xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c (revision 3fc7c707)
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 	char fw_name[30];
216 	int err;
217 
218 	DRM_DEBUG("\n");
219 
220 	switch (adev->asic_type) {
221 	case CHIP_TONGA:
222 		chip_name = "tonga";
223 		break;
224 	case CHIP_POLARIS11:
225 		if (ASICID_IS_P21(adev->pdev->device, adev->pdev->revision) ||
226 		    ASICID_IS_P31(adev->pdev->device, adev->pdev->revision))
227 			chip_name = "polaris11_k";
228 		else
229 			chip_name = "polaris11";
230 		break;
231 	case CHIP_POLARIS10:
232 		if (ASICID_IS_P30(adev->pdev->device, adev->pdev->revision))
233 			chip_name = "polaris10_k";
234 		else
235 			chip_name = "polaris10";
236 		break;
237 	case CHIP_POLARIS12:
238 		if (ASICID_IS_P23(adev->pdev->device, adev->pdev->revision)) {
239 			chip_name = "polaris12_k";
240 		} else {
241 			WREG32(mmMC_SEQ_IO_DEBUG_INDEX, ixMC_IO_DEBUG_UP_159);
242 			/* Polaris12 32bit ASIC needs a special MC firmware */
243 			if (RREG32(mmMC_SEQ_IO_DEBUG_DATA) == 0x05b4dc40)
244 				chip_name = "polaris12_32";
245 			else
246 				chip_name = "polaris12";
247 		}
248 		break;
249 	case CHIP_FIJI:
250 	case CHIP_CARRIZO:
251 	case CHIP_STONEY:
252 	case CHIP_VEGAM:
253 		return 0;
254 	default:
255 		return -EINVAL;
256 	}
257 
258 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mc.bin", chip_name);
259 	err = amdgpu_ucode_request(adev, &adev->gmc.fw, fw_name);
260 	if (err) {
261 		pr_err("mc: Failed to load firmware \"%s\"\n", fw_name);
262 		amdgpu_ucode_release(&adev->gmc.fw);
263 	}
264 	return err;
265 }
266 
267 /**
268  * gmc_v8_0_tonga_mc_load_microcode - load tonga MC ucode into the hw
269  *
270  * @adev: amdgpu_device pointer
271  *
272  * Load the GDDR MC ucode into the hw (VI).
273  * Returns 0 on success, error on failure.
274  */
275 static int gmc_v8_0_tonga_mc_load_microcode(struct amdgpu_device *adev)
276 {
277 	const struct mc_firmware_header_v1_0 *hdr;
278 	const __le32 *fw_data = NULL;
279 	const __le32 *io_mc_regs = NULL;
280 	u32 running;
281 	int i, ucode_size, regs_size;
282 
283 	/* Skip MC ucode loading on SR-IOV capable boards.
284 	 * vbios does this for us in asic_init in that case.
285 	 * Skip MC ucode loading on VF, because hypervisor will do that
286 	 * for this adaptor.
287 	 */
288 	if (amdgpu_sriov_bios(adev))
289 		return 0;
290 
291 	if (!adev->gmc.fw)
292 		return -EINVAL;
293 
294 	hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
295 	amdgpu_ucode_print_mc_hdr(&hdr->header);
296 
297 	adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
298 	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
299 	io_mc_regs = (const __le32 *)
300 		(adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
301 	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
302 	fw_data = (const __le32 *)
303 		(adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
304 
305 	running = REG_GET_FIELD(RREG32(mmMC_SEQ_SUP_CNTL), MC_SEQ_SUP_CNTL, RUN);
306 
307 	if (running == 0) {
308 		/* reset the engine and set to writable */
309 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
310 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
311 
312 		/* load mc io regs */
313 		for (i = 0; i < regs_size; i++) {
314 			WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
315 			WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
316 		}
317 		/* load the MC ucode */
318 		for (i = 0; i < ucode_size; i++)
319 			WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
320 
321 		/* put the engine back into the active state */
322 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
323 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
324 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
325 
326 		/* wait for training to complete */
327 		for (i = 0; i < adev->usec_timeout; i++) {
328 			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
329 					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D0))
330 				break;
331 			udelay(1);
332 		}
333 		for (i = 0; i < adev->usec_timeout; i++) {
334 			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
335 					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D1))
336 				break;
337 			udelay(1);
338 		}
339 	}
340 
341 	return 0;
342 }
343 
344 static int gmc_v8_0_polaris_mc_load_microcode(struct amdgpu_device *adev)
345 {
346 	const struct mc_firmware_header_v1_0 *hdr;
347 	const __le32 *fw_data = NULL;
348 	const __le32 *io_mc_regs = NULL;
349 	u32 data;
350 	int i, ucode_size, regs_size;
351 
352 	/* Skip MC ucode loading on SR-IOV capable boards.
353 	 * vbios does this for us in asic_init in that case.
354 	 * Skip MC ucode loading on VF, because hypervisor will do that
355 	 * for this adaptor.
356 	 */
357 	if (amdgpu_sriov_bios(adev))
358 		return 0;
359 
360 	if (!adev->gmc.fw)
361 		return -EINVAL;
362 
363 	hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
364 	amdgpu_ucode_print_mc_hdr(&hdr->header);
365 
366 	adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
367 	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
368 	io_mc_regs = (const __le32 *)
369 		(adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
370 	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
371 	fw_data = (const __le32 *)
372 		(adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
373 
374 	data = RREG32(mmMC_SEQ_MISC0);
375 	data &= ~(0x40);
376 	WREG32(mmMC_SEQ_MISC0, data);
377 
378 	/* load mc io regs */
379 	for (i = 0; i < regs_size; i++) {
380 		WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
381 		WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
382 	}
383 
384 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
385 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
386 
387 	/* load the MC ucode */
388 	for (i = 0; i < ucode_size; i++)
389 		WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
390 
391 	/* put the engine back into the active state */
392 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
393 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
394 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
395 
396 	/* wait for training to complete */
397 	for (i = 0; i < adev->usec_timeout; i++) {
398 		data = RREG32(mmMC_SEQ_MISC0);
399 		if (data & 0x80)
400 			break;
401 		udelay(1);
402 	}
403 
404 	return 0;
405 }
406 
407 static void gmc_v8_0_vram_gtt_location(struct amdgpu_device *adev,
408 				       struct amdgpu_gmc *mc)
409 {
410 	u64 base = 0;
411 
412 	if (!amdgpu_sriov_vf(adev))
413 		base = RREG32(mmMC_VM_FB_LOCATION) & 0xFFFF;
414 	base <<= 24;
415 
416 	amdgpu_gmc_vram_location(adev, mc, base);
417 	amdgpu_gmc_gart_location(adev, mc);
418 }
419 
420 /**
421  * gmc_v8_0_mc_program - program the GPU memory controller
422  *
423  * @adev: amdgpu_device pointer
424  *
425  * Set the location of vram, gart, and AGP in the GPU's
426  * physical address space (VI).
