xref: /linux/drivers/gpu/drm/i915/display/skl_watermark.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright © 2022 Intel Corporation
4  */
5 
6 #include <linux/debugfs.h>
7 
8 #include <drm/drm_blend.h>
9 
10 #include "i915_drv.h"
11 #include "i915_reg.h"
12 #include "i9xx_wm.h"
13 #include "intel_atomic.h"
14 #include "intel_atomic_plane.h"
15 #include "intel_bw.h"
16 #include "intel_cdclk.h"
17 #include "intel_crtc.h"
18 #include "intel_cursor_regs.h"
19 #include "intel_de.h"
20 #include "intel_display.h"
21 #include "intel_display_power.h"
22 #include "intel_display_types.h"
23 #include "intel_fb.h"
24 #include "intel_fixed.h"
25 #include "intel_pcode.h"
26 #include "intel_wm.h"
27 #include "skl_universal_plane_regs.h"
28 #include "skl_watermark.h"
29 #include "skl_watermark_regs.h"
30 
31 /*It is expected that DSB can do posted writes to every register in
32  * the pipe and planes within 100us. For flip queue use case, the
33  * recommended DSB execution time is 100us + one SAGV block time.
34  */
35 #define DSB_EXE_TIME 100
36 
37 static void skl_sagv_disable(struct drm_i915_private *i915);
38 
39 /* Stores plane specific WM parameters */
40 struct skl_wm_params {
41 	bool x_tiled, y_tiled;
42 	bool rc_surface;
43 	bool is_planar;
44 	u32 width;
45 	u8 cpp;
46 	u32 plane_pixel_rate;
47 	u32 y_min_scanlines;
48 	u32 plane_bytes_per_line;
49 	uint_fixed_16_16_t plane_blocks_per_line;
50 	uint_fixed_16_16_t y_tile_minimum;
51 	u32 linetime_us;
52 	u32 dbuf_block_size;
53 };
54 
55 u8 intel_enabled_dbuf_slices_mask(struct drm_i915_private *i915)
56 {
57 	u8 enabled_slices = 0;
58 	enum dbuf_slice slice;
59 
60 	for_each_dbuf_slice(i915, slice) {
61 		if (intel_de_read(i915, DBUF_CTL_S(slice)) & DBUF_POWER_STATE)
62 			enabled_slices |= BIT(slice);
63 	}
64 
65 	return enabled_slices;
66 }
67 
68 /*
69  * FIXME: We still don't have the proper code detect if we need to apply the WA,
70  * so assume we'll always need it in order to avoid underruns.
71  */
72 static bool skl_needs_memory_bw_wa(struct drm_i915_private *i915)
73 {
74 	return DISPLAY_VER(i915) == 9;
75 }
76 
77 bool
78 intel_has_sagv(struct drm_i915_private *i915)
79 {
80 	return HAS_SAGV(i915) &&
81 		i915->display.sagv.status != I915_SAGV_NOT_CONTROLLED;
82 }
83 
84 static u32
85 intel_sagv_block_time(struct drm_i915_private *i915)
86 {
87 	if (DISPLAY_VER(i915) >= 14) {
88 		u32 val;
89 
90 		val = intel_de_read(i915, MTL_LATENCY_SAGV);
91 
92 		return REG_FIELD_GET(MTL_LATENCY_QCLK_SAGV, val);
93 	} else if (DISPLAY_VER(i915) >= 12) {
94 		u32 val = 0;
95 		int ret;
96 
97 		ret = snb_pcode_read(&i915->uncore,
98 				     GEN12_PCODE_READ_SAGV_BLOCK_TIME_US,
99 				     &val, NULL);
100 		if (ret) {
101 			drm_dbg_kms(&i915->drm, "Couldn't read SAGV block time!\n");
102 			return 0;
103 		}
104 
105 		return val;
106 	} else if (DISPLAY_VER(i915) == 11) {
107 		return 10;
108 	} else if (HAS_SAGV(i915)) {
109 		return 30;
110 	} else {
111 		return 0;
112 	}
113 }
114 
115 static void intel_sagv_init(struct drm_i915_private *i915)
116 {
117 	if (!HAS_SAGV(i915))
118 		i915->display.sagv.status = I915_SAGV_NOT_CONTROLLED;
119 
120 	/*
121 	 * Probe to see if we have working SAGV control.
122 	 * For icl+ this was already determined by intel_bw_init_hw().
123 	 */
124 	if (DISPLAY_VER(i915) < 11)
125 		skl_sagv_disable(i915);
126 
127 	drm_WARN_ON(&i915->drm, i915->display.sagv.status == I915_SAGV_UNKNOWN);
128 
129 	i915->display.sagv.block_time_us = intel_sagv_block_time(i915);
130 
131 	drm_dbg_kms(&i915->drm, "SAGV supported: %s, original SAGV block time: %u us\n",
132 		    str_yes_no(intel_has_sagv(i915)), i915->display.sagv.block_time_us);
133 
134 	/* avoid overflow when adding with wm0 latency/etc. */
135 	if (drm_WARN(&i915->drm, i915->display.sagv.block_time_us > U16_MAX,
136 		     "Excessive SAGV block time %u, ignoring\n",
137 		     i915->display.sagv.block_time_us))
138 		i915->display.sagv.block_time_us = 0;
139 
140 	if (!intel_has_sagv(i915))
141 		i915->display.sagv.block_time_us = 0;
142 }
143 
144 /*
145  * SAGV dynamically adjusts the system agent voltage and clock frequencies
146  * depending on power and performance requirements. The display engine access
147  * to system memory is blocked during the adjustment time. Because of the
148  * blocking time, having this enabled can cause full system hangs and/or pipe
149  * underruns if we don't meet all of the following requirements:
150  *
151  *  - <= 1 pipe enabled
152  *  - All planes can enable watermarks for latencies >= SAGV engine block time
153  *  - We're not using an interlaced display configuration
154  */
155 static void skl_sagv_enable(struct drm_i915_private *i915)
156 {
157 	int ret;
158 
159 	if (!intel_has_sagv(i915))
160 		return;
161 
162 	if (i915->display.sagv.status == I915_SAGV_ENABLED)
163 		return;
164 
165 	drm_dbg_kms(&i915->drm, "Enabling SAGV\n");
166 	ret = snb_pcode_write(&i915->uncore, GEN9_PCODE_SAGV_CONTROL,
167 			      GEN9_SAGV_ENABLE);
168 
169 	/* We don't need to wait for SAGV when enabling */
170 
171 	/*
172 	 * Some skl systems, pre-release machines in particular,
173 	 * don't actually have SAGV.
174 	 */
175 	if (IS_SKYLAKE(i915) && ret == -ENXIO) {
176 		drm_dbg(&i915->drm, "No SAGV found on system, ignoring\n");
177 		i915->display.sagv.status = I915_SAGV_NOT_CONTROLLED;
178 		return;
179 	} else if (ret < 0) {
180 		drm_err(&i915->drm, "Failed to enable SAGV\n");
181 		return;
182 	}
183 
184 	i915->display.sagv.status = I915_SAGV_ENABLED;
185 }
186 
187 static void skl_sagv_disable(struct drm_i915_private *i915)
188 {
189 	int ret;
190 
191 	if (!intel_has_sagv(i915))
192 		return;
193 
194 	if (i915->display.sagv.status == I915_SAGV_DISABLED)
195 		return;
196 
197 	drm_dbg_kms(&i915->drm, "Disabling SAGV\n");
198 	/* bspec says to keep retrying for at least 1 ms */
199 	ret = skl_pcode_request(&i915->uncore, GEN9_PCODE_SAGV_CONTROL,
200 				GEN9_SAGV_DISABLE,
201 				GEN9_SAGV_IS_DISABLED, GEN9_SAGV_IS_DISABLED,
202 				1);
203 	/*
204 	 * Some skl systems, pre-release machines in particular,
205 	 * don't actually have SAGV.
206 	 */
207 	if (IS_SKYLAKE(i915) && ret == -ENXIO) {
208 		drm_dbg(&i915->drm, "No SAGV found on system, ignoring\n");
209 		i915->display.sagv.status = I915_SAGV_NOT_CONTROLLED;
210 		return;
211 	} else if (ret < 0) {
212 		drm_err(&i915->drm, "Failed to disable SAGV (%d)\n", ret);
213 		return;
214 	}
215 
216 	i915->display.sagv.status = I915_SAGV_DISABLED;
217 }
218 
219 static void skl_sagv_pre_plane_update(struct intel_atomic_state *state)
220 {
221 	struct drm_i915_private *i915 = to_i915(state->base.dev);
222 	const struct intel_bw_state *new_bw_state =
223 		intel_atomic_get_new_bw_state(state);
224 
225 	if (!new_bw_state)
226 		return;
227 
228 	if (!intel_can_enable_sagv(i915, new_bw_state))
229 		skl_sagv_disable(i915);
230 }
231 
232 static void skl_sagv_post_plane_update(struct intel_atomic_state *state)
233 {
234 	struct drm_i915_private *i915 = to_i915(state->base.dev);
235 	const struct intel_bw_state *new_bw_state =
236 		intel_atomic_get_new_bw_state(state);
237 
238 	if (!new_bw_state)
239 		return;
240 
241 	if (intel_can_enable_sagv(i915, new_bw_state))
242 		skl_sagv_enable(i915);
243 }
244 
245 static void icl_sagv_pre_plane_update(struct intel_atomic_state *state)
246 {
247 	struct drm_i915_private *i915 = to_i915(state->base.dev);
248 	const struct intel_bw_state *old_bw_state =
249 		intel_atomic_get_old_bw_state(state);
250 	const struct intel_bw_state *new_bw_state =
251 		intel_atomic_get_new_bw_state(state);
252 	u16 old_mask, new_mask;
253 
254 	if (!new_bw_state)
255 		return;
256 
257 	old_mask = old_bw_state->qgv_points_mask;
258 	new_mask = old_bw_state->qgv_points_mask | new_bw_state->qgv_points_mask;
259 
260 	if (old_mask == new_mask)
261 		return;
262 
263 	WARN_ON(!new_bw_state->base.changed);
264 
265 	drm_dbg_kms(&i915->drm, "Restricting QGV points: 0x%x -> 0x%x\n",
266 		    old_mask, new_mask);
267 
268 	/*
269 	 * Restrict required qgv points before updating the configuration.
270 	 * According to BSpec we can't mask and unmask qgv points at the same
271 	 * time. Also masking should be done before updating the configuration
272 	 * and unmasking afterwards.
273 	 */
274 	icl_pcode_restrict_qgv_points(i915, new_mask);
275 }
276 
277 static void icl_sagv_post_plane_update(struct intel_atomic_state *state)
278 {
279 	struct drm_i915_private *i915 = to_i915(state->base.dev);
280 	const struct intel_bw_state *old_bw_state =
281 		intel_atomic_get_old_bw_state(state);
282 	const struct intel_bw_state *new_bw_state =
283 		intel_atomic_get_new_bw_state(state);
284 	u16 old_mask, new_mask;
285 
286 	if (!new_bw_state)
287 		return;
288 
289 	old_mask = old_bw_state->qgv_points_mask | new_bw_state->qgv_points_mask;
290 	new_mask = new_bw_state->qgv_points_mask;
291 
292 	if (old_mask == new_mask)
293 		return;
294 
295 	WARN_ON(!new_bw_state->base.changed);
296 
297 	drm_dbg_kms(&i915->drm, "Relaxing QGV points: 0x%x -> 0x%x\n",
298 		    old_mask, new_mask);
299 
300 	/*
301 	 * Allow required qgv points after updating the configuration.
302 	 * According to BSpec we can't mask and unmask qgv points at the same
303 	 * time. Also masking should be done before updating the configuration
304 	 * and unmasking afterwards.
305 	 */
306 	icl_pcode_restrict_qgv_points(i915, new_mask);
307 }
308 
309 void intel_sagv_pre_plane_update(struct intel_atomic_state *state)
310 {
311 	struct drm_i915_private *i915 = to_i915(state->base.dev);
312 
313 	/*
314 	 * Just return if we can't control SAGV or don't have it.
315 	 * This is different from situation when we have SAGV but just can't
316 	 * afford it due to DBuf limitation - in case if SAGV is completely
317 	 * disabled in a BIOS, we are not even allowed to send a PCode request,
318 	 * as it will throw an error. So have to check it here.
319 	 */
320 	if (!intel_has_sagv(i915))
321 		return;
322 
323 	if (DISPLAY_VER(i915) >= 11)
324 		icl_sagv_pre_plane_update(state);
325 	else
326 		skl_sagv_pre_plane_update(state);
327 }
328 
329 void intel_sagv_post_plane_update(struct intel_atomic_state *state)
330 {
331 	struct drm_i915_private *i915 = to_i915(state->base.dev);
332 
333 	/*
334 	 * Just return if we can't control SAGV or don't have it.
335 	 * This is different from situation when we have SAGV but just can't
336 	 * afford it due to DBuf limitation - in case if SAGV is completely
337 	 * disabled in a BIOS, we are not even allowed to send a PCode request,
338 	 * as it will throw an error. So have to check it here.
339 	 */
340 	if (!intel_has_sagv(i915))
341 		return;
342 
343 	if (DISPLAY_VER(i915) >= 11)
344 		icl_sagv_post_plane_update(state);
345 	else
346 		skl_sagv_post_plane_update(state);
347 }
348 
349 static bool skl_crtc_can_enable_sagv(const struct intel_crtc_state *crtc_state)
350 {
351 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
352 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
353 	enum plane_id plane_id;
354 	int max_level = INT_MAX;
355 
356 	if (!intel_has_sagv(i915))
357 		return false;
358 
359 	if (!crtc_state->hw.active)
360 		return true;
361 
362 	if (crtc_state->hw.pipe_mode.flags & DRM_MODE_FLAG_INTERLACE)
363 		return false;
364 
365 	for_each_plane_id_on_crtc(crtc, plane_id) {
366 		const struct skl_plane_wm *wm =
367 			&crtc_state->wm.skl.optimal.planes[plane_id];
368 		int level;
369 
370 		/* Skip this plane if it's not enabled */
371 		if (!wm->wm[0].enable)
372 			continue;
373 
374 		/* Find the highest enabled wm level for this plane */
375 		for (level = i915->display.wm.num_levels - 1;
376 		     !wm->wm[level].enable; --level)
377 		     { }
378 
379 		/* Highest common enabled wm level for all planes */
380 		max_level = min(level, max_level);
381 	}
382 
383 	/* No enabled planes? */
384 	if (max_level == INT_MAX)
385 		return true;
386 
387 	for_each_plane_id_on_crtc(crtc, plane_id) {
388 		const struct skl_plane_wm *wm =
389 			&crtc_state->wm.skl.optimal.planes[plane_id];
390 
391 		/*
392 		 * All enabled planes must have enabled a common wm level that
393 		 * can tolerate memory latencies higher than sagv_block_time_us
394 		 */
395 		if (wm->wm[0].enable && !wm->wm[max_level].can_sagv)
396 			return false;
397 	}
398 
399 	return true;
400 }
401 
402 static bool tgl_crtc_can_enable_sagv(const struct intel_crtc_state *crtc_state)
403 {
404 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
405 	enum plane_id plane_id;
406 
407 	if (!crtc_state->hw.active)
408 		return true;
409 
410 	for_each_plane_id_on_crtc(crtc, plane_id) {
411 		const struct skl_plane_wm *wm =
412 			&crtc_state->wm.skl.optimal.planes[plane_id];
413 
414 		if (wm->wm[0].enable && !wm->sagv.wm0.enable)
415 			return false;
416 	}
417 
418 	return true;
419 }
420 
421 static bool intel_crtc_can_enable_sagv(const struct intel_crtc_state *crtc_state)
422 {
423 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
424 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
425 
426 	if (!i915->display.params.enable_sagv)
427 		return false;
428 
429 	if (DISPLAY_VER(i915) >= 12)
430 		return tgl_crtc_can_enable_sagv(crtc_state);
431 	else
432 		return skl_crtc_can_enable_sagv(crtc_state);
433 }
434 
435 bool intel_can_enable_sagv(struct drm_i915_private *i915,
436 			   const struct intel_bw_state *bw_state)
437 {
438 	if (DISPLAY_VER(i915) < 11 &&
439 	    bw_state->active_pipes && !is_power_of_2(bw_state->active_pipes))
440 		return false;
441 
442 	return bw_state->pipe_sagv_reject == 0;
443 }
444 
445 static int intel_compute_sagv_mask(struct intel_atomic_state *state)
446 {
447 	struct drm_i915_private *i915 = to_i915(state->base.dev);
448 	int ret;
449 	struct intel_crtc *crtc;
450 	struct intel_crtc_state *new_crtc_state;
451 	struct intel_bw_state *new_bw_state = NULL;
452 	const struct intel_bw_state *old_bw_state = NULL;
453 	int i;
454 
455 	for_each_new_intel_crtc_in_state(state, crtc,
456 					 new_crtc_state, i) {
457 		struct skl_pipe_wm *pipe_wm = &new_crtc_state->wm.skl.optimal;
458 
459 		new_bw_state = intel_atomic_get_bw_state(state);
460 		if (IS_ERR(new_bw_state))
461 			return PTR_ERR(new_bw_state);
462 
463 		old_bw_state = intel_atomic_get_old_bw_state(state);
464 
465 		/*
466 		 * We store use_sagv_wm in the crtc state rather than relying on
467 		 * that bw state since we have no convenient way to get at the
468 		 * latter from the plane commit hooks (especially in the legacy
469 		 * cursor case).
470 		 *
471 		 * drm_atomic_check_only() gets upset if we pull more crtcs
472 		 * into the state, so we have to calculate this based on the
473 		 * individual intel_crtc_can_enable_sagv() rather than
474 		 * the overall intel_can_enable_sagv(). Otherwise the
475 		 * crtcs not included in the commit would not switch to the
476 		 * SAGV watermarks when we are about to enable SAGV, and that
477 		 * would lead to underruns. This does mean extra power draw
478 		 * when only a subset of the crtcs are blocking SAGV as the
479 		 * other crtcs can't be allowed to use the more optimal
480 		 * normal (ie. non-SAGV) watermarks.
481 		 */
482 		pipe_wm->use_sagv_wm = !HAS_HW_SAGV_WM(i915) &&
483 			DISPLAY_VER(i915) >= 12 &&
484 			intel_crtc_can_enable_sagv(new_crtc_state);
485 
486 		if (intel_crtc_can_enable_sagv(new_crtc_state))
487 			new_bw_state->pipe_sagv_reject &= ~BIT(crtc->pipe);
488 		else
489 			new_bw_state->pipe_sagv_reject |= BIT(crtc->pipe);
490 	}
491 
492 	if (!new_bw_state)
493 		return 0;
494 
495 	new_bw_state->active_pipes =
496 		intel_calc_active_pipes(state, old_bw_state->active_pipes);
497 
498 	if (new_bw_state->active_pipes != old_bw_state->active_pipes) {
499 		ret = intel_atomic_lock_global_state(&new_bw_state->base);
500 		if (ret)
501 			return ret;
502 	}
503 
504 	if (intel_can_enable_sagv(i915, new_bw_state) !=
505 	    intel_can_enable_sagv(i915, old_bw_state)) {
506 		ret = intel_atomic_serialize_global_state(&new_bw_state->base);
507 		if (ret)
508 			return ret;
509 	} else if (new_bw_state->pipe_sagv_reject != old_bw_state->pipe_sagv_reject) {
510 		ret = intel_atomic_lock_global_state(&new_bw_state->base);
511 		if (ret)
512 			return ret;
513 	}
514 
515 	return 0;
516 }
517 
518 static u16 skl_ddb_entry_init(struct skl_ddb_entry *entry,
519 			      u16 start, u16 end)
520 {
521 	entry->start = start;
522 	entry->end = end;
523 
524 	return end;
525 }
526 
527 static int intel_dbuf_slice_size(struct drm_i915_private *i915)
528 {
529 	return DISPLAY_INFO(i915)->dbuf.size /
530 		hweight8(DISPLAY_INFO(i915)->dbuf.slice_mask);
531 }
532 
533 static void
534 skl_ddb_entry_for_slices(struct drm_i915_private *i915, u8 slice_mask,
535 			 struct skl_ddb_entry *ddb)
536 {
537 	int slice_size = intel_dbuf_slice_size(i915);
538 
539 	if (!slice_mask) {
540 		ddb->start = 0;
541 		ddb->end = 0;
542 		return;
543 	}
544 
545 	ddb->start = (ffs(slice_mask) - 1) * slice_size;
546 	ddb->end = fls(slice_mask) * slice_size;
547 
548 	WARN_ON(ddb->start >= ddb->end);
549 	WARN_ON(ddb->end > DISPLAY_INFO(i915)->dbuf.size);
550 }
551 
552 static unsigned int mbus_ddb_offset(struct drm_i915_private *i915, u8 slice_mask)
553 {
554 	struct skl_ddb_entry ddb;
555 
556 	if (slice_mask & (BIT(DBUF_S1) | BIT(DBUF_S2)))
557 		slice_mask = BIT(DBUF_S1);
558 	else if (slice_mask & (BIT(DBUF_S3) | BIT(DBUF_S4)))
559 		slice_mask = BIT(DBUF_S3);
560 
561 	skl_ddb_entry_for_slices(i915, slice_mask, &ddb);
562 
563 	return ddb.start;
564 }
565 
566 u32 skl_ddb_dbuf_slice_mask(struct drm_i915_private *i915,
567 			    const struct skl_ddb_entry *entry)
568 {
569 	int slice_size = intel_dbuf_slice_size(i915);
570 	enum dbuf_slice start_slice, end_slice;
571 	u8 slice_mask = 0;
572 
573 	if (!skl_ddb_entry_size(entry))
574 		return 0;
575 
576 	start_slice = entry->start / slice_size;
577 	end_slice = (entry->end - 1) / slice_size;
578 
579 	/*
580 	 * Per plane DDB entry can in a really worst case be on multiple slices
581 	 * but single entry is anyway contigious.
