3 * Copyright 2014 Advanced Micro Devices, Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
19 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
21 * OTHER DEALINGS IN THE SOFTWARE.
24 #include <linux/firmware.h>
28 #include "amdgpu_ucode.h"
31 static void gmc_v6_0_set_gart_funcs(struct amdgpu_device *adev);
32 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev);
33 static int gmc_v6_0_wait_for_idle(void *handle);
35 MODULE_FIRMWARE("radeon/tahiti_mc.bin");
36 MODULE_FIRMWARE("radeon/pitcairn_mc.bin");
37 MODULE_FIRMWARE("radeon/verde_mc.bin");
38 MODULE_FIRMWARE("radeon/oland_mc.bin");
40 static const u32 crtc_offsets[6] =
42 SI_CRTC0_REGISTER_OFFSET,
43 SI_CRTC1_REGISTER_OFFSET,
44 SI_CRTC2_REGISTER_OFFSET,
45 SI_CRTC3_REGISTER_OFFSET,
46 SI_CRTC4_REGISTER_OFFSET,
47 SI_CRTC5_REGISTER_OFFSET
50 static void gmc_v6_0_mc_stop(struct amdgpu_device *adev,
51 struct amdgpu_mode_mc_save *save)
55 if (adev->mode_info.num_crtc)
56 amdgpu_display_stop_mc_access(adev, save);
58 gmc_v6_0_wait_for_idle((void *)adev);
60 blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
61 if (REG_GET_FIELD(blackout, mmMC_SHARED_BLACKOUT_CNTL, xxBLACKOUT_MODE) != 1) {
62 /* Block CPU access */
65 blackout = REG_SET_FIELD(blackout,
66 mmMC_SHARED_BLACKOUT_CNTL, xxBLACKOUT_MODE, 0);
67 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout | 1);
69 /* wait for the MC to settle */
74 static void gmc_v6_0_mc_resume(struct amdgpu_device *adev,
75 struct amdgpu_mode_mc_save *save)
79 /* unblackout the MC */
80 tmp = RREG32(MC_SHARED_BLACKOUT_CNTL);
81 tmp = REG_SET_FIELD(tmp, mmMC_SHARED_BLACKOUT_CNTL, xxBLACKOUT_MODE, 0);
82 WREG32(MC_SHARED_BLACKOUT_CNTL, tmp);
83 /* allow CPU access */
84 tmp = REG_SET_FIELD(0, mmBIF_FB_EN, xxFB_READ_EN, 1);
85 tmp = REG_SET_FIELD(tmp, mmBIF_FB_EN, xxFB_WRITE_EN, 1);
86 WREG32(BIF_FB_EN, tmp);
88 if (adev->mode_info.num_crtc)
89 amdgpu_display_resume_mc_access(adev, save);
93 static int gmc_v6_0_init_microcode(struct amdgpu_device *adev)
95 const char *chip_name;
101 switch (adev->asic_type) {
103 chip_name = "tahiti";
106 chip_name = "pitcairn";
115 chip_name = "hainan";
120 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
121 err = request_firmware(&adev->mc.fw, fw_name, adev->dev);
125 err = amdgpu_ucode_validate(adev->mc.fw);
130 "si_mc: Failed to load firmware \"%s\"\n",
132 release_firmware(adev->mc.fw);
138 static int gmc_v6_0_mc_load_microcode(struct amdgpu_device *adev)
140 const __le32 *new_fw_data = NULL;
142 const __le32 *new_io_mc_regs = NULL;
143 int i, regs_size, ucode_size;
144 const struct mc_firmware_header_v1_0 *hdr;
149 hdr = (const struct mc_firmware_header_v1_0 *)adev->mc.fw->data;
151 amdgpu_ucode_print_mc_hdr(&hdr->header);
153 adev->mc.fw_version = le32_to_cpu(hdr->header.ucode_version);
154 regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
155 new_io_mc_regs = (const __le32 *)
156 (adev->mc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
157 ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
158 new_fw_data = (const __le32 *)
159 (adev->mc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
161 running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK;
165 /* reset the engine and set to writable */
166 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
167 WREG32(MC_SEQ_SUP_CNTL, 0x00000010);
169 /* load mc io regs */
170 for (i = 0; i < regs_size; i++) {
171 WREG32(MC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(new_io_mc_regs++));