427  */
428 static void gmc_v8_0_mc_program(struct amdgpu_device *adev)
429 {
430 	u32 tmp;
431 	int i, j;
432 
433 	/* Initialize HDP */
434 	for (i = 0, j = 0; i < 32; i++, j += 0x6) {
435 		WREG32((0xb05 + j), 0x00000000);
436 		WREG32((0xb06 + j), 0x00000000);
437 		WREG32((0xb07 + j), 0x00000000);
438 		WREG32((0xb08 + j), 0x00000000);
439 		WREG32((0xb09 + j), 0x00000000);
440 	}
441 	WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
442 
443 	if (gmc_v8_0_wait_for_idle((void *)adev))
444 		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
445 
446 	if (adev->mode_info.num_crtc) {
447 		/* Lockout access through VGA aperture*/
448 		tmp = RREG32(mmVGA_HDP_CONTROL);
449 		tmp = REG_SET_FIELD(tmp, VGA_HDP_CONTROL, VGA_MEMORY_DISABLE, 1);
450 		WREG32(mmVGA_HDP_CONTROL, tmp);
451 
452 		/* disable VGA render */
453 		tmp = RREG32(mmVGA_RENDER_CONTROL);
454 		tmp = REG_SET_FIELD(tmp, VGA_RENDER_CONTROL, VGA_VSTATUS_CNTL, 0);
455 		WREG32(mmVGA_RENDER_CONTROL, tmp);
456 	}
457 	/* Update configuration */
458 	WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
459 	       adev->gmc.vram_start >> 12);
460 	WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
461 	       adev->gmc.vram_end >> 12);
462 	WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
463 	       adev->mem_scratch.gpu_addr >> 12);
464 
465 	if (amdgpu_sriov_vf(adev)) {
466 		tmp = ((adev->gmc.vram_end >> 24) & 0xFFFF) << 16;
467 		tmp |= ((adev->gmc.vram_start >> 24) & 0xFFFF);
468 		WREG32(mmMC_VM_FB_LOCATION, tmp);
469 		/* XXX double check these! */
470 		WREG32(mmHDP_NONSURFACE_BASE, (adev->gmc.vram_start >> 8));
471 		WREG32(mmHDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
472 		WREG32(mmHDP_NONSURFACE_SIZE, 0x3FFFFFFF);
473 	}
474 
475 	WREG32(mmMC_VM_AGP_BASE, 0);
476 	WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
477 	WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
478 	if (gmc_v8_0_wait_for_idle((void *)adev))
479 		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
480 
481 	WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK);
482 
483 	tmp = RREG32(mmHDP_MISC_CNTL);
484 	tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 0);
485 	WREG32(mmHDP_MISC_CNTL, tmp);
486 
487 	tmp = RREG32(mmHDP_HOST_PATH_CNTL);
488 	WREG32(mmHDP_HOST_PATH_CNTL, tmp);
489 }
490 
491 /**
492  * gmc_v8_0_mc_init - initialize the memory controller driver params
493  *
494  * @adev: amdgpu_device pointer
495  *
496  * Look up the amount of vram, vram width, and decide how to place
497  * vram and gart within the GPU's physical address space (VI).
498  * Returns 0 for success.
499  */
500 static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
501 {
502 	int r;
503 	u32 tmp;
504 
505 	adev->gmc.vram_width = amdgpu_atombios_get_vram_width(adev);
506 	if (!adev->gmc.vram_width) {
507 		int chansize, numchan;
508 
509 		/* Get VRAM informations */
510 		tmp = RREG32(mmMC_ARB_RAMCFG);
511 		if (REG_GET_FIELD(tmp, MC_ARB_RAMCFG, CHANSIZE))
512 			chansize = 64;
513 		else
514 			chansize = 32;
515 
516 		tmp = RREG32(mmMC_SHARED_CHMAP);
517 		switch (REG_GET_FIELD(tmp, MC_SHARED_CHMAP, NOOFCHAN)) {
518 		case 0:
519 		default:
520 			numchan = 1;
521 			break;
522 		case 1:
523 			numchan = 2;
524 			break;
525 		case 2:
526 			numchan = 4;
527 			break;
528 		case 3:
529 			numchan = 8;
530 			break;
531 		case 4:
532 			numchan = 3;
533 			break;
534 		case 5:
535 			numchan = 6;
536 			break;
537 		case 6:
538 			numchan = 10;
539 			break;
540 		case 7:
541 			numchan = 12;
542 			break;
543 		case 8:
544 			numchan = 16;
545 			break;
546 		}
547 		adev->gmc.vram_width = numchan * chansize;
548 	}
549 	/* size in MB on si */
550 	tmp = RREG32(mmCONFIG_MEMSIZE);
551 	/* some boards may have garbage in the upper 16 bits */
552 	if (tmp & 0xffff0000) {
553 		DRM_INFO("Probable bad vram size: 0x%08x\n", tmp);
554 		if (tmp & 0xffff)
555 			tmp &= 0xffff;
556 	}
557 	adev->gmc.mc_vram_size = tmp * 1024ULL * 1024ULL;
558 	adev->gmc.real_vram_size = adev->gmc.mc_vram_size;
559 
560 	if (!(adev->flags & AMD_IS_APU)) {
561 		r = amdgpu_device_resize_fb_bar(adev);
562 		if (r)
563 			return r;
564 	}
565 	adev->gmc.aper_base = pci_resource_start(adev->pdev, 0);
566 	adev->gmc.aper_size = pci_resource_len(adev->pdev, 0);
567 
568 #ifdef CONFIG_X86_64
569 	if ((adev->flags & AMD_IS_APU) && !amdgpu_passthrough(adev)) {
570 		adev->gmc.aper_base = ((u64)RREG32(mmMC_VM_FB_OFFSET)) << 22;
571 		adev->gmc.aper_size = adev->gmc.real_vram_size;
572 	}
573 #endif
574 
575 	adev->gmc.visible_vram_size = adev->gmc.aper_size;
576 
577 	/* set the gart size */
578 	if (amdgpu_gart_size == -1) {
579 		switch (adev->asic_type) {
580 		case CHIP_POLARIS10: /* all engines support GPUVM */
581 		case CHIP_POLARIS11: /* all engines support GPUVM */
582 		case CHIP_POLARIS12: /* all engines support GPUVM */
583 		case CHIP_VEGAM:     /* all engines support GPUVM */
584 		default:
585 			adev->gmc.gart_size = 256ULL << 20;
586 			break;
587 		case CHIP_TONGA:   /* UVD, VCE do not support GPUVM */
588 		case CHIP_FIJI:    /* UVD, VCE do not support GPUVM */
589 		case CHIP_CARRIZO: /* UVD, VCE do not support GPUVM, DCE SG support */
590 		case CHIP_STONEY:  /* UVD does not support GPUVM, DCE SG support */
591 			adev->gmc.gart_size = 1024ULL << 20;
592 			break;
593 		}
594 	} else {
595 		adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
596 	}
597 
598 	adev->gmc.gart_size += adev->pm.smu_prv_buffer_size;
599 	gmc_v8_0_vram_gtt_location(adev, &adev->gmc);
600 
601 	return 0;
602 }
603 
604 /**
605  * gmc_v8_0_flush_gpu_tlb_pasid - tlb flush via pasid
606  *
607  * @adev: amdgpu_device pointer
608  * @pasid: pasid to be flush
609  * @flush_type: type of flush
610  * @all_hub: flush all hubs
611  * @inst: is used to select which instance of KIQ to use for the invalidation
612  *
613  * Flush the TLB for the requested pasid.
