xref: /linux/drivers/gpu/drm/vc4/vc4_hvs.c (revision 2dc73b48665411a08c4e5f0f823dea8510761603)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2015 Broadcom
4  */
5 
6 /**
7  * DOC: VC4 HVS module.
8  *
9  * The Hardware Video Scaler (HVS) is the piece of hardware that does
10  * translation, scaling, colorspace conversion, and compositing of
11  * pixels stored in framebuffers into a FIFO of pixels going out to
12  * the Pixel Valve (CRTC).  It operates at the system clock rate (the
13  * system audio clock gate, specifically), which is much higher than
14  * the pixel clock rate.
15  *
16  * There is a single global HVS, with multiple output FIFOs that can
17  * be consumed by the PVs.  This file just manages the resources for
18  * the HVS, while the vc4_crtc.c code actually drives HVS setup for
19  * each CRTC.
20  */
21 
22 #include <linux/bitfield.h>
23 #include <linux/clk.h>
24 #include <linux/component.h>
25 #include <linux/platform_device.h>
26 
27 #include <drm/drm_atomic_helper.h>
28 #include <drm/drm_vblank.h>
29 
30 #include "vc4_drv.h"
31 #include "vc4_regs.h"
32 
33 static const struct debugfs_reg32 hvs_regs[] = {
34 	VC4_REG32(SCALER_DISPCTRL),
35 	VC4_REG32(SCALER_DISPSTAT),
36 	VC4_REG32(SCALER_DISPID),
37 	VC4_REG32(SCALER_DISPECTRL),
38 	VC4_REG32(SCALER_DISPPROF),
39 	VC4_REG32(SCALER_DISPDITHER),
40 	VC4_REG32(SCALER_DISPEOLN),
41 	VC4_REG32(SCALER_DISPLIST0),
42 	VC4_REG32(SCALER_DISPLIST1),
43 	VC4_REG32(SCALER_DISPLIST2),
44 	VC4_REG32(SCALER_DISPLSTAT),
45 	VC4_REG32(SCALER_DISPLACT0),
46 	VC4_REG32(SCALER_DISPLACT1),
47 	VC4_REG32(SCALER_DISPLACT2),
48 	VC4_REG32(SCALER_DISPCTRL0),
49 	VC4_REG32(SCALER_DISPBKGND0),
50 	VC4_REG32(SCALER_DISPSTAT0),
51 	VC4_REG32(SCALER_DISPBASE0),
52 	VC4_REG32(SCALER_DISPCTRL1),
53 	VC4_REG32(SCALER_DISPBKGND1),
54 	VC4_REG32(SCALER_DISPSTAT1),
55 	VC4_REG32(SCALER_DISPBASE1),
56 	VC4_REG32(SCALER_DISPCTRL2),
57 	VC4_REG32(SCALER_DISPBKGND2),
58 	VC4_REG32(SCALER_DISPSTAT2),
59 	VC4_REG32(SCALER_DISPBASE2),
60 	VC4_REG32(SCALER_DISPALPHA2),
61 	VC4_REG32(SCALER_OLEDOFFS),
62 	VC4_REG32(SCALER_OLEDCOEF0),
63 	VC4_REG32(SCALER_OLEDCOEF1),
64 	VC4_REG32(SCALER_OLEDCOEF2),
65 };
66 
67 void vc4_hvs_dump_state(struct vc4_hvs *hvs)
68 {
69 	struct drm_printer p = drm_info_printer(&hvs->pdev->dev);
70 	int i;
71 
72 	drm_print_regset32(&p, &hvs->regset);
73 
74 	DRM_INFO("HVS ctx:\n");
75 	for (i = 0; i < 64; i += 4) {
76 		DRM_INFO("0x%08x (%s): 0x%08x 0x%08x 0x%08x 0x%08x\n",
77 			 i * 4, i < HVS_BOOTLOADER_DLIST_END ? "B" : "D",
78 			 readl((u32 __iomem *)hvs->dlist + i + 0),
79 			 readl((u32 __iomem *)hvs->dlist + i + 1),
80 			 readl((u32 __iomem *)hvs->dlist + i + 2),
81 			 readl((u32 __iomem *)hvs->dlist + i + 3));
82 	}
83 }
84 
85 static int vc4_hvs_debugfs_underrun(struct seq_file *m, void *data)
86 {
87 	struct drm_info_node *node = m->private;
88 	struct drm_device *dev = node->minor->dev;
89 	struct vc4_dev *vc4 = to_vc4_dev(dev);
90 	struct drm_printer p = drm_seq_file_printer(m);
91 
92 	drm_printf(&p, "%d\n", atomic_read(&vc4->underrun));
93 
94 	return 0;
95 }
96 
97 static int vc4_hvs_debugfs_dlist(struct seq_file *m, void *data)
98 {
99 	struct drm_info_node *node = m->private;
100 	struct drm_device *dev = node->minor->dev;
101 	struct vc4_dev *vc4 = to_vc4_dev(dev);
102 	struct vc4_hvs *hvs = vc4->hvs;
103 	struct drm_printer p = drm_seq_file_printer(m);
104 	unsigned int next_entry_start = 0;
105 	unsigned int i, j;
106 	u32 dlist_word, dispstat;