172 WREG32(MC_SEQ_IO_DEBUG_DATA, le32_to_cpup(new_io_mc_regs++));
174 /* load the MC ucode */
175 for (i = 0; i < ucode_size; i++) {
176 WREG32(MC_SEQ_SUP_PGM, le32_to_cpup(new_fw_data++));
179 /* put the engine back into the active state */
180 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
181 WREG32(MC_SEQ_SUP_CNTL, 0x00000004);
182 WREG32(MC_SEQ_SUP_CNTL, 0x00000001);
184 /* wait for training to complete */
185 for (i = 0; i < adev->usec_timeout; i++) {
186 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D0)
190 for (i = 0; i < adev->usec_timeout; i++) {
191 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D1)
201 static void gmc_v6_0_vram_gtt_location(struct amdgpu_device *adev,
202 struct amdgpu_mc *mc)
204 if (mc->mc_vram_size > 0xFFC0000000ULL) {
205 dev_warn(adev->dev, "limiting VRAM\n");
206 mc->real_vram_size = 0xFFC0000000ULL;
207 mc->mc_vram_size = 0xFFC0000000ULL;
209 amdgpu_vram_location(adev, &adev->mc, 0);
210 adev->mc.gtt_base_align = 0;
211 amdgpu_gtt_location(adev, mc);
214 static void gmc_v6_0_mc_program(struct amdgpu_device *adev)
216 struct amdgpu_mode_mc_save save;
221 for (i = 0, j = 0; i < 32; i++, j += 0x6) {
222 WREG32((0xb05 + j), 0x00000000);
223 WREG32((0xb06 + j), 0x00000000);
224 WREG32((0xb07 + j), 0x00000000);
225 WREG32((0xb08 + j), 0x00000000);
226 WREG32((0xb09 + j), 0x00000000);
228 WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
230 gmc_v6_0_mc_stop(adev, &save);
232 if (gmc_v6_0_wait_for_idle((void *)adev)) {
233 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
236 WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
237 /* Update configuration */
238 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
239 adev->mc.vram_start >> 12);
240 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
241 adev->mc.vram_end >> 12);
242 WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
243 adev->vram_scratch.gpu_addr >> 12);
244 tmp = ((adev->mc.vram_end >> 24) & 0xFFFF) << 16;
245 tmp |= ((adev->mc.vram_start >> 24) & 0xFFFF);
246 WREG32(MC_VM_FB_LOCATION, tmp);
247 /* XXX double check these! */
248 WREG32(HDP_NONSURFACE_BASE, (adev->mc.vram_start >> 8));
249 WREG32(HDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
250 WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF);
251 WREG32(MC_VM_AGP_BASE, 0);
252 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
253 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
255 if (gmc_v6_0_wait_for_idle((void *)adev)) {
256 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
258 gmc_v6_0_mc_resume(adev, &save);
259 amdgpu_display_set_vga_render_state(adev, false);
262 static int gmc_v6_0_mc_init(struct amdgpu_device *adev)
266 int chansize, numchan;
268 tmp = RREG32(MC_ARB_RAMCFG);
269 if (tmp & CHANSIZE_OVERRIDE) {
271 } else if (tmp & CHANSIZE_MASK) {
276 tmp = RREG32(MC_SHARED_CHMAP);
277 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
307 adev->mc.vram_width = numchan * chansize;
308 /* Could aper size report 0 ? */
309 adev->mc.aper_base = pci_resource_start(adev->pdev, 0);
310 adev->mc.aper_size = pci_resource_len(adev->pdev, 0);
311 /* size in MB on si */
312 adev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024ULL * 1024ULL;
313 adev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024ULL * 1024ULL;
314 adev->mc.visible_vram_size = adev->mc.aper_size;
316 /* unless the user had overridden it, set the gart
317 * size equal to the 1024 or vram, whichever is larger.
