xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c (revision ca55b2fe)
1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
3  *
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:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
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.
21  *
22  */
23 #include <linux/firmware.h>
24 #include "drmP.h"
25 #include "amdgpu.h"
26 #include "cikd.h"
27 #include "cik.h"
28 #include "gmc_v7_0.h"
29 #include "amdgpu_ucode.h"
30 
31 #include "bif/bif_4_1_d.h"
32 #include "bif/bif_4_1_sh_mask.h"
33 
34 #include "gmc/gmc_7_1_d.h"
35 #include "gmc/gmc_7_1_sh_mask.h"
36 
37 #include "oss/oss_2_0_d.h"
38 #include "oss/oss_2_0_sh_mask.h"
39 
40 static void gmc_v7_0_set_gart_funcs(struct amdgpu_device *adev);
41 static void gmc_v7_0_set_irq_funcs(struct amdgpu_device *adev);
42 
43 MODULE_FIRMWARE("radeon/boniare_mc.bin");
44 MODULE_FIRMWARE("radeon/hawaii_mc.bin");
45 
46 /**
47  * gmc8_mc_wait_for_idle - wait for MC idle callback.
48  *
49  * @adev: amdgpu_device pointer
50  *
51  * Wait for the MC (memory controller) to be idle.
52  * (evergreen+).
53  * Returns 0 if the MC is idle, -1 if not.
54  */
55 int gmc_v7_0_mc_wait_for_idle(struct amdgpu_device *adev)
56 {
57 	unsigned i;
58 	u32 tmp;
59 
60 	for (i = 0; i < adev->usec_timeout; i++) {
61 		/* read MC_STATUS */
62 		tmp = RREG32(mmSRBM_STATUS) & 0x1F00;
63 		if (!tmp)
64 			return 0;
65 		udelay(1);
66 	}
67 	return -1;
68 }
69 
70 void gmc_v7_0_mc_stop(struct amdgpu_device *adev,
71 		      struct amdgpu_mode_mc_save *save)
72 {
73 	u32 blackout;
74 
75 	if (adev->mode_info.num_crtc)
76 		amdgpu_display_stop_mc_access(adev, save);
77 
78 	amdgpu_asic_wait_for_mc_idle(adev);
79 
80 	blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
81 	if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
82 		/* Block CPU access */
83 		WREG32(mmBIF_FB_EN, 0);
84 		/* blackout the MC */
85 		blackout = REG_SET_FIELD(blackout,
86 					 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
87 		WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout | 1);
88 	}
89 	/* wait for the MC to settle */
90 	udelay(100);
91 }
92 
93 void gmc_v7_0_mc_resume(struct amdgpu_device *adev,
94 			struct amdgpu_mode_mc_save *save)
95 {
96 	u32 tmp;
97 
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);
106 
107 	if (adev->mode_info.num_crtc)
108 		amdgpu_display_resume_mc_access(adev, save);
109 }
110 
111 /**
112  * gmc_v7_0_init_microcode - load ucode images from disk
113  *
114  * @adev: amdgpu_device pointer
115  *
116  * Use the firmware interface to load the ucode images into
117  * the driver (not loaded into hw).
118  * Returns 0 on success, error on failure.
119  */
120 static int gmc_v7_0_init_microcode(struct amdgpu_device *adev)
121 {
122 	const char *chip_name;
123 	char fw_name[30];
124 	int err;
125 
126 	DRM_DEBUG("\n");
127 
128 	switch (adev->asic_type) {
129 	case CHIP_BONAIRE:
130 		chip_name = "bonaire";
131 		break;
132 	case CHIP_HAWAII:
133 		chip_name = "hawaii";
134 		break;
135 	case CHIP_KAVERI:
136 	case CHIP_KABINI:
137 		return 0;
138 	default: BUG();
139 	}
140 
141 	snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
142 	err = request_firmware(&adev->mc.fw, fw_name, adev->dev);
143 	if (err)
144 		goto out;
145 	err = amdgpu_ucode_validate(adev->mc.fw);
146 
147 out:
148 	if (err) {
149 		printk(KERN_ERR
150 		       "cik_mc: Failed to load firmware \"%s\"\n",
151 		       fw_name);
152 		release_firmware(adev->mc.fw);
153 		adev->mc.fw = NULL;
154 	}
155 	return err;
156 }
157 
158 /**
159  * gmc_v7_0_mc_load_microcode - load MC ucode into the hw
160  *
161  * @adev: amdgpu_device pointer
162  *
163  * Load the GDDR MC ucode into the hw (CIK).
164  * Returns 0 on success, error on failure.
165  */
166 static int gmc_v7_0_mc_load_microcode(struct amdgpu_device *adev)
167 {
168 	const struct mc_firmware_header_v1_0 *hdr;
169 	const __le32 *fw_data = NULL;
170 	const __le32 *io_mc_regs = NULL;
171 	u32 running, blackout = 0;
172 	int i, ucode_size, regs_size;
173 
174 	if (!adev->mc.fw)
175 		return -EINVAL;
176 
177 	hdr = (const struct mc_firmware_header_v1_0 *)adev->mc.fw->data;
178 	amdgpu_ucode_print_mc_hdr(&hdr->header);
179 
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 	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 	fw_data = (const __le32 *)
186 		(adev->mc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
187 
188 	running = REG_GET_FIELD(RREG32(mmMC_SEQ_SUP_CNTL), MC_SEQ_SUP_CNTL, RUN);
189 
190 	if (running == 0) {
191 		if (running) {
192 			blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
193 			WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout | 1);
194 		}
195 
196 		/* reset the engine and set to writable */
197 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
198 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
199 
200 		/* load mc io regs */
201 		for (i = 0; i < regs_size; i++) {
202 			WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
203 			WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
204 		}
205 		/* load the MC ucode */
206 		for (i = 0; i < ucode_size; i++)
207 			WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
208 
209 		/* put the engine back into the active state */
210 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
211 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
212 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
213 
214 		/* wait for training to complete */
215 		for (i = 0; i < adev->usec_timeout; i++) {
216 			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
217 					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D0))
218 				break;
219 			udelay(1);
220 		}
221 		for (i = 0; i < adev->usec_timeout; i++) {
222 			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
223 					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D1))
224 				break;
225 			udelay(1);
226 		}
227 
228 		if (running)
229 			WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout);
230 	}
231 
232 	return 0;
233 }
234 
235 static void gmc_v7_0_vram_gtt_location(struct amdgpu_device *adev,
236 				       struct amdgpu_mc *mc)
237 {
238 	if (mc->mc_vram_size > 0xFFC0000000ULL) {
239 		/* leave room for at least 1024M GTT */
240 		dev_warn(adev->dev, "limiting VRAM\n");
241 		mc->real_vram_size = 0xFFC0000000ULL;
242 		mc->mc_vram_size = 0xFFC0000000ULL;
243 	}
244 	amdgpu_vram_location(adev, &adev->mc, 0);
245 	adev->mc.gtt_base_align = 0;
246 	amdgpu_gtt_location(adev, mc);
247 }
248 
249 /**
250  * gmc_v7_0_mc_program - program the GPU memory controller
251  *
252  * @adev: amdgpu_device pointer
253  *
254  * Set the location of vram, gart, and AGP in the GPU's
255  * physical address space (CIK).
