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)
73 gmc_v6_0_wait_for_idle((void *)adev);
75 blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
76 if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
77 /* Block CPU access */
78 WREG32(mmBIF_FB_EN, 0);
80 blackout = REG_SET_FIELD(blackout,
81 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
82 WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout | 1);
84 /* wait for the MC to settle */
89 static void gmc_v6_0_mc_resume(struct amdgpu_device *adev)
93 /* unblackout the MC */
94 tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
95 tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
96 WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
97 /* allow CPU access */
98 tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
99 tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
100 WREG32(mmBIF_FB_EN, tmp);
103 static int gmc_v6_0_init_microcode(struct amdgpu_device *adev)
105 const char *chip_name;
108 bool is_58_fw = false;
112 switch (adev->asic_type) {
114 chip_name = "tahiti";
117 chip_name = "pitcairn";
126 chip_name = "hainan";
131 /* this memory configuration requires special firmware */
132 if (((RREG32(mmMC_SEQ_MISC0) & 0xff000000) >> 24) == 0x58)
136 snprintf(fw_name, sizeof(fw_name), "radeon/si58_mc.bin");
138 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
139 err = request_firmware(&adev->mc.fw, fw_name, adev->dev);
143 err = amdgpu_ucode_validate(adev->mc.fw);
148 "si_mc: Failed to load firmware \"%s\"\n",
150 release_firmware(adev->mc.fw);
156 static int gmc_v6_0_mc_load_microcode(struct amdgpu_device *adev)
158 const __le32 *new_fw_data = NULL;
160 const __le32 *new_io_mc_regs = NULL;
161 int i, regs_size, ucode_size;
162 const struct mc_firmware_header_v1_0 *hdr;
167 hdr = (const struct mc_firmware_header_v1_0 *)adev->mc.fw->data;
169 amdgpu_ucode_print_mc_hdr(&hdr->header);
171 adev->mc.fw_version = le32_to_cpu(hdr->header.ucode_version);
172 regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
173 new_io_mc_regs = (const __le32 *)
174 (adev->mc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
175 ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
176 new_fw_data = (const __le32 *)
177 (adev->mc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
179 running = RREG32(mmMC_SEQ_SUP_CNTL) & MC_SEQ_SUP_CNTL__RUN_MASK;
183 /* reset the engine and set to writable */
184 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
185 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
187 /* load mc io regs */
188 for (i = 0; i < regs_size; i++) {
189 WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(new_io_mc_regs++));
190 WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(new_io_mc_regs++));
192 /* load the MC ucode */
193 for (i = 0; i < ucode_size; i++) {
194 WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(new_fw_data++));
197 /* put the engine back into the active state */
198 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
199 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
200 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
202 /* wait for training to complete */
203 for (i = 0; i < adev->usec_timeout; i++) {
204 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D0_MASK)
208 for (i = 0; i < adev->usec_timeout; i++) {
209 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D1_MASK)
219 static void gmc_v6_0_vram_gtt_location(struct amdgpu_device *adev,
220 struct amdgpu_mc *mc)
222 u64 base = RREG32(mmMC_VM_FB_LOCATION) & 0xFFFF;
225 if (mc->mc_vram_size > 0xFFC0000000ULL) {
226 dev_warn(adev->dev, "limiting VRAM\n");
227 mc->real_vram_size = 0xFFC0000000ULL;
228 mc->mc_vram_size = 0xFFC0000000ULL;
230 amdgpu_vram_location(adev, &adev->mc, base);
231 amdgpu_gart_location(adev, mc);
234 static void gmc_v6_0_mc_program(struct amdgpu_device *adev)
239 for (i = 0, j = 0; i < 32; i++, j += 0x6) {
240 WREG32((0xb05 + j), 0x00000000);
241 WREG32((0xb06 + j), 0x00000000);
242 WREG32((0xb07 + j), 0x00000000);
243 WREG32((0xb08 + j), 0x00000000);
244 WREG32((0xb09 + j), 0x00000000);
246 WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
248 if (gmc_v6_0_wait_for_idle((void *)adev)) {
249 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
252 if (adev->mode_info.num_crtc) {
255 /* Lockout access through VGA aperture*/
256 tmp = RREG32(mmVGA_HDP_CONTROL);
257 tmp |= VGA_HDP_CONTROL__VGA_MEMORY_DISABLE_MASK;
258 WREG32(mmVGA_HDP_CONTROL, tmp);
260 /* disable VGA render */
261 tmp = RREG32(mmVGA_RENDER_CONTROL);
