xref: /linux/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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 
24 #include <linux/firmware.h>
25 #include <linux/module.h>
26 #include <linux/pci.h>
27 
28 #include <drm/drm_cache.h>
29 #include "amdgpu.h"
30 #include "gmc_v8_0.h"
31 #include "amdgpu_ucode.h"
32 #include "amdgpu_amdkfd.h"
33 #include "amdgpu_gem.h"
34 
35 #include "gmc/gmc_8_1_d.h"
36 #include "gmc/gmc_8_1_sh_mask.h"
37 
38 #include "bif/bif_5_0_d.h"
39 #include "bif/bif_5_0_sh_mask.h"
40 
41 #include "oss/oss_3_0_d.h"
42 #include "oss/oss_3_0_sh_mask.h"
43 
44 #include "dce/dce_10_0_d.h"
45 #include "dce/dce_10_0_sh_mask.h"
46 
47 #include "vid.h"
48 #include "vi.h"
49 
50 #include "amdgpu_atombios.h"
51 
52 #include "ivsrcid/ivsrcid_vislands30.h"
53 
54 static void gmc_v8_0_set_gmc_funcs(struct amdgpu_device *adev);
55 static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev);
56 static int gmc_v8_0_wait_for_idle(void *handle);
57 
58 MODULE_FIRMWARE("amdgpu/tonga_mc.bin");
59 MODULE_FIRMWARE("amdgpu/polaris11_mc.bin");
60 MODULE_FIRMWARE("amdgpu/polaris10_mc.bin");
61 MODULE_FIRMWARE("amdgpu/polaris12_mc.bin");
62 MODULE_FIRMWARE("amdgpu/polaris12_32_mc.bin");
63 MODULE_FIRMWARE("amdgpu/polaris11_k_mc.bin");
64 MODULE_FIRMWARE("amdgpu/polaris10_k_mc.bin");
65 MODULE_FIRMWARE("amdgpu/polaris12_k_mc.bin");
66 
67 static const u32 golden_settings_tonga_a11[] = {
68 	mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
69 	mmMC_HUB_RDREQ_DMIF_LIMIT, 0x0000007f, 0x00000028,
70 	mmMC_HUB_WDP_UMC, 0x00007fb6, 0x00000991,
71 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
72 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
73 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
74 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
75 };
76 
77 static const u32 tonga_mgcg_cgcg_init[] = {
78 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
79 };
80 
81 static const u32 golden_settings_fiji_a10[] = {
82 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
83 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
84 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
85 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
86 };
87 
88 static const u32 fiji_mgcg_cgcg_init[] = {
89 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
90 };
91 
92 static const u32 golden_settings_polaris11_a11[] = {
93 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
94 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
95 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
96 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff
97 };
98 
99 static const u32 golden_settings_polaris10_a11[] = {
100 	mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
101 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
102 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
103 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
104 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff
105 };
106 
107 static const u32 cz_mgcg_cgcg_init[] = {
108 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
109 };
110 
111 static const u32 stoney_mgcg_cgcg_init[] = {
112 	mmATC_MISC_CG, 0xffffffff, 0x000c0200,
113 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
114 };
115 
116 static const u32 golden_settings_stoney_common[] = {
117 	mmMC_HUB_RDREQ_UVD, MC_HUB_RDREQ_UVD__PRESCALE_MASK, 0x00000004,
118 	mmMC_RD_GRP_OTH, MC_RD_GRP_OTH__UVD_MASK, 0x00600000
119 };
120 
121 static void gmc_v8_0_init_golden_registers(struct amdgpu_device *adev)
122 {
123 	switch (adev->asic_type) {
124 	case CHIP_FIJI:
125 		amdgpu_device_program_register_sequence(adev,
126 							fiji_mgcg_cgcg_init,
127 							ARRAY_SIZE(fiji_mgcg_cgcg_init));
128 		amdgpu_device_program_register_sequence(adev,
129 							golden_settings_fiji_a10,
130 							ARRAY_SIZE(golden_settings_fiji_a10));
131 		break;
132 	case CHIP_TONGA:
133 		amdgpu_device_program_register_sequence(adev,
134 							tonga_mgcg_cgcg_init,
135 							ARRAY_SIZE(tonga_mgcg_cgcg_init));
136 		amdgpu_device_program_register_sequence(adev,
137 							golden_settings_tonga_a11,
138 							ARRAY_SIZE(golden_settings_tonga_a11));
139 		break;
140 	case CHIP_POLARIS11:
141 	case CHIP_POLARIS12:
142 	case CHIP_VEGAM:
143 		amdgpu_device_program_register_sequence(adev,
144 							golden_settings_polaris11_a11,
145 							ARRAY_SIZE(golden_settings_polaris11_a11));
146 		break;
147 	case CHIP_POLARIS10:
148 		amdgpu_device_program_register_sequence(adev,
149 							golden_settings_polaris10_a11,
150 							ARRAY_SIZE(golden_settings_polaris10_a11));
151 		break;
152 	case CHIP_CARRIZO:
153 		amdgpu_device_program_register_sequence(adev,
154 							cz_mgcg_cgcg_init,
155 							ARRAY_SIZE(cz_mgcg_cgcg_init));
156 		break;
157 	case CHIP_STONEY:
158 		amdgpu_device_program_register_sequence(adev,
159 							stoney_mgcg_cgcg_init,
160 							ARRAY_SIZE(stoney_mgcg_cgcg_init));
161 		amdgpu_device_program_register_sequence(adev,
162 							golden_settings_stoney_common,
163 							ARRAY_SIZE(golden_settings_stoney_common));
164 		break;
165 	default:
166 		break;
167 	}
168 }
169 
170 static void gmc_v8_0_mc_stop(struct amdgpu_device *adev)
171 {
172 	u32 blackout;
173 
174 	gmc_v8_0_wait_for_idle(adev);
175 
176 	blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
177 	if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
178 		/* Block CPU access */
179 		WREG32(mmBIF_FB_EN, 0);
180 		/* blackout the MC */
181 		blackout = REG_SET_FIELD(blackout,
182 					 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 1);
183 		WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout);
184 	}
185 	/* wait for the MC to settle */
186 	udelay(100);
187 }
188 
189 static void gmc_v8_0_mc_resume(struct amdgpu_device *adev)
190 {
191 	u32 tmp;
192 
193 	/* unblackout the MC */
194 	tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
195 	tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
196 	WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
197 	/* allow CPU access */
198 	tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
199 	tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
200 	WREG32(mmBIF_FB_EN, tmp);
201 }
202 
203 /**
204  * gmc_v8_0_init_microcode - load ucode images from disk
205  *
206  * @adev: amdgpu_device pointer
207  *
208  * Use the firmware interface to load the ucode images into
209  * the driver (not loaded into hw).
210  * Returns 0 on success, error on failure.