256  */
257 static void gmc_v7_0_mc_program(struct amdgpu_device *adev)
258 {
259 	struct amdgpu_mode_mc_save save;
260 	u32 tmp;
261 	int i, j;
262 
263 	/* Initialize HDP */
264 	for (i = 0, j = 0; i < 32; i++, j += 0x6) {
265 		WREG32((0xb05 + j), 0x00000000);
266 		WREG32((0xb06 + j), 0x00000000);
267 		WREG32((0xb07 + j), 0x00000000);
268 		WREG32((0xb08 + j), 0x00000000);
269 		WREG32((0xb09 + j), 0x00000000);
270 	}
271 	WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
272 
273 	if (adev->mode_info.num_crtc)
274 		amdgpu_display_set_vga_render_state(adev, false);
275 
276 	gmc_v7_0_mc_stop(adev, &save);
277 	if (amdgpu_asic_wait_for_mc_idle(adev)) {
278 		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
279 	}
280 	/* Update configuration */
281 	WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
282 	       adev->mc.vram_start >> 12);
283 	WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
284 	       adev->mc.vram_end >> 12);
285 	WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
286 	       adev->vram_scratch.gpu_addr >> 12);
287 	tmp = ((adev->mc.vram_end >> 24) & 0xFFFF) << 16;
288 	tmp |= ((adev->mc.vram_start >> 24) & 0xFFFF);
289 	WREG32(mmMC_VM_FB_LOCATION, tmp);
290 	/* XXX double check these! */
291 	WREG32(mmHDP_NONSURFACE_BASE, (adev->mc.vram_start >> 8));
292 	WREG32(mmHDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
293 	WREG32(mmHDP_NONSURFACE_SIZE, 0x3FFFFFFF);
294 	WREG32(mmMC_VM_AGP_BASE, 0);
295 	WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
296 	WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
297 	if (amdgpu_asic_wait_for_mc_idle(adev)) {
298 		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
299 	}
300 	gmc_v7_0_mc_resume(adev, &save);
301 
302 	WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK);
303 
304 	tmp = RREG32(mmHDP_MISC_CNTL);
305 	tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 1);
306 	WREG32(mmHDP_MISC_CNTL, tmp);
307 
308 	tmp = RREG32(mmHDP_HOST_PATH_CNTL);
309 	WREG32(mmHDP_HOST_PATH_CNTL, tmp);
310 }
311 
312 /**
313  * gmc_v7_0_mc_init - initialize the memory controller driver params
314  *
315  * @adev: amdgpu_device pointer
316  *
317  * Look up the amount of vram, vram width, and decide how to place
318  * vram and gart within the GPU's physical address space (CIK).
319  * Returns 0 for success.
320  */
321 static int gmc_v7_0_mc_init(struct amdgpu_device *adev)
322 {
323 	u32 tmp;
324 	int chansize, numchan;
325 
326 	/* Get VRAM informations */
327 	tmp = RREG32(mmMC_ARB_RAMCFG);
328 	if (REG_GET_FIELD(tmp, MC_ARB_RAMCFG, CHANSIZE)) {
329 		chansize = 64;
330 	} else {
331 		chansize = 32;
332 	}
333 	tmp = RREG32(mmMC_SHARED_CHMAP);
334 	switch (REG_GET_FIELD(tmp, MC_SHARED_CHMAP, NOOFCHAN)) {
335 	case 0:
336 	default:
337 		numchan = 1;
338 		break;
339 	case 1:
340 		numchan = 2;
341 		break;
342 	case 2:
343 		numchan = 4;
344 		break;
345 	case 3:
346 		numchan = 8;
347 		break;
348 	case 4:
349 		numchan = 3;
350 		break;
351 	case 5:
352 		numchan = 6;
353 		break;
354 	case 6:
355 		numchan = 10;
356 		break;
357 	case 7:
358 		numchan = 12;
359 		break;
360 	case 8:
361 		numchan = 16;
362 		break;
363 	}
364 	adev->mc.vram_width = numchan * chansize;
365 	/* Could aper size report 0 ? */
366 	adev->mc.aper_base = pci_resource_start(adev->pdev, 0);
367 	adev->mc.aper_size = pci_resource_len(adev->pdev, 0);
368 	/* size in MB on si */
369 	adev->mc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
370 	adev->mc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
371 	adev->mc.visible_vram_size = adev->mc.aper_size;
372 
373 	/* unless the user had overridden it, set the gart
374 	 * size equal to the 1024 or vram, whichever is larger.
