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/si58_mc.bin");
51 #define MC_SEQ_MISC0__MT__MASK 0xf0000000
52 #define MC_SEQ_MISC0__MT__GDDR1 0x10000000
53 #define MC_SEQ_MISC0__MT__DDR2 0x20000000
54 #define MC_SEQ_MISC0__MT__GDDR3 0x30000000
55 #define MC_SEQ_MISC0__MT__GDDR4 0x40000000
56 #define MC_SEQ_MISC0__MT__GDDR5 0x50000000
57 #define MC_SEQ_MISC0__MT__HBM 0x60000000
58 #define MC_SEQ_MISC0__MT__DDR3 0xB0000000
61 static const u32 crtc_offsets[6] =
63 SI_CRTC0_REGISTER_OFFSET,
64 SI_CRTC1_REGISTER_OFFSET,
65 SI_CRTC2_REGISTER_OFFSET,
66 SI_CRTC3_REGISTER_OFFSET,
67 SI_CRTC4_REGISTER_OFFSET,
68 SI_CRTC5_REGISTER_OFFSET
71 static void gmc_v6_0_mc_stop(struct amdgpu_device *adev)
75 gmc_v6_0_wait_for_idle((void *)adev);
77 blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
78 if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
79 /* Block CPU access */
80 WREG32(mmBIF_FB_EN, 0);
82 blackout = REG_SET_FIELD(blackout,
83 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
84 WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout | 1);
86 /* wait for the MC to settle */
91 static void gmc_v6_0_mc_resume(struct amdgpu_device *adev)
95 /* unblackout the MC */
96 tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
97 tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
98 WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
99 /* allow CPU access */
100 tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
101 tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
102 WREG32(mmBIF_FB_EN, tmp);
105 static int gmc_v6_0_init_microcode(struct amdgpu_device *adev)
107 const char *chip_name;
110 bool is_58_fw = false;
114 switch (adev->asic_type) {
116 chip_name = "tahiti";
119 chip_name = "pitcairn";
128 chip_name = "hainan";
133 /* this memory configuration requires special firmware */
134 if (((RREG32(mmMC_SEQ_MISC0) & 0xff000000) >> 24) == 0x58)
138 snprintf(fw_name, sizeof(fw_name), "amdgpu/si58_mc.bin");
140 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mc.bin", chip_name);
141 err = request_firmware(&adev->gmc.fw, fw_name, adev->dev);
145 err = amdgpu_ucode_validate(adev->gmc.fw);
150 "si_mc: Failed to load firmware \"%s\"\n",
152 release_firmware(adev->gmc.fw);
158 static int gmc_v6_0_mc_load_microcode(struct amdgpu_device *adev)
160 const __le32 *new_fw_data = NULL;
162 const __le32 *new_io_mc_regs = NULL;
163 int i, regs_size, ucode_size;
164 const struct mc_firmware_header_v1_0 *hdr;
169 hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
171 amdgpu_ucode_print_mc_hdr(&hdr->header);
173 adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
174 regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
175 new_io_mc_regs = (const __le32 *)
176 (adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
177 ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
178 new_fw_data = (const __le32 *)
179 (adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
181 running = RREG32(mmMC_SEQ_SUP_CNTL) & MC_SEQ_SUP_CNTL__RUN_MASK;
185 /* reset the engine and set to writable */
186 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
187 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
189 /* load mc io regs */
190 for (i = 0; i < regs_size; i++) {
191 WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(new_io_mc_regs++));
192 WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(new_io_mc_regs++));
194 /* load the MC ucode */
195 for (i = 0; i < ucode_size; i++) {
196 WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(new_fw_data++));
199 /* put the engine back into the active state */
200 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
201 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
202 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
204 /* wait for training to complete */
205 for (i = 0; i < adev->usec_timeout; i++) {
206 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D0_MASK)
210 for (i = 0; i < adev->usec_timeout; i++) {
211 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D1_MASK)
221 static void gmc_v6_0_vram_gtt_location(struct amdgpu_device *adev,
222 struct amdgpu_gmc *mc)