211  */
212 static int gmc_v8_0_init_microcode(struct amdgpu_device *adev)
213 {
214 	const char *chip_name;
215 	int err;
216 
217 	DRM_DEBUG("\n");
218 
219 	switch (adev->asic_type) {
220 	case CHIP_TONGA:
221 		chip_name = "tonga";
222 		break;
223 	case CHIP_POLARIS11:
224 		if (ASICID_IS_P21(adev->pdev->device, adev->pdev->revision) ||
225 		    ASICID_IS_P31(adev->pdev->device, adev->pdev->revision))
226 			chip_name = "polaris11_k";
227 		else
228 			chip_name = "polaris11";
229 		break;
230 	case CHIP_POLARIS10:
231 		if (ASICID_IS_P30(adev->pdev->device, adev->pdev->revision))
232 			chip_name = "polaris10_k";
233 		else
234 			chip_name = "polaris10";
235 		break;
236 	case CHIP_POLARIS12:
237 		if (ASICID_IS_P23(adev->pdev->device, adev->pdev->revision)) {
238 			chip_name = "polaris12_k";
239 		} else {
240 			WREG32(mmMC_SEQ_IO_DEBUG_INDEX, ixMC_IO_DEBUG_UP_159);
241 			/* Polaris12 32bit ASIC needs a special MC firmware */
242 			if (RREG32(mmMC_SEQ_IO_DEBUG_DATA) == 0x05b4dc40)
243 				chip_name = "polaris12_32";
244 			else
245 				chip_name = "polaris12";
246 		}
247 		break;
248 	case CHIP_FIJI:
249 	case CHIP_CARRIZO:
250 	case CHIP_STONEY:
251 	case CHIP_VEGAM:
252 		return 0;
253 	default:
254 		return -EINVAL;
255 	}
256 
257 	err = amdgpu_ucode_request(adev, &adev->gmc.fw, "amdgpu/%s_mc.bin", chip_name);
258 	if (err) {
259 		pr_err("mc: Failed to load firmware \"%s_mc.bin\"\n", chip_name);
260 		amdgpu_ucode_release(&adev->gmc.fw);
261 	}
262 	return err;
263 }
264 
265 /**
266  * gmc_v8_0_tonga_mc_load_microcode - load tonga MC ucode into the hw
267  *
268  * @adev: amdgpu_device pointer
269  *
270  * Load the GDDR MC ucode into the hw (VI).
271  * Returns 0 on success, error on failure.
272  */
273 static int gmc_v8_0_tonga_mc_load_microcode(struct amdgpu_device *adev)
274 {
275 	const struct mc_firmware_header_v1_0 *hdr;
276 	const __le32 *fw_data = NULL;
277 	const __le32 *io_mc_regs = NULL;
278 	u32 running;
279 	int i, ucode_size, regs_size;
280 
281 	/* Skip MC ucode loading on SR-IOV capable boards.
282 	 * vbios does this for us in asic_init in that case.
283 	 * Skip MC ucode loading on VF, because hypervisor will do that
284 	 * for this adaptor.
285 	 */
286 	if (amdgpu_sriov_bios(adev))
287 		return 0;
288 
289 	if (!adev->gmc.fw)
290 		return -EINVAL;
291 
292 	hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
293 	amdgpu_ucode_print_mc_hdr(&hdr->header);
294 
295 	adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
296 	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
297 	io_mc_regs = (const __le32 *)
298 		(adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
299 	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
300 	fw_data = (const __le32 *)
301 		(adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
302 
303 	running = REG_GET_FIELD(RREG32(mmMC_SEQ_SUP_CNTL), MC_SEQ_SUP_CNTL, RUN);
304 
305 	if (running == 0) {
306 		/* reset the engine and set to writable */
307 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
308 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
309 
310 		/* load mc io regs */
311 		for (i = 0; i < regs_size; i++) {
312 			WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
313 			WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
314 		}
315 		/* load the MC ucode */
316 		for (i = 0; i < ucode_size; i++)
317 			WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
318 
319 		/* put the engine back into the active state */
320 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
321 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
322 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
323 
324 		/* wait for training to complete */
325 		for (i = 0; i < adev->usec_timeout; i++) {
326 			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
327 					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D0))
328 				break;
329 			udelay(1);
330 		}
331 		for (i = 0; i < adev->usec_timeout; i++) {
332 			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
333 					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D1))
334 				break;
335 			udelay(1);
336 		}
337 	}
338 
339 	return 0;
340 }
341 
342 static int gmc_v8_0_polaris_mc_load_microcode(struct amdgpu_device *adev)
343 {
344 	const struct mc_firmware_header_v1_0 *hdr;
345 	const __le32 *fw_data = NULL;
346 	const __le32 *io_mc_regs = NULL;
347 	u32 data;
348 	int i, ucode_size, regs_size;
349 
350 	/* Skip MC ucode loading on SR-IOV capable boards.
351 	 * vbios does this for us in asic_init in that case.
352 	 * Skip MC ucode loading on VF, because hypervisor will do that
353 	 * for this adaptor.
354 	 */
355 	if (amdgpu_sriov_bios(adev))
356 		return 0;
357 
358 	if (!adev->gmc.fw)
359 		return -EINVAL;
360 
361 	hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
362 	amdgpu_ucode_print_mc_hdr(&hdr->header);
363 
364 	adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
365 	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
366 	io_mc_regs = (const __le32 *)
367 		(adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
368 	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
369 	fw_data = (const __le32 *)
370 		(adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
371 
372 	data = RREG32(mmMC_SEQ_MISC0);
373 	data &= ~(0x40);
374 	WREG32(mmMC_SEQ_MISC0, data);
375 
376 	/* load mc io regs */
377 	for (i = 0; i < regs_size; i++) {
378 		WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
379 		WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
380 	}
381 
382 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
383 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
384 
385 	/* load the MC ucode */
386 	for (i = 0; i < ucode_size; i++)
387 		WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
388 
389 	/* put the engine back into the active state */
390 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
391 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
392 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
393 
394 	/* wait for training to complete */
395 	for (i = 0; i < adev->usec_timeout; i++) {
396 		data = RREG32(mmMC_SEQ_MISC0);
397 		if (data & 0x80)
398 			break;
399 		udelay(1);
400 	}
401 
402 	return 0;
403 }
404 
405 static void gmc_v8_0_vram_gtt_location(struct amdgpu_device *adev,
406 				       struct amdgpu_gmc *mc)
407 {
408 	u64 base = 0;
409 
410 	if (!amdgpu_sriov_vf(adev))
411 		base = RREG32(mmMC_VM_FB_LOCATION) & 0xFFFF;
412 	base <<= 24;
413 
414 	amdgpu_gmc_set_agp_default(adev, mc);
415 	amdgpu_gmc_vram_location(adev, mc, base);
416 	amdgpu_gmc_gart_location(adev, mc, AMDGPU_GART_PLACEMENT_BEST_FIT);
417 }
418 
419 /**
420  * gmc_v8_0_mc_program - program the GPU memory controller
421  *
422  * @adev: amdgpu_device pointer
423  *
424  * Set the location of vram, gart, and AGP in the GPU's
425  * physical address space (VI).
