xref: /linux/drivers/gpu/drm/i915/display/skl_watermark.c (revision f5c31bcf604db54470868f3118a60dc4a9ba8813)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright © 2022 Intel Corporation
4  */
5 
6 #include <drm/drm_blend.h>
7 
8 #include "i915_drv.h"
9 #include "i915_fixed.h"
10 #include "i915_reg.h"
11 #include "i9xx_wm.h"
12 #include "intel_atomic.h"
13 #include "intel_atomic_plane.h"
14 #include "intel_bw.h"
15 #include "intel_crtc.h"
16 #include "intel_de.h"
17 #include "intel_display.h"
18 #include "intel_display_power.h"
19 #include "intel_display_types.h"
20 #include "intel_fb.h"
21 #include "intel_pcode.h"
22 #include "intel_wm.h"
23 #include "skl_watermark.h"
24 #include "skl_watermark_regs.h"
25 
26 /*It is expected that DSB can do posted writes to every register in
27  * the pipe and planes within 100us. For flip queue use case, the
28  * recommended DSB execution time is 100us + one SAGV block time.
29  */
30 #define DSB_EXE_TIME 100
31 
32 static void skl_sagv_disable(struct drm_i915_private *i915);
33 
34 /* Stores plane specific WM parameters */
35 struct skl_wm_params {
36 	bool x_tiled, y_tiled;
37 	bool rc_surface;
38 	bool is_planar;
39 	u32 width;
40 	u8 cpp;
41 	u32 plane_pixel_rate;
42 	u32 y_min_scanlines;
43 	u32 plane_bytes_per_line;
44 	uint_fixed_16_16_t plane_blocks_per_line;
45 	uint_fixed_16_16_t y_tile_minimum;
46 	u32 linetime_us;
47 	u32 dbuf_block_size;
48 };
49 
50 u8 intel_enabled_dbuf_slices_mask(struct drm_i915_private *i915)
51 {
52 	u8 enabled_slices = 0;
53 	enum dbuf_slice slice;
54 
55 	for_each_dbuf_slice(i915, slice) {
56 		if (intel_de_read(i915, DBUF_CTL_S(slice)) & DBUF_POWER_STATE)
57 			enabled_slices |= BIT(slice);
58 	}
59 
60 	return enabled_slices;
61 }
62 
63 /*
64  * FIXME: We still don't have the proper code detect if we need to apply the WA,
65  * so assume we'll always need it in order to avoid underruns.
66  */
67 static bool skl_needs_memory_bw_wa(struct drm_i915_private *i915)
68 {
69 	return DISPLAY_VER(i915) == 9;
70 }
71 
72 static bool
73 intel_has_sagv(struct drm_i915_private *i915)
74 {
75 	return HAS_SAGV(i915) &&
76 		i915->display.sagv.status != I915_SAGV_NOT_CONTROLLED;
77 }
78 
79 static u32
80 intel_sagv_block_time(struct drm_i915_private *i915)
81 {
82 	if (DISPLAY_VER(i915) >= 14) {
83 		u32 val;
84 
85 		val = intel_de_read(i915, MTL_LATENCY_SAGV);
86 
87 		return REG_FIELD_GET(MTL_LATENCY_QCLK_SAGV, val);
88 	} else if (DISPLAY_VER(i915) >= 12) {
89 		u32 val = 0;
90 		int ret;
91 
92 		ret = snb_pcode_read(&i915->uncore,
93 				     GEN12_PCODE_READ_SAGV_BLOCK_TIME_US,
94 				     &val, NULL);
95 		if (ret) {
96 			drm_dbg_kms(&i915->drm, "Couldn't read SAGV block time!\n");
97 			return 0;
98 		}
99 
100 		return val;
101 	} else if (DISPLAY_VER(i915) == 11) {
102 		return 10;
103 	} else if (HAS_SAGV(i915)) {
104 		return 30;
105 	} else {
106 		return 0;
107 	}
108 }
109 
110 static void intel_sagv_init(struct drm_i915_private *i915)
111 {
112 	if (!HAS_SAGV(i915))
113 		i915->display.sagv.status = I915_SAGV_NOT_CONTROLLED;
114 
115 	/*
116 	 * Probe to see if we have working SAGV control.
117 	 * For icl+ this was already determined by intel_bw_init_hw().
118 	 */
119 	if (DISPLAY_VER(i915) < 11)
120 		skl_sagv_disable(i915);
121 
122 	drm_WARN_ON(&i915->drm, i915->display.sagv.status == I915_SAGV_UNKNOWN);
123 
124 	i915->display.sagv.block_time_us = intel_sagv_block_time(i915);
125 
126 	drm_dbg_kms(&i915->drm, "SAGV supported: %s, original SAGV block time: %u us\n",
127 		    str_yes_no(intel_has_sagv(i915)), i915->display.sagv.block_time_us);
128 
129 	/* avoid overflow when adding with wm0 latency/etc. */
130 	if (drm_WARN(&i915->drm, i915->display.sagv.block_time_us > U16_MAX,
131 		     "Excessive SAGV block time %u, ignoring\n",
132 		     i915->display.sagv.block_time_us))
133 		i915->display.sagv.block_time_us = 0;
134 
135 	if (!intel_has_sagv(i915))
136 		i915->display.sagv.block_time_us = 0;
137 }
138 
139 /*
140  * SAGV dynamically adjusts the system agent voltage and clock frequencies
141  * depending on power and performance requirements. The display engine access
142  * to system memory is blocked during the adjustment time. Because of the
143  * blocking time, having this enabled can cause full system hangs and/or pipe
144  * underruns if we don't meet all of the following requirements:
145  *
146  *  - <= 1 pipe enabled
147  *  - All planes can enable watermarks for latencies >= SAGV engine block time
148  *  - We're not using an interlaced display configuration
149  */
150 static void skl_sagv_enable(struct drm_i915_private *i915)
151 {
152 	int ret;
153 
154 	if (!intel_has_sagv(i915))
155 		return;
156 
157 	if (i915->display.sagv.status == I915_SAGV_ENABLED)
158 		return;
159 
160 	drm_dbg_kms(&i915->drm, "Enabling SAGV\n");
161 	ret = snb_pcode_write(&i915->uncore, GEN9_PCODE_SAGV_CONTROL,
162 			      GEN9_SAGV_ENABLE);
163 
164 	/* We don't need to wait for SAGV when enabling */
165 
166 	/*
167 	 * Some skl systems, pre-release machines in particular,
168 	 * don't actually have SAGV.
169 	 */
170 	if (IS_SKYLAKE(i915) && ret == -ENXIO) {
171 		drm_dbg(&i915->drm, "No SAGV found on system, ignoring\n");
172 		i915->display.sagv.status = I915_SAGV_NOT_CONTROLLED;
173 		return;
174 	} else if (ret < 0) {
175 		drm_err(&i915->drm, "Failed to enable SAGV\n");
176 		return;
177 	}
178 
179 	i915->display.sagv.status = I915_SAGV_ENABLED;
180 }
181 
182 static void skl_sagv_disable(struct drm_i915_private *i915)
183 {
184 	int ret;
185 
186 	if (!intel_has_sagv(i915))
187 		return;
188 
189 	if (i915->display.sagv.status == I915_SAGV_DISABLED)
190 		return;
191 
192 	drm_dbg_kms(&i915->drm, "Disabling SAGV\n");
193 	/* bspec says to keep retrying for at least 1 ms */
194 	ret = skl_pcode_request(&i915->uncore, GEN9_PCODE_SAGV_CONTROL,
195 				GEN9_SAGV_DISABLE,
196 				GEN9_SAGV_IS_DISABLED, GEN9_SAGV_IS_DISABLED,
197 				1);
198 	/*
199 	 * Some skl systems, pre-release machines in particular,
200 	 * don't actually have SAGV.
201 	 */
202 	if (IS_SKYLAKE(i915) && ret == -ENXIO) {
203 		drm_dbg(&i915->drm, "No SAGV found on system, ignoring\n");
204 		i915->display.sagv.status = I915_SAGV_NOT_CONTROLLED;
205 		return;
206 	} else if (ret < 0) {
207 		drm_err(&i915->drm, "Failed to disable SAGV (%d)\n", ret);
208 		return;
209 	}
210 
211 	i915->display.sagv.status = I915_SAGV_DISABLED;
212 }
213 
214 static void skl_sagv_pre_plane_update(struct intel_atomic_state *state)
215 {
216 	struct drm_i915_private *i915 = to_i915(state->base.dev);
217 	const struct intel_bw_state *new_bw_state =
218 		intel_atomic_get_new_bw_state(state);
219 
220 	if (!new_bw_state)
221 		return;
222 
223 	if (!intel_can_enable_sagv(i915, new_bw_state))
224 		skl_sagv_disable(i915);
225 }
226 
227 static void skl_sagv_post_plane_update(struct intel_atomic_state *state)
228 {
229 	struct drm_i915_private *i915 = to_i915(state->base.dev);
230 	const struct intel_bw_state *new_bw_state =
231 		intel_atomic_get_new_bw_state(state);
232 
233 	if (!new_bw_state)
234 		return;
235 
236 	if (intel_can_enable_sagv(i915, new_bw_state))
237 		skl_sagv_enable(i915);
238 }
239 
240 static void icl_sagv_pre_plane_update(struct intel_atomic_state *state)
241 {
242 	struct drm_i915_private *i915 = to_i915(state->base.dev);
243 	const struct intel_bw_state *old_bw_state =
244 		intel_atomic_get_old_bw_state(state);
245 	const struct intel_bw_state *new_bw_state =
246 		intel_atomic_get_new_bw_state(state);
247 	u16 old_mask, new_mask;
248 
249 	if (!new_bw_state)
250 		return;
251 
252 	old_mask = old_bw_state->qgv_points_mask;
253 	new_mask = old_bw_state->qgv_points_mask | new_bw_state->qgv_points_mask;
254 
255 	if (old_mask == new_mask)
256 		return;
257 
258 	WARN_ON(!new_bw_state->base.changed);
259 
260 	drm_dbg_kms(&i915->drm, "Restricting QGV points: 0x%x -> 0x%x\n",
261 		    old_mask, new_mask);
262 
263 	/*
264 	 * Restrict required qgv points before updating the configuration.
265 	 * According to BSpec we can't mask and unmask qgv points at the same
266 	 * time. Also masking should be done before updating the configuration
267 	 * and unmasking afterwards.
268 	 */
269 	icl_pcode_restrict_qgv_points(i915, new_mask);
270 }
271 
272 static void icl_sagv_post_plane_update(struct intel_atomic_state *state)
273 {
274 	struct drm_i915_private *i915 = to_i915(state->base.dev);
275 	const struct intel_bw_state *old_bw_state =
276 		intel_atomic_get_old_bw_state(state);
277 	const struct intel_bw_state *new_bw_state =
278 		intel_atomic_get_new_bw_state(state);
279 	u16 old_mask, new_mask;
280 
281 	if (!new_bw_state)
282 		return;
283 
284 	old_mask = old_bw_state->qgv_points_mask | new_bw_state->qgv_points_mask;
285 	new_mask = new_bw_state->qgv_points_mask;
286 
287 	if (old_mask == new_mask)
288 		return;
289 
290 	WARN_ON(!new_bw_state->base.changed);
291 
292 	drm_dbg_kms(&i915->drm, "Relaxing QGV points: 0x%x -> 0x%x\n",
293 		    old_mask, new_mask);
294 
295 	/*
296 	 * Allow required qgv points after updating the configuration.
297 	 * According to BSpec we can't mask and unmask qgv points at the same
298 	 * time. Also masking should be done before updating the configuration
299 	 * and unmasking afterwards.
300 	 */
301 	icl_pcode_restrict_qgv_points(i915, new_mask);
302 }
303 
304 void intel_sagv_pre_plane_update(struct intel_atomic_state *state)
305 {
306 	struct drm_i915_private *i915 = to_i915(state->base.dev);
307 
308 	/*
309 	 * Just return if we can't control SAGV or don't have it.
310 	 * This is different from situation when we have SAGV but just can't
311 	 * afford it due to DBuf limitation - in case if SAGV is completely
312 	 * disabled in a BIOS, we are not even allowed to send a PCode request,
313 	 * as it will throw an error. So have to check it here.
314 	 */
315 	if (!intel_has_sagv(i915))
316 		return;
317 
318 	if (DISPLAY_VER(i915) >= 11)
319 		icl_sagv_pre_plane_update(state);
320 	else
321 		skl_sagv_pre_plane_update(state);
322 }
323 
324 void intel_sagv_post_plane_update(struct intel_atomic_state *state)
325 {
326 	struct drm_i915_private *i915 = to_i915(state->base.dev);
327 
328 	/*
329 	 * Just return if we can't control SAGV or don't have it.
330 	 * This is different from situation when we have SAGV but just can't
331 	 * afford it due to DBuf limitation - in case if SAGV is completely
332 	 * disabled in a BIOS, we are not even allowed to send a PCode request,
333 	 * as it will throw an error. So have to check it here.
334 	 */
335 	if (!intel_has_sagv(i915))
336 		return;
337 
338 	if (DISPLAY_VER(i915) >= 11)
339 		icl_sagv_post_plane_update(state);
340 	else
341 		skl_sagv_post_plane_update(state);
342 }
343 
344 static bool skl_crtc_can_enable_sagv(const struct intel_crtc_state *crtc_state)
345 {
346 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
347 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
348 	enum plane_id plane_id;
349 	int max_level = INT_MAX;
350 
351 	if (!intel_has_sagv(i915))
352 		return false;
353 
354 	if (!crtc_state->hw.active)
355 		return true;
356 
357 	if (crtc_state->hw.pipe_mode.flags & DRM_MODE_FLAG_INTERLACE)
358 		return false;
359 
360 	for_each_plane_id_on_crtc(crtc, plane_id) {
361 		const struct skl_plane_wm *wm =
362 			&crtc_state->wm.skl.optimal.planes[plane_id];
363 		int level;
364 
365 		/* Skip this plane if it's not enabled */
366 		if (!wm->wm[0].enable)
367 			continue;
368 
369 		/* Find the highest enabled wm level for this plane */
370 		for (level = i915->display.wm.num_levels - 1;
371 		     !wm->wm[level].enable; --level)
372 		     { }
373 
374 		/* Highest common enabled wm level for all planes */
375 		max_level = min(level, max_level);
376 	}
377 
378 	/* No enabled planes? */
379 	if (max_level == INT_MAX)
380 		return true;
381 
382 	for_each_plane_id_on_crtc(crtc, plane_id) {
383 		const struct skl_plane_wm *wm =
384 			&crtc_state->wm.skl.optimal.planes[plane_id];
385 
386 		/*
387 		 * All enabled planes must have enabled a common wm level that
388 		 * can tolerate memory latencies higher than sagv_block_time_us
389 		 */
390 		if (wm->wm[0].enable && !wm->wm[max_level].can_sagv)
391 			return false;
392 	}
393 
394 	return true;
395 }
396 
397 static bool tgl_crtc_can_enable_sagv(const struct intel_crtc_state *crtc_state)
398 {
399 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
400 	enum plane_id plane_id;
401 
402 	if (!crtc_state->hw.active)
403 		return true;
404 
405 	for_each_plane_id_on_crtc(crtc, plane_id) {
406 		const struct skl_plane_wm *wm =
407 			&crtc_state->wm.skl.optimal.planes[plane_id];
408 
409 		if (wm->wm[0].enable && !wm->sagv.wm0.enable)
410 			return false;
411 	}
412 
413 	return true;
414 }
415 
416 static bool intel_crtc_can_enable_sagv(const struct intel_crtc_state *crtc_state)
417 {
418 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
419 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
420 
421 	if (!i915->display.params.enable_sagv)
422 		return false;
423 
424 	if (DISPLAY_VER(i915) >= 12)
425 		return tgl_crtc_can_enable_sagv(crtc_state);
426 	else
427 		return skl_crtc_can_enable_sagv(crtc_state);
428 }
429 
430 bool intel_can_enable_sagv(struct drm_i915_private *i915,
431 			   const struct intel_bw_state *bw_state)
432 {
433 	if (DISPLAY_VER(i915) < 11 &&
434 	    bw_state->active_pipes && !is_power_of_2(bw_state->active_pipes))
435 		return false;
436 
437 	return bw_state->pipe_sagv_reject == 0;
438 }
439 
440 static int intel_compute_sagv_mask(struct intel_atomic_state *state)
441 {
442 	struct drm_i915_private *i915 = to_i915(state->base.dev);
443 	int ret;
444 	struct intel_crtc *crtc;
445 	struct intel_crtc_state *new_crtc_state;
446 	struct intel_bw_state *new_bw_state = NULL;
447 	const struct intel_bw_state *old_bw_state = NULL;
448 	int i;
449 
450 	for_each_new_intel_crtc_in_state(state, crtc,
451 					 new_crtc_state, i) {
452 		struct skl_pipe_wm *pipe_wm = &new_crtc_state->wm.skl.optimal;
453 
454 		new_bw_state = intel_atomic_get_bw_state(state);
455 		if (IS_ERR(new_bw_state))
456 			return PTR_ERR(new_bw_state);
457 
458 		old_bw_state = intel_atomic_get_old_bw_state(state);
459 
460 		/*
461 		 * We store use_sagv_wm in the crtc state rather than relying on
462 		 * that bw state since we have no convenient way to get at the
463 		 * latter from the plane commit hooks (especially in the legacy
464 		 * cursor case).
465 		 *
466 		 * drm_atomic_check_only() gets upset if we pull more crtcs
467 		 * into the state, so we have to calculate this based on the
468 		 * individual intel_crtc_can_enable_sagv() rather than
469 		 * the overall intel_can_enable_sagv(). Otherwise the
470 		 * crtcs not included in the commit would not switch to the
471 		 * SAGV watermarks when we are about to enable SAGV, and that
472 		 * would lead to underruns. This does mean extra power draw
473 		 * when only a subset of the crtcs are blocking SAGV as the
474 		 * other crtcs can't be allowed to use the more optimal
475 		 * normal (ie. non-SAGV) watermarks.
476 		 */
477 		pipe_wm->use_sagv_wm = !HAS_HW_SAGV_WM(i915) &&
478 			DISPLAY_VER(i915) >= 12 &&
479 			intel_crtc_can_enable_sagv(new_crtc_state);
480 
481 		if (intel_crtc_can_enable_sagv(new_crtc_state))
482 			new_bw_state->pipe_sagv_reject &= ~BIT(crtc->pipe);
483 		else
484 			new_bw_state->pipe_sagv_reject |= BIT(crtc->pipe);
485 	}
486 
487 	if (!new_bw_state)
488 		return 0;
489 
490 	new_bw_state->active_pipes =
491 		intel_calc_active_pipes(state, old_bw_state->active_pipes);
492 
493 	if (new_bw_state->active_pipes != old_bw_state->active_pipes) {
494 		ret = intel_atomic_lock_global_state(&new_bw_state->base);
495 		if (ret)
496 			return ret;
497 	}
498 
499 	if (intel_can_enable_sagv(i915, new_bw_state) !=
500 	    intel_can_enable_sagv(i915, old_bw_state)) {
501 		ret = intel_atomic_serialize_global_state(&new_bw_state->base);
502 		if (ret)
503 			return ret;
504 	} else if (new_bw_state->pipe_sagv_reject != old_bw_state->pipe_sagv_reject) {
505 		ret = intel_atomic_lock_global_state(&new_bw_state->base);
506 		if (ret)
507 			return ret;
508 	}
509 
510 	return 0;
511 }
512 
513 static u16 skl_ddb_entry_init(struct skl_ddb_entry *entry,
514 			      u16 start, u16 end)
515 {
516 	entry->start = start;
517 	entry->end = end;
518 
519 	return end;
520 }
521 
522 static int intel_dbuf_slice_size(struct drm_i915_private *i915)
523 {
524 	return DISPLAY_INFO(i915)->dbuf.size /
525 		hweight8(DISPLAY_INFO(i915)->dbuf.slice_mask);
526 }
527 
528 static void
529 skl_ddb_entry_for_slices(struct drm_i915_private *i915, u8 slice_mask,
530 			 struct skl_ddb_entry *ddb)
531 {
532 	int slice_size = intel_dbuf_slice_size(i915);
533 
534 	if (!slice_mask) {
535 		ddb->start = 0;
536 		ddb->end = 0;
537 		return;
538 	}
539 
540 	ddb->start = (ffs(slice_mask) - 1) * slice_size;
541 	ddb->end = fls(slice_mask) * slice_size;
542 
543 	WARN_ON(ddb->start >= ddb->end);
544 	WARN_ON(ddb->end > DISPLAY_INFO(i915)->dbuf.size);
545 }
546 
547 static unsigned int mbus_ddb_offset(struct drm_i915_private *i915, u8 slice_mask)
548 {
549 	struct skl_ddb_entry ddb;
550 
551 	if (slice_mask & (BIT(DBUF_S1) | BIT(DBUF_S2)))
552 		slice_mask = BIT(DBUF_S1);
553 	else if (slice_mask & (BIT(DBUF_S3) | BIT(DBUF_S4)))
554 		slice_mask = BIT(DBUF_S3);
555 
556 	skl_ddb_entry_for_slices(i915, slice_mask, &ddb);
557 
558 	return ddb.start;
559 }
560 
561 u32 skl_ddb_dbuf_slice_mask(struct drm_i915_private *i915,
562 			    const struct skl_ddb_entry *entry)
563 {
564 	int slice_size = intel_dbuf_slice_size(i915);
565 	enum dbuf_slice start_slice, end_slice;
566 	u8 slice_mask = 0;
567 
568 	if (!skl_ddb_entry_size(entry))
569 		return 0;
570 
571 	start_slice = entry->start / slice_size;
572 	end_slice = (entry->end - 1) / slice_size;
573 
574 	/*
575 	 * Per plane DDB entry can in a really worst case be on multiple slices
576 	 * but single entry is anyway contigious.