224 u64 base = RREG32(mmMC_VM_FB_LOCATION) & 0xFFFF;
227 amdgpu_gmc_vram_location(adev, &adev->gmc, base);
228 amdgpu_gmc_gart_location(adev, mc);
231 static void gmc_v6_0_mc_program(struct amdgpu_device *adev)
236 for (i = 0, j = 0; i < 32; i++, j += 0x6) {
237 WREG32((0xb05 + j), 0x00000000);
238 WREG32((0xb06 + j), 0x00000000);
239 WREG32((0xb07 + j), 0x00000000);
240 WREG32((0xb08 + j), 0x00000000);
241 WREG32((0xb09 + j), 0x00000000);
243 WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
245 if (gmc_v6_0_wait_for_idle((void *)adev)) {
246 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
249 if (adev->mode_info.num_crtc) {
252 /* Lockout access through VGA aperture*/
253 tmp = RREG32(mmVGA_HDP_CONTROL);
254 tmp |= VGA_HDP_CONTROL__VGA_MEMORY_DISABLE_MASK;
255 WREG32(mmVGA_HDP_CONTROL, tmp);
257 /* disable VGA render */
258 tmp = RREG32(mmVGA_RENDER_CONTROL);
259 tmp &= ~VGA_VSTATUS_CNTL;
260 WREG32(mmVGA_RENDER_CONTROL, tmp);
262 /* Update configuration */
263 WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
264 adev->gmc.vram_start >> 12);
265 WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
266 adev->gmc.vram_end >> 12);
267 WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
268 adev->vram_scratch.gpu_addr >> 12);
269 WREG32(mmMC_VM_AGP_BASE, 0);
270 WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
271 WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
273 if (gmc_v6_0_wait_for_idle((void *)adev)) {
274 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
278 static int gmc_v6_0_mc_init(struct amdgpu_device *adev)
282 int chansize, numchan;
285 tmp = RREG32(mmMC_ARB_RAMCFG);
286 if (tmp & (1 << 11)) {
288 } else if (tmp & MC_ARB_RAMCFG__CHANSIZE_MASK) {
293 tmp = RREG32(mmMC_SHARED_CHMAP);
294 switch ((tmp & MC_SHARED_CHMAP__NOOFCHAN_MASK) >> MC_SHARED_CHMAP__NOOFCHAN__SHIFT) {
324 adev->gmc.vram_width = numchan * chansize;
325 /* size in MB on si */
326 adev->gmc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
327 adev->gmc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
329 if (!(adev->flags & AMD_IS_APU)) {
330 r = amdgpu_device_resize_fb_bar(adev);
334 adev->gmc.aper_base = pci_resource_start(adev->pdev, 0);
335 adev->gmc.aper_size = pci_resource_len(adev->pdev, 0);
336 adev->gmc.visible_vram_size = adev->gmc.aper_size;
338 /* set the gart size */
339 if (amdgpu_gart_size == -1) {
340 switch (adev->asic_type) {
341 case CHIP_HAINAN: /* no MM engines */
343 adev->gmc.gart_size = 256ULL << 20;
345 case CHIP_VERDE: /* UVD, VCE do not support GPUVM */
346 case CHIP_TAHITI: /* UVD, VCE do not support GPUVM */
347 case CHIP_PITCAIRN: /* UVD, VCE do not support GPUVM */
348 case CHIP_OLAND: /* UVD, VCE do not support GPUVM */
349 adev->gmc.gart_size = 1024ULL << 20;
353 adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
356 gmc_v6_0_vram_gtt_location(adev, &adev->gmc);
361 static void gmc_v6_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid)
363 WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
366 static uint64_t gmc_v6_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
367 unsigned vmid, uint64_t pd_addr)
371 /* write new base address */
373 reg = mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid;
375 reg = mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (vmid - 8);
376 amdgpu_ring_emit_wreg(ring, reg, pd_addr >> 12);
378 /* bits 0-15 are the VM contexts0-15 */
379 amdgpu_ring_emit_wreg(ring, mmVM_INVALIDATE_REQUEST, 1 << vmid);
384 static int gmc_v6_0_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr,
385 uint32_t gpu_page_idx, uint64_t addr,
388 void __iomem *ptr = (void *)cpu_pt_addr;
391 value = addr & 0xFFFFFFFFFFFFF000ULL;
393 writeq(value, ptr + (gpu_page_idx * 8));
398 static uint64_t gmc_v6_0_get_vm_pte_flags(struct amdgpu_device *adev,
401 uint64_t pte_flag = 0;
403 if (flags & AMDGPU_VM_PAGE_READABLE)
404 pte_flag |= AMDGPU_PTE_READABLE;
405 if (flags & AMDGPU_VM_PAGE_WRITEABLE)
406 pte_flag |= AMDGPU_PTE_WRITEABLE;
407 if (flags & AMDGPU_VM_PAGE_PRT)
408 pte_flag |= AMDGPU_PTE_PRT;
413 static void gmc_v6_0_get_vm_pde(struct amdgpu_device *adev, int level,