426  */
427 static void gmc_v8_0_mc_program(struct amdgpu_device *adev)
428 {
429 	u32 tmp;
430 	int i, j;
431 
432 	/* Initialize HDP */
433 	for (i = 0, j = 0; i < 32; i++, j += 0x6) {
434 		WREG32((0xb05 + j), 0x00000000);
435 		WREG32((0xb06 + j), 0x00000000);
436 		WREG32((0xb07 + j), 0x00000000);
437 		WREG32((0xb08 + j), 0x00000000);
438 		WREG32((0xb09 + j), 0x00000000);
439 	}
440 	WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
441 
442 	if (gmc_v8_0_wait_for_idle((void *)adev))
443 		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
444 
445 	if (adev->mode_info.num_crtc) {
446 		/* Lockout access through VGA aperture*/
447 		tmp = RREG32(mmVGA_HDP_CONTROL);
448 		tmp = REG_SET_FIELD(tmp, VGA_HDP_CONTROL, VGA_MEMORY_DISABLE, 1);
449 		WREG32(mmVGA_HDP_CONTROL, tmp);
450 
451 		/* disable VGA render */
452 		tmp = RREG32(mmVGA_RENDER_CONTROL);
453 		tmp = REG_SET_FIELD(tmp, VGA_RENDER_CONTROL, VGA_VSTATUS_CNTL, 0);
454 		WREG32(mmVGA_RENDER_CONTROL, tmp);
455 	}
456 	/* Update configuration */
457 	WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
458 	       adev->gmc.vram_start >> 12);
459 	WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
460 	       adev->gmc.vram_end >> 12);
461 	WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
462 	       adev->mem_scratch.gpu_addr >> 12);
463 
464 	if (amdgpu_sriov_vf(adev)) {
465 		tmp = ((adev->gmc.vram_end >> 24) & 0xFFFF) << 16;
466 		tmp |= ((adev->gmc.vram_start >> 24) & 0xFFFF);
467 		WREG32(mmMC_VM_FB_LOCATION, tmp);
468 		/* XXX double check these! */
469 		WREG32(mmHDP_NONSURFACE_BASE, (adev->gmc.vram_start >> 8));
470 		WREG32(mmHDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
471 		WREG32(mmHDP_NONSURFACE_SIZE, 0x3FFFFFFF);
472 	}
473 
474 	WREG32(mmMC_VM_AGP_BASE, 0);
475 	WREG32(mmMC_VM_AGP_TOP, adev->gmc.agp_end >> 22);
476 	WREG32(mmMC_VM_AGP_BOT, adev->gmc.agp_start >> 22);
477 	if (gmc_v8_0_wait_for_idle((void *)adev))
478 		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
479 
480 	WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK);
481 
482 	tmp = RREG32(mmHDP_MISC_CNTL);
483 	tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 0);
484 	WREG32(mmHDP_MISC_CNTL, tmp);
485 
486 	tmp = RREG32(mmHDP_HOST_PATH_CNTL);
487 	WREG32(mmHDP_HOST_PATH_CNTL, tmp);
488 }
489 
490 /**
491  * gmc_v8_0_mc_init - initialize the memory controller driver params
492  *
493  * @adev: amdgpu_device pointer
494  *
495  * Look up the amount of vram, vram width, and decide how to place
496  * vram and gart within the GPU's physical address space (VI).
497  * Returns 0 for success.
498  */
499 static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
500 {
501 	int r;
502 	u32 tmp;
503 
504 	adev->gmc.vram_width = amdgpu_atombios_get_vram_width(adev);
505 	if (!adev->gmc.vram_width) {
506 		int chansize, numchan;
507 
508 		/* Get VRAM informations */
509 		tmp = RREG32(mmMC_ARB_RAMCFG);
510 		if (REG_GET_FIELD(tmp, MC_ARB_RAMCFG, CHANSIZE))
511 			chansize = 64;
512 		else
513 			chansize = 32;
514 
515 		tmp = RREG32(mmMC_SHARED_CHMAP);
516 		switch (REG_GET_FIELD(tmp, MC_SHARED_CHMAP, NOOFCHAN)) {
517 		case 0:
518 		default:
519 			numchan = 1;
520 			break;
521 		case 1:
522 			numchan = 2;
523 			break;
524 		case 2:
525 			numchan = 4;
526 			break;
527 		case 3:
528 			numchan = 8;
529 			break;
530 		case 4:
531 			numchan = 3;
532 			break;
533 		case 5:
534 			numchan = 6;
535 			break;
536 		case 6:
537 			numchan = 10;
538 			break;
539 		case 7:
540 			numchan = 12;
541 			break;
542 		case 8:
543 			numchan = 16;
544 			break;
545 		}
546 		adev->gmc.vram_width = numchan * chansize;
547 	}
548 	/* size in MB on si */
549 	tmp = RREG32(mmCONFIG_MEMSIZE);
550 	/* some boards may have garbage in the upper 16 bits */
551 	if (tmp & 0xffff0000) {
552 		DRM_INFO("Probable bad vram size: 0x%08x\n", tmp);
553 		if (tmp & 0xffff)
554 			tmp &= 0xffff;
555 	}
556 	adev->gmc.mc_vram_size = tmp * 1024ULL * 1024ULL;
557 	adev->gmc.real_vram_size = adev->gmc.mc_vram_size;
558 
559 	if (!(adev->flags & AMD_IS_APU)) {
560 		r = amdgpu_device_resize_fb_bar(adev);
561 		if (r)
562 			return r;
563 	}
564 	adev->gmc.aper_base = pci_resource_start(adev->pdev, 0);
565 	adev->gmc.aper_size = pci_resource_len(adev->pdev, 0);
566 
567 #ifdef CONFIG_X86_64
568 	if ((adev->flags & AMD_IS_APU) && !amdgpu_passthrough(adev)) {
569 		adev->gmc.aper_base = ((u64)RREG32(mmMC_VM_FB_OFFSET)) << 22;
570 		adev->gmc.aper_size = adev->gmc.real_vram_size;
571 	}
572 #endif
573 
574 	adev->gmc.visible_vram_size = adev->gmc.aper_size;
575 
576 	/* set the gart size */
577 	if (amdgpu_gart_size == -1) {
578 		switch (adev->asic_type) {
579 		case CHIP_POLARIS10: /* all engines support GPUVM */
580 		case CHIP_POLARIS11: /* all engines support GPUVM */
581 		case CHIP_POLARIS12: /* all engines support GPUVM */
582 		case CHIP_VEGAM:     /* all engines support GPUVM */
583 		default:
584 			adev->gmc.gart_size = 256ULL << 20;
585 			break;
586 		case CHIP_TONGA:   /* UVD, VCE do not support GPUVM */
587 		case CHIP_FIJI:    /* UVD, VCE do not support GPUVM */
588 		case CHIP_CARRIZO: /* UVD, VCE do not support GPUVM, DCE SG support */
589 		case CHIP_STONEY:  /* UVD does not support GPUVM, DCE SG support */
590 			adev->gmc.gart_size = 1024ULL << 20;
591 			break;
592 		}
593 	} else {
594 		adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
595 	}
596 
597 	adev->gmc.gart_size += adev->pm.smu_prv_buffer_size;
598 	gmc_v8_0_vram_gtt_location(adev, &adev->gmc);
599 
600 	return 0;
601 }
602 
603 /**
604  * gmc_v8_0_flush_gpu_tlb_pasid - tlb flush via pasid
605  *
606  * @adev: amdgpu_device pointer
607  * @pasid: pasid to be flush
608  * @flush_type: type of flush
609  * @all_hub: flush all hubs
610  * @inst: is used to select which instance of KIQ to use for the invalidation
611  *
612  * Flush the TLB for the requested pasid.
613  */
614 static void gmc_v8_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
615 					 uint16_t pasid, uint32_t flush_type,
616 					 bool all_hub, uint32_t inst)
617 {
618 	u32 mask = 0x0;
619 	int vmid;
620 
621 	for (vmid = 1; vmid < 16; vmid++) {
622 		u32 tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
623 
624 		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
625 		    (tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid)
626 			mask |= 1 << vmid;
627 	}
628 
629 	WREG32(mmVM_INVALIDATE_REQUEST, mask);
630 	RREG32(mmVM_INVALIDATE_RESPONSE);
631 }
632 
633 /*
634  * GART
635  * VMID 0 is the physical GPU addresses as used by the kernel.
636  * VMIDs 1-15 are used for userspace clients and are handled
637  * by the amdgpu vm/hsa code.
638  */
639 
640 /**
641  * gmc_v8_0_flush_gpu_tlb - gart tlb flush callback
642  *
643  * @adev: amdgpu_device pointer
644  * @vmid: vm instance to flush
645  * @vmhub: which hub to flush
646  * @flush_type: type of flush
647  *
648  * Flush the TLB for the requested page table (VI).