262 tmp &= ~VGA_VSTATUS_CNTL;
263 WREG32(mmVGA_RENDER_CONTROL, tmp);
265 /* Update configuration */
266 WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
267 adev->mc.vram_start >> 12);
268 WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
269 adev->mc.vram_end >> 12);
270 WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
271 adev->vram_scratch.gpu_addr >> 12);
272 WREG32(mmMC_VM_AGP_BASE, 0);
273 WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
274 WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
276 if (gmc_v6_0_wait_for_idle((void *)adev)) {
277 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
281 static int gmc_v6_0_mc_init(struct amdgpu_device *adev)
285 int chansize, numchan;
288 tmp = RREG32(mmMC_ARB_RAMCFG);
289 if (tmp & (1 << 11)) {
291 } else if (tmp & MC_ARB_RAMCFG__CHANSIZE_MASK) {
296 tmp = RREG32(mmMC_SHARED_CHMAP);
297 switch ((tmp & MC_SHARED_CHMAP__NOOFCHAN_MASK) >> MC_SHARED_CHMAP__NOOFCHAN__SHIFT) {
327 adev->mc.vram_width = numchan * chansize;
328 /* size in MB on si */
329 adev->mc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
330 adev->mc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
332 if (!(adev->flags & AMD_IS_APU)) {
333 r = amdgpu_device_resize_fb_bar(adev);
337 adev->mc.aper_base = pci_resource_start(adev->pdev, 0);
338 adev->mc.aper_size = pci_resource_len(adev->pdev, 0);
339 adev->mc.visible_vram_size = adev->mc.aper_size;
341 /* set the gart size */
342 if (amdgpu_gart_size == -1) {
343 switch (adev->asic_type) {
344 case CHIP_HAINAN: /* no MM engines */
346 adev->mc.gart_size = 256ULL << 20;
348 case CHIP_VERDE: /* UVD, VCE do not support GPUVM */
349 case CHIP_TAHITI: /* UVD, VCE do not support GPUVM */
350 case CHIP_PITCAIRN: /* UVD, VCE do not support GPUVM */
351 case CHIP_OLAND: /* UVD, VCE do not support GPUVM */
352 adev->mc.gart_size = 1024ULL << 20;
356 adev->mc.gart_size = (u64)amdgpu_gart_size << 20;
359 gmc_v6_0_vram_gtt_location(adev, &adev->mc);
364 static void gmc_v6_0_gart_flush_gpu_tlb(struct amdgpu_device *adev,
367 WREG32(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0);
369 WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
372 static int gmc_v6_0_gart_set_pte_pde(struct amdgpu_device *adev,
374 uint32_t gpu_page_idx,
378 void __iomem *ptr = (void *)cpu_pt_addr;
381 value = addr & 0xFFFFFFFFFFFFF000ULL;
383 writeq(value, ptr + (gpu_page_idx * 8));
388 static uint64_t gmc_v6_0_get_vm_pte_flags(struct amdgpu_device *adev,
391 uint64_t pte_flag = 0;
393 if (flags & AMDGPU_VM_PAGE_READABLE)
394 pte_flag |= AMDGPU_PTE_READABLE;
395 if (flags & AMDGPU_VM_PAGE_WRITEABLE)
396 pte_flag |= AMDGPU_PTE_WRITEABLE;
397 if (flags & AMDGPU_VM_PAGE_PRT)
398 pte_flag |= AMDGPU_PTE_PRT;
403 static uint64_t gmc_v6_0_get_vm_pde(struct amdgpu_device *adev, uint64_t addr)
405 BUG_ON(addr & 0xFFFFFF0000000FFFULL);
409 static void gmc_v6_0_set_fault_enable_default(struct amdgpu_device *adev,
414 tmp = RREG32(mmVM_CONTEXT1_CNTL);
415 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
416 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
417 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
418 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
419 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
420 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
421 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
422 VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
423 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
424 READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
425 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
426 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
427 WREG32(mmVM_CONTEXT1_CNTL, tmp);
431 + * gmc_v8_0_set_prt - set PRT VM fault
433 + * @adev: amdgpu_device pointer
434 + * @enable: enable/disable VM fault handling for PRT
436 static void gmc_v6_0_set_prt(struct amdgpu_device *adev, bool enable)
440 if (enable && !adev->mc.prt_warning) {
441 dev_warn(adev->dev, "Disabling VM faults because of PRT request!\n");
442 adev->mc.prt_warning = true;
445 tmp = RREG32(mmVM_PRT_CNTL);
446 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
447 CB_DISABLE_FAULT_ON_UNMAPPED_ACCESS,
449 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
450 TC_DISABLE_FAULT_ON_UNMAPPED_ACCESS,
452 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
453 L2_CACHE_STORE_INVALID_ENTRIES,
455 tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
456 L1_TLB_STORE_INVALID_ENTRIES,
458 WREG32(mmVM_PRT_CNTL, tmp);
