2 * Copyright 2014 Advanced Micro Devices, Inc.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
23 #include <linux/firmware.h>
27 #include "amdgpu_ucode.h"
29 #include "bif/bif_3_0_d.h"
30 #include "bif/bif_3_0_sh_mask.h"
31 #include "oss/oss_1_0_d.h"
32 #include "oss/oss_1_0_sh_mask.h"
33 #include "gmc/gmc_6_0_d.h"
34 #include "gmc/gmc_6_0_sh_mask.h"
35 #include "dce/dce_6_0_d.h"
36 #include "dce/dce_6_0_sh_mask.h"
39 static void gmc_v6_0_set_gart_funcs(struct amdgpu_device *adev);
40 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev);
41 static int gmc_v6_0_wait_for_idle(void *handle);
43 MODULE_FIRMWARE("radeon/tahiti_mc.bin");
44 MODULE_FIRMWARE("radeon/pitcairn_mc.bin");
45 MODULE_FIRMWARE("radeon/verde_mc.bin");
46 MODULE_FIRMWARE("radeon/oland_mc.bin");
47 MODULE_FIRMWARE("radeon/si58_mc.bin");
49 #define MC_SEQ_MISC0__MT__MASK 0xf0000000
50 #define MC_SEQ_MISC0__MT__GDDR1 0x10000000
51 #define MC_SEQ_MISC0__MT__DDR2 0x20000000
52 #define MC_SEQ_MISC0__MT__GDDR3 0x30000000
53 #define MC_SEQ_MISC0__MT__GDDR4 0x40000000
54 #define MC_SEQ_MISC0__MT__GDDR5 0x50000000
55 #define MC_SEQ_MISC0__MT__HBM 0x60000000
56 #define MC_SEQ_MISC0__MT__DDR3 0xB0000000
59 static const u32 crtc_offsets[6] =
61 SI_CRTC0_REGISTER_OFFSET,
62 SI_CRTC1_REGISTER_OFFSET,
63 SI_CRTC2_REGISTER_OFFSET,
64 SI_CRTC3_REGISTER_OFFSET,
65 SI_CRTC4_REGISTER_OFFSET,
66 SI_CRTC5_REGISTER_OFFSET
69 static void gmc_v6_0_mc_stop(struct amdgpu_device *adev,
70 struct amdgpu_mode_mc_save *save)
74 if (adev->mode_info.num_crtc)
75 amdgpu_display_stop_mc_access(adev, save);
77 gmc_v6_0_wait_for_idle((void *)adev);
79 blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
80 if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
81 /* Block CPU access */
82 WREG32(mmBIF_FB_EN, 0);
84 blackout = REG_SET_FIELD(blackout,
85 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
86 WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout | 1);
88 /* wait for the MC to settle */
93 static void gmc_v6_0_mc_resume(struct amdgpu_device *adev,
94 struct amdgpu_mode_mc_save *save)
98 /* unblackout the MC */
99 tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
100 tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
101 WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
102 /* allow CPU access */
103 tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
104 tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
105 WREG32(mmBIF_FB_EN, tmp);
107 if (adev->mode_info.num_crtc)
108 amdgpu_display_resume_mc_access(adev, save);
112 static int gmc_v6_0_init_microcode(struct amdgpu_device *adev)
114 const char *chip_name;
117 bool is_58_fw = false;
121 switch (adev->asic_type) {
123 chip_name = "tahiti";
126 chip_name = "pitcairn";
135 chip_name = "hainan";
140 /* this memory configuration requires special firmware */
141 if (((RREG32(mmMC_SEQ_MISC0) & 0xff000000) >> 24) == 0x58)
145 snprintf(fw_name, sizeof(fw_name), "radeon/si58_mc.bin");
147 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
148 err = request_firmware(&adev->mc.fw, fw_name, adev->dev);
152 err = amdgpu_ucode_validate(adev->mc.fw);
157 "si_mc: Failed to load firmware \"%s\"\n",
159 release_firmware(adev->mc.fw);
165 static int gmc_v6_0_mc_load_microcode(struct amdgpu_device *adev)
167 const __le32 *new_fw_data = NULL;
169 const __le32 *new_io_mc_regs = NULL;
170 int i, regs_size, ucode_size;
171 const struct mc_firmware_header_v1_0 *hdr;
176 hdr = (const struct mc_firmware_header_v1_0 *)adev->mc.fw->data;
178 amdgpu_ucode_print_mc_hdr(&hdr->header);
180 adev->mc.fw_version = le32_to_cpu(hdr->header.ucode_version);
181 regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
182 new_io_mc_regs = (const __le32 *)
183 (adev->mc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
184 ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
185 new_fw_data = (const __le32 *)
186 (adev->mc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