375 	 */
376 	if (amdgpu_gart_size == -1)
377 		adev->mc.gtt_size = max((1024ULL << 20), adev->mc.mc_vram_size);
378 	else
379 		adev->mc.gtt_size = (uint64_t)amdgpu_gart_size << 20;
380 
381 	gmc_v7_0_vram_gtt_location(adev, &adev->mc);
382 
383 	return 0;
384 }
385 
386 /*
387  * GART
388  * VMID 0 is the physical GPU addresses as used by the kernel.
389  * VMIDs 1-15 are used for userspace clients and are handled
390  * by the amdgpu vm/hsa code.
391  */
392 
393 /**
394  * gmc_v7_0_gart_flush_gpu_tlb - gart tlb flush callback
395  *
396  * @adev: amdgpu_device pointer
397  * @vmid: vm instance to flush
398  *
399  * Flush the TLB for the requested page table (CIK).
400  */
401 static void gmc_v7_0_gart_flush_gpu_tlb(struct amdgpu_device *adev,
402 					uint32_t vmid)
403 {
404 	/* flush hdp cache */
405 	WREG32(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0);
406 
407 	/* bits 0-15 are the VM contexts0-15 */
408 	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
409 }
410 
411 /**
412  * gmc_v7_0_gart_set_pte_pde - update the page tables using MMIO
413  *
414  * @adev: amdgpu_device pointer
415  * @cpu_pt_addr: cpu address of the page table
416  * @gpu_page_idx: entry in the page table to update
417  * @addr: dst addr to write into pte/pde
418  * @flags: access flags
419  *
420  * Update the page tables using the CPU.
421  */
422 static int gmc_v7_0_gart_set_pte_pde(struct amdgpu_device *adev,
423 				     void *cpu_pt_addr,
424 				     uint32_t gpu_page_idx,
425 				     uint64_t addr,
426 				     uint32_t flags)
427 {
428 	void __iomem *ptr = (void *)cpu_pt_addr;
429 	uint64_t value;
430 
431 	value = addr & 0xFFFFFFFFFFFFF000ULL;
432 	value |= flags;
433 	writeq(value, ptr + (gpu_page_idx * 8));
434 
435 	return 0;
436 }
437 
438 /**
439  * gmc_v7_0_gart_enable - gart enable
440  *
441  * @adev: amdgpu_device pointer
442  *
443  * This sets up the TLBs, programs the page tables for VMID0,
444  * sets up the hw for VMIDs 1-15 which are allocated on
445  * demand, and sets up the global locations for the LDS, GDS,
446  * and GPUVM for FSA64 clients (CIK).
447  * Returns 0 for success, errors for failure.
448  */
449 static int gmc_v7_0_gart_enable(struct amdgpu_device *adev)
450 {
451 	int r, i;
452 	u32 tmp;
453 
454 	if (adev->gart.robj == NULL) {
455 		dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
456 		return -EINVAL;
457 	}
458 	r = amdgpu_gart_table_vram_pin(adev);
459 	if (r)
460 		return r;
461 	/* Setup TLB control */
462 	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
463 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 1);
464 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 1);
465 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE, 3);
466 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 1);
467 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
468 	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
469 	/* Setup L2 cache */
470 	tmp = RREG32(mmVM_L2_CNTL);
471 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
472 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
473 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE, 1);
474 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1);
475 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7);
476 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
477 	WREG32(mmVM_L2_CNTL, tmp);
478 	tmp = REG_SET_FIELD(0, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
479 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
480 	WREG32(mmVM_L2_CNTL2, tmp);
481 	tmp = RREG32(mmVM_L2_CNTL3);
482 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY, 1);
483 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 4);
484 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_FRAGMENT_SIZE, 4);
485 	WREG32(mmVM_L2_CNTL3, tmp);
486 	/* setup context0 */
487 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->mc.gtt_start >> 12);
488 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, (adev->mc.gtt_end >> 12) - 1);
489 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12);
490 	WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
491 			(u32)(adev->dummy_page.addr >> 12));
492 	WREG32(mmVM_CONTEXT0_CNTL2, 0);
493 	tmp = RREG32(mmVM_CONTEXT0_CNTL);
494 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
495 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH, 0);
496 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
497 	WREG32(mmVM_CONTEXT0_CNTL, tmp);
498 
499 	WREG32(0x575, 0);
500 	WREG32(0x576, 0);
501 	WREG32(0x577, 0);
502 
503 	/* empty context1-15 */
504 	/* FIXME start with 4G, once using 2 level pt switch to full
505 	 * vm size space
506 	 */
507 	/* set vm size, must be a multiple of 4 */
508 	WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
509 	WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
510 	for (i = 1; i < 16; i++) {
511 		if (i < 8)
512 			WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
513 			       adev->gart.table_addr >> 12);
514 		else
515 			WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
516 			       adev->gart.table_addr >> 12);
517 	}
518 
519 	/* enable context1-15 */
520 	WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
521 	       (u32)(adev->dummy_page.addr >> 12));
522 	WREG32(mmVM_CONTEXT1_CNTL2, 4);
523 	tmp = RREG32(mmVM_CONTEXT1_CNTL);
524 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
525 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH, 1);
526 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
527 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
528 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
529 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
530 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
531 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
532 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE,
533 			    amdgpu_vm_block_size - 9);
534 	WREG32(mmVM_CONTEXT1_CNTL, tmp);
535 
536 	if (adev->asic_type == CHIP_KAVERI) {
537 		tmp = RREG32(mmCHUB_CONTROL);
538 		tmp &= ~BYPASS_VM;
539 		WREG32(mmCHUB_CONTROL, tmp);
540 	}
541 
542 	gmc_v7_0_gart_flush_gpu_tlb(adev, 0);
543 	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
544 		 (unsigned)(adev->mc.gtt_size >> 20),
545 		 (unsigned long long)adev->gart.table_addr);
546 	adev->gart.ready = true;
547 	return 0;
548 }
549 
550 static int gmc_v7_0_gart_init(struct amdgpu_device *adev)
551 {
552 	int r;
553 
554 	if (adev->gart.robj) {
555 		WARN(1, "R600 PCIE GART already initialized\n");
556 		return 0;
557 	}
558 	/* Initialize common gart structure */
559 	r = amdgpu_gart_init(adev);
560 	if (r)
561 		return r;
562 	adev->gart.table_size = adev->gart.num_gpu_pages * 8;
563 	return amdgpu_gart_table_vram_alloc(adev);
564 }
565 
566 /**
567  * gmc_v7_0_gart_disable - gart disable
568  *
569  * @adev: amdgpu_device pointer
570  *
571  * This disables all VM page table (CIK).