107 
108 	for (i = 0; i < SCALER_CHANNELS_COUNT; i++) {
109 		dispstat = VC4_GET_FIELD(HVS_READ(SCALER_DISPSTATX(i)),
110 					 SCALER_DISPSTATX_MODE);
111 		if (dispstat == SCALER_DISPSTATX_MODE_DISABLED ||
112 		    dispstat == SCALER_DISPSTATX_MODE_EOF) {
113 			drm_printf(&p, "HVS chan %u disabled\n", i);
114 			continue;
115 		}
116 
117 		drm_printf(&p, "HVS chan %u:\n", i);
118 
119 		for (j = HVS_READ(SCALER_DISPLISTX(i)); j < 256; j++) {
120 			dlist_word = readl((u32 __iomem *)vc4->hvs->dlist + j);
121 			drm_printf(&p, "dlist: %02d: 0x%08x\n", j,
122 				   dlist_word);
123 			if (!next_entry_start ||
124 			    next_entry_start == j) {
125 				if (dlist_word & SCALER_CTL0_END)
126 					break;
127 				next_entry_start = j +
128 					VC4_GET_FIELD(dlist_word,
129 						      SCALER_CTL0_SIZE);
130 			}
131 		}
132 	}
133 
134 	return 0;
135 }
136 
137 /* The filter kernel is composed of dwords each containing 3 9-bit
138  * signed integers packed next to each other.
139  */
140 #define VC4_INT_TO_COEFF(coeff) (coeff & 0x1ff)
141 #define VC4_PPF_FILTER_WORD(c0, c1, c2)				\
142 	((((c0) & 0x1ff) << 0) |				\
143 	 (((c1) & 0x1ff) << 9) |				\
144 	 (((c2) & 0x1ff) << 18))
145 
146 /* The whole filter kernel is arranged as the coefficients 0-16 going
147  * up, then a pad, then 17-31 going down and reversed within the
148  * dwords.  This means that a linear phase kernel (where it's
149  * symmetrical at the boundary between 15 and 16) has the last 5
150  * dwords matching the first 5, but reversed.
151  */
152 #define VC4_LINEAR_PHASE_KERNEL(c0, c1, c2, c3, c4, c5, c6, c7, c8,	\
153 				c9, c10, c11, c12, c13, c14, c15)	\
154 	{VC4_PPF_FILTER_WORD(c0, c1, c2),				\
155 	 VC4_PPF_FILTER_WORD(c3, c4, c5),				\
156 	 VC4_PPF_FILTER_WORD(c6, c7, c8),				\
157 	 VC4_PPF_FILTER_WORD(c9, c10, c11),				\
158 	 VC4_PPF_FILTER_WORD(c12, c13, c14),				\
159 	 VC4_PPF_FILTER_WORD(c15, c15, 0)}
160 
161 #define VC4_LINEAR_PHASE_KERNEL_DWORDS 6
162 #define VC4_KERNEL_DWORDS (VC4_LINEAR_PHASE_KERNEL_DWORDS * 2 - 1)
163 
164 /* Recommended B=1/3, C=1/3 filter choice from Mitchell/Netravali.
165  * http://www.cs.utexas.edu/~fussell/courses/cs384g/lectures/mitchell/Mitchell.pdf
166  */
167 static const u32 mitchell_netravali_1_3_1_3_kernel[] =
168 	VC4_LINEAR_PHASE_KERNEL(0, -2, -6, -8, -10, -8, -3, 2, 18,
169 				50, 82, 119, 155, 187, 213, 227);
170 
171 static int vc4_hvs_upload_linear_kernel(struct vc4_hvs *hvs,
172 					struct drm_mm_node *space,
173 					const u32 *kernel)
174 {
175 	int ret, i;
176 	u32 __iomem *dst_kernel;
177 
178 	ret = drm_mm_insert_node(&hvs->dlist_mm, space, VC4_KERNEL_DWORDS);
179 	if (ret) {
180 		DRM_ERROR("Failed to allocate space for filter kernel: %d\n",
181 			  ret);
182 		return ret;
183 	}
184 
185 	dst_kernel = hvs->dlist + space->start;
186 
187 	for (i = 0; i < VC4_KERNEL_DWORDS; i++) {
188 		if (i < VC4_LINEAR_PHASE_KERNEL_DWORDS)
189 			writel(kernel[i], &dst_kernel[i]);
190 		else {
191 			writel(kernel[VC4_KERNEL_DWORDS - i - 1],
192 			       &dst_kernel[i]);
193 		}
194 	}
195 
196 	return 0;
197 }
198 
199 static void vc4_hvs_lut_load(struct vc4_hvs *hvs,
200 			     struct vc4_crtc *vc4_crtc)
201 {
202 	struct drm_crtc *crtc = &vc4_crtc->base;
203 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
204 	u32 i;
205 
206 	/* The LUT memory is laid out with each HVS channel in order,
207 	 * each of which takes 256 writes for R, 256 for G, then 256
208 	 * for B.