319 if (amdgpu_gart_size == -1)
320 adev->mc.gtt_size = amdgpu_ttm_get_gtt_mem_size(adev);
322 adev->mc.gtt_size = (uint64_t)amdgpu_gart_size << 20;
324 gmc_v6_0_vram_gtt_location(adev, &adev->mc);
329 static void gmc_v6_0_gart_flush_gpu_tlb(struct amdgpu_device *adev,
332 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0);
334 WREG32(VM_INVALIDATE_REQUEST, 1 << vmid);
337 static int gmc_v6_0_gart_set_pte_pde(struct amdgpu_device *adev,
339 uint32_t gpu_page_idx,
343 void __iomem *ptr = (void *)cpu_pt_addr;
346 value = addr & 0xFFFFFFFFFFFFF000ULL;
348 writeq(value, ptr + (gpu_page_idx * 8));
353 static void gmc_v6_0_set_fault_enable_default(struct amdgpu_device *adev,
358 tmp = RREG32(VM_CONTEXT1_CNTL);
359 tmp = REG_SET_FIELD(tmp, mmVM_CONTEXT1_CNTL,
360 xxRANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
361 tmp = REG_SET_FIELD(tmp, mmVM_CONTEXT1_CNTL,
362 xxDUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
363 tmp = REG_SET_FIELD(tmp, mmVM_CONTEXT1_CNTL,
364 xxPDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
365 tmp = REG_SET_FIELD(tmp, mmVM_CONTEXT1_CNTL,
366 xxVALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
367 tmp = REG_SET_FIELD(tmp, mmVM_CONTEXT1_CNTL,
368 xxREAD_PROTECTION_FAULT_ENABLE_DEFAULT, value);
369 tmp = REG_SET_FIELD(tmp, mmVM_CONTEXT1_CNTL,
370 xxWRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
371 WREG32(VM_CONTEXT1_CNTL, tmp);
374 static int gmc_v6_0_gart_enable(struct amdgpu_device *adev)
378 if (adev->gart.robj == NULL) {
379 dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
382 r = amdgpu_gart_table_vram_pin(adev);
385 /* Setup TLB control */
386 WREG32(MC_VM_MX_L1_TLB_CNTL,
389 ENABLE_L1_FRAGMENT_PROCESSING |
390 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
391 ENABLE_ADVANCED_DRIVER_MODEL |
392 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
394 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE |
395 ENABLE_L2_FRAGMENT_PROCESSING |
396 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
397 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
398 EFFECTIVE_L2_QUEUE_SIZE(7) |
399 CONTEXT1_IDENTITY_ACCESS_MODE(1));
400 WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE);
401 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
403 L2_CACHE_BIGK_FRAGMENT_SIZE(4));
405 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->mc.gtt_start >> 12);
406 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->mc.gtt_end >> 12);
407 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12);
408 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
409 (u32)(adev->dummy_page.addr >> 12));
410 WREG32(VM_CONTEXT0_CNTL2, 0);
411 WREG32(VM_CONTEXT0_CNTL, (ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
412 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT));
418 /* empty context1-15 */
419 /* set vm size, must be a multiple of 4 */
420 WREG32(VM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
421 WREG32(VM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
422 /* Assign the pt base to something valid for now; the pts used for
423 * the VMs are determined by the application and setup and assigned
424 * on the fly in the vm part of radeon_gart.c
426 for (i = 1; i < 16; i++) {
428 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
429 adev->gart.table_addr >> 12);
431 WREG32(VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
432 adev->gart.table_addr >> 12);
435 /* enable context1-15 */
436 WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
437 (u32)(adev->dummy_page.addr >> 12));