414 uint64_t *addr, uint64_t *flags)
416 BUG_ON(*addr & 0xFFFFFF0000000FFFULL);
419 static void gmc_v6_0_set_fault_enable_default(struct amdgpu_device *adev,
424 tmp = RREG32(mmVM_CONTEXT1_CNTL);
425 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
426 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
427 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
428 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
429 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
430 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
431 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
432 VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
433 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
434 READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
435 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
436 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
437 WREG32(mmVM_CONTEXT1_CNTL, tmp);
441 + * gmc_v8_0_set_prt - set PRT VM fault
443 + * @adev: amdgpu_device pointer
444 + * @enable: enable/disable VM fault handling for PRT
446 static void gmc_v6_0_set_prt(struct amdgpu_device *adev, bool enable)
450 if (enable && !adev->gmc.prt_warning) {
451 dev_warn(adev->dev, "Disabling VM faults because of PRT request!\n");
452 adev->gmc.prt_warning = true;
455 tmp = RREG32(mmVM_PRT_CNTL);
456 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
457 CB_DISABLE_FAULT_ON_UNMAPPED_ACCESS,
459 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
460 TC_DISABLE_FAULT_ON_UNMAPPED_ACCESS,
462 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
463 L2_CACHE_STORE_INVALID_ENTRIES,
465 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
466 L1_TLB_STORE_INVALID_ENTRIES,
468 WREG32(mmVM_PRT_CNTL, tmp);
471 uint32_t low = AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT;
472 uint32_t high = adev->vm_manager.max_pfn -
473 (AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT);
475 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, low);
476 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, low);
477 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, low);
478 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, low);
479 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, high);
480 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, high);
481 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, high);
482 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, high);
484 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, 0xfffffff);
485 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, 0xfffffff);
486 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, 0xfffffff);
487 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, 0xfffffff);
488 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, 0x0);
489 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, 0x0);
490 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, 0x0);
491 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, 0x0);
495 static int gmc_v6_0_gart_enable(struct amdgpu_device *adev)
501 if (adev->gart.bo == NULL) {
502 dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
505 r = amdgpu_gart_table_vram_pin(adev);
509 table_addr = amdgpu_bo_gpu_offset(adev->gart.bo);
511 /* Setup TLB control */
512 WREG32(mmMC_VM_MX_L1_TLB_CNTL,
514 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_TLB_MASK |
515 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_FRAGMENT_PROCESSING_MASK |
516 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
517 MC_VM_MX_L1_TLB_CNTL__ENABLE_ADVANCED_DRIVER_MODEL_MASK |
518 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
521 VM_L2_CNTL__ENABLE_L2_CACHE_MASK |
522 VM_L2_CNTL__ENABLE_L2_FRAGMENT_PROCESSING_MASK |
523 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
524 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
525 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
526 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
527 WREG32(mmVM_L2_CNTL2,
528 VM_L2_CNTL2__INVALIDATE_ALL_L1_TLBS_MASK |
529 VM_L2_CNTL2__INVALIDATE_L2_CACHE_MASK);
531 field = adev->vm_manager.fragment_size;