209 	 */
210 	HVS_WRITE(SCALER_GAMADDR,
211 		  SCALER_GAMADDR_AUTOINC |
212 		  (vc4_state->assigned_channel * 3 * crtc->gamma_size));
213 
214 	for (i = 0; i < crtc->gamma_size; i++)
215 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_r[i]);
216 	for (i = 0; i < crtc->gamma_size; i++)
217 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_g[i]);
218 	for (i = 0; i < crtc->gamma_size; i++)
219 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_b[i]);
220 }
221 
222 static void vc4_hvs_update_gamma_lut(struct vc4_hvs *hvs,
223 				     struct vc4_crtc *vc4_crtc)
224 {
225 	struct drm_crtc_state *crtc_state = vc4_crtc->base.state;
226 	struct drm_color_lut *lut = crtc_state->gamma_lut->data;
227 	u32 length = drm_color_lut_size(crtc_state->gamma_lut);
228 	u32 i;
229 
230 	for (i = 0; i < length; i++) {
231 		vc4_crtc->lut_r[i] = drm_color_lut_extract(lut[i].red, 8);
232 		vc4_crtc->lut_g[i] = drm_color_lut_extract(lut[i].green, 8);
233 		vc4_crtc->lut_b[i] = drm_color_lut_extract(lut[i].blue, 8);
234 	}
235 
236 	vc4_hvs_lut_load(hvs, vc4_crtc);
237 }
238 
239 u8 vc4_hvs_get_fifo_frame_count(struct vc4_hvs *hvs, unsigned int fifo)
240 {
241 	u8 field = 0;
242 
243 	switch (fifo) {
244 	case 0:
245 		field = VC4_GET_FIELD(HVS_READ(SCALER_DISPSTAT1),
246 				      SCALER_DISPSTAT1_FRCNT0);
247 		break;
248 	case 1:
249 		field = VC4_GET_FIELD(HVS_READ(SCALER_DISPSTAT1),
250 				      SCALER_DISPSTAT1_FRCNT1);
251 		break;
252 	case 2:
253 		field = VC4_GET_FIELD(HVS_READ(SCALER_DISPSTAT2),
254 				      SCALER_DISPSTAT2_FRCNT2);
255 		break;
256 	}
257 
258 	return field;
259 }
260 
261 int vc4_hvs_get_fifo_from_output(struct vc4_hvs *hvs, unsigned int output)
262 {
263 	struct vc4_dev *vc4 = hvs->vc4;
264 	u32 reg;
265 	int ret;
266 
267 	if (!vc4->is_vc5)
268 		return output;
269 
270 	switch (output) {
271 	case 0:
272 		return 0;
273 
274 	case 1:
275 		return 1;
276 
277 	case 2:
278 		reg = HVS_READ(SCALER_DISPECTRL);
279 		ret = FIELD_GET(SCALER_DISPECTRL_DSP2_MUX_MASK, reg);
280 		if (ret == 0)
281 			return 2;
282 
283 		return 0;
284 
285 	case 3:
286 		reg = HVS_READ(SCALER_DISPCTRL);
287 		ret = FIELD_GET(SCALER_DISPCTRL_DSP3_MUX_MASK, reg);
288 		if (ret == 3)
289 			return -EPIPE;
290 
291 		return ret;
292 
293 	case 4:
294 		reg = HVS_READ(SCALER_DISPEOLN);
295 		ret = FIELD_GET(SCALER_DISPEOLN_DSP4_MUX_MASK, reg);
296 		if (ret == 3)
297 			return -EPIPE;
298 
299 		return ret;
300 
301 	case 5:
302 		reg = HVS_READ(SCALER_DISPDITHER);
303 		ret = FIELD_GET(SCALER_DISPDITHER_DSP5_MUX_MASK, reg);
304 		if (ret == 3)
305 			return -EPIPE;
306 
307 		return ret;
308 
309 	default:
310 		return -EPIPE;
311 	}
312 }
313 
314 static int vc4_hvs_init_channel(struct vc4_hvs *hvs, struct drm_crtc *crtc,
315 				struct drm_display_mode *mode, bool oneshot)
316 {
317 	struct vc4_dev *vc4 = hvs->vc4;
318 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
319 	struct vc4_crtc_state *vc4_crtc_state = to_vc4_crtc_state(crtc->state);
320 	unsigned int chan = vc4_crtc_state->assigned_channel;
321 	bool interlace = mode->flags & DRM_MODE_FLAG_INTERLACE;
322 	u32 dispbkgndx;
323 	u32 dispctrl;
324 
325 	HVS_WRITE(SCALER_DISPCTRLX(chan), 0);
326 	HVS_WRITE(SCALER_DISPCTRLX(chan), SCALER_DISPCTRLX_RESET);
327 	HVS_WRITE(SCALER_DISPCTRLX(chan), 0);
328 
329 	/* Turn on the scaler, which will wait for vstart to start
330 	 * compositing.