582 	 */
583 	while (start_slice <= end_slice) {
584 		slice_mask |= BIT(start_slice);
585 		start_slice++;
586 	}
587 
588 	return slice_mask;
589 }
590 
591 static unsigned int intel_crtc_ddb_weight(const struct intel_crtc_state *crtc_state)
592 {
593 	const struct drm_display_mode *pipe_mode = &crtc_state->hw.pipe_mode;
594 	int hdisplay, vdisplay;
595 
596 	if (!crtc_state->hw.active)
597 		return 0;
598 
599 	/*
600 	 * Watermark/ddb requirement highly depends upon width of the
601 	 * framebuffer, So instead of allocating DDB equally among pipes
602 	 * distribute DDB based on resolution/width of the display.
603 	 */
604 	drm_mode_get_hv_timing(pipe_mode, &hdisplay, &vdisplay);
605 
606 	return hdisplay;
607 }
608 
609 static void intel_crtc_dbuf_weights(const struct intel_dbuf_state *dbuf_state,
610 				    enum pipe for_pipe,
611 				    unsigned int *weight_start,
612 				    unsigned int *weight_end,
613 				    unsigned int *weight_total)
614 {
615 	struct drm_i915_private *i915 =
616 		to_i915(dbuf_state->base.state->base.dev);
617 	enum pipe pipe;
618 
619 	*weight_start = 0;
620 	*weight_end = 0;
621 	*weight_total = 0;
622 
623 	for_each_pipe(i915, pipe) {
624 		int weight = dbuf_state->weight[pipe];
625 
626 		/*
627 		 * Do not account pipes using other slice sets
628 		 * luckily as of current BSpec slice sets do not partially
629 		 * intersect(pipes share either same one slice or same slice set
630 		 * i.e no partial intersection), so it is enough to check for
631 		 * equality for now.
632 		 */
633 		if (dbuf_state->slices[pipe] != dbuf_state->slices[for_pipe])
634 			continue;
635 
636 		*weight_total += weight;
637 		if (pipe < for_pipe) {
638 			*weight_start += weight;
639 			*weight_end += weight;
640 		} else if (pipe == for_pipe) {
641 			*weight_end += weight;
642 		}
643 	}
644 }
645 
646 static int
647 skl_crtc_allocate_ddb(struct intel_atomic_state *state, struct intel_crtc *crtc)
648 {
649 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
650 	unsigned int weight_total, weight_start, weight_end;
651 	const struct intel_dbuf_state *old_dbuf_state =
652 		intel_atomic_get_old_dbuf_state(state);
653 	struct intel_dbuf_state *new_dbuf_state =
654 		intel_atomic_get_new_dbuf_state(state);
655 	struct intel_crtc_state *crtc_state;
656 	struct skl_ddb_entry ddb_slices;
657 	enum pipe pipe = crtc->pipe;
658 	unsigned int mbus_offset = 0;
659 	u32 ddb_range_size;
660 	u32 dbuf_slice_mask;
661 	u32 start, end;
662 	int ret;
663 
664 	if (new_dbuf_state->weight[pipe] == 0) {
665 		skl_ddb_entry_init(&new_dbuf_state->ddb[pipe], 0, 0);
666 		goto out;
667 	}
668 
669 	dbuf_slice_mask = new_dbuf_state->slices[pipe];
670 
671 	skl_ddb_entry_for_slices(i915, dbuf_slice_mask, &ddb_slices);
672 	mbus_offset = mbus_ddb_offset(i915, dbuf_slice_mask);
673 	ddb_range_size = skl_ddb_entry_size(&ddb_slices);
674 
675 	intel_crtc_dbuf_weights(new_dbuf_state, pipe,
676 				&weight_start, &weight_end, &weight_total);
677 
678 	start = ddb_range_size * weight_start / weight_total;
679 	end = ddb_range_size * weight_end / weight_total;
680 
681 	skl_ddb_entry_init(&new_dbuf_state->ddb[pipe],
682 			   ddb_slices.start - mbus_offset + start,
683 			   ddb_slices.start - mbus_offset + end);
684 
685 out:
686 	if (old_dbuf_state->slices[pipe] == new_dbuf_state->slices[pipe] &&
687 	    skl_ddb_entry_equal(&old_dbuf_state->ddb[pipe],
688 				&new_dbuf_state->ddb[pipe]))
689 		return 0;
690 
691 	ret = intel_atomic_lock_global_state(&new_dbuf_state->base);
692 	if (ret)
693 		return ret;
694 
695 	crtc_state = intel_atomic_get_crtc_state(&state->base, crtc);
696 	if (IS_ERR(crtc_state))
697 		return PTR_ERR(crtc_state);
698 
699 	/*
700 	 * Used for checking overlaps, so we need absolute
701 	 * offsets instead of MBUS relative offsets.
702 	 */
703 	crtc_state->wm.skl.ddb.start = mbus_offset + new_dbuf_state->ddb[pipe].start;
704 	crtc_state->wm.skl.ddb.end = mbus_offset + new_dbuf_state->ddb[pipe].end;
705 
706 	drm_dbg_kms(&i915->drm,
707 		    "[CRTC:%d:%s] dbuf slices 0x%x -> 0x%x, ddb (%d - %d) -> (%d - %d), active pipes 0x%x -> 0x%x\n",
708 		    crtc->base.base.id, crtc->base.name,
709 		    old_dbuf_state->slices[pipe], new_dbuf_state->slices[pipe],
710 		    old_dbuf_state->ddb[pipe].start, old_dbuf_state->ddb[pipe].end,
711 		    new_dbuf_state->ddb[pipe].start, new_dbuf_state->ddb[pipe].end,
712 		    old_dbuf_state->active_pipes, new_dbuf_state->active_pipes);
713 
714 	return 0;
715 }
716 
717 static int skl_compute_wm_params(const struct intel_crtc_state *crtc_state,
718 				 int width, const struct drm_format_info *format,
719 				 u64 modifier, unsigned int rotation,
720 				 u32 plane_pixel_rate, struct skl_wm_params *wp,
721 				 int color_plane, unsigned int pan_x);
722 
723 static void skl_compute_plane_wm(const struct intel_crtc_state *crtc_state,
724 				 struct intel_plane *plane,
725 				 int level,
726 				 unsigned int latency,
727 				 const struct skl_wm_params *wp,
728 				 const struct skl_wm_level *result_prev,
729 				 struct skl_wm_level *result /* out */);
730 
731 static unsigned int skl_wm_latency(struct drm_i915_private *i915, int level,
732 				   const struct skl_wm_params *wp)
733 {
734 	unsigned int latency = i915->display.wm.skl_latency[level];
735 
736 	if (latency == 0)
737 		return 0;
738 
739 	/*
740 	 * WaIncreaseLatencyIPCEnabled: kbl,cfl
741 	 * Display WA #1141: kbl,cfl
742 	 */
743 	if ((IS_KABYLAKE(i915) || IS_COFFEELAKE(i915) || IS_COMETLAKE(i915)) &&
744 	    skl_watermark_ipc_enabled(i915))
745 		latency += 4;
746 
747 	if (skl_needs_memory_bw_wa(i915) && wp && wp->x_tiled)
748 		latency += 15;
749 
750 	return latency;
751 }
752 
753 static unsigned int
754 skl_cursor_allocation(const struct intel_crtc_state *crtc_state,
755 		      int num_active)
756 {
757 	struct intel_plane *plane = to_intel_plane(crtc_state->uapi.crtc->cursor);
758 	struct drm_i915_private *i915 = to_i915(crtc_state->uapi.crtc->dev);
759 	struct skl_wm_level wm = {};
760 	int ret, min_ddb_alloc = 0;
761 	struct skl_wm_params wp;
762 	int level;
763 
764 	ret = skl_compute_wm_params(crtc_state, 256,
765 				    drm_format_info(DRM_FORMAT_ARGB8888),
766 				    DRM_FORMAT_MOD_LINEAR,
767 				    DRM_MODE_ROTATE_0,
768 				    crtc_state->pixel_rate, &wp, 0, 0);
769 	drm_WARN_ON(&i915->drm, ret);
770 
771 	for (level = 0; level < i915->display.wm.num_levels; level++) {
772 		unsigned int latency = skl_wm_latency(i915, level, &wp);
773 
774 		skl_compute_plane_wm(crtc_state, plane, level, latency, &wp, &wm, &wm);
775 		if (wm.min_ddb_alloc == U16_MAX)
776 			break;
777 
778 		min_ddb_alloc = wm.min_ddb_alloc;
779 	}
780 
781 	return max(num_active == 1 ? 32 : 8, min_ddb_alloc);
782 }
783 
784 static void skl_ddb_entry_init_from_hw(struct skl_ddb_entry *entry, u32 reg)
785 {
786 	skl_ddb_entry_init(entry,
787 			   REG_FIELD_GET(PLANE_BUF_START_MASK, reg),
788 			   REG_FIELD_GET(PLANE_BUF_END_MASK, reg));
789 	if (entry->end)
790 		entry->end++;
791 }
792 
793 static void
794 skl_ddb_get_hw_plane_state(struct drm_i915_private *i915,
795 			   const enum pipe pipe,
796 			   const enum plane_id plane_id,
797 			   struct skl_ddb_entry *ddb,
798 			   struct skl_ddb_entry *ddb_y)
799 {
800 	u32 val;
801 
802 	/* Cursor doesn't support NV12/planar, so no extra calculation needed */
803 	if (plane_id == PLANE_CURSOR) {
804 		val = intel_de_read(i915, CUR_BUF_CFG(pipe));
805 		skl_ddb_entry_init_from_hw(ddb, val);
806 		return;
807 	}
808 
809 	val = intel_de_read(i915, PLANE_BUF_CFG(pipe, plane_id));
810 	skl_ddb_entry_init_from_hw(ddb, val);
811 
812 	if (DISPLAY_VER(i915) >= 11)
813 		return;
814 
815 	val = intel_de_read(i915, PLANE_NV12_BUF_CFG(pipe, plane_id));
816 	skl_ddb_entry_init_from_hw(ddb_y, val);
817 }
818 
819 static void skl_pipe_ddb_get_hw_state(struct intel_crtc *crtc,
820 				      struct skl_ddb_entry *ddb,
821 				      struct skl_ddb_entry *ddb_y)
822 {
823 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
824 	enum intel_display_power_domain power_domain;
825 	enum pipe pipe = crtc->pipe;
826 	intel_wakeref_t wakeref;
827 	enum plane_id plane_id;
828 
829 	power_domain = POWER_DOMAIN_PIPE(pipe);
830 	wakeref = intel_display_power_get_if_enabled(i915, power_domain);
831 	if (!wakeref)
832 		return;
833 
834 	for_each_plane_id_on_crtc(crtc, plane_id)
835 		skl_ddb_get_hw_plane_state(i915, pipe,
836 					   plane_id,
837 					   &ddb[plane_id],
838 					   &ddb_y[plane_id]);
839 
840 	intel_display_power_put(i915, power_domain, wakeref);
841 }
842 
843 struct dbuf_slice_conf_entry {
844 	u8 active_pipes;
845 	u8 dbuf_mask[I915_MAX_PIPES];
846 	bool join_mbus;
847 };
848 
849 /*
850  * Table taken from Bspec 12716
851  * Pipes do have some preferred DBuf slice affinity,
852  * plus there are some hardcoded requirements on how
853  * those should be distributed for multipipe scenarios.
854  * For more DBuf slices algorithm can get even more messy
855  * and less readable, so decided to use a table almost
856  * as is from BSpec itself - that way it is at least easier
857  * to compare, change and check.
858  */
859 static const struct dbuf_slice_conf_entry icl_allowed_dbufs[] =
860 /* Autogenerated with igt/tools/intel_dbuf_map tool: */
861 {
862 	{
863 		.active_pipes = BIT(PIPE_A),
864 		.dbuf_mask = {
865 			[PIPE_A] = BIT(DBUF_S1),
866 		},
867 	},
868 	{
869 		.active_pipes = BIT(PIPE_B),
870 		.dbuf_mask = {
871 			[PIPE_B] = BIT(DBUF_S1),
872 		},
873 	},
874 	{
875 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B),
876 		.dbuf_mask = {
877 			[PIPE_A] = BIT(DBUF_S1),
878 			[PIPE_B] = BIT(DBUF_S2),
879 		},
880 	},
881 	{
882 		.active_pipes = BIT(PIPE_C),
883 		.dbuf_mask = {
884 			[PIPE_C] = BIT(DBUF_S2),
885 		},
886 	},
887 	{
888 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C),
889 		.dbuf_mask = {
890 			[PIPE_A] = BIT(DBUF_S1),
891 			[PIPE_C] = BIT(DBUF_S2),
892 		},
893 	},
894 	{
895 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C),
896 		.dbuf_mask = {
897 			[PIPE_B] = BIT(DBUF_S1),
898 			[PIPE_C] = BIT(DBUF_S2),
899 		},
900 	},
901 	{
902 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C),
903 		.dbuf_mask = {
904 			[PIPE_A] = BIT(DBUF_S1),
905 			[PIPE_B] = BIT(DBUF_S1),
906 			[PIPE_C] = BIT(DBUF_S2),
907 		},
908 	},
909 	{}
910 };
911 
912 /*
913  * Table taken from Bspec 49255
914  * Pipes do have some preferred DBuf slice affinity,
915  * plus there are some hardcoded requirements on how
916  * those should be distributed for multipipe scenarios.
917  * For more DBuf slices algorithm can get even more messy
918  * and less readable, so decided to use a table almost
919  * as is from BSpec itself - that way it is at least easier
920  * to compare, change and check.
921  */
922 static const struct dbuf_slice_conf_entry tgl_allowed_dbufs[] =
923 /* Autogenerated with igt/tools/intel_dbuf_map tool: */
924 {
925 	{
926 		.active_pipes = BIT(PIPE_A),
927 		.dbuf_mask = {
928 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
929 		},
930 	},
931 	{
932 		.active_pipes = BIT(PIPE_B),
933 		.dbuf_mask = {
934 			[PIPE_B] = BIT(DBUF_S1) | BIT(DBUF_S2),
935 		},
936 	},
937 	{
938 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B),
939 		.dbuf_mask = {
940 			[PIPE_A] = BIT(DBUF_S2),
941 			[PIPE_B] = BIT(DBUF_S1),
942 		},
943 	},
944 	{
945 		.active_pipes = BIT(PIPE_C),
946 		.dbuf_mask = {
947 			[PIPE_C] = BIT(DBUF_S2) | BIT(DBUF_S1),
948 		},
949 	},
950 	{
951 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C),
952 		.dbuf_mask = {
953 			[PIPE_A] = BIT(DBUF_S1),
954 			[PIPE_C] = BIT(DBUF_S2),
955 		},
956 	},
957 	{
958 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C),
959 		.dbuf_mask = {
960 			[PIPE_B] = BIT(DBUF_S1),
961 			[PIPE_C] = BIT(DBUF_S2),
962 		},
963 	},
964 	{
965 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C),
966 		.dbuf_mask = {
967 			[PIPE_A] = BIT(DBUF_S1),
968 			[PIPE_B] = BIT(DBUF_S1),
969 			[PIPE_C] = BIT(DBUF_S2),
970 		},
971 	},
972 	{
973 		.active_pipes = BIT(PIPE_D),
974 		.dbuf_mask = {
975 			[PIPE_D] = BIT(DBUF_S2) | BIT(DBUF_S1),
976 		},
977 	},
978 	{
979 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_D),
980 		.dbuf_mask = {
981 			[PIPE_A] = BIT(DBUF_S1),
982 			[PIPE_D] = BIT(DBUF_S2),
983 		},
984 	},
985 	{
986 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_D),
987 		.dbuf_mask = {
988 			[PIPE_B] = BIT(DBUF_S1),
989 			[PIPE_D] = BIT(DBUF_S2),
990 		},
991 	},
992 	{
993 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_D),
994 		.dbuf_mask = {
995 			[PIPE_A] = BIT(DBUF_S1),
996 			[PIPE_B] = BIT(DBUF_S1),
997 			[PIPE_D] = BIT(DBUF_S2),
998 		},
999 	},
1000 	{
1001 		.active_pipes = BIT(PIPE_C) | BIT(PIPE_D),
1002 		.dbuf_mask = {
1003 			[PIPE_C] = BIT(DBUF_S1),
1004 			[PIPE_D] = BIT(DBUF_S2),
1005 		},
1006 	},
1007 	{
1008 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C) | BIT(PIPE_D),
1009 		.dbuf_mask = {
1010 			[PIPE_A] = BIT(DBUF_S1),
1011 			[PIPE_C] = BIT(DBUF_S2),
1012 			[PIPE_D] = BIT(DBUF_S2),
1013 		},
1014 	},
1015 	{
1016 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
1017 		.dbuf_mask = {
1018 			[PIPE_B] = BIT(DBUF_S1),
1019 			[PIPE_C] = BIT(DBUF_S2),
1020 			[PIPE_D] = BIT(DBUF_S2),
1021 		},
1022 	},
1023 	{
1024 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
1025 		.dbuf_mask = {
1026 			[PIPE_A] = BIT(DBUF_S1),
1027 			[PIPE_B] = BIT(DBUF_S1),
1028 			[PIPE_C] = BIT(DBUF_S2),
1029 			[PIPE_D] = BIT(DBUF_S2),
1030 		},
1031 	},
1032 	{}
1033 };
1034 
1035 static const struct dbuf_slice_conf_entry dg2_allowed_dbufs[] = {
1036 	{
1037 		.active_pipes = BIT(PIPE_A),
1038 		.dbuf_mask = {
1039 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1040 		},
1041 	},
1042 	{
1043 		.active_pipes = BIT(PIPE_B),
1044 		.dbuf_mask = {
1045 			[PIPE_B] = BIT(DBUF_S1) | BIT(DBUF_S2),
1046 		},
1047 	},
1048 	{
1049 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B),
1050 		.dbuf_mask = {
1051 			[PIPE_A] = BIT(DBUF_S1),
1052 			[PIPE_B] = BIT(DBUF_S2),
1053 		},
1054 	},
1055 	{
1056 		.active_pipes = BIT(PIPE_C),
1057 		.dbuf_mask = {
1058 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1059 		},
1060 	},
1061 	{
1062 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C),
1063 		.dbuf_mask = {
1064 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1065 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1066 		},
1067 	},
1068 	{
1069 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C),
1070 		.dbuf_mask = {
1071 			[PIPE_B] = BIT(DBUF_S1) | BIT(DBUF_S2),
1072 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1073 		},
1074 	},
1075 	{
1076 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C),
1077 		.dbuf_mask = {
1078 			[PIPE_A] = BIT(DBUF_S1),
1079 			[PIPE_B] = BIT(DBUF_S2),
1080 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1081 		},
1082 	},
1083 	{
1084 		.active_pipes = BIT(PIPE_D),
1085 		.dbuf_mask = {
1086 			[PIPE_D] = BIT(DBUF_S3) | BIT(DBUF_S4),
1087 		},
1088 	},
1089 	{
1090 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_D),
1091 		.dbuf_mask = {
1092 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1093 			[PIPE_D] = BIT(DBUF_S3) | BIT(DBUF_S4),
1094 		},
1095 	},
1096 	{
1097 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_D),
1098 		.dbuf_mask = {
1099 			[PIPE_B] = BIT(DBUF_S1) | BIT(DBUF_S2),
1100 			[PIPE_D] = BIT(DBUF_S3) | BIT(DBUF_S4),
1101 		},
1102 	},
1103 	{
1104 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_D),
1105 		.dbuf_mask = {
1106 			[PIPE_A] = BIT(DBUF_S1),
1107 			[PIPE_B] = BIT(DBUF_S2),
1108 			[PIPE_D] = BIT(DBUF_S3) | BIT(DBUF_S4),
1109 		},
1110 	},
1111 	{
1112 		.active_pipes = BIT(PIPE_C) | BIT(PIPE_D),
1113 		.dbuf_mask = {
1114 			[PIPE_C] = BIT(DBUF_S3),
1115 			[PIPE_D] = BIT(DBUF_S4),
1116 		},
1117 	},
1118 	{
1119 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C) | BIT(PIPE_D),
1120 		.dbuf_mask = {
1121 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1122 			[PIPE_C] = BIT(DBUF_S3),
1123 			[PIPE_D] = BIT(DBUF_S4),
1124 		},
1125 	},
1126 	{
1127 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
1128 		.dbuf_mask = {
1129 			[PIPE_B] = BIT(DBUF_S1) | BIT(DBUF_S2),
1130 			[PIPE_C] = BIT(DBUF_S3),
1131 			[PIPE_D] = BIT(DBUF_S4),
1132 		},
1133 	},
1134 	{
1135 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
1136 		.dbuf_mask = {
1137 			[PIPE_A] = BIT(DBUF_S1),
1138 			[PIPE_B] = BIT(DBUF_S2),
1139 			[PIPE_C] = BIT(DBUF_S3),
1140 			[PIPE_D] = BIT(DBUF_S4),
1141 		},
1142 	},
1143 	{}
1144 };
1145 
1146 static const struct dbuf_slice_conf_entry adlp_allowed_dbufs[] = {
1147 	/*
1148 	 * Keep the join_mbus cases first so check_mbus_joined()
1149 	 * will prefer them over the !join_mbus cases.
