xref: /openbmc/linux/drivers/gpu/drm/vc4/vc4_hvs.c (revision aaa746ad)
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_drv.h>
29 #include <drm/drm_vblank.h>
30 
31 #include <soc/bcm2835/raspberrypi-firmware.h>
32 
33 #include "vc4_drv.h"
34 #include "vc4_regs.h"
35 
36 static const struct debugfs_reg32 hvs_regs[] = {
37 	VC4_REG32(SCALER_DISPCTRL),
38 	VC4_REG32(SCALER_DISPSTAT),
39 	VC4_REG32(SCALER_DISPID),
40 	VC4_REG32(SCALER_DISPECTRL),
41 	VC4_REG32(SCALER_DISPPROF),
42 	VC4_REG32(SCALER_DISPDITHER),
43 	VC4_REG32(SCALER_DISPEOLN),
44 	VC4_REG32(SCALER_DISPLIST0),
45 	VC4_REG32(SCALER_DISPLIST1),
46 	VC4_REG32(SCALER_DISPLIST2),
47 	VC4_REG32(SCALER_DISPLSTAT),
48 	VC4_REG32(SCALER_DISPLACT0),
49 	VC4_REG32(SCALER_DISPLACT1),
50 	VC4_REG32(SCALER_DISPLACT2),
51 	VC4_REG32(SCALER_DISPCTRL0),
52 	VC4_REG32(SCALER_DISPBKGND0),
53 	VC4_REG32(SCALER_DISPSTAT0),
54 	VC4_REG32(SCALER_DISPBASE0),
55 	VC4_REG32(SCALER_DISPCTRL1),
56 	VC4_REG32(SCALER_DISPBKGND1),
57 	VC4_REG32(SCALER_DISPSTAT1),
58 	VC4_REG32(SCALER_DISPBASE1),
59 	VC4_REG32(SCALER_DISPCTRL2),
60 	VC4_REG32(SCALER_DISPBKGND2),
61 	VC4_REG32(SCALER_DISPSTAT2),
62 	VC4_REG32(SCALER_DISPBASE2),
63 	VC4_REG32(SCALER_DISPALPHA2),
64 	VC4_REG32(SCALER_OLEDOFFS),
65 	VC4_REG32(SCALER_OLEDCOEF0),
66 	VC4_REG32(SCALER_OLEDCOEF1),
67 	VC4_REG32(SCALER_OLEDCOEF2),
68 };
69 
70 void vc4_hvs_dump_state(struct vc4_hvs *hvs)
71 {
72 	struct drm_device *drm = &hvs->vc4->base;
73 	struct drm_printer p = drm_info_printer(&hvs->pdev->dev);
74 	int idx, i;
75 
76 	if (!drm_dev_enter(drm, &idx))
77 		return;
78 
79 	drm_print_regset32(&p, &hvs->regset);
80 
81 	DRM_INFO("HVS ctx:\n");
82 	for (i = 0; i < 64; i += 4) {
83 		DRM_INFO("0x%08x (%s): 0x%08x 0x%08x 0x%08x 0x%08x\n",
84 			 i * 4, i < HVS_BOOTLOADER_DLIST_END ? "B" : "D",
85 			 readl((u32 __iomem *)hvs->dlist + i + 0),
86 			 readl((u32 __iomem *)hvs->dlist + i + 1),
87 			 readl((u32 __iomem *)hvs->dlist + i + 2),
88 			 readl((u32 __iomem *)hvs->dlist + i + 3));
89 	}
90 
91 	drm_dev_exit(idx);
92 }
93 
94 static int vc4_hvs_debugfs_underrun(struct seq_file *m, void *data)
95 {
96 	struct drm_info_node *node = m->private;
97 	struct drm_device *dev = node->minor->dev;
98 	struct vc4_dev *vc4 = to_vc4_dev(dev);
99 	struct drm_printer p = drm_seq_file_printer(m);
100 
101 	drm_printf(&p, "%d\n", atomic_read(&vc4->underrun));
102 
103 	return 0;
104 }
105 
106 static int vc4_hvs_debugfs_dlist(struct seq_file *m, void *data)
107 {
108 	struct drm_info_node *node = m->private;
109 	struct drm_device *dev = node->minor->dev;
110 	struct vc4_dev *vc4 = to_vc4_dev(dev);
111 	struct vc4_hvs *hvs = vc4->hvs;
112 	struct drm_printer p = drm_seq_file_printer(m);
113 	unsigned int next_entry_start = 0;
114 	unsigned int i, j;
115 	u32 dlist_word, dispstat;
116 
117 	for (i = 0; i < SCALER_CHANNELS_COUNT; i++) {
118 		dispstat = VC4_GET_FIELD(HVS_READ(SCALER_DISPSTATX(i)),
119 					 SCALER_DISPSTATX_MODE);
120 		if (dispstat == SCALER_DISPSTATX_MODE_DISABLED ||
121 		    dispstat == SCALER_DISPSTATX_MODE_EOF) {
122 			drm_printf(&p, "HVS chan %u disabled\n", i);
123 			continue;
124 		}
125 
126 		drm_printf(&p, "HVS chan %u:\n", i);
127 
128 		for (j = HVS_READ(SCALER_DISPLISTX(i)); j < 256; j++) {
129 			dlist_word = readl((u32 __iomem *)vc4->hvs->dlist + j);
130 			drm_printf(&p, "dlist: %02d: 0x%08x\n", j,
131 				   dlist_word);
132 			if (!next_entry_start ||
133 			    next_entry_start == j) {
134 				if (dlist_word & SCALER_CTL0_END)
135 					break;
136 				next_entry_start = j +
137 					VC4_GET_FIELD(dlist_word,
138 						      SCALER_CTL0_SIZE);
139 			}
140 		}
141 	}
142 
143 	return 0;
144 }
145 
146 /* The filter kernel is composed of dwords each containing 3 9-bit
147  * signed integers packed next to each other.