461 uint32_t low = AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT;
462 uint32_t high = adev->vm_manager.max_pfn;
464 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, low);
465 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, low);
466 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, low);
467 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, low);
468 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, high);
469 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, high);
470 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, high);
471 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, high);
473 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, 0xfffffff);
474 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, 0xfffffff);
475 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, 0xfffffff);
476 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, 0xfffffff);
477 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, 0x0);
478 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, 0x0);
479 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, 0x0);
480 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, 0x0);
484 static int gmc_v6_0_gart_enable(struct amdgpu_device *adev)
489 if (adev->gart.robj == NULL) {
490 dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
494 /* Setup TLB control */
495 WREG32(mmMC_VM_MX_L1_TLB_CNTL,
497 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_TLB_MASK |
498 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_FRAGMENT_PROCESSING_MASK |
499 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
500 MC_VM_MX_L1_TLB_CNTL__ENABLE_ADVANCED_DRIVER_MODEL_MASK |
501 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
504 VM_L2_CNTL__ENABLE_L2_CACHE_MASK |
505 VM_L2_CNTL__ENABLE_L2_FRAGMENT_PROCESSING_MASK |
506 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
507 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
508 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
509 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
510 WREG32(mmVM_L2_CNTL2,
511 VM_L2_CNTL2__INVALIDATE_ALL_L1_TLBS_MASK |
512 VM_L2_CNTL2__INVALIDATE_L2_CACHE_MASK);
514 field = adev->vm_manager.fragment_size;
515 WREG32(mmVM_L2_CNTL3,
516 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
517 (field << VM_L2_CNTL3__BANK_SELECT__SHIFT) |
518 (field << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
520 WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->mc.gart_start >> 12);
521 WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->mc.gart_end >> 12);
522 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12);
523 WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
524 (u32)(adev->dummy_page.addr >> 12));
525 WREG32(mmVM_CONTEXT0_CNTL2, 0);
526 WREG32(mmVM_CONTEXT0_CNTL,
527 VM_CONTEXT0_CNTL__ENABLE_CONTEXT_MASK |
528 (0UL << VM_CONTEXT0_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
529 VM_CONTEXT0_CNTL__RANGE_PROTECTION_FAULT_ENABLE_DEFAULT_MASK);
535 /* empty context1-15 */
536 /* set vm size, must be a multiple of 4 */
537 WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
538 WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
539 /* Assign the pt base to something valid for now; the pts used for
540 * the VMs are determined by the application and setup and assigned
541 * on the fly in the vm part of radeon_gart.c
543 for (i = 1; i < 16; i++) {
545 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
546 adev->gart.table_addr >> 12);
548 WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
549 adev->gart.table_addr >> 12);
552 /* enable context1-15 */
553 WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
554 (u32)(adev->dummy_page.addr >> 12));
555 WREG32(mmVM_CONTEXT1_CNTL2, 4);
556 WREG32(mmVM_CONTEXT1_CNTL,
557 VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK |
558 (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
559 ((adev->vm_manager.block_size - 9)
560 << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT));
561 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
562 gmc_v6_0_set_fault_enable_default(adev, false);
564 gmc_v6_0_set_fault_enable_default(adev, true);
566 gmc_v6_0_gart_flush_gpu_tlb(adev, 0);
567 dev_info(adev->dev, "PCIE GART of %uM enabled (table at 0x%016llX).\n",
568 (unsigned)(adev->mc.gart_size >> 20),
569 (unsigned long long)adev->gart.table_addr);
570 adev->gart.ready = true;
574 static int gmc_v6_0_gart_init(struct amdgpu_device *adev)
578 if (adev->gart.robj) {
579 dev_warn(adev->dev, "gmc_v6_0 PCIE GART already initialized\n");