532 WREG32(mmVM_L2_CNTL3,
533 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
534 (field << VM_L2_CNTL3__BANK_SELECT__SHIFT) |
535 (field << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
537 WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->gmc.gart_start >> 12);
538 WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->gmc.gart_end >> 12);
539 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, table_addr >> 12);
540 WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
541 (u32)(adev->dummy_page_addr >> 12));
542 WREG32(mmVM_CONTEXT0_CNTL2, 0);
543 WREG32(mmVM_CONTEXT0_CNTL,
544 VM_CONTEXT0_CNTL__ENABLE_CONTEXT_MASK |
545 (0UL << VM_CONTEXT0_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
546 VM_CONTEXT0_CNTL__RANGE_PROTECTION_FAULT_ENABLE_DEFAULT_MASK);
552 /* empty context1-15 */
553 /* set vm size, must be a multiple of 4 */
554 WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
555 WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
556 /* Assign the pt base to something valid for now; the pts used for
557 * the VMs are determined by the application and setup and assigned
558 * on the fly in the vm part of radeon_gart.c
560 for (i = 1; i < 16; i++) {
562 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
565 WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
569 /* enable context1-15 */
570 WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
571 (u32)(adev->dummy_page_addr >> 12));
572 WREG32(mmVM_CONTEXT1_CNTL2, 4);
573 WREG32(mmVM_CONTEXT1_CNTL,
574 VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK |
575 (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
576 ((adev->vm_manager.block_size - 9)
577 << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT));
578 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
579 gmc_v6_0_set_fault_enable_default(adev, false);
581 gmc_v6_0_set_fault_enable_default(adev, true);
583 gmc_v6_0_flush_gpu_tlb(adev, 0);
584 dev_info(adev->dev, "PCIE GART of %uM enabled (table at 0x%016llX).\n",
585 (unsigned)(adev->gmc.gart_size >> 20),
586 (unsigned long long)table_addr);
587 adev->gart.ready = true;
591 static int gmc_v6_0_gart_init(struct amdgpu_device *adev)
596 dev_warn(adev->dev, "gmc_v6_0 PCIE GART already initialized\n");
599 r = amdgpu_gart_init(adev);
602 adev->gart.table_size = adev->gart.num_gpu_pages * 8;
603 adev->gart.gart_pte_flags = 0;
604 return amdgpu_gart_table_vram_alloc(adev);
607 static void gmc_v6_0_gart_disable(struct amdgpu_device *adev)
611 for (i = 1; i < 16; ++i) {
614 reg = VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i ;
616 reg = VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (i - 8);
617 adev->vm_manager.saved_table_addr[i] = RREG32(reg);
620 /* Disable all tables */
621 WREG32(mmVM_CONTEXT0_CNTL, 0);
622 WREG32(mmVM_CONTEXT1_CNTL, 0);
623 /* Setup TLB control */
624 WREG32(mmMC_VM_MX_L1_TLB_CNTL,
625 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
626 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
629 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
630 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
631 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
632 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
633 WREG32(mmVM_L2_CNTL2, 0);
634 WREG32(mmVM_L2_CNTL3,
635 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
636 (0UL << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
637 amdgpu_gart_table_vram_unpin(adev);
640 static void gmc_v6_0_vm_decode_fault(struct amdgpu_device *adev,
641 u32 status, u32 addr, u32 mc_client)
644 u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
645 u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
647 char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
648 (mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
650 mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
653 dev_err(adev->dev, "VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
654 protections, vmid, addr,
655 REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
657 "write" : "read", block, mc_client, mc_id);
661 static const u32 mc_cg_registers[] = {