331 	 * When feeding the transposer, we should operate in oneshot
332 	 * mode.
333 	 */
334 	dispctrl = SCALER_DISPCTRLX_ENABLE;
335 
336 	if (!vc4->is_vc5)
337 		dispctrl |= VC4_SET_FIELD(mode->hdisplay,
338 					  SCALER_DISPCTRLX_WIDTH) |
339 			    VC4_SET_FIELD(mode->vdisplay,
340 					  SCALER_DISPCTRLX_HEIGHT) |
341 			    (oneshot ? SCALER_DISPCTRLX_ONESHOT : 0);
342 	else
343 		dispctrl |= VC4_SET_FIELD(mode->hdisplay,
344 					  SCALER5_DISPCTRLX_WIDTH) |
345 			    VC4_SET_FIELD(mode->vdisplay,
346 					  SCALER5_DISPCTRLX_HEIGHT) |
347 			    (oneshot ? SCALER5_DISPCTRLX_ONESHOT : 0);
348 
349 	HVS_WRITE(SCALER_DISPCTRLX(chan), dispctrl);
350 
351 	dispbkgndx = HVS_READ(SCALER_DISPBKGNDX(chan));
352 	dispbkgndx &= ~SCALER_DISPBKGND_GAMMA;
353 	dispbkgndx &= ~SCALER_DISPBKGND_INTERLACE;
354 
355 	HVS_WRITE(SCALER_DISPBKGNDX(chan), dispbkgndx |
356 		  SCALER_DISPBKGND_AUTOHS |
357 		  ((!vc4->is_vc5) ? SCALER_DISPBKGND_GAMMA : 0) |
358 		  (interlace ? SCALER_DISPBKGND_INTERLACE : 0));
359 
360 	/* Reload the LUT, since the SRAMs would have been disabled if
361 	 * all CRTCs had SCALER_DISPBKGND_GAMMA unset at once.
362 	 */
363 	vc4_hvs_lut_load(hvs, vc4_crtc);
364 
365 	return 0;
366 }
367 
368 void vc4_hvs_stop_channel(struct vc4_hvs *hvs, unsigned int chan)
369 {
370 	if (HVS_READ(SCALER_DISPCTRLX(chan)) & SCALER_DISPCTRLX_ENABLE)
371 		return;
372 
373 	HVS_WRITE(SCALER_DISPCTRLX(chan),
374 		  HVS_READ(SCALER_DISPCTRLX(chan)) | SCALER_DISPCTRLX_RESET);
375 	HVS_WRITE(SCALER_DISPCTRLX(chan),
376 		  HVS_READ(SCALER_DISPCTRLX(chan)) & ~SCALER_DISPCTRLX_ENABLE);
377 
378 	/* Once we leave, the scaler should be disabled and its fifo empty. */
379 	WARN_ON_ONCE(HVS_READ(SCALER_DISPCTRLX(chan)) & SCALER_DISPCTRLX_RESET);
380 
381 	WARN_ON_ONCE(VC4_GET_FIELD(HVS_READ(SCALER_DISPSTATX(chan)),
382 				   SCALER_DISPSTATX_MODE) !=
383 		     SCALER_DISPSTATX_MODE_DISABLED);
384 
385 	WARN_ON_ONCE((HVS_READ(SCALER_DISPSTATX(chan)) &
386 		      (SCALER_DISPSTATX_FULL | SCALER_DISPSTATX_EMPTY)) !=
387 		     SCALER_DISPSTATX_EMPTY);
388 }
389 
390 int vc4_hvs_atomic_check(struct drm_crtc *crtc, struct drm_atomic_state *state)
391 {
392 	struct drm_crtc_state *crtc_state = drm_atomic_get_new_crtc_state(state, crtc);
393 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc_state);
394 	struct drm_device *dev = crtc->dev;
395 	struct vc4_dev *vc4 = to_vc4_dev(dev);
396 	struct drm_plane *plane;
397 	unsigned long flags;
398 	const struct drm_plane_state *plane_state;
399 	u32 dlist_count = 0;
400 	int ret;
401 
402 	/* The pixelvalve can only feed one encoder (and encoders are
403 	 * 1:1 with connectors.)