582 r = amdgpu_gart_init(adev);
585 adev->gart.table_size = adev->gart.num_gpu_pages * 8;
586 adev->gart.gart_pte_flags = 0;
587 return amdgpu_gart_table_vram_alloc(adev);
590 static void gmc_v6_0_gart_disable(struct amdgpu_device *adev)
594 for (i = 1; i < 16; ++i) {
597 reg = VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i ;
599 reg = VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (i - 8);
600 adev->vm_manager.saved_table_addr[i] = RREG32(reg);
603 /* Disable all tables */
604 WREG32(mmVM_CONTEXT0_CNTL, 0);
605 WREG32(mmVM_CONTEXT1_CNTL, 0);
606 /* Setup TLB control */
607 WREG32(mmMC_VM_MX_L1_TLB_CNTL,
608 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
609 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
612 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
613 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
614 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
615 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
616 WREG32(mmVM_L2_CNTL2, 0);
617 WREG32(mmVM_L2_CNTL3,
618 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
619 (0UL << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
622 static void gmc_v6_0_gart_fini(struct amdgpu_device *adev)
624 amdgpu_gart_table_vram_free(adev);
625 amdgpu_gart_fini(adev);
628 static void gmc_v6_0_vm_decode_fault(struct amdgpu_device *adev,
629 u32 status, u32 addr, u32 mc_client)
632 u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
633 u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
635 char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
636 (mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
638 mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
641 dev_err(adev->dev, "VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
642 protections, vmid, addr,
643 REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
645 "write" : "read", block, mc_client, mc_id);
649 static const u32 mc_cg_registers[] = {
661 static const u32 mc_cg_ls_en[] = {
662 MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK,
663 MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK,
664 MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK,
665 MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK,
666 ATC_MISC_CG__MEM_LS_ENABLE_MASK,
667 MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK,
668 MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK,
669 MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK,
670 VM_L2_CG__MEM_LS_ENABLE_MASK,
673 static const u32 mc_cg_en[] = {
674 MC_HUB_MISC_HUB_CG__ENABLE_MASK,
675 MC_HUB_MISC_SIP_CG__ENABLE_MASK,
676 MC_HUB_MISC_VM_CG__ENABLE_MASK,
677 MC_XPB_CLK_GAT__ENABLE_MASK,
678 ATC_MISC_CG__ENABLE_MASK,
679 MC_CITF_MISC_WR_CG__ENABLE_MASK,
680 MC_CITF_MISC_RD_CG__ENABLE_MASK,
681 MC_CITF_MISC_VM_CG__ENABLE_MASK,
682 VM_L2_CG__ENABLE_MASK,
685 static void gmc_v6_0_enable_mc_ls(struct amdgpu_device *adev,
691 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
692 orig = data = RREG32(mc_cg_registers[i]);
693 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_LS))
694 data |= mc_cg_ls_en[i];
696 data &= ~mc_cg_ls_en[i];
698 WREG32(mc_cg_registers[i], data);
702 static void gmc_v6_0_enable_mc_mgcg(struct amdgpu_device *adev,
708 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
709 orig = data = RREG32(mc_cg_registers[i]);
710 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_MGCG))
713 data &= ~mc_cg_en[i];
715 WREG32(mc_cg_registers[i], data);
719 static void gmc_v6_0_enable_bif_mgls(struct amdgpu_device *adev,
724 orig = data = RREG32_PCIE(ixPCIE_CNTL2);
726 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_BIF_LS)) {
727 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 1);
728 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 1);
729 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 1);
730 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 1);
732 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 0);
733 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 0);
734 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 0);
735 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 0);
739 WREG32_PCIE(ixPCIE_CNTL2, data);
742 static void gmc_v6_0_enable_hdp_mgcg(struct amdgpu_device *adev,
747 orig = data = RREG32(mmHDP_HOST_PATH_CNTL);
749 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_MGCG))
750 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 0);
752 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 1);
755 WREG32(mmHDP_HOST_PATH_CNTL, data);