438 WREG32(VM_CONTEXT1_CNTL2, 4);
439 WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) |
440 PAGE_TABLE_BLOCK_SIZE(amdgpu_vm_block_size - 9) |
441 RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
442 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT |
443 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
444 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT |
445 PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT |
446 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT |
447 VALID_PROTECTION_FAULT_ENABLE_INTERRUPT |
448 VALID_PROTECTION_FAULT_ENABLE_DEFAULT |
449 READ_PROTECTION_FAULT_ENABLE_INTERRUPT |
450 READ_PROTECTION_FAULT_ENABLE_DEFAULT |
451 WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT |
452 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT);
454 gmc_v6_0_gart_flush_gpu_tlb(adev, 0);
455 dev_info(adev->dev, "PCIE GART of %uM enabled (table at 0x%016llX).\n",
456 (unsigned)(adev->mc.gtt_size >> 20),
457 (unsigned long long)adev->gart.table_addr);
458 adev->gart.ready = true;
462 static int gmc_v6_0_gart_init(struct amdgpu_device *adev)
466 if (adev->gart.robj) {
467 dev_warn(adev->dev, "gmc_v6_0 PCIE GART already initialized\n");
470 r = amdgpu_gart_init(adev);
473 adev->gart.table_size = adev->gart.num_gpu_pages * 8;
474 return amdgpu_gart_table_vram_alloc(adev);
477 static void gmc_v6_0_gart_disable(struct amdgpu_device *adev)
481 for (i = 1; i < 16; ++i) {
484 reg = VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i ;
486 reg = VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (i - 8);
487 adev->vm_manager.saved_table_addr[i] = RREG32(reg);
490 /* Disable all tables */
491 WREG32(VM_CONTEXT0_CNTL, 0);
492 WREG32(VM_CONTEXT1_CNTL, 0);
493 /* Setup TLB control */
494 WREG32(MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE_NOT_IN_SYS |
495 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
497 WREG32(VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
498 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
499 EFFECTIVE_L2_QUEUE_SIZE(7) |
500 CONTEXT1_IDENTITY_ACCESS_MODE(1));
501 WREG32(VM_L2_CNTL2, 0);
502 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
503 L2_CACHE_BIGK_FRAGMENT_SIZE(0));
504 amdgpu_gart_table_vram_unpin(adev);
507 static void gmc_v6_0_gart_fini(struct amdgpu_device *adev)
509 amdgpu_gart_table_vram_free(adev);
510 amdgpu_gart_fini(adev);
513 static int gmc_v6_0_vm_init(struct amdgpu_device *adev)
517 * VMID 0 is reserved for System
518 * amdgpu graphics/compute will use VMIDs 1-7
519 * amdkfd will use VMIDs 8-15
521 adev->vm_manager.num_ids = AMDGPU_NUM_OF_VMIDS;
522 amdgpu_vm_manager_init(adev);
524 /* base offset of vram pages */
525 if (adev->flags & AMD_IS_APU) {
526 u64 tmp = RREG32(MC_VM_FB_OFFSET);
528 adev->vm_manager.vram_base_offset = tmp;
530 adev->vm_manager.vram_base_offset = 0;
535 static void gmc_v6_0_vm_fini(struct amdgpu_device *adev)
539 static void gmc_v6_0_vm_decode_fault(struct amdgpu_device *adev,
540 u32 status, u32 addr, u32 mc_client)
543 u32 vmid = REG_GET_FIELD(status, mmVM_CONTEXT1_PROTECTION_FAULT_STATUS, xxVMID);
544 u32 protections = REG_GET_FIELD(status, mmVM_CONTEXT1_PROTECTION_FAULT_STATUS,
546 char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
547 (mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
549 mc_id = REG_GET_FIELD(status, mmVM_CONTEXT1_PROTECTION_FAULT_STATUS,
552 dev_err(adev->dev, "VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
553 protections, vmid, addr,
554 REG_GET_FIELD(status, mmVM_CONTEXT1_PROTECTION_FAULT_STATUS,
555 xxMEMORY_CLIENT_RW) ?