614  */
615 static int gmc_v8_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
616 					uint16_t pasid, uint32_t flush_type,
617 					bool all_hub, uint32_t inst)
618 {
619 	int vmid;
620 	unsigned int tmp;
621 
622 	if (amdgpu_in_reset(adev))
623 		return -EIO;
624 
625 	for (vmid = 1; vmid < 16; vmid++) {
626 
627 		tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
628 		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
629 			(tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
630 			WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
631 			RREG32(mmVM_INVALIDATE_RESPONSE);
632 			break;
633 		}
634 	}
635 
636 	return 0;
637 
638 }
639 
640 /*
641  * GART
642  * VMID 0 is the physical GPU addresses as used by the kernel.
643  * VMIDs 1-15 are used for userspace clients and are handled
644  * by the amdgpu vm/hsa code.
645  */
646 
647 /**
648  * gmc_v8_0_flush_gpu_tlb - gart tlb flush callback
649  *
650  * @adev: amdgpu_device pointer
651  * @vmid: vm instance to flush
652  * @vmhub: which hub to flush
653  * @flush_type: type of flush
654  *
655  * Flush the TLB for the requested page table (VI).
656  */
657 static void gmc_v8_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
658 					uint32_t vmhub, uint32_t flush_type)
659 {
660 	/* bits 0-15 are the VM contexts0-15 */
661 	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
662 }
663 
664 static uint64_t gmc_v8_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
665 					    unsigned int vmid, uint64_t pd_addr)
666 {
667 	uint32_t reg;
668 
669 	if (vmid < 8)
670 		reg = mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid;
671 	else
672 		reg = mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8;
673 	amdgpu_ring_emit_wreg(ring, reg, pd_addr >> 12);
674 
675 	/* bits 0-15 are the VM contexts0-15 */
676 	amdgpu_ring_emit_wreg(ring, mmVM_INVALIDATE_REQUEST, 1 << vmid);
677 
678 	return pd_addr;
679 }
680 
681 static void gmc_v8_0_emit_pasid_mapping(struct amdgpu_ring *ring, unsigned int vmid,
682 					unsigned int pasid)
683 {
684 	amdgpu_ring_emit_wreg(ring, mmIH_VMID_0_LUT + vmid, pasid);
685 }
686 
687 /*
688  * PTE format on VI:
689  * 63:40 reserved
690  * 39:12 4k physical page base address
691  * 11:7 fragment
692  * 6 write
693  * 5 read
694  * 4 exe
695  * 3 reserved
696  * 2 snooped
697  * 1 system
698  * 0 valid
699  *
700  * PDE format on VI:
701  * 63:59 block fragment size
702  * 58:40 reserved
703  * 39:1 physical base address of PTE
704  * bits 5:1 must be 0.
705  * 0 valid
706  */
707 
708 static void gmc_v8_0_get_vm_pde(struct amdgpu_device *adev, int level,
709 				uint64_t *addr, uint64_t *flags)
710 {
711 	BUG_ON(*addr & 0xFFFFFF0000000FFFULL);
712 }
713 
714 static void gmc_v8_0_get_vm_pte(struct amdgpu_device *adev,
715 				struct amdgpu_bo_va_mapping *mapping,
716 				uint64_t *flags)
717 {
718 	*flags &= ~AMDGPU_PTE_EXECUTABLE;
719 	*flags |= mapping->flags & AMDGPU_PTE_EXECUTABLE;
720 	*flags &= ~AMDGPU_PTE_PRT;
721 }
722 
723 /**
724  * gmc_v8_0_set_fault_enable_default - update VM fault handling
725  *
726  * @adev: amdgpu_device pointer
727  * @value: true redirects VM faults to the default page
728  */
729 static void gmc_v8_0_set_fault_enable_default(struct amdgpu_device *adev,
730 					      bool value)
731 {
732 	u32 tmp;
733 
734 	tmp = RREG32(mmVM_CONTEXT1_CNTL);
735 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
736 			    RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
737 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
738 			    DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
739 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
740 			    PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
741 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
742 			    VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
743 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
744 			    READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
745 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
746 			    WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
747 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
748 			    EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
749 	WREG32(mmVM_CONTEXT1_CNTL, tmp);
750 }
751 
752 /**
753  * gmc_v8_0_set_prt() - set PRT VM fault
754  *
755  * @adev: amdgpu_device pointer
756  * @enable: enable/disable VM fault handling for PRT
757  */
758 static void gmc_v8_0_set_prt(struct amdgpu_device *adev, bool enable)
759 {
760 	u32 tmp;
761 
762 	if (enable && !adev->gmc.prt_warning) {
763 		dev_warn(adev->dev, "Disabling VM faults because of PRT request!\n");
764 		adev->gmc.prt_warning = true;
765 	}
766 
767 	tmp = RREG32(mmVM_PRT_CNTL);
768 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
769 			    CB_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
770 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
771 			    CB_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
772 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
773 			    TC_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
774 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
775 			    TC_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
776 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
777 			    L2_CACHE_STORE_INVALID_ENTRIES, enable);
778 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
779 			    L1_TLB_STORE_INVALID_ENTRIES, enable);
780 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
781 			    MASK_PDE0_FAULT, enable);
782 	WREG32(mmVM_PRT_CNTL, tmp);
783 
784 	if (enable) {
785 		uint32_t low = AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT;
786 		uint32_t high = adev->vm_manager.max_pfn -
787 			(AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT);
788 
789 		WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, low);
790 		WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, low);
791 		WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, low);
792 		WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, low);
793 		WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, high);
794 		WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, high);
795 		WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, high);
796 		WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, high);
797 	} else {
798 		WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, 0xfffffff);
799 		WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, 0xfffffff);
800 		WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, 0xfffffff);
801 		WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, 0xfffffff);
802 		WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, 0x0);
803 		WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, 0x0);
804 		WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, 0x0);
805 		WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, 0x0);
806 	}
807 }
808 
809 /**
810  * gmc_v8_0_gart_enable - gart enable
811  *
812  * @adev: amdgpu_device pointer
813  *
814  * This sets up the TLBs, programs the page tables for VMID0,
815  * sets up the hw for VMIDs 1-15 which are allocated on
816  * demand, and sets up the global locations for the LDS, GDS,
817  * and GPUVM for FSA64 clients (VI).