572  */
573 static void gmc_v7_0_gart_disable(struct amdgpu_device *adev)
574 {
575 	u32 tmp;
576 
577 	/* Disable all tables */
578 	WREG32(mmVM_CONTEXT0_CNTL, 0);
579 	WREG32(mmVM_CONTEXT1_CNTL, 0);
580 	/* Setup TLB control */
581 	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
582 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
583 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 0);
584 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 0);
585 	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
586 	/* Setup L2 cache */
587 	tmp = RREG32(mmVM_L2_CNTL);
588 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
589 	WREG32(mmVM_L2_CNTL, tmp);
590 	WREG32(mmVM_L2_CNTL2, 0);
591 	amdgpu_gart_table_vram_unpin(adev);
592 }
593 
594 /**
595  * gmc_v7_0_gart_fini - vm fini callback
596  *
597  * @adev: amdgpu_device pointer
598  *
599  * Tears down the driver GART/VM setup (CIK).
600  */
601 static void gmc_v7_0_gart_fini(struct amdgpu_device *adev)
602 {
603 	amdgpu_gart_table_vram_free(adev);
604 	amdgpu_gart_fini(adev);
605 }
606 
607 /*
608  * vm
609  * VMID 0 is the physical GPU addresses as used by the kernel.
610  * VMIDs 1-15 are used for userspace clients and are handled
611  * by the amdgpu vm/hsa code.
612  */
613 /**
614  * gmc_v7_0_vm_init - cik vm init callback
615  *
616  * @adev: amdgpu_device pointer
617  *
618  * Inits cik specific vm parameters (number of VMs, base of vram for
619  * VMIDs 1-15) (CIK).
620  * Returns 0 for success.
621  */
622 static int gmc_v7_0_vm_init(struct amdgpu_device *adev)
623 {
624 	/*
625 	 * number of VMs
626 	 * VMID 0 is reserved for System
627 	 * amdgpu graphics/compute will use VMIDs 1-7
628 	 * amdkfd will use VMIDs 8-15
629 	 */
630 	adev->vm_manager.nvm = AMDGPU_NUM_OF_VMIDS;
631 
632 	/* base offset of vram pages */
633 	if (adev->flags & AMD_IS_APU) {
634 		u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
635 		tmp <<= 22;
636 		adev->vm_manager.vram_base_offset = tmp;
637 	} else
638 		adev->vm_manager.vram_base_offset = 0;
639 
640 	return 0;
641 }
642 
643 /**
644  * gmc_v7_0_vm_fini - cik vm fini callback
645  *
646  * @adev: amdgpu_device pointer
647  *
648  * Tear down any asic specific VM setup (CIK).
649  */
650 static void gmc_v7_0_vm_fini(struct amdgpu_device *adev)
651 {
652 }
653 
654 /**
655  * gmc_v7_0_vm_decode_fault - print human readable fault info
656  *
657  * @adev: amdgpu_device pointer
658  * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
659  * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
660  *
661  * Print human readable fault information (CIK).
662  */
663 static void gmc_v7_0_vm_decode_fault(struct amdgpu_device *adev,
664 				     u32 status, u32 addr, u32 mc_client)
665 {
666 	u32 mc_id;
667 	u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
668 	u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
669 					PROTECTIONS);
670 	char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
671 		(mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
672 
673 	mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
674 			      MEMORY_CLIENT_ID);
675 
676 	printk("VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
677 	       protections, vmid, addr,
678 	       REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
679 			     MEMORY_CLIENT_RW) ?
680 	       "write" : "read", block, mc_client, mc_id);
681 }
682 
683 
684 static const u32 mc_cg_registers[] = {
685 	mmMC_HUB_MISC_HUB_CG,
686 	mmMC_HUB_MISC_SIP_CG,
687 	mmMC_HUB_MISC_VM_CG,
688 	mmMC_XPB_CLK_GAT,
689 	mmATC_MISC_CG,
690 	mmMC_CITF_MISC_WR_CG,
691 	mmMC_CITF_MISC_RD_CG,
692 	mmMC_CITF_MISC_VM_CG,
693 	mmVM_L2_CG,
694 };
695 
696 static const u32 mc_cg_ls_en[] = {
697 	MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK,
698 	MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK,
699 	MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK,
700 	MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK,
701 	ATC_MISC_CG__MEM_LS_ENABLE_MASK,
702 	MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK,
703 	MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK,
704 	MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK,
705 	VM_L2_CG__MEM_LS_ENABLE_MASK,
706 };
707 
708 static const u32 mc_cg_en[] = {
709 	MC_HUB_MISC_HUB_CG__ENABLE_MASK,
710 	MC_HUB_MISC_SIP_CG__ENABLE_MASK,
711 	MC_HUB_MISC_VM_CG__ENABLE_MASK,
712 	MC_XPB_CLK_GAT__ENABLE_MASK,
713 	ATC_MISC_CG__ENABLE_MASK,
714 	MC_CITF_MISC_WR_CG__ENABLE_MASK,
715 	MC_CITF_MISC_RD_CG__ENABLE_MASK,
716 	MC_CITF_MISC_VM_CG__ENABLE_MASK,
717 	VM_L2_CG__ENABLE_MASK,
718 };
719 
720 static void gmc_v7_0_enable_mc_ls(struct amdgpu_device *adev,
721 				  bool enable)
722 {
723 	int i;
724 	u32 orig, data;
725 
726 	for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