1150 	 */
1151 	{
1152 		.active_pipes = BIT(PIPE_A),
1153 		.dbuf_mask = {
1154 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2) | BIT(DBUF_S3) | BIT(DBUF_S4),
1155 		},
1156 		.join_mbus = true,
1157 	},
1158 	{
1159 		.active_pipes = BIT(PIPE_B),
1160 		.dbuf_mask = {
1161 			[PIPE_B] = BIT(DBUF_S1) | BIT(DBUF_S2) | BIT(DBUF_S3) | BIT(DBUF_S4),
1162 		},
1163 		.join_mbus = true,
1164 	},
1165 	{
1166 		.active_pipes = BIT(PIPE_A),
1167 		.dbuf_mask = {
1168 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1169 		},
1170 		.join_mbus = false,
1171 	},
1172 	{
1173 		.active_pipes = BIT(PIPE_B),
1174 		.dbuf_mask = {
1175 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1176 		},
1177 		.join_mbus = false,
1178 	},
1179 	{
1180 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B),
1181 		.dbuf_mask = {
1182 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1183 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1184 		},
1185 	},
1186 	{
1187 		.active_pipes = BIT(PIPE_C),
1188 		.dbuf_mask = {
1189 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1190 		},
1191 	},
1192 	{
1193 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C),
1194 		.dbuf_mask = {
1195 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1196 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1197 		},
1198 	},
1199 	{
1200 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C),
1201 		.dbuf_mask = {
1202 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1203 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1204 		},
1205 	},
1206 	{
1207 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C),
1208 		.dbuf_mask = {
1209 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1210 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1211 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1212 		},
1213 	},
1214 	{
1215 		.active_pipes = BIT(PIPE_D),
1216 		.dbuf_mask = {
1217 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1218 		},
1219 	},
1220 	{
1221 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_D),
1222 		.dbuf_mask = {
1223 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1224 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1225 		},
1226 	},
1227 	{
1228 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_D),
1229 		.dbuf_mask = {
1230 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1231 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1232 		},
1233 	},
1234 	{
1235 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_D),
1236 		.dbuf_mask = {
1237 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1238 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1239 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1240 		},
1241 	},
1242 	{
1243 		.active_pipes = BIT(PIPE_C) | BIT(PIPE_D),
1244 		.dbuf_mask = {
1245 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1246 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1247 		},
1248 	},
1249 	{
1250 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C) | BIT(PIPE_D),
1251 		.dbuf_mask = {
1252 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1253 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1254 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1255 		},
1256 	},
1257 	{
1258 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
1259 		.dbuf_mask = {
1260 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1261 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1262 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1263 		},
1264 	},
1265 	{
1266 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
1267 		.dbuf_mask = {
1268 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1269 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1270 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1271 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1272 		},
1273 	},
1274 	{}
1275 
1276 };
1277 
1278 static bool check_mbus_joined(u8 active_pipes,
1279 			      const struct dbuf_slice_conf_entry *dbuf_slices)
1280 {
1281 	int i;
1282 
1283 	for (i = 0; dbuf_slices[i].active_pipes != 0; i++) {
1284 		if (dbuf_slices[i].active_pipes == active_pipes)
1285 			return dbuf_slices[i].join_mbus;
1286 	}
1287 	return false;
1288 }
1289 
1290 static bool adlp_check_mbus_joined(u8 active_pipes)
1291 {
1292 	return check_mbus_joined(active_pipes, adlp_allowed_dbufs);
1293 }
1294 
1295 static u8 compute_dbuf_slices(enum pipe pipe, u8 active_pipes, bool join_mbus,
1296 			      const struct dbuf_slice_conf_entry *dbuf_slices)
1297 {
1298 	int i;
1299 
1300 	for (i = 0; dbuf_slices[i].active_pipes != 0; i++) {
1301 		if (dbuf_slices[i].active_pipes == active_pipes &&
1302 		    dbuf_slices[i].join_mbus == join_mbus)
1303 			return dbuf_slices[i].dbuf_mask[pipe];
1304 	}
1305 	return 0;
1306 }
1307 
1308 /*
1309  * This function finds an entry with same enabled pipe configuration and
1310  * returns correspondent DBuf slice mask as stated in BSpec for particular
1311  * platform.
1312  */
1313 static u8 icl_compute_dbuf_slices(enum pipe pipe, u8 active_pipes, bool join_mbus)
1314 {
1315 	/*
1316 	 * FIXME: For ICL this is still a bit unclear as prev BSpec revision
1317 	 * required calculating "pipe ratio" in order to determine
1318 	 * if one or two slices can be used for single pipe configurations
1319 	 * as additional constraint to the existing table.
1320 	 * However based on recent info, it should be not "pipe ratio"
1321 	 * but rather ratio between pixel_rate and cdclk with additional
1322 	 * constants, so for now we are using only table until this is
1323 	 * clarified. Also this is the reason why crtc_state param is
1324 	 * still here - we will need it once those additional constraints
1325 	 * pop up.
1326 	 */
1327 	return compute_dbuf_slices(pipe, active_pipes, join_mbus,
1328 				   icl_allowed_dbufs);
1329 }
1330 
1331 static u8 tgl_compute_dbuf_slices(enum pipe pipe, u8 active_pipes, bool join_mbus)
1332 {
1333 	return compute_dbuf_slices(pipe, active_pipes, join_mbus,
1334 				   tgl_allowed_dbufs);
1335 }
1336 
1337 static u8 adlp_compute_dbuf_slices(enum pipe pipe, u8 active_pipes, bool join_mbus)
1338 {
1339 	return compute_dbuf_slices(pipe, active_pipes, join_mbus,
1340 				   adlp_allowed_dbufs);
1341 }
1342 
1343 static u8 dg2_compute_dbuf_slices(enum pipe pipe, u8 active_pipes, bool join_mbus)
1344 {
1345 	return compute_dbuf_slices(pipe, active_pipes, join_mbus,
1346 				   dg2_allowed_dbufs);
1347 }
1348 
1349 static u8 skl_compute_dbuf_slices(struct intel_crtc *crtc, u8 active_pipes, bool join_mbus)
1350 {
1351 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
1352 	enum pipe pipe = crtc->pipe;
1353 
1354 	if (IS_DG2(i915))
1355 		return dg2_compute_dbuf_slices(pipe, active_pipes, join_mbus);
1356 	else if (DISPLAY_VER(i915) >= 13)
1357 		return adlp_compute_dbuf_slices(pipe, active_pipes, join_mbus);
1358 	else if (DISPLAY_VER(i915) == 12)
1359 		return tgl_compute_dbuf_slices(pipe, active_pipes, join_mbus);
1360 	else if (DISPLAY_VER(i915) == 11)
1361 		return icl_compute_dbuf_slices(pipe, active_pipes, join_mbus);
1362 	/*
1363 	 * For anything else just return one slice yet.
1364 	 * Should be extended for other platforms.
1365 	 */
1366 	return active_pipes & BIT(pipe) ? BIT(DBUF_S1) : 0;
1367 }
1368 
1369 static bool
1370 use_minimal_wm0_only(const struct intel_crtc_state *crtc_state,
1371 		     struct intel_plane *plane)
1372 {
1373 	struct drm_i915_private *i915 = to_i915(plane->base.dev);
1374 
1375 	return DISPLAY_VER(i915) >= 13 &&
1376 	       crtc_state->uapi.async_flip &&
1377 	       plane->async_flip;
1378 }
1379 
1380 static u64
1381 skl_total_relative_data_rate(const struct intel_crtc_state *crtc_state)
1382 {
1383 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
1384 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
1385 	enum plane_id plane_id;
1386 	u64 data_rate = 0;
1387 
1388 	for_each_plane_id_on_crtc(crtc, plane_id) {
1389 		if (plane_id == PLANE_CURSOR)
1390 			continue;
1391 
1392 		data_rate += crtc_state->rel_data_rate[plane_id];
1393 
1394 		if (DISPLAY_VER(i915) < 11)
1395 			data_rate += crtc_state->rel_data_rate_y[plane_id];
1396 	}
1397 
1398 	return data_rate;
1399 }
1400 
1401 const struct skl_wm_level *
1402 skl_plane_wm_level(const struct skl_pipe_wm *pipe_wm,
1403 		   enum plane_id plane_id,
1404 		   int level)
1405 {
1406 	const struct skl_plane_wm *wm = &pipe_wm->planes[plane_id];
1407 
1408 	if (level == 0 && pipe_wm->use_sagv_wm)
1409 		return &wm->sagv.wm0;
1410 
1411 	return &wm->wm[level];
1412 }
1413 
1414 const struct skl_wm_level *
1415 skl_plane_trans_wm(const struct skl_pipe_wm *pipe_wm,
1416 		   enum plane_id plane_id)
1417 {
1418 	const struct skl_plane_wm *wm = &pipe_wm->planes[plane_id];
1419 
1420 	if (pipe_wm->use_sagv_wm)
1421 		return &wm->sagv.trans_wm;
1422 
1423 	return &wm->trans_wm;
1424 }
1425 
1426 /*
1427  * We only disable the watermarks for each plane if
1428  * they exceed the ddb allocation of said plane. This
1429  * is done so that we don't end up touching cursor
1430  * watermarks needlessly when some other plane reduces
1431  * our max possible watermark level.
1432  *
1433  * Bspec has this to say about the PLANE_WM enable bit:
1434  * "All the watermarks at this level for all enabled
1435  *  planes must be enabled before the level will be used."
1436  * So this is actually safe to do.
1437  */
1438 static void
1439 skl_check_wm_level(struct skl_wm_level *wm, const struct skl_ddb_entry *ddb)
1440 {
1441 	if (wm->min_ddb_alloc > skl_ddb_entry_size(ddb))
1442 		memset(wm, 0, sizeof(*wm));
1443 }
1444 
1445 static void
1446 skl_check_nv12_wm_level(struct skl_wm_level *wm, struct skl_wm_level *uv_wm,
1447 			const struct skl_ddb_entry *ddb_y, const struct skl_ddb_entry *ddb)
1448 {
1449 	if (wm->min_ddb_alloc > skl_ddb_entry_size(ddb_y) ||
1450 	    uv_wm->min_ddb_alloc > skl_ddb_entry_size(ddb)) {
1451 		memset(wm, 0, sizeof(*wm));
1452 		memset(uv_wm, 0, sizeof(*uv_wm));
1453 	}
1454 }
1455 
1456 static bool skl_need_wm_copy_wa(struct drm_i915_private *i915, int level,
1457 				const struct skl_plane_wm *wm)
1458 {
1459 	/*
1460 	 * Wa_1408961008:icl, ehl
1461 	 * Wa_14012656716:tgl, adl
1462 	 * Wa_14017887344:icl
1463 	 * Wa_14017868169:adl, tgl
1464 	 * Due to some power saving optimizations, different subsystems
1465 	 * like PSR, might still use even disabled wm level registers,
1466 	 * for "reference", so lets keep at least the values sane.
1467 	 * Considering amount of WA requiring us to do similar things, was
1468 	 * decided to simply do it for all of the platforms, as those wm
1469 	 * levels are disabled, this isn't going to do harm anyway.
1470 	 */
1471 	return level > 0 && !wm->wm[level].enable;
1472 }
1473 
1474 struct skl_plane_ddb_iter {
1475 	u64 data_rate;
1476 	u16 start, size;
1477 };
1478 
1479 static void
1480 skl_allocate_plane_ddb(struct skl_plane_ddb_iter *iter,
1481 		       struct skl_ddb_entry *ddb,
1482 		       const struct skl_wm_level *wm,
1483 		       u64 data_rate)
1484 {
1485 	u16 size, extra = 0;
1486 
1487 	if (data_rate) {
1488 		extra = min_t(u16, iter->size,
1489 			      DIV64_U64_ROUND_UP(iter->size * data_rate,
1490 						 iter->data_rate));
1491 		iter->size -= extra;
1492 		iter->data_rate -= data_rate;
1493 	}
1494 
1495 	/*
1496 	 * Keep ddb entry of all disabled planes explicitly zeroed
1497 	 * to avoid skl_ddb_add_affected_planes() adding them to
1498 	 * the state when other planes change their allocations.
1499 	 */
1500 	size = wm->min_ddb_alloc + extra;
1501 	if (size)
1502 		iter->start = skl_ddb_entry_init(ddb, iter->start,
1503 						 iter->start + size);
1504 }
1505 
1506 static int
1507 skl_crtc_allocate_plane_ddb(struct intel_atomic_state *state,
1508 			    struct intel_crtc *crtc)
1509 {
1510 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
1511 	struct intel_crtc_state *crtc_state =
1512 		intel_atomic_get_new_crtc_state(state, crtc);
1513 	const struct intel_dbuf_state *dbuf_state =
1514 		intel_atomic_get_new_dbuf_state(state);
1515 	const struct skl_ddb_entry *alloc = &dbuf_state->ddb[crtc->pipe];
1516 	int num_active = hweight8(dbuf_state->active_pipes);
1517 	struct skl_plane_ddb_iter iter;
1518 	enum plane_id plane_id;
1519 	u16 cursor_size;
1520 	u32 blocks;
1521 	int level;
1522 
1523 	/* Clear the partitioning for disabled planes. */
1524 	memset(crtc_state->wm.skl.plane_ddb, 0, sizeof(crtc_state->wm.skl.plane_ddb));
1525 	memset(crtc_state->wm.skl.plane_ddb_y, 0, sizeof(crtc_state->wm.skl.plane_ddb_y));
1526 
1527 	if (!crtc_state->hw.active)
1528 		return 0;
1529 
1530 	iter.start = alloc->start;
1531 	iter.size = skl_ddb_entry_size(alloc);
1532 	if (iter.size == 0)
1533 		return 0;
1534 
1535 	/* Allocate fixed number of blocks for cursor. */
1536 	cursor_size = skl_cursor_allocation(crtc_state, num_active);
1537 	iter.size -= cursor_size;
1538 	skl_ddb_entry_init(&crtc_state->wm.skl.plane_ddb[PLANE_CURSOR],
1539 			   alloc->end - cursor_size, alloc->end);
1540 
1541 	iter.data_rate = skl_total_relative_data_rate(crtc_state);
1542 
1543 	/*
1544 	 * Find the highest watermark level for which we can satisfy the block
1545 	 * requirement of active planes.
1546 	 */
1547 	for (level = i915->display.wm.num_levels - 1; level >= 0; level--) {
1548 		blocks = 0;
1549 		for_each_plane_id_on_crtc(crtc, plane_id) {
1550 			const struct skl_plane_wm *wm =
1551 				&crtc_state->wm.skl.optimal.planes[plane_id];
1552 
1553 			if (plane_id == PLANE_CURSOR) {
1554 				const struct skl_ddb_entry *ddb =
1555 					&crtc_state->wm.skl.plane_ddb[plane_id];
1556 
1557 				if (wm->wm[level].min_ddb_alloc > skl_ddb_entry_size(ddb)) {
1558 					drm_WARN_ON(&i915->drm,
1559 						    wm->wm[level].min_ddb_alloc != U16_MAX);
1560 					blocks = U32_MAX;
1561 					break;
1562 				}
1563 				continue;
1564 			}
1565 
1566 			blocks += wm->wm[level].min_ddb_alloc;
1567 			blocks += wm->uv_wm[level].min_ddb_alloc;
1568 		}
1569 
1570 		if (blocks <= iter.size) {
1571 			iter.size -= blocks;
1572 			break;
1573 		}
1574 	}
1575 
1576 	if (level < 0) {
1577 		drm_dbg_kms(&i915->drm,
1578 			    "Requested display configuration exceeds system DDB limitations");
1579 		drm_dbg_kms(&i915->drm, "minimum required %d/%d\n",
1580 			    blocks, iter.size);
1581 		return -EINVAL;
1582 	}
1583 
1584 	/* avoid the WARN later when we don't allocate any extra DDB */
1585 	if (iter.data_rate == 0)
1586 		iter.size = 0;
1587 
1588 	/*
1589 	 * Grant each plane the blocks it requires at the highest achievable
1590 	 * watermark level, plus an extra share of the leftover blocks
1591 	 * proportional to its relative data rate.
1592 	 */
1593 	for_each_plane_id_on_crtc(crtc, plane_id) {
1594 		struct skl_ddb_entry *ddb =
1595 			&crtc_state->wm.skl.plane_ddb[plane_id];
1596 		struct skl_ddb_entry *ddb_y =
1597 			&crtc_state->wm.skl.plane_ddb_y[plane_id];
1598 		const struct skl_plane_wm *wm =
1599 			&crtc_state->wm.skl.optimal.planes[plane_id];
1600 
1601 		if (plane_id == PLANE_CURSOR)
1602 			continue;
1603 
1604 		if (DISPLAY_VER(i915) < 11 &&
1605 		    crtc_state->nv12_planes & BIT(plane_id)) {
1606 			skl_allocate_plane_ddb(&iter, ddb_y, &wm->wm[level],
1607 					       crtc_state->rel_data_rate_y[plane_id]);
1608 			skl_allocate_plane_ddb(&iter, ddb, &wm->uv_wm[level],
1609 					       crtc_state->rel_data_rate[plane_id]);
1610 		} else {
1611 			skl_allocate_plane_ddb(&iter, ddb, &wm->wm[level],
1612 					       crtc_state->rel_data_rate[plane_id]);
1613 		}
1614 	}
1615 	drm_WARN_ON(&i915->drm, iter.size != 0 || iter.data_rate != 0);
1616 
1617 	/*
1618 	 * When we calculated watermark values we didn't know how high
1619 	 * of a level we'd actually be able to hit, so we just marked
1620 	 * all levels as "enabled."  Go back now and disable the ones
1621 	 * that aren't actually possible.