556 "write" : "read", block, mc_client, mc_id);
560 static const u32 mc_cg_registers[] = {
572 static const u32 mc_cg_ls_en[] = {
573 MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK,
574 MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK,
575 MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK,
576 MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK,
577 ATC_MISC_CG__MEM_LS_ENABLE_MASK,
578 MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK,
579 MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK,
580 MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK,
581 VM_L2_CG__MEM_LS_ENABLE_MASK,
584 static const u32 mc_cg_en[] = {
585 MC_HUB_MISC_HUB_CG__ENABLE_MASK,
586 MC_HUB_MISC_SIP_CG__ENABLE_MASK,
587 MC_HUB_MISC_VM_CG__ENABLE_MASK,
588 MC_XPB_CLK_GAT__ENABLE_MASK,
589 ATC_MISC_CG__ENABLE_MASK,
590 MC_CITF_MISC_WR_CG__ENABLE_MASK,
591 MC_CITF_MISC_RD_CG__ENABLE_MASK,
592 MC_CITF_MISC_VM_CG__ENABLE_MASK,
593 VM_L2_CG__ENABLE_MASK,
596 static void gmc_v6_0_enable_mc_ls(struct amdgpu_device *adev,
602 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
603 orig = data = RREG32(mc_cg_registers[i]);
604 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_LS))
605 data |= mc_cg_ls_en[i];
607 data &= ~mc_cg_ls_en[i];
609 WREG32(mc_cg_registers[i], data);
613 static void gmc_v6_0_enable_mc_mgcg(struct amdgpu_device *adev,
619 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
620 orig = data = RREG32(mc_cg_registers[i]);
621 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_MGCG))
624 data &= ~mc_cg_en[i];
626 WREG32(mc_cg_registers[i], data);
630 static void gmc_v6_0_enable_bif_mgls(struct amdgpu_device *adev,
635 orig = data = RREG32_PCIE(ixPCIE_CNTL2);
637 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_BIF_LS)) {
638 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 1);
639 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 1);
640 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 1);
641 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 1);
643 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 0);
644 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 0);
645 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 0);
646 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 0);
650 WREG32_PCIE(ixPCIE_CNTL2, data);
653 static void gmc_v6_0_enable_hdp_mgcg(struct amdgpu_device *adev,
658 orig = data = RREG32(HDP_HOST_PATH_CNTL);
660 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_MGCG))
661 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 0);
663 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 1);
666 WREG32(HDP_HOST_PATH_CNTL, data);
669 static void gmc_v6_0_enable_hdp_ls(struct amdgpu_device *adev,
674 orig = data = RREG32(HDP_MEM_POWER_LS);
676 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_LS))
677 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 1);
679 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 0);
682 WREG32(HDP_MEM_POWER_LS, data);
686 static int gmc_v6_0_convert_vram_type(int mc_seq_vram_type)
688 switch (mc_seq_vram_type) {
689 case MC_SEQ_MISC0__MT__GDDR1:
690 return AMDGPU_VRAM_TYPE_GDDR1;
691 case MC_SEQ_MISC0__MT__DDR2:
692 return AMDGPU_VRAM_TYPE_DDR2;
693 case MC_SEQ_MISC0__MT__GDDR3:
694 return AMDGPU_VRAM_TYPE_GDDR3;
695 case MC_SEQ_MISC0__MT__GDDR4:
696 return AMDGPU_VRAM_TYPE_GDDR4;
697 case MC_SEQ_MISC0__MT__GDDR5:
698 return AMDGPU_VRAM_TYPE_GDDR5;
699 case MC_SEQ_MISC0__MT__DDR3:
700 return AMDGPU_VRAM_TYPE_DDR3;
702 return AMDGPU_VRAM_TYPE_UNKNOWN;
706 static int gmc_v6_0_early_init(void *handle)
708 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
710 gmc_v6_0_set_gart_funcs(adev);
711 gmc_v6_0_set_irq_funcs(adev);
713 if (adev->flags & AMD_IS_APU) {