188 running = RREG32(mmMC_SEQ_SUP_CNTL) & MC_SEQ_SUP_CNTL__RUN_MASK;
192 /* reset the engine and set to writable */
193 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
194 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
196 /* load mc io regs */
197 for (i = 0; i < regs_size; i++) {
198 WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(new_io_mc_regs++));
199 WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(new_io_mc_regs++));
201 /* load the MC ucode */
202 for (i = 0; i < ucode_size; i++) {
203 WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(new_fw_data++));
206 /* put the engine back into the active state */
207 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
208 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
209 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
211 /* wait for training to complete */
212 for (i = 0; i < adev->usec_timeout; i++) {
213 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D0_MASK)
217 for (i = 0; i < adev->usec_timeout; i++) {
218 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D1_MASK)
228 static void gmc_v6_0_vram_gtt_location(struct amdgpu_device *adev,
229 struct amdgpu_mc *mc)
231 if (mc->mc_vram_size > 0xFFC0000000ULL) {
232 dev_warn(adev->dev, "limiting VRAM\n");
233 mc->real_vram_size = 0xFFC0000000ULL;
234 mc->mc_vram_size = 0xFFC0000000ULL;
236 amdgpu_vram_location(adev, &adev->mc, 0);
237 adev->mc.gtt_base_align = 0;
238 amdgpu_gtt_location(adev, mc);
241 static void gmc_v6_0_mc_program(struct amdgpu_device *adev)
243 struct amdgpu_mode_mc_save save;
248 for (i = 0, j = 0; i < 32; i++, j += 0x6) {
249 WREG32((0xb05 + j), 0x00000000);
250 WREG32((0xb06 + j), 0x00000000);
251 WREG32((0xb07 + j), 0x00000000);
252 WREG32((0xb08 + j), 0x00000000);
253 WREG32((0xb09 + j), 0x00000000);
255 WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
257 if (adev->mode_info.num_crtc)
258 amdgpu_display_set_vga_render_state(adev, false);
260 gmc_v6_0_mc_stop(adev, &save);
262 if (gmc_v6_0_wait_for_idle((void *)adev)) {
263 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
266 WREG32(mmVGA_HDP_CONTROL, VGA_HDP_CONTROL__VGA_MEMORY_DISABLE_MASK);
267 /* Update configuration */
268 WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
269 adev->mc.vram_start >> 12);
270 WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
271 adev->mc.vram_end >> 12);
272 WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
273 adev->vram_scratch.gpu_addr >> 12);
274 tmp = ((adev->mc.vram_end >> 24) & 0xFFFF) << 16;
275 tmp |= ((adev->mc.vram_start >> 24) & 0xFFFF);
276 WREG32(mmMC_VM_FB_LOCATION, tmp);
277 /* XXX double check these! */
278 WREG32(mmHDP_NONSURFACE_BASE, (adev->mc.vram_start >> 8));
279 WREG32(mmHDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
280 WREG32(mmHDP_NONSURFACE_SIZE, 0x3FFFFFFF);
281 WREG32(mmMC_VM_AGP_BASE, 0);
282 WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
283 WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
285 if (gmc_v6_0_wait_for_idle((void *)adev)) {
286 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
288 gmc_v6_0_mc_resume(adev, &save);
291 static int gmc_v6_0_mc_init(struct amdgpu_device *adev)
295 int chansize, numchan;
297 tmp = RREG32(mmMC_ARB_RAMCFG);
298 if (tmp & (1 << 11)) {
300 } else if (tmp & MC_ARB_RAMCFG__CHANSIZE_MASK) {
305 tmp = RREG32(mmMC_SHARED_CHMAP);
306 switch ((tmp & MC_SHARED_CHMAP__NOOFCHAN_MASK) >> MC_SHARED_CHMAP__NOOFCHAN__SHIFT) {
336 adev->mc.vram_width = numchan * chansize;
337 /* Could aper size report 0 ? */
338 adev->mc.aper_base = pci_resource_start(adev->pdev, 0);
339 adev->mc.aper_size = pci_resource_len(adev->pdev, 0);
340 /* size in MB on si */
341 adev->mc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
342 adev->mc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
343 adev->mc.visible_vram_size = adev->mc.aper_size;
345 /* unless the user had overridden it, set the gart
346 * size equal to the 1024 or vram, whichever is larger.