1622 	 */
1623 	for (level++; level < i915->display.wm.num_levels; level++) {
1624 		for_each_plane_id_on_crtc(crtc, plane_id) {
1625 			const struct skl_ddb_entry *ddb =
1626 				&crtc_state->wm.skl.plane_ddb[plane_id];
1627 			const struct skl_ddb_entry *ddb_y =
1628 				&crtc_state->wm.skl.plane_ddb_y[plane_id];
1629 			struct skl_plane_wm *wm =
1630 				&crtc_state->wm.skl.optimal.planes[plane_id];
1631 
1632 			if (DISPLAY_VER(i915) < 11 &&
1633 			    crtc_state->nv12_planes & BIT(plane_id))
1634 				skl_check_nv12_wm_level(&wm->wm[level],
1635 							&wm->uv_wm[level],
1636 							ddb_y, ddb);
1637 			else
1638 				skl_check_wm_level(&wm->wm[level], ddb);
1639 
1640 			if (skl_need_wm_copy_wa(i915, level, wm)) {
1641 				wm->wm[level].blocks = wm->wm[level - 1].blocks;
1642 				wm->wm[level].lines = wm->wm[level - 1].lines;
1643 				wm->wm[level].ignore_lines = wm->wm[level - 1].ignore_lines;
1644 			}
1645 		}
1646 	}
1647 
1648 	/*
1649 	 * Go back and disable the transition and SAGV watermarks
1650 	 * if it turns out we don't have enough DDB blocks for them.
1651 	 */
1652 	for_each_plane_id_on_crtc(crtc, plane_id) {
1653 		const struct skl_ddb_entry *ddb =
1654 			&crtc_state->wm.skl.plane_ddb[plane_id];
1655 		const struct skl_ddb_entry *ddb_y =
1656 			&crtc_state->wm.skl.plane_ddb_y[plane_id];
1657 		struct skl_plane_wm *wm =
1658 			&crtc_state->wm.skl.optimal.planes[plane_id];
1659 
1660 		if (DISPLAY_VER(i915) < 11 &&
1661 		    crtc_state->nv12_planes & BIT(plane_id)) {
1662 			skl_check_wm_level(&wm->trans_wm, ddb_y);
1663 		} else {
1664 			WARN_ON(skl_ddb_entry_size(ddb_y));
1665 
1666 			skl_check_wm_level(&wm->trans_wm, ddb);
1667 		}
1668 
1669 		skl_check_wm_level(&wm->sagv.wm0, ddb);
1670 		skl_check_wm_level(&wm->sagv.trans_wm, ddb);
1671 	}
1672 
1673 	return 0;
1674 }
1675 
1676 /*
1677  * The max latency should be 257 (max the punit can code is 255 and we add 2us
1678  * for the read latency) and cpp should always be <= 8, so that
1679  * should allow pixel_rate up to ~2 GHz which seems sufficient since max
1680  * 2xcdclk is 1350 MHz and the pixel rate should never exceed that.
1681  */
1682 static uint_fixed_16_16_t
1683 skl_wm_method1(const struct drm_i915_private *i915, u32 pixel_rate,
1684 	       u8 cpp, u32 latency, u32 dbuf_block_size)
1685 {
1686 	u32 wm_intermediate_val;
1687 	uint_fixed_16_16_t ret;
1688 
1689 	if (latency == 0)
1690 		return FP_16_16_MAX;
1691 
1692 	wm_intermediate_val = latency * pixel_rate * cpp;
1693 	ret = div_fixed16(wm_intermediate_val, 1000 * dbuf_block_size);
1694 
1695 	if (DISPLAY_VER(i915) >= 10)
1696 		ret = add_fixed16_u32(ret, 1);
1697 
1698 	return ret;
1699 }
1700 
1701 static uint_fixed_16_16_t
1702 skl_wm_method2(u32 pixel_rate, u32 pipe_htotal, u32 latency,
1703 	       uint_fixed_16_16_t plane_blocks_per_line)
1704 {
1705 	u32 wm_intermediate_val;
1706 	uint_fixed_16_16_t ret;
1707 
1708 	if (latency == 0)
1709 		return FP_16_16_MAX;
1710 
1711 	wm_intermediate_val = latency * pixel_rate;
1712 	wm_intermediate_val = DIV_ROUND_UP(wm_intermediate_val,
1713 					   pipe_htotal * 1000);
1714 	ret = mul_u32_fixed16(wm_intermediate_val, plane_blocks_per_line);
1715 	return ret;
1716 }
1717 
1718 static uint_fixed_16_16_t
1719 intel_get_linetime_us(const struct intel_crtc_state *crtc_state)
1720 {
1721 	struct drm_i915_private *i915 = to_i915(crtc_state->uapi.crtc->dev);
1722 	u32 pixel_rate;
1723 	u32 crtc_htotal;
1724 	uint_fixed_16_16_t linetime_us;
1725 
1726 	if (!crtc_state->hw.active)
1727 		return u32_to_fixed16(0);
1728 
1729 	pixel_rate = crtc_state->pixel_rate;
1730 
1731 	if (drm_WARN_ON(&i915->drm, pixel_rate == 0))
1732 		return u32_to_fixed16(0);
1733 
1734 	crtc_htotal = crtc_state->hw.pipe_mode.crtc_htotal;
1735 	linetime_us = div_fixed16(crtc_htotal * 1000, pixel_rate);
1736 
1737 	return linetime_us;
1738 }
1739 
1740 static int
1741 skl_compute_wm_params(const struct intel_crtc_state *crtc_state,
1742 		      int width, const struct drm_format_info *format,
1743 		      u64 modifier, unsigned int rotation,
1744 		      u32 plane_pixel_rate, struct skl_wm_params *wp,
1745 		      int color_plane, unsigned int pan_x)
1746 {
1747 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
1748 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
1749 	u32 interm_pbpl;
1750 
1751 	/* only planar format has two planes */
1752 	if (color_plane == 1 &&
1753 	    !intel_format_info_is_yuv_semiplanar(format, modifier)) {
1754 		drm_dbg_kms(&i915->drm,
1755 			    "Non planar format have single plane\n");
1756 		return -EINVAL;
1757 	}
1758 
1759 	wp->x_tiled = modifier == I915_FORMAT_MOD_X_TILED;
1760 	wp->y_tiled = modifier != I915_FORMAT_MOD_X_TILED &&
1761 		intel_fb_is_tiled_modifier(modifier);
1762 	wp->rc_surface = intel_fb_is_ccs_modifier(modifier);
1763 	wp->is_planar = intel_format_info_is_yuv_semiplanar(format, modifier);
1764 
1765 	wp->width = width;
1766 	if (color_plane == 1 && wp->is_planar)
1767 		wp->width /= 2;
1768 
1769 	wp->cpp = format->cpp[color_plane];
1770 	wp->plane_pixel_rate = plane_pixel_rate;
1771 
1772 	if (DISPLAY_VER(i915) >= 11 &&
1773 	    modifier == I915_FORMAT_MOD_Yf_TILED  && wp->cpp == 1)
1774 		wp->dbuf_block_size = 256;
1775 	else
1776 		wp->dbuf_block_size = 512;
1777 
1778 	if (drm_rotation_90_or_270(rotation)) {
1779 		switch (wp->cpp) {
1780 		case 1:
1781 			wp->y_min_scanlines = 16;
1782 			break;
1783 		case 2:
1784 			wp->y_min_scanlines = 8;
1785 			break;
1786 		case 4:
1787 			wp->y_min_scanlines = 4;
1788 			break;
1789 		default:
1790 			MISSING_CASE(wp->cpp);
1791 			return -EINVAL;
1792 		}
1793 	} else {
1794 		wp->y_min_scanlines = 4;
1795 	}
1796 
1797 	if (skl_needs_memory_bw_wa(i915))
1798 		wp->y_min_scanlines *= 2;
1799 
1800 	wp->plane_bytes_per_line = wp->width * wp->cpp;
1801 	if (wp->y_tiled) {
1802 		interm_pbpl = DIV_ROUND_UP(wp->plane_bytes_per_line *
1803 					   wp->y_min_scanlines,
1804 					   wp->dbuf_block_size);
1805 
1806 		if (DISPLAY_VER(i915) >= 30)
1807 			interm_pbpl += (pan_x != 0);
1808 		else if (DISPLAY_VER(i915) >= 10)
1809 			interm_pbpl++;
1810 
1811 		wp->plane_blocks_per_line = div_fixed16(interm_pbpl,
1812 							wp->y_min_scanlines);
1813 	} else {
1814 		interm_pbpl = DIV_ROUND_UP(wp->plane_bytes_per_line,
1815 					   wp->dbuf_block_size);
1816 
1817 		if (!wp->x_tiled || DISPLAY_VER(i915) >= 10)
1818 			interm_pbpl++;
1819 
1820 		wp->plane_blocks_per_line = u32_to_fixed16(interm_pbpl);
1821 	}
1822 
1823 	wp->y_tile_minimum = mul_u32_fixed16(wp->y_min_scanlines,
1824 					     wp->plane_blocks_per_line);
1825 
1826 	wp->linetime_us = fixed16_to_u32_round_up(intel_get_linetime_us(crtc_state));
1827 
1828 	return 0;
1829 }
1830 
1831 static int
1832 skl_compute_plane_wm_params(const struct intel_crtc_state *crtc_state,
1833 			    const struct intel_plane_state *plane_state,
1834 			    struct skl_wm_params *wp, int color_plane)
1835 {
1836 	const struct drm_framebuffer *fb = plane_state->hw.fb;
1837 	int width;
1838 
1839 	/*
1840 	 * Src coordinates are already rotated by 270 degrees for
1841 	 * the 90/270 degree plane rotation cases (to match the
1842 	 * GTT mapping), hence no need to account for rotation here.
1843 	 */
1844 	width = drm_rect_width(&plane_state->uapi.src) >> 16;
1845 
1846 	return skl_compute_wm_params(crtc_state, width,
1847 				     fb->format, fb->modifier,
1848 				     plane_state->hw.rotation,
1849 				     intel_plane_pixel_rate(crtc_state, plane_state),
1850 				     wp, color_plane,
1851 				     plane_state->uapi.src.x1);
1852 }
1853 
1854 static bool skl_wm_has_lines(struct drm_i915_private *i915, int level)
1855 {
1856 	if (DISPLAY_VER(i915) >= 10)
1857 		return true;
1858 
1859 	/* The number of lines are ignored for the level 0 watermark. */
1860 	return level > 0;
1861 }
1862 
1863 static int skl_wm_max_lines(struct drm_i915_private *i915)
1864 {
1865 	if (DISPLAY_VER(i915) >= 13)
1866 		return 255;
1867 	else
1868 		return 31;
1869 }
1870 
1871 static void skl_compute_plane_wm(const struct intel_crtc_state *crtc_state,
1872 				 struct intel_plane *plane,
1873 				 int level,
1874 				 unsigned int latency,
1875 				 const struct skl_wm_params *wp,
1876 				 const struct skl_wm_level *result_prev,
1877 				 struct skl_wm_level *result /* out */)
1878 {
1879 	struct drm_i915_private *i915 = to_i915(crtc_state->uapi.crtc->dev);
1880 	uint_fixed_16_16_t method1, method2;
1881 	uint_fixed_16_16_t selected_result;
1882 	u32 blocks, lines, min_ddb_alloc = 0;
1883 
1884 	if (latency == 0 ||
1885 	    (use_minimal_wm0_only(crtc_state, plane) && level > 0)) {
1886 		/* reject it */
1887 		result->min_ddb_alloc = U16_MAX;
1888 		return;
1889 	}
1890 
1891 	method1 = skl_wm_method1(i915, wp->plane_pixel_rate,
1892 				 wp->cpp, latency, wp->dbuf_block_size);
1893 	method2 = skl_wm_method2(wp->plane_pixel_rate,
1894 				 crtc_state->hw.pipe_mode.crtc_htotal,
1895 				 latency,
1896 				 wp->plane_blocks_per_line);
1897 
1898 	if (wp->y_tiled) {
1899 		selected_result = max_fixed16(method2, wp->y_tile_minimum);
1900 	} else {
1901 		if ((wp->cpp * crtc_state->hw.pipe_mode.crtc_htotal /
1902 		     wp->dbuf_block_size < 1) &&
1903 		     (wp->plane_bytes_per_line / wp->dbuf_block_size < 1)) {
1904 			selected_result = method2;
1905 		} else if (latency >= wp->linetime_us) {
1906 			if (DISPLAY_VER(i915) == 9)
1907 				selected_result = min_fixed16(method1, method2);
1908 			else
1909 				selected_result = method2;
1910 		} else {
1911 			selected_result = method1;
1912 		}
1913 	}
1914 
1915 	blocks = fixed16_to_u32_round_up(selected_result);
1916 	if (DISPLAY_VER(i915) < 30)
1917 		blocks++;
1918 
1919 	/*
1920 	 * Lets have blocks at minimum equivalent to plane_blocks_per_line
1921 	 * as there will be at minimum one line for lines configuration. This
1922 	 * is a work around for FIFO underruns observed with resolutions like
1923 	 * 4k 60 Hz in single channel DRAM configurations.
1924 	 *
1925 	 * As per the Bspec 49325, if the ddb allocation can hold at least
1926 	 * one plane_blocks_per_line, we should have selected method2 in
1927 	 * the above logic. Assuming that modern versions have enough dbuf
1928 	 * and method2 guarantees blocks equivalent to at least 1 line,
1929 	 * select the blocks as plane_blocks_per_line.
1930 	 *
1931 	 * TODO: Revisit the logic when we have better understanding on DRAM
1932 	 * channels' impact on the level 0 memory latency and the relevant
1933 	 * wm calculations.
1934 	 */
1935 	if (skl_wm_has_lines(i915, level))
1936 		blocks = max(blocks,
1937 			     fixed16_to_u32_round_up(wp->plane_blocks_per_line));
1938 	lines = div_round_up_fixed16(selected_result,
1939 				     wp->plane_blocks_per_line);
1940 
1941 	if (DISPLAY_VER(i915) == 9) {
1942 		/* Display WA #1125: skl,bxt,kbl */
1943 		if (level == 0 && wp->rc_surface)
1944 			blocks += fixed16_to_u32_round_up(wp->y_tile_minimum);
1945 
1946 		/* Display WA #1126: skl,bxt,kbl */
1947 		if (level >= 1 && level <= 7) {
1948 			if (wp->y_tiled) {
1949 				blocks += fixed16_to_u32_round_up(wp->y_tile_minimum);
1950 				lines += wp->y_min_scanlines;
1951 			} else {
1952 				blocks++;
1953 			}
1954 
1955 			/*
1956 			 * Make sure result blocks for higher latency levels are
1957 			 * at least as high as level below the current level.
1958 			 * Assumption in DDB algorithm optimization for special
1959 			 * cases. Also covers Display WA #1125 for RC.
1960 			 */
1961 			if (result_prev->blocks > blocks)
1962 				blocks = result_prev->blocks;
1963 		}
1964 	}
1965 
1966 	if (DISPLAY_VER(i915) >= 11) {
1967 		if (wp->y_tiled) {
1968 			int extra_lines;
1969 
1970 			if (lines % wp->y_min_scanlines == 0)
1971 				extra_lines = wp->y_min_scanlines;
1972 			else
1973 				extra_lines = wp->y_min_scanlines * 2 -
1974 					lines % wp->y_min_scanlines;
1975 
1976 			min_ddb_alloc = mul_round_up_u32_fixed16(lines + extra_lines,
1977 								 wp->plane_blocks_per_line);
1978 		} else {
1979 			min_ddb_alloc = blocks + DIV_ROUND_UP(blocks, 10);
1980 		}
1981 	}
1982 
1983 	if (!skl_wm_has_lines(i915, level))
1984 		lines = 0;
1985 
1986 	if (lines > skl_wm_max_lines(i915)) {
1987 		/* reject it */
1988 		result->min_ddb_alloc = U16_MAX;
1989 		return;
1990 	}
1991 
1992 	/*
1993 	 * If lines is valid, assume we can use this watermark level
1994 	 * for now.  We'll come back and disable it after we calculate the
1995 	 * DDB allocation if it turns out we don't actually have enough
1996 	 * blocks to satisfy it.
1997 	 */
1998 	result->blocks = blocks;
1999 	result->lines = lines;
2000 	/* Bspec says: value >= plane ddb allocation -> invalid, hence the +1 here */
2001 	result->min_ddb_alloc = max(min_ddb_alloc, blocks) + 1;
2002 	result->enable = true;
2003 
2004 	if (DISPLAY_VER(i915) < 12 && i915->display.sagv.block_time_us)
2005 		result->can_sagv = latency >= i915->display.sagv.block_time_us;
2006 }
2007 
2008 static void
2009 skl_compute_wm_levels(const struct intel_crtc_state *crtc_state,
2010 		      struct intel_plane *plane,
2011 		      const struct skl_wm_params *wm_params,
2012 		      struct skl_wm_level *levels)
2013 {
2014 	struct drm_i915_private *i915 = to_i915(crtc_state->uapi.crtc->dev);
2015 	struct skl_wm_level *result_prev = &levels[0];
2016 	int level;
2017 
2018 	for (level = 0; level < i915->display.wm.num_levels; level++) {
2019 		struct skl_wm_level *result = &levels[level];
2020 		unsigned int latency = skl_wm_latency(i915, level, wm_params);
2021 
2022 		skl_compute_plane_wm(crtc_state, plane, level, latency,
2023 				     wm_params, result_prev, result);
2024 
2025 		result_prev = result;
2026 	}
2027 }
2028 
2029 static void tgl_compute_sagv_wm(const struct intel_crtc_state *crtc_state,
2030 				struct intel_plane *plane,
2031 				const struct skl_wm_params *wm_params,
2032 				struct skl_plane_wm *plane_wm)
2033 {
2034 	struct drm_i915_private *i915 = to_i915(crtc_state->uapi.crtc->dev);
2035 	struct skl_wm_level *sagv_wm = &plane_wm->sagv.wm0;
2036 	struct skl_wm_level *levels = plane_wm->wm;
2037 	unsigned int latency = 0;
2038 
2039 	if (i915->display.sagv.block_time_us)
2040 		latency = i915->display.sagv.block_time_us +
2041 			skl_wm_latency(i915, 0, wm_params);
2042 
2043 	skl_compute_plane_wm(crtc_state, plane, 0, latency,
2044 			     wm_params, &levels[0],
2045 			     sagv_wm);
2046 }
2047 
2048 static void skl_compute_transition_wm(struct drm_i915_private *i915,
2049 				      struct skl_wm_level *trans_wm,
2050 				      const struct skl_wm_level *wm0,
2051 				      const struct skl_wm_params *wp)
2052 {
2053 	u16 trans_min, trans_amount, trans_y_tile_min;
2054 	u16 wm0_blocks, trans_offset, blocks;
2055 
2056 	/* Transition WM don't make any sense if ipc is disabled */
2057 	if (!skl_watermark_ipc_enabled(i915))
2058 		return;
2059 
2060 	/*
2061 	 * WaDisableTWM:skl,kbl,cfl,bxt
2062 	 * Transition WM are not recommended by HW team for GEN9
2063 	 */
2064 	if (DISPLAY_VER(i915) == 9)
2065 		return;
2066 
2067 	if (DISPLAY_VER(i915) >= 11)
2068 		trans_min = 4;
2069 	else
2070 		trans_min = 14;
2071 
2072 	/* Display WA #1140: glk,cnl */
2073 	if (DISPLAY_VER(i915) == 10)
2074 		trans_amount = 0;
2075 	else
2076 		trans_amount = 10; /* This is configurable amount */
2077 
2078 	trans_offset = trans_min + trans_amount;
2079 
2080 	/*
2081 	 * The spec asks for Selected Result Blocks for wm0 (the real value),
2082 	 * not Result Blocks (the integer value). Pay attention to the capital
2083 	 * letters. The value wm_l0->blocks is actually Result Blocks, but
2084 	 * since Result Blocks is the ceiling of Selected Result Blocks plus 1,
2085 	 * and since we later will have to get the ceiling of the sum in the
2086 	 * transition watermarks calculation, we can just pretend Selected
2087 	 * Result Blocks is Result Blocks minus 1 and it should work for the
2088 	 * current platforms.