577 	 */
578 	while (start_slice <= end_slice) {
579 		slice_mask |= BIT(start_slice);
580 		start_slice++;
581 	}
582 
583 	return slice_mask;
584 }
585 
586 static unsigned int intel_crtc_ddb_weight(const struct intel_crtc_state *crtc_state)
587 {
588 	const struct drm_display_mode *pipe_mode = &crtc_state->hw.pipe_mode;
589 	int hdisplay, vdisplay;
590 
591 	if (!crtc_state->hw.active)
592 		return 0;
593 
594 	/*
595 	 * Watermark/ddb requirement highly depends upon width of the
596 	 * framebuffer, So instead of allocating DDB equally among pipes
597 	 * distribute DDB based on resolution/width of the display.
598 	 */
599 	drm_mode_get_hv_timing(pipe_mode, &hdisplay, &vdisplay);
600 
601 	return hdisplay;
602 }
603 
604 static void intel_crtc_dbuf_weights(const struct intel_dbuf_state *dbuf_state,
605 				    enum pipe for_pipe,
606 				    unsigned int *weight_start,
607 				    unsigned int *weight_end,
608 				    unsigned int *weight_total)
609 {
610 	struct drm_i915_private *i915 =
611 		to_i915(dbuf_state->base.state->base.dev);
612 	enum pipe pipe;
613 
614 	*weight_start = 0;
615 	*weight_end = 0;
616 	*weight_total = 0;
617 
618 	for_each_pipe(i915, pipe) {
619 		int weight = dbuf_state->weight[pipe];
620 
621 		/*
622 		 * Do not account pipes using other slice sets
623 		 * luckily as of current BSpec slice sets do not partially
624 		 * intersect(pipes share either same one slice or same slice set
625 		 * i.e no partial intersection), so it is enough to check for
626 		 * equality for now.
627 		 */
628 		if (dbuf_state->slices[pipe] != dbuf_state->slices[for_pipe])
629 			continue;
630 
631 		*weight_total += weight;
632 		if (pipe < for_pipe) {
633 			*weight_start += weight;
634 			*weight_end += weight;
635 		} else if (pipe == for_pipe) {
636 			*weight_end += weight;
637 		}
638 	}
639 }
640 
641 static int
642 skl_crtc_allocate_ddb(struct intel_atomic_state *state, struct intel_crtc *crtc)
643 {
644 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
645 	unsigned int weight_total, weight_start, weight_end;
646 	const struct intel_dbuf_state *old_dbuf_state =
647 		intel_atomic_get_old_dbuf_state(state);
648 	struct intel_dbuf_state *new_dbuf_state =
649 		intel_atomic_get_new_dbuf_state(state);
650 	struct intel_crtc_state *crtc_state;
651 	struct skl_ddb_entry ddb_slices;
652 	enum pipe pipe = crtc->pipe;
653 	unsigned int mbus_offset = 0;
654 	u32 ddb_range_size;
655 	u32 dbuf_slice_mask;
656 	u32 start, end;
657 	int ret;
658 
659 	if (new_dbuf_state->weight[pipe] == 0) {
660 		skl_ddb_entry_init(&new_dbuf_state->ddb[pipe], 0, 0);
661 		goto out;
662 	}
663 
664 	dbuf_slice_mask = new_dbuf_state->slices[pipe];
665 
666 	skl_ddb_entry_for_slices(i915, dbuf_slice_mask, &ddb_slices);
667 	mbus_offset = mbus_ddb_offset(i915, dbuf_slice_mask);
668 	ddb_range_size = skl_ddb_entry_size(&ddb_slices);
669 
670 	intel_crtc_dbuf_weights(new_dbuf_state, pipe,
671 				&weight_start, &weight_end, &weight_total);
672 
673 	start = ddb_range_size * weight_start / weight_total;
674 	end = ddb_range_size * weight_end / weight_total;
675 
676 	skl_ddb_entry_init(&new_dbuf_state->ddb[pipe],
677 			   ddb_slices.start - mbus_offset + start,
678 			   ddb_slices.start - mbus_offset + end);
679 
680 out:
681 	if (old_dbuf_state->slices[pipe] == new_dbuf_state->slices[pipe] &&
682 	    skl_ddb_entry_equal(&old_dbuf_state->ddb[pipe],
683 				&new_dbuf_state->ddb[pipe]))
684 		return 0;
685 
686 	ret = intel_atomic_lock_global_state(&new_dbuf_state->base);
687 	if (ret)
688 		return ret;
689 
690 	crtc_state = intel_atomic_get_crtc_state(&state->base, crtc);
691 	if (IS_ERR(crtc_state))
692 		return PTR_ERR(crtc_state);
693 
694 	/*
695 	 * Used for checking overlaps, so we need absolute
696 	 * offsets instead of MBUS relative offsets.
697 	 */
698 	crtc_state->wm.skl.ddb.start = mbus_offset + new_dbuf_state->ddb[pipe].start;
699 	crtc_state->wm.skl.ddb.end = mbus_offset + new_dbuf_state->ddb[pipe].end;
700 
701 	drm_dbg_kms(&i915->drm,
702 		    "[CRTC:%d:%s] dbuf slices 0x%x -> 0x%x, ddb (%d - %d) -> (%d - %d), active pipes 0x%x -> 0x%x\n",
703 		    crtc->base.base.id, crtc->base.name,
704 		    old_dbuf_state->slices[pipe], new_dbuf_state->slices[pipe],
705 		    old_dbuf_state->ddb[pipe].start, old_dbuf_state->ddb[pipe].end,
706 		    new_dbuf_state->ddb[pipe].start, new_dbuf_state->ddb[pipe].end,
707 		    old_dbuf_state->active_pipes, new_dbuf_state->active_pipes);
708 
709 	return 0;
710 }
711 
712 static int skl_compute_wm_params(const struct intel_crtc_state *crtc_state,
713 				 int width, const struct drm_format_info *format,
714 				 u64 modifier, unsigned int rotation,
715 				 u32 plane_pixel_rate, struct skl_wm_params *wp,
716 				 int color_plane);
717 
718 static void skl_compute_plane_wm(const struct intel_crtc_state *crtc_state,
719 				 struct intel_plane *plane,
720 				 int level,
721 				 unsigned int latency,
722 				 const struct skl_wm_params *wp,
723 				 const struct skl_wm_level *result_prev,
724 				 struct skl_wm_level *result /* out */);
725 
726 static unsigned int skl_wm_latency(struct drm_i915_private *i915, int level,
727 				   const struct skl_wm_params *wp)
728 {
729 	unsigned int latency = i915->display.wm.skl_latency[level];
730 
731 	if (latency == 0)
732 		return 0;
733 
734 	/*
735 	 * WaIncreaseLatencyIPCEnabled: kbl,cfl
736 	 * Display WA #1141: kbl,cfl
737 	 */
738 	if ((IS_KABYLAKE(i915) || IS_COFFEELAKE(i915) || IS_COMETLAKE(i915)) &&
739 	    skl_watermark_ipc_enabled(i915))
740 		latency += 4;
741 
742 	if (skl_needs_memory_bw_wa(i915) && wp && wp->x_tiled)
743 		latency += 15;
744 
745 	return latency;
746 }
747 
748 static unsigned int
749 skl_cursor_allocation(const struct intel_crtc_state *crtc_state,
750 		      int num_active)
751 {
752 	struct intel_plane *plane = to_intel_plane(crtc_state->uapi.crtc->cursor);
753 	struct drm_i915_private *i915 = to_i915(crtc_state->uapi.crtc->dev);
754 	struct skl_wm_level wm = {};
755 	int ret, min_ddb_alloc = 0;
756 	struct skl_wm_params wp;
757 	int level;
758 
759 	ret = skl_compute_wm_params(crtc_state, 256,
760 				    drm_format_info(DRM_FORMAT_ARGB8888),
761 				    DRM_FORMAT_MOD_LINEAR,
762 				    DRM_MODE_ROTATE_0,
763 				    crtc_state->pixel_rate, &wp, 0);
764 	drm_WARN_ON(&i915->drm, ret);
765 
766 	for (level = 0; level < i915->display.wm.num_levels; level++) {
767 		unsigned int latency = skl_wm_latency(i915, level, &wp);
768 
769 		skl_compute_plane_wm(crtc_state, plane, level, latency, &wp, &wm, &wm);
770 		if (wm.min_ddb_alloc == U16_MAX)
771 			break;
772 
773 		min_ddb_alloc = wm.min_ddb_alloc;
774 	}
775 
776 	return max(num_active == 1 ? 32 : 8, min_ddb_alloc);
777 }
778 
779 static void skl_ddb_entry_init_from_hw(struct skl_ddb_entry *entry, u32 reg)
780 {
781 	skl_ddb_entry_init(entry,
782 			   REG_FIELD_GET(PLANE_BUF_START_MASK, reg),
783 			   REG_FIELD_GET(PLANE_BUF_END_MASK, reg));
784 	if (entry->end)
785 		entry->end++;
786 }
787 
788 static void
789 skl_ddb_get_hw_plane_state(struct drm_i915_private *i915,
790 			   const enum pipe pipe,
791 			   const enum plane_id plane_id,
792 			   struct skl_ddb_entry *ddb,
793 			   struct skl_ddb_entry *ddb_y)
794 {
795 	u32 val;
796 
797 	/* Cursor doesn't support NV12/planar, so no extra calculation needed */
798 	if (plane_id == PLANE_CURSOR) {
799 		val = intel_de_read(i915, CUR_BUF_CFG(pipe));
800 		skl_ddb_entry_init_from_hw(ddb, val);
801 		return;
802 	}
803 
804 	val = intel_de_read(i915, PLANE_BUF_CFG(pipe, plane_id));
805 	skl_ddb_entry_init_from_hw(ddb, val);
806 
807 	if (DISPLAY_VER(i915) >= 11)
808 		return;
809 
810 	val = intel_de_read(i915, PLANE_NV12_BUF_CFG(pipe, plane_id));
811 	skl_ddb_entry_init_from_hw(ddb_y, val);
812 }
813 
814 static void skl_pipe_ddb_get_hw_state(struct intel_crtc *crtc,
815 				      struct skl_ddb_entry *ddb,
816 				      struct skl_ddb_entry *ddb_y)
817 {
818 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
819 	enum intel_display_power_domain power_domain;
820 	enum pipe pipe = crtc->pipe;
821 	intel_wakeref_t wakeref;
822 	enum plane_id plane_id;
823 
824 	power_domain = POWER_DOMAIN_PIPE(pipe);
825 	wakeref = intel_display_power_get_if_enabled(i915, power_domain);
826 	if (!wakeref)
827 		return;
828 
829 	for_each_plane_id_on_crtc(crtc, plane_id)
830 		skl_ddb_get_hw_plane_state(i915, pipe,
831 					   plane_id,
832 					   &ddb[plane_id],
833 					   &ddb_y[plane_id]);
834 
835 	intel_display_power_put(i915, power_domain, wakeref);
836 }
837 
838 struct dbuf_slice_conf_entry {
839 	u8 active_pipes;
840 	u8 dbuf_mask[I915_MAX_PIPES];
841 	bool join_mbus;
842 };
843 
844 /*
845  * Table taken from Bspec 12716
846  * Pipes do have some preferred DBuf slice affinity,
847  * plus there are some hardcoded requirements on how
848  * those should be distributed for multipipe scenarios.
849  * For more DBuf slices algorithm can get even more messy
850  * and less readable, so decided to use a table almost
851  * as is from BSpec itself - that way it is at least easier
852  * to compare, change and check.
853  */
854 static const struct dbuf_slice_conf_entry icl_allowed_dbufs[] =
855 /* Autogenerated with igt/tools/intel_dbuf_map tool: */
856 {
857 	{
858 		.active_pipes = BIT(PIPE_A),
859 		.dbuf_mask = {
860 			[PIPE_A] = BIT(DBUF_S1),
861 		},
862 	},
863 	{
864 		.active_pipes = BIT(PIPE_B),
865 		.dbuf_mask = {
866 			[PIPE_B] = BIT(DBUF_S1),
867 		},
868 	},
869 	{
870 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B),
871 		.dbuf_mask = {
872 			[PIPE_A] = BIT(DBUF_S1),
873 			[PIPE_B] = BIT(DBUF_S2),
874 		},
875 	},
876 	{
877 		.active_pipes = BIT(PIPE_C),
878 		.dbuf_mask = {
879 			[PIPE_C] = BIT(DBUF_S2),
880 		},
881 	},
882 	{
883 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C),
884 		.dbuf_mask = {
885 			[PIPE_A] = BIT(DBUF_S1),
886 			[PIPE_C] = BIT(DBUF_S2),
887 		},
888 	},
889 	{
890 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C),
891 		.dbuf_mask = {
892 			[PIPE_B] = BIT(DBUF_S1),
893 			[PIPE_C] = BIT(DBUF_S2),
894 		},
895 	},
896 	{
897 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C),
898 		.dbuf_mask = {
899 			[PIPE_A] = BIT(DBUF_S1),
900 			[PIPE_B] = BIT(DBUF_S1),
901 			[PIPE_C] = BIT(DBUF_S2),
902 		},
903 	},
904 	{}
905 };
906 
907 /*
908  * Table taken from Bspec 49255
909  * Pipes do have some preferred DBuf slice affinity,
910  * plus there are some hardcoded requirements on how
911  * those should be distributed for multipipe scenarios.
912  * For more DBuf slices algorithm can get even more messy
913  * and less readable, so decided to use a table almost
914  * as is from BSpec itself - that way it is at least easier
915  * to compare, change and check.
916  */
917 static const struct dbuf_slice_conf_entry tgl_allowed_dbufs[] =
918 /* Autogenerated with igt/tools/intel_dbuf_map tool: */
919 {
920 	{
921 		.active_pipes = BIT(PIPE_A),
922 		.dbuf_mask = {
923 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
924 		},
925 	},
926 	{
927 		.active_pipes = BIT(PIPE_B),
928 		.dbuf_mask = {
929 			[PIPE_B] = BIT(DBUF_S1) | BIT(DBUF_S2),
930 		},
931 	},
932 	{
933 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B),
934 		.dbuf_mask = {
935 			[PIPE_A] = BIT(DBUF_S2),
936 			[PIPE_B] = BIT(DBUF_S1),
937 		},
938 	},
939 	{
940 		.active_pipes = BIT(PIPE_C),
941 		.dbuf_mask = {
942 			[PIPE_C] = BIT(DBUF_S2) | BIT(DBUF_S1),
943 		},
944 	},
945 	{
946 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C),
947 		.dbuf_mask = {
948 			[PIPE_A] = BIT(DBUF_S1),
949 			[PIPE_C] = BIT(DBUF_S2),
950 		},
951 	},
952 	{
953 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C),
954 		.dbuf_mask = {
955 			[PIPE_B] = BIT(DBUF_S1),
956 			[PIPE_C] = BIT(DBUF_S2),
957 		},
958 	},
959 	{
960 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C),
961 		.dbuf_mask = {
962 			[PIPE_A] = BIT(DBUF_S1),
963 			[PIPE_B] = BIT(DBUF_S1),
964 			[PIPE_C] = BIT(DBUF_S2),
965 		},
966 	},
967 	{
968 		.active_pipes = BIT(PIPE_D),
969 		.dbuf_mask = {
970 			[PIPE_D] = BIT(DBUF_S2) | BIT(DBUF_S1),
971 		},
972 	},
973 	{
974 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_D),
975 		.dbuf_mask = {
976 			[PIPE_A] = BIT(DBUF_S1),
977 			[PIPE_D] = BIT(DBUF_S2),
978 		},
979 	},
980 	{
981 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_D),
982 		.dbuf_mask = {
983 			[PIPE_B] = BIT(DBUF_S1),
984 			[PIPE_D] = BIT(DBUF_S2),
985 		},
986 	},
987 	{
988 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_D),
989 		.dbuf_mask = {
990 			[PIPE_A] = BIT(DBUF_S1),
991 			[PIPE_B] = BIT(DBUF_S1),
992 			[PIPE_D] = BIT(DBUF_S2),
993 		},
994 	},
995 	{
996 		.active_pipes = BIT(PIPE_C) | BIT(PIPE_D),
997 		.dbuf_mask = {
998 			[PIPE_C] = BIT(DBUF_S1),
999 			[PIPE_D] = BIT(DBUF_S2),
1000 		},
1001 	},
1002 	{
1003 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C) | BIT(PIPE_D),
1004 		.dbuf_mask = {
1005 			[PIPE_A] = BIT(DBUF_S1),
1006 			[PIPE_C] = BIT(DBUF_S2),
1007 			[PIPE_D] = BIT(DBUF_S2),
1008 		},
1009 	},
1010 	{
1011 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
1012 		.dbuf_mask = {
1013 			[PIPE_B] = BIT(DBUF_S1),
1014 			[PIPE_C] = BIT(DBUF_S2),
1015 			[PIPE_D] = BIT(DBUF_S2),
1016 		},
1017 	},
1018 	{
1019 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
1020 		.dbuf_mask = {
1021 			[PIPE_A] = BIT(DBUF_S1),
1022 			[PIPE_B] = BIT(DBUF_S1),
1023 			[PIPE_C] = BIT(DBUF_S2),
1024 			[PIPE_D] = BIT(DBUF_S2),
1025 		},
1026 	},
1027 	{}
1028 };
1029 
1030 static const struct dbuf_slice_conf_entry dg2_allowed_dbufs[] = {
1031 	{
1032 		.active_pipes = BIT(PIPE_A),
1033 		.dbuf_mask = {
1034 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1035 		},
1036 	},
1037 	{
1038 		.active_pipes = BIT(PIPE_B),
1039 		.dbuf_mask = {
1040 			[PIPE_B] = BIT(DBUF_S1) | BIT(DBUF_S2),
1041 		},
1042 	},
1043 	{
1044 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B),
1045 		.dbuf_mask = {
1046 			[PIPE_A] = BIT(DBUF_S1),
1047 			[PIPE_B] = BIT(DBUF_S2),
1048 		},
1049 	},
1050 	{
1051 		.active_pipes = BIT(PIPE_C),
1052 		.dbuf_mask = {
1053 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1054 		},
1055 	},
1056 	{
1057 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C),
1058 		.dbuf_mask = {
1059 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1060 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1061 		},
1062 	},
1063 	{
1064 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C),
1065 		.dbuf_mask = {
1066 			[PIPE_B] = BIT(DBUF_S1) | BIT(DBUF_S2),
1067 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1068 		},
1069 	},
1070 	{
1071 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C),
1072 		.dbuf_mask = {
1073 			[PIPE_A] = BIT(DBUF_S1),
1074 			[PIPE_B] = BIT(DBUF_S2),
1075 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1076 		},
1077 	},
1078 	{
1079 		.active_pipes = BIT(PIPE_D),
1080 		.dbuf_mask = {
1081 			[PIPE_D] = BIT(DBUF_S3) | BIT(DBUF_S4),
1082 		},
1083 	},
1084 	{
1085 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_D),
1086 		.dbuf_mask = {
1087 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1088 			[PIPE_D] = BIT(DBUF_S3) | BIT(DBUF_S4),
1089 		},
1090 	},
1091 	{
1092 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_D),
1093 		.dbuf_mask = {
1094 			[PIPE_B] = BIT(DBUF_S1) | BIT(DBUF_S2),
1095 			[PIPE_D] = BIT(DBUF_S3) | BIT(DBUF_S4),
1096 		},
1097 	},
1098 	{
1099 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_D),
1100 		.dbuf_mask = {
1101 			[PIPE_A] = BIT(DBUF_S1),
1102 			[PIPE_B] = BIT(DBUF_S2),
1103 			[PIPE_D] = BIT(DBUF_S3) | BIT(DBUF_S4),
1104 		},
1105 	},
1106 	{
1107 		.active_pipes = BIT(PIPE_C) | BIT(PIPE_D),
1108 		.dbuf_mask = {
1109 			[PIPE_C] = BIT(DBUF_S3),
1110 			[PIPE_D] = BIT(DBUF_S4),
1111 		},
1112 	},
1113 	{
1114 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C) | BIT(PIPE_D),
1115 		.dbuf_mask = {
1116 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1117 			[PIPE_C] = BIT(DBUF_S3),
1118 			[PIPE_D] = BIT(DBUF_S4),
1119 		},
1120 	},
1121 	{
1122 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
1123 		.dbuf_mask = {
1124 			[PIPE_B] = BIT(DBUF_S1) | BIT(DBUF_S2),
1125 			[PIPE_C] = BIT(DBUF_S3),
1126 			[PIPE_D] = BIT(DBUF_S4),
1127 		},
1128 	},
1129 	{
1130 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
1131 		.dbuf_mask = {
1132 			[PIPE_A] = BIT(DBUF_S1),
1133 			[PIPE_B] = BIT(DBUF_S2),
1134 			[PIPE_C] = BIT(DBUF_S3),
1135 			[PIPE_D] = BIT(DBUF_S4),
1136 		},
1137 	},
1138 	{}
1139 };
1140 
1141 static const struct dbuf_slice_conf_entry adlp_allowed_dbufs[] = {
1142 	/*
1143 	 * Keep the join_mbus cases first so check_mbus_joined()
1144 	 * will prefer them over the !join_mbus cases.