404 	 */
405 	if (hweight32(crtc_state->connector_mask) > 1)
406 		return -EINVAL;
407 
408 	drm_atomic_crtc_state_for_each_plane_state(plane, plane_state, crtc_state)
409 		dlist_count += vc4_plane_dlist_size(plane_state);
410 
411 	dlist_count++; /* Account for SCALER_CTL0_END. */
412 
413 	spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
414 	ret = drm_mm_insert_node(&vc4->hvs->dlist_mm, &vc4_state->mm,
415 				 dlist_count);
416 	spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
417 	if (ret)
418 		return ret;
419 
420 	return 0;
421 }
422 
423 static void vc4_hvs_install_dlist(struct drm_crtc *crtc)
424 {
425 	struct drm_device *dev = crtc->dev;
426 	struct vc4_dev *vc4 = to_vc4_dev(dev);
427 	struct vc4_hvs *hvs = vc4->hvs;
428 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
429 
430 	HVS_WRITE(SCALER_DISPLISTX(vc4_state->assigned_channel),
431 		  vc4_state->mm.start);
432 }
433 
434 static void vc4_hvs_update_dlist(struct drm_crtc *crtc)
435 {
436 	struct drm_device *dev = crtc->dev;
437 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
438 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
439 	unsigned long flags;
440 
441 	if (crtc->state->event) {
442 		crtc->state->event->pipe = drm_crtc_index(crtc);
443 
444 		WARN_ON(drm_crtc_vblank_get(crtc) != 0);
445 
446 		spin_lock_irqsave(&dev->event_lock, flags);
447 
448 		if (!vc4_crtc->feeds_txp || vc4_state->txp_armed) {
449 			vc4_crtc->event = crtc->state->event;
450 			crtc->state->event = NULL;
451 		}
452 
453 		spin_unlock_irqrestore(&dev->event_lock, flags);
454 	}
455 
456 	spin_lock_irqsave(&vc4_crtc->irq_lock, flags);
457 	vc4_crtc->current_dlist = vc4_state->mm.start;
458 	spin_unlock_irqrestore(&vc4_crtc->irq_lock, flags);
459 }
460 
461 void vc4_hvs_atomic_begin(struct drm_crtc *crtc,
462 			  struct drm_atomic_state *state)
463 {
464 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
465 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
466 	unsigned long flags;
467 
468 	spin_lock_irqsave(&vc4_crtc->irq_lock, flags);
469 	vc4_crtc->current_hvs_channel = vc4_state->assigned_channel;
470 	spin_unlock_irqrestore(&vc4_crtc->irq_lock, flags);
471 }
472 
473 void vc4_hvs_atomic_enable(struct drm_crtc *crtc,
474 			   struct drm_atomic_state *state)
475 {
476 	struct drm_device *dev = crtc->dev;
477 	struct vc4_dev *vc4 = to_vc4_dev(dev);
478 	struct drm_display_mode *mode = &crtc->state->adjusted_mode;
479 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
480 	bool oneshot = vc4_crtc->feeds_txp;
481 
482 	vc4_hvs_install_dlist(crtc);
483 	vc4_hvs_update_dlist(crtc);
484 	vc4_hvs_init_channel(vc4->hvs, crtc, mode, oneshot);
485 }
486 
487 void vc4_hvs_atomic_disable(struct drm_crtc *crtc,
488 			    struct drm_atomic_state *state)
489 {
490 	struct drm_device *dev = crtc->dev;
491 	struct vc4_dev *vc4 = to_vc4_dev(dev);
492 	struct drm_crtc_state *old_state = drm_atomic_get_old_crtc_state(state, crtc);
493 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(old_state);
494 	unsigned int chan = vc4_state->assigned_channel;
495 
496 	vc4_hvs_stop_channel(vc4->hvs, chan);
497 }
498 
499 void vc4_hvs_atomic_flush(struct drm_crtc *crtc,
500 			  struct drm_atomic_state *state)
501 {
502 	struct drm_crtc_state *old_state = drm_atomic_get_old_crtc_state(state,
503 									 crtc);
504 	struct drm_device *dev = crtc->dev;
505 	struct vc4_dev *vc4 = to_vc4_dev(dev);
506 	struct vc4_hvs *hvs = vc4->hvs;
507 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
508 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
509 	unsigned int channel = vc4_state->assigned_channel;
510 	struct drm_plane *plane;
511 	struct vc4_plane_state *vc4_plane_state;
512 	bool debug_dump_regs = false;
513 	bool enable_bg_fill = false;
514 	u32 __iomem *dlist_start = vc4->hvs->dlist + vc4_state->mm.start;
515 	u32 __iomem *dlist_next = dlist_start;
516 
517 	if (debug_dump_regs) {
518 		DRM_INFO("CRTC %d HVS before:\n", drm_crtc_index(crtc));
519 		vc4_hvs_dump_state(hvs);
520 	}
521 
522 	/* Copy all the active planes' dlist contents to the hardware dlist. */
523 	drm_atomic_crtc_for_each_plane(plane, crtc) {
524 		/* Is this the first active plane? */
525 		if (dlist_next == dlist_start) {
526 			/* We need to enable background fill when a plane
527 			 * could be alpha blending from the background, i.e.