758 static void gmc_v6_0_enable_hdp_ls(struct amdgpu_device *adev,
763 orig = data = RREG32(mmHDP_MEM_POWER_LS);
765 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_LS))
766 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 1);
768 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 0);
771 WREG32(mmHDP_MEM_POWER_LS, data);
775 static int gmc_v6_0_convert_vram_type(int mc_seq_vram_type)
777 switch (mc_seq_vram_type) {
778 case MC_SEQ_MISC0__MT__GDDR1:
779 return AMDGPU_VRAM_TYPE_GDDR1;
780 case MC_SEQ_MISC0__MT__DDR2:
781 return AMDGPU_VRAM_TYPE_DDR2;
782 case MC_SEQ_MISC0__MT__GDDR3:
783 return AMDGPU_VRAM_TYPE_GDDR3;
784 case MC_SEQ_MISC0__MT__GDDR4:
785 return AMDGPU_VRAM_TYPE_GDDR4;
786 case MC_SEQ_MISC0__MT__GDDR5:
787 return AMDGPU_VRAM_TYPE_GDDR5;
788 case MC_SEQ_MISC0__MT__DDR3:
789 return AMDGPU_VRAM_TYPE_DDR3;
791 return AMDGPU_VRAM_TYPE_UNKNOWN;
795 static int gmc_v6_0_early_init(void *handle)
797 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
799 gmc_v6_0_set_gart_funcs(adev);
800 gmc_v6_0_set_irq_funcs(adev);
805 static int gmc_v6_0_late_init(void *handle)
807 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
809 if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
810 return amdgpu_irq_get(adev, &adev->mc.vm_fault, 0);
815 static int gmc_v6_0_sw_init(void *handle)
819 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
821 if (adev->flags & AMD_IS_APU) {
822 adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
824 u32 tmp = RREG32(mmMC_SEQ_MISC0);
825 tmp &= MC_SEQ_MISC0__MT__MASK;
826 adev->mc.vram_type = gmc_v6_0_convert_vram_type(tmp);
829 r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_LEGACY, 146, &adev->mc.vm_fault);
833 r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_LEGACY, 147, &adev->mc.vm_fault);
837 amdgpu_vm_adjust_size(adev, 64, 9);
839 adev->mc.mc_mask = 0xffffffffffULL;
841 adev->mc.stolen_size = 256 * 1024;
843 adev->need_dma32 = false;
844 dma_bits = adev->need_dma32 ? 32 : 40;
845 r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
847 adev->need_dma32 = true;
849 dev_warn(adev->dev, "amdgpu: No suitable DMA available.\n");
851 r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
853 pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
854 dev_warn(adev->dev, "amdgpu: No coherent DMA available.\n");
857 r = gmc_v6_0_init_microcode(adev);
859 dev_err(adev->dev, "Failed to load mc firmware!\n");
863 r = gmc_v6_0_mc_init(adev);
867 r = amdgpu_bo_init(adev);
871 r = gmc_v6_0_gart_init(adev);
877 * VMID 0 is reserved for System
878 * amdgpu graphics/compute will use VMIDs 1-7
879 * amdkfd will use VMIDs 8-15
881 adev->vm_manager.id_mgr[0].num_ids = AMDGPU_NUM_OF_VMIDS;
882 adev->vm_manager.num_level = 1;
883 amdgpu_vm_manager_init(adev);
885 /* base offset of vram pages */
886 if (adev->flags & AMD_IS_APU) {
887 u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
890 adev->vm_manager.vram_base_offset = tmp;
892 adev->vm_manager.vram_base_offset = 0;
898 static int gmc_v6_0_sw_fini(void *handle)
900 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
902 amdgpu_gem_force_release(adev);
903 amdgpu_vm_manager_fini(adev);
904 gmc_v6_0_gart_fini(adev);
905 amdgpu_bo_fini(adev);
906 release_firmware(adev->mc.fw);
912 static int gmc_v6_0_hw_init(void *handle)
915 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
917 gmc_v6_0_mc_program(adev);
919 if (!(adev->flags & AMD_IS_APU)) {
920 r = gmc_v6_0_mc_load_microcode(adev);
922 dev_err(adev->dev, "Failed to load MC firmware!\n");
927 r = gmc_v6_0_gart_enable(adev);
934 static int gmc_v6_0_hw_fini(void *handle)
936 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
938 amdgpu_irq_put(adev, &adev->mc.vm_fault, 0);
939 gmc_v6_0_gart_disable(adev);
944 static int gmc_v6_0_suspend(void *handle)
946 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
948 gmc_v6_0_hw_fini(adev);
953 static int gmc_v6_0_resume(void *handle)
956 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
958 r = gmc_v6_0_hw_init(adev);
962 amdgpu_vm_reset_all_ids(adev);
967 static bool gmc_v6_0_is_idle(void *handle)
969 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
970 u32 tmp = RREG32(mmSRBM_STATUS);
972 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
973 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
979 static int gmc_v6_0_wait_for_idle(void *handle)
982 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
984 for (i = 0; i < adev->usec_timeout; i++) {
985 if (gmc_v6_0_is_idle(handle))
993 static int gmc_v6_0_soft_reset(void *handle)