727 		orig = data = RREG32(mc_cg_registers[i]);
728 		if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_LS))
729 			data |= mc_cg_ls_en[i];
730 		else
731 			data &= ~mc_cg_ls_en[i];
732 		if (data != orig)
733 			WREG32(mc_cg_registers[i], data);
734 	}
735 }
736 
737 static void gmc_v7_0_enable_mc_mgcg(struct amdgpu_device *adev,
738 				    bool enable)
739 {
740 	int i;
741 	u32 orig, data;
742 
743 	for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
744 		orig = data = RREG32(mc_cg_registers[i]);
745 		if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_MGCG))
746 			data |= mc_cg_en[i];
747 		else
748 			data &= ~mc_cg_en[i];
749 		if (data != orig)
750 			WREG32(mc_cg_registers[i], data);
751 	}
752 }
753 
754 static void gmc_v7_0_enable_bif_mgls(struct amdgpu_device *adev,
755 				     bool enable)
756 {
757 	u32 orig, data;
758 
759 	orig = data = RREG32_PCIE(ixPCIE_CNTL2);
760 
761 	if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_BIF_LS)) {
762 		data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 1);
763 		data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 1);
764 		data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 1);
765 		data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 1);
766 	} else {
767 		data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 0);
768 		data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 0);
769 		data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 0);
770 		data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 0);
771 	}
772 
773 	if (orig != data)
774 		WREG32_PCIE(ixPCIE_CNTL2, data);
775 }
776 
777 static void gmc_v7_0_enable_hdp_mgcg(struct amdgpu_device *adev,
778 				     bool enable)
779 {
780 	u32 orig, data;
781 
782 	orig = data = RREG32(mmHDP_HOST_PATH_CNTL);
783 
784 	if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_MGCG))
785 		data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 0);
786 	else
787 		data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 1);
788 
789 	if (orig != data)
790 		WREG32(mmHDP_HOST_PATH_CNTL, data);
791 }
792 
793 static void gmc_v7_0_enable_hdp_ls(struct amdgpu_device *adev,
794 				   bool enable)
795 {
796 	u32 orig, data;
797 
798 	orig = data = RREG32(mmHDP_MEM_POWER_LS);
799 
800 	if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_LS))
801 		data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 1);
802 	else
803 		data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 0);
804 
805 	if (orig != data)
806 		WREG32(mmHDP_MEM_POWER_LS, data);
807 }
808 
809 static int gmc_v7_0_convert_vram_type(int mc_seq_vram_type)
810 {
811 	switch (mc_seq_vram_type) {
812 	case MC_SEQ_MISC0__MT__GDDR1:
813 		return AMDGPU_VRAM_TYPE_GDDR1;
814 	case MC_SEQ_MISC0__MT__DDR2:
815 		return AMDGPU_VRAM_TYPE_DDR2;
816 	case MC_SEQ_MISC0__MT__GDDR3:
817 		return AMDGPU_VRAM_TYPE_GDDR3;
818 	case MC_SEQ_MISC0__MT__GDDR4:
819 		return AMDGPU_VRAM_TYPE_GDDR4;
820 	case MC_SEQ_MISC0__MT__GDDR5:
821 		return AMDGPU_VRAM_TYPE_GDDR5;
822 	case MC_SEQ_MISC0__MT__HBM:
823 		return AMDGPU_VRAM_TYPE_HBM;
824 	case MC_SEQ_MISC0__MT__DDR3:
825 		return AMDGPU_VRAM_TYPE_DDR3;
826 	default:
827 		return AMDGPU_VRAM_TYPE_UNKNOWN;
828 	}
829 }
830 
831 static int gmc_v7_0_early_init(void *handle)
832 {
833 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
834 
835 	gmc_v7_0_set_gart_funcs(adev);
836 	gmc_v7_0_set_irq_funcs(adev);
837 
838 	if (adev->flags & AMD_IS_APU) {
839 		adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
840 	} else {
841 		u32 tmp = RREG32(mmMC_SEQ_MISC0);
842 		tmp &= MC_SEQ_MISC0__MT__MASK;
843 		adev->mc.vram_type = gmc_v7_0_convert_vram_type(tmp);
844 	}
845 
846 	return 0;
847 }
848 
849 static int gmc_v7_0_late_init(void *handle)
850 {
851 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
852 
853 	return amdgpu_irq_get(adev, &adev->mc.vm_fault, 0);
854 }
855 
856 static int gmc_v7_0_sw_init(void *handle)
857 {
858 	int r;
859 	int dma_bits;
860 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
861 
862 	r = amdgpu_gem_init(adev);
863 	if (r)
864 		return r;
865 
866 	r = amdgpu_irq_add_id(adev, 146, &adev->mc.vm_fault);
867 	if (r)
868 		return r;
869 
870 	r = amdgpu_irq_add_id(adev, 147, &adev->mc.vm_fault);
871 	if (r)
872 		return r;
873 
874 	/* Adjust VM size here.
875 	 * Currently set to 4GB ((1 << 20) 4k pages).
876 	 * Max GPUVM size for cayman and SI is 40 bits.
877 	 */
878 	adev->vm_manager.max_pfn = amdgpu_vm_size << 18;
879 
880 	/* Set the internal MC address mask
881 	 * This is the max address of the GPU's
882 	 * internal address space.
883 	 */
884 	adev->mc.mc_mask = 0xffffffffffULL; /* 40 bit MC */
885 
886 	/* set DMA mask + need_dma32 flags.
887 	 * PCIE - can handle 40-bits.