673 static const u32 mc_cg_ls_en[] = {
674 MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK,
675 MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK,
676 MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK,
677 MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK,
678 ATC_MISC_CG__MEM_LS_ENABLE_MASK,
679 MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK,
680 MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK,
681 MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK,
682 VM_L2_CG__MEM_LS_ENABLE_MASK,
685 static const u32 mc_cg_en[] = {
686 MC_HUB_MISC_HUB_CG__ENABLE_MASK,
687 MC_HUB_MISC_SIP_CG__ENABLE_MASK,
688 MC_HUB_MISC_VM_CG__ENABLE_MASK,
689 MC_XPB_CLK_GAT__ENABLE_MASK,
690 ATC_MISC_CG__ENABLE_MASK,
691 MC_CITF_MISC_WR_CG__ENABLE_MASK,
692 MC_CITF_MISC_RD_CG__ENABLE_MASK,
693 MC_CITF_MISC_VM_CG__ENABLE_MASK,
694 VM_L2_CG__ENABLE_MASK,
697 static void gmc_v6_0_enable_mc_ls(struct amdgpu_device *adev,
703 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
704 orig = data = RREG32(mc_cg_registers[i]);
705 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_LS))
706 data |= mc_cg_ls_en[i];
708 data &= ~mc_cg_ls_en[i];
710 WREG32(mc_cg_registers[i], data);
714 static void gmc_v6_0_enable_mc_mgcg(struct amdgpu_device *adev,
720 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
721 orig = data = RREG32(mc_cg_registers[i]);
722 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_MGCG))
725 data &= ~mc_cg_en[i];
727 WREG32(mc_cg_registers[i], data);
731 static void gmc_v6_0_enable_bif_mgls(struct amdgpu_device *adev,
736 orig = data = RREG32_PCIE(ixPCIE_CNTL2);
738 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_BIF_LS)) {
739 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 1);
740 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 1);
741 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 1);
742 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 1);
744 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 0);
745 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 0);
746 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 0);
747 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 0);
751 WREG32_PCIE(ixPCIE_CNTL2, data);
754 static void gmc_v6_0_enable_hdp_mgcg(struct amdgpu_device *adev,
759 orig = data = RREG32(mmHDP_HOST_PATH_CNTL);
761 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_MGCG))
762 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 0);
764 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 1);
767 WREG32(mmHDP_HOST_PATH_CNTL, data);
770 static void gmc_v6_0_enable_hdp_ls(struct amdgpu_device *adev,
775 orig = data = RREG32(mmHDP_MEM_POWER_LS);
777 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_LS))
778 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 1);
780 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 0);
783 WREG32(mmHDP_MEM_POWER_LS, data);
787 static int gmc_v6_0_convert_vram_type(int mc_seq_vram_type)
789 switch (mc_seq_vram_type) {
790 case MC_SEQ_MISC0__MT__GDDR1:
791 return AMDGPU_VRAM_TYPE_GDDR1;
792 case MC_SEQ_MISC0__MT__DDR2:
793 return AMDGPU_VRAM_TYPE_DDR2;
794 case MC_SEQ_MISC0__MT__GDDR3:
795 return AMDGPU_VRAM_TYPE_GDDR3;
796 case MC_SEQ_MISC0__MT__GDDR4:
797 return AMDGPU_VRAM_TYPE_GDDR4;
798 case MC_SEQ_MISC0__MT__GDDR5:
799 return AMDGPU_VRAM_TYPE_GDDR5;
800 case MC_SEQ_MISC0__MT__DDR3:
801 return AMDGPU_VRAM_TYPE_DDR3;
803 return AMDGPU_VRAM_TYPE_UNKNOWN;
807 static int gmc_v6_0_early_init(void *handle)
809 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
811 gmc_v6_0_set_gmc_funcs(adev);
812 gmc_v6_0_set_irq_funcs(adev);
817 static int gmc_v6_0_late_init(void *handle)
819 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
821 amdgpu_bo_late_init(adev);
823 if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
824 return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
829 static unsigned gmc_v6_0_get_vbios_fb_size(struct amdgpu_device *adev)
831 u32 d1vga_control = RREG32(mmD1VGA_CONTROL);
834 if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
835 size = 9 * 1024 * 1024; /* reserve 8MB for vga emulator and 1 MB for FB */