2089 	 */
2090 	wm0_blocks = wm0->blocks - 1;
2091 
2092 	if (wp->y_tiled) {
2093 		trans_y_tile_min =
2094 			(u16)mul_round_up_u32_fixed16(2, wp->y_tile_minimum);
2095 		blocks = max(wm0_blocks, trans_y_tile_min) + trans_offset;
2096 	} else {
2097 		blocks = wm0_blocks + trans_offset;
2098 	}
2099 	blocks++;
2100 
2101 	/*
2102 	 * Just assume we can enable the transition watermark.  After
2103 	 * computing the DDB we'll come back and disable it if that
2104 	 * assumption turns out to be false.
2105 	 */
2106 	trans_wm->blocks = blocks;
2107 	trans_wm->min_ddb_alloc = max_t(u16, wm0->min_ddb_alloc, blocks + 1);
2108 	trans_wm->enable = true;
2109 }
2110 
2111 static int skl_build_plane_wm_single(struct intel_crtc_state *crtc_state,
2112 				     const struct intel_plane_state *plane_state,
2113 				     struct intel_plane *plane, int color_plane)
2114 {
2115 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
2116 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
2117 	struct skl_plane_wm *wm = &crtc_state->wm.skl.raw.planes[plane->id];
2118 	struct skl_wm_params wm_params;
2119 	int ret;
2120 
2121 	ret = skl_compute_plane_wm_params(crtc_state, plane_state,
2122 					  &wm_params, color_plane);
2123 	if (ret)
2124 		return ret;
2125 
2126 	skl_compute_wm_levels(crtc_state, plane, &wm_params, wm->wm);
2127 
2128 	skl_compute_transition_wm(i915, &wm->trans_wm,
2129 				  &wm->wm[0], &wm_params);
2130 
2131 	if (DISPLAY_VER(i915) >= 12) {
2132 		tgl_compute_sagv_wm(crtc_state, plane, &wm_params, wm);
2133 
2134 		skl_compute_transition_wm(i915, &wm->sagv.trans_wm,
2135 					  &wm->sagv.wm0, &wm_params);
2136 	}
2137 
2138 	return 0;
2139 }
2140 
2141 static int skl_build_plane_wm_uv(struct intel_crtc_state *crtc_state,
2142 				 const struct intel_plane_state *plane_state,
2143 				 struct intel_plane *plane)
2144 {
2145 	struct skl_plane_wm *wm = &crtc_state->wm.skl.raw.planes[plane->id];
2146 	struct skl_wm_params wm_params;
2147 	int ret;
2148 
2149 	wm->is_planar = true;
2150 
2151 	/* uv plane watermarks must also be validated for NV12/Planar */
2152 	ret = skl_compute_plane_wm_params(crtc_state, plane_state,
2153 					  &wm_params, 1);
2154 	if (ret)
2155 		return ret;
2156 
2157 	skl_compute_wm_levels(crtc_state, plane, &wm_params, wm->uv_wm);
2158 
2159 	return 0;
2160 }
2161 
2162 static int skl_build_plane_wm(struct intel_crtc_state *crtc_state,
2163 			      const struct intel_plane_state *plane_state)
2164 {
2165 	struct intel_plane *plane = to_intel_plane(plane_state->uapi.plane);
2166 	enum plane_id plane_id = plane->id;
2167 	struct skl_plane_wm *wm = &crtc_state->wm.skl.raw.planes[plane_id];
2168 	const struct drm_framebuffer *fb = plane_state->hw.fb;
2169 	int ret;
2170 
2171 	memset(wm, 0, sizeof(*wm));
2172 
2173 	if (!intel_wm_plane_visible(crtc_state, plane_state))
2174 		return 0;
2175 
2176 	ret = skl_build_plane_wm_single(crtc_state, plane_state,
2177 					plane, 0);
2178 	if (ret)
2179 		return ret;
2180 
2181 	if (fb->format->is_yuv && fb->format->num_planes > 1) {
2182 		ret = skl_build_plane_wm_uv(crtc_state, plane_state,
2183 					    plane);
2184 		if (ret)
2185 			return ret;
2186 	}
2187 
2188 	return 0;
2189 }
2190 
2191 static int icl_build_plane_wm(struct intel_crtc_state *crtc_state,
2192 			      const struct intel_plane_state *plane_state)
2193 {
2194 	struct intel_plane *plane = to_intel_plane(plane_state->uapi.plane);
2195 	struct drm_i915_private *i915 = to_i915(plane->base.dev);
2196 	enum plane_id plane_id = plane->id;
2197 	struct skl_plane_wm *wm = &crtc_state->wm.skl.raw.planes[plane_id];
2198 	int ret;
2199 
2200 	/* Watermarks calculated in master */
2201 	if (plane_state->planar_slave)
2202 		return 0;
2203 
2204 	memset(wm, 0, sizeof(*wm));
2205 
2206 	if (plane_state->planar_linked_plane) {
2207 		const struct drm_framebuffer *fb = plane_state->hw.fb;
2208 
2209 		drm_WARN_ON(&i915->drm,
2210 			    !intel_wm_plane_visible(crtc_state, plane_state));
2211 		drm_WARN_ON(&i915->drm, !fb->format->is_yuv ||
2212 			    fb->format->num_planes == 1);
2213 
2214 		ret = skl_build_plane_wm_single(crtc_state, plane_state,
2215 						plane_state->planar_linked_plane, 0);
2216 		if (ret)
2217 			return ret;
2218 
2219 		ret = skl_build_plane_wm_single(crtc_state, plane_state,
2220 						plane, 1);
2221 		if (ret)
2222 			return ret;
2223 	} else if (intel_wm_plane_visible(crtc_state, plane_state)) {
2224 		ret = skl_build_plane_wm_single(crtc_state, plane_state,
2225 						plane, 0);
2226 		if (ret)
2227 			return ret;
2228 	}
2229 
2230 	return 0;
2231 }
2232 
2233 static bool
2234 skl_is_vblank_too_short(const struct intel_crtc_state *crtc_state,
2235 			int wm0_lines, int latency)
2236 {
2237 	const struct drm_display_mode *adjusted_mode =
2238 		&crtc_state->hw.adjusted_mode;
2239 
2240 	/* FIXME missing scaler and DSC pre-fill time */
2241 	return crtc_state->framestart_delay +
2242 		intel_usecs_to_scanlines(adjusted_mode, latency) +
2243 		wm0_lines >
2244 		adjusted_mode->crtc_vtotal - adjusted_mode->crtc_vblank_start;
2245 }
2246 
2247 static int skl_max_wm0_lines(const struct intel_crtc_state *crtc_state)
2248 {
2249 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
2250 	enum plane_id plane_id;
2251 	int wm0_lines = 0;
2252 
2253 	for_each_plane_id_on_crtc(crtc, plane_id) {
2254 		const struct skl_plane_wm *wm = &crtc_state->wm.skl.optimal.planes[plane_id];
2255 
2256 		/* FIXME what about !skl_wm_has_lines() platforms? */
2257 		wm0_lines = max_t(int, wm0_lines, wm->wm[0].lines);
2258 	}
2259 
2260 	return wm0_lines;
2261 }
2262 
2263 static int skl_max_wm_level_for_vblank(struct intel_crtc_state *crtc_state,
2264 				       int wm0_lines)
2265 {
2266 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
2267 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
2268 	int level;
2269 
2270 	for (level = i915->display.wm.num_levels - 1; level >= 0; level--) {
2271 		int latency;
2272 
2273 		/* FIXME should we care about the latency w/a's? */
2274 		latency = skl_wm_latency(i915, level, NULL);
2275 		if (latency == 0)
2276 			continue;
2277 
2278 		/* FIXME is it correct to use 0 latency for wm0 here? */
2279 		if (level == 0)
2280 			latency = 0;
2281 
2282 		if (!skl_is_vblank_too_short(crtc_state, wm0_lines, latency))
2283 			return level;
2284 	}
2285 
2286 	return -EINVAL;
2287 }
2288 
2289 static int skl_wm_check_vblank(struct intel_crtc_state *crtc_state)
2290 {
2291 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
2292 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
2293 	int wm0_lines, level;
2294 
2295 	if (!crtc_state->hw.active)
2296 		return 0;
2297 
2298 	wm0_lines = skl_max_wm0_lines(crtc_state);
2299 
2300 	level = skl_max_wm_level_for_vblank(crtc_state, wm0_lines);
2301 	if (level < 0)
2302 		return level;
2303 
2304 	/*
2305 	 * PSR needs to toggle LATENCY_REPORTING_REMOVED_PIPE_*
2306 	 * based on whether we're limited by the vblank duration.
2307 	 */
2308 	crtc_state->wm_level_disabled = level < i915->display.wm.num_levels - 1;
2309 
2310 	for (level++; level < i915->display.wm.num_levels; level++) {
2311 		enum plane_id plane_id;
2312 
2313 		for_each_plane_id_on_crtc(crtc, plane_id) {
2314 			struct skl_plane_wm *wm =
2315 				&crtc_state->wm.skl.optimal.planes[plane_id];
2316 
2317 			/*
2318 			 * FIXME just clear enable or flag the entire
2319 			 * thing as bad via min_ddb_alloc=U16_MAX?
2320 			 */
2321 			wm->wm[level].enable = false;
2322 			wm->uv_wm[level].enable = false;
2323 		}
2324 	}
2325 
2326 	if (DISPLAY_VER(i915) >= 12 &&
2327 	    i915->display.sagv.block_time_us &&
2328 	    skl_is_vblank_too_short(crtc_state, wm0_lines,
2329 				    i915->display.sagv.block_time_us)) {
2330 		enum plane_id plane_id;
2331 
2332 		for_each_plane_id_on_crtc(crtc, plane_id) {
2333 			struct skl_plane_wm *wm =
2334 				&crtc_state->wm.skl.optimal.planes[plane_id];
2335 
2336 			wm->sagv.wm0.enable = false;
2337 			wm->sagv.trans_wm.enable = false;
2338 		}
2339 	}
2340 
2341 	return 0;
2342 }
2343 
2344 static int skl_build_pipe_wm(struct intel_atomic_state *state,
2345 			     struct intel_crtc *crtc)
2346 {
2347 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
2348 	struct intel_crtc_state *crtc_state =
2349 		intel_atomic_get_new_crtc_state(state, crtc);
2350 	const struct intel_plane_state *plane_state;
2351 	struct intel_plane *plane;
2352 	int ret, i;
2353 
2354 	for_each_new_intel_plane_in_state(state, plane, plane_state, i) {
2355 		/*
2356 		 * FIXME should perhaps check {old,new}_plane_crtc->hw.crtc
2357 		 * instead but we don't populate that correctly for NV12 Y
2358 		 * planes so for now hack this.
2359 		 */
2360 		if (plane->pipe != crtc->pipe)
2361 			continue;
2362 
2363 		if (DISPLAY_VER(i915) >= 11)
2364 			ret = icl_build_plane_wm(crtc_state, plane_state);
2365 		else
2366 			ret = skl_build_plane_wm(crtc_state, plane_state);
2367 		if (ret)
2368 			return ret;
2369 	}
2370 
2371 	crtc_state->wm.skl.optimal = crtc_state->wm.skl.raw;
2372 
2373 	return skl_wm_check_vblank(crtc_state);
2374 }
2375 
2376 static bool skl_wm_level_equals(const struct skl_wm_level *l1,
2377 				const struct skl_wm_level *l2)
2378 {
2379 	return l1->enable == l2->enable &&
2380 		l1->ignore_lines == l2->ignore_lines &&
2381 		l1->lines == l2->lines &&
2382 		l1->blocks == l2->blocks;
2383 }
2384 
2385 static bool skl_plane_wm_equals(struct drm_i915_private *i915,
2386 				const struct skl_plane_wm *wm1,
2387 				const struct skl_plane_wm *wm2)
2388 {
2389 	int level;
2390 
2391 	for (level = 0; level < i915->display.wm.num_levels; level++) {
2392 		/*
2393 		 * We don't check uv_wm as the hardware doesn't actually
2394 		 * use it. It only gets used for calculating the required
2395 		 * ddb allocation.
2396 		 */
2397 		if (!skl_wm_level_equals(&wm1->wm[level], &wm2->wm[level]))
2398 			return false;
2399 	}
2400 
2401 	return skl_wm_level_equals(&wm1->trans_wm, &wm2->trans_wm) &&
2402 		skl_wm_level_equals(&wm1->sagv.wm0, &wm2->sagv.wm0) &&
2403 		skl_wm_level_equals(&wm1->sagv.trans_wm, &wm2->sagv.trans_wm);
2404 }
2405 
2406 static bool skl_ddb_entries_overlap(const struct skl_ddb_entry *a,
2407 				    const struct skl_ddb_entry *b)
2408 {
2409 	return a->start < b->end && b->start < a->end;
2410 }
2411 
2412 static void skl_ddb_entry_union(struct skl_ddb_entry *a,
2413 				const struct skl_ddb_entry *b)
2414 {
2415 	if (a->end && b->end) {
2416 		a->start = min(a->start, b->start);
2417 		a->end = max(a->end, b->end);
2418 	} else if (b->end) {
2419 		a->start = b->start;
2420 		a->end = b->end;
2421 	}
2422 }
2423 
2424 bool skl_ddb_allocation_overlaps(const struct skl_ddb_entry *ddb,
2425 				 const struct skl_ddb_entry *entries,
2426 				 int num_entries, int ignore_idx)
2427 {
2428 	int i;
2429 
2430 	for (i = 0; i < num_entries; i++) {
2431 		if (i != ignore_idx &&
2432 		    skl_ddb_entries_overlap(ddb, &entries[i]))
2433 			return true;
2434 	}
2435 
2436 	return false;
2437 }
2438 
2439 static int
2440 skl_ddb_add_affected_planes(struct intel_atomic_state *state,
2441 			    struct intel_crtc *crtc)
2442 {
2443 	struct drm_i915_private *i915 = to_i915(state->base.dev);
2444 	const struct intel_crtc_state *old_crtc_state =
2445 		intel_atomic_get_old_crtc_state(state, crtc);
2446 	struct intel_crtc_state *new_crtc_state =
2447 		intel_atomic_get_new_crtc_state(state, crtc);
2448 	struct intel_plane *plane;
2449 
2450 	for_each_intel_plane_on_crtc(&i915->drm, crtc, plane) {
2451 		struct intel_plane_state *plane_state;
2452 		enum plane_id plane_id = plane->id;
2453 
2454 		if (skl_ddb_entry_equal(&old_crtc_state->wm.skl.plane_ddb[plane_id],
2455 					&new_crtc_state->wm.skl.plane_ddb[plane_id]) &&
2456 		    skl_ddb_entry_equal(&old_crtc_state->wm.skl.plane_ddb_y[plane_id],
2457 					&new_crtc_state->wm.skl.plane_ddb_y[plane_id]))
2458 			continue;
2459 
2460 		if (new_crtc_state->do_async_flip) {
2461 			drm_dbg_kms(&i915->drm, "[PLANE:%d:%s] Can't change DDB during async flip\n",
2462 				    plane->base.base.id, plane->base.name);
2463 			return -EINVAL;
2464 		}
2465 
2466 		plane_state = intel_atomic_get_plane_state(state, plane);
2467 		if (IS_ERR(plane_state))
2468 			return PTR_ERR(plane_state);
2469 
2470 		new_crtc_state->update_planes |= BIT(plane_id);
2471 		new_crtc_state->async_flip_planes = 0;
2472 		new_crtc_state->do_async_flip = false;
2473 	}
2474 
2475 	return 0;
2476 }
2477 
2478 static u8 intel_dbuf_enabled_slices(const struct intel_dbuf_state *dbuf_state)
2479 {
2480 	struct drm_i915_private *i915 = to_i915(dbuf_state->base.state->base.dev);
2481 	u8 enabled_slices;
2482 	enum pipe pipe;
2483 
2484 	/*
2485 	 * FIXME: For now we always enable slice S1 as per
2486 	 * the Bspec display initialization sequence.