1145 	 */
1146 	{
1147 		.active_pipes = BIT(PIPE_A),
1148 		.dbuf_mask = {
1149 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2) | BIT(DBUF_S3) | BIT(DBUF_S4),
1150 		},
1151 		.join_mbus = true,
1152 	},
1153 	{
1154 		.active_pipes = BIT(PIPE_B),
1155 		.dbuf_mask = {
1156 			[PIPE_B] = BIT(DBUF_S1) | BIT(DBUF_S2) | BIT(DBUF_S3) | BIT(DBUF_S4),
1157 		},
1158 		.join_mbus = true,
1159 	},
1160 	{
1161 		.active_pipes = BIT(PIPE_A),
1162 		.dbuf_mask = {
1163 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1164 		},
1165 		.join_mbus = false,
1166 	},
1167 	{
1168 		.active_pipes = BIT(PIPE_B),
1169 		.dbuf_mask = {
1170 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1171 		},
1172 		.join_mbus = false,
1173 	},
1174 	{
1175 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B),
1176 		.dbuf_mask = {
1177 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1178 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1179 		},
1180 	},
1181 	{
1182 		.active_pipes = BIT(PIPE_C),
1183 		.dbuf_mask = {
1184 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1185 		},
1186 	},
1187 	{
1188 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C),
1189 		.dbuf_mask = {
1190 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1191 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1192 		},
1193 	},
1194 	{
1195 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C),
1196 		.dbuf_mask = {
1197 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1198 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1199 		},
1200 	},
1201 	{
1202 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C),
1203 		.dbuf_mask = {
1204 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1205 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1206 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1207 		},
1208 	},
1209 	{
1210 		.active_pipes = BIT(PIPE_D),
1211 		.dbuf_mask = {
1212 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1213 		},
1214 	},
1215 	{
1216 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_D),
1217 		.dbuf_mask = {
1218 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1219 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1220 		},
1221 	},
1222 	{
1223 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_D),
1224 		.dbuf_mask = {
1225 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1226 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1227 		},
1228 	},
1229 	{
1230 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_D),
1231 		.dbuf_mask = {
1232 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1233 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1234 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1235 		},
1236 	},
1237 	{
1238 		.active_pipes = BIT(PIPE_C) | BIT(PIPE_D),
1239 		.dbuf_mask = {
1240 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1241 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1242 		},
1243 	},
1244 	{
1245 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_C) | BIT(PIPE_D),
1246 		.dbuf_mask = {
1247 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1248 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1249 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1250 		},
1251 	},
1252 	{
1253 		.active_pipes = BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
1254 		.dbuf_mask = {
1255 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1256 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1257 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1258 		},
1259 	},
1260 	{
1261 		.active_pipes = BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
1262 		.dbuf_mask = {
1263 			[PIPE_A] = BIT(DBUF_S1) | BIT(DBUF_S2),
1264 			[PIPE_B] = BIT(DBUF_S3) | BIT(DBUF_S4),
1265 			[PIPE_C] = BIT(DBUF_S3) | BIT(DBUF_S4),
1266 			[PIPE_D] = BIT(DBUF_S1) | BIT(DBUF_S2),
1267 		},
1268 	},
1269 	{}
1270 
1271 };
1272 
1273 static bool check_mbus_joined(u8 active_pipes,
1274 			      const struct dbuf_slice_conf_entry *dbuf_slices)
1275 {
1276 	int i;
1277 
1278 	for (i = 0; dbuf_slices[i].active_pipes != 0; i++) {
1279 		if (dbuf_slices[i].active_pipes == active_pipes)
1280 			return dbuf_slices[i].join_mbus;
1281 	}
1282 	return false;
1283 }
1284 
1285 static bool adlp_check_mbus_joined(u8 active_pipes)
1286 {
1287 	return check_mbus_joined(active_pipes, adlp_allowed_dbufs);
1288 }
1289 
1290 static u8 compute_dbuf_slices(enum pipe pipe, u8 active_pipes, bool join_mbus,
1291 			      const struct dbuf_slice_conf_entry *dbuf_slices)
1292 {
1293 	int i;
1294 
1295 	for (i = 0; dbuf_slices[i].active_pipes != 0; i++) {
1296 		if (dbuf_slices[i].active_pipes == active_pipes &&
1297 		    dbuf_slices[i].join_mbus == join_mbus)
1298 			return dbuf_slices[i].dbuf_mask[pipe];
1299 	}
1300 	return 0;
1301 }
1302 
1303 /*
1304  * This function finds an entry with same enabled pipe configuration and
1305  * returns correspondent DBuf slice mask as stated in BSpec for particular
1306  * platform.
1307  */
1308 static u8 icl_compute_dbuf_slices(enum pipe pipe, u8 active_pipes, bool join_mbus)
1309 {
1310 	/*
1311 	 * FIXME: For ICL this is still a bit unclear as prev BSpec revision
1312 	 * required calculating "pipe ratio" in order to determine
1313 	 * if one or two slices can be used for single pipe configurations
1314 	 * as additional constraint to the existing table.
1315 	 * However based on recent info, it should be not "pipe ratio"
1316 	 * but rather ratio between pixel_rate and cdclk with additional
1317 	 * constants, so for now we are using only table until this is
1318 	 * clarified. Also this is the reason why crtc_state param is
1319 	 * still here - we will need it once those additional constraints
1320 	 * pop up.
1321 	 */
1322 	return compute_dbuf_slices(pipe, active_pipes, join_mbus,
1323 				   icl_allowed_dbufs);
1324 }
1325 
1326 static u8 tgl_compute_dbuf_slices(enum pipe pipe, u8 active_pipes, bool join_mbus)
1327 {
1328 	return compute_dbuf_slices(pipe, active_pipes, join_mbus,
1329 				   tgl_allowed_dbufs);
1330 }
1331 
1332 static u8 adlp_compute_dbuf_slices(enum pipe pipe, u8 active_pipes, bool join_mbus)
1333 {
1334 	return compute_dbuf_slices(pipe, active_pipes, join_mbus,
1335 				   adlp_allowed_dbufs);
1336 }
1337 
1338 static u8 dg2_compute_dbuf_slices(enum pipe pipe, u8 active_pipes, bool join_mbus)
1339 {
1340 	return compute_dbuf_slices(pipe, active_pipes, join_mbus,
1341 				   dg2_allowed_dbufs);
1342 }
1343 
1344 static u8 skl_compute_dbuf_slices(struct intel_crtc *crtc, u8 active_pipes, bool join_mbus)
1345 {
1346 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
1347 	enum pipe pipe = crtc->pipe;
1348 
1349 	if (IS_DG2(i915))
1350 		return dg2_compute_dbuf_slices(pipe, active_pipes, join_mbus);
1351 	else if (DISPLAY_VER(i915) >= 13)
1352 		return adlp_compute_dbuf_slices(pipe, active_pipes, join_mbus);
1353 	else if (DISPLAY_VER(i915) == 12)
1354 		return tgl_compute_dbuf_slices(pipe, active_pipes, join_mbus);
1355 	else if (DISPLAY_VER(i915) == 11)
1356 		return icl_compute_dbuf_slices(pipe, active_pipes, join_mbus);
1357 	/*
1358 	 * For anything else just return one slice yet.
1359 	 * Should be extended for other platforms.
1360 	 */
1361 	return active_pipes & BIT(pipe) ? BIT(DBUF_S1) : 0;
1362 }
1363 
1364 static bool
1365 use_minimal_wm0_only(const struct intel_crtc_state *crtc_state,
1366 		     struct intel_plane *plane)
1367 {
1368 	struct drm_i915_private *i915 = to_i915(plane->base.dev);
1369 
1370 	return DISPLAY_VER(i915) >= 13 &&
1371 	       crtc_state->uapi.async_flip &&
1372 	       plane->async_flip;
1373 }
1374 
1375 static u64
1376 skl_total_relative_data_rate(const struct intel_crtc_state *crtc_state)
1377 {
1378 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
1379 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
1380 	enum plane_id plane_id;
1381 	u64 data_rate = 0;
1382 
1383 	for_each_plane_id_on_crtc(crtc, plane_id) {
1384 		if (plane_id == PLANE_CURSOR)
1385 			continue;
1386 
1387 		data_rate += crtc_state->rel_data_rate[plane_id];
1388 
1389 		if (DISPLAY_VER(i915) < 11)
1390 			data_rate += crtc_state->rel_data_rate_y[plane_id];
1391 	}
1392 
1393 	return data_rate;
1394 }
1395 
1396 static const struct skl_wm_level *
1397 skl_plane_wm_level(const struct skl_pipe_wm *pipe_wm,
1398 		   enum plane_id plane_id,
1399 		   int level)
1400 {
1401 	const struct skl_plane_wm *wm = &pipe_wm->planes[plane_id];
1402 
1403 	if (level == 0 && pipe_wm->use_sagv_wm)
1404 		return &wm->sagv.wm0;
1405 
1406 	return &wm->wm[level];
1407 }
1408 
1409 static const struct skl_wm_level *
1410 skl_plane_trans_wm(const struct skl_pipe_wm *pipe_wm,
1411 		   enum plane_id plane_id)
1412 {
1413 	const struct skl_plane_wm *wm = &pipe_wm->planes[plane_id];
1414 
1415 	if (pipe_wm->use_sagv_wm)
1416 		return &wm->sagv.trans_wm;
1417 
1418 	return &wm->trans_wm;
1419 }
1420 
1421 /*
1422  * We only disable the watermarks for each plane if
1423  * they exceed the ddb allocation of said plane. This
1424  * is done so that we don't end up touching cursor
1425  * watermarks needlessly when some other plane reduces
1426  * our max possible watermark level.
1427  *
1428  * Bspec has this to say about the PLANE_WM enable bit:
1429  * "All the watermarks at this level for all enabled
1430  *  planes must be enabled before the level will be used."
1431  * So this is actually safe to do.
1432  */
1433 static void
1434 skl_check_wm_level(struct skl_wm_level *wm, const struct skl_ddb_entry *ddb)
1435 {
1436 	if (wm->min_ddb_alloc > skl_ddb_entry_size(ddb))
1437 		memset(wm, 0, sizeof(*wm));
1438 }
1439 
1440 static void
1441 skl_check_nv12_wm_level(struct skl_wm_level *wm, struct skl_wm_level *uv_wm,
1442 			const struct skl_ddb_entry *ddb_y, const struct skl_ddb_entry *ddb)
1443 {
1444 	if (wm->min_ddb_alloc > skl_ddb_entry_size(ddb_y) ||
1445 	    uv_wm->min_ddb_alloc > skl_ddb_entry_size(ddb)) {
1446 		memset(wm, 0, sizeof(*wm));
1447 		memset(uv_wm, 0, sizeof(*uv_wm));
1448 	}
1449 }
1450 
1451 static bool skl_need_wm_copy_wa(struct drm_i915_private *i915, int level,
1452 				const struct skl_plane_wm *wm)
1453 {
1454 	/*
1455 	 * Wa_1408961008:icl, ehl
1456 	 * Wa_14012656716:tgl, adl
1457 	 * Wa_14017887344:icl
1458 	 * Wa_14017868169:adl, tgl
1459 	 * Due to some power saving optimizations, different subsystems
1460 	 * like PSR, might still use even disabled wm level registers,
1461 	 * for "reference", so lets keep at least the values sane.
1462 	 * Considering amount of WA requiring us to do similar things, was
1463 	 * decided to simply do it for all of the platforms, as those wm
1464 	 * levels are disabled, this isn't going to do harm anyway.
1465 	 */
1466 	return level > 0 && !wm->wm[level].enable;
1467 }
1468 
1469 struct skl_plane_ddb_iter {
1470 	u64 data_rate;
1471 	u16 start, size;
1472 };
1473 
1474 static void
1475 skl_allocate_plane_ddb(struct skl_plane_ddb_iter *iter,
1476 		       struct skl_ddb_entry *ddb,
1477 		       const struct skl_wm_level *wm,
1478 		       u64 data_rate)
1479 {
1480 	u16 size, extra = 0;
1481 
1482 	if (data_rate) {
1483 		extra = min_t(u16, iter->size,
1484 			      DIV64_U64_ROUND_UP(iter->size * data_rate,
1485 						 iter->data_rate));
1486 		iter->size -= extra;
1487 		iter->data_rate -= data_rate;
1488 	}
1489 
1490 	/*
1491 	 * Keep ddb entry of all disabled planes explicitly zeroed
1492 	 * to avoid skl_ddb_add_affected_planes() adding them to
1493 	 * the state when other planes change their allocations.
1494 	 */
1495 	size = wm->min_ddb_alloc + extra;
1496 	if (size)
1497 		iter->start = skl_ddb_entry_init(ddb, iter->start,
1498 						 iter->start + size);
1499 }
1500 
1501 static int
1502 skl_crtc_allocate_plane_ddb(struct intel_atomic_state *state,
1503 			    struct intel_crtc *crtc)
1504 {
1505 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
1506 	struct intel_crtc_state *crtc_state =
1507 		intel_atomic_get_new_crtc_state(state, crtc);
1508 	const struct intel_dbuf_state *dbuf_state =
1509 		intel_atomic_get_new_dbuf_state(state);
1510 	const struct skl_ddb_entry *alloc = &dbuf_state->ddb[crtc->pipe];
1511 	int num_active = hweight8(dbuf_state->active_pipes);
1512 	struct skl_plane_ddb_iter iter;
1513 	enum plane_id plane_id;
1514 	u16 cursor_size;
1515 	u32 blocks;
1516 	int level;
1517 
1518 	/* Clear the partitioning for disabled planes. */
1519 	memset(crtc_state->wm.skl.plane_ddb, 0, sizeof(crtc_state->wm.skl.plane_ddb));
1520 	memset(crtc_state->wm.skl.plane_ddb_y, 0, sizeof(crtc_state->wm.skl.plane_ddb_y));
1521 
1522 	if (!crtc_state->hw.active)
1523 		return 0;
1524 
1525 	iter.start = alloc->start;
1526 	iter.size = skl_ddb_entry_size(alloc);
1527 	if (iter.size == 0)
1528 		return 0;
1529 
1530 	/* Allocate fixed number of blocks for cursor. */
1531 	cursor_size = skl_cursor_allocation(crtc_state, num_active);
1532 	iter.size -= cursor_size;
1533 	skl_ddb_entry_init(&crtc_state->wm.skl.plane_ddb[PLANE_CURSOR],
1534 			   alloc->end - cursor_size, alloc->end);
1535 
1536 	iter.data_rate = skl_total_relative_data_rate(crtc_state);
1537 
1538 	/*
1539 	 * Find the highest watermark level for which we can satisfy the block
1540 	 * requirement of active planes.
1541 	 */
1542 	for (level = i915->display.wm.num_levels - 1; level >= 0; level--) {
1543 		blocks = 0;
1544 		for_each_plane_id_on_crtc(crtc, plane_id) {
1545 			const struct skl_plane_wm *wm =
1546 				&crtc_state->wm.skl.optimal.planes[plane_id];
1547 
1548 			if (plane_id == PLANE_CURSOR) {
1549 				const struct skl_ddb_entry *ddb =
1550 					&crtc_state->wm.skl.plane_ddb[plane_id];
1551 
1552 				if (wm->wm[level].min_ddb_alloc > skl_ddb_entry_size(ddb)) {
1553 					drm_WARN_ON(&i915->drm,
1554 						    wm->wm[level].min_ddb_alloc != U16_MAX);
1555 					blocks = U32_MAX;
1556 					break;
1557 				}
1558 				continue;
1559 			}
1560 
1561 			blocks += wm->wm[level].min_ddb_alloc;
1562 			blocks += wm->uv_wm[level].min_ddb_alloc;
1563 		}
1564 
1565 		if (blocks <= iter.size) {
1566 			iter.size -= blocks;
1567 			break;
1568 		}
1569 	}
1570 
1571 	if (level < 0) {
1572 		drm_dbg_kms(&i915->drm,
1573 			    "Requested display configuration exceeds system DDB limitations");
1574 		drm_dbg_kms(&i915->drm, "minimum required %d/%d\n",
1575 			    blocks, iter.size);
1576 		return -EINVAL;
1577 	}
1578 
1579 	/* avoid the WARN later when we don't allocate any extra DDB */
1580 	if (iter.data_rate == 0)
1581 		iter.size = 0;
1582 
1583 	/*
1584 	 * Grant each plane the blocks it requires at the highest achievable
1585 	 * watermark level, plus an extra share of the leftover blocks
1586 	 * proportional to its relative data rate.