888 	 * IGP - can handle 40-bits
889 	 * PCI - dma32 for legacy pci gart, 40 bits on newer asics
890 	 */
891 	adev->need_dma32 = false;
892 	dma_bits = adev->need_dma32 ? 32 : 40;
893 	r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
894 	if (r) {
895 		adev->need_dma32 = true;
896 		dma_bits = 32;
897 		printk(KERN_WARNING "amdgpu: No suitable DMA available.\n");
898 	}
899 	r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
900 	if (r) {
901 		pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
902 		printk(KERN_WARNING "amdgpu: No coherent DMA available.\n");
903 	}
904 
905 	r = gmc_v7_0_init_microcode(adev);
906 	if (r) {
907 		DRM_ERROR("Failed to load mc firmware!\n");
908 		return r;
909 	}
910 
911 	r = gmc_v7_0_mc_init(adev);
912 	if (r)
913 		return r;
914 
915 	/* Memory manager */
916 	r = amdgpu_bo_init(adev);
917 	if (r)
918 		return r;
919 
920 	r = gmc_v7_0_gart_init(adev);
921 	if (r)
922 		return r;
923 
924 	if (!adev->vm_manager.enabled) {
925 		r = gmc_v7_0_vm_init(adev);
926 		if (r) {
927 			dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
928 			return r;
929 		}
930 		adev->vm_manager.enabled = true;
931 	}
932 
933 	return r;
934 }
935 
936 static int gmc_v7_0_sw_fini(void *handle)
937 {
938 	int i;
939 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
940 
941 	if (adev->vm_manager.enabled) {
942 		for (i = 0; i < AMDGPU_NUM_VM; ++i)
943 			amdgpu_fence_unref(&adev->vm_manager.active[i]);
944 		gmc_v7_0_vm_fini(adev);
945 		adev->vm_manager.enabled = false;
946 	}
947 	gmc_v7_0_gart_fini(adev);
948 	amdgpu_gem_fini(adev);
949 	amdgpu_bo_fini(adev);
950 
951 	return 0;
952 }
953 
954 static int gmc_v7_0_hw_init(void *handle)
955 {
956 	int r;
957 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
958 
959 	gmc_v7_0_mc_program(adev);
960 
961 	if (!(adev->flags & AMD_IS_APU)) {
962 		r = gmc_v7_0_mc_load_microcode(adev);
963 		if (r) {
964 			DRM_ERROR("Failed to load MC firmware!\n");
965 			return r;
966 		}
967 	}
968 
969 	r = gmc_v7_0_gart_enable(adev);
970 	if (r)
971 		return r;
972 
973 	return r;
974 }
975 
976 static int gmc_v7_0_hw_fini(void *handle)
977 {
978 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
979 
980 	amdgpu_irq_put(adev, &adev->mc.vm_fault, 0);
981 	gmc_v7_0_gart_disable(adev);
982 
983 	return 0;
984 }
985 
986 static int gmc_v7_0_suspend(void *handle)
987 {
988 	int i;
989 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
990 
991 	if (adev->vm_manager.enabled) {
992 		for (i = 0; i < AMDGPU_NUM_VM; ++i)
993 			amdgpu_fence_unref(&adev->vm_manager.active[i]);
994 		gmc_v7_0_vm_fini(adev);
995 		adev->vm_manager.enabled = false;
996 	}
997 	gmc_v7_0_hw_fini(adev);
998 
999 	return 0;
1000 }
1001 
1002 static int gmc_v7_0_resume(void *handle)
1003 {
1004 	int r;
1005 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1006 
1007 	r = gmc_v7_0_hw_init(adev);
1008 	if (r)
1009 		return r;
1010 
1011 	if (!adev->vm_manager.enabled) {
1012 		r = gmc_v7_0_vm_init(adev);
1013 		if (r) {
1014 			dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
1015 			return r;
1016 		}
1017 		adev->vm_manager.enabled = true;
1018 	}
1019 
1020 	return r;
1021 }
1022 
1023 static bool gmc_v7_0_is_idle(void *handle)
1024 {
1025 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1026 	u32 tmp = RREG32(mmSRBM_STATUS);
1027 
1028 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1029 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
1030 		return false;
1031 
1032 	return true;
1033 }
1034 
1035 static int gmc_v7_0_wait_for_idle(void *handle)
1036 {
1037 	unsigned i;
1038 	u32 tmp;
1039 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1040 
1041 	for (i = 0; i < adev->usec_timeout; i++) {
1042 		/* read MC_STATUS */
1043 		tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
1044 					       SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1045 					       SRBM_STATUS__MCC_BUSY_MASK |
1046 					       SRBM_STATUS__MCD_BUSY_MASK |
1047 					       SRBM_STATUS__VMC_BUSY_MASK);
1048 		if (!tmp)
1049 			return 0;
1050 		udelay(1);
1051 	}
1052 	return -ETIMEDOUT;
1053 
1054 }
1055 
1056 static void gmc_v7_0_print_status(void *handle)
1057 {
1058 	int i, j;
1059 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1060 
1061 	dev_info(adev->dev, "GMC 8.x registers\n");
1062 	dev_info(adev->dev, "  SRBM_STATUS=0x%08X\n",
1063 		RREG32(mmSRBM_STATUS));
1064 	dev_info(adev->dev, "  SRBM_STATUS2=0x%08X\n",
1065 		RREG32(mmSRBM_STATUS2));
1066 
1067 	dev_info(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1068 		 RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR));
1069 	dev_info(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1070 		 RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS));
1071 	dev_info(adev->dev, "  MC_VM_MX_L1_TLB_CNTL=0x%08X\n",
1072 		 RREG32(mmMC_VM_MX_L1_TLB_CNTL));
1073 	dev_info(adev->dev, "  VM_L2_CNTL=0x%08X\n",
1074 		 RREG32(mmVM_L2_CNTL));
1075 	dev_info(adev->dev, "  VM_L2_CNTL2=0x%08X\n",
1076 		 RREG32(mmVM_L2_CNTL2));
1077 	dev_info(adev->dev, "  VM_L2_CNTL3=0x%08X\n",
