xref: /linux/drivers/gpu/drm/i915/gt/intel_rc6.c (revision 815e260a18a3af4dab59025ee99a7156c0e8b5e0)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright © 2019 Intel Corporation
4  */
5 
6 #include <linux/pm_runtime.h>
7 #include <linux/string_helpers.h>
8 
9 #include <drm/drm_print.h>
10 
11 #include "display/vlv_clock.h"
12 #include "gem/i915_gem_region.h"
13 #include "i915_drv.h"
14 #include "i915_reg.h"
15 #include "i915_vgpu.h"
16 #include "intel_engine_regs.h"
17 #include "intel_gt.h"
18 #include "intel_gt_pm.h"
19 #include "intel_gt_regs.h"
20 #include "intel_pcode.h"
21 #include "intel_rc6.h"
22 
23 /**
24  * DOC: RC6
25  *
26  * RC6 is a special power stage which allows the GPU to enter an very
27  * low-voltage mode when idle, using down to 0V while at this stage.  This
28  * stage is entered automatically when the GPU is idle when RC6 support is
29  * enabled, and as soon as new workload arises GPU wakes up automatically as
30  * well.
31  *
32  * There are different RC6 modes available in Intel GPU, which differentiate
33  * among each other with the latency required to enter and leave RC6 and
34  * voltage consumed by the GPU in different states.
35  *
36  * The combination of the following flags define which states GPU is allowed
37  * to enter, while RC6 is the normal RC6 state, RC6p is the deep RC6, and
38  * RC6pp is deepest RC6. Their support by hardware varies according to the
39  * GPU, BIOS, chipset and platform. RC6 is usually the safest one and the one
40  * which brings the most power savings; deeper states save more power, but
41  * require higher latency to switch to and wake up.
42  */
43 
44 static struct intel_gt *rc6_to_gt(struct intel_rc6 *rc6)
45 {
46 	return container_of(rc6, struct intel_gt, rc6);
47 }
48 
49 static struct intel_uncore *rc6_to_uncore(struct intel_rc6 *rc)
50 {
51 	return rc6_to_gt(rc)->uncore;
52 }
53 
54 static struct drm_i915_private *rc6_to_i915(struct intel_rc6 *rc)
55 {
56 	return rc6_to_gt(rc)->i915;
57 }
58 
59 static void gen11_rc6_enable(struct intel_rc6 *rc6)
60 {
61 	struct intel_gt *gt = rc6_to_gt(rc6);
62 	struct intel_uncore *uncore = gt->uncore;
63 	struct intel_engine_cs *engine;
64 	enum intel_engine_id id;
65 	u32 pg_enable;
66 	int i;
67 
68 	/*
69 	 * With GuCRC, these parameters are set by GuC
70 	 */
71 	if (!intel_uc_uses_guc_rc(&gt->uc)) {
72 		/* 2b: Program RC6 thresholds.*/
73 		intel_uncore_write_fw(uncore, GEN6_RC6_WAKE_RATE_LIMIT, 54 << 16 | 85);
74 		intel_uncore_write_fw(uncore, GEN10_MEDIA_WAKE_RATE_LIMIT, 150);
75 
76 		intel_uncore_write_fw(uncore, GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
77 		intel_uncore_write_fw(uncore, GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
78 		for_each_engine(engine, rc6_to_gt(rc6), id)
79 			intel_uncore_write_fw(uncore, RING_MAX_IDLE(engine->mmio_base), 10);
80 
81 		intel_uncore_write_fw(uncore, GUC_MAX_IDLE_COUNT, 0xA);
82 
83 		intel_uncore_write_fw(uncore, GEN6_RC_SLEEP, 0);
84 
85 		intel_uncore_write_fw(uncore, GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */
86 	}
87 
88 	/*
89 	 * 2c: Program Coarse Power Gating Policies.
90 	 *
91 	 * Bspec's guidance is to use 25us (really 25 * 1280ns) here. What we
92 	 * use instead is a more conservative estimate for the maximum time
93 	 * it takes us to service a CS interrupt and submit a new ELSP - that
94 	 * is the time which the GPU is idle waiting for the CPU to select the
95 	 * next request to execute. If the idle hysteresis is less than that
96 	 * interrupt service latency, the hardware will automatically gate
97 	 * the power well and we will then incur the wake up cost on top of
98 	 * the service latency. A similar guide from plane_state is that we
99 	 * do not want the enable hysteresis to less than the wakeup latency.
100 	 *
101 	 * igt/gem_exec_nop/sequential provides a rough estimate for the
102 	 * service latency, and puts it under 10us for Icelake, similar to
103 	 * Broadwell+, To be conservative, we want to factor in a context
104 	 * switch on top (due to ksoftirqd).
105 	 */
106 	intel_uncore_write_fw(uncore, GEN9_MEDIA_PG_IDLE_HYSTERESIS, 60);
107 	intel_uncore_write_fw(uncore, GEN9_RENDER_PG_IDLE_HYSTERESIS, 60);
108 
109 	/* 3a: Enable RC6
110 	 *
111 	 * With GuCRC, we do not enable bit 31 of RC_CTL,
112 	 * thus allowing GuC to control RC6 entry/exit fully instead.