818  * Returns 0 for success, errors for failure.
819  */
820 static int gmc_v8_0_gart_enable(struct amdgpu_device *adev)
821 {
822 	uint64_t table_addr;
823 	u32 tmp, field;
824 	int i;
825 
826 	if (adev->gart.bo == NULL) {
827 		dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
828 		return -EINVAL;
829 	}
830 	amdgpu_gtt_mgr_recover(&adev->mman.gtt_mgr);
831 	table_addr = amdgpu_bo_gpu_offset(adev->gart.bo);
832 
833 	/* Setup TLB control */
834 	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
835 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 1);
836 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 1);
837 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE, 3);
838 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 1);
839 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
840 	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
841 	/* Setup L2 cache */
842 	tmp = RREG32(mmVM_L2_CNTL);
843 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
844 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
845 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE, 1);
846 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1);
847 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7);
848 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
849 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1);
850 	WREG32(mmVM_L2_CNTL, tmp);
851 	tmp = RREG32(mmVM_L2_CNTL2);
852 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
853 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
854 	WREG32(mmVM_L2_CNTL2, tmp);
855 
856 	field = adev->vm_manager.fragment_size;
857 	tmp = RREG32(mmVM_L2_CNTL3);
858 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY, 1);
859 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, field);
860 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_FRAGMENT_SIZE, field);
861 	WREG32(mmVM_L2_CNTL3, tmp);
862 	/* XXX: set to enable PTE/PDE in system memory */
863 	tmp = RREG32(mmVM_L2_CNTL4);
864 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_PHYSICAL, 0);
865 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SHARED, 0);
866 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SNOOP, 0);
867 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_PHYSICAL, 0);
868 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SHARED, 0);
869 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SNOOP, 0);
870 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_PHYSICAL, 0);
871 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SHARED, 0);
872 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SNOOP, 0);
873 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_PHYSICAL, 0);
874 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SHARED, 0);
875 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SNOOP, 0);
876 	WREG32(mmVM_L2_CNTL4, tmp);
877 	/* setup context0 */
878 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->gmc.gart_start >> 12);
879 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->gmc.gart_end >> 12);
880 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, table_addr >> 12);
881 	WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
882 			(u32)(adev->dummy_page_addr >> 12));
883 	WREG32(mmVM_CONTEXT0_CNTL2, 0);
884 	tmp = RREG32(mmVM_CONTEXT0_CNTL);
885 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
886 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH, 0);
887 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
888 	WREG32(mmVM_CONTEXT0_CNTL, tmp);
889 
890 	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR, 0);
891 	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR, 0);
892 	WREG32(mmVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET, 0);
893 
894 	/* empty context1-15 */
895 	/* FIXME start with 4G, once using 2 level pt switch to full
896 	 * vm size space
897 	 */
898 	/* set vm size, must be a multiple of 4 */
899 	WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
900 	WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
901 	for (i = 1; i < AMDGPU_NUM_VMID; i++) {
902 		if (i < 8)
903 			WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
904 			       table_addr >> 12);
905 		else
906 			WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
907 			       table_addr >> 12);
908 	}
909 
910 	/* enable context1-15 */
911 	WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
912 	       (u32)(adev->dummy_page_addr >> 12));
913 	WREG32(mmVM_CONTEXT1_CNTL2, 4);
914 	tmp = RREG32(mmVM_CONTEXT1_CNTL);
915 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
916 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH, 1);
917 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
918 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
919 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
920 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
921 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
922 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
923 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
924 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE,
925 			    adev->vm_manager.block_size - 9);
926 	WREG32(mmVM_CONTEXT1_CNTL, tmp);
927 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
928 		gmc_v8_0_set_fault_enable_default(adev, false);
929 	else
930 		gmc_v8_0_set_fault_enable_default(adev, true);
931 
932 	gmc_v8_0_flush_gpu_tlb(adev, 0, 0, 0);
933 	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
934 		 (unsigned int)(adev->gmc.gart_size >> 20),
935 		 (unsigned long long)table_addr);
936 	return 0;
937 }
938 
939 static int gmc_v8_0_gart_init(struct amdgpu_device *adev)
940 {
941 	int r;
942 
943 	if (adev->gart.bo) {
944 		WARN(1, "R600 PCIE GART already initialized\n");
945 		return 0;
946 	}
947 	/* Initialize common gart structure */
948 	r = amdgpu_gart_init(adev);
949 	if (r)
950 		return r;
951 	adev->gart.table_size = adev->gart.num_gpu_pages * 8;
952 	adev->gart.gart_pte_flags = AMDGPU_PTE_EXECUTABLE;
953 	return amdgpu_gart_table_vram_alloc(adev);
954 }
955 
956 /**
957  * gmc_v8_0_gart_disable - gart disable
958  *
959  * @adev: amdgpu_device pointer
960  *
961  * This disables all VM page table (VI).
962  */
963 static void gmc_v8_0_gart_disable(struct amdgpu_device *adev)
964 {
965 	u32 tmp;
966 
967 	/* Disable all tables */
968 	WREG32(mmVM_CONTEXT0_CNTL, 0);
969 	WREG32(mmVM_CONTEXT1_CNTL, 0);
970 	/* Setup TLB control */
971 	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
972 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
973 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 0);
974 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 0);
975 	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
976 	/* Setup L2 cache */
977 	tmp = RREG32(mmVM_L2_CNTL);
978 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
979 	WREG32(mmVM_L2_CNTL, tmp);
980 	WREG32(mmVM_L2_CNTL2, 0);
981 }
982 
983 /**
984  * gmc_v8_0_vm_decode_fault - print human readable fault info
985  *
986  * @adev: amdgpu_device pointer
987  * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
988  * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
989  * @mc_client: VM_CONTEXT1_PROTECTION_FAULT_MCCLIENT register value
990  * @pasid: debug logging only - no functional use
991  *
992  * Print human readable fault information (VI).