348 if (amdgpu_gart_size == -1)
349 adev->mc.gtt_size = max((1024ULL << 20), adev->mc.mc_vram_size);
351 adev->mc.gtt_size = (uint64_t)amdgpu_gart_size << 20;
353 gmc_v6_0_vram_gtt_location(adev, &adev->mc);
358 static void gmc_v6_0_gart_flush_gpu_tlb(struct amdgpu_device *adev,
361 WREG32(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0);
363 WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
366 static int gmc_v6_0_gart_set_pte_pde(struct amdgpu_device *adev,
368 uint32_t gpu_page_idx,
372 void __iomem *ptr = (void *)cpu_pt_addr;
375 value = addr & 0xFFFFFFFFFFFFF000ULL;
377 writeq(value, ptr + (gpu_page_idx * 8));
382 static void gmc_v6_0_set_fault_enable_default(struct amdgpu_device *adev,
387 tmp = RREG32(mmVM_CONTEXT1_CNTL);
388 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
389 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
390 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
391 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
392 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
393 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
394 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
395 VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
396 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
397 READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
398 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
399 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
400 WREG32(mmVM_CONTEXT1_CNTL, tmp);
403 static int gmc_v6_0_gart_enable(struct amdgpu_device *adev)
407 if (adev->gart.robj == NULL) {
408 dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
411 r = amdgpu_gart_table_vram_pin(adev);
414 /* Setup TLB control */
415 WREG32(mmMC_VM_MX_L1_TLB_CNTL,
417 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_TLB_MASK |
418 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_FRAGMENT_PROCESSING_MASK |
419 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
420 MC_VM_MX_L1_TLB_CNTL__ENABLE_ADVANCED_DRIVER_MODEL_MASK |
421 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
424 VM_L2_CNTL__ENABLE_L2_CACHE_MASK |
425 VM_L2_CNTL__ENABLE_L2_FRAGMENT_PROCESSING_MASK |
426 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
427 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
428 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
429 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
430 WREG32(mmVM_L2_CNTL2,
431 VM_L2_CNTL2__INVALIDATE_ALL_L1_TLBS_MASK |
432 VM_L2_CNTL2__INVALIDATE_L2_CACHE_MASK);
433 WREG32(mmVM_L2_CNTL3,
434 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
435 (4UL << VM_L2_CNTL3__BANK_SELECT__SHIFT) |
436 (4UL << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
438 WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->mc.gtt_start >> 12);
439 WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->mc.gtt_end >> 12);
440 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12);
441 WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
442 (u32)(adev->dummy_page.addr >> 12));
443 WREG32(mmVM_CONTEXT0_CNTL2, 0);
444 WREG32(mmVM_CONTEXT0_CNTL,
445 VM_CONTEXT0_CNTL__ENABLE_CONTEXT_MASK |
446 (0UL << VM_CONTEXT0_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
447 VM_CONTEXT0_CNTL__RANGE_PROTECTION_FAULT_ENABLE_DEFAULT_MASK);
453 /* empty context1-15 */
454 /* set vm size, must be a multiple of 4 */
455 WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
456 WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
457 /* Assign the pt base to something valid for now; the pts used for
458 * the VMs are determined by the application and setup and assigned
459 * on the fly in the vm part of radeon_gart.c
461 for (i = 1; i < 16; i++) {
463 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
464 adev->gart.table_addr >> 12);