113 	 * We will not set the HW ENABLE and EI bits
114 	 */
115 	if (!intel_guc_rc_enable(gt_to_guc(gt)))
116 		rc6->ctl_enable = GEN6_RC_CTL_RC6_ENABLE;
117 	else
118 		rc6->ctl_enable =
119 			GEN6_RC_CTL_HW_ENABLE |
120 			GEN6_RC_CTL_RC6_ENABLE |
121 			GEN6_RC_CTL_EI_MODE(1);
122 
123 	pg_enable =
124 		GEN9_RENDER_PG_ENABLE |
125 		GEN9_MEDIA_PG_ENABLE |
126 		GEN11_MEDIA_SAMPLER_PG_ENABLE;
127 
128 	if (GRAPHICS_VER(gt->i915) >= 12 && !IS_DG1(gt->i915)) {
129 		for (i = 0; i < I915_MAX_VCS; i++)
130 			if (HAS_ENGINE(gt, _VCS(i)))
131 				pg_enable |= (VDN_HCP_POWERGATE_ENABLE(i) |
132 					      VDN_MFX_POWERGATE_ENABLE(i));
133 	}
134 
135 	intel_uncore_write_fw(uncore, GEN9_PG_ENABLE, pg_enable);
136 }
137 
138 static void gen9_rc6_enable(struct intel_rc6 *rc6)
139 {
140 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
141 	struct intel_engine_cs *engine;
142 	enum intel_engine_id id;
143 
144 	/* 2b: Program RC6 thresholds.*/
145 	if (GRAPHICS_VER(rc6_to_i915(rc6)) >= 11) {
146 		intel_uncore_write_fw(uncore, GEN6_RC6_WAKE_RATE_LIMIT, 54 << 16 | 85);
147 		intel_uncore_write_fw(uncore, GEN10_MEDIA_WAKE_RATE_LIMIT, 150);
148 	} else if (IS_SKYLAKE(rc6_to_i915(rc6))) {
149 		/*
150 		 * WaRsDoubleRc6WrlWithCoarsePowerGating:skl Doubling WRL only
151 		 * when CPG is enabled
152 		 */
153 		intel_uncore_write_fw(uncore, GEN6_RC6_WAKE_RATE_LIMIT, 108 << 16);
154 	} else {
155 		intel_uncore_write_fw(uncore, GEN6_RC6_WAKE_RATE_LIMIT, 54 << 16);
156 	}
157 
158 	intel_uncore_write_fw(uncore, GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
159 	intel_uncore_write_fw(uncore, GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
160 	for_each_engine(engine, rc6_to_gt(rc6), id)
161 		intel_uncore_write_fw(uncore, RING_MAX_IDLE(engine->mmio_base), 10);
162 
163 	intel_uncore_write_fw(uncore, GUC_MAX_IDLE_COUNT, 0xA);
164 
165 	intel_uncore_write_fw(uncore, GEN6_RC_SLEEP, 0);
166 
167 	/*
168 	 * 2c: Program Coarse Power Gating Policies.
169 	 *
170 	 * Bspec's guidance is to use 25us (really 25 * 1280ns) here. What we
171 	 * use instead is a more conservative estimate for the maximum time
172 	 * it takes us to service a CS interrupt and submit a new ELSP - that
173 	 * is the time which the GPU is idle waiting for the CPU to select the
174 	 * next request to execute. If the idle hysteresis is less than that
175 	 * interrupt service latency, the hardware will automatically gate
176 	 * the power well and we will then incur the wake up cost on top of
177 	 * the service latency. A similar guide from plane_state is that we
178 	 * do not want the enable hysteresis to less than the wakeup latency.
179 	 *
180 	 * igt/gem_exec_nop/sequential provides a rough estimate for the
181 	 * service latency, and puts it around 10us for Broadwell (and other
182 	 * big core) and around 40us for Broxton (and other low power cores).
183 	 * [Note that for legacy ringbuffer submission, this is less than 1us!]
184 	 * However, the wakeup latency on Broxton is closer to 100us. To be
185 	 * conservative, we have to factor in a context switch on top (due
186 	 * to ksoftirqd).
187 	 */
188 	intel_uncore_write_fw(uncore, GEN9_MEDIA_PG_IDLE_HYSTERESIS, 250);
189 	intel_uncore_write_fw(uncore, GEN9_RENDER_PG_IDLE_HYSTERESIS, 250);
190 
191 	/* 3a: Enable RC6 */
192 	intel_uncore_write_fw(uncore, GEN6_RC6_THRESHOLD, 37500); /* 37.5/125ms per EI */
193 
194 	rc6->ctl_enable =
195 		GEN6_RC_CTL_HW_ENABLE |
196 		GEN6_RC_CTL_RC6_ENABLE |
197 		GEN6_RC_CTL_EI_MODE(1);
198 
199 	/*
200 	 * WaRsDisableCoarsePowerGating:skl,cnl
201 	 *   - Render/Media PG need to be disabled with RC6.