714 adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
716 u32 tmp = RREG32(MC_SEQ_MISC0);
717 tmp &= MC_SEQ_MISC0__MT__MASK;
718 adev->mc.vram_type = gmc_v6_0_convert_vram_type(tmp);
724 static int gmc_v6_0_late_init(void *handle)
726 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
728 return amdgpu_irq_get(adev, &adev->mc.vm_fault, 0);
731 static int gmc_v6_0_sw_init(void *handle)
735 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
737 r = amdgpu_irq_add_id(adev, 146, &adev->mc.vm_fault);
741 r = amdgpu_irq_add_id(adev, 147, &adev->mc.vm_fault);
745 adev->vm_manager.max_pfn = amdgpu_vm_size << 18;
747 adev->mc.mc_mask = 0xffffffffffULL;
749 adev->need_dma32 = false;
750 dma_bits = adev->need_dma32 ? 32 : 40;
751 r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
753 adev->need_dma32 = true;
755 dev_warn(adev->dev, "amdgpu: No suitable DMA available.\n");
757 r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
759 pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
760 dev_warn(adev->dev, "amdgpu: No coherent DMA available.\n");
763 r = gmc_v6_0_init_microcode(adev);
765 dev_err(adev->dev, "Failed to load mc firmware!\n");
769 r = amdgpu_ttm_global_init(adev);
774 r = gmc_v6_0_mc_init(adev);
778 r = amdgpu_bo_init(adev);
782 r = gmc_v6_0_gart_init(adev);
786 if (!adev->vm_manager.enabled) {
787 r = gmc_v6_0_vm_init(adev);
789 dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
792 adev->vm_manager.enabled = true;
798 static int gmc_v6_0_sw_fini(void *handle)
800 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
802 if (adev->vm_manager.enabled) {
803 gmc_v6_0_vm_fini(adev);
804 adev->vm_manager.enabled = false;
806 gmc_v6_0_gart_fini(adev);
807 amdgpu_gem_force_release(adev);
808 amdgpu_bo_fini(adev);
813 static int gmc_v6_0_hw_init(void *handle)
816 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
818 gmc_v6_0_mc_program(adev);
820 if (!(adev->flags & AMD_IS_APU)) {
821 r = gmc_v6_0_mc_load_microcode(adev);
823 dev_err(adev->dev, "Failed to load MC firmware!\n");
828 r = gmc_v6_0_gart_enable(adev);
835 static int gmc_v6_0_hw_fini(void *handle)
837 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
839 amdgpu_irq_put(adev, &adev->mc.vm_fault, 0);
840 gmc_v6_0_gart_disable(adev);
845 static int gmc_v6_0_suspend(void *handle)
847 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
849 if (adev->vm_manager.enabled) {
850 gmc_v6_0_vm_fini(adev);
851 adev->vm_manager.enabled = false;
853 gmc_v6_0_hw_fini(adev);
858 static int gmc_v6_0_resume(void *handle)
861 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
863 r = gmc_v6_0_hw_init(adev);
867 if (!adev->vm_manager.enabled) {
868 r = gmc_v6_0_vm_init(adev);
870 dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
873 adev->vm_manager.enabled = true;
879 static bool gmc_v6_0_is_idle(void *handle)
881 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
882 u32 tmp = RREG32(SRBM_STATUS);
884 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
885 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
891 static int gmc_v6_0_wait_for_idle(void *handle)
895 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
897 for (i = 0; i < adev->usec_timeout; i++) {
898 tmp = RREG32(SRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
899 SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
900 SRBM_STATUS__MCC_BUSY_MASK |
901 SRBM_STATUS__MCD_BUSY_MASK |
902 SRBM_STATUS__VMC_BUSY_MASK);
911 static int gmc_v6_0_soft_reset(void *handle)
913 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
914 struct amdgpu_mode_mc_save save;
915 u32 srbm_soft_reset = 0;
916 u32 tmp = RREG32(SRBM_STATUS);
918 if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
919 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