2487 	 */
2488 	enabled_slices = BIT(DBUF_S1);
2489 
2490 	for_each_pipe(i915, pipe)
2491 		enabled_slices |= dbuf_state->slices[pipe];
2492 
2493 	return enabled_slices;
2494 }
2495 
2496 static int
2497 skl_compute_ddb(struct intel_atomic_state *state)
2498 {
2499 	struct drm_i915_private *i915 = to_i915(state->base.dev);
2500 	const struct intel_dbuf_state *old_dbuf_state;
2501 	struct intel_dbuf_state *new_dbuf_state = NULL;
2502 	struct intel_crtc_state *new_crtc_state;
2503 	struct intel_crtc *crtc;
2504 	int ret, i;
2505 
2506 	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
2507 		new_dbuf_state = intel_atomic_get_dbuf_state(state);
2508 		if (IS_ERR(new_dbuf_state))
2509 			return PTR_ERR(new_dbuf_state);
2510 
2511 		old_dbuf_state = intel_atomic_get_old_dbuf_state(state);
2512 		break;
2513 	}
2514 
2515 	if (!new_dbuf_state)
2516 		return 0;
2517 
2518 	new_dbuf_state->active_pipes =
2519 		intel_calc_active_pipes(state, old_dbuf_state->active_pipes);
2520 
2521 	if (old_dbuf_state->active_pipes != new_dbuf_state->active_pipes) {
2522 		ret = intel_atomic_lock_global_state(&new_dbuf_state->base);
2523 		if (ret)
2524 			return ret;
2525 	}
2526 
2527 	if (HAS_MBUS_JOINING(i915)) {
2528 		new_dbuf_state->joined_mbus =
2529 			adlp_check_mbus_joined(new_dbuf_state->active_pipes);
2530 
2531 		if (old_dbuf_state->joined_mbus != new_dbuf_state->joined_mbus) {
2532 			ret = intel_cdclk_state_set_joined_mbus(state, new_dbuf_state->joined_mbus);
2533 			if (ret)
2534 				return ret;
2535 		}
2536 	}
2537 
2538 	for_each_intel_crtc(&i915->drm, crtc) {
2539 		enum pipe pipe = crtc->pipe;
2540 
2541 		new_dbuf_state->slices[pipe] =
2542 			skl_compute_dbuf_slices(crtc, new_dbuf_state->active_pipes,
2543 						new_dbuf_state->joined_mbus);
2544 
2545 		if (old_dbuf_state->slices[pipe] == new_dbuf_state->slices[pipe])
2546 			continue;
2547 
2548 		ret = intel_atomic_lock_global_state(&new_dbuf_state->base);
2549 		if (ret)
2550 			return ret;
2551 	}
2552 
2553 	new_dbuf_state->enabled_slices = intel_dbuf_enabled_slices(new_dbuf_state);
2554 
2555 	if (old_dbuf_state->enabled_slices != new_dbuf_state->enabled_slices ||
2556 	    old_dbuf_state->joined_mbus != new_dbuf_state->joined_mbus) {
2557 		ret = intel_atomic_serialize_global_state(&new_dbuf_state->base);
2558 		if (ret)
2559 			return ret;
2560 
2561 		drm_dbg_kms(&i915->drm,
2562 			    "Enabled dbuf slices 0x%x -> 0x%x (total dbuf slices 0x%x), mbus joined? %s->%s\n",
2563 			    old_dbuf_state->enabled_slices,
2564 			    new_dbuf_state->enabled_slices,
2565 			    DISPLAY_INFO(i915)->dbuf.slice_mask,
2566 			    str_yes_no(old_dbuf_state->joined_mbus),
2567 			    str_yes_no(new_dbuf_state->joined_mbus));
2568 	}
2569 
2570 	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
2571 		enum pipe pipe = crtc->pipe;
2572 
2573 		new_dbuf_state->weight[pipe] = intel_crtc_ddb_weight(new_crtc_state);
2574 
2575 		if (old_dbuf_state->weight[pipe] == new_dbuf_state->weight[pipe])
2576 			continue;
2577 
2578 		ret = intel_atomic_lock_global_state(&new_dbuf_state->base);
2579 		if (ret)
2580 			return ret;
2581 	}
2582 
2583 	for_each_intel_crtc(&i915->drm, crtc) {
2584 		ret = skl_crtc_allocate_ddb(state, crtc);
2585 		if (ret)
2586 			return ret;
2587 	}
2588 
2589 	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
2590 		ret = skl_crtc_allocate_plane_ddb(state, crtc);
2591 		if (ret)
2592 			return ret;
2593 
2594 		ret = skl_ddb_add_affected_planes(state, crtc);
2595 		if (ret)
2596 			return ret;
2597 	}
2598 
2599 	return 0;
2600 }
2601 
2602 static char enast(bool enable)
2603 {
2604 	return enable ? '*' : ' ';
2605 }
2606 
2607 static void
2608 skl_print_wm_changes(struct intel_atomic_state *state)
2609 {
2610 	struct drm_i915_private *i915 = to_i915(state->base.dev);
2611 	const struct intel_crtc_state *old_crtc_state;
2612 	const struct intel_crtc_state *new_crtc_state;
2613 	struct intel_plane *plane;
2614 	struct intel_crtc *crtc;
2615 	int i;
2616 
2617 	if (!drm_debug_enabled(DRM_UT_KMS))
2618 		return;
2619 
2620 	for_each_oldnew_intel_crtc_in_state(state, crtc, old_crtc_state,
2621 					    new_crtc_state, i) {
2622 		const struct skl_pipe_wm *old_pipe_wm, *new_pipe_wm;
2623 
2624 		old_pipe_wm = &old_crtc_state->wm.skl.optimal;
2625 		new_pipe_wm = &new_crtc_state->wm.skl.optimal;
2626 
2627 		for_each_intel_plane_on_crtc(&i915->drm, crtc, plane) {
2628 			enum plane_id plane_id = plane->id;
2629 			const struct skl_ddb_entry *old, *new;
2630 
2631 			old = &old_crtc_state->wm.skl.plane_ddb[plane_id];
2632 			new = &new_crtc_state->wm.skl.plane_ddb[plane_id];
2633 
2634 			if (skl_ddb_entry_equal(old, new))
2635 				continue;
2636 
2637 			drm_dbg_kms(&i915->drm,
2638 				    "[PLANE:%d:%s] ddb (%4d - %4d) -> (%4d - %4d), size %4d -> %4d\n",
2639 				    plane->base.base.id, plane->base.name,
2640 				    old->start, old->end, new->start, new->end,
2641 				    skl_ddb_entry_size(old), skl_ddb_entry_size(new));
2642 		}
2643 
2644 		for_each_intel_plane_on_crtc(&i915->drm, crtc, plane) {
2645 			enum plane_id plane_id = plane->id;
2646 			const struct skl_plane_wm *old_wm, *new_wm;
2647 
2648 			old_wm = &old_pipe_wm->planes[plane_id];
2649 			new_wm = &new_pipe_wm->planes[plane_id];
2650 
2651 			if (skl_plane_wm_equals(i915, old_wm, new_wm))
2652 				continue;
2653 
2654 			drm_dbg_kms(&i915->drm,
2655 				    "[PLANE:%d:%s]   level %cwm0,%cwm1,%cwm2,%cwm3,%cwm4,%cwm5,%cwm6,%cwm7,%ctwm,%cswm,%cstwm"
2656 				    " -> %cwm0,%cwm1,%cwm2,%cwm3,%cwm4,%cwm5,%cwm6,%cwm7,%ctwm,%cswm,%cstwm\n",
2657 				    plane->base.base.id, plane->base.name,
2658 				    enast(old_wm->wm[0].enable), enast(old_wm->wm[1].enable),
2659 				    enast(old_wm->wm[2].enable), enast(old_wm->wm[3].enable),
2660 				    enast(old_wm->wm[4].enable), enast(old_wm->wm[5].enable),
2661 				    enast(old_wm->wm[6].enable), enast(old_wm->wm[7].enable),
2662 				    enast(old_wm->trans_wm.enable),
2663 				    enast(old_wm->sagv.wm0.enable),
2664 				    enast(old_wm->sagv.trans_wm.enable),
2665 				    enast(new_wm->wm[0].enable), enast(new_wm->wm[1].enable),
2666 				    enast(new_wm->wm[2].enable), enast(new_wm->wm[3].enable),
2667 				    enast(new_wm->wm[4].enable), enast(new_wm->wm[5].enable),
2668 				    enast(new_wm->wm[6].enable), enast(new_wm->wm[7].enable),
2669 				    enast(new_wm->trans_wm.enable),
2670 				    enast(new_wm->sagv.wm0.enable),
2671 				    enast(new_wm->sagv.trans_wm.enable));
2672 
2673 			drm_dbg_kms(&i915->drm,
2674 				    "[PLANE:%d:%s]   lines %c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%4d"
2675 				      " -> %c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%4d\n",
2676 				    plane->base.base.id, plane->base.name,
2677 				    enast(old_wm->wm[0].ignore_lines), old_wm->wm[0].lines,
2678 				    enast(old_wm->wm[1].ignore_lines), old_wm->wm[1].lines,
2679 				    enast(old_wm->wm[2].ignore_lines), old_wm->wm[2].lines,
2680 				    enast(old_wm->wm[3].ignore_lines), old_wm->wm[3].lines,
2681 				    enast(old_wm->wm[4].ignore_lines), old_wm->wm[4].lines,
2682 				    enast(old_wm->wm[5].ignore_lines), old_wm->wm[5].lines,
2683 				    enast(old_wm->wm[6].ignore_lines), old_wm->wm[6].lines,
2684 				    enast(old_wm->wm[7].ignore_lines), old_wm->wm[7].lines,
2685 				    enast(old_wm->trans_wm.ignore_lines), old_wm->trans_wm.lines,
2686 				    enast(old_wm->sagv.wm0.ignore_lines), old_wm->sagv.wm0.lines,
2687 				    enast(old_wm->sagv.trans_wm.ignore_lines), old_wm->sagv.trans_wm.lines,
2688 				    enast(new_wm->wm[0].ignore_lines), new_wm->wm[0].lines,
2689 				    enast(new_wm->wm[1].ignore_lines), new_wm->wm[1].lines,
2690 				    enast(new_wm->wm[2].ignore_lines), new_wm->wm[2].lines,
2691 				    enast(new_wm->wm[3].ignore_lines), new_wm->wm[3].lines,
2692 				    enast(new_wm->wm[4].ignore_lines), new_wm->wm[4].lines,
2693 				    enast(new_wm->wm[5].ignore_lines), new_wm->wm[5].lines,
2694 				    enast(new_wm->wm[6].ignore_lines), new_wm->wm[6].lines,
2695 				    enast(new_wm->wm[7].ignore_lines), new_wm->wm[7].lines,
2696 				    enast(new_wm->trans_wm.ignore_lines), new_wm->trans_wm.lines,
2697 				    enast(new_wm->sagv.wm0.ignore_lines), new_wm->sagv.wm0.lines,
2698 				    enast(new_wm->sagv.trans_wm.ignore_lines), new_wm->sagv.trans_wm.lines);
2699 
2700 			drm_dbg_kms(&i915->drm,
2701 				    "[PLANE:%d:%s]  blocks %4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%5d"
2702 				    " -> %4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%5d\n",
2703 				    plane->base.base.id, plane->base.name,
2704 				    old_wm->wm[0].blocks, old_wm->wm[1].blocks,
2705 				    old_wm->wm[2].blocks, old_wm->wm[3].blocks,
2706 				    old_wm->wm[4].blocks, old_wm->wm[5].blocks,
2707 				    old_wm->wm[6].blocks, old_wm->wm[7].blocks,
2708 				    old_wm->trans_wm.blocks,
2709 				    old_wm->sagv.wm0.blocks,
2710 				    old_wm->sagv.trans_wm.blocks,
2711 				    new_wm->wm[0].blocks, new_wm->wm[1].blocks,
2712 				    new_wm->wm[2].blocks, new_wm->wm[3].blocks,
2713 				    new_wm->wm[4].blocks, new_wm->wm[5].blocks,
2714 				    new_wm->wm[6].blocks, new_wm->wm[7].blocks,
2715 				    new_wm->trans_wm.blocks,
2716 				    new_wm->sagv.wm0.blocks,
2717 				    new_wm->sagv.trans_wm.blocks);
2718 
2719 			drm_dbg_kms(&i915->drm,
2720 				    "[PLANE:%d:%s] min_ddb %4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%5d"
2721 				    " -> %4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%5d\n",
2722 				    plane->base.base.id, plane->base.name,
2723 				    old_wm->wm[0].min_ddb_alloc, old_wm->wm[1].min_ddb_alloc,
2724 				    old_wm->wm[2].min_ddb_alloc, old_wm->wm[3].min_ddb_alloc,
2725 				    old_wm->wm[4].min_ddb_alloc, old_wm->wm[5].min_ddb_alloc,
2726 				    old_wm->wm[6].min_ddb_alloc, old_wm->wm[7].min_ddb_alloc,
2727 				    old_wm->trans_wm.min_ddb_alloc,
2728 				    old_wm->sagv.wm0.min_ddb_alloc,
2729 				    old_wm->sagv.trans_wm.min_ddb_alloc,
2730 				    new_wm->wm[0].min_ddb_alloc, new_wm->wm[1].min_ddb_alloc,
2731 				    new_wm->wm[2].min_ddb_alloc, new_wm->wm[3].min_ddb_alloc,
2732 				    new_wm->wm[4].min_ddb_alloc, new_wm->wm[5].min_ddb_alloc,
2733 				    new_wm->wm[6].min_ddb_alloc, new_wm->wm[7].min_ddb_alloc,
2734 				    new_wm->trans_wm.min_ddb_alloc,
2735 				    new_wm->sagv.wm0.min_ddb_alloc,
2736 				    new_wm->sagv.trans_wm.min_ddb_alloc);
2737 		}
2738 	}
2739 }
2740 
2741 static bool skl_plane_selected_wm_equals(struct intel_plane *plane,
2742 					 const struct skl_pipe_wm *old_pipe_wm,
2743 					 const struct skl_pipe_wm *new_pipe_wm)
2744 {
2745 	struct drm_i915_private *i915 = to_i915(plane->base.dev);
2746 	int level;
2747 
2748 	for (level = 0; level < i915->display.wm.num_levels; level++) {
2749 		/*
2750 		 * We don't check uv_wm as the hardware doesn't actually
2751 		 * use it. It only gets used for calculating the required
2752 		 * ddb allocation.
2753 		 */
2754 		if (!skl_wm_level_equals(skl_plane_wm_level(old_pipe_wm, plane->id, level),
2755 					 skl_plane_wm_level(new_pipe_wm, plane->id, level)))
2756 			return false;
2757 	}
2758 
2759 	if (HAS_HW_SAGV_WM(i915)) {
2760 		const struct skl_plane_wm *old_wm = &old_pipe_wm->planes[plane->id];
2761 		const struct skl_plane_wm *new_wm = &new_pipe_wm->planes[plane->id];
2762 
2763 		if (!skl_wm_level_equals(&old_wm->sagv.wm0, &new_wm->sagv.wm0) ||
2764 		    !skl_wm_level_equals(&old_wm->sagv.trans_wm, &new_wm->sagv.trans_wm))
2765 			return false;
2766 	}
2767 
2768 	return skl_wm_level_equals(skl_plane_trans_wm(old_pipe_wm, plane->id),
2769 				   skl_plane_trans_wm(new_pipe_wm, plane->id));
2770 }
2771 
2772 /*
2773  * To make sure the cursor watermark registers are always consistent
2774  * with our computed state the following scenario needs special
2775  * treatment:
2776  *
2777  * 1. enable cursor
2778  * 2. move cursor entirely offscreen
2779  * 3. disable cursor
2780  *
2781  * Step 2. does call .disable_plane() but does not zero the watermarks
2782  * (since we consider an offscreen cursor still active for the purposes
2783  * of watermarks). Step 3. would not normally call .disable_plane()
2784  * because the actual plane visibility isn't changing, and we don't
2785  * deallocate the cursor ddb until the pipe gets disabled. So we must
2786  * force step 3. to call .disable_plane() to update the watermark
2787  * registers properly.
2788  *
2789  * Other planes do not suffer from this issues as their watermarks are
2790  * calculated based on the actual plane visibility. The only time this
2791  * can trigger for the other planes is during the initial readout as the
2792  * default value of the watermarks registers is not zero.
2793  */
2794 static int skl_wm_add_affected_planes(struct intel_atomic_state *state,
2795 				      struct intel_crtc *crtc)
2796 {
2797 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
2798 	const struct intel_crtc_state *old_crtc_state =
2799 		intel_atomic_get_old_crtc_state(state, crtc);
2800 	struct intel_crtc_state *new_crtc_state =
2801 		intel_atomic_get_new_crtc_state(state, crtc);
2802 	struct intel_plane *plane;
2803 
2804 	for_each_intel_plane_on_crtc(&i915->drm, crtc, plane) {
2805 		struct intel_plane_state *plane_state;
2806 		enum plane_id plane_id = plane->id;
2807 
2808 		/*
2809 		 * Force a full wm update for every plane on modeset.
2810 		 * Required because the reset value of the wm registers
2811 		 * is non-zero, whereas we want all disabled planes to
2812 		 * have zero watermarks. So if we turn off the relevant
2813 		 * power well the hardware state will go out of sync
2814 		 * with the software state.
2815 		 */
2816 		if (!intel_crtc_needs_modeset(new_crtc_state) &&
2817 		    skl_plane_selected_wm_equals(plane,
2818 						 &old_crtc_state->wm.skl.optimal,
2819 						 &new_crtc_state->wm.skl.optimal))
2820 			continue;
2821 
2822 		if (new_crtc_state->do_async_flip) {
2823 			drm_dbg_kms(&i915->drm, "[PLANE:%d:%s] Can't change watermarks during async flip\n",
2824 				    plane->base.base.id, plane->base.name);
2825 			return -EINVAL;
2826 		}
2827 
2828 		plane_state = intel_atomic_get_plane_state(state, plane);
2829 		if (IS_ERR(plane_state))
2830 			return PTR_ERR(plane_state);
2831 
2832 		new_crtc_state->update_planes |= BIT(plane_id);
2833 		new_crtc_state->async_flip_planes = 0;
2834 		new_crtc_state->do_async_flip = false;
2835 	}
2836 
2837 	return 0;
2838 }
2839 
2840 /*
2841  * If Fixed Refresh Rate or For VRR case Vmin = Vmax = Flipline:
2842  * Program DEEP PKG_C_LATENCY Pkg C with highest valid latency from
2843  * watermark level1 and up and above. If watermark level 1 is
2844  * invalid program it with all 1's.
2845  * Program PKG_C_LATENCY Added Wake Time = DSB execution time
2846  * If Variable Refresh Rate where Vmin != Vmax != Flipline:
2847  * Program DEEP PKG_C_LATENCY Pkg C with all 1's.
2848  * Program PKG_C_LATENCY Added Wake Time = 0
2849  */
2850 static void
2851 skl_program_dpkgc_latency(struct drm_i915_private *i915, bool enable_dpkgc)
2852 {
2853 	u32 max_latency = 0;
2854 	u32 clear = 0, val = 0;
2855 	u32 added_wake_time = 0;
2856 
2857 	if (DISPLAY_VER(i915) < 20)
2858 		return;
2859 
2860 	if (enable_dpkgc) {
2861 		max_latency = skl_watermark_max_latency(i915, 1);
2862 		if (max_latency == 0)
2863 			max_latency = LNL_PKG_C_LATENCY_MASK;
2864 		added_wake_time = DSB_EXE_TIME +
2865 			i915->display.sagv.block_time_us;
2866 	} else {
2867 		max_latency = LNL_PKG_C_LATENCY_MASK;
2868 		added_wake_time = 0;
2869 	}
2870 
2871 	clear |= LNL_ADDED_WAKE_TIME_MASK | LNL_PKG_C_LATENCY_MASK;
2872 	val |= REG_FIELD_PREP(LNL_PKG_C_LATENCY_MASK, max_latency);
2873 	val |= REG_FIELD_PREP(LNL_ADDED_WAKE_TIME_MASK, added_wake_time);
2874 
2875 	intel_uncore_rmw(&i915->uncore, LNL_PKG_C_LATENCY, clear, val);
2876 }
2877 
2878 static int
2879 skl_compute_wm(struct intel_atomic_state *state)
2880 {
2881 	struct intel_crtc *crtc;
2882 	struct intel_crtc_state __maybe_unused *new_crtc_state;
2883 	int ret, i;
2884 	bool enable_dpkgc = false;
2885 
2886 	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
2887 		ret = skl_build_pipe_wm(state, crtc);
2888 		if (ret)
2889 			return ret;
2890 	}
2891 
2892 	ret = skl_compute_ddb(state);
2893 	if (ret)
2894 		return ret;
2895 
2896 	ret = intel_compute_sagv_mask(state);
2897 	if (ret)
2898 		return ret;
2899 
2900 	/*
2901 	 * skl_compute_ddb() will have adjusted the final watermarks
2902 	 * based on how much ddb is available. Now we can actually
2903 	 * check if the final watermarks changed.
2904 	 */
2905 	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
2906 		ret = skl_wm_add_affected_planes(state, crtc);
2907 		if (ret)
2908 			return ret;
2909 
2910 		if ((new_crtc_state->vrr.vmin == new_crtc_state->vrr.vmax &&
2911 		     new_crtc_state->vrr.vmin == new_crtc_state->vrr.flipline) ||
2912 		    !new_crtc_state->vrr.enable)
2913 			enable_dpkgc = true;
2914 	}
2915 
2916 	skl_program_dpkgc_latency(to_i915(state->base.dev), enable_dpkgc);
2917 
2918 	skl_print_wm_changes(state);
2919 
2920 	return 0;
2921 }
2922 
2923 static void skl_wm_level_from_reg_val(u32 val, struct skl_wm_level *level)
2924 {
2925 	level->enable = val & PLANE_WM_EN;
2926 	level->ignore_lines = val & PLANE_WM_IGNORE_LINES;
2927 	level->blocks = REG_FIELD_GET(PLANE_WM_BLOCKS_MASK, val);
2928 	level->lines = REG_FIELD_GET(PLANE_WM_LINES_MASK, val);
2929 }
2930 
2931 static void skl_pipe_wm_get_hw_state(struct intel_crtc *crtc,
2932 				     struct skl_pipe_wm *out)
2933 {
2934 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
2935 	enum pipe pipe = crtc->pipe;
2936 	enum plane_id plane_id;
2937 	int level;
2938 	u32 val;
2939 
2940 	for_each_plane_id_on_crtc(crtc, plane_id) {
2941 		struct skl_plane_wm *wm = &out->planes[plane_id];
2942 
2943 		for (level = 0; level < i915->display.wm.num_levels; level++) {
2944 			if (plane_id != PLANE_CURSOR)
2945 				val = intel_de_read(i915, PLANE_WM(pipe, plane_id, level));
2946 			else
2947 				val = intel_de_read(i915, CUR_WM(pipe, level));
2948 
2949 			skl_wm_level_from_reg_val(val, &wm->wm[level]);
2950 		}
2951 
2952 		if (plane_id != PLANE_CURSOR)
2953 			val = intel_de_read(i915, PLANE_WM_TRANS(pipe, plane_id));
2954 		else
2955 			val = intel_de_read(i915, CUR_WM_TRANS(pipe));
2956 
2957 		skl_wm_level_from_reg_val(val, &wm->trans_wm);
2958 
2959 		if (HAS_HW_SAGV_WM(i915)) {
2960 			if (plane_id != PLANE_CURSOR)
2961 				val = intel_de_read(i915, PLANE_WM_SAGV(pipe, plane_id));
2962 			else
2963 				val = intel_de_read(i915, CUR_WM_SAGV(pipe));
2964 
2965 			skl_wm_level_from_reg_val(val, &wm->sagv.wm0);
2966 
2967 			if (plane_id != PLANE_CURSOR)
2968 				val = intel_de_read(i915, PLANE_WM_SAGV_TRANS(pipe, plane_id));
2969 			else
2970 				val = intel_de_read(i915, CUR_WM_SAGV_TRANS(pipe));
2971 
2972 			skl_wm_level_from_reg_val(val, &wm->sagv.trans_wm);
2973 		} else if (DISPLAY_VER(i915) >= 12) {
2974 			wm->sagv.wm0 = wm->wm[0];
2975 			wm->sagv.trans_wm = wm->trans_wm;
2976 		}
2977 	}
2978 }
2979 
2980 static void skl_wm_get_hw_state(struct drm_i915_private *i915)
2981 {
2982 	struct intel_display *display = &i915->display;
2983 	struct intel_dbuf_state *dbuf_state =
2984 		to_intel_dbuf_state(i915->display.dbuf.obj.state);
2985 	struct intel_crtc *crtc;
2986 
2987 	if (HAS_MBUS_JOINING(i915))
2988 		dbuf_state->joined_mbus = intel_de_read(i915, MBUS_CTL) & MBUS_JOIN;
2989 
2990 	dbuf_state->mdclk_cdclk_ratio = intel_mdclk_cdclk_ratio(display, &display->cdclk.hw);
2991 
2992 	for_each_intel_crtc(&i915->drm, crtc) {
2993 		struct intel_crtc_state *crtc_state =
2994 			to_intel_crtc_state(crtc->base.state);
2995 		enum pipe pipe = crtc->pipe;
2996 		unsigned int mbus_offset;
2997 		enum plane_id plane_id;
2998 		u8 slices;
2999 
3000 		memset(&crtc_state->wm.skl.optimal, 0,
3001 		       sizeof(crtc_state->wm.skl.optimal));
3002 		if (crtc_state->hw.active)
3003 			skl_pipe_wm_get_hw_state(crtc, &crtc_state->wm.skl.optimal);
3004 		crtc_state->wm.skl.raw = crtc_state->wm.skl.optimal;
3005 
3006 		memset(&dbuf_state->ddb[pipe], 0, sizeof(dbuf_state->ddb[pipe]));
3007 
3008 		for_each_plane_id_on_crtc(crtc, plane_id) {
3009 			struct skl_ddb_entry *ddb =
3010 				&crtc_state->wm.skl.plane_ddb[plane_id];
3011 			struct skl_ddb_entry *ddb_y =
3012 				&crtc_state->wm.skl.plane_ddb_y[plane_id];
3013 
3014 			if (!crtc_state->hw.active)
3015 				continue;
3016 
3017 			skl_ddb_get_hw_plane_state(i915, crtc->pipe,
3018 						   plane_id, ddb, ddb_y);
3019 
3020 			skl_ddb_entry_union(&dbuf_state->ddb[pipe], ddb);
3021 			skl_ddb_entry_union(&dbuf_state->ddb[pipe], ddb_y);
3022 		}
3023 
3024 		dbuf_state->weight[pipe] = intel_crtc_ddb_weight(crtc_state);
3025 
3026 		/*
3027 		 * Used for checking overlaps, so we need absolute
3028 		 * offsets instead of MBUS relative offsets.