202 	 */
203 	if (!NEEDS_WaRsDisableCoarsePowerGating(rc6_to_i915(rc6)))
204 		intel_uncore_write_fw(uncore, GEN9_PG_ENABLE,
205 				      GEN9_RENDER_PG_ENABLE | GEN9_MEDIA_PG_ENABLE);
206 }
207 
208 static void gen8_rc6_enable(struct intel_rc6 *rc6)
209 {
210 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
211 	struct intel_engine_cs *engine;
212 	enum intel_engine_id id;
213 
214 	/* 2b: Program RC6 thresholds.*/
215 	intel_uncore_write_fw(uncore, GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16);
216 	intel_uncore_write_fw(uncore, GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
217 	intel_uncore_write_fw(uncore, GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
218 	for_each_engine(engine, rc6_to_gt(rc6), id)
219 		intel_uncore_write_fw(uncore, RING_MAX_IDLE(engine->mmio_base), 10);
220 	intel_uncore_write_fw(uncore, GEN6_RC_SLEEP, 0);
221 	intel_uncore_write_fw(uncore, GEN6_RC6_THRESHOLD, 625); /* 800us/1.28 for TO */
222 
223 	/* 3: Enable RC6 */
224 	rc6->ctl_enable =
225 	    GEN6_RC_CTL_HW_ENABLE |
226 	    GEN7_RC_CTL_TO_MODE |
227 	    GEN6_RC_CTL_RC6_ENABLE;
228 }
229 
230 static void gen6_rc6_enable(struct intel_rc6 *rc6)
231 {
232 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
233 	struct drm_i915_private *i915 = rc6_to_i915(rc6);
234 	struct intel_engine_cs *engine;
235 	enum intel_engine_id id;
236 	u32 rc6vids, rc6_mask;
237 	int ret;
238 
239 	intel_uncore_write_fw(uncore, GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16);
240 	intel_uncore_write_fw(uncore, GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30);
241 	intel_uncore_write_fw(uncore, GEN6_RC6pp_WAKE_RATE_LIMIT, 30);
242 	intel_uncore_write_fw(uncore, GEN6_RC_EVALUATION_INTERVAL, 125000);
243 	intel_uncore_write_fw(uncore, GEN6_RC_IDLE_HYSTERSIS, 25);
244 
245 	for_each_engine(engine, rc6_to_gt(rc6), id)
246 		intel_uncore_write_fw(uncore, RING_MAX_IDLE(engine->mmio_base), 10);
247 
248 	intel_uncore_write_fw(uncore, GEN6_RC_SLEEP, 0);
249 	intel_uncore_write_fw(uncore, GEN6_RC1e_THRESHOLD, 1000);
250 	intel_uncore_write_fw(uncore, GEN6_RC6_THRESHOLD, 50000);
251 	intel_uncore_write_fw(uncore, GEN6_RC6p_THRESHOLD, 150000);
252 	intel_uncore_write_fw(uncore, GEN6_RC6pp_THRESHOLD, 64000); /* unused */
253 
254 	/* We don't use those on Haswell */
255 	rc6_mask = GEN6_RC_CTL_RC6_ENABLE;
256 	if (HAS_RC6p(i915))
257 		rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE;
258 	if (HAS_RC6pp(i915))
259 		rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE;
260 	rc6->ctl_enable =
261 	    rc6_mask |
262 	    GEN6_RC_CTL_EI_MODE(1) |
263 	    GEN6_RC_CTL_HW_ENABLE;
264 
265 	rc6vids = 0;
266 	ret = snb_pcode_read(rc6_to_gt(rc6)->uncore, GEN6_PCODE_READ_RC6VIDS, &rc6vids, NULL);
267 	if (GRAPHICS_VER(i915) == 6 && ret) {
268 		drm_dbg(&i915->drm, "Couldn't check for BIOS workaround\n");
269 	} else if (GRAPHICS_VER(i915) == 6 &&
270 		   (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) {
271 		drm_dbg(&i915->drm,
272 			"You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
273 			GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450);
274 		rc6vids &= 0xffff00;
275 		rc6vids |= GEN6_ENCODE_RC6_VID(450);
276 		ret = snb_pcode_write(rc6_to_gt(rc6)->uncore, GEN6_PCODE_WRITE_RC6VIDS, rc6vids);
277 		if (ret)
278 			drm_err(&i915->drm,
279 				"Couldn't fix incorrect rc6 voltage\n");
280 	}
281 }
282 
283 /* Check that the pcbr address is not empty. */
284 static int chv_rc6_init(struct intel_rc6 *rc6)
285 {
286 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
287 	struct drm_i915_private *i915 = rc6_to_i915(rc6);