466 WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
467 adev->gart.table_addr >> 12);
470 /* enable context1-15 */
471 WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
472 (u32)(adev->dummy_page.addr >> 12));
473 WREG32(mmVM_CONTEXT1_CNTL2, 4);
474 WREG32(mmVM_CONTEXT1_CNTL,
475 VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK |
476 (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
477 ((amdgpu_vm_block_size - 9) << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT));
478 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
479 gmc_v6_0_set_fault_enable_default(adev, false);
481 gmc_v6_0_set_fault_enable_default(adev, true);
483 gmc_v6_0_gart_flush_gpu_tlb(adev, 0);
484 dev_info(adev->dev, "PCIE GART of %uM enabled (table at 0x%016llX).\n",
485 (unsigned)(adev->mc.gtt_size >> 20),
486 (unsigned long long)adev->gart.table_addr);
487 adev->gart.ready = true;
491 static int gmc_v6_0_gart_init(struct amdgpu_device *adev)
495 if (adev->gart.robj) {
496 dev_warn(adev->dev, "gmc_v6_0 PCIE GART already initialized\n");
499 r = amdgpu_gart_init(adev);
502 adev->gart.table_size = adev->gart.num_gpu_pages * 8;
503 return amdgpu_gart_table_vram_alloc(adev);
506 static void gmc_v6_0_gart_disable(struct amdgpu_device *adev)
510 for (i = 1; i < 16; ++i) {
513 reg = VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i ;
515 reg = VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (i - 8);
516 adev->vm_manager.saved_table_addr[i] = RREG32(reg);
519 /* Disable all tables */
520 WREG32(mmVM_CONTEXT0_CNTL, 0);
521 WREG32(mmVM_CONTEXT1_CNTL, 0);
522 /* Setup TLB control */
523 WREG32(mmMC_VM_MX_L1_TLB_CNTL,
524 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
525 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
528 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
529 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
530 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
531 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
532 WREG32(mmVM_L2_CNTL2, 0);
533 WREG32(mmVM_L2_CNTL3,
534 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
535 (0UL << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
536 amdgpu_gart_table_vram_unpin(adev);
539 static void gmc_v6_0_gart_fini(struct amdgpu_device *adev)
541 amdgpu_gart_table_vram_free(adev);
542 amdgpu_gart_fini(adev);
545 static int gmc_v6_0_vm_init(struct amdgpu_device *adev)
549 * VMID 0 is reserved for System
550 * amdgpu graphics/compute will use VMIDs 1-7
551 * amdkfd will use VMIDs 8-15
553 adev->vm_manager.num_ids = AMDGPU_NUM_OF_VMIDS;
554 amdgpu_vm_manager_init(adev);
556 /* base offset of vram pages */
557 if (adev->flags & AMD_IS_APU) {
558 u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
560 adev->vm_manager.vram_base_offset = tmp;
562 adev->vm_manager.vram_base_offset = 0;
567 static void gmc_v6_0_vm_fini(struct amdgpu_device *adev)
571 static void gmc_v6_0_vm_decode_fault(struct amdgpu_device *adev,
572 u32 status, u32 addr, u32 mc_client)
575 u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
576 u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
578 char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
579 (mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
581 mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
584 dev_err(adev->dev, "VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
585 protections, vmid, addr,
586 REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
588 "write" : "read", block, mc_client, mc_id);
592 static const u32 mc_cg_registers[] = {
604 static const u32 mc_cg_ls_en[] = {
605 MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK,
606 MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK,
607 MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK,
