CNTL3,
636 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
637 (0UL << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
638 amdgpu_gart_table_vram_unpin(adev);
641 static void gmc_v6_0_vm_decode_fault(struct amdgpu_device *adev,
642 u32 status, u32 addr, u32 mc_client)
645 u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
646 u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
648 char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
649 (mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
651 mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
654 dev_err(adev->dev, "VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
655 protections, vmid, addr,
656 REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
658 "write" : "read", block, mc_client, mc_id);
662 static const u32 mc_cg_registers[] = {
674 static const u32 mc_cg_ls_en[] = {
675 MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK,
676 MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK,
677 MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK,
678 MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK,
679 ATC_MISC_CG__MEM_LS_ENABLE_MASK,
680 MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK,
681 MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK,
682 MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK,
683 VM_L2_CG__MEM_LS_ENABLE_MASK,
686 static const u32 mc_cg_en[] = {
687 MC_HUB_MISC_HUB_CG__ENABLE_MASK,
688 MC_HUB_MISC_SIP_CG__ENABLE_MASK,
689 MC_HUB_MISC_VM_CG__ENABLE_MASK,
690 MC_XPB_CLK_GAT__ENABLE_MASK,
691 ATC_MISC_CG__ENABLE_MASK,
692 MC_CITF_MISC_WR_CG__ENABLE_MASK,
693 MC_CITF_MISC_RD_CG__ENABLE_MASK,
694 MC_CITF_MISC_VM_CG__ENABLE_MASK,
695 VM_L2_CG__ENABLE_MASK,
698 static void gmc_v6_0_enable_mc_ls(struct amdgpu_device *adev,
704 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
705 orig = data = RREG32(mc_cg_registers[i]);
706 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_LS))
707 data |= mc_cg_ls_en[i];
709 data &= ~mc_cg_ls_en[i];
711 WREG32(mc_cg_registers[i], data);
715 static void gmc_v6_0_enable_mc_mgcg(struct amdgpu_device *adev,
721 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
722 orig = data = RREG32(mc_cg_registers[i]);
723 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_MGCG))
726 data &= ~mc_cg_en[i];
728 WREG32(mc_cg_registers[i], data);
732 static void gmc_v6_0_enable_bif_mgls(struct amdgpu_device *adev,
737 orig = data = RREG32_PCIE(ixPCIE_CNTL2);
739 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_BIF_LS)) {
740 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 1);
741 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 1);
742 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 1);
743 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 1);
745 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 0);
746 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 0);
747 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 0);
748 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 0);
752 WREG32_PCIE(ixPCIE_CNTL2, data);
755 static void gmc_v6_0_enable_hdp_mgcg(struct amdgpu_device *adev,
760 orig = data = RREG32(mmHDP_HOST_PATH_CNTL);
762 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_MGCG))
763 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 0);
765 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 1);
768 WREG32(mmHDP_HOST_PATH_CNTL, data);
771 static void gmc_v6_0_enable_hdp_ls(struct amdgpu_device *adev,
776 orig = data = RREG32(mmHDP_MEM_POWER_LS);
778 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_LS))
779 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 1);
781 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 0);
784 WREG32(mmHDP_MEM_POWER_LS, data);
788 static int gmc_v6_0_convert_vram_type(int mc_seq_vram_type)
790 switch (mc_seq_vram_type) {
791 case MC_SEQ_MISC0__MT__GDDR1:
792 return AMDGPU_VRAM_TYPE_GDDR1;
793 case MC_SEQ_MISC0__MT__DDR2:
794 return AMDGPU_VRAM_TYPE_DDR2;
795 case MC_SEQ_MISC0__MT__GDDR3:
796 return AMDGPU_VRAM_TYPE_GDDR3;
797 case MC_SEQ_MISC0__MT__GDDR4:
798 return AMDGPU_VRAM_TYPE_GDDR4;
799 case MC_SEQ_MISC0__MT__GDDR5:
800 return AMDGPU_VRAM_TYPE_GDDR5;
801 case MC_SEQ_MISC0__MT__DDR3:
802 return AMDGPU_VRAM_TYPE_DDR3;
804 return AMDGPU_VRAM_TYPE_UNKNOWN;
808 static int gmc_v6_0_early_init(void *handle)
810 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