288 	resource_size_t pctx_paddr, paddr;
289 	resource_size_t pctx_size = 32 * SZ_1K;
290 	u32 pcbr;
291 
292 	pcbr = intel_uncore_read(uncore, VLV_PCBR);
293 	if ((pcbr >> VLV_PCBR_ADDR_SHIFT) == 0) {
294 		drm_dbg(&i915->drm, "BIOS didn't set up PCBR, fixing up\n");
295 		paddr = i915->dsm.stolen.end + 1 - pctx_size;
296 		GEM_BUG_ON(paddr > U32_MAX);
297 
298 		pctx_paddr = (paddr & ~4095);
299 		intel_uncore_write(uncore, VLV_PCBR, pctx_paddr);
300 	}
301 
302 	return 0;
303 }
304 
305 static int vlv_rc6_init(struct intel_rc6 *rc6)
306 {
307 	struct drm_i915_private *i915 = rc6_to_i915(rc6);
308 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
309 	struct drm_i915_gem_object *pctx;
310 	resource_size_t pctx_paddr;
311 	resource_size_t pctx_size = 24 * SZ_1K;
312 	u32 pcbr;
313 
314 	pcbr = intel_uncore_read(uncore, VLV_PCBR);
315 	if (pcbr) {
316 		/* BIOS set it up already, grab the pre-alloc'd space */
317 		resource_size_t pcbr_offset;
318 
319 		pcbr_offset = (pcbr & ~4095) - i915->dsm.stolen.start;
320 		pctx = i915_gem_object_create_region_at(i915->mm.stolen_region,
321 							pcbr_offset,
322 							pctx_size,
323 							0);
324 		if (IS_ERR(pctx))
325 			return PTR_ERR(pctx);
326 
327 		goto out;
328 	}
329 
330 	drm_dbg(&i915->drm, "BIOS didn't set up PCBR, fixing up\n");
331 
332 	/*
333 	 * From the Gunit register HAS:
334 	 * The Gfx driver is expected to program this register and ensure
335 	 * proper allocation within Gfx stolen memory.  For example, this
336 	 * register should be programmed such than the PCBR range does not
337 	 * overlap with other ranges, such as the frame buffer, protected
338 	 * memory, or any other relevant ranges.
339 	 */
340 	pctx = i915_gem_object_create_stolen(i915, pctx_size);
341 	if (IS_ERR(pctx)) {
342 		drm_dbg(&i915->drm,
343 			"not enough stolen space for PCTX, disabling\n");
344 		return PTR_ERR(pctx);
345 	}
346 
347 	GEM_BUG_ON(range_end_overflows_t(u64,
348 					 i915->dsm.stolen.start,
349 					 pctx->stolen->start,
350 					 U32_MAX));
351 	pctx_paddr = i915->dsm.stolen.start + pctx->stolen->start;
352 	intel_uncore_write(uncore, VLV_PCBR, pctx_paddr);
353 
354 out:
355 	rc6->pctx = pctx;
356 	return 0;
357 }
358 
359 static void chv_rc6_enable(struct intel_rc6 *rc6)
360 {
361 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
362 	struct intel_engine_cs *engine;
363 	enum intel_engine_id id;
364 
365 	/* 2a: Program RC6 thresholds.*/
366 	intel_uncore_write_fw(uncore, GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16);
367 	intel_uncore_write_fw(uncore, GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
368 	intel_uncore_write_fw(uncore, GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
369 
370 	for_each_engine(engine, rc6_to_gt(rc6), id)
371 		intel_uncore_write_fw(uncore, RING_MAX_IDLE(engine->mmio_base), 10);
372 	intel_uncore_write_fw(uncore, GEN6_RC_SLEEP, 0);
373 
374 	/* TO threshold set to 500 us (0x186 * 1.28 us) */
375 	intel_uncore_write_fw(uncore, GEN6_RC6_THRESHOLD, 0x186);
376 
377 	/* Allows RC6 residency counter to work */
378 	intel_uncore_write_fw(uncore, VLV_COUNTER_CONTROL,
379 			      _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH |
380 						 VLV_MEDIA_RC6_COUNT_EN |
381 						 VLV_RENDER_RC6_COUNT_EN));
382 
383 	/* 3: Enable RC6 */
384 	rc6->ctl_enable = GEN7_RC_CTL_TO_MODE;
385 }
386 
387 static void vlv_rc6_enable(struct intel_rc6 *rc6)
388 {
389 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
390 	struct intel_engine_cs *engine;
391 	enum intel_engine_id id;
392 