608 MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK,
609 ATC_MISC_CG__MEM_LS_ENABLE_MASK,
610 MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK,
611 MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK,
612 MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK,
613 VM_L2_CG__MEM_LS_ENABLE_MASK,
616 static const u32 mc_cg_en[] = {
617 MC_HUB_MISC_HUB_CG__ENABLE_MASK,
618 MC_HUB_MISC_SIP_CG__ENABLE_MASK,
619 MC_HUB_MISC_VM_CG__ENABLE_MASK,
620 MC_XPB_CLK_GAT__ENABLE_MASK,
621 ATC_MISC_CG__ENABLE_MASK,
622 MC_CITF_MISC_WR_CG__ENABLE_MASK,
623 MC_CITF_MISC_RD_CG__ENABLE_MASK,
624 MC_CITF_MISC_VM_CG__ENABLE_MASK,
625 VM_L2_CG__ENABLE_MASK,
628 static void gmc_v6_0_enable_mc_ls(struct amdgpu_device *adev,
634 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
635 orig = data = RREG32(mc_cg_registers[i]);
636 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_LS))
637 data |= mc_cg_ls_en[i];
639 data &= ~mc_cg_ls_en[i];
641 WREG32(mc_cg_registers[i], data);
645 static void gmc_v6_0_enable_mc_mgcg(struct amdgpu_device *adev,
651 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
652 orig = data = RREG32(mc_cg_registers[i]);
653 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_MGCG))
656 data &= ~mc_cg_en[i];
658 WREG32(mc_cg_registers[i], data);
662 static void gmc_v6_0_enable_bif_mgls(struct amdgpu_device *adev,
667 orig = data = RREG32_PCIE(ixPCIE_CNTL2);
669 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_BIF_LS)) {
670 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 1);
671 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 1);
672 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 1);
673 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 1);
675 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 0);
676 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 0);
677 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 0);
678 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 0);
682 WREG32_PCIE(ixPCIE_CNTL2, data);
685 static void gmc_v6_0_enable_hdp_mgcg(struct amdgpu_device *adev,
690 orig = data = RREG32(mmHDP_HOST_PATH_CNTL);
692 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_MGCG))
693 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 0);
695 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 1);
698 WREG32(mmHDP_HOST_PATH_CNTL, data);
701 static void gmc_v6_0_enable_hdp_ls(struct amdgpu_device *adev,
706 orig = data = RREG32(mmHDP_MEM_POWER_LS);
708 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_LS))
709 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 1);
711 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 0);
714 WREG32(mmHDP_MEM_POWER_LS, data);
718 static int gmc_v6_0_convert_vram_type(int mc_seq_vram_type)
720 switch (mc_seq_vram_type) {
721 case MC_SEQ_MISC0__MT__GDDR1:
722 return AMDGPU_VRAM_TYPE_GDDR1;
723 case MC_SEQ_MISC0__MT__DDR2:
724 return AMDGPU_VRAM_TYPE_DDR2;
725 case MC_SEQ_MISC0__MT__GDDR3:
726 return AMDGPU_VRAM_TYPE_GDDR3;
727 case MC_SEQ_MISC0__MT__GDDR4:
728 return AMDGPU_VRAM_TYPE_GDDR4;
729 case MC_SEQ_MISC0__MT__GDDR5:
730 return AMDGPU_VRAM_TYPE_GDDR5;
731 case MC_SEQ_MISC0__MT__DDR3:
732 return AMDGPU_VRAM_TYPE_DDR3;
734 return AMDGPU_VRAM_TYPE_UNKNOWN;
738 static int gmc_v6_0_early_init(void *handle)
740 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
742 gmc_v6_0_set_gart_funcs(adev);
743 gmc_v6_0_set_irq_funcs(adev);
745 if (adev->flags & AMD_IS_APU) {
746 adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
748 u32 tmp = RREG32(mmMC_SEQ_MISC0);
749 tmp &= MC_SEQ_MISC0__MT__MASK;
750 adev->mc.vram_type = gmc_v6_0_convert_vram_type(tmp);
756 static int gmc_v6_0_late_init(void *handle)
758 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
760 if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