649  */
650 static void gmc_v8_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
651 					uint32_t vmhub, uint32_t flush_type)
652 {
653 	/* bits 0-15 are the VM contexts0-15 */
654 	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
655 }
656 
657 static uint64_t gmc_v8_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
658 					    unsigned int vmid, uint64_t pd_addr)
659 {
660 	uint32_t reg;
661 
662 	if (vmid < 8)
663 		reg = mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid;
664 	else
665 		reg = mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8;
666 	amdgpu_ring_emit_wreg(ring, reg, pd_addr >> 12);
667 
668 	/* bits 0-15 are the VM contexts0-15 */
669 	amdgpu_ring_emit_wreg(ring, mmVM_INVALIDATE_REQUEST, 1 << vmid);
670 
671 	return pd_addr;
672 }
673 
674 static void gmc_v8_0_emit_pasid_mapping(struct amdgpu_ring *ring, unsigned int vmid,
675 					unsigned int pasid)
676 {
677 	amdgpu_ring_emit_wreg(ring, mmIH_VMID_0_LUT + vmid, pasid);
678 }
679 
680 /*
681  * PTE format on VI:
682  * 63:40 reserved
683  * 39:12 4k physical page base address
684  * 11:7 fragment
685  * 6 write
686  * 5 read
687  * 4 exe
688  * 3 reserved
689  * 2 snooped
690  * 1 system
691  * 0 valid
692  *
693  * PDE format on VI:
694  * 63:59 block fragment size
695  * 58:40 reserved
696  * 39:1 physical base address of PTE
697  * bits 5:1 must be 0.
698  * 0 valid
699  */
700 
701 static void gmc_v8_0_get_vm_pde(struct amdgpu_device *adev, int level,
702 				uint64_t *addr, uint64_t *flags)
703 {
704 	BUG_ON(*addr & 0xFFFFFF0000000FFFULL);
705 }
706 
707 static void gmc_v8_0_get_vm_pte(struct amdgpu_device *adev,
708 				struct amdgpu_bo_va_mapping *mapping,
709 				uint64_t *flags)
710 {
711 	*flags &= ~AMDGPU_PTE_EXECUTABLE;
712 	*flags |= mapping->flags & AMDGPU_PTE_EXECUTABLE;
713 	*flags &= ~AMDGPU_PTE_PRT;
714 }
715 
716 /**
717  * gmc_v8_0_set_fault_enable_default - update VM fault handling
718  *
719  * @adev: amdgpu_device pointer
720  * @value: true redirects VM faults to the default page
721  */
722 static void gmc_v8_0_set_fault_enable_default(struct amdgpu_device *adev,
723 					      bool value)
724 {
725 	u32 tmp;
726 
727 	tmp = RREG32(mmVM_CONTEXT1_CNTL);
728 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
729 			    RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
730 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
731 			    DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
732 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
733 			    PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
734 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
735 			    VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
736 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
737 			    READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
738 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
739 			    WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
740 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
741 			    EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
742 	WREG32(mmVM_CONTEXT1_CNTL, tmp);
743 }
744 
745 /**
746  * gmc_v8_0_set_prt() - set PRT VM fault
747  *
748  * @adev: amdgpu_device pointer
749  * @enable: enable/disable VM fault handling for PRT
750  */
751 static void gmc_v8_0_set_prt(struct amdgpu_device *adev, bool enable)
752 {
753 	u32 tmp;
754 
755 	if (enable && !adev->gmc.prt_warning) {
756 		dev_warn(adev->dev, "Disabling VM faults because of PRT request!\n");
757 		adev->gmc.prt_warning = true;
758 	}
759 
760 	tmp = RREG32(mmVM_PRT_CNTL);
761 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
762 			    CB_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
763 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
764 			    CB_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
765 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
766 			    TC_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
767 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
768 			    TC_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
769 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
770 			    L2_CACHE_STORE_INVALID_ENTRIES, enable);
771 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
772 			    L1_TLB_STORE_INVALID_ENTRIES, enable);
773 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
774 			    MASK_PDE0_FAULT, enable);
775 	WREG32(mmVM_PRT_CNTL, tmp);
776 
777 	if (enable) {
778 		uint32_t low = AMDGPU_VA_RESERVED_BOTTOM >>
779 			AMDGPU_GPU_PAGE_SHIFT;
780 		uint32_t high = adev->vm_manager.max_pfn -
781 			(AMDGPU_VA_RESERVED_TOP >> AMDGPU_GPU_PAGE_SHIFT);
782 
783 		WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, low);
784 		WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, low);
785 		WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, low);
786 		WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, low);
787 		WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, high);
788 		WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, high);
789 		WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, high);
790 		WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, high);
791 	} else {
792 		WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, 0xfffffff);
793 		WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, 0xfffffff);
794 		WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, 0xfffffff);
795 		WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, 0xfffffff);
796 		WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, 0x0);
797 		WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, 0x0);
798 		WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, 0x0);
799 		WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, 0x0);
800 	}
801 }
802 
803 /**
804  * gmc_v8_0_gart_enable - gart enable
805  *
806  * @adev: amdgpu_device pointer
807  *
808  * This sets up the TLBs, programs the page tables for VMID0,
809  * sets up the hw for VMIDs 1-15 which are allocated on
810  * demand, and sets up the global locations for the LDS, GDS,
811  * and GPUVM for FSA64 clients (VI).
812  * Returns 0 for success, errors for failure.
813  */
814 static int gmc_v8_0_gart_enable(struct amdgpu_device *adev)
815 {
816 	uint64_t table_addr;
817 	u32 tmp, field;
818 	int i;
819 
820 	if (adev->gart.bo == NULL) {
821 		dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
822 		return -EINVAL;
823 	}
824 	amdgpu_gtt_mgr_recover(&adev->mman.gtt_mgr);
825 	table_addr = amdgpu_bo_gpu_offset(adev->gart.bo);
826 
827 	/* Setup TLB control */
828 	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
829 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 1);
830 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 1);
831 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE, 3);
832 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 1);
833 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
834 	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
835 	/* Setup L2 cache */
836 	tmp = RREG32(mmVM_L2_CNTL);
837 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
838 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
839 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE, 1);
840 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1);
841 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7);
842 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
843 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1);
844 	WREG32(mmVM_L2_CNTL, tmp);
845 	tmp = RREG32(mmVM_L2_CNTL2);
846 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
847 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
848 	WREG32(mmVM_L2_CNTL2, tmp);
849 
850 	field = adev->vm_manager.fragment_size;
851 	tmp = RREG32(mmVM_L2_CNTL3);
852 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY, 1);
853 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, field);
854 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_FRAGMENT_SIZE, field);
855 	WREG32(mmVM_L2_CNTL3, tmp);
856 	/* XXX: set to enable PTE/PDE in system memory */
857 	tmp = RREG32(mmVM_L2_CNTL4);
858 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_PHYSICAL, 0);
859 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SHARED, 0);
860 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SNOOP, 0);
861 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_PHYSICAL, 0);
862 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SHARED, 0);
863 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SNOOP, 0);
864 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_PHYSICAL, 0);
865 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SHARED, 0);
866 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SNOOP, 0);
867 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_PHYSICAL, 0);
868 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SHARED, 0);
869 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SNOOP, 0);
870 	WREG32(mmVM_L2_CNTL4, tmp);
871 	/* setup context0 */
872 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->gmc.gart_start >> 12);
873 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->gmc.gart_end >> 12);
874 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, table_addr >> 12);
875 	WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
876 			(u32)(adev->dummy_page_addr >> 12));
877 	WREG32(mmVM_CONTEXT0_CNTL2, 0);
878 	tmp = RREG32(mmVM_CONTEXT0_CNTL);
879 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
880 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH, 0);
881 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
882 	WREG32(mmVM_CONTEXT0_CNTL, tmp);
883 
884 	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR, 0);
885 	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR, 0);
886 	WREG32(mmVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET, 0);
887 
888 	/* empty context1-15 */
889 	/* FIXME start with 4G, once using 2 level pt switch to full
890 	 * vm size space
891 	 */
892 	/* set vm size, must be a multiple of 4 */
893 	WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
894 	WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
895 	for (i = 1; i < AMDGPU_NUM_VMID; i++) {
896 		if (i < 8)
897 			WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
898 			       table_addr >> 12);
899 		else
900 			WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
901 			       table_addr >> 12);
902 	}
903 
904 	/* enable context1-15 */
905 	WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
906 	       (u32)(adev->dummy_page_addr >> 12));
907 	WREG32(mmVM_CONTEXT1_CNTL2, 4);
908 	tmp = RREG32(mmVM_CONTEXT1_CNTL);
909 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
910 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH, 1);
911 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
912 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
913 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
914 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
915 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
916 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
917 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
918 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE,
919 			    adev->vm_manager.block_size - 9);
920 	WREG32(mmVM_CONTEXT1_CNTL, tmp);
921 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
922 		gmc_v8_0_set_fault_enable_default(adev, false);
923 	else
924 		gmc_v8_0_set_fault_enable_default(adev, true);
925 
926 	gmc_v8_0_flush_gpu_tlb(adev, 0, 0, 0);
927 	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
928 		 (unsigned int)(adev->gmc.gart_size >> 20),
929 		 (unsigned long long)table_addr);
930 	return 0;
931 }
932 
933 static int gmc_v8_0_gart_init(struct amdgpu_device *adev)
934 {
935 	int r;
936 
937 	if (adev->gart.bo) {
938 		WARN(1, "R600 PCIE GART already initialized\n");
939 		return 0;
940 	}
941 	/* Initialize common gart structure */
942 	r = amdgpu_gart_init(adev);
943 	if (r)
944 		return r;
945 	adev->gart.table_size = adev->gart.num_gpu_pages * 8;
946 	adev->gart.gart_pte_flags = AMDGPU_PTE_EXECUTABLE;
947 	return amdgpu_gart_table_vram_alloc(adev);
948 }
949 
950 /**
951  * gmc_v8_0_gart_disable - gart disable
952  *
953  * @adev: amdgpu_device pointer
954  *
955  * This disables all VM page table (VI).