393 	intel_uncore_write_fw(uncore, GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000);
394 	intel_uncore_write_fw(uncore, GEN6_RC_EVALUATION_INTERVAL, 125000);
395 	intel_uncore_write_fw(uncore, GEN6_RC_IDLE_HYSTERSIS, 25);
396 
397 	for_each_engine(engine, rc6_to_gt(rc6), id)
398 		intel_uncore_write_fw(uncore, RING_MAX_IDLE(engine->mmio_base), 10);
399 
400 	intel_uncore_write_fw(uncore, GEN6_RC6_THRESHOLD, 0x557);
401 
402 	/* Allows RC6 residency counter to work */
403 	intel_uncore_write_fw(uncore, VLV_COUNTER_CONTROL,
404 			      _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH |
405 						 VLV_MEDIA_RC0_COUNT_EN |
406 						 VLV_RENDER_RC0_COUNT_EN |
407 						 VLV_MEDIA_RC6_COUNT_EN |
408 						 VLV_RENDER_RC6_COUNT_EN));
409 
410 	rc6->ctl_enable =
411 	    GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL;
412 }
413 
414 bool intel_check_bios_c6_setup(struct intel_rc6 *rc6)
415 {
416 	if (!rc6->bios_state_captured) {
417 		struct intel_uncore *uncore = rc6_to_uncore(rc6);
418 		intel_wakeref_t wakeref;
419 
420 		with_intel_runtime_pm(uncore->rpm, wakeref)
421 			rc6->bios_rc_state = intel_uncore_read(uncore, GEN6_RC_STATE);
422 
423 		rc6->bios_state_captured = true;
424 	}
425 
426 	return rc6->bios_rc_state & RC_SW_TARGET_STATE_MASK;
427 }
428 
429 static bool bxt_check_bios_rc6_setup(struct intel_rc6 *rc6)
430 {
431 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
432 	struct drm_i915_private *i915 = rc6_to_i915(rc6);
433 	u32 rc6_ctx_base, rc_ctl, rc_sw_target;
434 	bool enable_rc6 = true;
435 
436 	rc_ctl = intel_uncore_read(uncore, GEN6_RC_CONTROL);
437 	rc_sw_target = intel_uncore_read(uncore, GEN6_RC_STATE);
438 	rc_sw_target &= RC_SW_TARGET_STATE_MASK;
439 	rc_sw_target >>= RC_SW_TARGET_STATE_SHIFT;
440 	drm_dbg(&i915->drm, "BIOS enabled RC states: "
441 			 "HW_CTRL %s HW_RC6 %s SW_TARGET_STATE %x\n",
442 			 str_on_off(rc_ctl & GEN6_RC_CTL_HW_ENABLE),
443 			 str_on_off(rc_ctl & GEN6_RC_CTL_RC6_ENABLE),
444 			 rc_sw_target);
445 
446 	if (!(intel_uncore_read(uncore, RC6_LOCATION) & RC6_CTX_IN_DRAM)) {
447 		drm_dbg(&i915->drm, "RC6 Base location not set properly.\n");
448 		enable_rc6 = false;
449 	}
450 
451 	/*
452 	 * The exact context size is not known for BXT, so assume a page size
453 	 * for this check.
454 	 */
455 	rc6_ctx_base =
456 		intel_uncore_read(uncore, RC6_CTX_BASE) & RC6_CTX_BASE_MASK;
457 	if (!(rc6_ctx_base >= i915->dsm.reserved.start &&
458 	      rc6_ctx_base + PAGE_SIZE < i915->dsm.reserved.end)) {
459 		drm_dbg(&i915->drm, "RC6 Base address not as expected.\n");
460 		enable_rc6 = false;
461 	}
462 
463 	if (!((intel_uncore_read(uncore, PWRCTX_MAXCNT(RENDER_RING_BASE)) & IDLE_TIME_MASK) > 1 &&
464 	      (intel_uncore_read(uncore, PWRCTX_MAXCNT(GEN6_BSD_RING_BASE)) & IDLE_TIME_MASK) > 1 &&
465 	      (intel_uncore_read(uncore, PWRCTX_MAXCNT(BLT_RING_BASE)) & IDLE_TIME_MASK) > 1 &&
466 	      (intel_uncore_read(uncore, PWRCTX_MAXCNT(VEBOX_RING_BASE)) & IDLE_TIME_MASK) > 1)) {
467 		drm_dbg(&i915->drm,
468 			"Engine Idle wait time not set properly.\n");
469 		enable_rc6 = false;
470 	}
471 
472 	if (!intel_uncore_read(uncore, GEN8_PUSHBUS_CONTROL) ||
473 	    !intel_uncore_read(uncore, GEN8_PUSHBUS_ENABLE) ||
474 	    !intel_uncore_read(uncore, GEN8_PUSHBUS_SHIFT)) {
475 		drm_dbg(&i915->drm, "Pushbus not setup properly.\n");
476 		enable_rc6 = false;
477 	}
478 
479 	if (!intel_uncore_read(uncore, GEN6_GFXPAUSE)) {
480 		drm_dbg(&i915->drm, "GFX pause not setup properly.\n");
481 		enable_rc6 = false;
482 	}
483 
484 	if (!intel_uncore_read(uncore, GEN8_MISC_CTRL0)) {
485 		drm_dbg(&i915->drm, "GPM control not setup properly.\n");