528 			 * where no other plane is underneath. It suffices to
529 			 * consider the first active plane here since we set
530 			 * needs_bg_fill such that either the first plane
531 			 * already needs it or all planes on top blend from
532 			 * the first or a lower plane.
533 			 */
534 			vc4_plane_state = to_vc4_plane_state(plane->state);
535 			enable_bg_fill = vc4_plane_state->needs_bg_fill;
536 		}
537 
538 		dlist_next += vc4_plane_write_dlist(plane, dlist_next);
539 	}
540 
541 	writel(SCALER_CTL0_END, dlist_next);
542 	dlist_next++;
543 
544 	WARN_ON_ONCE(dlist_next - dlist_start != vc4_state->mm.size);
545 
546 	if (enable_bg_fill)
547 		/* This sets a black background color fill, as is the case
548 		 * with other DRM drivers.
549 		 */
550 		HVS_WRITE(SCALER_DISPBKGNDX(channel),
551 			  HVS_READ(SCALER_DISPBKGNDX(channel)) |
552 			  SCALER_DISPBKGND_FILL);
553 
554 	/* Only update DISPLIST if the CRTC was already running and is not
555 	 * being disabled.
556 	 * vc4_crtc_enable() takes care of updating the dlist just after
557 	 * re-enabling VBLANK interrupts and before enabling the engine.
558 	 * If the CRTC is being disabled, there's no point in updating this
559 	 * information.
560 	 */
561 	if (crtc->state->active && old_state->active) {
562 		vc4_hvs_install_dlist(crtc);
563 		vc4_hvs_update_dlist(crtc);
564 	}
565 
566 	if (crtc->state->color_mgmt_changed) {
567 		u32 dispbkgndx = HVS_READ(SCALER_DISPBKGNDX(channel));
568 
569 		if (crtc->state->gamma_lut) {
570 			vc4_hvs_update_gamma_lut(hvs, vc4_crtc);
571 			dispbkgndx |= SCALER_DISPBKGND_GAMMA;
572 		} else {
573 			/* Unsetting DISPBKGND_GAMMA skips the gamma lut step
574 			 * in hardware, which is the same as a linear lut that
575 			 * DRM expects us to use in absence of a user lut.
576 			 */
577 			dispbkgndx &= ~SCALER_DISPBKGND_GAMMA;
578 		}
579 		HVS_WRITE(SCALER_DISPBKGNDX(channel), dispbkgndx);
580 	}
581 
582 	if (debug_dump_regs) {
583 		DRM_INFO("CRTC %d HVS after:\n", drm_crtc_index(crtc));
584 		vc4_hvs_dump_state(hvs);
585 	}
586 }
587 
588 void vc4_hvs_mask_underrun(struct vc4_hvs *hvs, int channel)
589 {
590 	u32 dispctrl = HVS_READ(SCALER_DISPCTRL);
591 
592 	dispctrl &= ~SCALER_DISPCTRL_DSPEISLUR(channel);
593 
594 	HVS_WRITE(SCALER_DISPCTRL, dispctrl);
595 }
596 
597 void vc4_hvs_unmask_underrun(struct vc4_hvs *hvs, int channel)
598 {
599 	u32 dispctrl = HVS_READ(SCALER_DISPCTRL);
600 
601 	dispctrl |= SCALER_DISPCTRL_DSPEISLUR(channel);
602 
603 	HVS_WRITE(SCALER_DISPSTAT,
604 		  SCALER_DISPSTAT_EUFLOW(channel));
605 	HVS_WRITE(SCALER_DISPCTRL, dispctrl);
606 }
607 
608 static void vc4_hvs_report_underrun(struct drm_device *dev)
609 {
610 	struct vc4_dev *vc4 = to_vc4_dev(dev);
611 
612 	atomic_inc(&vc4->underrun);
613 	DRM_DEV_ERROR(dev->dev, "HVS underrun\n");
614 }
615 
616 static irqreturn_t vc4_hvs_irq_handler(int irq, void *data)
617 {
618 	struct drm_device *dev = data;
619 	struct vc4_dev *vc4 = to_vc4_dev(dev);
620 	struct vc4_hvs *hvs = vc4->hvs;
621 	irqreturn_t irqret = IRQ_NONE;