1078 		 RREG32(mmVM_L2_CNTL3));
1079 	dev_info(adev->dev, "  VM_CONTEXT0_PAGE_TABLE_START_ADDR=0x%08X\n",
1080 		 RREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR));
1081 	dev_info(adev->dev, "  VM_CONTEXT0_PAGE_TABLE_END_ADDR=0x%08X\n",
1082 		 RREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR));
1083 	dev_info(adev->dev, "  VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR=0x%08X\n",
1084 		 RREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR));
1085 	dev_info(adev->dev, "  VM_CONTEXT0_CNTL2=0x%08X\n",
1086 		 RREG32(mmVM_CONTEXT0_CNTL2));
1087 	dev_info(adev->dev, "  VM_CONTEXT0_CNTL=0x%08X\n",
1088 		 RREG32(mmVM_CONTEXT0_CNTL));
1089 	dev_info(adev->dev, "  0x15D4=0x%08X\n",
1090 		 RREG32(0x575));
1091 	dev_info(adev->dev, "  0x15D8=0x%08X\n",
1092 		 RREG32(0x576));
1093 	dev_info(adev->dev, "  0x15DC=0x%08X\n",
1094 		 RREG32(0x577));
1095 	dev_info(adev->dev, "  VM_CONTEXT1_PAGE_TABLE_START_ADDR=0x%08X\n",
1096 		 RREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR));
1097 	dev_info(adev->dev, "  VM_CONTEXT1_PAGE_TABLE_END_ADDR=0x%08X\n",
1098 		 RREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR));
1099 	dev_info(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR=0x%08X\n",
1100 		 RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR));
1101 	dev_info(adev->dev, "  VM_CONTEXT1_CNTL2=0x%08X\n",
1102 		 RREG32(mmVM_CONTEXT1_CNTL2));
1103 	dev_info(adev->dev, "  VM_CONTEXT1_CNTL=0x%08X\n",
1104 		 RREG32(mmVM_CONTEXT1_CNTL));
1105 	for (i = 0; i < 16; i++) {
1106 		if (i < 8)
1107 			dev_info(adev->dev, "  VM_CONTEXT%d_PAGE_TABLE_BASE_ADDR=0x%08X\n",
1108 				 i, RREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i));
1109 		else
1110 			dev_info(adev->dev, "  VM_CONTEXT%d_PAGE_TABLE_BASE_ADDR=0x%08X\n",
1111 				 i, RREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8));
1112 	}
1113 	dev_info(adev->dev, "  MC_VM_SYSTEM_APERTURE_LOW_ADDR=0x%08X\n",
1114 		 RREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR));
1115 	dev_info(adev->dev, "  MC_VM_SYSTEM_APERTURE_HIGH_ADDR=0x%08X\n",
1116 		 RREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR));
1117 	dev_info(adev->dev, "  MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR=0x%08X\n",
1118 		 RREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR));
1119 	dev_info(adev->dev, "  MC_VM_FB_LOCATION=0x%08X\n",
1120 		 RREG32(mmMC_VM_FB_LOCATION));
1121 	dev_info(adev->dev, "  MC_VM_AGP_BASE=0x%08X\n",
1122 		 RREG32(mmMC_VM_AGP_BASE));
1123 	dev_info(adev->dev, "  MC_VM_AGP_TOP=0x%08X\n",
1124 		 RREG32(mmMC_VM_AGP_TOP));
1125 	dev_info(adev->dev, "  MC_VM_AGP_BOT=0x%08X\n",
1126 		 RREG32(mmMC_VM_AGP_BOT));
1127 
1128 	if (adev->asic_type == CHIP_KAVERI) {
1129 		dev_info(adev->dev, "  CHUB_CONTROL=0x%08X\n",
1130 			 RREG32(mmCHUB_CONTROL));
1131 	}
1132 
1133 	dev_info(adev->dev, "  HDP_REG_COHERENCY_FLUSH_CNTL=0x%08X\n",
1134 		 RREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL));
1135 	dev_info(adev->dev, "  HDP_NONSURFACE_BASE=0x%08X\n",
1136 		 RREG32(mmHDP_NONSURFACE_BASE));
1137 	dev_info(adev->dev, "  HDP_NONSURFACE_INFO=0x%08X\n",
1138 		 RREG32(mmHDP_NONSURFACE_INFO));
1139 	dev_info(adev->dev, "  HDP_NONSURFACE_SIZE=0x%08X\n",
1140 		 RREG32(mmHDP_NONSURFACE_SIZE));
1141 	dev_info(adev->dev, "  HDP_MISC_CNTL=0x%08X\n",
1142 		 RREG32(mmHDP_MISC_CNTL));
1143 	dev_info(adev->dev, "  HDP_HOST_PATH_CNTL=0x%08X\n",
1144 		 RREG32(mmHDP_HOST_PATH_CNTL));
1145 
1146 	for (i = 0, j = 0; i < 32; i++, j += 0x6) {
1147 		dev_info(adev->dev, "  %d:\n", i);
1148 		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1149 			 0xb05 + j, RREG32(0xb05 + j));
1150 		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1151 			 0xb06 + j, RREG32(0xb06 + j));
1152 		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1153 			 0xb07 + j, RREG32(0xb07 + j));
1154 		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1155 			 0xb08 + j, RREG32(0xb08 + j));
1156 		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1157 			 0xb09 + j, RREG32(0xb09 + j));
1158 	}
1159 
1160 	dev_info(adev->dev, "  BIF_FB_EN=0x%08X\n",
1161 		 RREG32(mmBIF_FB_EN));
1162 }
1163 
1164 static int gmc_v7_0_soft_reset(void *handle)
1165 {
1166 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1167 	struct amdgpu_mode_mc_save save;
1168 	u32 srbm_soft_reset = 0;
1169 	u32 tmp = RREG32(mmSRBM_STATUS);
1170 
1171 	if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1172 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1173 						SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1174 
1175 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1176 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1177 		if (!(adev->flags & AMD_IS_APU))
1178 			srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1179 							SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1180 	}
1181 
1182 	if (srbm_soft_reset) {
1183 		gmc_v7_0_print_status((void *)adev);
1184 
1185 		gmc_v7_0_mc_stop(adev, &save);
1186 		if (gmc_v7_0_wait_for_idle(adev)) {
1187 			dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1188 		}