1587 	 */
1588 	for_each_plane_id_on_crtc(crtc, plane_id) {
1589 		struct skl_ddb_entry *ddb =
1590 			&crtc_state->wm.skl.plane_ddb[plane_id];
1591 		struct skl_ddb_entry *ddb_y =
1592 			&crtc_state->wm.skl.plane_ddb_y[plane_id];
1593 		const struct skl_plane_wm *wm =
1594 			&crtc_state->wm.skl.optimal.planes[plane_id];
1595 
1596 		if (plane_id == PLANE_CURSOR)
1597 			continue;
1598 
1599 		if (DISPLAY_VER(i915) < 11 &&
1600 		    crtc_state->nv12_planes & BIT(plane_id)) {
1601 			skl_allocate_plane_ddb(&iter, ddb_y, &wm->wm[level],
1602 					       crtc_state->rel_data_rate_y[plane_id]);
1603 			skl_allocate_plane_ddb(&iter, ddb, &wm->uv_wm[level],
1604 					       crtc_state->rel_data_rate[plane_id]);
1605 		} else {
1606 			skl_allocate_plane_ddb(&iter, ddb, &wm->wm[level],
1607 					       crtc_state->rel_data_rate[plane_id]);
1608 		}
1609 	}
1610 	drm_WARN_ON(&i915->drm, iter.size != 0 || iter.data_rate != 0);
1611 
1612 	/*
1613 	 * When we calculated watermark values we didn't know how high
1614 	 * of a level we'd actually be able to hit, so we just marked
1615 	 * all levels as "enabled."  Go back now and disable the ones
1616 	 * that aren't actually possible.
1617 	 */
1618 	for (level++; level < i915->display.wm.num_levels; level++) {
1619 		for_each_plane_id_on_crtc(crtc, plane_id) {
1620 			const struct skl_ddb_entry *ddb =
1621 				&crtc_state->wm.skl.plane_ddb[plane_id];
1622 			const struct skl_ddb_entry *ddb_y =
1623 				&crtc_state->wm.skl.plane_ddb_y[plane_id];
1624 			struct skl_plane_wm *wm =
1625 				&crtc_state->wm.skl.optimal.planes[plane_id];
1626 
1627 			if (DISPLAY_VER(i915) < 11 &&
1628 			    crtc_state->nv12_planes & BIT(plane_id))
1629 				skl_check_nv12_wm_level(&wm->wm[level],
1630 							&wm->uv_wm[level],
1631 							ddb_y, ddb);
1632 			else
1633 				skl_check_wm_level(&wm->wm[level], ddb);
1634 
1635 			if (skl_need_wm_copy_wa(i915, level, wm)) {
1636 				wm->wm[level].blocks = wm->wm[level - 1].blocks;
1637 				wm->wm[level].lines = wm->wm[level - 1].lines;
1638 				wm->wm[level].ignore_lines = wm->wm[level - 1].ignore_lines;
1639 			}
1640 		}
1641 	}
1642 
1643 	/*
1644 	 * Go back and disable the transition and SAGV watermarks
1645 	 * if it turns out we don't have enough DDB blocks for them.
1646 	 */
1647 	for_each_plane_id_on_crtc(crtc, plane_id) {
1648 		const struct skl_ddb_entry *ddb =
1649 			&crtc_state->wm.skl.plane_ddb[plane_id];
1650 		const struct skl_ddb_entry *ddb_y =
1651 			&crtc_state->wm.skl.plane_ddb_y[plane_id];
1652 		struct skl_plane_wm *wm =
1653 			&crtc_state->wm.skl.optimal.planes[plane_id];
1654 
1655 		if (DISPLAY_VER(i915) < 11 &&
1656 		    crtc_state->nv12_planes & BIT(plane_id)) {
1657 			skl_check_wm_level(&wm->trans_wm, ddb_y);
1658 		} else {
1659 			WARN_ON(skl_ddb_entry_size(ddb_y));
1660 
1661 			skl_check_wm_level(&wm->trans_wm, ddb);
1662 		}
1663 
1664 		skl_check_wm_level(&wm->sagv.wm0, ddb);
1665 		skl_check_wm_level(&wm->sagv.trans_wm, ddb);
1666 	}
1667 
1668 	return 0;
1669 }
1670 
1671 /*
1672  * The max latency should be 257 (max the punit can code is 255 and we add 2us
1673  * for the read latency) and cpp should always be <= 8, so that
1674  * should allow pixel_rate up to ~2 GHz which seems sufficient since max
1675  * 2xcdclk is 1350 MHz and the pixel rate should never exceed that.
1676  */
1677 static uint_fixed_16_16_t
1678 skl_wm_method1(const struct drm_i915_private *i915, u32 pixel_rate,
1679 	       u8 cpp, u32 latency, u32 dbuf_block_size)
1680 {
1681 	u32 wm_intermediate_val;
1682 	uint_fixed_16_16_t ret;
1683 
1684 	if (latency == 0)
1685 		return FP_16_16_MAX;
1686 
1687 	wm_intermediate_val = latency * pixel_rate * cpp;
1688 	ret = div_fixed16(wm_intermediate_val, 1000 * dbuf_block_size);
1689 
1690 	if (DISPLAY_VER(i915) >= 10)
1691 		ret = add_fixed16_u32(ret, 1);
1692 
1693 	return ret;
1694 }
1695 
1696 static uint_fixed_16_16_t
1697 skl_wm_method2(u32 pixel_rate, u32 pipe_htotal, u32 latency,
1698 	       uint_fixed_16_16_t plane_blocks_per_line)
1699 {
1700 	u32 wm_intermediate_val;
1701 	uint_fixed_16_16_t ret;
1702 
1703 	if (latency == 0)
1704 		return FP_16_16_MAX;
1705 
1706 	wm_intermediate_val = latency * pixel_rate;
1707 	wm_intermediate_val = DIV_ROUND_UP(wm_intermediate_val,
1708 					   pipe_htotal * 1000);
1709 	ret = mul_u32_fixed16(wm_intermediate_val, plane_blocks_per_line);
1710 	return ret;
1711 }
1712 
1713 static uint_fixed_16_16_t
1714 intel_get_linetime_us(const struct intel_crtc_state *crtc_state)
1715 {
1716 	struct drm_i915_private *i915 = to_i915(crtc_state->uapi.crtc->dev);
1717 	u32 pixel_rate;
1718 	u32 crtc_htotal;
1719 	uint_fixed_16_16_t linetime_us;
1720 
1721 	if (!crtc_state->hw.active)
1722 		return u32_to_fixed16(0);
1723 
1724 	pixel_rate = crtc_state->pixel_rate;
1725 
1726 	if (drm_WARN_ON(&i915->drm, pixel_rate == 0))
1727 		return u32_to_fixed16(0);
1728 
1729 	crtc_htotal = crtc_state->hw.pipe_mode.crtc_htotal;
1730 	linetime_us = div_fixed16(crtc_htotal * 1000, pixel_rate);
1731 
1732 	return linetime_us;
1733 }
1734 
1735 static int
1736 skl_compute_wm_params(const struct intel_crtc_state *crtc_state,
1737 		      int width, const struct drm_format_info *format,
1738 		      u64 modifier, unsigned int rotation,
1739 		      u32 plane_pixel_rate, struct skl_wm_params *wp,
1740 		      int color_plane)
1741 {
1742 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
1743 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
1744 	u32 interm_pbpl;
1745 
1746 	/* only planar format has two planes */
1747 	if (color_plane == 1 &&
1748 	    !intel_format_info_is_yuv_semiplanar(format, modifier)) {
1749 		drm_dbg_kms(&i915->drm,
1750 			    "Non planar format have single plane\n");
1751 		return -EINVAL;
1752 	}
1753 
1754 	wp->x_tiled = modifier == I915_FORMAT_MOD_X_TILED;
1755 	wp->y_tiled = modifier != I915_FORMAT_MOD_X_TILED &&
1756 		intel_fb_is_tiled_modifier(modifier);
1757 	wp->rc_surface = intel_fb_is_ccs_modifier(modifier);
1758 	wp->is_planar = intel_format_info_is_yuv_semiplanar(format, modifier);
1759 
1760 	wp->width = width;
1761 	if (color_plane == 1 && wp->is_planar)
1762 		wp->width /= 2;
1763 
1764 	wp->cpp = format->cpp[color_plane];
1765 	wp->plane_pixel_rate = plane_pixel_rate;
1766 
1767 	if (DISPLAY_VER(i915) >= 11 &&
1768 	    modifier == I915_FORMAT_MOD_Yf_TILED  && wp->cpp == 1)
1769 		wp->dbuf_block_size = 256;
1770 	else
1771 		wp->dbuf_block_size = 512;
1772 
1773 	if (drm_rotation_90_or_270(rotation)) {
1774 		switch (wp->cpp) {
1775 		case 1:
1776 			wp->y_min_scanlines = 16;
1777 			break;
1778 		case 2:
1779 			wp->y_min_scanlines = 8;
1780 			break;
1781 		case 4:
1782 			wp->y_min_scanlines = 4;
1783 			break;
1784 		default:
1785 			MISSING_CASE(wp->cpp);
1786 			return -EINVAL;
1787 		}
1788 	} else {
1789 		wp->y_min_scanlines = 4;
1790 	}
1791 
1792 	if (skl_needs_memory_bw_wa(i915))
1793 		wp->y_min_scanlines *= 2;
1794 
1795 	wp->plane_bytes_per_line = wp->width * wp->cpp;
1796 	if (wp->y_tiled) {
1797 		interm_pbpl = DIV_ROUND_UP(wp->plane_bytes_per_line *
1798 					   wp->y_min_scanlines,
1799 					   wp->dbuf_block_size);
1800 
1801 		if (DISPLAY_VER(i915) >= 10)
1802 			interm_pbpl++;
1803 
1804 		wp->plane_blocks_per_line = div_fixed16(interm_pbpl,
1805 							wp->y_min_scanlines);
1806 	} else {
1807 		interm_pbpl = DIV_ROUND_UP(wp->plane_bytes_per_line,
1808 					   wp->dbuf_block_size);
1809 
1810 		if (!wp->x_tiled || DISPLAY_VER(i915) >= 10)
1811 			interm_pbpl++;
1812 
1813 		wp->plane_blocks_per_line = u32_to_fixed16(interm_pbpl);
1814 	}
1815 
1816 	wp->y_tile_minimum = mul_u32_fixed16(wp->y_min_scanlines,
1817 					     wp->plane_blocks_per_line);
1818 
1819 	wp->linetime_us = fixed16_to_u32_round_up(intel_get_linetime_us(crtc_state));
1820 
1821 	return 0;
1822 }
1823 
1824 static int
1825 skl_compute_plane_wm_params(const struct intel_crtc_state *crtc_state,
1826 			    const struct intel_plane_state *plane_state,
1827 			    struct skl_wm_params *wp, int color_plane)
1828 {
1829 	const struct drm_framebuffer *fb = plane_state->hw.fb;
1830 	int width;
1831 
1832 	/*
1833 	 * Src coordinates are already rotated by 270 degrees for
1834 	 * the 90/270 degree plane rotation cases (to match the
1835 	 * GTT mapping), hence no need to account for rotation here.
1836 	 */
1837 	width = drm_rect_width(&plane_state->uapi.src) >> 16;
1838 
1839 	return skl_compute_wm_params(crtc_state, width,
1840 				     fb->format, fb->modifier,
1841 				     plane_state->hw.rotation,
1842 				     intel_plane_pixel_rate(crtc_state, plane_state),
1843 				     wp, color_plane);
1844 }
1845 
1846 static bool skl_wm_has_lines(struct drm_i915_private *i915, int level)
1847 {
1848 	if (DISPLAY_VER(i915) >= 10)
1849 		return true;
1850 
1851 	/* The number of lines are ignored for the level 0 watermark. */
1852 	return level > 0;
1853 }
1854 
1855 static int skl_wm_max_lines(struct drm_i915_private *i915)
1856 {
1857 	if (DISPLAY_VER(i915) >= 13)
1858 		return 255;
1859 	else
1860 		return 31;
1861 }
1862 
1863 static void skl_compute_plane_wm(const struct intel_crtc_state *crtc_state,
1864 				 struct intel_plane *plane,
1865 				 int level,
1866 				 unsigned int latency,
1867 				 const struct skl_wm_params *wp,
1868 				 const struct skl_wm_level *result_prev,
1869 				 struct skl_wm_level *result /* out */)
1870 {
1871 	struct drm_i915_private *i915 = to_i915(crtc_state->uapi.crtc->dev);
1872 	uint_fixed_16_16_t method1, method2;
1873 	uint_fixed_16_16_t selected_result;
1874 	u32 blocks, lines, min_ddb_alloc = 0;
1875 
1876 	if (latency == 0 ||
1877 	    (use_minimal_wm0_only(crtc_state, plane) && level > 0)) {
1878 		/* reject it */
1879 		result->min_ddb_alloc = U16_MAX;
1880 		return;
1881 	}
1882 
1883 	method1 = skl_wm_method1(i915, wp->plane_pixel_rate,
1884 				 wp->cpp, latency, wp->dbuf_block_size);
1885 	method2 = skl_wm_method2(wp->plane_pixel_rate,
1886 				 crtc_state->hw.pipe_mode.crtc_htotal,
1887 				 latency,
1888 				 wp->plane_blocks_per_line);
1889 
1890 	if (wp->y_tiled) {
1891 		selected_result = max_fixed16(method2, wp->y_tile_minimum);
1892 	} else {
1893 		if ((wp->cpp * crtc_state->hw.pipe_mode.crtc_htotal /
1894 		     wp->dbuf_block_size < 1) &&
1895 		     (wp->plane_bytes_per_line / wp->dbuf_block_size < 1)) {
1896 			selected_result = method2;
1897 		} else if (latency >= wp->linetime_us) {
1898 			if (DISPLAY_VER(i915) == 9)
1899 				selected_result = min_fixed16(method1, method2);
1900 			else
1901 				selected_result = method2;
1902 		} else {
1903 			selected_result = method1;
1904 		}
1905 	}
1906 
1907 	blocks = fixed16_to_u32_round_up(selected_result) + 1;
1908 	/*
1909 	 * Lets have blocks at minimum equivalent to plane_blocks_per_line
1910 	 * as there will be at minimum one line for lines configuration. This
1911 	 * is a work around for FIFO underruns observed with resolutions like
1912 	 * 4k 60 Hz in single channel DRAM configurations.
1913 	 *
1914 	 * As per the Bspec 49325, if the ddb allocation can hold at least
1915 	 * one plane_blocks_per_line, we should have selected method2 in
1916 	 * the above logic. Assuming that modern versions have enough dbuf
1917 	 * and method2 guarantees blocks equivalent to at least 1 line,
1918 	 * select the blocks as plane_blocks_per_line.
1919 	 *
1920 	 * TODO: Revisit the logic when we have better understanding on DRAM
1921 	 * channels' impact on the level 0 memory latency and the relevant
1922 	 * wm calculations.
1923 	 */
1924 	if (skl_wm_has_lines(i915, level))
1925 		blocks = max(blocks,
1926 			     fixed16_to_u32_round_up(wp->plane_blocks_per_line));
1927 	lines = div_round_up_fixed16(selected_result,
1928 				     wp->plane_blocks_per_line);
1929 
1930 	if (DISPLAY_VER(i915) == 9) {
1931 		/* Display WA #1125: skl,bxt,kbl */
1932 		if (level == 0 && wp->rc_surface)
1933 			blocks += fixed16_to_u32_round_up(wp->y_tile_minimum);
1934 
1935 		/* Display WA #1126: skl,bxt,kbl */
1936 		if (level >= 1 && level <= 7) {
1937 			if (wp->y_tiled) {
1938 				blocks += fixed16_to_u32_round_up(wp->y_tile_minimum);
1939 				lines += wp->y_min_scanlines;
1940 			} else {
1941 				blocks++;
1942 			}
1943 
1944 			/*
1945 			 * Make sure result blocks for higher latency levels are
1946 			 * at least as high as level below the current level.
1947 			 * Assumption in DDB algorithm optimization for special
1948 			 * cases. Also covers Display WA #1125 for RC.
1949 			 */
1950 			if (result_prev->blocks > blocks)
1951 				blocks = result_prev->blocks;
1952 		}
1953 	}
1954 
1955 	if (DISPLAY_VER(i915) >= 11) {
1956 		if (wp->y_tiled) {
1957 			int extra_lines;
1958 
1959 			if (lines % wp->y_min_scanlines == 0)
1960 				extra_lines = wp->y_min_scanlines;
1961 			else
1962 				extra_lines = wp->y_min_scanlines * 2 -
1963 					lines % wp->y_min_scanlines;
1964 
1965 			min_ddb_alloc = mul_round_up_u32_fixed16(lines + extra_lines,
1966 								 wp->plane_blocks_per_line);
1967 		} else {
1968 			min_ddb_alloc = blocks + DIV_ROUND_UP(blocks, 10);
1969 		}
1970 	}
1971 
1972 	if (!skl_wm_has_lines(i915, level))
1973 		lines = 0;
1974 
1975 	if (lines > skl_wm_max_lines(i915)) {
1976 		/* reject it */
1977 		result->min_ddb_alloc = U16_MAX;
1978 		return;
1979 	}
1980 
1981 	/*
1982 	 * If lines is valid, assume we can use this watermark level
1983 	 * for now.  We'll come back and disable it after we calculate the
1984 	 * DDB allocation if it turns out we don't actually have enough
1985 	 * blocks to satisfy it.
1986 	 */
1987 	result->blocks = blocks;
1988 	result->lines = lines;
1989 	/* Bspec says: value >= plane ddb allocation -> invalid, hence the +1 here */
1990 	result->min_ddb_alloc = max(min_ddb_alloc, blocks) + 1;
1991 	result->enable = true;
1992 
1993 	if (DISPLAY_VER(i915) < 12 && i915->display.sagv.block_time_us)
1994 		result->can_sagv = latency >= i915->display.sagv.block_time_us;
1995 }
1996 
1997 static void
1998 skl_compute_wm_levels(const struct intel_crtc_state *crtc_state,
1999 		      struct intel_plane *plane,
2000 		      const struct skl_wm_params *wm_params,
2001 		      struct skl_wm_level *levels)
2002 {
2003 	struct drm_i915_private *i915 = to_i915(crtc_state->uapi.crtc->dev);
2004 	struct skl_wm_level *result_prev = &levels[0];
2005 	int level;
2006 
2007 	for (level = 0; level < i915->display.wm.num_levels; level++) {
2008 		struct skl_wm_level *result = &levels[level];
2009 		unsigned int latency = skl_wm_latency(i915, level, wm_params);
2010 
2011 		skl_compute_plane_wm(crtc_state, plane, level, latency,
2012 				     wm_params, result_prev, result);
2013 
2014 		result_prev = result;
2015 	}
2016 }
2017 
2018 static void tgl_compute_sagv_wm(const struct intel_crtc_state *crtc_state,
2019 				struct intel_plane *plane,
2020 				const struct skl_wm_params *wm_params,
2021 				struct skl_plane_wm *plane_wm)
2022 {
2023 	struct drm_i915_private *i915 = to_i915(crtc_state->uapi.crtc->dev);
2024 	struct skl_wm_level *sagv_wm = &plane_wm->sagv.wm0;
2025 	struct skl_wm_level *levels = plane_wm->wm;
2026 	unsigned int latency = 0;
2027 
2028 	if (i915->display.sagv.block_time_us)
2029 		latency = i915->display.sagv.block_time_us +
2030 			skl_wm_latency(i915, 0, wm_params);
2031 
2032 	skl_compute_plane_wm(crtc_state, plane, 0, latency,
2033 			     wm_params, &levels[0],
2034 			     sagv_wm);
2035 }
2036 
2037 static void skl_compute_transition_wm(struct drm_i915_private *i915,
2038 				      struct skl_wm_level *trans_wm,
2039 				      const struct skl_wm_level *wm0,
2040 				      const struct skl_wm_params *wp)
2041 {
2042 	u16 trans_min, trans_amount, trans_y_tile_min;
2043 	u16 wm0_blocks, trans_offset, blocks;
2044 
2045 	/* Transition WM don't make any sense if ipc is disabled */
2046 	if (!skl_watermark_ipc_enabled(i915))
2047 		return;
2048 
2049 	/*
2050 	 * WaDisableTWM:skl,kbl,cfl,bxt
2051 	 * Transition WM are not recommended by HW team for GEN9
2052 	 */
2053 	if (DISPLAY_VER(i915) == 9)
2054 		return;
2055 
2056 	if (DISPLAY_VER(i915) >= 11)
2057 		trans_min = 4;
2058 	else
2059 		trans_min = 14;
2060 
2061 	/* Display WA #1140: glk,cnl */
2062 	if (DISPLAY_VER(i915) == 10)
2063 		trans_amount = 0;
2064 	else
2065 		trans_amount = 10; /* This is configurable amount */
2066 
2067 	trans_offset = trans_min + trans_amount;
2068 
2069 	/*
2070 	 * The spec asks for Selected Result Blocks for wm0 (the real value),
2071 	 * not Result Blocks (the integer value). Pay attention to the capital
2072 	 * letters. The value wm_l0->blocks is actually Result Blocks, but
2073 	 * since Result Blocks is the ceiling of Selected Result Blocks plus 1,
2074 	 * and since we later will have to get the ceiling of the sum in the
2075 	 * transition watermarks calculation, we can just pretend Selected
2076 	 * Result Blocks is Result Blocks minus 1 and it should work for the
2077 	 * current platforms.