148  */
149 #define VC4_INT_TO_COEFF(coeff) (coeff & 0x1ff)
150 #define VC4_PPF_FILTER_WORD(c0, c1, c2)				\
151 	((((c0) & 0x1ff) << 0) |				\
152 	 (((c1) & 0x1ff) << 9) |				\
153 	 (((c2) & 0x1ff) << 18))
154 
155 /* The whole filter kernel is arranged as the coefficients 0-16 going
156  * up, then a pad, then 17-31 going down and reversed within the
157  * dwords.  This means that a linear phase kernel (where it's
158  * symmetrical at the boundary between 15 and 16) has the last 5
159  * dwords matching the first 5, but reversed.
160  */
161 #define VC4_LINEAR_PHASE_KERNEL(c0, c1, c2, c3, c4, c5, c6, c7, c8,	\
162 				c9, c10, c11, c12, c13, c14, c15)	\
163 	{VC4_PPF_FILTER_WORD(c0, c1, c2),				\
164 	 VC4_PPF_FILTER_WORD(c3, c4, c5),				\
165 	 VC4_PPF_FILTER_WORD(c6, c7, c8),				\
166 	 VC4_PPF_FILTER_WORD(c9, c10, c11),				\
167 	 VC4_PPF_FILTER_WORD(c12, c13, c14),				\
168 	 VC4_PPF_FILTER_WORD(c15, c15, 0)}
169 
170 #define VC4_LINEAR_PHASE_KERNEL_DWORDS 6
171 #define VC4_KERNEL_DWORDS (VC4_LINEAR_PHASE_KERNEL_DWORDS * 2 - 1)
172 
173 /* Recommended B=1/3, C=1/3 filter choice from Mitchell/Netravali.
174  * http://www.cs.utexas.edu/~fussell/courses/cs384g/lectures/mitchell/Mitchell.pdf
175  */
176 static const u32 mitchell_netravali_1_3_1_3_kernel[] =
177 	VC4_LINEAR_PHASE_KERNEL(0, -2, -6, -8, -10, -8, -3, 2, 18,
178 				50, 82, 119, 155, 187, 213, 227);
179 
180 static int vc4_hvs_upload_linear_kernel(struct vc4_hvs *hvs,
181 					struct drm_mm_node *space,
182 					const u32 *kernel)
183 {
184 	int ret, i;
185 	u32 __iomem *dst_kernel;
186 
187 	/*
188 	 * NOTE: We don't need a call to drm_dev_enter()/drm_dev_exit()
189 	 * here since that function is only called from vc4_hvs_bind().
190 	 */
191 
192 	ret = drm_mm_insert_node(&hvs->dlist_mm, space, VC4_KERNEL_DWORDS);
193 	if (ret) {
194 		DRM_ERROR("Failed to allocate space for filter kernel: %d\n",
195 			  ret);
196 		return ret;
197 	}
198 
199 	dst_kernel = hvs->dlist + space->start;
200 
201 	for (i = 0; i < VC4_KERNEL_DWORDS; i++) {
202 		if (i < VC4_LINEAR_PHASE_KERNEL_DWORDS)
203 			writel(kernel[i], &dst_kernel[i]);
204 		else {
205 			writel(kernel[VC4_KERNEL_DWORDS - i - 1],
206 			       &dst_kernel[i]);
207 		}
208 	}
209 
210 	return 0;
211 }
212 
213 static void vc4_hvs_lut_load(struct vc4_hvs *hvs,
214 			     struct vc4_crtc *vc4_crtc)
215 {
216 	struct drm_device *drm = &hvs->vc4->base;
217 	struct drm_crtc *crtc = &vc4_crtc->base;
218 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
219 	int idx;
220 	u32 i;
221 
222 	if (!drm_dev_enter(drm, &idx))
223 		return;
224 
225 	/* The LUT memory is laid out with each HVS channel in order,
226 	 * each of which takes 256 writes for R, 256 for G, then 256
227 	 * for B.