486 		enable_rc6 = false;
487 	}
488 
489 	return enable_rc6;
490 }
491 
492 static bool rc6_supported(struct intel_rc6 *rc6)
493 {
494 	struct drm_i915_private *i915 = rc6_to_i915(rc6);
495 	struct intel_gt *gt = rc6_to_gt(rc6);
496 
497 	if (!HAS_RC6(i915))
498 		return false;
499 
500 	if (intel_vgpu_active(i915))
501 		return false;
502 
503 	if (is_mock_gt(rc6_to_gt(rc6)))
504 		return false;
505 
506 	if (IS_GEN9_LP(i915) && !bxt_check_bios_rc6_setup(rc6)) {
507 		drm_notice(&i915->drm,
508 			   "RC6 and powersaving disabled by BIOS\n");
509 		return false;
510 	}
511 
512 	if (IS_METEORLAKE(gt->i915) &&
513 	    !intel_check_bios_c6_setup(rc6)) {
514 		drm_notice(&i915->drm,
515 			   "C6 disabled by BIOS\n");
516 		return false;
517 	}
518 
519 	if (IS_MEDIA_GT_IP_STEP(gt, IP_VER(13, 0), STEP_A0, STEP_B0)) {
520 		drm_notice(&i915->drm,
521 			   "Media RC6 disabled on A step\n");
522 		return false;
523 	}
524 
525 	return true;
526 }
527 
528 static void rpm_get(struct intel_rc6 *rc6)
529 {
530 	GEM_BUG_ON(rc6->wakeref);
531 	pm_runtime_get_sync(rc6_to_i915(rc6)->drm.dev);
532 	rc6->wakeref = true;
533 }
534 
535 static void rpm_put(struct intel_rc6 *rc6)
536 {
537 	GEM_BUG_ON(!rc6->wakeref);
538 	pm_runtime_put(rc6_to_i915(rc6)->drm.dev);
539 	rc6->wakeref = false;
540 }
541 
542 static bool pctx_corrupted(struct intel_rc6 *rc6)
543 {
544 	struct drm_i915_private *i915 = rc6_to_i915(rc6);
545 
546 	if (!NEEDS_RC6_CTX_CORRUPTION_WA(i915))
547 		return false;
548 
549 	if (intel_uncore_read(rc6_to_uncore(rc6), GEN8_RC6_CTX_INFO))
550 		return false;
551 
552 	drm_notice(&i915->drm,
553 		   "RC6 context corruption, disabling runtime power management\n");
554 	return true;
555 }
556 
557 static void __intel_rc6_disable(struct intel_rc6 *rc6)
558 {
559 	struct drm_i915_private *i915 = rc6_to_i915(rc6);
560 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
561 	struct intel_gt *gt = rc6_to_gt(rc6);
562 
563 	/* Take control of RC6 back from GuC */
564 	intel_guc_rc_disable(gt_to_guc(gt));
565 
566 	intel_uncore_forcewake_get(uncore, FORCEWAKE_ALL);
567 	if (GRAPHICS_VER(i915) >= 9)
568 		intel_uncore_write_fw(uncore, GEN9_PG_ENABLE, 0);
569 	intel_uncore_write_fw(uncore, GEN6_RC_CONTROL, 0);
570 	intel_uncore_write_fw(uncore, GEN6_RC_STATE, 0);
571 	intel_uncore_forcewake_put(uncore, FORCEWAKE_ALL);
572 }
573 
574 static void rc6_res_reg_init(struct intel_rc6 *rc6)
575 {
576 	i915_reg_t res_reg[INTEL_RC6_RES_MAX] = {
577 		[0 ... INTEL_RC6_RES_MAX - 1] = INVALID_MMIO_REG,
578 	};
579 
580 	switch (rc6_to_gt(rc6)->type) {
581 	case GT_MEDIA:
582 		res_reg[INTEL_RC6_RES_RC6] = MTL_MEDIA_MC6;
583 		break;
584 	default:
585 		res_reg[INTEL_RC6_RES_RC6_LOCKED] = GEN6_GT_GFX_RC6_LOCKED;
586 		res_reg[INTEL_RC6_RES_RC6] = GEN6_GT_GFX_RC6;
587 		res_reg[INTEL_RC6_RES_RC6p] = GEN6_GT_GFX_RC6p;
588 		res_reg[INTEL_RC6_RES_RC6pp] = GEN6_GT_GFX_RC6pp;
589 		break;
590 	}
591 
592 	memcpy(rc6->res_reg, res_reg, sizeof(res_reg));
593 }
594 
595 void intel_rc6_init(struct intel_rc6 *rc6)
596 {
597 	struct drm_i915_private *i915 = rc6_to_i915(rc6);
598 	int err;
599 
600 	/* Disable runtime-pm until we can save the GPU state with rc6 pctx */
601 	rpm_get(rc6);
602 
603 	if (!rc6_supported(rc6))
604 		return;
605 
606 	rc6_res_reg_init(rc6);
607 
608 	if (IS_CHERRYVIEW(i915))
609 		err = chv_rc6_init(rc6);
610 	else if (IS_VALLEYVIEW(i915))
611 		err = vlv_rc6_init(rc6);
612 	else
613 		err = 0;
614 
615 	/* Sanitize rc6, ensure it is disabled before we are ready. */
616 	__intel_rc6_disable(rc6);
617 
618 	rc6->supported = err == 0;