812 gmc_v6_0_set_gmc_funcs(adev);
813 gmc_v6_0_set_irq_funcs(adev);
818 static int gmc_v6_0_late_init(void *handle)
820 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
822 amdgpu_bo_late_init(adev);
824 if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
825 return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
830 static unsigned gmc_v6_0_get_vbios_fb_size(struct amdgpu_device *adev)
832 u32 d1vga_control = RREG32(mmD1VGA_CONTROL);
835 if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
836 size = 9 * 1024 * 1024; /* reserve 8MB for vga emulator and 1 MB for FB */
838 u32 viewport = RREG32(mmVIEWPORT_SIZE);
839 size = (REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
840 REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_WIDTH) *
843 /* return 0 if the pre-OS buffer uses up most of vram */
844 if ((adev->gmc.real_vram_size - size) < (8 * 1024 * 1024))
849 static int gmc_v6_0_sw_init(void *handle)
853 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
855 if (adev->flags & AMD_IS_APU) {
856 adev->gmc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
858 u32 tmp = RREG32(mmMC_SEQ_MISC0);
859 tmp &= MC_SEQ_MISC0__MT__MASK;
860 adev->gmc.vram_type = gmc_v6_0_convert_vram_type(tmp);
863 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 146, &adev->gmc.vm_fault);
867 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 147, &adev->gmc.vm_fault);
871 amdgpu_vm_adjust_size(adev, 64, 9, 1, 40);
873 adev->gmc.mc_mask = 0xffffffffffULL;
875 adev->need_dma32 = false;
876 dma_bits = adev->need_dma32 ? 32 : 40;
877 r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
879 adev->need_dma32 = true;
881 dev_warn(adev->dev, "amdgpu: No suitable DMA available.\n");
883 r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
885 pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
886 dev_warn(adev->dev, "amdgpu: No coherent DMA available.\n");
888 adev->need_swiotlb = drm_get_max_iomem() > ((u64)1 << dma_bits);
890 r = gmc_v6_0_init_microcode(adev);
892 dev_err(adev->dev, "Failed to load mc firmware!\n");
896 r = gmc_v6_0_mc_init(adev);
900 adev->gmc.stolen_size = gmc_v6_0_get_vbios_fb_size(adev);
902 r = amdgpu_bo_init(adev);
906 r = gmc_v6_0_gart_init(adev);
912 * VMID 0 is reserved for System
913 * amdgpu graphics/compute will use VMIDs 1-7
914 * amdkfd will use VMIDs 8-15
916 adev->vm_manager.id_mgr[0].num_ids = AMDGPU_NUM_OF_VMIDS;
917 amdgpu_vm_manager_init(adev);
919 /* base offset of vram pages */
920 if (adev->flags & AMD_IS_APU) {
921 u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
924 adev->vm_manager.vram_base_offset = tmp;
926 adev->vm_manager.vram_base_offset = 0;
932 static int gmc_v6_0_sw_fini(void *handle)
934 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
936 amdgpu_gem_force_release(adev);
937 amdgpu_vm_manager_fini(adev);
938 amdgpu_gart_table_vram_free(adev);
939 amdgpu_bo_fini(adev);
940 amdgpu_gart_fini(adev);
941 release_firmware(adev->gmc.fw);
947 static int gmc_v6_0_hw_init(void *handle)
950 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
952 gmc_v6_0_mc_program(adev);
954 if (!(adev->flags & AMD_IS_APU)) {
955 r = gmc_v6_0_mc_load_microcode(adev);
957 dev_err(adev->dev, "Failed to load MC firmware!\n");
962 r = gmc_v6_0_gart_enable(adev);
969 static int gmc_v6_0_hw_fini(void *handle)
971 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
973 amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
974 gmc_v6_0_gart_disable(adev);
979 static int gmc_v6_0_suspend(void *handle)
981 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
983 gmc_v6_0_hw_fini(adev);
988 static int gmc_v6_0_resume(void *handle)
991 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
993 r = gmc_v6_0_hw_init(adev);
997 amdgpu_vmid_reset_all(adev);
1002 static bool gmc_v6_0_is_idle(void *handle)
1004 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1005 u32 tmp = RREG32(mmSRBM_STATUS);
1007 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1008 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
1014 static int gmc_v6_0_wait_for_idle(void *handle)
1017 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1019 for (i = 0; i < adev->usec_timeout; i++) {
1020 if (gmc_v6_0_is_idle(handle))
1028 static int gmc_v6_0_soft_reset(void *handle)
1030 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1031 u32 srbm_soft_reset = 0;
1032 u32 tmp = RREG32(mmSRBM_STATUS);