228 	 */
229 	HVS_WRITE(SCALER_GAMADDR,
230 		  SCALER_GAMADDR_AUTOINC |
231 		  (vc4_state->assigned_channel * 3 * crtc->gamma_size));
232 
233 	for (i = 0; i < crtc->gamma_size; i++)
234 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_r[i]);
235 	for (i = 0; i < crtc->gamma_size; i++)
236 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_g[i]);
237 	for (i = 0; i < crtc->gamma_size; i++)
238 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_b[i]);
239 
240 	drm_dev_exit(idx);
241 }
242 
243 static void vc4_hvs_update_gamma_lut(struct vc4_hvs *hvs,
244 				     struct vc4_crtc *vc4_crtc)
245 {
246 	struct drm_crtc_state *crtc_state = vc4_crtc->base.state;
247 	struct drm_color_lut *lut = crtc_state->gamma_lut->data;
248 	u32 length = drm_color_lut_size(crtc_state->gamma_lut);
249 	u32 i;
250 
251 	for (i = 0; i < length; i++) {
252 		vc4_crtc->lut_r[i] = drm_color_lut_extract(lut[i].red, 8);
253 		vc4_crtc->lut_g[i] = drm_color_lut_extract(lut[i].green, 8);
254 		vc4_crtc->lut_b[i] = drm_color_lut_extract(lut[i].blue, 8);
255 	}
256 
257 	vc4_hvs_lut_load(hvs, vc4_crtc);
258 }
259 
260 u8 vc4_hvs_get_fifo_frame_count(struct vc4_hvs *hvs, unsigned int fifo)
261 {
262 	struct drm_device *drm = &hvs->vc4->base;
263 	u8 field = 0;
264 	int idx;
265 
266 	if (!drm_dev_enter(drm, &idx))
267 		return 0;
268 
269 	switch (fifo) {
270 	case 0:
271 		field = VC4_GET_FIELD(HVS_READ(SCALER_DISPSTAT1),
272 				      SCALER_DISPSTAT1_FRCNT0);
273 		break;
274 	case 1:
275 		field = VC4_GET_FIELD(HVS_READ(SCALER_DISPSTAT1),
276 				      SCALER_DISPSTAT1_FRCNT1);
277 		break;
278 	case 2:
279 		field = VC4_GET_FIELD(HVS_READ(SCALER_DISPSTAT2),
280 				      SCALER_DISPSTAT2_FRCNT2);
281 		break;
282 	}
283 
284 	drm_dev_exit(idx);
285 	return field;
286 }
287 
288 int vc4_hvs_get_fifo_from_output(struct vc4_hvs *hvs, unsigned int output)
289 {
290 	struct vc4_dev *vc4 = hvs->vc4;
291 	u32 reg;
292 	int ret;
293 
294 	if (!vc4->is_vc5)
295 		return output;
296 
297 	/*
298 	 * NOTE: We should probably use drm_dev_enter()/drm_dev_exit()
299 	 * here, but this function is only used during the DRM device
300 	 * initialization, so we should be fine.