622 	int channel;
623 	u32 control;
624 	u32 status;
625 
626 	status = HVS_READ(SCALER_DISPSTAT);
627 	control = HVS_READ(SCALER_DISPCTRL);
628 
629 	for (channel = 0; channel < SCALER_CHANNELS_COUNT; channel++) {
630 		/* Interrupt masking is not always honored, so check it here. */
631 		if (status & SCALER_DISPSTAT_EUFLOW(channel) &&
632 		    control & SCALER_DISPCTRL_DSPEISLUR(channel)) {
633 			vc4_hvs_mask_underrun(hvs, channel);
634 			vc4_hvs_report_underrun(dev);
635 
636 			irqret = IRQ_HANDLED;
637 		}
638 	}
639 
640 	/* Clear every per-channel interrupt flag. */
641 	HVS_WRITE(SCALER_DISPSTAT, SCALER_DISPSTAT_IRQMASK(0) |
642 				   SCALER_DISPSTAT_IRQMASK(1) |
643 				   SCALER_DISPSTAT_IRQMASK(2));
644 
645 	return irqret;
646 }
647 
648 static int vc4_hvs_bind(struct device *dev, struct device *master, void *data)
649 {
650 	struct platform_device *pdev = to_platform_device(dev);
651 	struct drm_device *drm = dev_get_drvdata(master);
652 	struct vc4_dev *vc4 = to_vc4_dev(drm);
653 	struct vc4_hvs *hvs = NULL;
654 	int ret;
655 	u32 dispctrl;
656 	u32 reg;
657 
658 	hvs = devm_kzalloc(&pdev->dev, sizeof(*hvs), GFP_KERNEL);
659 	if (!hvs)
660 		return -ENOMEM;
661 
662 	hvs->vc4 = vc4;
663 	hvs->pdev = pdev;
664 
665 	hvs->regs = vc4_ioremap_regs(pdev, 0);
666 	if (IS_ERR(hvs->regs))
667 		return PTR_ERR(hvs->regs);
668 
669 	hvs->regset.base = hvs->regs;
670 	hvs->regset.regs = hvs_regs;
671 	hvs->regset.nregs = ARRAY_SIZE(hvs_regs);
672 
673 	if (vc4->is_vc5) {
674 		hvs->core_clk = devm_clk_get(&pdev->dev, NULL);
675 		if (IS_ERR(hvs->core_clk)) {
676 			dev_err(&pdev->dev, "Couldn't get core clock\n");
677 			return PTR_ERR(hvs->core_clk);
678 		}
679 
680 		ret = clk_prepare_enable(hvs->core_clk);
681 		if (ret) {
682 			dev_err(&pdev->dev, "Couldn't enable the core clock\n");
683 			return ret;
684 		}
685 	}
686 
687 	if (!vc4->is_vc5)
688 		hvs->dlist = hvs->regs + SCALER_DLIST_START;
689 	else
690 		hvs->dlist = hvs->regs + SCALER5_DLIST_START;
691 
692 	spin_lock_init(&hvs->mm_lock);
693 
694 	/* Set up the HVS display list memory manager.  We never
695 	 * overwrite the setup from the bootloader (just 128b out of
696 	 * our 16K), since we don't want to scramble the screen when
697 	 * transitioning from the firmware's boot setup to runtime.
698 	 */
699 	drm_mm_init(&hvs->dlist_mm,
700 		    HVS_BOOTLOADER_DLIST_END,
701 		    (SCALER_DLIST_SIZE >> 2) - HVS_BOOTLOADER_DLIST_END);
702 
703 	/* Set up the HVS LBM memory manager.  We could have some more
704 	 * complicated data structure that allowed reuse of LBM areas
705 	 * between planes when they don't overlap on the screen, but
706 	 * for now we just allocate globally.
707 	 */
708 	if (!vc4->is_vc5)
709 		/* 48k words of 2x12-bit pixels */
710 		drm_mm_init(&hvs->lbm_mm, 0, 48 * 1024);
711 	else
712 		/* 60k words of 4x12-bit pixels */
713 		drm_mm_init(&hvs->lbm_mm, 0, 60 * 1024);
714 
715 	/* Upload filter kernels.  We only have the one for now, so we
716 	 * keep it around for the lifetime of the driver.