2078 	 */
2079 	wm0_blocks = wm0->blocks - 1;
2080 
2081 	if (wp->y_tiled) {
2082 		trans_y_tile_min =
2083 			(u16)mul_round_up_u32_fixed16(2, wp->y_tile_minimum);
2084 		blocks = max(wm0_blocks, trans_y_tile_min) + trans_offset;
2085 	} else {
2086 		blocks = wm0_blocks + trans_offset;
2087 	}
2088 	blocks++;
2089 
2090 	/*
2091 	 * Just assume we can enable the transition watermark.  After
2092 	 * computing the DDB we'll come back and disable it if that
2093 	 * assumption turns out to be false.
2094 	 */
2095 	trans_wm->blocks = blocks;
2096 	trans_wm->min_ddb_alloc = max_t(u16, wm0->min_ddb_alloc, blocks + 1);
2097 	trans_wm->enable = true;
2098 }
2099 
2100 static int skl_build_plane_wm_single(struct intel_crtc_state *crtc_state,
2101 				     const struct intel_plane_state *plane_state,
2102 				     struct intel_plane *plane, int color_plane)
2103 {
2104 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
2105 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
2106 	struct skl_plane_wm *wm = &crtc_state->wm.skl.raw.planes[plane->id];
2107 	struct skl_wm_params wm_params;
2108 	int ret;
2109 
2110 	ret = skl_compute_plane_wm_params(crtc_state, plane_state,
2111 					  &wm_params, color_plane);
2112 	if (ret)
2113 		return ret;
2114 
2115 	skl_compute_wm_levels(crtc_state, plane, &wm_params, wm->wm);
2116 
2117 	skl_compute_transition_wm(i915, &wm->trans_wm,
2118 				  &wm->wm[0], &wm_params);
2119 
2120 	if (DISPLAY_VER(i915) >= 12) {
2121 		tgl_compute_sagv_wm(crtc_state, plane, &wm_params, wm);
2122 
2123 		skl_compute_transition_wm(i915, &wm->sagv.trans_wm,
2124 					  &wm->sagv.wm0, &wm_params);
2125 	}
2126 
2127 	return 0;
2128 }
2129 
2130 static int skl_build_plane_wm_uv(struct intel_crtc_state *crtc_state,
2131 				 const struct intel_plane_state *plane_state,
2132 				 struct intel_plane *plane)
2133 {
2134 	struct skl_plane_wm *wm = &crtc_state->wm.skl.raw.planes[plane->id];
2135 	struct skl_wm_params wm_params;
2136 	int ret;
2137 
2138 	wm->is_planar = true;
2139 
2140 	/* uv plane watermarks must also be validated for NV12/Planar */
2141 	ret = skl_compute_plane_wm_params(crtc_state, plane_state,
2142 					  &wm_params, 1);
2143 	if (ret)
2144 		return ret;
2145 
2146 	skl_compute_wm_levels(crtc_state, plane, &wm_params, wm->uv_wm);
2147 
2148 	return 0;
2149 }
2150 
2151 static int skl_build_plane_wm(struct intel_crtc_state *crtc_state,
2152 			      const struct intel_plane_state *plane_state)
2153 {
2154 	struct intel_plane *plane = to_intel_plane(plane_state->uapi.plane);
2155 	enum plane_id plane_id = plane->id;
2156 	struct skl_plane_wm *wm = &crtc_state->wm.skl.raw.planes[plane_id];
2157 	const struct drm_framebuffer *fb = plane_state->hw.fb;
2158 	int ret;
2159 
2160 	memset(wm, 0, sizeof(*wm));
2161 
2162 	if (!intel_wm_plane_visible(crtc_state, plane_state))
2163 		return 0;
2164 
2165 	ret = skl_build_plane_wm_single(crtc_state, plane_state,
2166 					plane, 0);
2167 	if (ret)
2168 		return ret;
2169 
2170 	if (fb->format->is_yuv && fb->format->num_planes > 1) {
2171 		ret = skl_build_plane_wm_uv(crtc_state, plane_state,
2172 					    plane);
2173 		if (ret)
2174 			return ret;
2175 	}
2176 
2177 	return 0;
2178 }
2179 
2180 static int icl_build_plane_wm(struct intel_crtc_state *crtc_state,
2181 			      const struct intel_plane_state *plane_state)
2182 {
2183 	struct intel_plane *plane = to_intel_plane(plane_state->uapi.plane);
2184 	struct drm_i915_private *i915 = to_i915(plane->base.dev);
2185 	enum plane_id plane_id = plane->id;
2186 	struct skl_plane_wm *wm = &crtc_state->wm.skl.raw.planes[plane_id];
2187 	int ret;
2188 
2189 	/* Watermarks calculated in master */
2190 	if (plane_state->planar_slave)
2191 		return 0;
2192 
2193 	memset(wm, 0, sizeof(*wm));
2194 
2195 	if (plane_state->planar_linked_plane) {
2196 		const struct drm_framebuffer *fb = plane_state->hw.fb;
2197 
2198 		drm_WARN_ON(&i915->drm,
2199 			    !intel_wm_plane_visible(crtc_state, plane_state));
2200 		drm_WARN_ON(&i915->drm, !fb->format->is_yuv ||
2201 			    fb->format->num_planes == 1);
2202 
2203 		ret = skl_build_plane_wm_single(crtc_state, plane_state,
2204 						plane_state->planar_linked_plane, 0);
2205 		if (ret)
2206 			return ret;
2207 
2208 		ret = skl_build_plane_wm_single(crtc_state, plane_state,
2209 						plane, 1);
2210 		if (ret)
2211 			return ret;
2212 	} else if (intel_wm_plane_visible(crtc_state, plane_state)) {
2213 		ret = skl_build_plane_wm_single(crtc_state, plane_state,
2214 						plane, 0);
2215 		if (ret)
2216 			return ret;
2217 	}
2218 
2219 	return 0;
2220 }
2221 
2222 static bool
2223 skl_is_vblank_too_short(const struct intel_crtc_state *crtc_state,
2224 			int wm0_lines, int latency)
2225 {
2226 	const struct drm_display_mode *adjusted_mode =
2227 		&crtc_state->hw.adjusted_mode;
2228 
2229 	/* FIXME missing scaler and DSC pre-fill time */
2230 	return crtc_state->framestart_delay +
2231 		intel_usecs_to_scanlines(adjusted_mode, latency) +
2232 		wm0_lines >
2233 		adjusted_mode->crtc_vtotal - adjusted_mode->crtc_vblank_start;
2234 }
2235 
2236 static int skl_max_wm0_lines(const struct intel_crtc_state *crtc_state)
2237 {
2238 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
2239 	enum plane_id plane_id;
2240 	int wm0_lines = 0;
2241 
2242 	for_each_plane_id_on_crtc(crtc, plane_id) {
2243 		const struct skl_plane_wm *wm = &crtc_state->wm.skl.optimal.planes[plane_id];
2244 
2245 		/* FIXME what about !skl_wm_has_lines() platforms? */
2246 		wm0_lines = max_t(int, wm0_lines, wm->wm[0].lines);
2247 	}
2248 
2249 	return wm0_lines;
2250 }
2251 
2252 static int skl_max_wm_level_for_vblank(struct intel_crtc_state *crtc_state,
2253 				       int wm0_lines)
2254 {
2255 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
2256 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
2257 	int level;
2258 
2259 	for (level = i915->display.wm.num_levels - 1; level >= 0; level--) {
2260 		int latency;
2261 
2262 		/* FIXME should we care about the latency w/a's? */
2263 		latency = skl_wm_latency(i915, level, NULL);
2264 		if (latency == 0)
2265 			continue;
2266 
2267 		/* FIXME is it correct to use 0 latency for wm0 here? */
2268 		if (level == 0)
2269 			latency = 0;
2270 
2271 		if (!skl_is_vblank_too_short(crtc_state, wm0_lines, latency))
2272 			return level;
2273 	}
2274 
2275 	return -EINVAL;
2276 }
2277 
2278 static int skl_wm_check_vblank(struct intel_crtc_state *crtc_state)
2279 {
2280 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
2281 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
2282 	int wm0_lines, level;
2283 
2284 	if (!crtc_state->hw.active)
2285 		return 0;
2286 
2287 	wm0_lines = skl_max_wm0_lines(crtc_state);
2288 
2289 	level = skl_max_wm_level_for_vblank(crtc_state, wm0_lines);
2290 	if (level < 0)
2291 		return level;
2292 
2293 	/*
2294 	 * PSR needs to toggle LATENCY_REPORTING_REMOVED_PIPE_*
2295 	 * based on whether we're limited by the vblank duration.
2296 	 */
2297 	crtc_state->wm_level_disabled = level < i915->display.wm.num_levels - 1;
2298 
2299 	for (level++; level < i915->display.wm.num_levels; level++) {
2300 		enum plane_id plane_id;
2301 
2302 		for_each_plane_id_on_crtc(crtc, plane_id) {
2303 			struct skl_plane_wm *wm =
2304 				&crtc_state->wm.skl.optimal.planes[plane_id];
2305 
2306 			/*
2307 			 * FIXME just clear enable or flag the entire
2308 			 * thing as bad via min_ddb_alloc=U16_MAX?
2309 			 */
2310 			wm->wm[level].enable = false;
2311 			wm->uv_wm[level].enable = false;
2312 		}
2313 	}
2314 
2315 	if (DISPLAY_VER(i915) >= 12 &&
2316 	    i915->display.sagv.block_time_us &&
2317 	    skl_is_vblank_too_short(crtc_state, wm0_lines,
2318 				    i915->display.sagv.block_time_us)) {
2319 		enum plane_id plane_id;
2320 
2321 		for_each_plane_id_on_crtc(crtc, plane_id) {
2322 			struct skl_plane_wm *wm =
2323 				&crtc_state->wm.skl.optimal.planes[plane_id];
2324 
2325 			wm->sagv.wm0.enable = false;
2326 			wm->sagv.trans_wm.enable = false;
2327 		}
2328 	}
2329 
2330 	return 0;
2331 }
2332 
2333 static int skl_build_pipe_wm(struct intel_atomic_state *state,
2334 			     struct intel_crtc *crtc)
2335 {
2336 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
2337 	struct intel_crtc_state *crtc_state =
2338 		intel_atomic_get_new_crtc_state(state, crtc);
2339 	const struct intel_plane_state *plane_state;
2340 	struct intel_plane *plane;
2341 	int ret, i;
2342 
2343 	for_each_new_intel_plane_in_state(state, plane, plane_state, i) {
2344 		/*
2345 		 * FIXME should perhaps check {old,new}_plane_crtc->hw.crtc
2346 		 * instead but we don't populate that correctly for NV12 Y
2347 		 * planes so for now hack this.
2348 		 */
2349 		if (plane->pipe != crtc->pipe)
2350 			continue;
2351 
2352 		if (DISPLAY_VER(i915) >= 11)
2353 			ret = icl_build_plane_wm(crtc_state, plane_state);
2354 		else
2355 			ret = skl_build_plane_wm(crtc_state, plane_state);
2356 		if (ret)
2357 			return ret;
2358 	}
2359 
2360 	crtc_state->wm.skl.optimal = crtc_state->wm.skl.raw;
2361 
2362 	return skl_wm_check_vblank(crtc_state);
2363 }
2364 
2365 static void skl_ddb_entry_write(struct drm_i915_private *i915,
2366 				i915_reg_t reg,
2367 				const struct skl_ddb_entry *entry)
2368 {
2369 	if (entry->end)
2370 		intel_de_write_fw(i915, reg,
2371 				  PLANE_BUF_END(entry->end - 1) |
2372 				  PLANE_BUF_START(entry->start));
2373 	else
2374 		intel_de_write_fw(i915, reg, 0);
2375 }
2376 
2377 static void skl_write_wm_level(struct drm_i915_private *i915,
2378 			       i915_reg_t reg,
2379 			       const struct skl_wm_level *level)
2380 {
2381 	u32 val = 0;
2382 
2383 	if (level->enable)
2384 		val |= PLANE_WM_EN;
2385 	if (level->ignore_lines)
2386 		val |= PLANE_WM_IGNORE_LINES;
2387 	val |= REG_FIELD_PREP(PLANE_WM_BLOCKS_MASK, level->blocks);
2388 	val |= REG_FIELD_PREP(PLANE_WM_LINES_MASK, level->lines);
2389 
2390 	intel_de_write_fw(i915, reg, val);
2391 }
2392 
2393 void skl_write_plane_wm(struct intel_plane *plane,
2394 			const struct intel_crtc_state *crtc_state)
2395 {
2396 	struct drm_i915_private *i915 = to_i915(plane->base.dev);
2397 	enum plane_id plane_id = plane->id;
2398 	enum pipe pipe = plane->pipe;
2399 	const struct skl_pipe_wm *pipe_wm = &crtc_state->wm.skl.optimal;
2400 	const struct skl_ddb_entry *ddb =
2401 		&crtc_state->wm.skl.plane_ddb[plane_id];
2402 	const struct skl_ddb_entry *ddb_y =
2403 		&crtc_state->wm.skl.plane_ddb_y[plane_id];
2404 	int level;
2405 
2406 	for (level = 0; level < i915->display.wm.num_levels; level++)
2407 		skl_write_wm_level(i915, PLANE_WM(pipe, plane_id, level),
2408 				   skl_plane_wm_level(pipe_wm, plane_id, level));
2409 
2410 	skl_write_wm_level(i915, PLANE_WM_TRANS(pipe, plane_id),
2411 			   skl_plane_trans_wm(pipe_wm, plane_id));
2412 
2413 	if (HAS_HW_SAGV_WM(i915)) {
2414 		const struct skl_plane_wm *wm = &pipe_wm->planes[plane_id];
2415 
2416 		skl_write_wm_level(i915, PLANE_WM_SAGV(pipe, plane_id),
2417 				   &wm->sagv.wm0);
2418 		skl_write_wm_level(i915, PLANE_WM_SAGV_TRANS(pipe, plane_id),
2419 				   &wm->sagv.trans_wm);
2420 	}
2421 
2422 	skl_ddb_entry_write(i915,
2423 			    PLANE_BUF_CFG(pipe, plane_id), ddb);
2424 
2425 	if (DISPLAY_VER(i915) < 11)
2426 		skl_ddb_entry_write(i915,
2427 				    PLANE_NV12_BUF_CFG(pipe, plane_id), ddb_y);
2428 }
2429 
2430 void skl_write_cursor_wm(struct intel_plane *plane,
2431 			 const struct intel_crtc_state *crtc_state)
2432 {
2433 	struct drm_i915_private *i915 = to_i915(plane->base.dev);
2434 	enum plane_id plane_id = plane->id;
2435 	enum pipe pipe = plane->pipe;
2436 	const struct skl_pipe_wm *pipe_wm = &crtc_state->wm.skl.optimal;
2437 	const struct skl_ddb_entry *ddb =
2438 		&crtc_state->wm.skl.plane_ddb[plane_id];
2439 	int level;
2440 
2441 	for (level = 0; level < i915->display.wm.num_levels; level++)
2442 		skl_write_wm_level(i915, CUR_WM(pipe, level),
2443 				   skl_plane_wm_level(pipe_wm, plane_id, level));
2444 
2445 	skl_write_wm_level(i915, CUR_WM_TRANS(pipe),
2446 			   skl_plane_trans_wm(pipe_wm, plane_id));
2447 
2448 	if (HAS_HW_SAGV_WM(i915)) {
2449 		const struct skl_plane_wm *wm = &pipe_wm->planes[plane_id];
2450 
2451 		skl_write_wm_level(i915, CUR_WM_SAGV(pipe),
2452 				   &wm->sagv.wm0);
2453 		skl_write_wm_level(i915, CUR_WM_SAGV_TRANS(pipe),
2454 				   &wm->sagv.trans_wm);
2455 	}
2456 
2457 	skl_ddb_entry_write(i915, CUR_BUF_CFG(pipe), ddb);
2458 }
2459 
2460 static bool skl_wm_level_equals(const struct skl_wm_level *l1,
2461 				const struct skl_wm_level *l2)
2462 {
2463 	return l1->enable == l2->enable &&
2464 		l1->ignore_lines == l2->ignore_lines &&
2465 		l1->lines == l2->lines &&
2466 		l1->blocks == l2->blocks;
2467 }
2468 
2469 static bool skl_plane_wm_equals(struct drm_i915_private *i915,
2470 				const struct skl_plane_wm *wm1,
2471 				const struct skl_plane_wm *wm2)
2472 {
2473 	int level;
2474 
2475 	for (level = 0; level < i915->display.wm.num_levels; level++) {
2476 		/*
2477 		 * We don't check uv_wm as the hardware doesn't actually
2478 		 * use it. It only gets used for calculating the required
2479 		 * ddb allocation.
2480 		 */
2481 		if (!skl_wm_level_equals(&wm1->wm[level], &wm2->wm[level]))
2482 			return false;
2483 	}
2484 
2485 	return skl_wm_level_equals(&wm1->trans_wm, &wm2->trans_wm) &&
2486 		skl_wm_level_equals(&wm1->sagv.wm0, &wm2->sagv.wm0) &&
2487 		skl_wm_level_equals(&wm1->sagv.trans_wm, &wm2->sagv.trans_wm);
2488 }
2489 
2490 static bool skl_ddb_entries_overlap(const struct skl_ddb_entry *a,
2491 				    const struct skl_ddb_entry *b)
2492 {
2493 	return a->start < b->end && b->start < a->end;
2494 }
2495 
2496 static void skl_ddb_entry_union(struct skl_ddb_entry *a,
2497 				const struct skl_ddb_entry *b)
2498 {
2499 	if (a->end && b->end) {
2500 		a->start = min(a->start, b->start);
2501 		a->end = max(a->end, b->end);
2502 	} else if (b->end) {
2503 		a->start = b->start;
2504 		a->end = b->end;
2505 	}
2506 }
2507 
2508 bool skl_ddb_allocation_overlaps(const struct skl_ddb_entry *ddb,
2509 				 const struct skl_ddb_entry *entries,
2510 				 int num_entries, int ignore_idx)
2511 {
2512 	int i;
2513 
2514 	for (i = 0; i < num_entries; i++) {
2515 		if (i != ignore_idx &&
2516 		    skl_ddb_entries_overlap(ddb, &entries[i]))
2517 			return true;
2518 	}
2519 
2520 	return false;
2521 }
2522 
2523 static int
2524 skl_ddb_add_affected_planes(const struct intel_crtc_state *old_crtc_state,
2525 			    struct intel_crtc_state *new_crtc_state)
2526 {
2527 	struct intel_atomic_state *state = to_intel_atomic_state(new_crtc_state->uapi.state);
2528 	struct intel_crtc *crtc = to_intel_crtc(new_crtc_state->uapi.crtc);
2529 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
2530 	struct intel_plane *plane;
2531 
2532 	for_each_intel_plane_on_crtc(&i915->drm, crtc, plane) {
2533 		struct intel_plane_state *plane_state;
2534 		enum plane_id plane_id = plane->id;
2535 
2536 		if (skl_ddb_entry_equal(&old_crtc_state->wm.skl.plane_ddb[plane_id],
2537 					&new_crtc_state->wm.skl.plane_ddb[plane_id]) &&
2538 		    skl_ddb_entry_equal(&old_crtc_state->wm.skl.plane_ddb_y[plane_id],
2539 					&new_crtc_state->wm.skl.plane_ddb_y[plane_id]))
2540 			continue;
2541 
2542 		plane_state = intel_atomic_get_plane_state(state, plane);
2543 		if (IS_ERR(plane_state))
2544 			return PTR_ERR(plane_state);
2545 
2546 		new_crtc_state->update_planes |= BIT(plane_id);
2547 		new_crtc_state->async_flip_planes = 0;
2548 		new_crtc_state->do_async_flip = false;
2549 	}
2550 
2551 	return 0;
2552 }
2553 
2554 static u8 intel_dbuf_enabled_slices(const struct intel_dbuf_state *dbuf_state)
2555 {
2556 	struct drm_i915_private *i915 = to_i915(dbuf_state->base.state->base.dev);
2557 	u8 enabled_slices;
2558 	enum pipe pipe;
2559 
2560 	/*
2561 	 * FIXME: For now we always enable slice S1 as per
2562 	 * the Bspec display initialization sequence.