761 return amdgpu_irq_get(adev, &adev->mc.vm_fault, 0);
766 static int gmc_v6_0_sw_init(void *handle)
770 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
772 r = amdgpu_irq_add_id(adev, 146, &adev->mc.vm_fault);
776 r = amdgpu_irq_add_id(adev, 147, &adev->mc.vm_fault);
780 adev->vm_manager.max_pfn = amdgpu_vm_size << 18;
782 adev->mc.mc_mask = 0xffffffffffULL;
784 adev->need_dma32 = false;
785 dma_bits = adev->need_dma32 ? 32 : 40;
786 r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
788 adev->need_dma32 = true;
790 dev_warn(adev->dev, "amdgpu: No suitable DMA available.\n");
792 r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
794 pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
795 dev_warn(adev->dev, "amdgpu: No coherent DMA available.\n");
798 r = gmc_v6_0_init_microcode(adev);
800 dev_err(adev->dev, "Failed to load mc firmware!\n");
804 r = gmc_v6_0_mc_init(adev);
808 r = amdgpu_bo_init(adev);
812 r = gmc_v6_0_gart_init(adev);
816 if (!adev->vm_manager.enabled) {
817 r = gmc_v6_0_vm_init(adev);
819 dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
822 adev->vm_manager.enabled = true;
828 static int gmc_v6_0_sw_fini(void *handle)
830 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
832 if (adev->vm_manager.enabled) {
833 gmc_v6_0_vm_fini(adev);
834 adev->vm_manager.enabled = false;
836 gmc_v6_0_gart_fini(adev);
837 amdgpu_gem_force_release(adev);
838 amdgpu_bo_fini(adev);
843 static int gmc_v6_0_hw_init(void *handle)
846 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
848 gmc_v6_0_mc_program(adev);
850 if (!(adev->flags & AMD_IS_APU)) {
851 r = gmc_v6_0_mc_load_microcode(adev);
853 dev_err(adev->dev, "Failed to load MC firmware!\n");
858 r = gmc_v6_0_gart_enable(adev);
865 static int gmc_v6_0_hw_fini(void *handle)
867 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
869 amdgpu_irq_put(adev, &adev->mc.vm_fault, 0);
870 gmc_v6_0_gart_disable(adev);
875 static int gmc_v6_0_suspend(void *handle)
877 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
879 if (adev->vm_manager.enabled) {
880 gmc_v6_0_vm_fini(adev);
881 adev->vm_manager.enabled = false;
883 gmc_v6_0_hw_fini(adev);
888 static int gmc_v6_0_resume(void *handle)
891 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
893 r = gmc_v6_0_hw_init(adev);
897 if (!adev->vm_manager.enabled) {
898 r = gmc_v6_0_vm_init(adev);
900 dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
903 adev->vm_manager.enabled = true;
909 static bool gmc_v6_0_is_idle(void *handle)
911 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
912 u32 tmp = RREG32(mmSRBM_STATUS);
914 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
915 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
921 static int gmc_v6_0_wait_for_idle(void *handle)
925 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
927 for (i = 0; i < adev->usec_timeout; i++) {
928 tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
929 SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
930 SRBM_STATUS__MCC_BUSY_MASK |
931 SRBM_STATUS__MCD_BUSY_MASK |
932 SRBM_STATUS__VMC_BUSY_MASK);
941 static int gmc_v6_0_soft_reset(void *handle)
943 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
944 struct amdgpu_mode_mc_save save;
945 u32 srbm_soft_reset = 0;
946 u32 tmp = RREG32(mmSRBM_STATUS);
948 if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
949 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
950 SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
952 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
953 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
954 if (!(adev->flags & AMD_IS_APU))
955 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
956 SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
959 if (srbm_soft_reset) {
960 gmc_v6_0_mc_stop(adev, &save);