956  */
957 static void gmc_v8_0_gart_disable(struct amdgpu_device *adev)
958 {
959 	u32 tmp;
960 
961 	/* Disable all tables */
962 	WREG32(mmVM_CONTEXT0_CNTL, 0);
963 	WREG32(mmVM_CONTEXT1_CNTL, 0);
964 	/* Setup TLB control */
965 	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
966 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
967 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 0);
968 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 0);
969 	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
970 	/* Setup L2 cache */
971 	tmp = RREG32(mmVM_L2_CNTL);
972 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
973 	WREG32(mmVM_L2_CNTL, tmp);
974 	WREG32(mmVM_L2_CNTL2, 0);
975 }
976 
977 /**
978  * gmc_v8_0_vm_decode_fault - print human readable fault info
979  *
980  * @adev: amdgpu_device pointer
981  * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
982  * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
983  * @mc_client: VM_CONTEXT1_PROTECTION_FAULT_MCCLIENT register value
984  * @pasid: debug logging only - no functional use
985  *
986  * Print human readable fault information (VI).
987  */
988 static void gmc_v8_0_vm_decode_fault(struct amdgpu_device *adev, u32 status,
989 				     u32 addr, u32 mc_client, unsigned int pasid)
990 {
991 	u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
992 	u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
993 					PROTECTIONS);
994 	char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
995 		(mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
996 	u32 mc_id;
997 
998 	mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
999 			      MEMORY_CLIENT_ID);
1000 
1001 	dev_err(adev->dev, "VM fault (0x%02x, vmid %d, pasid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
1002 	       protections, vmid, pasid, addr,
1003 	       REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1004 			     MEMORY_CLIENT_RW) ?
1005 	       "write" : "read", block, mc_client, mc_id);
1006 }
1007 
1008 static int gmc_v8_0_convert_vram_type(int mc_seq_vram_type)
1009 {
1010 	switch (mc_seq_vram_type) {
1011 	case MC_SEQ_MISC0__MT__GDDR1:
1012 		return AMDGPU_VRAM_TYPE_GDDR1;
1013 	case MC_SEQ_MISC0__MT__DDR2:
1014 		return AMDGPU_VRAM_TYPE_DDR2;
1015 	case MC_SEQ_MISC0__MT__GDDR3:
1016 		return AMDGPU_VRAM_TYPE_GDDR3;
1017 	case MC_SEQ_MISC0__MT__GDDR4:
1018 		return AMDGPU_VRAM_TYPE_GDDR4;
1019 	case MC_SEQ_MISC0__MT__GDDR5:
1020 		return AMDGPU_VRAM_TYPE_GDDR5;
1021 	case MC_SEQ_MISC0__MT__HBM:
1022 		return AMDGPU_VRAM_TYPE_HBM;
1023 	case MC_SEQ_MISC0__MT__DDR3:
1024 		return AMDGPU_VRAM_TYPE_DDR3;
1025 	default:
1026 		return AMDGPU_VRAM_TYPE_UNKNOWN;
1027 	}
1028 }
1029 
1030 static int gmc_v8_0_early_init(void *handle)
1031 {
1032 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1033 
1034 	gmc_v8_0_set_gmc_funcs(adev);
1035 	gmc_v8_0_set_irq_funcs(adev);
1036 
1037 	adev->gmc.shared_aperture_start = 0x2000000000000000ULL;
1038 	adev->gmc.shared_aperture_end =
1039 		adev->gmc.shared_aperture_start + (4ULL << 30) - 1;
1040 	adev->gmc.private_aperture_start =
1041 		adev->gmc.shared_aperture_end + 1;
1042 	adev->gmc.private_aperture_end =
1043 		adev->gmc.private_aperture_start + (4ULL << 30) - 1;
1044 	adev->gmc.noretry_flags = AMDGPU_VM_NORETRY_FLAGS_TF;
1045 
1046 	return 0;
1047 }
1048 
1049 static int gmc_v8_0_late_init(void *handle)
1050 {
1051 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1052 
1053 	if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
1054 		return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
1055 	else
1056 		return 0;
1057 }
1058 
1059 static unsigned int gmc_v8_0_get_vbios_fb_size(struct amdgpu_device *adev)
1060 {
1061 	u32 d1vga_control = RREG32(mmD1VGA_CONTROL);
1062 	unsigned int size;
1063 
1064 	if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
1065 		size = AMDGPU_VBIOS_VGA_ALLOCATION;
1066 	} else {
1067 		u32 viewport = RREG32(mmVIEWPORT_SIZE);
1068 
1069 		size = (REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
1070 			REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_WIDTH) *
1071 			4);
1072 	}
1073 
1074 	return size;
1075 }
1076 
1077 #define mmMC_SEQ_MISC0_FIJI 0xA71
1078 
1079 static int gmc_v8_0_sw_init(void *handle)
1080 {
1081 	int r;
1082 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1083 
1084 	set_bit(AMDGPU_GFXHUB(0), adev->vmhubs_mask);
1085 
1086 	if (adev->flags & AMD_IS_APU) {
1087 		adev->gmc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
1088 	} else {
1089 		u32 tmp;
1090 
1091 		if ((adev->asic_type == CHIP_FIJI) ||
1092 		    (adev->asic_type == CHIP_VEGAM))
1093 			tmp = RREG32(mmMC_SEQ_MISC0_FIJI);
1094 		else
1095 			tmp = RREG32(mmMC_SEQ_MISC0);
1096 		tmp &= MC_SEQ_MISC0__MT__MASK;
1097 		adev->gmc.vram_type = gmc_v8_0_convert_vram_type(tmp);
1098 	}
1099 
1100 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_GFX_PAGE_INV_FAULT, &adev->gmc.vm_fault);
1101 	if (r)
1102 		return r;
1103 
1104 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_GFX_MEM_PROT_FAULT, &adev->gmc.vm_fault);
1105 	if (r)
1106 		return r;
1107 
1108 	/* Adjust VM size here.