3029 		 */
3030 		slices = skl_compute_dbuf_slices(crtc, dbuf_state->active_pipes,
3031 						 dbuf_state->joined_mbus);
3032 		mbus_offset = mbus_ddb_offset(i915, slices);
3033 		crtc_state->wm.skl.ddb.start = mbus_offset + dbuf_state->ddb[pipe].start;
3034 		crtc_state->wm.skl.ddb.end = mbus_offset + dbuf_state->ddb[pipe].end;
3035 
3036 		/* The slices actually used by the planes on the pipe */
3037 		dbuf_state->slices[pipe] =
3038 			skl_ddb_dbuf_slice_mask(i915, &crtc_state->wm.skl.ddb);
3039 
3040 		drm_dbg_kms(&i915->drm,
3041 			    "[CRTC:%d:%s] dbuf slices 0x%x, ddb (%d - %d), active pipes 0x%x, mbus joined: %s\n",
3042 			    crtc->base.base.id, crtc->base.name,
3043 			    dbuf_state->slices[pipe], dbuf_state->ddb[pipe].start,
3044 			    dbuf_state->ddb[pipe].end, dbuf_state->active_pipes,
3045 			    str_yes_no(dbuf_state->joined_mbus));
3046 	}
3047 
3048 	dbuf_state->enabled_slices = i915->display.dbuf.enabled_slices;
3049 }
3050 
3051 static bool skl_dbuf_is_misconfigured(struct drm_i915_private *i915)
3052 {
3053 	const struct intel_dbuf_state *dbuf_state =
3054 		to_intel_dbuf_state(i915->display.dbuf.obj.state);
3055 	struct skl_ddb_entry entries[I915_MAX_PIPES] = {};
3056 	struct intel_crtc *crtc;
3057 
3058 	for_each_intel_crtc(&i915->drm, crtc) {
3059 		const struct intel_crtc_state *crtc_state =
3060 			to_intel_crtc_state(crtc->base.state);
3061 
3062 		entries[crtc->pipe] = crtc_state->wm.skl.ddb;
3063 	}
3064 
3065 	for_each_intel_crtc(&i915->drm, crtc) {
3066 		const struct intel_crtc_state *crtc_state =
3067 			to_intel_crtc_state(crtc->base.state);
3068 		u8 slices;
3069 
3070 		slices = skl_compute_dbuf_slices(crtc, dbuf_state->active_pipes,
3071 						 dbuf_state->joined_mbus);
3072 		if (dbuf_state->slices[crtc->pipe] & ~slices)
3073 			return true;
3074 
3075 		if (skl_ddb_allocation_overlaps(&crtc_state->wm.skl.ddb, entries,
3076 						I915_MAX_PIPES, crtc->pipe))
3077 			return true;
3078 	}
3079 
3080 	return false;
3081 }
3082 
3083 static void skl_wm_sanitize(struct drm_i915_private *i915)
3084 {
3085 	struct intel_crtc *crtc;
3086 
3087 	/*
3088 	 * On TGL/RKL (at least) the BIOS likes to assign the planes
3089 	 * to the wrong DBUF slices. This will cause an infinite loop
3090 	 * in skl_commit_modeset_enables() as it can't find a way to
3091 	 * transition between the old bogus DBUF layout to the new
3092 	 * proper DBUF layout without DBUF allocation overlaps between
3093 	 * the planes (which cannot be allowed or else the hardware
3094 	 * may hang). If we detect a bogus DBUF layout just turn off
3095 	 * all the planes so that skl_commit_modeset_enables() can
3096 	 * simply ignore them.
3097 	 */
3098 	if (!skl_dbuf_is_misconfigured(i915))
3099 		return;
3100 
3101 	drm_dbg_kms(&i915->drm, "BIOS has misprogrammed the DBUF, disabling all planes\n");
3102 
3103 	for_each_intel_crtc(&i915->drm, crtc) {
3104 		struct intel_plane *plane = to_intel_plane(crtc->base.primary);
3105 		const struct intel_plane_state *plane_state =
3106 			to_intel_plane_state(plane->base.state);
3107 		struct intel_crtc_state *crtc_state =
3108 			to_intel_crtc_state(crtc->base.state);
3109 
3110 		if (plane_state->uapi.visible)
3111 			intel_plane_disable_noatomic(crtc, plane);
3112 
3113 		drm_WARN_ON(&i915->drm, crtc_state->active_planes != 0);
3114 
3115 		memset(&crtc_state->wm.skl.ddb, 0, sizeof(crtc_state->wm.skl.ddb));
3116 	}
3117 }
3118 
3119 static void skl_wm_get_hw_state_and_sanitize(struct drm_i915_private *i915)
3120 {
3121 	skl_wm_get_hw_state(i915);
3122 	skl_wm_sanitize(i915);
3123 }
3124 
3125 void intel_wm_state_verify(struct intel_atomic_state *state,
3126 			   struct intel_crtc *crtc)
3127 {
3128 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3129 	const struct intel_crtc_state *new_crtc_state =
3130 		intel_atomic_get_new_crtc_state(state, crtc);
3131 	struct skl_hw_state {
3132 		struct skl_ddb_entry ddb[I915_MAX_PLANES];
3133 		struct skl_ddb_entry ddb_y[I915_MAX_PLANES];
3134 		struct skl_pipe_wm wm;
3135 	} *hw;
3136 	const struct skl_pipe_wm *sw_wm = &new_crtc_state->wm.skl.optimal;
3137 	struct intel_plane *plane;
3138 	u8 hw_enabled_slices;
3139 	int level;
3140 
3141 	if (DISPLAY_VER(i915) < 9 || !new_crtc_state->hw.active)
3142 		return;
3143 
3144 	hw = kzalloc(sizeof(*hw), GFP_KERNEL);
3145 	if (!hw)
3146 		return;
3147 
3148 	skl_pipe_wm_get_hw_state(crtc, &hw->wm);
3149 
3150 	skl_pipe_ddb_get_hw_state(crtc, hw->ddb, hw->ddb_y);
3151 
3152 	hw_enabled_slices = intel_enabled_dbuf_slices_mask(i915);
3153 
3154 	if (DISPLAY_VER(i915) >= 11 &&
3155 	    hw_enabled_slices != i915->display.dbuf.enabled_slices)
3156 		drm_err(&i915->drm,
3157 			"mismatch in DBUF Slices (expected 0x%x, got 0x%x)\n",
3158 			i915->display.dbuf.enabled_slices,
3159 			hw_enabled_slices);
3160 
3161 	for_each_intel_plane_on_crtc(&i915->drm, crtc, plane) {
3162 		const struct skl_ddb_entry *hw_ddb_entry, *sw_ddb_entry;
3163 		const struct skl_wm_level *hw_wm_level, *sw_wm_level;
3164 
3165 		/* Watermarks */
3166 		for (level = 0; level < i915->display.wm.num_levels; level++) {
3167 			hw_wm_level = &hw->wm.planes[plane->id].wm[level];
3168 			sw_wm_level = skl_plane_wm_level(sw_wm, plane->id, level);
3169 
3170 			if (skl_wm_level_equals(hw_wm_level, sw_wm_level))
3171 				continue;
3172 
3173 			drm_err(&i915->drm,
3174 				"[PLANE:%d:%s] mismatch in WM%d (expected e=%d b=%u l=%u, got e=%d b=%u l=%u)\n",
3175 				plane->base.base.id, plane->base.name, level,
3176 				sw_wm_level->enable,
3177 				sw_wm_level->blocks,
3178 				sw_wm_level->lines,
3179 				hw_wm_level->enable,
3180 				hw_wm_level->blocks,
3181 				hw_wm_level->lines);
3182 		}
3183 
3184 		hw_wm_level = &hw->wm.planes[plane->id].trans_wm;
3185 		sw_wm_level = skl_plane_trans_wm(sw_wm, plane->id);
3186 
3187 		if (!skl_wm_level_equals(hw_wm_level, sw_wm_level)) {
3188 			drm_err(&i915->drm,
3189 				"[PLANE:%d:%s] mismatch in trans WM (expected e=%d b=%u l=%u, got e=%d b=%u l=%u)\n",
3190 				plane->base.base.id, plane->base.name,
3191 				sw_wm_level->enable,
3192 				sw_wm_level->blocks,
3193 				sw_wm_level->lines,
3194 				hw_wm_level->enable,
3195 				hw_wm_level->blocks,
3196 				hw_wm_level->lines);
3197 		}
3198 
3199 		hw_wm_level = &hw->wm.planes[plane->id].sagv.wm0;
3200 		sw_wm_level = &sw_wm->planes[plane->id].sagv.wm0;
3201 
3202 		if (HAS_HW_SAGV_WM(i915) &&
3203 		    !skl_wm_level_equals(hw_wm_level, sw_wm_level)) {
3204 			drm_err(&i915->drm,
3205 				"[PLANE:%d:%s] mismatch in SAGV WM (expected e=%d b=%u l=%u, got e=%d b=%u l=%u)\n",
3206 				plane->base.base.id, plane->base.name,
3207 				sw_wm_level->enable,
3208 				sw_wm_level->blocks,
3209 				sw_wm_level->lines,
3210 				hw_wm_level->enable,
3211 				hw_wm_level->blocks,
3212 				hw_wm_level->lines);
3213 		}
3214 
3215 		hw_wm_level = &hw->wm.planes[plane->id].sagv.trans_wm;
3216 		sw_wm_level = &sw_wm->planes[plane->id].sagv.trans_wm;
3217 
3218 		if (HAS_HW_SAGV_WM(i915) &&
3219 		    !skl_wm_level_equals(hw_wm_level, sw_wm_level)) {
3220 			drm_err(&i915->drm,
3221 				"[PLANE:%d:%s] mismatch in SAGV trans WM (expected e=%d b=%u l=%u, got e=%d b=%u l=%u)\n",
3222 				plane->base.base.id, plane->base.name,
3223 				sw_wm_level->enable,
3224 				sw_wm_level->blocks,
3225 				sw_wm_level->lines,
3226 				hw_wm_level->enable,
3227 				hw_wm_level->blocks,
3228 				hw_wm_level->lines);
3229 		}
3230 
3231 		/* DDB */
3232 		hw_ddb_entry = &hw->ddb[PLANE_CURSOR];
3233 		sw_ddb_entry = &new_crtc_state->wm.skl.plane_ddb[PLANE_CURSOR];
3234 
3235 		if (!skl_ddb_entry_equal(hw_ddb_entry, sw_ddb_entry)) {
3236 			drm_err(&i915->drm,
3237 				"[PLANE:%d:%s] mismatch in DDB (expected (%u,%u), found (%u,%u))\n",
3238 				plane->base.base.id, plane->base.name,
3239 				sw_ddb_entry->start, sw_ddb_entry->end,
3240 				hw_ddb_entry->start, hw_ddb_entry->end);
3241 		}
3242 	}
3243 
3244 	kfree(hw);
3245 }
3246 
3247 bool skl_watermark_ipc_enabled(struct drm_i915_private *i915)
3248 {
3249 	return i915->display.wm.ipc_enabled;
3250 }
3251 
3252 void skl_watermark_ipc_update(struct drm_i915_private *i915)
3253 {
3254 	if (!HAS_IPC(i915))
3255 		return;
3256 
3257 	intel_de_rmw(i915, DISP_ARB_CTL2, DISP_IPC_ENABLE,
3258 		     skl_watermark_ipc_enabled(i915) ? DISP_IPC_ENABLE : 0);
3259 }
3260 
3261 static bool skl_watermark_ipc_can_enable(struct drm_i915_private *i915)
3262 {
3263 	/* Display WA #0477 WaDisableIPC: skl */
3264 	if (IS_SKYLAKE(i915))
3265 		return false;
3266 
3267 	/* Display WA #1141: SKL:all KBL:all CFL */
3268 	if (IS_KABYLAKE(i915) ||
3269 	    IS_COFFEELAKE(i915) ||
3270 	    IS_COMETLAKE(i915))
3271 		return i915->dram_info.symmetric_memory;
3272 
3273 	return true;
3274 }
3275 
3276 void skl_watermark_ipc_init(struct drm_i915_private *i915)
3277 {
3278 	if (!HAS_IPC(i915))
3279 		return;
3280 
3281 	i915->display.wm.ipc_enabled = skl_watermark_ipc_can_enable(i915);
3282 
3283 	skl_watermark_ipc_update(i915);
3284 }
3285 
3286 static void
3287 adjust_wm_latency(struct drm_i915_private *i915,
3288 		  u16 wm[], int num_levels, int read_latency)
3289 {
3290 	bool wm_lv_0_adjust_needed = i915->dram_info.wm_lv_0_adjust_needed;
3291 	int i, level;
3292 
3293 	/*
3294 	 * If a level n (n > 1) has a 0us latency, all levels m (m >= n)
3295 	 * need to be disabled. We make sure to sanitize the values out
3296 	 * of the punit to satisfy this requirement.
3297 	 */
3298 	for (level = 1; level < num_levels; level++) {
3299 		if (wm[level] == 0) {
3300 			for (i = level + 1; i < num_levels; i++)
3301 				wm[i] = 0;
3302 
3303 			num_levels = level;
3304 			break;
3305 		}
3306 	}
3307 
3308 	/*
3309 	 * WaWmMemoryReadLatency
3310 	 *
3311 	 * punit doesn't take into account the read latency so we need
3312 	 * to add proper adjustement to each valid level we retrieve
3313 	 * from the punit when level 0 response data is 0us.
3314 	 */
3315 	if (wm[0] == 0) {
3316 		for (level = 0; level < num_levels; level++)
3317 			wm[level] += read_latency;
3318 	}
3319 
3320 	/*
3321 	 * WA Level-0 adjustment for 16GB DIMMs: SKL+
3322 	 * If we could not get dimm info enable this WA to prevent from
3323 	 * any underrun. If not able to get Dimm info assume 16GB dimm
3324 	 * to avoid any underrun.