837 u32 viewport = RREG32(mmVIEWPORT_SIZE);
838 size = (REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
839 REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_WIDTH) *
842 /* return 0 if the pre-OS buffer uses up most of vram */
843 if ((adev->gmc.real_vram_size - size) < (8 * 1024 * 1024))
848 static int gmc_v6_0_sw_init(void *handle)
852 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
854 if (adev->flags & AMD_IS_APU) {
855 adev->gmc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
857 u32 tmp = RREG32(mmMC_SEQ_MISC0);
858 tmp &= MC_SEQ_MISC0__MT__MASK;
859 adev->gmc.vram_type = gmc_v6_0_convert_vram_type(tmp);
862 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 146, &adev->gmc.vm_fault);
866 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 147, &adev->gmc.vm_fault);
870 amdgpu_vm_adjust_size(adev, 64, 9, 1, 40);
872 adev->gmc.mc_mask = 0xffffffffffULL;
874 adev->need_dma32 = false;
875 dma_bits = adev->need_dma32 ? 32 : 40;
876 r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
878 adev->need_dma32 = true;
880 dev_warn(adev->dev, "amdgpu: No suitable DMA available.\n");
882 r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
884 pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
885 dev_warn(adev->dev, "amdgpu: No coherent DMA available.\n");
887 adev->need_swiotlb = drm_get_max_iomem() > ((u64)1 << dma_bits);
889 r = gmc_v6_0_init_microcode(adev);
891 dev_err(adev->dev, "Failed to load mc firmware!\n");
895 r = gmc_v6_0_mc_init(adev);
899 adev->gmc.stolen_size = gmc_v6_0_get_vbios_fb_size(adev);
901 r = amdgpu_bo_init(adev);
905 r = gmc_v6_0_gart_init(adev);
911 * VMID 0 is reserved for System
912 * amdgpu graphics/compute will use VMIDs 1-7
913 * amdkfd will use VMIDs 8-15
915 adev->vm_manager.id_mgr[0].num_ids = AMDGPU_NUM_OF_VMIDS;
916 amdgpu_vm_manager_init(adev);
918 /* base offset of vram pages */
919 if (adev->flags & AMD_IS_APU) {
920 u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
923 adev->vm_manager.vram_base_offset = tmp;
925 adev->vm_manager.vram_base_offset = 0;
931 static int gmc_v6_0_sw_fini(void *handle)
933 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
935 amdgpu_gem_force_release(adev);
936 amdgpu_vm_manager_fini(adev);
937 amdgpu_gart_table_vram_free(adev);
938 amdgpu_bo_fini(adev);
939 amdgpu_gart_fini(adev);
940 release_firmware(adev->gmc.fw);
946 static int gmc_v6_0_hw_init(void *handle)
949 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
951 gmc_v6_0_mc_program(adev);
953 if (!(adev->flags & AMD_IS_APU)) {
954 r = gmc_v6_0_mc_load_microcode(adev);
956 dev_err(adev->dev, "Failed to load MC firmware!\n");
961 r = gmc_v6_0_gart_enable(adev);
968 static int gmc_v6_0_hw_fini(void *handle)
970 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
972 amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
973 gmc_v6_0_gart_disable(adev);
978 static int gmc_v6_0_suspend(void *handle)
980 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
982 gmc_v6_0_hw_fini(adev);
987 static int gmc_v6_0_resume(void *handle)
990 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
992 r = gmc_v6_0_hw_init(adev);
996 amdgpu_vmid_reset_all(adev);
1001 static bool gmc_v6_0_is_idle(void *handle)
1003 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1004 u32 tmp = RREG32(mmSRBM_STATUS);
1006 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1007 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
1013 static int gmc_v6_0_wait_for_idle(void *handle)
1016 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1018 for (i = 0; i < adev->usec_timeout; i++) {
1019 if (gmc_v6_0_is_idle(handle))
1027 static int gmc_v6_0_soft_reset(void *handle)
1029 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1030 u32 srbm_soft_reset = 0;
1031 u32 tmp = RREG32(mmSRBM_STATUS);
1033 if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1034 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1035 SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1037 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1038 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1039 if (!(adev->flags & AMD_IS_APU))