993  */
994 static void gmc_v8_0_vm_decode_fault(struct amdgpu_device *adev, u32 status,
995 				     u32 addr, u32 mc_client, unsigned int pasid)
996 {
997 	u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
998 	u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
999 					PROTECTIONS);
1000 	char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
1001 		(mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
1002 	u32 mc_id;
1003 
1004 	mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1005 			      MEMORY_CLIENT_ID);
1006 
1007 	dev_err(adev->dev, "VM fault (0x%02x, vmid %d, pasid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
1008 	       protections, vmid, pasid, addr,
1009 	       REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1010 			     MEMORY_CLIENT_RW) ?
1011 	       "write" : "read", block, mc_client, mc_id);
1012 }
1013 
1014 static int gmc_v8_0_convert_vram_type(int mc_seq_vram_type)
1015 {
1016 	switch (mc_seq_vram_type) {
1017 	case MC_SEQ_MISC0__MT__GDDR1:
1018 		return AMDGPU_VRAM_TYPE_GDDR1;
1019 	case MC_SEQ_MISC0__MT__DDR2:
1020 		return AMDGPU_VRAM_TYPE_DDR2;
1021 	case MC_SEQ_MISC0__MT__GDDR3:
1022 		return AMDGPU_VRAM_TYPE_GDDR3;
1023 	case MC_SEQ_MISC0__MT__GDDR4:
1024 		return AMDGPU_VRAM_TYPE_GDDR4;
1025 	case MC_SEQ_MISC0__MT__GDDR5:
1026 		return AMDGPU_VRAM_TYPE_GDDR5;
1027 	case MC_SEQ_MISC0__MT__HBM:
1028 		return AMDGPU_VRAM_TYPE_HBM;
1029 	case MC_SEQ_MISC0__MT__DDR3:
1030 		return AMDGPU_VRAM_TYPE_DDR3;
1031 	default:
1032 		return AMDGPU_VRAM_TYPE_UNKNOWN;
1033 	}
1034 }
1035 
1036 static int gmc_v8_0_early_init(void *handle)
1037 {
1038 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1039 
1040 	gmc_v8_0_set_gmc_funcs(adev);
1041 	gmc_v8_0_set_irq_funcs(adev);
1042 
1043 	adev->gmc.shared_aperture_start = 0x2000000000000000ULL;
1044 	adev->gmc.shared_aperture_end =
1045 		adev->gmc.shared_aperture_start + (4ULL << 30) - 1;
1046 	adev->gmc.private_aperture_start =
1047 		adev->gmc.shared_aperture_end + 1;
1048 	adev->gmc.private_aperture_end =
1049 		adev->gmc.private_aperture_start + (4ULL << 30) - 1;
1050 	adev->gmc.noretry_flags = AMDGPU_VM_NORETRY_FLAGS_TF;
1051 
1052 	return 0;
1053 }
1054 
1055 static int gmc_v8_0_late_init(void *handle)
1056 {
1057 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1058 
1059 	if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
1060 		return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
1061 	else
1062 		return 0;
1063 }
1064 
1065 static unsigned int gmc_v8_0_get_vbios_fb_size(struct amdgpu_device *adev)
1066 {
1067 	u32 d1vga_control = RREG32(mmD1VGA_CONTROL);
1068 	unsigned int size;
1069 
1070 	if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
1071 		size = AMDGPU_VBIOS_VGA_ALLOCATION;
1072 	} else {
1073 		u32 viewport = RREG32(mmVIEWPORT_SIZE);
1074 
1075 		size = (REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
1076 			REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_WIDTH) *
1077 			4);
1078 	}
1079 
1080 	return size;
1081 }
1082 
1083 #define mmMC_SEQ_MISC0_FIJI 0xA71
1084 
1085 static int gmc_v8_0_sw_init(void *handle)
1086 {
1087 	int r;
1088 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1089 
1090 	set_bit(AMDGPU_GFXHUB(0), adev->vmhubs_mask);
1091 
1092 	if (adev->flags & AMD_IS_APU) {
1093 		adev->gmc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
1094 	} else {
1095 		u32 tmp;
1096 
1097 		if ((adev->asic_type == CHIP_FIJI) ||
1098 		    (adev->asic_type == CHIP_VEGAM))
1099 			tmp = RREG32(mmMC_SEQ_MISC0_FIJI);
1100 		else
1101 			tmp = RREG32(mmMC_SEQ_MISC0);
1102 		tmp &= MC_SEQ_MISC0__MT__MASK;
1103 		adev->gmc.vram_type = gmc_v8_0_convert_vram_type(tmp);
1104 	}
1105 
1106 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_GFX_PAGE_INV_FAULT, &adev->gmc.vm_fault);
1107 	if (r)
1108 		return r;
1109 
1110 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_GFX_MEM_PROT_FAULT, &adev->gmc.vm_fault);
1111 	if (r)
1112 		return r;
1113 
1114 	/* Adjust VM size here.
1115 	 * Currently set to 4GB ((1 << 20) 4k pages).
1116 	 * Max GPUVM size for cayman and SI is 40 bits.
1117 	 */
1118 	amdgpu_vm_adjust_size(adev, 64, 9, 1, 40);
1119 
1120 	/* Set the internal MC address mask
1121 	 * This is the max address of the GPU's
1122 	 * internal address space.