995 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
996 u32 srbm_soft_reset = 0;
997 u32 tmp = RREG32(mmSRBM_STATUS);
999 if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1000 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1001 SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1003 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1004 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1005 if (!(adev->flags & AMD_IS_APU))
1006 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1007 SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1010 if (srbm_soft_reset) {
1011 gmc_v6_0_mc_stop(adev);
1012 if (gmc_v6_0_wait_for_idle(adev)) {
1013 dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1017 tmp = RREG32(mmSRBM_SOFT_RESET);
1018 tmp |= srbm_soft_reset;
1019 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1020 WREG32(mmSRBM_SOFT_RESET, tmp);
1021 tmp = RREG32(mmSRBM_SOFT_RESET);
1025 tmp &= ~srbm_soft_reset;
1026 WREG32(mmSRBM_SOFT_RESET, tmp);
1027 tmp = RREG32(mmSRBM_SOFT_RESET);
1031 gmc_v6_0_mc_resume(adev);
1038 static int gmc_v6_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1039 struct amdgpu_irq_src *src,
1041 enum amdgpu_interrupt_state state)
1044 u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1045 VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1046 VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1047 VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1048 VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1049 VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1052 case AMDGPU_IRQ_STATE_DISABLE:
1053 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1055 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1056 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1058 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1060 case AMDGPU_IRQ_STATE_ENABLE:
1061 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1063 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1064 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1066 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1075 static int gmc_v6_0_process_interrupt(struct amdgpu_device *adev,
1076 struct amdgpu_irq_src *source,
1077 struct amdgpu_iv_entry *entry)
1081 addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1082 status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1083 WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1085 if (!addr && !status)
1088 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1089 gmc_v6_0_set_fault_enable_default(adev, false);
1091 if (printk_ratelimit()) {
1092 dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1093 entry->src_id, entry->src_data[0]);
1094 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
1096 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1098 gmc_v6_0_vm_decode_fault(adev, status, addr, 0);
1104 static int gmc_v6_0_set_clockgating_state(void *handle,
1105 enum amd_clockgating_state state)
1110 static int gmc_v6_0_set_powergating_state(void *handle,
1111 enum amd_powergating_state state)
1116 static const struct amd_ip_funcs gmc_v6_0_ip_funcs = {
1118 .early_init = gmc_v6_0_early_init,
1119 .late_init = gmc_v6_0_late_init,
1120 .sw_init = gmc_v6_0_sw_init,
1121 .sw_fini = gmc_v6_0_sw_fini,
1122 .hw_init = gmc_v6_0_hw_init,
1123 .hw_fini = gmc_v6_0_hw_fini,
1124 .suspend = gmc_v6_0_suspend,
1125 .resume = gmc_v6_0_resume,
1126 .is_idle = gmc_v6_0_is_idle,
1127 .wait_for_idle = gmc_v6_0_wait_for_idle,
1128 .soft_reset = gmc_v6_0_soft_reset,
1129 .set_clockgating_state = gmc_v6_0_set_clockgating_state,
1130 .set_powergating_state = gmc_v6_0_set_powergating_state,
1133 static const struct amdgpu_gart_funcs gmc_v6_0_gart_funcs = {
1134 .flush_gpu_tlb = gmc_v6_0_gart_flush_gpu_tlb,
1135 .set_pte_pde = gmc_v6_0_gart_set_pte_pde,
1136 .set_prt = gmc_v6_0_set_prt,
1137 .get_vm_pde = gmc_v6_0_get_vm_pde,
1138 .get_vm_pte_flags = gmc_v6_0_get_vm_pte_flags
1141 static const struct amdgpu_irq_src_funcs gmc_v6_0_irq_funcs = {
1142 .set = gmc_v6_0_vm_fault_interrupt_state,
1143 .process = gmc_v6_0_process_interrupt,
1146 static void gmc_v6_0_set_gart_funcs(struct amdgpu_device *adev)
1148 if (adev->gart.gart_funcs == NULL)
1149 adev->gart.gart_funcs = &gmc_v6_0_gart_funcs;
1152 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev)
1154 adev->mc.vm_fault.num_types = 1;
1155 adev->mc.vm_fault.funcs = &gmc_v6_0_irq_funcs;
1158 const struct amdgpu_ip_block_version gmc_v6_0_ip_block =
1160 .type = AMD_IP_BLOCK_TYPE_GMC,
1164 .funcs = &gmc_v6_0_ip_funcs,