1109 	 * Currently set to 4GB ((1 << 20) 4k pages).
1110 	 * Max GPUVM size for cayman and SI is 40 bits.
1111 	 */
1112 	amdgpu_vm_adjust_size(adev, 64, 9, 1, 40);
1113 
1114 	/* Set the internal MC address mask
1115 	 * This is the max address of the GPU's
1116 	 * internal address space.
1117 	 */
1118 	adev->gmc.mc_mask = 0xffffffffffULL; /* 40 bit MC */
1119 
1120 	r = dma_set_mask_and_coherent(adev->dev, DMA_BIT_MASK(40));
1121 	if (r) {
1122 		pr_warn("No suitable DMA available\n");
1123 		return r;
1124 	}
1125 	adev->need_swiotlb = drm_need_swiotlb(40);
1126 
1127 	r = gmc_v8_0_init_microcode(adev);
1128 	if (r) {
1129 		DRM_ERROR("Failed to load mc firmware!\n");
1130 		return r;
1131 	}
1132 
1133 	r = gmc_v8_0_mc_init(adev);
1134 	if (r)
1135 		return r;
1136 
1137 	amdgpu_gmc_get_vbios_allocations(adev);
1138 
1139 	/* Memory manager */
1140 	r = amdgpu_bo_init(adev);
1141 	if (r)
1142 		return r;
1143 
1144 	r = gmc_v8_0_gart_init(adev);
1145 	if (r)
1146 		return r;
1147 
1148 	/*
1149 	 * number of VMs
1150 	 * VMID 0 is reserved for System
1151 	 * amdgpu graphics/compute will use VMIDs 1-7
1152 	 * amdkfd will use VMIDs 8-15
1153 	 */
1154 	adev->vm_manager.first_kfd_vmid = 8;
1155 	amdgpu_vm_manager_init(adev);
1156 
1157 	/* base offset of vram pages */
1158 	if (adev->flags & AMD_IS_APU) {
1159 		u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
1160 
1161 		tmp <<= 22;
1162 		adev->vm_manager.vram_base_offset = tmp;
1163 	} else {
1164 		adev->vm_manager.vram_base_offset = 0;
1165 	}
1166 
1167 	adev->gmc.vm_fault_info = kmalloc(sizeof(struct kfd_vm_fault_info),
1168 					GFP_KERNEL);
1169 	if (!adev->gmc.vm_fault_info)
1170 		return -ENOMEM;
1171 	atomic_set(&adev->gmc.vm_fault_info_updated, 0);
1172 
1173 	return 0;
1174 }
1175 
1176 static int gmc_v8_0_sw_fini(void *handle)
1177 {
1178 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1179 
1180 	amdgpu_gem_force_release(adev);
1181 	amdgpu_vm_manager_fini(adev);
1182 	kfree(adev->gmc.vm_fault_info);
1183 	amdgpu_gart_table_vram_free(adev);
1184 	amdgpu_bo_fini(adev);
1185 	amdgpu_ucode_release(&adev->gmc.fw);
1186 
1187 	return 0;
1188 }
1189 
1190 static int gmc_v8_0_hw_init(void *handle)
1191 {
1192 	int r;
1193 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1194 
1195 	gmc_v8_0_init_golden_registers(adev);
1196 
1197 	gmc_v8_0_mc_program(adev);
1198 
1199 	if (adev->asic_type == CHIP_TONGA) {
1200 		r = gmc_v8_0_tonga_mc_load_microcode(adev);
1201 		if (r) {
1202 			DRM_ERROR("Failed to load MC firmware!\n");
1203 			return r;
1204 		}
1205 	} else if (adev->asic_type == CHIP_POLARIS11 ||
1206 			adev->asic_type == CHIP_POLARIS10 ||
1207 			adev->asic_type == CHIP_POLARIS12) {
1208 		r = gmc_v8_0_polaris_mc_load_microcode(adev);
1209 		if (r) {
1210 			DRM_ERROR("Failed to load MC firmware!\n");
1211 			return r;
1212 		}
1213 	}
1214 
1215 	r = gmc_v8_0_gart_enable(adev);
1216 	if (r)
1217 		return r;
1218 
1219 	if (amdgpu_emu_mode == 1)
1220 		return amdgpu_gmc_vram_checking(adev);
1221 
1222 	return 0;
1223 }
1224 
1225 static int gmc_v8_0_hw_fini(void *handle)
1226 {
1227 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1228 
1229 	amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
1230 	gmc_v8_0_gart_disable(adev);
1231 
1232 	return 0;
1233 }
1234 
1235 static int gmc_v8_0_suspend(void *handle)
1236 {
1237 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1238 
1239 	gmc_v8_0_hw_fini(adev);
1240 
1241 	return 0;
1242 }
1243 
1244 static int gmc_v8_0_resume(void *handle)
1245 {
1246 	int r;
1247 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1248 
1249 	r = gmc_v8_0_hw_init(adev);
1250 	if (r)
1251 		return r;
1252 
1253 	amdgpu_vmid_reset_all(adev);
1254 
1255 	return 0;
1256 }
1257 
1258 static bool gmc_v8_0_is_idle(void *handle)
1259 {
1260 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1261 	u32 tmp = RREG32(mmSRBM_STATUS);
1262 
1263 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1264 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
1265 		return false;
1266 
1267 	return true;
1268 }
1269 
1270 static int gmc_v8_0_wait_for_idle(void *handle)
1271 {
1272 	unsigned int i;
1273 	u32 tmp;
1274 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1275 
1276 	for (i = 0; i < adev->usec_timeout; i++) {
1277 		/* read MC_STATUS */
1278 		tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
1279 					       SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1280 					       SRBM_STATUS__MCC_BUSY_MASK |
1281 					       SRBM_STATUS__MCD_BUSY_MASK |
1282 					       SRBM_STATUS__VMC_BUSY_MASK |
1283 					       SRBM_STATUS__VMC1_BUSY_MASK);
1284 		if (!tmp)
1285 			return 0;
1286 		udelay(1);
1287 	}
1288 	return -ETIMEDOUT;
1289 
1290 }
1291 
1292 static bool gmc_v8_0_check_soft_reset(void *handle)
1293 {
1294 	u32 srbm_soft_reset = 0;
1295 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1296 	u32 tmp = RREG32(mmSRBM_STATUS);
1297 
1298 	if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1299 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1300 						SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1301 
1302 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1303 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1304 		if (!(adev->flags & AMD_IS_APU))
1305 			srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1306 							SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1307 	}
1308 
1309 	if (srbm_soft_reset) {
1310 		adev->gmc.srbm_soft_reset = srbm_soft_reset;
1311 		return true;
1312 	}
1313 
1314 	adev->gmc.srbm_soft_reset = 0;
1315 
1316 	return false;
1317 }
1318 
1319 static int gmc_v8_0_pre_soft_reset(void *handle)
1320 {
1321 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1322 
1323 	if (!adev->gmc.srbm_soft_reset)
1324 		return 0;
1325 
1326 	gmc_v8_0_mc_stop(adev);
1327 	if (gmc_v8_0_wait_for_idle(adev))
1328 		dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1329 
1330 	return 0;
1331 }
1332 
1333 static int gmc_v8_0_soft_reset(void *handle)
1334 {
1335 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1336 	u32 srbm_soft_reset;
1337 
1338 	if (!adev->gmc.srbm_soft_reset)
1339 		return 0;
1340 	srbm_soft_reset = adev->gmc.srbm_soft_reset;
1341 
1342 	if (srbm_soft_reset) {
1343 		u32 tmp;
1344 
1345 		tmp = RREG32(mmSRBM_SOFT_RESET);
1346 		tmp |= srbm_soft_reset;
1347 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1348 		WREG32(mmSRBM_SOFT_RESET, tmp);
1349 		tmp = RREG32(mmSRBM_SOFT_RESET);
1350 
1351 		udelay(50);
1352 
1353 		tmp &= ~srbm_soft_reset;
1354 		WREG32(mmSRBM_SOFT_RESET, tmp);
1355 		tmp = RREG32(mmSRBM_SOFT_RESET);
1356 
1357 		/* Wait a little for things to settle down */
1358 		udelay(50);
1359 	}
1360 
1361 	return 0;
1362 }
1363 
1364 static int gmc_v8_0_post_soft_reset(void *handle)
1365 {
1366 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1367 
1368 	if (!adev->gmc.srbm_soft_reset)
1369 		return 0;
1370 
1371 	gmc_v8_0_mc_resume(adev);
1372 	return 0;
1373 }
1374 
1375 static int gmc_v8_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1376 					     struct amdgpu_irq_src *src,
1377 					     unsigned int type,