1123 	 */
1124 	adev->gmc.mc_mask = 0xffffffffffULL; /* 40 bit MC */
1125 
1126 	r = dma_set_mask_and_coherent(adev->dev, DMA_BIT_MASK(40));
1127 	if (r) {
1128 		pr_warn("No suitable DMA available\n");
1129 		return r;
1130 	}
1131 	adev->need_swiotlb = drm_need_swiotlb(40);
1132 
1133 	r = gmc_v8_0_init_microcode(adev);
1134 	if (r) {
1135 		DRM_ERROR("Failed to load mc firmware!\n");
1136 		return r;
1137 	}
1138 
1139 	r = gmc_v8_0_mc_init(adev);
1140 	if (r)
1141 		return r;
1142 
1143 	amdgpu_gmc_get_vbios_allocations(adev);
1144 
1145 	/* Memory manager */
1146 	r = amdgpu_bo_init(adev);
1147 	if (r)
1148 		return r;
1149 
1150 	r = gmc_v8_0_gart_init(adev);
1151 	if (r)
1152 		return r;
1153 
1154 	/*
1155 	 * number of VMs
1156 	 * VMID 0 is reserved for System
1157 	 * amdgpu graphics/compute will use VMIDs 1-7
1158 	 * amdkfd will use VMIDs 8-15
1159 	 */
1160 	adev->vm_manager.first_kfd_vmid = 8;
1161 	amdgpu_vm_manager_init(adev);
1162 
1163 	/* base offset of vram pages */
1164 	if (adev->flags & AMD_IS_APU) {
1165 		u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
1166 
1167 		tmp <<= 22;
1168 		adev->vm_manager.vram_base_offset = tmp;
1169 	} else {
1170 		adev->vm_manager.vram_base_offset = 0;
1171 	}
1172 
1173 	adev->gmc.vm_fault_info = kmalloc(sizeof(struct kfd_vm_fault_info),
1174 					GFP_KERNEL);
1175 	if (!adev->gmc.vm_fault_info)
1176 		return -ENOMEM;
1177 	atomic_set(&adev->gmc.vm_fault_info_updated, 0);
1178 
1179 	return 0;
1180 }
1181 
1182 static int gmc_v8_0_sw_fini(void *handle)
1183 {
1184 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1185 
1186 	amdgpu_gem_force_release(adev);
1187 	amdgpu_vm_manager_fini(adev);
1188 	kfree(adev->gmc.vm_fault_info);
1189 	amdgpu_gart_table_vram_free(adev);
1190 	amdgpu_bo_fini(adev);
1191 	amdgpu_ucode_release(&adev->gmc.fw);
1192 
1193 	return 0;
1194 }
1195 
1196 static int gmc_v8_0_hw_init(void *handle)
1197 {
1198 	int r;
1199 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1200 
1201 	gmc_v8_0_init_golden_registers(adev);
1202 
1203 	gmc_v8_0_mc_program(adev);
1204 
1205 	if (adev->asic_type == CHIP_TONGA) {
1206 		r = gmc_v8_0_tonga_mc_load_microcode(adev);
1207 		if (r) {
1208 			DRM_ERROR("Failed to load MC firmware!\n");
1209 			return r;
1210 		}
1211 	} else if (adev->asic_type == CHIP_POLARIS11 ||
1212 			adev->asic_type == CHIP_POLARIS10 ||
1213 			adev->asic_type == CHIP_POLARIS12) {
1214 		r = gmc_v8_0_polaris_mc_load_microcode(adev);
1215 		if (r) {
1216 			DRM_ERROR("Failed to load MC firmware!\n");
1217 			return r;
1218 		}
1219 	}
1220 
1221 	r = gmc_v8_0_gart_enable(adev);
1222 	if (r)
1223 		return r;
1224 
1225 	if (amdgpu_emu_mode == 1)
1226 		return amdgpu_gmc_vram_checking(adev);
1227 	else
1228 		return r;
1229 }
1230 
1231 static int gmc_v8_0_hw_fini(void *handle)
1232 {
1233 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1234 
1235 	amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
1236 	gmc_v8_0_gart_disable(adev);
1237 
1238 	return 0;
1239 }
1240 
1241 static int gmc_v8_0_suspend(void *handle)
1242 {
1243 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1244 
1245 	gmc_v8_0_hw_fini(adev);
1246 
1247 	return 0;
1248 }
1249 
1250 static int gmc_v8_0_resume(void *handle)
1251 {
1252 	int r;
1253 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1254 
1255 	r = gmc_v8_0_hw_init(adev);
1256 	if (r)
1257 		return r;
1258 
1259 	amdgpu_vmid_reset_all(adev);
1260 
1261 	return 0;
1262 }
1263 
1264 static bool gmc_v8_0_is_idle(void *handle)
1265 {
1266 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1267 	u32 tmp = RREG32(mmSRBM_STATUS);
1268 
1269 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1270 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
1271 		return false;
1272 
1273 	return true;
1274 }
1275 
1276 static int gmc_v8_0_wait_for_idle(void *handle)
1277 {
1278 	unsigned int i;
1279 	u32 tmp;
1280 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1281 
1282 	for (i = 0; i < adev->usec_timeout; i++) {
1283 		/* read MC_STATUS */
1284 		tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
1285 					       SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1286 					       SRBM_STATUS__MCC_BUSY_MASK |
1287 					       SRBM_STATUS__MCD_BUSY_MASK |
1288 					       SRBM_STATUS__VMC_BUSY_MASK |
1289 					       SRBM_STATUS__VMC1_BUSY_MASK);
1290 		if (!tmp)
1291 			return 0;
1292 		udelay(1);
1293 	}
1294 	return -ETIMEDOUT;
1295 
1296 }
1297 
1298 static bool gmc_v8_0_check_soft_reset(void *handle)
1299 {
1300 	u32 srbm_soft_reset = 0;
1301 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1302 	u32 tmp = RREG32(mmSRBM_STATUS);
1303 
1304 	if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1305 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1306 						SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1307 
1308 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1309 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1310 		if (!(adev->flags & AMD_IS_APU))
1311 			srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1312 							SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1313 	}
1314 
1315 	if (srbm_soft_reset) {
1316 		adev->gmc.srbm_soft_reset = srbm_soft_reset;
1317 		return true;
1318 	}
1319 
1320 	adev->gmc.srbm_soft_reset = 0;
1321 
1322 	return false;
1323 }
1324 
1325 static int gmc_v8_0_pre_soft_reset(void *handle)
1326 {
1327 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1328 
1329 	if (!adev->gmc.srbm_soft_reset)
1330 		return 0;
1331 
1332 	gmc_v8_0_mc_stop(adev);
1333 	if (gmc_v8_0_wait_for_idle(adev))
1334 		dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1335 
1336 	return 0;
1337 }
1338 
1339 static int gmc_v8_0_soft_reset(void *handle)
1340 {
1341 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1342 	u32 srbm_soft_reset;
1343 
1344 	if (!adev->gmc.srbm_soft_reset)
1345 		return 0;
1346 	srbm_soft_reset = adev->gmc.srbm_soft_reset;
1347 
1348 	if (srbm_soft_reset) {
1349 		u32 tmp;
1350 
1351 		tmp = RREG32(mmSRBM_SOFT_RESET);
1352 		tmp |= srbm_soft_reset;
1353 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1354 		WREG32(mmSRBM_SOFT_RESET, tmp);
1355 		tmp = RREG32(mmSRBM_SOFT_RESET);
1356 
1357 		udelay(50);
1358 
1359 		tmp &= ~srbm_soft_reset;
1360 		WREG32(mmSRBM_SOFT_RESET, tmp);
1361 		tmp = RREG32(mmSRBM_SOFT_RESET);
1362 
1363 		/* Wait a little for things to settle down */
1364 		udelay(50);
1365 	}
1366 
1367 	return 0;
1368 }
1369 
1370 static int gmc_v8_0_post_soft_reset(void *handle)
1371 {
1372 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1373 
1374 	if (!adev->gmc.srbm_soft_reset)
1375 		return 0;
1376 
1377 	gmc_v8_0_mc_resume(adev);
1378 	return 0;
1379 }
1380 
1381 static int gmc_v8_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1382 					     struct amdgpu_irq_src *src,
1383 					     unsigned int type,
1384 					     enum amdgpu_interrupt_state state)
1385 {
1386 	u32 tmp;