3325 	 */
3326 	if (wm_lv_0_adjust_needed)
3327 		wm[0] += 1;
3328 }
3329 
3330 static void mtl_read_wm_latency(struct drm_i915_private *i915, u16 wm[])
3331 {
3332 	int num_levels = i915->display.wm.num_levels;
3333 	u32 val;
3334 
3335 	val = intel_de_read(i915, MTL_LATENCY_LP0_LP1);
3336 	wm[0] = REG_FIELD_GET(MTL_LATENCY_LEVEL_EVEN_MASK, val);
3337 	wm[1] = REG_FIELD_GET(MTL_LATENCY_LEVEL_ODD_MASK, val);
3338 
3339 	val = intel_de_read(i915, MTL_LATENCY_LP2_LP3);
3340 	wm[2] = REG_FIELD_GET(MTL_LATENCY_LEVEL_EVEN_MASK, val);
3341 	wm[3] = REG_FIELD_GET(MTL_LATENCY_LEVEL_ODD_MASK, val);
3342 
3343 	val = intel_de_read(i915, MTL_LATENCY_LP4_LP5);
3344 	wm[4] = REG_FIELD_GET(MTL_LATENCY_LEVEL_EVEN_MASK, val);
3345 	wm[5] = REG_FIELD_GET(MTL_LATENCY_LEVEL_ODD_MASK, val);
3346 
3347 	adjust_wm_latency(i915, wm, num_levels, 6);
3348 }
3349 
3350 static void skl_read_wm_latency(struct drm_i915_private *i915, u16 wm[])
3351 {
3352 	int num_levels = i915->display.wm.num_levels;
3353 	int read_latency = DISPLAY_VER(i915) >= 12 ? 3 : 2;
3354 	int mult = IS_DG2(i915) ? 2 : 1;
3355 	u32 val;
3356 	int ret;
3357 
3358 	/* read the first set of memory latencies[0:3] */
3359 	val = 0; /* data0 to be programmed to 0 for first set */
3360 	ret = snb_pcode_read(&i915->uncore, GEN9_PCODE_READ_MEM_LATENCY, &val, NULL);
3361 	if (ret) {
3362 		drm_err(&i915->drm, "SKL Mailbox read error = %d\n", ret);
3363 		return;
3364 	}
3365 
3366 	wm[0] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_0_4_MASK, val) * mult;
3367 	wm[1] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_1_5_MASK, val) * mult;
3368 	wm[2] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_2_6_MASK, val) * mult;
3369 	wm[3] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_3_7_MASK, val) * mult;
3370 
3371 	/* read the second set of memory latencies[4:7] */
3372 	val = 1; /* data0 to be programmed to 1 for second set */
3373 	ret = snb_pcode_read(&i915->uncore, GEN9_PCODE_READ_MEM_LATENCY, &val, NULL);
3374 	if (ret) {
3375 		drm_err(&i915->drm, "SKL Mailbox read error = %d\n", ret);
3376 		return;
3377 	}
3378 
3379 	wm[4] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_0_4_MASK, val) * mult;
3380 	wm[5] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_1_5_MASK, val) * mult;
3381 	wm[6] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_2_6_MASK, val) * mult;
3382 	wm[7] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_3_7_MASK, val) * mult;
3383 
3384 	adjust_wm_latency(i915, wm, num_levels, read_latency);
3385 }
3386 
3387 static void skl_setup_wm_latency(struct drm_i915_private *i915)
3388 {
3389 	if (HAS_HW_SAGV_WM(i915))
3390 		i915->display.wm.num_levels = 6;
3391 	else
3392 		i915->display.wm.num_levels = 8;
3393 
3394 	if (DISPLAY_VER(i915) >= 14)
3395 		mtl_read_wm_latency(i915, i915->display.wm.skl_latency);
3396 	else
3397 		skl_read_wm_latency(i915, i915->display.wm.skl_latency);
3398 
3399 	intel_print_wm_latency(i915, "Gen9 Plane", i915->display.wm.skl_latency);
3400 }
3401 
3402 static const struct intel_wm_funcs skl_wm_funcs = {
3403 	.compute_global_watermarks = skl_compute_wm,
3404 	.get_hw_state = skl_wm_get_hw_state_and_sanitize,
3405 };
3406 
3407 void skl_wm_init(struct drm_i915_private *i915)
3408 {
3409 	intel_sagv_init(i915);
3410 
3411 	skl_setup_wm_latency(i915);
3412 
3413 	i915->display.funcs.wm = &skl_wm_funcs;
3414 }
3415 
3416 static struct intel_global_state *intel_dbuf_duplicate_state(struct intel_global_obj *obj)
3417 {
3418 	struct intel_dbuf_state *dbuf_state;
3419 
3420 	dbuf_state = kmemdup(obj->state, sizeof(*dbuf_state), GFP_KERNEL);
3421 	if (!dbuf_state)
3422 		return NULL;
3423 
3424 	return &dbuf_state->base;
3425 }
3426 
3427 static void intel_dbuf_destroy_state(struct intel_global_obj *obj,
3428 				     struct intel_global_state *state)
3429 {
3430 	kfree(state);
3431 }
3432 
3433 static const struct intel_global_state_funcs intel_dbuf_funcs = {
3434 	.atomic_duplicate_state = intel_dbuf_duplicate_state,
3435 	.atomic_destroy_state = intel_dbuf_destroy_state,
3436 };
3437 
3438 struct intel_dbuf_state *
3439 intel_atomic_get_dbuf_state(struct intel_atomic_state *state)
3440 {
3441 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3442 	struct intel_global_state *dbuf_state;
3443 
3444 	dbuf_state = intel_atomic_get_global_obj_state(state, &i915->display.dbuf.obj);
3445 	if (IS_ERR(dbuf_state))
3446 		return ERR_CAST(dbuf_state);
3447 
3448 	return to_intel_dbuf_state(dbuf_state);
3449 }
3450 
3451 int intel_dbuf_init(struct drm_i915_private *i915)
3452 {
3453 	struct intel_dbuf_state *dbuf_state;
3454 
3455 	dbuf_state = kzalloc(sizeof(*dbuf_state), GFP_KERNEL);
3456 	if (!dbuf_state)
3457 		return -ENOMEM;
3458 
3459 	intel_atomic_global_obj_init(i915, &i915->display.dbuf.obj,
3460 				     &dbuf_state->base, &intel_dbuf_funcs);
3461 
3462 	return 0;
3463 }
3464 
3465 static bool xelpdp_is_only_pipe_per_dbuf_bank(enum pipe pipe, u8 active_pipes)
3466 {
3467 	switch (pipe) {
3468 	case PIPE_A:
3469 		return !(active_pipes & BIT(PIPE_D));
3470 	case PIPE_D:
3471 		return !(active_pipes & BIT(PIPE_A));
3472 	case PIPE_B:
3473 		return !(active_pipes & BIT(PIPE_C));
3474 	case PIPE_C:
3475 		return !(active_pipes & BIT(PIPE_B));
3476 	default: /* to suppress compiler warning */
3477 		MISSING_CASE(pipe);
3478 		break;
3479 	}
3480 
3481 	return false;
3482 }
3483 
3484 static void intel_mbus_dbox_update(struct intel_atomic_state *state)
3485 {
3486 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3487 	const struct intel_dbuf_state *new_dbuf_state, *old_dbuf_state;
3488 	const struct intel_crtc *crtc;
3489 	u32 val = 0;
3490 
3491 	if (DISPLAY_VER(i915) < 11)
3492 		return;
3493 
3494 	new_dbuf_state = intel_atomic_get_new_dbuf_state(state);
3495 	old_dbuf_state = intel_atomic_get_old_dbuf_state(state);
3496 	if (!new_dbuf_state ||
3497 	    (new_dbuf_state->joined_mbus == old_dbuf_state->joined_mbus &&
3498 	     new_dbuf_state->active_pipes == old_dbuf_state->active_pipes))
3499 		return;
3500 
3501 	if (DISPLAY_VER(i915) >= 14)
3502 		val |= MBUS_DBOX_I_CREDIT(2);
3503 
3504 	if (DISPLAY_VER(i915) >= 12) {
3505 		val |= MBUS_DBOX_B2B_TRANSACTIONS_MAX(16);
3506 		val |= MBUS_DBOX_B2B_TRANSACTIONS_DELAY(1);
3507 		val |= MBUS_DBOX_REGULATE_B2B_TRANSACTIONS_EN;
3508 	}
3509 
3510 	if (DISPLAY_VER(i915) >= 14)
3511 		val |= new_dbuf_state->joined_mbus ? MBUS_DBOX_A_CREDIT(12) :
3512 						     MBUS_DBOX_A_CREDIT(8);
3513 	else if (IS_ALDERLAKE_P(i915))
3514 		/* Wa_22010947358:adl-p */
3515 		val |= new_dbuf_state->joined_mbus ? MBUS_DBOX_A_CREDIT(6) :
3516 						     MBUS_DBOX_A_CREDIT(4);
3517 	else
3518 		val |= MBUS_DBOX_A_CREDIT(2);
3519 
3520 	if (DISPLAY_VER(i915) >= 14) {
3521 		val |= MBUS_DBOX_B_CREDIT(0xA);
3522 	} else if (IS_ALDERLAKE_P(i915)) {
3523 		val |= MBUS_DBOX_BW_CREDIT(2);
3524 		val |= MBUS_DBOX_B_CREDIT(8);
3525 	} else if (DISPLAY_VER(i915) >= 12) {
3526 		val |= MBUS_DBOX_BW_CREDIT(2);
3527 		val |= MBUS_DBOX_B_CREDIT(12);
3528 	} else {
3529 		val |= MBUS_DBOX_BW_CREDIT(1);
3530 		val |= MBUS_DBOX_B_CREDIT(8);
3531 	}
3532 
3533 	for_each_intel_crtc_in_pipe_mask(&i915->drm, crtc, new_dbuf_state->active_pipes) {
3534 		u32 pipe_val = val;
3535 
3536 		if (DISPLAY_VERx100(i915) == 1400) {
3537 			if (xelpdp_is_only_pipe_per_dbuf_bank(crtc->pipe,
3538 							      new_dbuf_state->active_pipes))
3539 				pipe_val |= MBUS_DBOX_BW_8CREDITS_MTL;
3540 			else
3541 				pipe_val |= MBUS_DBOX_BW_4CREDITS_MTL;
3542 		}
3543 
3544 		intel_de_write(i915, PIPE_MBUS_DBOX_CTL(crtc->pipe), pipe_val);
3545 	}
3546 }
3547 
3548 int intel_dbuf_state_set_mdclk_cdclk_ratio(struct intel_atomic_state *state,
3549 					   int ratio)
3550 {
3551 	struct intel_dbuf_state *dbuf_state;
3552 
3553 	dbuf_state = intel_atomic_get_dbuf_state(state);
3554 	if (IS_ERR(dbuf_state))
3555 		return PTR_ERR(dbuf_state);
3556 
3557 	dbuf_state->mdclk_cdclk_ratio = ratio;
3558 
3559 	return intel_atomic_lock_global_state(&dbuf_state->base);
3560 }
3561 
3562 void intel_dbuf_mdclk_cdclk_ratio_update(struct drm_i915_private *i915,
3563 					 int ratio, bool joined_mbus)
3564 {
3565 	enum dbuf_slice slice;
3566 
3567 	if (!HAS_MBUS_JOINING(i915))
3568 		return;
3569 
3570 	if (DISPLAY_VER(i915) >= 20)
3571 		intel_de_rmw(i915, MBUS_CTL, MBUS_TRANSLATION_THROTTLE_MIN_MASK,
3572 			     MBUS_TRANSLATION_THROTTLE_MIN(ratio - 1));
3573 
3574 	if (joined_mbus)
3575 		ratio *= 2;
3576 
3577 	drm_dbg_kms(&i915->drm, "Updating dbuf ratio to %d (mbus joined: %s)\n",
3578 		    ratio, str_yes_no(joined_mbus));
3579 
3580 	for_each_dbuf_slice(i915, slice)
3581 		intel_de_rmw(i915, DBUF_CTL_S(slice),
3582 			     DBUF_MIN_TRACKER_STATE_SERVICE_MASK,
3583 			     DBUF_MIN_TRACKER_STATE_SERVICE(ratio - 1));
3584 }
3585 
3586 static void intel_dbuf_mdclk_min_tracker_update(struct intel_atomic_state *state)
3587 {
3588 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3589 	const struct intel_dbuf_state *old_dbuf_state =
3590 		intel_atomic_get_old_dbuf_state(state);
3591 	const struct intel_dbuf_state *new_dbuf_state =
3592 		intel_atomic_get_new_dbuf_state(state);
3593 	int mdclk_cdclk_ratio;
3594 
3595 	if (intel_cdclk_is_decreasing_later(state)) {
3596 		/* cdclk/mdclk will be changed later by intel_set_cdclk_post_plane_update() */
3597 		mdclk_cdclk_ratio = old_dbuf_state->mdclk_cdclk_ratio;
3598 	} else {
3599 		/* cdclk/mdclk already changed by intel_set_cdclk_pre_plane_update() */
3600 		mdclk_cdclk_ratio = new_dbuf_state->mdclk_cdclk_ratio;
3601 	}
3602 
3603 	intel_dbuf_mdclk_cdclk_ratio_update(i915, mdclk_cdclk_ratio,
3604 					    new_dbuf_state->joined_mbus);
3605 }
3606 
3607 static enum pipe intel_mbus_joined_pipe(struct intel_atomic_state *state,
3608 					const struct intel_dbuf_state *dbuf_state)
3609 {
3610 	struct intel_display *display = to_intel_display(state);
3611 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3612 	enum pipe pipe = ffs(dbuf_state->active_pipes) - 1;
3613 	const struct intel_crtc_state *new_crtc_state;
3614 	struct intel_crtc *crtc;
3615 
3616 	drm_WARN_ON(&i915->drm, !dbuf_state->joined_mbus);
3617 	drm_WARN_ON(&i915->drm, !is_power_of_2(dbuf_state->active_pipes));
3618 
3619 	crtc = intel_crtc_for_pipe(display, pipe);
3620 	new_crtc_state = intel_atomic_get_new_crtc_state(state, crtc);
3621 
3622 	if (new_crtc_state && !intel_crtc_needs_modeset(new_crtc_state))
3623 		return pipe;
3624 	else
3625 		return INVALID_PIPE;
3626 }
3627 
3628 static void intel_dbuf_mbus_join_update(struct intel_atomic_state *state,
3629 					enum pipe pipe)
3630 {
3631 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3632 	const struct intel_dbuf_state *old_dbuf_state =
3633 		intel_atomic_get_old_dbuf_state(state);
3634 	const struct intel_dbuf_state *new_dbuf_state =
3635 		intel_atomic_get_new_dbuf_state(state);
3636 	u32 mbus_ctl;
3637 
3638 	drm_dbg_kms(&i915->drm, "Changing mbus joined: %s -> %s (pipe: %c)\n",
3639 		    str_yes_no(old_dbuf_state->joined_mbus),
3640 		    str_yes_no(new_dbuf_state->joined_mbus),
3641 		    pipe != INVALID_PIPE ? pipe_name(pipe) : '*');
3642 
3643 	if (new_dbuf_state->joined_mbus)
3644 		mbus_ctl = MBUS_HASHING_MODE_1x4 | MBUS_JOIN;
3645 	else
3646 		mbus_ctl = MBUS_HASHING_MODE_2x2;
3647 
3648 	if (pipe != INVALID_PIPE)
3649 		mbus_ctl |= MBUS_JOIN_PIPE_SELECT(pipe);
3650 	else
3651 		mbus_ctl |= MBUS_JOIN_PIPE_SELECT_NONE;
3652 
3653 	intel_de_rmw(i915, MBUS_CTL,
3654 		     MBUS_HASHING_MODE_MASK | MBUS_JOIN |
3655 		     MBUS_JOIN_PIPE_SELECT_MASK, mbus_ctl);
3656 }
3657 
3658 void intel_dbuf_mbus_pre_ddb_update(struct intel_atomic_state *state)
3659 {
3660 	const struct intel_dbuf_state *new_dbuf_state =
3661 		intel_atomic_get_new_dbuf_state(state);
3662 	const struct intel_dbuf_state *old_dbuf_state =
3663 		intel_atomic_get_old_dbuf_state(state);
3664 
3665 	if (!new_dbuf_state)
3666 		return;
3667 
3668 	if (!old_dbuf_state->joined_mbus && new_dbuf_state->joined_mbus) {
3669 		enum pipe pipe = intel_mbus_joined_pipe(state, new_dbuf_state);
3670 
3671 		WARN_ON(!new_dbuf_state->base.changed);
3672 
3673 		intel_dbuf_mbus_join_update(state, pipe);
3674 		intel_mbus_dbox_update(state);
3675 		intel_dbuf_mdclk_min_tracker_update(state);
3676 	}
3677 }
3678 
3679 void intel_dbuf_mbus_post_ddb_update(struct intel_atomic_state *state)
3680 {
3681 	struct intel_display *display = to_intel_display(state);
3682 	const struct intel_dbuf_state *new_dbuf_state =
3683 		intel_atomic_get_new_dbuf_state(state);
3684 	const struct intel_dbuf_state *old_dbuf_state =
3685 		intel_atomic_get_old_dbuf_state(state);
3686 
3687 	if (!new_dbuf_state)
3688 		return;
3689 
3690 	if (old_dbuf_state->joined_mbus && !new_dbuf_state->joined_mbus) {
3691 		enum pipe pipe = intel_mbus_joined_pipe(state, old_dbuf_state);
3692 
3693 		WARN_ON(!new_dbuf_state->base.changed);
3694 
3695 		intel_dbuf_mdclk_min_tracker_update(state);
3696 		intel_mbus_dbox_update(state);
3697 		intel_dbuf_mbus_join_update(state, pipe);
3698 
3699 		if (pipe != INVALID_PIPE) {
3700 			struct intel_crtc *crtc = intel_crtc_for_pipe(display, pipe);
3701 
3702 			intel_crtc_wait_for_next_vblank(crtc);
3703 		}
3704 	} else if (old_dbuf_state->joined_mbus == new_dbuf_state->joined_mbus &&
3705 		   old_dbuf_state->active_pipes != new_dbuf_state->active_pipes) {
3706 		WARN_ON(!new_dbuf_state->base.changed);
3707 
3708 		intel_dbuf_mdclk_min_tracker_update(state);
3709 		intel_mbus_dbox_update(state);
3710 	}
3711 
3712 }
3713 
3714 void intel_dbuf_pre_plane_update(struct intel_atomic_state *state)
3715 {
3716 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3717 	const struct intel_dbuf_state *new_dbuf_state =
3718 		intel_atomic_get_new_dbuf_state(state);
3719 	const struct intel_dbuf_state *old_dbuf_state =
3720 		intel_atomic_get_old_dbuf_state(state);
3721 	u8 old_slices, new_slices;
3722 
3723 	if (!new_dbuf_state)
3724 		return;
3725 
3726 	old_slices = old_dbuf_state->enabled_slices;
3727 	new_slices = old_dbuf_state->enabled_slices | new_dbuf_state->enabled_slices;
3728 
3729 	if (old_slices == new_slices)
3730 		return;
3731 
3732 	WARN_ON(!new_dbuf_state->base.changed);
3733 
3734 	gen9_dbuf_slices_update(i915, new_slices);
3735 }
3736 
3737 void intel_dbuf_post_plane_update(struct intel_atomic_state *state)
3738 {
3739 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3740 	const struct intel_dbuf_state *new_dbuf_state =
3741 		intel_atomic_get_new_dbuf_state(state);
3742 	const struct intel_dbuf_state *old_dbuf_state =
3743 		intel_atomic_get_old_dbuf_state(state);
3744 	u8 old_slices, new_slices;
3745 
3746 	if (!new_dbuf_state)
3747 		return;
3748 
3749 	old_slices = old_dbuf_state->enabled_slices | new_dbuf_state->enabled_slices;
3750 	new_slices = new_dbuf_state->enabled_slices;
3751 
3752 	if (old_slices == new_slices)
3753 		return;
3754 
3755 	WARN_ON(!new_dbuf_state->base.changed);
3756 
3757 	gen9_dbuf_slices_update(i915, new_slices);
3758 }
3759 
3760 static int skl_watermark_ipc_status_show(struct seq_file *m, void *data)
3761 {
3762 	struct drm_i915_private *i915 = m->private;
3763 
3764 	seq_printf(m, "Isochronous Priority Control: %s\n",
3765 		   str_yes_no(skl_watermark_ipc_enabled(i915)));
3766 	return 0;
3767 }
3768 
3769 static int skl_watermark_ipc_status_open(struct inode *inode, struct file *file)
3770 {
3771 	struct drm_i915_private *i915 = inode->i_private;
3772 
3773 	return single_open(file, skl_watermark_ipc_status_show, i915);
3774 }
3775 
3776 static ssize_t skl_watermark_ipc_status_write(struct file *file,
3777 					      const char __user *ubuf,
3778 					      size_t len, loff_t *offp)
3779 {
3780 	struct seq_file *m = file->private_data;
3781 	struct drm_i915_private *i915 = m->private;
3782 	intel_wakeref_t wakeref;
3783 	bool enable;
3784 	int ret;
3785 
3786 	ret = kstrtobool_from_user(ubuf, len, &enable);
3787 	if (ret < 0)
3788 		return ret;
3789 
3790 	with_intel_runtime_pm(&i915->runtime_pm, wakeref) {
3791 		if (!skl_watermark_ipc_enabled(i915) && enable)
3792 			drm_info(&i915->drm,
3793 				 "Enabling IPC: WM will be proper only after next commit\n");
3794 		i915->display.wm.ipc_enabled = enable;
3795 		skl_watermark_ipc_update(i915);
3796 	}
3797 
3798 	return len;
3799 }
3800 
3801 static const struct file_operations skl_watermark_ipc_status_fops = {
3802 	.owner = THIS_MODULE,
3803 	.open = skl_watermark_ipc_status_open,
3804 	.read = seq_read,
3805 	.llseek = seq_lseek,
3806 	.release = single_release,
3807 	.write = skl_watermark_ipc_status_write
3808 };
3809 
3810 static int intel_sagv_status_show(struct seq_file *m, void *unused)
3811 {
3812 	struct drm_i915_private *i915 = m->private;
3813 	static const char * const sagv_status[] = {
3814 		[I915_SAGV_UNKNOWN] = "unknown",
3815 		[I915_SAGV_DISABLED] = "disabled",
3816 		[I915_SAGV_ENABLED] = "enabled",
3817 		[I915_SAGV_NOT_CONTROLLED] = "not controlled",
3818 	};
3819 
3820 	seq_printf(m, "SAGV available: %s\n", str_yes_no(intel_has_sagv(i915)));
3821 	seq_printf(m, "SAGV modparam: %s\n",
3822 		   str_enabled_disabled(i915->display.params.enable_sagv));
3823 	seq_printf(m, "SAGV status: %s\n", sagv_status[i915->display.sagv.status]);
3824 	seq_printf(m, "SAGV block time: %d usec\n", i915->display.sagv.block_time_us);
3825 
3826 	return 0;
3827 }
3828 
3829 DEFINE_SHOW_ATTRIBUTE(intel_sagv_status);
3830 
3831 void skl_watermark_debugfs_register(struct drm_i915_private *i915)
3832 {
3833 	struct drm_minor *minor = i915->drm.primary;
3834 
3835 	if (HAS_IPC(i915))
3836 		debugfs_create_file("i915_ipc_status", 0644, minor->debugfs_root, i915,
3837 				    &skl_watermark_ipc_status_fops);
3838 
3839 	if (HAS_SAGV(i915))
3840 		debugfs_create_file("i915_sagv_status", 0444, minor->debugfs_root, i915,
3841 				    &intel_sagv_status_fops);
3842 }
3843 
3844 unsigned int skl_watermark_max_latency(struct drm_i915_private *i915, int initial_wm_level)
3845 {
3846 	int level;
3847 
3848 	for (level = i915->display.wm.num_levels - 1; level >= initial_wm_level; level--) {
3849 		unsigned int latency = skl_wm_latency(i915, level, NULL);
3850 
3851 		if (latency)
3852 			return latency;
3853 	}
3854 
3855 	return 0;
3856 }
3857