1040 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1041 SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1044 if (srbm_soft_reset) {
1045 gmc_v6_0_mc_stop(adev);
1046 if (gmc_v6_0_wait_for_idle(adev)) {
1047 dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1051 tmp = RREG32(mmSRBM_SOFT_RESET);
1052 tmp |= srbm_soft_reset;
1053 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1054 WREG32(mmSRBM_SOFT_RESET, tmp);
1055 tmp = RREG32(mmSRBM_SOFT_RESET);
1059 tmp &= ~srbm_soft_reset;
1060 WREG32(mmSRBM_SOFT_RESET, tmp);
1061 tmp = RREG32(mmSRBM_SOFT_RESET);
1065 gmc_v6_0_mc_resume(adev);
1072 static int gmc_v6_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1073 struct amdgpu_irq_src *src,
1075 enum amdgpu_interrupt_state state)
1078 u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1079 VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1080 VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1081 VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1082 VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1083 VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1086 case AMDGPU_IRQ_STATE_DISABLE:
1087 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1089 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1090 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1092 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1094 case AMDGPU_IRQ_STATE_ENABLE:
1095 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1097 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1098 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1100 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1109 static int gmc_v6_0_process_interrupt(struct amdgpu_device *adev,
1110 struct amdgpu_irq_src *source,
1111 struct amdgpu_iv_entry *entry)
1115 addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1116 status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1117 WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1119 if (!addr && !status)
1122 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1123 gmc_v6_0_set_fault_enable_default(adev, false);
1125 if (printk_ratelimit()) {
1126 dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1127 entry->src_id, entry->src_data[0]);
1128 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
1130 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1132 gmc_v6_0_vm_decode_fault(adev, status, addr, 0);
1138 static int gmc_v6_0_set_clockgating_state(void *handle,
1139 enum amd_clockgating_state state)
1144 static int gmc_v6_0_set_powergating_state(void *handle,
1145 enum amd_powergating_state state)
1150 static const struct amd_ip_funcs gmc_v6_0_ip_funcs = {
1152 .early_init = gmc_v6_0_early_init,
1153 .late_init = gmc_v6_0_late_init,
1154 .sw_init = gmc_v6_0_sw_init,
1155 .sw_fini = gmc_v6_0_sw_fini,
1156 .hw_init = gmc_v6_0_hw_init,
1157 .hw_fini = gmc_v6_0_hw_fini,
1158 .suspend = gmc_v6_0_suspend,
1159 .resume = gmc_v6_0_resume,
1160 .is_idle = gmc_v6_0_is_idle,
1161 .wait_for_idle = gmc_v6_0_wait_for_idle,
1162 .soft_reset = gmc_v6_0_soft_reset,
1163 .set_clockgating_state = gmc_v6_0_set_clockgating_state,
1164 .set_powergating_state = gmc_v6_0_set_powergating_state,
1167 static const struct amdgpu_gmc_funcs gmc_v6_0_gmc_funcs = {
1168 .flush_gpu_tlb = gmc_v6_0_flush_gpu_tlb,
1169 .emit_flush_gpu_tlb = gmc_v6_0_emit_flush_gpu_tlb,
1170 .set_pte_pde = gmc_v6_0_set_pte_pde,
1171 .set_prt = gmc_v6_0_set_prt,
1172 .get_vm_pde = gmc_v6_0_get_vm_pde,
1173 .get_vm_pte_flags = gmc_v6_0_get_vm_pte_flags
1176 static const struct amdgpu_irq_src_funcs gmc_v6_0_irq_funcs = {
1177 .set = gmc_v6_0_vm_fault_interrupt_state,
1178 .process = gmc_v6_0_process_interrupt,
1181 static void gmc_v6_0_set_gmc_funcs(struct amdgpu_device *adev)
1183 adev->gmc.gmc_funcs = &gmc_v6_0_gmc_funcs;
1186 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev)
1188 adev->gmc.vm_fault.num_types = 1;
1189 adev->gmc.vm_fault.funcs = &gmc_v6_0_irq_funcs;
1192 const struct amdgpu_ip_block_version gmc_v6_0_ip_block =
1194 .type = AMD_IP_BLOCK_TYPE_GMC,
1198 .funcs = &gmc_v6_0_ip_funcs,