619 }
620 
621 void intel_rc6_sanitize(struct intel_rc6 *rc6)
622 {
623 	memset(rc6->prev_hw_residency, 0, sizeof(rc6->prev_hw_residency));
624 
625 	if (rc6->enabled) { /* unbalanced suspend/resume */
626 		rpm_get(rc6);
627 		rc6->enabled = false;
628 	}
629 
630 	if (rc6->supported)
631 		__intel_rc6_disable(rc6);
632 }
633 
634 void intel_rc6_enable(struct intel_rc6 *rc6)
635 {
636 	struct drm_i915_private *i915 = rc6_to_i915(rc6);
637 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
638 
639 	if (!rc6->supported)
640 		return;
641 
642 	GEM_BUG_ON(rc6->enabled);
643 
644 	intel_uncore_forcewake_get(uncore, FORCEWAKE_ALL);
645 
646 	if (IS_CHERRYVIEW(i915))
647 		chv_rc6_enable(rc6);
648 	else if (IS_VALLEYVIEW(i915))
649 		vlv_rc6_enable(rc6);
650 	else if (GRAPHICS_VER(i915) >= 11)
651 		gen11_rc6_enable(rc6);
652 	else if (GRAPHICS_VER(i915) >= 9)
653 		gen9_rc6_enable(rc6);
654 	else if (IS_BROADWELL(i915))
655 		gen8_rc6_enable(rc6);
656 	else if (GRAPHICS_VER(i915) >= 6)
657 		gen6_rc6_enable(rc6);
658 
659 	rc6->manual = rc6->ctl_enable & GEN6_RC_CTL_RC6_ENABLE;
660 	if (NEEDS_RC6_CTX_CORRUPTION_WA(i915))
661 		rc6->ctl_enable = 0;
662 
663 	intel_uncore_forcewake_put(uncore, FORCEWAKE_ALL);
664 
665 	if (unlikely(pctx_corrupted(rc6)))
666 		return;
667 
668 	/* rc6 is ready, runtime-pm is go! */
669 	rpm_put(rc6);
670 	rc6->enabled = true;
671 }
672 
673 void intel_rc6_unpark(struct intel_rc6 *rc6)
674 {
675 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
676 
677 	if (!rc6->enabled)
678 		return;
679 
680 	/* Restore HW timers for automatic RC6 entry while busy */
681 	intel_uncore_write_fw(uncore, GEN6_RC_CONTROL, rc6->ctl_enable);
682 }
683 
684 void intel_rc6_park(struct intel_rc6 *rc6)
685 {
686 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
687 	unsigned int target;
688 
689 	if (!rc6->enabled)
690 		return;
691 
692 	if (unlikely(pctx_corrupted(rc6))) {
693 		intel_rc6_disable(rc6);
694 		return;
695 	}
696 
697 	if (!rc6->manual)
698 		return;
699 
700 	/* Turn off the HW timers and go directly to rc6 */
701 	intel_uncore_write_fw(uncore, GEN6_RC_CONTROL, GEN6_RC_CTL_RC6_ENABLE);
702 
703 	if (HAS_RC6pp(rc6_to_i915(rc6)))
704 		target = 0x6; /* deepest rc6 */
705 	else if (HAS_RC6p(rc6_to_i915(rc6)))
706 		target = 0x5; /* deep rc6 */
707 	else
708 		target = 0x4; /* normal rc6 */
709 	intel_uncore_write_fw(uncore, GEN6_RC_STATE, target << RC_SW_TARGET_STATE_SHIFT);
710 }
711 
712 void intel_rc6_disable(struct intel_rc6 *rc6)
713 {
714 	if (!rc6->enabled)
715 		return;
716 
717 	rpm_get(rc6);
718 	rc6->enabled = false;
719 
720 	__intel_rc6_disable(rc6);
721 }
722 
723 void intel_rc6_fini(struct intel_rc6 *rc6)
724 {
725 	struct drm_i915_gem_object *pctx;
726 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
727 
728 	intel_rc6_disable(rc6);
729 
730 	/* We want the BIOS C6 state preserved across loads for MTL */
731 	if (IS_METEORLAKE(rc6_to_i915(rc6)) && rc6->bios_state_captured)
732 		intel_uncore_write_fw(uncore, GEN6_RC_STATE, rc6->bios_rc_state);
733 
734 	pctx = fetch_and_zero(&rc6->pctx);
735 	if (pctx)
736 		i915_gem_object_put(pctx);
737 
738 	if (rc6->wakeref)
739 		rpm_put(rc6);
740 }
741 
742 static u64 vlv_residency_raw(struct intel_uncore *uncore, const i915_reg_t reg)
743 {
744 	u32 lower, upper, tmp;
745 	int loop = 2;
746 
747 	/*
748 	 * The register accessed do not need forcewake. We borrow
749 	 * uncore lock to prevent concurrent access to range reg.
750 	 */
751 	lockdep_assert_held(&uncore->lock);
752 
753 	/*
754 	 * vlv and chv residency counters are 40 bits in width.