1387 	u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1388 		    VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1389 		    VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1390 		    VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1391 		    VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1392 		    VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1393 		    VM_CONTEXT1_CNTL__EXECUTE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1394 
1395 	switch (state) {
1396 	case AMDGPU_IRQ_STATE_DISABLE:
1397 		/* system context */
1398 		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1399 		tmp &= ~bits;
1400 		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1401 		/* VMs */
1402 		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1403 		tmp &= ~bits;
1404 		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1405 		break;
1406 	case AMDGPU_IRQ_STATE_ENABLE:
1407 		/* system context */
1408 		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1409 		tmp |= bits;
1410 		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1411 		/* VMs */
1412 		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1413 		tmp |= bits;
1414 		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1415 		break;
1416 	default:
1417 		break;
1418 	}
1419 
1420 	return 0;
1421 }
1422 
1423 static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
1424 				      struct amdgpu_irq_src *source,
1425 				      struct amdgpu_iv_entry *entry)
1426 {
1427 	u32 addr, status, mc_client, vmid;
1428 
1429 	if (amdgpu_sriov_vf(adev)) {
1430 		dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1431 			entry->src_id, entry->src_data[0]);
1432 		dev_err(adev->dev, " Can't decode VM fault info here on SRIOV VF\n");
1433 		return 0;
1434 	}
1435 
1436 	addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1437 	status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1438 	mc_client = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_MCCLIENT);
1439 	/* reset addr and status */
1440 	WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1441 
1442 	if (!addr && !status)
1443 		return 0;
1444 
1445 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1446 		gmc_v8_0_set_fault_enable_default(adev, false);
1447 
1448 	if (printk_ratelimit()) {
1449 		struct amdgpu_task_info task_info;
1450 
1451 		memset(&task_info, 0, sizeof(struct amdgpu_task_info));
1452 		amdgpu_vm_get_task_info(adev, entry->pasid, &task_info);
1453 
1454 		dev_err(adev->dev, "GPU fault detected: %d 0x%08x for process %s pid %d thread %s pid %d\n",
1455 			entry->src_id, entry->src_data[0], task_info.process_name,
1456 			task_info.tgid, task_info.task_name, task_info.pid);
1457 		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1458 			addr);
1459 		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1460 			status);
1461 		gmc_v8_0_vm_decode_fault(adev, status, addr, mc_client,
1462 					 entry->pasid);
1463 	}
1464 
1465 	vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1466 			     VMID);
1467 	if (amdgpu_amdkfd_is_kfd_vmid(adev, vmid)
1468 		&& !atomic_read(&adev->gmc.vm_fault_info_updated)) {
1469 		struct kfd_vm_fault_info *info = adev->gmc.vm_fault_info;
1470 		u32 protections = REG_GET_FIELD(status,
1471 					VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1472 					PROTECTIONS);
1473 
1474 		info->vmid = vmid;
1475 		info->mc_id = REG_GET_FIELD(status,
1476 					    VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1477 					    MEMORY_CLIENT_ID);
1478 		info->status = status;
1479 		info->page_addr = addr;
1480 		info->prot_valid = protections & 0x7 ? true : false;
1481 		info->prot_read = protections & 0x8 ? true : false;
1482 		info->prot_write = protections & 0x10 ? true : false;
1483 		info->prot_exec = protections & 0x20 ? true : false;
1484 		mb();
1485 		atomic_set(&adev->gmc.vm_fault_info_updated, 1);
1486 	}
1487 
1488 	return 0;
1489 }
1490 
1491 static void fiji_update_mc_medium_grain_clock_gating(struct amdgpu_device *adev,
1492 						     bool enable)
1493 {
1494 	uint32_t data;
1495 
1496 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG)) {
1497 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1498 		data |= MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1499 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1500 
1501 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1502 		data |= MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1503 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1504 
1505 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1506 		data |= MC_HUB_MISC_VM_CG__ENABLE_MASK;
1507 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1508 
1509 		data = RREG32(mmMC_XPB_CLK_GAT);
1510 		data |= MC_XPB_CLK_GAT__ENABLE_MASK;
1511 		WREG32(mmMC_XPB_CLK_GAT, data);
1512 
1513 		data = RREG32(mmATC_MISC_CG);
1514 		data |= ATC_MISC_CG__ENABLE_MASK;
1515 		WREG32(mmATC_MISC_CG, data);
1516 
1517 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1518 		data |= MC_CITF_MISC_WR_CG__ENABLE_MASK;
1519 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1520 
1521 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1522 		data |= MC_CITF_MISC_RD_CG__ENABLE_MASK;
1523 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1524 
1525 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1526 		data |= MC_CITF_MISC_VM_CG__ENABLE_MASK;
1527 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1528 
1529 		data = RREG32(mmVM_L2_CG);
1530 		data |= VM_L2_CG__ENABLE_MASK;
1531 		WREG32(mmVM_L2_CG, data);
1532 	} else {
1533 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1534 		data &= ~MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1535 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1536 
1537 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1538 		data &= ~MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1539 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1540 
1541 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1542 		data &= ~MC_HUB_MISC_VM_CG__ENABLE_MASK;
1543 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1544 
1545 		data = RREG32(mmMC_XPB_CLK_GAT);
1546 		data &= ~MC_XPB_CLK_GAT__ENABLE_MASK;
1547 		WREG32(mmMC_XPB_CLK_GAT, data);
1548 
1549 		data = RREG32(mmATC_MISC_CG);
1550 		data &= ~ATC_MISC_CG__ENABLE_MASK;
1551 		WREG32(mmATC_MISC_CG, data);
1552 
1553 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1554 		data &= ~MC_CITF_MISC_WR_CG__ENABLE_MASK;
1555 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1556 
1557 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1558 		data &= ~MC_CITF_MISC_RD_CG__ENABLE_MASK;
1559 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1560 
1561 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1562 		data &= ~MC_CITF_MISC_VM_CG__ENABLE_MASK;
1563 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1564 
1565 		data = RREG32(mmVM_L2_CG);
1566 		data &= ~VM_L2_CG__ENABLE_MASK;
1567 		WREG32(mmVM_L2_CG, data);
1568 	}
1569 }
1570 