2563 	 */
2564 	enabled_slices = BIT(DBUF_S1);
2565 
2566 	for_each_pipe(i915, pipe)
2567 		enabled_slices |= dbuf_state->slices[pipe];
2568 
2569 	return enabled_slices;
2570 }
2571 
2572 static int
2573 skl_compute_ddb(struct intel_atomic_state *state)
2574 {
2575 	struct drm_i915_private *i915 = to_i915(state->base.dev);
2576 	const struct intel_dbuf_state *old_dbuf_state;
2577 	struct intel_dbuf_state *new_dbuf_state = NULL;
2578 	const struct intel_crtc_state *old_crtc_state;
2579 	struct intel_crtc_state *new_crtc_state;
2580 	struct intel_crtc *crtc;
2581 	int ret, i;
2582 
2583 	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
2584 		new_dbuf_state = intel_atomic_get_dbuf_state(state);
2585 		if (IS_ERR(new_dbuf_state))
2586 			return PTR_ERR(new_dbuf_state);
2587 
2588 		old_dbuf_state = intel_atomic_get_old_dbuf_state(state);
2589 		break;
2590 	}
2591 
2592 	if (!new_dbuf_state)
2593 		return 0;
2594 
2595 	new_dbuf_state->active_pipes =
2596 		intel_calc_active_pipes(state, old_dbuf_state->active_pipes);
2597 
2598 	if (old_dbuf_state->active_pipes != new_dbuf_state->active_pipes) {
2599 		ret = intel_atomic_lock_global_state(&new_dbuf_state->base);
2600 		if (ret)
2601 			return ret;
2602 	}
2603 
2604 	if (HAS_MBUS_JOINING(i915))
2605 		new_dbuf_state->joined_mbus =
2606 			adlp_check_mbus_joined(new_dbuf_state->active_pipes);
2607 
2608 	for_each_intel_crtc(&i915->drm, crtc) {
2609 		enum pipe pipe = crtc->pipe;
2610 
2611 		new_dbuf_state->slices[pipe] =
2612 			skl_compute_dbuf_slices(crtc, new_dbuf_state->active_pipes,
2613 						new_dbuf_state->joined_mbus);
2614 
2615 		if (old_dbuf_state->slices[pipe] == new_dbuf_state->slices[pipe])
2616 			continue;
2617 
2618 		ret = intel_atomic_lock_global_state(&new_dbuf_state->base);
2619 		if (ret)
2620 			return ret;
2621 	}
2622 
2623 	new_dbuf_state->enabled_slices = intel_dbuf_enabled_slices(new_dbuf_state);
2624 
2625 	if (old_dbuf_state->enabled_slices != new_dbuf_state->enabled_slices ||
2626 	    old_dbuf_state->joined_mbus != new_dbuf_state->joined_mbus) {
2627 		ret = intel_atomic_serialize_global_state(&new_dbuf_state->base);
2628 		if (ret)
2629 			return ret;
2630 
2631 		if (old_dbuf_state->joined_mbus != new_dbuf_state->joined_mbus) {
2632 			/* TODO: Implement vblank synchronized MBUS joining changes */
2633 			ret = intel_modeset_all_pipes_late(state, "MBUS joining change");
2634 			if (ret)
2635 				return ret;
2636 		}
2637 
2638 		drm_dbg_kms(&i915->drm,
2639 			    "Enabled dbuf slices 0x%x -> 0x%x (total dbuf slices 0x%x), mbus joined? %s->%s\n",
2640 			    old_dbuf_state->enabled_slices,
2641 			    new_dbuf_state->enabled_slices,
2642 			    DISPLAY_INFO(i915)->dbuf.slice_mask,
2643 			    str_yes_no(old_dbuf_state->joined_mbus),
2644 			    str_yes_no(new_dbuf_state->joined_mbus));
2645 	}
2646 
2647 	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
2648 		enum pipe pipe = crtc->pipe;
2649 
2650 		new_dbuf_state->weight[pipe] = intel_crtc_ddb_weight(new_crtc_state);
2651 
2652 		if (old_dbuf_state->weight[pipe] == new_dbuf_state->weight[pipe])
2653 			continue;
2654 
2655 		ret = intel_atomic_lock_global_state(&new_dbuf_state->base);
2656 		if (ret)
2657 			return ret;
2658 	}
2659 
2660 	for_each_intel_crtc(&i915->drm, crtc) {
2661 		ret = skl_crtc_allocate_ddb(state, crtc);
2662 		if (ret)
2663 			return ret;
2664 	}
2665 
2666 	for_each_oldnew_intel_crtc_in_state(state, crtc, old_crtc_state,
2667 					    new_crtc_state, i) {
2668 		ret = skl_crtc_allocate_plane_ddb(state, crtc);
2669 		if (ret)
2670 			return ret;
2671 
2672 		ret = skl_ddb_add_affected_planes(old_crtc_state,
2673 						  new_crtc_state);
2674 		if (ret)
2675 			return ret;
2676 	}
2677 
2678 	return 0;
2679 }
2680 
2681 static char enast(bool enable)
2682 {
2683 	return enable ? '*' : ' ';
2684 }
2685 
2686 static void
2687 skl_print_wm_changes(struct intel_atomic_state *state)
2688 {
2689 	struct drm_i915_private *i915 = to_i915(state->base.dev);
2690 	const struct intel_crtc_state *old_crtc_state;
2691 	const struct intel_crtc_state *new_crtc_state;
2692 	struct intel_plane *plane;
2693 	struct intel_crtc *crtc;
2694 	int i;
2695 
2696 	if (!drm_debug_enabled(DRM_UT_KMS))
2697 		return;
2698 
2699 	for_each_oldnew_intel_crtc_in_state(state, crtc, old_crtc_state,
2700 					    new_crtc_state, i) {
2701 		const struct skl_pipe_wm *old_pipe_wm, *new_pipe_wm;
2702 
2703 		old_pipe_wm = &old_crtc_state->wm.skl.optimal;
2704 		new_pipe_wm = &new_crtc_state->wm.skl.optimal;
2705 
2706 		for_each_intel_plane_on_crtc(&i915->drm, crtc, plane) {
2707 			enum plane_id plane_id = plane->id;
2708 			const struct skl_ddb_entry *old, *new;
2709 
2710 			old = &old_crtc_state->wm.skl.plane_ddb[plane_id];
2711 			new = &new_crtc_state->wm.skl.plane_ddb[plane_id];
2712 
2713 			if (skl_ddb_entry_equal(old, new))
2714 				continue;
2715 
2716 			drm_dbg_kms(&i915->drm,
2717 				    "[PLANE:%d:%s] ddb (%4d - %4d) -> (%4d - %4d), size %4d -> %4d\n",
2718 				    plane->base.base.id, plane->base.name,
2719 				    old->start, old->end, new->start, new->end,
2720 				    skl_ddb_entry_size(old), skl_ddb_entry_size(new));
2721 		}
2722 
2723 		for_each_intel_plane_on_crtc(&i915->drm, crtc, plane) {
2724 			enum plane_id plane_id = plane->id;
2725 			const struct skl_plane_wm *old_wm, *new_wm;
2726 
2727 			old_wm = &old_pipe_wm->planes[plane_id];
2728 			new_wm = &new_pipe_wm->planes[plane_id];
2729 
2730 			if (skl_plane_wm_equals(i915, old_wm, new_wm))
2731 				continue;
2732 
2733 			drm_dbg_kms(&i915->drm,
2734 				    "[PLANE:%d:%s]   level %cwm0,%cwm1,%cwm2,%cwm3,%cwm4,%cwm5,%cwm6,%cwm7,%ctwm,%cswm,%cstwm"
2735 				    " -> %cwm0,%cwm1,%cwm2,%cwm3,%cwm4,%cwm5,%cwm6,%cwm7,%ctwm,%cswm,%cstwm\n",
2736 				    plane->base.base.id, plane->base.name,
2737 				    enast(old_wm->wm[0].enable), enast(old_wm->wm[1].enable),
2738 				    enast(old_wm->wm[2].enable), enast(old_wm->wm[3].enable),
2739 				    enast(old_wm->wm[4].enable), enast(old_wm->wm[5].enable),
2740 				    enast(old_wm->wm[6].enable), enast(old_wm->wm[7].enable),
2741 				    enast(old_wm->trans_wm.enable),
2742 				    enast(old_wm->sagv.wm0.enable),
2743 				    enast(old_wm->sagv.trans_wm.enable),
2744 				    enast(new_wm->wm[0].enable), enast(new_wm->wm[1].enable),
2745 				    enast(new_wm->wm[2].enable), enast(new_wm->wm[3].enable),
2746 				    enast(new_wm->wm[4].enable), enast(new_wm->wm[5].enable),
2747 				    enast(new_wm->wm[6].enable), enast(new_wm->wm[7].enable),
2748 				    enast(new_wm->trans_wm.enable),
2749 				    enast(new_wm->sagv.wm0.enable),
2750 				    enast(new_wm->sagv.trans_wm.enable));
2751 
2752 			drm_dbg_kms(&i915->drm,
2753 				    "[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"
2754 				      " -> %c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%4d\n",
2755 				    plane->base.base.id, plane->base.name,
2756 				    enast(old_wm->wm[0].ignore_lines), old_wm->wm[0].lines,
2757 				    enast(old_wm->wm[1].ignore_lines), old_wm->wm[1].lines,
2758 				    enast(old_wm->wm[2].ignore_lines), old_wm->wm[2].lines,
2759 				    enast(old_wm->wm[3].ignore_lines), old_wm->wm[3].lines,
2760 				    enast(old_wm->wm[4].ignore_lines), old_wm->wm[4].lines,
2761 				    enast(old_wm->wm[5].ignore_lines), old_wm->wm[5].lines,
2762 				    enast(old_wm->wm[6].ignore_lines), old_wm->wm[6].lines,
2763 				    enast(old_wm->wm[7].ignore_lines), old_wm->wm[7].lines,
2764 				    enast(old_wm->trans_wm.ignore_lines), old_wm->trans_wm.lines,
2765 				    enast(old_wm->sagv.wm0.ignore_lines), old_wm->sagv.wm0.lines,
2766 				    enast(old_wm->sagv.trans_wm.ignore_lines), old_wm->sagv.trans_wm.lines,
2767 				    enast(new_wm->wm[0].ignore_lines), new_wm->wm[0].lines,
2768 				    enast(new_wm->wm[1].ignore_lines), new_wm->wm[1].lines,
2769 				    enast(new_wm->wm[2].ignore_lines), new_wm->wm[2].lines,
2770 				    enast(new_wm->wm[3].ignore_lines), new_wm->wm[3].lines,
2771 				    enast(new_wm->wm[4].ignore_lines), new_wm->wm[4].lines,
2772 				    enast(new_wm->wm[5].ignore_lines), new_wm->wm[5].lines,
2773 				    enast(new_wm->wm[6].ignore_lines), new_wm->wm[6].lines,
2774 				    enast(new_wm->wm[7].ignore_lines), new_wm->wm[7].lines,
2775 				    enast(new_wm->trans_wm.ignore_lines), new_wm->trans_wm.lines,
2776 				    enast(new_wm->sagv.wm0.ignore_lines), new_wm->sagv.wm0.lines,
2777 				    enast(new_wm->sagv.trans_wm.ignore_lines), new_wm->sagv.trans_wm.lines);
2778 
2779 			drm_dbg_kms(&i915->drm,
2780 				    "[PLANE:%d:%s]  blocks %4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%5d"
2781 				    " -> %4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%5d\n",
2782 				    plane->base.base.id, plane->base.name,
2783 				    old_wm->wm[0].blocks, old_wm->wm[1].blocks,
2784 				    old_wm->wm[2].blocks, old_wm->wm[3].blocks,
2785 				    old_wm->wm[4].blocks, old_wm->wm[5].blocks,
2786 				    old_wm->wm[6].blocks, old_wm->wm[7].blocks,
2787 				    old_wm->trans_wm.blocks,
2788 				    old_wm->sagv.wm0.blocks,
2789 				    old_wm->sagv.trans_wm.blocks,
2790 				    new_wm->wm[0].blocks, new_wm->wm[1].blocks,
2791 				    new_wm->wm[2].blocks, new_wm->wm[3].blocks,
2792 				    new_wm->wm[4].blocks, new_wm->wm[5].blocks,
2793 				    new_wm->wm[6].blocks, new_wm->wm[7].blocks,
2794 				    new_wm->trans_wm.blocks,
2795 				    new_wm->sagv.wm0.blocks,
2796 				    new_wm->sagv.trans_wm.blocks);
2797 
2798 			drm_dbg_kms(&i915->drm,
2799 				    "[PLANE:%d:%s] min_ddb %4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%5d"
2800 				    " -> %4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%5d\n",
2801 				    plane->base.base.id, plane->base.name,
2802 				    old_wm->wm[0].min_ddb_alloc, old_wm->wm[1].min_ddb_alloc,
2803 				    old_wm->wm[2].min_ddb_alloc, old_wm->wm[3].min_ddb_alloc,
2804 				    old_wm->wm[4].min_ddb_alloc, old_wm->wm[5].min_ddb_alloc,
2805 				    old_wm->wm[6].min_ddb_alloc, old_wm->wm[7].min_ddb_alloc,
2806 				    old_wm->trans_wm.min_ddb_alloc,
2807 				    old_wm->sagv.wm0.min_ddb_alloc,
2808 				    old_wm->sagv.trans_wm.min_ddb_alloc,
2809 				    new_wm->wm[0].min_ddb_alloc, new_wm->wm[1].min_ddb_alloc,
2810 				    new_wm->wm[2].min_ddb_alloc, new_wm->wm[3].min_ddb_alloc,
2811 				    new_wm->wm[4].min_ddb_alloc, new_wm->wm[5].min_ddb_alloc,
2812 				    new_wm->wm[6].min_ddb_alloc, new_wm->wm[7].min_ddb_alloc,
2813 				    new_wm->trans_wm.min_ddb_alloc,
2814 				    new_wm->sagv.wm0.min_ddb_alloc,
2815 				    new_wm->sagv.trans_wm.min_ddb_alloc);
2816 		}
2817 	}
2818 }
2819 
2820 static bool skl_plane_selected_wm_equals(struct intel_plane *plane,
2821 					 const struct skl_pipe_wm *old_pipe_wm,
2822 					 const struct skl_pipe_wm *new_pipe_wm)
2823 {
2824 	struct drm_i915_private *i915 = to_i915(plane->base.dev);
2825 	int level;
2826 
2827 	for (level = 0; level < i915->display.wm.num_levels; level++) {
2828 		/*
2829 		 * We don't check uv_wm as the hardware doesn't actually
2830 		 * use it. It only gets used for calculating the required
2831 		 * ddb allocation.
2832 		 */
2833 		if (!skl_wm_level_equals(skl_plane_wm_level(old_pipe_wm, plane->id, level),
2834 					 skl_plane_wm_level(new_pipe_wm, plane->id, level)))
2835 			return false;
2836 	}
2837 
2838 	if (HAS_HW_SAGV_WM(i915)) {
2839 		const struct skl_plane_wm *old_wm = &old_pipe_wm->planes[plane->id];
2840 		const struct skl_plane_wm *new_wm = &new_pipe_wm->planes[plane->id];
2841 
2842 		if (!skl_wm_level_equals(&old_wm->sagv.wm0, &new_wm->sagv.wm0) ||
2843 		    !skl_wm_level_equals(&old_wm->sagv.trans_wm, &new_wm->sagv.trans_wm))
2844 			return false;
2845 	}
2846 
2847 	return skl_wm_level_equals(skl_plane_trans_wm(old_pipe_wm, plane->id),
2848 				   skl_plane_trans_wm(new_pipe_wm, plane->id));
2849 }
2850 
2851 /*
2852  * To make sure the cursor watermark registers are always consistent
2853  * with our computed state the following scenario needs special
2854  * treatment:
2855  *
2856  * 1. enable cursor
2857  * 2. move cursor entirely offscreen
2858  * 3. disable cursor
2859  *
2860  * Step 2. does call .disable_plane() but does not zero the watermarks
2861  * (since we consider an offscreen cursor still active for the purposes
2862  * of watermarks). Step 3. would not normally call .disable_plane()
2863  * because the actual plane visibility isn't changing, and we don't
2864  * deallocate the cursor ddb until the pipe gets disabled. So we must
2865  * force step 3. to call .disable_plane() to update the watermark
2866  * registers properly.
2867  *
2868  * Other planes do not suffer from this issues as their watermarks are
2869  * calculated based on the actual plane visibility. The only time this
2870  * can trigger for the other planes is during the initial readout as the
2871  * default value of the watermarks registers is not zero.
2872  */
2873 static int skl_wm_add_affected_planes(struct intel_atomic_state *state,
2874 				      struct intel_crtc *crtc)
2875 {
2876 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
2877 	const struct intel_crtc_state *old_crtc_state =
2878 		intel_atomic_get_old_crtc_state(state, crtc);
2879 	struct intel_crtc_state *new_crtc_state =
2880 		intel_atomic_get_new_crtc_state(state, crtc);
2881 	struct intel_plane *plane;
2882 
2883 	for_each_intel_plane_on_crtc(&i915->drm, crtc, plane) {
2884 		struct intel_plane_state *plane_state;
2885 		enum plane_id plane_id = plane->id;
2886 
2887 		/*
2888 		 * Force a full wm update for every plane on modeset.
2889 		 * Required because the reset value of the wm registers
2890 		 * is non-zero, whereas we want all disabled planes to
2891 		 * have zero watermarks. So if we turn off the relevant
2892 		 * power well the hardware state will go out of sync
2893 		 * with the software state.
2894 		 */
2895 		if (!intel_crtc_needs_modeset(new_crtc_state) &&
2896 		    skl_plane_selected_wm_equals(plane,
2897 						 &old_crtc_state->wm.skl.optimal,
2898 						 &new_crtc_state->wm.skl.optimal))
2899 			continue;
2900 
2901 		plane_state = intel_atomic_get_plane_state(state, plane);
2902 		if (IS_ERR(plane_state))
2903 			return PTR_ERR(plane_state);
2904 
2905 		new_crtc_state->update_planes |= BIT(plane_id);
2906 		new_crtc_state->async_flip_planes = 0;
2907 		new_crtc_state->do_async_flip = false;
2908 	}
2909 
2910 	return 0;
2911 }
2912 
2913 /*
2914  * If Fixed Refresh Rate:
2915  * Program DEEP PKG_C_LATENCY Pkg C with highest valid latency from
2916  * watermark level1 and up and above. If watermark level 1 is
2917  * invalid program it with all 1's.
2918  * Program PKG_C_LATENCY Added Wake Time = DSB execution time
2919  * If Variable Refresh Rate:
2920  * Program DEEP PKG_C_LATENCY Pkg C with all 1's.