1034 if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1035 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1036 SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1038 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1039 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1040 if (!(adev->flags & AMD_IS_APU))
1041 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1042 SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1045 if (srbm_soft_reset) {
1046 gmc_v6_0_mc_stop(adev);
1047 if (gmc_v6_0_wait_for_idle(adev)) {
1048 dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1052 tmp = RREG32(mmSRBM_SOFT_RESET);
1053 tmp |= srbm_soft_reset;
1054 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1055 WREG32(mmSRBM_SOFT_RESET, tmp);
1056 tmp = RREG32(mmSRBM_SOFT_RESET);
1060 tmp &= ~srbm_soft_reset;
1061 WREG32(mmSRBM_SOFT_RESET, tmp);
1062 tmp = RREG32(mmSRBM_SOFT_RESET);
1066 gmc_v6_0_mc_resume(adev);
1073 static int gmc_v6_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1074 struct amdgpu_irq_src *src,
1076 enum amdgpu_interrupt_state state)
1079 u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1080 VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1081 VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1082 VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1083 VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1084 VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1087 case AMDGPU_IRQ_STATE_DISABLE:
1088 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1090 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1091 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1093 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1095 case AMDGPU_IRQ_STATE_ENABLE:
1096 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1098 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1099 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1101 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1110 static int gmc_v6_0_process_interrupt(struct amdgpu_device *adev,
1111 struct amdgpu_irq_src *source,
1112 struct amdgpu_iv_entry *entry)
1116 addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1117 status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1118 WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1120 if (!addr && !status)
1123 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1124 gmc_v6_0_set_fault_enable_default(adev, false);
1126 if (printk_ratelimit()) {
1127 dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1128 entry->src_id, entry->src_data[0]);
1129 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
1131 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1133 gmc_v6_0_vm_decode_fault(adev, status, addr, 0);
1139 static int gmc_v6_0_set_clockgating_state(void *handle,
1140 enum amd_clockgating_state state)
1145 static int gmc_v6_0_set_powergating_state(void *handle,
1146 enum amd_powergating_state state)
1151 static const struct amd_ip_funcs gmc_v6_0_ip_funcs = {
1153 .early_init = gmc_v6_0_early_init,
1154 .late_init = gmc_v6_0_late_init,
1155 .sw_init = gmc_v6_0_sw_init,
1156 .sw_fini = gmc_v6_0_sw_fini,
1157 .hw_init = gmc_v6_0_hw_init,
1158 .hw_fini = gmc_v6_0_hw_fini,
1159 .suspend = gmc_v6_0_suspend,
1160 .resume = gmc_v6_0_resume,
1161 .is_idle = gmc_v6_0_is_idle,
1162 .wait_for_idle = gmc_v6_0_wait_for_idle,
1163 .soft_reset = gmc_v6_0_soft_reset,
1164 .set_clockgating_state = gmc_v6_0_set_clockgating_state,
1165 .set_powergating_state = gmc_v6_0_set_powergating_state,
1168 static const struct amdgpu_gmc_funcs gmc_v6_0_gmc_funcs = {
1169 .flush_gpu_tlb = gmc_v6_0_flush_gpu_tlb,
1170 .emit_flush_gpu_tlb = gmc_v6_0_emit_flush_gpu_tlb,
1171 .set_pte_pde = gmc_v6_0_set_pte_pde,
1172 .set_prt = gmc_v6_0_set_prt,
1173 .get_vm_pde = gmc_v6_0_get_vm_pde,
1174 .get_vm_pte_flags = gmc_v6_0_get_vm_pte_flags
1177 static const struct amdgpu_irq_src_funcs gmc_v6_0_irq_funcs = {
1178 .set = gmc_v6_0_vm_fault_interrupt_state,
1179 .process = gmc_v6_0_process_interrupt,
1182 static void gmc_v6_0_set_gmc_funcs(struct amdgpu_device *adev)
1184 adev->gmc.gmc_funcs = &gmc_v6_0_gmc_funcs;
1187 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev)
1189 adev->gmc.vm_fault.num_types = 1;
1190 adev->gmc.vm_fault.funcs = &gmc_v6_0_irq_funcs;
1193 const struct amdgpu_ip_block_version gmc_v6_0_ip_block =
1195 .type = AMD_IP_BLOCK_TYPE_GMC,
1199 .funcs = &gmc_v6_0_ip_funcs,