755 	 * With a control bit, we can choose between upper or lower
756 	 * 32bit window into this counter.
757 	 *
758 	 * Although we always use the counter in high-range mode elsewhere,
759 	 * userspace may attempt to read the value before rc6 is initialised,
760 	 * before we have set the default VLV_COUNTER_CONTROL value. So always
761 	 * set the high bit to be safe.
762 	 */
763 	intel_uncore_write_fw(uncore, VLV_COUNTER_CONTROL,
764 			      _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH));
765 	upper = intel_uncore_read_fw(uncore, reg);
766 	do {
767 		tmp = upper;
768 
769 		intel_uncore_write_fw(uncore, VLV_COUNTER_CONTROL,
770 				      _MASKED_BIT_DISABLE(VLV_COUNT_RANGE_HIGH));
771 		lower = intel_uncore_read_fw(uncore, reg);
772 
773 		intel_uncore_write_fw(uncore, VLV_COUNTER_CONTROL,
774 				      _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH));
775 		upper = intel_uncore_read_fw(uncore, reg);
776 	} while (upper != tmp && --loop);
777 
778 	/*
779 	 * Everywhere else we always use VLV_COUNTER_CONTROL with the
780 	 * VLV_COUNT_RANGE_HIGH bit set - so it is safe to leave it set
781 	 * now.
782 	 */
783 
784 	return lower | (u64)upper << 8;
785 }
786 
787 u64 intel_rc6_residency_ns(struct intel_rc6 *rc6, enum intel_rc6_res_type id)
788 {
789 	struct drm_i915_private *i915 = rc6_to_i915(rc6);
790 	struct intel_uncore *uncore = rc6_to_uncore(rc6);
791 	u64 time_hw, prev_hw, overflow_hw;
792 	i915_reg_t reg = rc6->res_reg[id];
793 	unsigned int fw_domains;
794 	unsigned long flags;
795 	u32 mul, div;
796 
797 	if (!rc6->supported)
798 		return 0;
799 
800 	fw_domains = intel_uncore_forcewake_for_reg(uncore, reg, FW_REG_READ);
801 
802 	spin_lock_irqsave(&uncore->lock, flags);
803 	intel_uncore_forcewake_get__locked(uncore, fw_domains);
804 
805 	/* On VLV and CHV, residency time is in CZ units rather than 1.28us */
806 	if (IS_VALLEYVIEW(i915) || IS_CHERRYVIEW(i915)) {
807 		mul = 1000000;
808 		div = vlv_clock_get_czclk(&i915->drm);
809 		overflow_hw = BIT_ULL(40);
810 		time_hw = vlv_residency_raw(uncore, reg);
811 	} else {
812 		/* 833.33ns units on Gen9LP, 1.28us elsewhere. */
813 		if (IS_GEN9_LP(i915)) {
814 			mul = 10000;
815 			div = 12;
816 		} else {
817 			mul = 1280;
818 			div = 1;
819 		}
820 
821 		overflow_hw = BIT_ULL(32);
822 		time_hw = intel_uncore_read_fw(uncore, reg);
823 	}
824 
825 	/*
826 	 * Counter wrap handling.
827 	 *
828 	 * Store previous hw counter values for counter wrap-around handling. But
829 	 * relying on a sufficient frequency of queries otherwise counters can still wrap.
830 	 */
831 	prev_hw = rc6->prev_hw_residency[id];
832 	rc6->prev_hw_residency[id] = time_hw;
833 
834 	/* RC6 delta from last sample. */
835 	if (time_hw >= prev_hw)
836 		time_hw -= prev_hw;
837 	else
838 		time_hw += overflow_hw - prev_hw;
839 
840 	/* Add delta to RC6 extended raw driver copy. */
841 	time_hw += rc6->cur_residency[id];
842 	rc6->cur_residency[id] = time_hw;
843 
844 	intel_uncore_forcewake_put__locked(uncore, fw_domains);
845 	spin_unlock_irqrestore(&uncore->lock, flags);
846 
847 	return mul_u64_u32_div(time_hw, mul, div);
848 }
849 
850 u64 intel_rc6_residency_us(struct intel_rc6 *rc6, enum intel_rc6_res_type id)
851 {
852 	return DIV_ROUND_UP_ULL(intel_rc6_residency_ns(rc6, id), 1000);
853 }
854 
855 void intel_rc6_print_residency(struct seq_file *m, const char *title,
856 			       enum intel_rc6_res_type id)
857 {
858 	struct intel_gt *gt = m->private;
859 	i915_reg_t reg = gt->rc6.res_reg[id];
860 	intel_wakeref_t wakeref;
861 
862 	with_intel_runtime_pm(gt->uncore->rpm, wakeref)
863 		seq_printf(m, "%s %u (%llu us)\n", title,
864 			   intel_uncore_read(gt->uncore, reg),
865 			   intel_rc6_residency_us(&gt->rc6, id));
866 }
867 
868 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
869 #include "selftest_rc6.c"
870 #endif
871