1189 
1190 
1191 		tmp = RREG32(mmSRBM_SOFT_RESET);
1192 		tmp |= srbm_soft_reset;
1193 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1194 		WREG32(mmSRBM_SOFT_RESET, tmp);
1195 		tmp = RREG32(mmSRBM_SOFT_RESET);
1196 
1197 		udelay(50);
1198 
1199 		tmp &= ~srbm_soft_reset;
1200 		WREG32(mmSRBM_SOFT_RESET, tmp);
1201 		tmp = RREG32(mmSRBM_SOFT_RESET);
1202 
1203 		/* Wait a little for things to settle down */
1204 		udelay(50);
1205 
1206 		gmc_v7_0_mc_resume(adev, &save);
1207 		udelay(50);
1208 
1209 		gmc_v7_0_print_status((void *)adev);
1210 	}
1211 
1212 	return 0;
1213 }
1214 
1215 static int gmc_v7_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1216 					     struct amdgpu_irq_src *src,
1217 					     unsigned type,
1218 					     enum amdgpu_interrupt_state state)
1219 {
1220 	u32 tmp;
1221 	u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1222 		    VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1223 		    VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1224 		    VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1225 		    VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1226 		    VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1227 
1228 	switch (state) {
1229 	case AMDGPU_IRQ_STATE_DISABLE:
1230 		/* system context */
1231 		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1232 		tmp &= ~bits;
1233 		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1234 		/* VMs */
1235 		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1236 		tmp &= ~bits;
1237 		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1238 		break;
1239 	case AMDGPU_IRQ_STATE_ENABLE:
1240 		/* system context */
1241 		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1242 		tmp |= bits;
1243 		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1244 		/* VMs */
1245 		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1246 		tmp |= bits;
1247 		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1248 		break;
1249 	default:
1250 		break;
1251 	}
1252 
1253 	return 0;
1254 }
1255 
1256 static int gmc_v7_0_process_interrupt(struct amdgpu_device *adev,
1257 				      struct amdgpu_irq_src *source,
1258 				      struct amdgpu_iv_entry *entry)
1259 {
1260 	u32 addr, status, mc_client;
1261 
1262 	addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1263 	status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1264 	mc_client = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_MCCLIENT);
1265 	/* reset addr and status */
1266 	WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1267 
1268 	if (!addr && !status)
1269 		return 0;
1270 
1271 	dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1272 		entry->src_id, entry->src_data);
1273 	dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1274 		addr);
1275 	dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1276 		status);
1277 	gmc_v7_0_vm_decode_fault(adev, status, addr, mc_client);
1278 
1279 	return 0;
1280 }
1281 
1282 static int gmc_v7_0_set_clockgating_state(void *handle,
1283 					  enum amd_clockgating_state state)
1284 {
1285 	bool gate = false;
1286 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1287 
1288 	if (state == AMD_CG_STATE_GATE)
1289 		gate = true;
1290 
1291 	if (!(adev->flags & AMD_IS_APU)) {
1292 		gmc_v7_0_enable_mc_mgcg(adev, gate);
1293 		gmc_v7_0_enable_mc_ls(adev, gate);
1294 	}
1295 	gmc_v7_0_enable_bif_mgls(adev, gate);
1296 	gmc_v7_0_enable_hdp_mgcg(adev, gate);
1297 	gmc_v7_0_enable_hdp_ls(adev, gate);
1298 
1299 	return 0;
1300 }
1301 
1302 static int gmc_v7_0_set_powergating_state(void *handle,
1303 					  enum amd_powergating_state state)
1304 {
1305 	return 0;
1306 }
1307 
1308 const struct amd_ip_funcs gmc_v7_0_ip_funcs = {
1309 	.early_init = gmc_v7_0_early_init,
1310 	.late_init = gmc_v7_0_late_init,
1311 	.sw_init = gmc_v7_0_sw_init,
1312 	.sw_fini = gmc_v7_0_sw_fini,
1313 	.hw_init = gmc_v7_0_hw_init,
1314 	.hw_fini = gmc_v7_0_hw_fini,
1315 	.suspend = gmc_v7_0_suspend,
1316 	.resume = gmc_v7_0_resume,
1317 	.is_idle = gmc_v7_0_is_idle,
1318 	.wait_for_idle = gmc_v7_0_wait_for_idle,
1319 	.soft_reset = gmc_v7_0_soft_reset,
1320 	.print_status = gmc_v7_0_print_status,
1321 	.set_clockgating_state = gmc_v7_0_set_clockgating_state,
1322 	.set_powergating_state = gmc_v7_0_set_powergating_state,
1323 };
1324 
1325 static const struct amdgpu_gart_funcs gmc_v7_0_gart_funcs = {
1326 	.flush_gpu_tlb = gmc_v7_0_gart_flush_gpu_tlb,
1327 	.set_pte_pde = gmc_v7_0_gart_set_pte_pde,
1328 };
1329 
1330 static const struct amdgpu_irq_src_funcs gmc_v7_0_irq_funcs = {
1331 	.set = gmc_v7_0_vm_fault_interrupt_state,
1332 	.process = gmc_v7_0_process_interrupt,
1333 };
1334 
1335 static void gmc_v7_0_set_gart_funcs(struct amdgpu_device *adev)
1336 {
1337 	if (adev->gart.gart_funcs == NULL)
1338 		adev->gart.gart_funcs = &gmc_v7_0_gart_funcs;
1339 }
1340 
1341 static void gmc_v7_0_set_irq_funcs(struct amdgpu_device *adev)
1342 {
1343 	adev->mc.vm_fault.num_types = 1;
1344 	adev->mc.vm_fault.funcs = &gmc_v7_0_irq_funcs;
1345 }
1346