2921  * Program PKG_C_LATENCY Added Wake Time = 0
2922  */
2923 static void
2924 skl_program_dpkgc_latency(struct drm_i915_private *i915, bool vrr_enabled)
2925 {
2926 	u32 max_latency = 0;
2927 	u32 clear = 0, val = 0;
2928 	u32 added_wake_time = 0;
2929 
2930 	if (DISPLAY_VER(i915) < 20)
2931 		return;
2932 
2933 	if (vrr_enabled) {
2934 		max_latency = LNL_PKG_C_LATENCY_MASK;
2935 		added_wake_time = 0;
2936 	} else {
2937 		max_latency = skl_watermark_max_latency(i915, 1);
2938 		if (max_latency == 0)
2939 			max_latency = LNL_PKG_C_LATENCY_MASK;
2940 		added_wake_time = DSB_EXE_TIME +
2941 			i915->display.sagv.block_time_us;
2942 	}
2943 
2944 	clear |= LNL_ADDED_WAKE_TIME_MASK | LNL_PKG_C_LATENCY_MASK;
2945 	val |= REG_FIELD_PREP(LNL_PKG_C_LATENCY_MASK, max_latency);
2946 	val |= REG_FIELD_PREP(LNL_ADDED_WAKE_TIME_MASK, added_wake_time);
2947 
2948 	intel_uncore_rmw(&i915->uncore, LNL_PKG_C_LATENCY, clear, val);
2949 }
2950 
2951 static int
2952 skl_compute_wm(struct intel_atomic_state *state)
2953 {
2954 	struct intel_crtc *crtc;
2955 	struct intel_crtc_state __maybe_unused *new_crtc_state;
2956 	int ret, i;
2957 	bool vrr_enabled = false;
2958 
2959 	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
2960 		ret = skl_build_pipe_wm(state, crtc);
2961 		if (ret)
2962 			return ret;
2963 	}
2964 
2965 	ret = skl_compute_ddb(state);
2966 	if (ret)
2967 		return ret;
2968 
2969 	ret = intel_compute_sagv_mask(state);
2970 	if (ret)
2971 		return ret;
2972 
2973 	/*
2974 	 * skl_compute_ddb() will have adjusted the final watermarks
2975 	 * based on how much ddb is available. Now we can actually
2976 	 * check if the final watermarks changed.
2977 	 */
2978 	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
2979 		ret = skl_wm_add_affected_planes(state, crtc);
2980 		if (ret)
2981 			return ret;
2982 
2983 		if (new_crtc_state->vrr.enable)
2984 			vrr_enabled = true;
2985 	}
2986 
2987 	skl_program_dpkgc_latency(to_i915(state->base.dev), vrr_enabled);
2988 
2989 	skl_print_wm_changes(state);
2990 
2991 	return 0;
2992 }
2993 
2994 static void skl_wm_level_from_reg_val(u32 val, struct skl_wm_level *level)
2995 {
2996 	level->enable = val & PLANE_WM_EN;
2997 	level->ignore_lines = val & PLANE_WM_IGNORE_LINES;
2998 	level->blocks = REG_FIELD_GET(PLANE_WM_BLOCKS_MASK, val);
2999 	level->lines = REG_FIELD_GET(PLANE_WM_LINES_MASK, val);
3000 }
3001 
3002 static void skl_pipe_wm_get_hw_state(struct intel_crtc *crtc,
3003 				     struct skl_pipe_wm *out)
3004 {
3005 	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
3006 	enum pipe pipe = crtc->pipe;
3007 	enum plane_id plane_id;
3008 	int level;
3009 	u32 val;
3010 
3011 	for_each_plane_id_on_crtc(crtc, plane_id) {
3012 		struct skl_plane_wm *wm = &out->planes[plane_id];
3013 
3014 		for (level = 0; level < i915->display.wm.num_levels; level++) {
3015 			if (plane_id != PLANE_CURSOR)
3016 				val = intel_de_read(i915, PLANE_WM(pipe, plane_id, level));
3017 			else
3018 				val = intel_de_read(i915, CUR_WM(pipe, level));
3019 
3020 			skl_wm_level_from_reg_val(val, &wm->wm[level]);
3021 		}
3022 
3023 		if (plane_id != PLANE_CURSOR)
3024 			val = intel_de_read(i915, PLANE_WM_TRANS(pipe, plane_id));
3025 		else
3026 			val = intel_de_read(i915, CUR_WM_TRANS(pipe));
3027 
3028 		skl_wm_level_from_reg_val(val, &wm->trans_wm);
3029 
3030 		if (HAS_HW_SAGV_WM(i915)) {
3031 			if (plane_id != PLANE_CURSOR)
3032 				val = intel_de_read(i915, PLANE_WM_SAGV(pipe, plane_id));
3033 			else
3034 				val = intel_de_read(i915, CUR_WM_SAGV(pipe));
3035 
3036 			skl_wm_level_from_reg_val(val, &wm->sagv.wm0);
3037 
3038 			if (plane_id != PLANE_CURSOR)
3039 				val = intel_de_read(i915, PLANE_WM_SAGV_TRANS(pipe, plane_id));
3040 			else
3041 				val = intel_de_read(i915, CUR_WM_SAGV_TRANS(pipe));
3042 
3043 			skl_wm_level_from_reg_val(val, &wm->sagv.trans_wm);
3044 		} else if (DISPLAY_VER(i915) >= 12) {
3045 			wm->sagv.wm0 = wm->wm[0];
3046 			wm->sagv.trans_wm = wm->trans_wm;
3047 		}
3048 	}
3049 }
3050 
3051 static void skl_wm_get_hw_state(struct drm_i915_private *i915)
3052 {
3053 	struct intel_dbuf_state *dbuf_state =
3054 		to_intel_dbuf_state(i915->display.dbuf.obj.state);
3055 	struct intel_crtc *crtc;
3056 
3057 	if (HAS_MBUS_JOINING(i915))
3058 		dbuf_state->joined_mbus = intel_de_read(i915, MBUS_CTL) & MBUS_JOIN;
3059 
3060 	for_each_intel_crtc(&i915->drm, crtc) {
3061 		struct intel_crtc_state *crtc_state =
3062 			to_intel_crtc_state(crtc->base.state);
3063 		enum pipe pipe = crtc->pipe;
3064 		unsigned int mbus_offset;
3065 		enum plane_id plane_id;
3066 		u8 slices;
3067 
3068 		memset(&crtc_state->wm.skl.optimal, 0,
3069 		       sizeof(crtc_state->wm.skl.optimal));
3070 		if (crtc_state->hw.active)
3071 			skl_pipe_wm_get_hw_state(crtc, &crtc_state->wm.skl.optimal);
3072 		crtc_state->wm.skl.raw = crtc_state->wm.skl.optimal;
3073 
3074 		memset(&dbuf_state->ddb[pipe], 0, sizeof(dbuf_state->ddb[pipe]));
3075 
3076 		for_each_plane_id_on_crtc(crtc, plane_id) {
3077 			struct skl_ddb_entry *ddb =
3078 				&crtc_state->wm.skl.plane_ddb[plane_id];
3079 			struct skl_ddb_entry *ddb_y =
3080 				&crtc_state->wm.skl.plane_ddb_y[plane_id];
3081 
3082 			if (!crtc_state->hw.active)
3083 				continue;
3084 
3085 			skl_ddb_get_hw_plane_state(i915, crtc->pipe,
3086 						   plane_id, ddb, ddb_y);
3087 
3088 			skl_ddb_entry_union(&dbuf_state->ddb[pipe], ddb);
3089 			skl_ddb_entry_union(&dbuf_state->ddb[pipe], ddb_y);
3090 		}
3091 
3092 		dbuf_state->weight[pipe] = intel_crtc_ddb_weight(crtc_state);
3093 
3094 		/*
3095 		 * Used for checking overlaps, so we need absolute
3096 		 * offsets instead of MBUS relative offsets.
3097 		 */
3098 		slices = skl_compute_dbuf_slices(crtc, dbuf_state->active_pipes,
3099 						 dbuf_state->joined_mbus);
3100 		mbus_offset = mbus_ddb_offset(i915, slices);
3101 		crtc_state->wm.skl.ddb.start = mbus_offset + dbuf_state->ddb[pipe].start;
3102 		crtc_state->wm.skl.ddb.end = mbus_offset + dbuf_state->ddb[pipe].end;
3103 
3104 		/* The slices actually used by the planes on the pipe */
3105 		dbuf_state->slices[pipe] =
3106 			skl_ddb_dbuf_slice_mask(i915, &crtc_state->wm.skl.ddb);
3107 
3108 		drm_dbg_kms(&i915->drm,
3109 			    "[CRTC:%d:%s] dbuf slices 0x%x, ddb (%d - %d), active pipes 0x%x, mbus joined: %s\n",
3110 			    crtc->base.base.id, crtc->base.name,
3111 			    dbuf_state->slices[pipe], dbuf_state->ddb[pipe].start,
3112 			    dbuf_state->ddb[pipe].end, dbuf_state->active_pipes,
3113 			    str_yes_no(dbuf_state->joined_mbus));
3114 	}
3115 
3116 	dbuf_state->enabled_slices = i915->display.dbuf.enabled_slices;
3117 }
3118 
3119 static bool skl_dbuf_is_misconfigured(struct drm_i915_private *i915)
3120 {
3121 	const struct intel_dbuf_state *dbuf_state =
3122 		to_intel_dbuf_state(i915->display.dbuf.obj.state);
3123 	struct skl_ddb_entry entries[I915_MAX_PIPES] = {};
3124 	struct intel_crtc *crtc;
3125 
3126 	for_each_intel_crtc(&i915->drm, crtc) {
3127 		const struct intel_crtc_state *crtc_state =
3128 			to_intel_crtc_state(crtc->base.state);
3129 
3130 		entries[crtc->pipe] = crtc_state->wm.skl.ddb;
3131 	}
3132 
3133 	for_each_intel_crtc(&i915->drm, crtc) {
3134 		const struct intel_crtc_state *crtc_state =
3135 			to_intel_crtc_state(crtc->base.state);
3136 		u8 slices;
3137 
3138 		slices = skl_compute_dbuf_slices(crtc, dbuf_state->active_pipes,
3139 						 dbuf_state->joined_mbus);
3140 		if (dbuf_state->slices[crtc->pipe] & ~slices)
3141 			return true;
3142 
3143 		if (skl_ddb_allocation_overlaps(&crtc_state->wm.skl.ddb, entries,
3144 						I915_MAX_PIPES, crtc->pipe))
3145 			return true;
3146 	}
3147 
3148 	return false;
3149 }
3150 
3151 static void skl_wm_sanitize(struct drm_i915_private *i915)
3152 {
3153 	struct intel_crtc *crtc;
3154 
3155 	/*
3156 	 * On TGL/RKL (at least) the BIOS likes to assign the planes
3157 	 * to the wrong DBUF slices. This will cause an infinite loop
3158 	 * in skl_commit_modeset_enables() as it can't find a way to
3159 	 * transition between the old bogus DBUF layout to the new
3160 	 * proper DBUF layout without DBUF allocation overlaps between
3161 	 * the planes (which cannot be allowed or else the hardware
3162 	 * may hang). If we detect a bogus DBUF layout just turn off
3163 	 * all the planes so that skl_commit_modeset_enables() can
3164 	 * simply ignore them.
3165 	 */
3166 	if (!skl_dbuf_is_misconfigured(i915))
3167 		return;
3168 
3169 	drm_dbg_kms(&i915->drm, "BIOS has misprogrammed the DBUF, disabling all planes\n");
3170 
3171 	for_each_intel_crtc(&i915->drm, crtc) {
3172 		struct intel_plane *plane = to_intel_plane(crtc->base.primary);
3173 		const struct intel_plane_state *plane_state =
3174 			to_intel_plane_state(plane->base.state);
3175 		struct intel_crtc_state *crtc_state =
3176 			to_intel_crtc_state(crtc->base.state);
3177 
3178 		if (plane_state->uapi.visible)
3179 			intel_plane_disable_noatomic(crtc, plane);
3180 
3181 		drm_WARN_ON(&i915->drm, crtc_state->active_planes != 0);
3182 
3183 		memset(&crtc_state->wm.skl.ddb, 0, sizeof(crtc_state->wm.skl.ddb));
3184 	}
3185 }
3186 
3187 static void skl_wm_get_hw_state_and_sanitize(struct drm_i915_private *i915)
3188 {
3189 	skl_wm_get_hw_state(i915);
3190 	skl_wm_sanitize(i915);
3191 }
3192 
3193 void intel_wm_state_verify(struct intel_atomic_state *state,
3194 			   struct intel_crtc *crtc)
3195 {
3196 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3197 	const struct intel_crtc_state *new_crtc_state =
3198 		intel_atomic_get_new_crtc_state(state, crtc);
3199 	struct skl_hw_state {
3200 		struct skl_ddb_entry ddb[I915_MAX_PLANES];
3201 		struct skl_ddb_entry ddb_y[I915_MAX_PLANES];
3202 		struct skl_pipe_wm wm;
3203 	} *hw;
3204 	const struct skl_pipe_wm *sw_wm = &new_crtc_state->wm.skl.optimal;
3205 	struct intel_plane *plane;
3206 	u8 hw_enabled_slices;
3207 	int level;
3208 
3209 	if (DISPLAY_VER(i915) < 9 || !new_crtc_state->hw.active)
3210 		return;
3211 
3212 	hw = kzalloc(sizeof(*hw), GFP_KERNEL);
3213 	if (!hw)
3214 		return;
3215 
3216 	skl_pipe_wm_get_hw_state(crtc, &hw->wm);
3217 
3218 	skl_pipe_ddb_get_hw_state(crtc, hw->ddb, hw->ddb_y);
3219 
3220 	hw_enabled_slices = intel_enabled_dbuf_slices_mask(i915);
3221 
3222 	if (DISPLAY_VER(i915) >= 11 &&
3223 	    hw_enabled_slices != i915->display.dbuf.enabled_slices)
3224 		drm_err(&i915->drm,
3225 			"mismatch in DBUF Slices (expected 0x%x, got 0x%x)\n",
3226 			i915->display.dbuf.enabled_slices,
3227 			hw_enabled_slices);
3228 
3229 	for_each_intel_plane_on_crtc(&i915->drm, crtc, plane) {
3230 		const struct skl_ddb_entry *hw_ddb_entry, *sw_ddb_entry;
3231 		const struct skl_wm_level *hw_wm_level, *sw_wm_level;
3232 
3233 		/* Watermarks */
3234 		for (level = 0; level < i915->display.wm.num_levels; level++) {
3235 			hw_wm_level = &hw->wm.planes[plane->id].wm[level];
3236 			sw_wm_level = skl_plane_wm_level(sw_wm, plane->id, level);
3237 
3238 			if (skl_wm_level_equals(hw_wm_level, sw_wm_level))
3239 				continue;
3240 
3241 			drm_err(&i915->drm,
3242 				"[PLANE:%d:%s] mismatch in WM%d (expected e=%d b=%u l=%u, got e=%d b=%u l=%u)\n",
3243 				plane->base.base.id, plane->base.name, level,
3244 				sw_wm_level->enable,
3245 				sw_wm_level->blocks,
3246 				sw_wm_level->lines,
3247 				hw_wm_level->enable,
3248 				hw_wm_level->blocks,
3249 				hw_wm_level->lines);
3250 		}
3251 
3252 		hw_wm_level = &hw->wm.planes[plane->id].trans_wm;
3253 		sw_wm_level = skl_plane_trans_wm(sw_wm, plane->id);
3254 
3255 		if (!skl_wm_level_equals(hw_wm_level, sw_wm_level)) {
3256 			drm_err(&i915->drm,
3257 				"[PLANE:%d:%s] mismatch in trans WM (expected e=%d b=%u l=%u, got e=%d b=%u l=%u)\n",
3258 				plane->base.base.id, plane->base.name,
3259 				sw_wm_level->enable,
3260 				sw_wm_level->blocks,
3261 				sw_wm_level->lines,
3262 				hw_wm_level->enable,
3263 				hw_wm_level->blocks,
3264 				hw_wm_level->lines);
3265 		}
3266 
3267 		hw_wm_level = &hw->wm.planes[plane->id].sagv.wm0;
3268 		sw_wm_level = &sw_wm->planes[plane->id].sagv.wm0;
3269 
3270 		if (HAS_HW_SAGV_WM(i915) &&
3271 		    !skl_wm_level_equals(hw_wm_level, sw_wm_level)) {
3272 			drm_err(&i915->drm,
3273 				"[PLANE:%d:%s] mismatch in SAGV WM (expected e=%d b=%u l=%u, got e=%d b=%u l=%u)\n",
3274 				plane->base.base.id, plane->base.name,
3275 				sw_wm_level->enable,
3276 				sw_wm_level->blocks,
3277 				sw_wm_level->lines,
3278 				hw_wm_level->enable,
3279 				hw_wm_level->blocks,
3280 				hw_wm_level->lines);
3281 		}
3282 
3283 		hw_wm_level = &hw->wm.planes[plane->id].sagv.trans_wm;
3284 		sw_wm_level = &sw_wm->planes[plane->id].sagv.trans_wm;
3285 
3286 		if (HAS_HW_SAGV_WM(i915) &&
3287 		    !skl_wm_level_equals(hw_wm_level, sw_wm_level)) {
3288 			drm_err(&i915->drm,
3289 				"[PLANE:%d:%s] mismatch in SAGV trans WM (expected e=%d b=%u l=%u, got e=%d b=%u l=%u)\n",
3290 				plane->base.base.id, plane->base.name,
3291 				sw_wm_level->enable,
3292 				sw_wm_level->blocks,
3293 				sw_wm_level->lines,
3294 				hw_wm_level->enable,
3295 				hw_wm_level->blocks,
3296 				hw_wm_level->lines);
3297 		}
3298 
3299 		/* DDB */
3300 		hw_ddb_entry = &hw->ddb[PLANE_CURSOR];
3301 		sw_ddb_entry = &new_crtc_state->wm.skl.plane_ddb[PLANE_CURSOR];
3302 
3303 		if (!skl_ddb_entry_equal(hw_ddb_entry, sw_ddb_entry)) {
3304 			drm_err(&i915->drm,
3305 				"[PLANE:%d:%s] mismatch in DDB (expected (%u,%u), found (%u,%u))\n",
3306 				plane->base.base.id, plane->base.name,
3307 				sw_ddb_entry->start, sw_ddb_entry->end,
3308 				hw_ddb_entry->start, hw_ddb_entry->end);
3309 		}
3310 	}
3311 
3312 	kfree(hw);
3313 }
3314 
3315 bool skl_watermark_ipc_enabled(struct drm_i915_private *i915)
3316 {
3317 	return i915->display.wm.ipc_enabled;
3318 }
3319 
3320 void skl_watermark_ipc_update(struct drm_i915_private *i915)
3321 {
3322 	if (!HAS_IPC(i915))
3323 		return;
3324 
3325 	intel_de_rmw(i915, DISP_ARB_CTL2, DISP_IPC_ENABLE,
3326 		     skl_watermark_ipc_enabled(i915) ? DISP_IPC_ENABLE : 0);
3327 }
3328 
3329 static bool skl_watermark_ipc_can_enable(struct drm_i915_private *i915)
3330 {
3331 	/* Display WA #0477 WaDisableIPC: skl */
3332 	if (IS_SKYLAKE(i915))
3333 		return false;
3334 
3335 	/* Display WA #1141: SKL:all KBL:all CFL */
3336 	if (IS_KABYLAKE(i915) ||
3337 	    IS_COFFEELAKE(i915) ||
3338 	    IS_COMETLAKE(i915))
3339 		return i915->dram_info.symmetric_memory;
3340 
3341 	return true;
3342 }
3343 
3344 void skl_watermark_ipc_init(struct drm_i915_private *i915)
3345 {
3346 	if (!HAS_IPC(i915))
3347 		return;
3348 
3349 	i915->display.wm.ipc_enabled = skl_watermark_ipc_can_enable(i915);
3350 
3351 	skl_watermark_ipc_update(i915);
3352 }
3353 
3354 static void
3355 adjust_wm_latency(struct drm_i915_private *i915,
3356 		  u16 wm[], int num_levels, int read_latency)
3357 {
3358 	bool wm_lv_0_adjust_needed = i915->dram_info.wm_lv_0_adjust_needed;
3359 	int i, level;
3360 
3361 	/*
3362 	 * If a level n (n > 1) has a 0us latency, all levels m (m >= n)
3363 	 * need to be disabled. We make sure to sanitize the values out
3364 	 * of the punit to satisfy this requirement.