1378 					     enum amdgpu_interrupt_state state)
1379 {
1380 	u32 tmp;
1381 	u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1382 		    VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1383 		    VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1384 		    VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1385 		    VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1386 		    VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1387 		    VM_CONTEXT1_CNTL__EXECUTE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1388 
1389 	switch (state) {
1390 	case AMDGPU_IRQ_STATE_DISABLE:
1391 		/* system context */
1392 		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1393 		tmp &= ~bits;
1394 		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1395 		/* VMs */
1396 		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1397 		tmp &= ~bits;
1398 		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1399 		break;
1400 	case AMDGPU_IRQ_STATE_ENABLE:
1401 		/* system context */
1402 		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1403 		tmp |= bits;
1404 		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1405 		/* VMs */
1406 		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1407 		tmp |= bits;
1408 		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1409 		break;
1410 	default:
1411 		break;
1412 	}
1413 
1414 	return 0;
1415 }
1416 
1417 static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
1418 				      struct amdgpu_irq_src *source,
1419 				      struct amdgpu_iv_entry *entry)
1420 {
1421 	u32 addr, status, mc_client, vmid;
1422 
1423 	if (amdgpu_sriov_vf(adev)) {
1424 		dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1425 			entry->src_id, entry->src_data[0]);
1426 		dev_err(adev->dev, " Can't decode VM fault info here on SRIOV VF\n");
1427 		return 0;
1428 	}
1429 
1430 	addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1431 	status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1432 	mc_client = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_MCCLIENT);
1433 	/* reset addr and status */
1434 	WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1435 
1436 	if (!addr && !status)
1437 		return 0;
1438 
1439 	amdgpu_vm_update_fault_cache(adev, entry->pasid,
1440 				     ((u64)addr) << AMDGPU_GPU_PAGE_SHIFT, status, AMDGPU_GFXHUB(0));
1441 
1442 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1443 		gmc_v8_0_set_fault_enable_default(adev, false);
1444 
1445 	if (printk_ratelimit()) {
1446 		struct amdgpu_task_info *task_info;
1447 
1448 		dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1449 			entry->src_id, entry->src_data[0]);
1450 
1451 		task_info = amdgpu_vm_get_task_info_pasid(adev, entry->pasid);
1452 		if (task_info) {
1453 			dev_err(adev->dev, " for process %s pid %d thread %s pid %d\n",
1454 				task_info->process_name, task_info->tgid,
1455 				task_info->task_name, task_info->pid);
1456 			amdgpu_vm_put_task_info(task_info);
1457 		}
1458 
1459 		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1460 				addr);
1461 		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1462 			status);
1463 
1464 		gmc_v8_0_vm_decode_fault(adev, status, addr, mc_client,
1465 					 entry->pasid);
1466 	}
1467 
1468 	vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1469 			     VMID);
1470 	if (amdgpu_amdkfd_is_kfd_vmid(adev, vmid)
1471 		&& !atomic_read(&adev->gmc.vm_fault_info_updated)) {
1472 		struct kfd_vm_fault_info *info = adev->gmc.vm_fault_info;
1473 		u32 protections = REG_GET_FIELD(status,
1474 					VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1475 					PROTECTIONS);
1476 
1477 		info->vmid = vmid;
1478 		info->mc_id = REG_GET_FIELD(status,
1479 					    VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1480 					    MEMORY_CLIENT_ID);
1481 		info->status = status;
1482 		info->page_addr = addr;
1483 		info->prot_valid = protections & 0x7 ? true : false;
1484 		info->prot_read = protections & 0x8 ? true : false;
1485 		info->prot_write = protections & 0x10 ? true : false;
1486 		info->prot_exec = protections & 0x20 ? true : false;
1487 		mb();
1488 		atomic_set(&adev->gmc.vm_fault_info_updated, 1);
1489 	}
1490 
1491 	return 0;
1492 }
1493 
1494 static void fiji_update_mc_medium_grain_clock_gating(struct amdgpu_device *adev,
1495 						     bool enable)
1496 {
1497 	uint32_t data;
1498 
1499 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG)) {
1500 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1501 		data |= MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1502 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1503 
1504 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1505 		data |= MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1506 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1507 
1508 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1509 		data |= MC_HUB_MISC_VM_CG__ENABLE_MASK;
1510 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1511 
1512 		data = RREG32(mmMC_XPB_CLK_GAT);
1513 		data |= MC_XPB_CLK_GAT__ENABLE_MASK;
1514 		WREG32(mmMC_XPB_CLK_GAT, data);
1515 
1516 		data = RREG32(mmATC_MISC_CG);
1517 		data |= ATC_MISC_CG__ENABLE_MASK;
1518 		WREG32(mmATC_MISC_CG, data);
1519 
1520 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1521 		data |= MC_CITF_MISC_WR_CG__ENABLE_MASK;
1522 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1523 
1524 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1525 		data |= MC_CITF_MISC_RD_CG__ENABLE_MASK;
1526 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1527 
1528 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1529 		data |= MC_CITF_MISC_VM_CG__ENABLE_MASK;
1530 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1531 
1532 		data = RREG32(mmVM_L2_CG);
1533 		data |= VM_L2_CG__ENABLE_MASK;
1534 		WREG32(mmVM_L2_CG, data);
1535 	} else {
1536 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1537 		data &= ~MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1538 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1539 
1540 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1541 		data &= ~MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1542 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1543 
1544 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1545 		data &= ~MC_HUB_MISC_VM_CG__ENABLE_MASK;
1546 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1547 
1548 		data = RREG32(mmMC_XPB_CLK_GAT);
1549 		data &= ~MC_XPB_CLK_GAT__ENABLE_MASK;
1550 		WREG32(mmMC_XPB_CLK_GAT, data);
1551 
1552 		data = RREG32(mmATC_MISC_CG);
1553 		data &= ~ATC_MISC_CG__ENABLE_MASK;
1554 		WREG32(mmATC_MISC_CG, data);
1555 
1556 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1557 		data &= ~MC_CITF_MISC_WR_CG__ENABLE_MASK;
1558 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1559 
1560 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1561 		data &= ~MC_CITF_MISC_RD_CG__ENABLE_MASK;
1562 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1563 
1564 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1565 		data &= ~MC_CITF_MISC_VM_CG__ENABLE_MASK;
1566 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1567 