961 if (gmc_v6_0_wait_for_idle(adev)) {
962 dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
966 tmp = RREG32(mmSRBM_SOFT_RESET);
967 tmp |= srbm_soft_reset;
968 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
969 WREG32(mmSRBM_SOFT_RESET, tmp);
970 tmp = RREG32(mmSRBM_SOFT_RESET);
974 tmp &= ~srbm_soft_reset;
975 WREG32(mmSRBM_SOFT_RESET, tmp);
976 tmp = RREG32(mmSRBM_SOFT_RESET);
980 gmc_v6_0_mc_resume(adev, &save);
987 static int gmc_v6_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
988 struct amdgpu_irq_src *src,
990 enum amdgpu_interrupt_state state)
993 u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
994 VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
995 VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
996 VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
997 VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
998 VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1001 case AMDGPU_IRQ_STATE_DISABLE:
1002 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1004 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1005 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1007 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1009 case AMDGPU_IRQ_STATE_ENABLE:
1010 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1012 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1013 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1015 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1024 static int gmc_v6_0_process_interrupt(struct amdgpu_device *adev,
1025 struct amdgpu_irq_src *source,
1026 struct amdgpu_iv_entry *entry)
1030 addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1031 status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1032 WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1034 if (!addr && !status)
1037 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1038 gmc_v6_0_set_fault_enable_default(adev, false);
1040 if (printk_ratelimit()) {
1041 dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1042 entry->src_id, entry->src_data);
1043 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
1045 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1047 gmc_v6_0_vm_decode_fault(adev, status, addr, 0);
1053 static int gmc_v6_0_set_clockgating_state(void *handle,
1054 enum amd_clockgating_state state)
1059 static int gmc_v6_0_set_powergating_state(void *handle,
1060 enum amd_powergating_state state)
1065 static const struct amd_ip_funcs gmc_v6_0_ip_funcs = {
1067 .early_init = gmc_v6_0_early_init,
1068 .late_init = gmc_v6_0_late_init,
1069 .sw_init = gmc_v6_0_sw_init,
1070 .sw_fini = gmc_v6_0_sw_fini,
1071 .hw_init = gmc_v6_0_hw_init,
1072 .hw_fini = gmc_v6_0_hw_fini,
1073 .suspend = gmc_v6_0_suspend,
1074 .resume = gmc_v6_0_resume,
1075 .is_idle = gmc_v6_0_is_idle,
1076 .wait_for_idle = gmc_v6_0_wait_for_idle,
1077 .soft_reset = gmc_v6_0_soft_reset,
1078 .set_clockgating_state = gmc_v6_0_set_clockgating_state,
1079 .set_powergating_state = gmc_v6_0_set_powergating_state,
1082 static const struct amdgpu_gart_funcs gmc_v6_0_gart_funcs = {
1083 .flush_gpu_tlb = gmc_v6_0_gart_flush_gpu_tlb,
1084 .set_pte_pde = gmc_v6_0_gart_set_pte_pde,
1087 static const struct amdgpu_irq_src_funcs gmc_v6_0_irq_funcs = {
1088 .set = gmc_v6_0_vm_fault_interrupt_state,
1089 .process = gmc_v6_0_process_interrupt,
1092 static void gmc_v6_0_set_gart_funcs(struct amdgpu_device *adev)
1094 if (adev->gart.gart_funcs == NULL)
1095 adev->gart.gart_funcs = &gmc_v6_0_gart_funcs;
1098 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev)
1100 adev->mc.vm_fault.num_types = 1;
1101 adev->mc.vm_fault.funcs = &gmc_v6_0_irq_funcs;
1104 const struct amdgpu_ip_block_version gmc_v6_0_ip_block =
1106 .type = AMD_IP_BLOCK_TYPE_GMC,
1110 .funcs = &gmc_v6_0_ip_funcs,