717 	 */
718 	ret = vc4_hvs_upload_linear_kernel(hvs,
719 					   &hvs->mitchell_netravali_filter,
720 					   mitchell_netravali_1_3_1_3_kernel);
721 	if (ret)
722 		return ret;
723 
724 	vc4->hvs = hvs;
725 
726 	reg = HVS_READ(SCALER_DISPECTRL);
727 	reg &= ~SCALER_DISPECTRL_DSP2_MUX_MASK;
728 	HVS_WRITE(SCALER_DISPECTRL,
729 		  reg | VC4_SET_FIELD(0, SCALER_DISPECTRL_DSP2_MUX));
730 
731 	reg = HVS_READ(SCALER_DISPCTRL);
732 	reg &= ~SCALER_DISPCTRL_DSP3_MUX_MASK;
733 	HVS_WRITE(SCALER_DISPCTRL,
734 		  reg | VC4_SET_FIELD(3, SCALER_DISPCTRL_DSP3_MUX));
735 
736 	reg = HVS_READ(SCALER_DISPEOLN);
737 	reg &= ~SCALER_DISPEOLN_DSP4_MUX_MASK;
738 	HVS_WRITE(SCALER_DISPEOLN,
739 		  reg | VC4_SET_FIELD(3, SCALER_DISPEOLN_DSP4_MUX));
740 
741 	reg = HVS_READ(SCALER_DISPDITHER);
742 	reg &= ~SCALER_DISPDITHER_DSP5_MUX_MASK;
743 	HVS_WRITE(SCALER_DISPDITHER,
744 		  reg | VC4_SET_FIELD(3, SCALER_DISPDITHER_DSP5_MUX));
745 
746 	dispctrl = HVS_READ(SCALER_DISPCTRL);
747 
748 	dispctrl |= SCALER_DISPCTRL_ENABLE;
749 	dispctrl |= SCALER_DISPCTRL_DISPEIRQ(0) |
750 		    SCALER_DISPCTRL_DISPEIRQ(1) |
751 		    SCALER_DISPCTRL_DISPEIRQ(2);
752 
753 	dispctrl &= ~(SCALER_DISPCTRL_DMAEIRQ |
754 		      SCALER_DISPCTRL_SLVWREIRQ |
755 		      SCALER_DISPCTRL_SLVRDEIRQ |
756 		      SCALER_DISPCTRL_DSPEIEOF(0) |
757 		      SCALER_DISPCTRL_DSPEIEOF(1) |
758 		      SCALER_DISPCTRL_DSPEIEOF(2) |
759 		      SCALER_DISPCTRL_DSPEIEOLN(0) |
760 		      SCALER_DISPCTRL_DSPEIEOLN(1) |
761 		      SCALER_DISPCTRL_DSPEIEOLN(2) |
762 		      SCALER_DISPCTRL_DSPEISLUR(0) |
763 		      SCALER_DISPCTRL_DSPEISLUR(1) |
764 		      SCALER_DISPCTRL_DSPEISLUR(2) |
765 		      SCALER_DISPCTRL_SCLEIRQ);
766 
767 	HVS_WRITE(SCALER_DISPCTRL, dispctrl);
768 
769 	ret = devm_request_irq(dev, platform_get_irq(pdev, 0),
770 			       vc4_hvs_irq_handler, 0, "vc4 hvs", drm);
771 	if (ret)
772 		return ret;
773 
774 	vc4_debugfs_add_regset32(drm, "hvs_regs", &hvs->regset);
775 	vc4_debugfs_add_file(drm, "hvs_underrun", vc4_hvs_debugfs_underrun,
776 			     NULL);
777 	vc4_debugfs_add_file(drm, "hvs_dlists", vc4_hvs_debugfs_dlist,
778 			     NULL);
779 
780 	return 0;
781 }
782 
783 static void vc4_hvs_unbind(struct device *dev, struct device *master,
784 			   void *data)
785 {
786 	struct drm_device *drm = dev_get_drvdata(master);
787 	struct vc4_dev *vc4 = to_vc4_dev(drm);
788 	struct vc4_hvs *hvs = vc4->hvs;
789 
790 	if (drm_mm_node_allocated(&vc4->hvs->mitchell_netravali_filter))
791 		drm_mm_remove_node(&vc4->hvs->mitchell_netravali_filter);
792 
793 	drm_mm_takedown(&vc4->hvs->dlist_mm);
794 	drm_mm_takedown(&vc4->hvs->lbm_mm);
795 
796 	clk_disable_unprepare(hvs->core_clk);
797 
798 	vc4->hvs = NULL;
799 }
800 
801 static const struct component_ops vc4_hvs_ops = {
802 	.bind   = vc4_hvs_bind,
803 	.unbind = vc4_hvs_unbind,
804 };
805 
806 static int vc4_hvs_dev_probe(struct platform_device *pdev)
807 {
808 	return component_add(&pdev->dev, &vc4_hvs_ops);
809 }
810 
811 static int vc4_hvs_dev_remove(struct platform_device *pdev)
812 {
813 	component_del(&pdev->dev, &vc4_hvs_ops);
814 	return 0;
815 }
816 
817 static const struct of_device_id vc4_hvs_dt_match[] = {
818 	{ .compatible = "brcm,bcm2711-hvs" },
819 	{ .compatible = "brcm,bcm2835-hvs" },
820 	{}
821 };
822 
823 struct platform_driver vc4_hvs_driver = {
824 	.probe = vc4_hvs_dev_probe,
825 	.remove = vc4_hvs_dev_remove,
826 	.driver = {
827 		.name = "vc4_hvs",
828 		.of_match_table = vc4_hvs_dt_match,
829 	},
830 };
831