1568 		data = RREG32(mmVM_L2_CG);
1569 		data &= ~VM_L2_CG__ENABLE_MASK;
1570 		WREG32(mmVM_L2_CG, data);
1571 	}
1572 }
1573 
1574 static void fiji_update_mc_light_sleep(struct amdgpu_device *adev,
1575 				       bool enable)
1576 {
1577 	uint32_t data;
1578 
1579 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_LS)) {
1580 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1581 		data |= MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1582 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1583 
1584 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1585 		data |= MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1586 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1587 
1588 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1589 		data |= MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1590 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1591 
1592 		data = RREG32(mmMC_XPB_CLK_GAT);
1593 		data |= MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1594 		WREG32(mmMC_XPB_CLK_GAT, data);
1595 
1596 		data = RREG32(mmATC_MISC_CG);
1597 		data |= ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1598 		WREG32(mmATC_MISC_CG, data);
1599 
1600 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1601 		data |= MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1602 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1603 
1604 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1605 		data |= MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1606 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1607 
1608 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1609 		data |= MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1610 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1611 
1612 		data = RREG32(mmVM_L2_CG);
1613 		data |= VM_L2_CG__MEM_LS_ENABLE_MASK;
1614 		WREG32(mmVM_L2_CG, data);
1615 	} else {
1616 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1617 		data &= ~MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1618 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1619 
1620 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1621 		data &= ~MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1622 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1623 
1624 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1625 		data &= ~MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1626 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1627 
1628 		data = RREG32(mmMC_XPB_CLK_GAT);
1629 		data &= ~MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1630 		WREG32(mmMC_XPB_CLK_GAT, data);
1631 
1632 		data = RREG32(mmATC_MISC_CG);
1633 		data &= ~ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1634 		WREG32(mmATC_MISC_CG, data);
1635 
1636 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1637 		data &= ~MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1638 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1639 
1640 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1641 		data &= ~MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1642 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1643 
1644 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1645 		data &= ~MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1646 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1647 
1648 		data = RREG32(mmVM_L2_CG);
1649 		data &= ~VM_L2_CG__MEM_LS_ENABLE_MASK;
1650 		WREG32(mmVM_L2_CG, data);
1651 	}
1652 }
1653 
1654 static int gmc_v8_0_set_clockgating_state(void *handle,
1655 					  enum amd_clockgating_state state)
1656 {
1657 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1658 
1659 	if (amdgpu_sriov_vf(adev))
1660 		return 0;
1661 
1662 	switch (adev->asic_type) {
1663 	case CHIP_FIJI:
1664 		fiji_update_mc_medium_grain_clock_gating(adev,
1665 				state == AMD_CG_STATE_GATE);
1666 		fiji_update_mc_light_sleep(adev,
1667 				state == AMD_CG_STATE_GATE);
1668 		break;
1669 	default:
1670 		break;
1671 	}
1672 	return 0;
1673 }
1674 
1675 static int gmc_v8_0_set_powergating_state(void *handle,
1676 					  enum amd_powergating_state state)
1677 {
1678 	return 0;
1679 }
1680 
1681 static void gmc_v8_0_get_clockgating_state(void *handle, u64 *flags)
1682 {
1683 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1684 	int data;
1685 
1686 	if (amdgpu_sriov_vf(adev))
1687 		*flags = 0;
1688 
1689 	/* AMD_CG_SUPPORT_MC_MGCG */
1690 	data = RREG32(mmMC_HUB_MISC_HUB_CG);
1691 	if (data & MC_HUB_MISC_HUB_CG__ENABLE_MASK)
1692 		*flags |= AMD_CG_SUPPORT_MC_MGCG;
1693 
1694 	/* AMD_CG_SUPPORT_MC_LS */
1695 	if (data & MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK)
1696 		*flags |= AMD_CG_SUPPORT_MC_LS;
1697 }
1698 
1699 static const struct amd_ip_funcs gmc_v8_0_ip_funcs = {
1700 	.name = "gmc_v8_0",
1701 	.early_init = gmc_v8_0_early_init,
1702 	.late_init = gmc_v8_0_late_init,
1703 	.sw_init = gmc_v8_0_sw_init,
1704 	.sw_fini = gmc_v8_0_sw_fini,
1705 	.hw_init = gmc_v8_0_hw_init,
1706 	.hw_fini = gmc_v8_0_hw_fini,
1707 	.suspend = gmc_v8_0_suspend,
1708 	.resume = gmc_v8_0_resume,
1709 	.is_idle = gmc_v8_0_is_idle,
1710 	.wait_for_idle = gmc_v8_0_wait_for_idle,
1711 	.check_soft_reset = gmc_v8_0_check_soft_reset,
1712 	.pre_soft_reset = gmc_v8_0_pre_soft_reset,
1713 	.soft_reset = gmc_v8_0_soft_reset,
1714 	.post_soft_reset = gmc_v8_0_post_soft_reset,
1715 	.set_clockgating_state = gmc_v8_0_set_clockgating_state,
1716 	.set_powergating_state = gmc_v8_0_set_powergating_state,
1717 	.get_clockgating_state = gmc_v8_0_get_clockgating_state,
1718 	.dump_ip_state = NULL,
1719 	.print_ip_state = NULL,
1720 };
1721 
1722 static const struct amdgpu_gmc_funcs gmc_v8_0_gmc_funcs = {
1723 	.flush_gpu_tlb = gmc_v8_0_flush_gpu_tlb,
1724 	.flush_gpu_tlb_pasid = gmc_v8_0_flush_gpu_tlb_pasid,
1725 	.emit_flush_gpu_tlb = gmc_v8_0_emit_flush_gpu_tlb,
1726 	.emit_pasid_mapping = gmc_v8_0_emit_pasid_mapping,
1727 	.set_prt = gmc_v8_0_set_prt,
1728 	.get_vm_pde = gmc_v8_0_get_vm_pde,
1729 	.get_vm_pte = gmc_v8_0_get_vm_pte,
1730 	.get_vbios_fb_size = gmc_v8_0_get_vbios_fb_size,
1731 };
1732 
1733 static const struct amdgpu_irq_src_funcs gmc_v8_0_irq_funcs = {
1734 	.set = gmc_v8_0_vm_fault_interrupt_state,
1735 	.process = gmc_v8_0_process_interrupt,
1736 };
1737 
1738 static void gmc_v8_0_set_gmc_funcs(struct amdgpu_device *adev)
1739 {
1740 	adev->gmc.gmc_funcs = &gmc_v8_0_gmc_funcs;
1741 }
1742 
1743 static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev)
1744 {
1745 	adev->gmc.vm_fault.num_types = 1;
1746 	adev->gmc.vm_fault.funcs = &gmc_v8_0_irq_funcs;
1747 }
1748 
1749 const struct amdgpu_ip_block_version gmc_v8_0_ip_block = {
1750 	.type = AMD_IP_BLOCK_TYPE_GMC,
1751 	.major = 8,
1752 	.minor = 0,
1753 	.rev = 0,
1754 	.funcs = &gmc_v8_0_ip_funcs,
1755 };
1756 
1757 const struct amdgpu_ip_block_version gmc_v8_1_ip_block = {
1758 	.type = AMD_IP_BLOCK_TYPE_GMC,
1759 	.major = 8,
1760 	.minor = 1,
1761 	.rev = 0,
1762 	.funcs = &gmc_v8_0_ip_funcs,
1763 };
1764 
1765 const struct amdgpu_ip_block_version gmc_v8_5_ip_block = {
1766 	.type = AMD_IP_BLOCK_TYPE_GMC,
1767 	.major = 8,
1768 	.minor = 5,
1769 	.rev = 0,
1770 	.funcs = &gmc_v8_0_ip_funcs,
1771 };
1772