920 mmSRBM_SOFT_RESET, xxSOFT_RESET_VMC, 1);
922 if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
923 SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
924 if (!(adev->flags & AMD_IS_APU))
925 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
926 mmSRBM_SOFT_RESET, xxSOFT_RESET_MC, 1);
929 if (srbm_soft_reset) {
930 gmc_v6_0_mc_stop(adev, &save);
931 if (gmc_v6_0_wait_for_idle(adev)) {
932 dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
936 tmp = RREG32(SRBM_SOFT_RESET);
937 tmp |= srbm_soft_reset;
938 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
939 WREG32(SRBM_SOFT_RESET, tmp);
940 tmp = RREG32(SRBM_SOFT_RESET);
944 tmp &= ~srbm_soft_reset;
945 WREG32(SRBM_SOFT_RESET, tmp);
946 tmp = RREG32(SRBM_SOFT_RESET);
950 gmc_v6_0_mc_resume(adev, &save);
957 static int gmc_v6_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
958 struct amdgpu_irq_src *src,
960 enum amdgpu_interrupt_state state)
963 u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
964 VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
965 VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
966 VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
967 VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
968 VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
971 case AMDGPU_IRQ_STATE_DISABLE:
972 tmp = RREG32(VM_CONTEXT0_CNTL);
974 WREG32(VM_CONTEXT0_CNTL, tmp);
975 tmp = RREG32(VM_CONTEXT1_CNTL);
977 WREG32(VM_CONTEXT1_CNTL, tmp);
979 case AMDGPU_IRQ_STATE_ENABLE:
980 tmp = RREG32(VM_CONTEXT0_CNTL);
982 WREG32(VM_CONTEXT0_CNTL, tmp);
983 tmp = RREG32(VM_CONTEXT1_CNTL);
985 WREG32(VM_CONTEXT1_CNTL, tmp);
994 static int gmc_v6_0_process_interrupt(struct amdgpu_device *adev,
995 struct amdgpu_irq_src *source,
996 struct amdgpu_iv_entry *entry)
1000 addr = RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR);
1001 status = RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS);
1002 WREG32_P(VM_CONTEXT1_CNTL2, 1, ~1);
1004 if (!addr && !status)
1007 if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1008 gmc_v6_0_set_fault_enable_default(adev, false);
1010 dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1011 entry->src_id, entry->src_data);
1012 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
1014 dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1016 gmc_v6_0_vm_decode_fault(adev, status, addr, 0);
1021 static int gmc_v6_0_set_clockgating_state(void *handle,
1022 enum amd_clockgating_state state)
1027 static int gmc_v6_0_set_powergating_state(void *handle,
1028 enum amd_powergating_state state)
1033 const struct amd_ip_funcs gmc_v6_0_ip_funcs = {
1035 .early_init = gmc_v6_0_early_init,
1036 .late_init = gmc_v6_0_late_init,
1037 .sw_init = gmc_v6_0_sw_init,
1038 .sw_fini = gmc_v6_0_sw_fini,
1039 .hw_init = gmc_v6_0_hw_init,
1040 .hw_fini = gmc_v6_0_hw_fini,
1041 .suspend = gmc_v6_0_suspend,
1042 .resume = gmc_v6_0_resume,
1043 .is_idle = gmc_v6_0_is_idle,
1044 .wait_for_idle = gmc_v6_0_wait_for_idle,
1045 .soft_reset = gmc_v6_0_soft_reset,
1046 .set_clockgating_state = gmc_v6_0_set_clockgating_state,
1047 .set_powergating_state = gmc_v6_0_set_powergating_state,
1050 static const struct amdgpu_gart_funcs gmc_v6_0_gart_funcs = {
1051 .flush_gpu_tlb = gmc_v6_0_gart_flush_gpu_tlb,
1052 .set_pte_pde = gmc_v6_0_gart_set_pte_pde,
1055 static const struct amdgpu_irq_src_funcs gmc_v6_0_irq_funcs = {
1056 .set = gmc_v6_0_vm_fault_interrupt_state,
1057 .process = gmc_v6_0_process_interrupt,
1060 static void gmc_v6_0_set_gart_funcs(struct amdgpu_device *adev)
1062 if (adev->gart.gart_funcs == NULL)
1063 adev->gart.gart_funcs = &gmc_v6_0_gart_funcs;
1066 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev)
1068 adev->mc.vm_fault.num_types = 1;
1069 adev->mc.vm_fault.funcs = &gmc_v6_0_irq_funcs;