1571 static void fiji_update_mc_light_sleep(struct amdgpu_device *adev,
1572 				       bool enable)
1573 {
1574 	uint32_t data;
1575 
1576 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_LS)) {
1577 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1578 		data |= MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1579 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1580 
1581 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1582 		data |= MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1583 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1584 
1585 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1586 		data |= MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1587 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1588 
1589 		data = RREG32(mmMC_XPB_CLK_GAT);
1590 		data |= MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1591 		WREG32(mmMC_XPB_CLK_GAT, data);
1592 
1593 		data = RREG32(mmATC_MISC_CG);
1594 		data |= ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1595 		WREG32(mmATC_MISC_CG, data);
1596 
1597 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1598 		data |= MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1599 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1600 
1601 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1602 		data |= MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1603 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1604 
1605 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1606 		data |= MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1607 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1608 
1609 		data = RREG32(mmVM_L2_CG);
1610 		data |= VM_L2_CG__MEM_LS_ENABLE_MASK;
1611 		WREG32(mmVM_L2_CG, data);
1612 	} else {
1613 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1614 		data &= ~MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1615 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1616 
1617 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1618 		data &= ~MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1619 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1620 
1621 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1622 		data &= ~MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1623 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1624 
1625 		data = RREG32(mmMC_XPB_CLK_GAT);
1626 		data &= ~MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1627 		WREG32(mmMC_XPB_CLK_GAT, data);
1628 
1629 		data = RREG32(mmATC_MISC_CG);
1630 		data &= ~ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1631 		WREG32(mmATC_MISC_CG, data);
1632 
1633 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1634 		data &= ~MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1635 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1636 
1637 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1638 		data &= ~MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1639 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1640 
1641 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1642 		data &= ~MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1643 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1644 
1645 		data = RREG32(mmVM_L2_CG);
1646 		data &= ~VM_L2_CG__MEM_LS_ENABLE_MASK;
1647 		WREG32(mmVM_L2_CG, data);
1648 	}
1649 }
1650 
1651 static int gmc_v8_0_set_clockgating_state(void *handle,
1652 					  enum amd_clockgating_state state)
1653 {
1654 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1655 
1656 	if (amdgpu_sriov_vf(adev))
1657 		return 0;
1658 
1659 	switch (adev->asic_type) {
1660 	case CHIP_FIJI:
1661 		fiji_update_mc_medium_grain_clock_gating(adev,
1662 				state == AMD_CG_STATE_GATE);
1663 		fiji_update_mc_light_sleep(adev,
1664 				state == AMD_CG_STATE_GATE);
1665 		break;
1666 	default:
1667 		break;
1668 	}
1669 	return 0;
1670 }
1671 
1672 static int gmc_v8_0_set_powergating_state(void *handle,
1673 					  enum amd_powergating_state state)
1674 {
1675 	return 0;
1676 }
1677 
1678 static void gmc_v8_0_get_clockgating_state(void *handle, u64 *flags)
1679 {
1680 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1681 	int data;
1682 
1683 	if (amdgpu_sriov_vf(adev))
1684 		*flags = 0;
1685 
1686 	/* AMD_CG_SUPPORT_MC_MGCG */
1687 	data = RREG32(mmMC_HUB_MISC_HUB_CG);
1688 	if (data & MC_HUB_MISC_HUB_CG__ENABLE_MASK)
1689 		*flags |= AMD_CG_SUPPORT_MC_MGCG;
1690 
1691 	/* AMD_CG_SUPPORT_MC_LS */
1692 	if (data & MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK)
1693 		*flags |= AMD_CG_SUPPORT_MC_LS;
1694 }
1695 
1696 static const struct amd_ip_funcs gmc_v8_0_ip_funcs = {
1697 	.name = "gmc_v8_0",
1698 	.early_init = gmc_v8_0_early_init,
1699 	.late_init = gmc_v8_0_late_init,
1700 	.sw_init = gmc_v8_0_sw_init,
1701 	.sw_fini = gmc_v8_0_sw_fini,
1702 	.hw_init = gmc_v8_0_hw_init,
1703 	.hw_fini = gmc_v8_0_hw_fini,
1704 	.suspend = gmc_v8_0_suspend,
1705 	.resume = gmc_v8_0_resume,
1706 	.is_idle = gmc_v8_0_is_idle,
1707 	.wait_for_idle = gmc_v8_0_wait_for_idle,
1708 	.check_soft_reset = gmc_v8_0_check_soft_reset,
1709 	.pre_soft_reset = gmc_v8_0_pre_soft_reset,
1710 	.soft_reset = gmc_v8_0_soft_reset,
1711 	.post_soft_reset = gmc_v8_0_post_soft_reset,
1712 	.set_clockgating_state = gmc_v8_0_set_clockgating_state,
1713 	.set_powergating_state = gmc_v8_0_set_powergating_state,
1714 	.get_clockgating_state = gmc_v8_0_get_clockgating_state,
1715 };
1716 
1717 static const struct amdgpu_gmc_funcs gmc_v8_0_gmc_funcs = {
1718 	.flush_gpu_tlb = gmc_v8_0_flush_gpu_tlb,
1719 	.flush_gpu_tlb_pasid = gmc_v8_0_flush_gpu_tlb_pasid,
1720 	.emit_flush_gpu_tlb = gmc_v8_0_emit_flush_gpu_tlb,
1721 	.emit_pasid_mapping = gmc_v8_0_emit_pasid_mapping,
1722 	.set_prt = gmc_v8_0_set_prt,
1723 	.get_vm_pde = gmc_v8_0_get_vm_pde,
1724 	.get_vm_pte = gmc_v8_0_get_vm_pte,
1725 	.get_vbios_fb_size = gmc_v8_0_get_vbios_fb_size,
1726 };
1727 
1728 static const struct amdgpu_irq_src_funcs gmc_v8_0_irq_funcs = {
1729 	.set = gmc_v8_0_vm_fault_interrupt_state,
1730 	.process = gmc_v8_0_process_interrupt,
1731 };
1732 
1733 static void gmc_v8_0_set_gmc_funcs(struct amdgpu_device *adev)
1734 {
1735 	adev->gmc.gmc_funcs = &gmc_v8_0_gmc_funcs;
1736 }
1737 
1738 static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev)
1739 {
1740 	adev->gmc.vm_fault.num_types = 1;
1741 	adev->gmc.vm_fault.funcs = &gmc_v8_0_irq_funcs;
1742 }
1743 
1744 const struct amdgpu_ip_block_version gmc_v8_0_ip_block = {
1745 	.type = AMD_IP_BLOCK_TYPE_GMC,
1746 	.major = 8,
1747 	.minor = 0,
1748 	.rev = 0,
1749 	.funcs = &gmc_v8_0_ip_funcs,
1750 };
1751 
1752 const struct amdgpu_ip_block_version gmc_v8_1_ip_block = {
1753 	.type = AMD_IP_BLOCK_TYPE_GMC,
1754 	.major = 8,
1755 	.minor = 1,
1756 	.rev = 0,
1757 	.funcs = &gmc_v8_0_ip_funcs,
1758 };
1759 
1760 const struct amdgpu_ip_block_version gmc_v8_5_ip_block = {
1761 	.type = AMD_IP_BLOCK_TYPE_GMC,
1762 	.major = 8,
1763 	.minor = 5,
1764 	.rev = 0,
1765 	.funcs = &gmc_v8_0_ip_funcs,
1766 };
1767