3365 	 */
3366 	for (level = 1; level < num_levels; level++) {
3367 		if (wm[level] == 0) {
3368 			for (i = level + 1; i < num_levels; i++)
3369 				wm[i] = 0;
3370 
3371 			num_levels = level;
3372 			break;
3373 		}
3374 	}
3375 
3376 	/*
3377 	 * WaWmMemoryReadLatency
3378 	 *
3379 	 * punit doesn't take into account the read latency so we need
3380 	 * to add proper adjustement to each valid level we retrieve
3381 	 * from the punit when level 0 response data is 0us.
3382 	 */
3383 	if (wm[0] == 0) {
3384 		for (level = 0; level < num_levels; level++)
3385 			wm[level] += read_latency;
3386 	}
3387 
3388 	/*
3389 	 * WA Level-0 adjustment for 16GB DIMMs: SKL+
3390 	 * If we could not get dimm info enable this WA to prevent from
3391 	 * any underrun. If not able to get Dimm info assume 16GB dimm
3392 	 * to avoid any underrun.
3393 	 */
3394 	if (wm_lv_0_adjust_needed)
3395 		wm[0] += 1;
3396 }
3397 
3398 static void mtl_read_wm_latency(struct drm_i915_private *i915, u16 wm[])
3399 {
3400 	int num_levels = i915->display.wm.num_levels;
3401 	u32 val;
3402 
3403 	val = intel_de_read(i915, MTL_LATENCY_LP0_LP1);
3404 	wm[0] = REG_FIELD_GET(MTL_LATENCY_LEVEL_EVEN_MASK, val);
3405 	wm[1] = REG_FIELD_GET(MTL_LATENCY_LEVEL_ODD_MASK, val);
3406 
3407 	val = intel_de_read(i915, MTL_LATENCY_LP2_LP3);
3408 	wm[2] = REG_FIELD_GET(MTL_LATENCY_LEVEL_EVEN_MASK, val);
3409 	wm[3] = REG_FIELD_GET(MTL_LATENCY_LEVEL_ODD_MASK, val);
3410 
3411 	val = intel_de_read(i915, MTL_LATENCY_LP4_LP5);
3412 	wm[4] = REG_FIELD_GET(MTL_LATENCY_LEVEL_EVEN_MASK, val);
3413 	wm[5] = REG_FIELD_GET(MTL_LATENCY_LEVEL_ODD_MASK, val);
3414 
3415 	adjust_wm_latency(i915, wm, num_levels, 6);
3416 }
3417 
3418 static void skl_read_wm_latency(struct drm_i915_private *i915, u16 wm[])
3419 {
3420 	int num_levels = i915->display.wm.num_levels;
3421 	int read_latency = DISPLAY_VER(i915) >= 12 ? 3 : 2;
3422 	int mult = IS_DG2(i915) ? 2 : 1;
3423 	u32 val;
3424 	int ret;
3425 
3426 	/* read the first set of memory latencies[0:3] */
3427 	val = 0; /* data0 to be programmed to 0 for first set */
3428 	ret = snb_pcode_read(&i915->uncore, GEN9_PCODE_READ_MEM_LATENCY, &val, NULL);
3429 	if (ret) {
3430 		drm_err(&i915->drm, "SKL Mailbox read error = %d\n", ret);
3431 		return;
3432 	}
3433 
3434 	wm[0] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_0_4_MASK, val) * mult;
3435 	wm[1] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_1_5_MASK, val) * mult;
3436 	wm[2] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_2_6_MASK, val) * mult;
3437 	wm[3] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_3_7_MASK, val) * mult;
3438 
3439 	/* read the second set of memory latencies[4:7] */
3440 	val = 1; /* data0 to be programmed to 1 for second set */
3441 	ret = snb_pcode_read(&i915->uncore, GEN9_PCODE_READ_MEM_LATENCY, &val, NULL);
3442 	if (ret) {
3443 		drm_err(&i915->drm, "SKL Mailbox read error = %d\n", ret);
3444 		return;
3445 	}
3446 
3447 	wm[4] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_0_4_MASK, val) * mult;
3448 	wm[5] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_1_5_MASK, val) * mult;
3449 	wm[6] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_2_6_MASK, val) * mult;
3450 	wm[7] = REG_FIELD_GET(GEN9_MEM_LATENCY_LEVEL_3_7_MASK, val) * mult;
3451 
3452 	adjust_wm_latency(i915, wm, num_levels, read_latency);
3453 }
3454 
3455 static void skl_setup_wm_latency(struct drm_i915_private *i915)
3456 {
3457 	if (HAS_HW_SAGV_WM(i915))
3458 		i915->display.wm.num_levels = 6;
3459 	else
3460 		i915->display.wm.num_levels = 8;
3461 
3462 	if (DISPLAY_VER(i915) >= 14)
3463 		mtl_read_wm_latency(i915, i915->display.wm.skl_latency);
3464 	else
3465 		skl_read_wm_latency(i915, i915->display.wm.skl_latency);
3466 
3467 	intel_print_wm_latency(i915, "Gen9 Plane", i915->display.wm.skl_latency);
3468 }
3469 
3470 static const struct intel_wm_funcs skl_wm_funcs = {
3471 	.compute_global_watermarks = skl_compute_wm,
3472 	.get_hw_state = skl_wm_get_hw_state_and_sanitize,
3473 };
3474 
3475 void skl_wm_init(struct drm_i915_private *i915)
3476 {
3477 	intel_sagv_init(i915);
3478 
3479 	skl_setup_wm_latency(i915);
3480 
3481 	i915->display.funcs.wm = &skl_wm_funcs;
3482 }
3483 
3484 static struct intel_global_state *intel_dbuf_duplicate_state(struct intel_global_obj *obj)
3485 {
3486 	struct intel_dbuf_state *dbuf_state;
3487 
3488 	dbuf_state = kmemdup(obj->state, sizeof(*dbuf_state), GFP_KERNEL);
3489 	if (!dbuf_state)
3490 		return NULL;
3491 
3492 	return &dbuf_state->base;
3493 }
3494 
3495 static void intel_dbuf_destroy_state(struct intel_global_obj *obj,
3496 				     struct intel_global_state *state)
3497 {
3498 	kfree(state);
3499 }
3500 
3501 static const struct intel_global_state_funcs intel_dbuf_funcs = {
3502 	.atomic_duplicate_state = intel_dbuf_duplicate_state,
3503 	.atomic_destroy_state = intel_dbuf_destroy_state,
3504 };
3505 
3506 struct intel_dbuf_state *
3507 intel_atomic_get_dbuf_state(struct intel_atomic_state *state)
3508 {
3509 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3510 	struct intel_global_state *dbuf_state;
3511 
3512 	dbuf_state = intel_atomic_get_global_obj_state(state, &i915->display.dbuf.obj);
3513 	if (IS_ERR(dbuf_state))
3514 		return ERR_CAST(dbuf_state);
3515 
3516 	return to_intel_dbuf_state(dbuf_state);
3517 }
3518 
3519 int intel_dbuf_init(struct drm_i915_private *i915)
3520 {
3521 	struct intel_dbuf_state *dbuf_state;
3522 
3523 	dbuf_state = kzalloc(sizeof(*dbuf_state), GFP_KERNEL);
3524 	if (!dbuf_state)
3525 		return -ENOMEM;
3526 
3527 	intel_atomic_global_obj_init(i915, &i915->display.dbuf.obj,
3528 				     &dbuf_state->base, &intel_dbuf_funcs);
3529 
3530 	return 0;
3531 }
3532 
3533 /*
3534  * Configure MBUS_CTL and all DBUF_CTL_S of each slice to join_mbus state before
3535  * update the request state of all DBUS slices.
3536  */
3537 static void update_mbus_pre_enable(struct intel_atomic_state *state)
3538 {
3539 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3540 	u32 mbus_ctl, dbuf_min_tracker_val;
3541 	enum dbuf_slice slice;
3542 	const struct intel_dbuf_state *dbuf_state =
3543 		intel_atomic_get_new_dbuf_state(state);
3544 
3545 	if (!HAS_MBUS_JOINING(i915))
3546 		return;
3547 
3548 	/*
3549 	 * TODO: Implement vblank synchronized MBUS joining changes.
3550 	 * Must be properly coordinated with dbuf reprogramming.
3551 	 */
3552 	if (dbuf_state->joined_mbus) {
3553 		mbus_ctl = MBUS_HASHING_MODE_1x4 | MBUS_JOIN |
3554 			MBUS_JOIN_PIPE_SELECT_NONE;
3555 		dbuf_min_tracker_val = DBUF_MIN_TRACKER_STATE_SERVICE(3);
3556 	} else {
3557 		mbus_ctl = MBUS_HASHING_MODE_2x2 |
3558 			MBUS_JOIN_PIPE_SELECT_NONE;
3559 		dbuf_min_tracker_val = DBUF_MIN_TRACKER_STATE_SERVICE(1);
3560 	}
3561 
3562 	intel_de_rmw(i915, MBUS_CTL,
3563 		     MBUS_HASHING_MODE_MASK | MBUS_JOIN |
3564 		     MBUS_JOIN_PIPE_SELECT_MASK, mbus_ctl);
3565 
3566 	for_each_dbuf_slice(i915, slice)
3567 		intel_de_rmw(i915, DBUF_CTL_S(slice),
3568 			     DBUF_MIN_TRACKER_STATE_SERVICE_MASK,
3569 			     dbuf_min_tracker_val);
3570 }
3571 
3572 void intel_dbuf_pre_plane_update(struct intel_atomic_state *state)
3573 {
3574 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3575 	const struct intel_dbuf_state *new_dbuf_state =
3576 		intel_atomic_get_new_dbuf_state(state);
3577 	const struct intel_dbuf_state *old_dbuf_state =
3578 		intel_atomic_get_old_dbuf_state(state);
3579 
3580 	if (!new_dbuf_state ||
3581 	    (new_dbuf_state->enabled_slices == old_dbuf_state->enabled_slices &&
3582 	     new_dbuf_state->joined_mbus == old_dbuf_state->joined_mbus))
3583 		return;
3584 
3585 	WARN_ON(!new_dbuf_state->base.changed);
3586 
3587 	update_mbus_pre_enable(state);
3588 	gen9_dbuf_slices_update(i915,
3589 				old_dbuf_state->enabled_slices |
3590 				new_dbuf_state->enabled_slices);
3591 }
3592 
3593 void intel_dbuf_post_plane_update(struct intel_atomic_state *state)
3594 {
3595 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3596 	const struct intel_dbuf_state *new_dbuf_state =
3597 		intel_atomic_get_new_dbuf_state(state);
3598 	const struct intel_dbuf_state *old_dbuf_state =
3599 		intel_atomic_get_old_dbuf_state(state);
3600 
3601 	if (!new_dbuf_state ||
3602 	    (new_dbuf_state->enabled_slices == old_dbuf_state->enabled_slices &&
3603 	     new_dbuf_state->joined_mbus == old_dbuf_state->joined_mbus))
3604 		return;
3605 
3606 	WARN_ON(!new_dbuf_state->base.changed);
3607 
3608 	gen9_dbuf_slices_update(i915,
3609 				new_dbuf_state->enabled_slices);
3610 }
3611 
3612 static bool xelpdp_is_only_pipe_per_dbuf_bank(enum pipe pipe, u8 active_pipes)
3613 {
3614 	switch (pipe) {
3615 	case PIPE_A:
3616 		return !(active_pipes & BIT(PIPE_D));
3617 	case PIPE_D:
3618 		return !(active_pipes & BIT(PIPE_A));
3619 	case PIPE_B:
3620 		return !(active_pipes & BIT(PIPE_C));
3621 	case PIPE_C:
3622 		return !(active_pipes & BIT(PIPE_B));
3623 	default: /* to suppress compiler warning */
3624 		MISSING_CASE(pipe);
3625 		break;
3626 	}
3627 
3628 	return false;
3629 }
3630 
3631 void intel_mbus_dbox_update(struct intel_atomic_state *state)
3632 {
3633 	struct drm_i915_private *i915 = to_i915(state->base.dev);
3634 	const struct intel_dbuf_state *new_dbuf_state, *old_dbuf_state;
3635 	const struct intel_crtc_state *new_crtc_state;
3636 	const struct intel_crtc *crtc;
3637 	u32 val = 0;
3638 	int i;
3639 
3640 	if (DISPLAY_VER(i915) < 11)
3641 		return;
3642 
3643 	new_dbuf_state = intel_atomic_get_new_dbuf_state(state);
3644 	old_dbuf_state = intel_atomic_get_old_dbuf_state(state);
3645 	if (!new_dbuf_state ||
3646 	    (new_dbuf_state->joined_mbus == old_dbuf_state->joined_mbus &&
3647 	     new_dbuf_state->active_pipes == old_dbuf_state->active_pipes))
3648 		return;
3649 
3650 	if (DISPLAY_VER(i915) >= 14)
3651 		val |= MBUS_DBOX_I_CREDIT(2);
3652 
3653 	if (DISPLAY_VER(i915) >= 12) {
3654 		val |= MBUS_DBOX_B2B_TRANSACTIONS_MAX(16);
3655 		val |= MBUS_DBOX_B2B_TRANSACTIONS_DELAY(1);
3656 		val |= MBUS_DBOX_REGULATE_B2B_TRANSACTIONS_EN;
3657 	}
3658 
3659 	if (DISPLAY_VER(i915) >= 14)
3660 		val |= new_dbuf_state->joined_mbus ? MBUS_DBOX_A_CREDIT(12) :
3661 						     MBUS_DBOX_A_CREDIT(8);
3662 	else if (IS_ALDERLAKE_P(i915))
3663 		/* Wa_22010947358:adl-p */
3664 		val |= new_dbuf_state->joined_mbus ? MBUS_DBOX_A_CREDIT(6) :
3665 						     MBUS_DBOX_A_CREDIT(4);
3666 	else
3667 		val |= MBUS_DBOX_A_CREDIT(2);
3668 
3669 	if (DISPLAY_VER(i915) >= 14) {
3670 		val |= MBUS_DBOX_B_CREDIT(0xA);
3671 	} else if (IS_ALDERLAKE_P(i915)) {
3672 		val |= MBUS_DBOX_BW_CREDIT(2);
3673 		val |= MBUS_DBOX_B_CREDIT(8);
3674 	} else if (DISPLAY_VER(i915) >= 12) {
3675 		val |= MBUS_DBOX_BW_CREDIT(2);
3676 		val |= MBUS_DBOX_B_CREDIT(12);
3677 	} else {
3678 		val |= MBUS_DBOX_BW_CREDIT(1);
3679 		val |= MBUS_DBOX_B_CREDIT(8);
3680 	}
3681 
3682 	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
3683 		u32 pipe_val = val;
3684 
3685 		if (!new_crtc_state->hw.active)
3686 			continue;
3687 
3688 		if (DISPLAY_VER(i915) >= 14) {
3689 			if (xelpdp_is_only_pipe_per_dbuf_bank(crtc->pipe,
3690 							      new_dbuf_state->active_pipes))
3691 				pipe_val |= MBUS_DBOX_BW_8CREDITS_MTL;
3692 			else
3693 				pipe_val |= MBUS_DBOX_BW_4CREDITS_MTL;
3694 		}
3695 
3696 		intel_de_write(i915, PIPE_MBUS_DBOX_CTL(crtc->pipe), pipe_val);
3697 	}
3698 }
3699 
3700 static int skl_watermark_ipc_status_show(struct seq_file *m, void *data)
3701 {
3702 	struct drm_i915_private *i915 = m->private;
3703 
3704 	seq_printf(m, "Isochronous Priority Control: %s\n",
3705 		   str_yes_no(skl_watermark_ipc_enabled(i915)));
3706 	return 0;
3707 }
3708 
3709 static int skl_watermark_ipc_status_open(struct inode *inode, struct file *file)
3710 {
3711 	struct drm_i915_private *i915 = inode->i_private;
3712 
3713 	return single_open(file, skl_watermark_ipc_status_show, i915);
3714 }
3715 
3716 static ssize_t skl_watermark_ipc_status_write(struct file *file,
3717 					      const char __user *ubuf,
3718 					      size_t len, loff_t *offp)
3719 {
3720 	struct seq_file *m = file->private_data;
3721 	struct drm_i915_private *i915 = m->private;
3722 	intel_wakeref_t wakeref;
3723 	bool enable;
3724 	int ret;
3725 
3726 	ret = kstrtobool_from_user(ubuf, len, &enable);
3727 	if (ret < 0)
3728 		return ret;
3729 
3730 	with_intel_runtime_pm(&i915->runtime_pm, wakeref) {
3731 		if (!skl_watermark_ipc_enabled(i915) && enable)
3732 			drm_info(&i915->drm,
3733 				 "Enabling IPC: WM will be proper only after next commit\n");
3734 		i915->display.wm.ipc_enabled = enable;
3735 		skl_watermark_ipc_update(i915);
3736 	}
3737 
3738 	return len;
3739 }
3740 
3741 static const struct file_operations skl_watermark_ipc_status_fops = {
3742 	.owner = THIS_MODULE,
3743 	.open = skl_watermark_ipc_status_open,
3744 	.read = seq_read,
3745 	.llseek = seq_lseek,
3746 	.release = single_release,
3747 	.write = skl_watermark_ipc_status_write
3748 };
3749 
3750 static int intel_sagv_status_show(struct seq_file *m, void *unused)
3751 {
3752 	struct drm_i915_private *i915 = m->private;
3753 	static const char * const sagv_status[] = {
3754 		[I915_SAGV_UNKNOWN] = "unknown",
3755 		[I915_SAGV_DISABLED] = "disabled",
3756 		[I915_SAGV_ENABLED] = "enabled",
3757 		[I915_SAGV_NOT_CONTROLLED] = "not controlled",
3758 	};
3759 
3760 	seq_printf(m, "SAGV available: %s\n", str_yes_no(intel_has_sagv(i915)));
3761 	seq_printf(m, "SAGV modparam: %s\n",
3762 		   str_enabled_disabled(i915->display.params.enable_sagv));
3763 	seq_printf(m, "SAGV status: %s\n", sagv_status[i915->display.sagv.status]);
3764 	seq_printf(m, "SAGV block time: %d usec\n", i915->display.sagv.block_time_us);
3765 
3766 	return 0;
3767 }
3768 
3769 DEFINE_SHOW_ATTRIBUTE(intel_sagv_status);
3770 
3771 void skl_watermark_debugfs_register(struct drm_i915_private *i915)
3772 {
3773 	struct drm_minor *minor = i915->drm.primary;
3774 
3775 	if (HAS_IPC(i915))
3776 		debugfs_create_file("i915_ipc_status", 0644, minor->debugfs_root, i915,
3777 				    &skl_watermark_ipc_status_fops);
3778 
3779 	if (HAS_SAGV(i915))
3780 		debugfs_create_file("i915_sagv_status", 0444, minor->debugfs_root, i915,
3781 				    &intel_sagv_status_fops);
3782 }
3783 
3784 unsigned int skl_watermark_max_latency(struct drm_i915_private *i915, int initial_wm_level)
3785 {
3786 	int level;
3787 
3788 	for (level = i915->display.wm.num_levels - 1; level >= initial_wm_level; level--) {
3789 		unsigned int latency = skl_wm_latency(i915, level, NULL);
3790 
3791 		if (latency)
3792 			return latency;
3793 	}
3794 
3795 	return 0;
3796 }
3797