301 	 */
302 
303 	switch (output) {
304 	case 0:
305 		return 0;
306 
307 	case 1:
308 		return 1;
309 
310 	case 2:
311 		reg = HVS_READ(SCALER_DISPECTRL);
312 		ret = FIELD_GET(SCALER_DISPECTRL_DSP2_MUX_MASK, reg);
313 		if (ret == 0)
314 			return 2;
315 
316 		return 0;
317 
318 	case 3:
319 		reg = HVS_READ(SCALER_DISPCTRL);
320 		ret = FIELD_GET(SCALER_DISPCTRL_DSP3_MUX_MASK, reg);
321 		if (ret == 3)
322 			return -EPIPE;
323 
324 		return ret;
325 
326 	case 4:
327 		reg = HVS_READ(SCALER_DISPEOLN);
328 		ret = FIELD_GET(SCALER_DISPEOLN_DSP4_MUX_MASK, reg);
329 		if (ret == 3)
330 			return -EPIPE;
331 
332 		return ret;
333 
334 	case 5:
335 		reg = HVS_READ(SCALER_DISPDITHER);
336 		ret = FIELD_GET(SCALER_DISPDITHER_DSP5_MUX_MASK, reg);
337 		if (ret == 3)
338 			return -EPIPE;
339 
340 		return ret;
341 
342 	default:
343 		return -EPIPE;
344 	}
345 }
346 
347 static int vc4_hvs_init_channel(struct vc4_hvs *hvs, struct drm_crtc *crtc,
348 				struct drm_display_mode *mode, bool oneshot)
349 {
350 	struct vc4_dev *vc4 = hvs->vc4;
351 	struct drm_device *drm = &vc4->base;
352 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
353 	struct vc4_crtc_state *vc4_crtc_state = to_vc4_crtc_state(crtc->state);
354 	unsigned int chan = vc4_crtc_state->assigned_channel;
355 	bool interlace = mode->flags & DRM_MODE_FLAG_INTERLACE;
356 	u32 dispbkgndx;
357 	u32 dispctrl;
358 	int idx;
359 
360 	if (!drm_dev_enter(drm, &idx))
361 		return -ENODEV;
362 
363 	HVS_WRITE(SCALER_DISPCTRLX(chan), 0);
364 	HVS_WRITE(SCALER_DISPCTRLX(chan), SCALER_DISPCTRLX_RESET);
365 	HVS_WRITE(SCALER_DISPCTRLX(chan), 0);
366 
367 	/* Turn on the scaler, which will wait for vstart to start
368 	 * compositing.
369 	 * When feeding the transposer, we should operate in oneshot
370 	 * mode.
371 	 */
372 	dispctrl = SCALER_DISPCTRLX_ENABLE;
373 
374 	if (!vc4->is_vc5)
375 		dispctrl |= VC4_SET_FIELD(mode->hdisplay,
376 					  SCALER_DISPCTRLX_WIDTH) |
377 			    VC4_SET_FIELD(mode->vdisplay,
378 					  SCALER_DISPCTRLX_HEIGHT) |
379 			    (oneshot ? SCALER_DISPCTRLX_ONESHOT : 0);
380 	else
381 		dispctrl |= VC4_SET_FIELD(mode->hdisplay,
382 					  SCALER5_DISPCTRLX_WIDTH) |
383 			    VC4_SET_FIELD(mode->vdisplay,
384 					  SCALER5_DISPCTRLX_HEIGHT) |
385 			    (oneshot ? SCALER5_DISPCTRLX_ONESHOT : 0);
386 
387 	HVS_WRITE(SCALER_DISPCTRLX(chan), dispctrl);
388 
389 	dispbkgndx = HVS_READ(SCALER_DISPBKGNDX(chan));
390 	dispbkgndx &= ~SCALER_DISPBKGND_GAMMA;
391 	dispbkgndx &= ~SCALER_DISPBKGND_INTERLACE;
392 
393 	HVS_WRITE(SCALER_DISPBKGNDX(chan), dispbkgndx |
394 		  SCALER_DISPBKGND_AUTOHS |
395 		  ((!vc4->is_vc5) ? SCALER_DISPBKGND_GAMMA : 0) |
396 		  (interlace ? SCALER_DISPBKGND_INTERLACE : 0));
397 
398 	/* Reload the LUT, since the SRAMs would have been disabled if
399 	 * all CRTCs had SCALER_DISPBKGND_GAMMA unset at once.
400 	 */
401 	vc4_hvs_lut_load(hvs, vc4_crtc);
402 
403 	drm_dev_exit(idx);
404 
405 	return 0;
406 }
407 
408 void vc4_hvs_stop_channel(struct vc4_hvs *hvs, unsigned int chan)
409 {
410 	struct drm_device *drm = &hvs->vc4->base;
411 	int idx;
412 
413 	if (!drm_dev_enter(drm, &idx))
414 		return;
415 
416 	if (HVS_READ(SCALER_DISPCTRLX(chan)) & SCALER_DISPCTRLX_ENABLE)
417 		goto out;
418 
419 	HVS_WRITE(SCALER_DISPCTRLX(chan),
420 		  HVS_READ(SCALER_DISPCTRLX(chan)) | SCALER_DISPCTRLX_RESET);
421 	HVS_WRITE(SCALER_DISPCTRLX(chan),
422 		  HVS_READ(SCALER_DISPCTRLX(chan)) & ~SCALER_DISPCTRLX_ENABLE);
423 
424 	/* Once we leave, the scaler should be disabled and its fifo empty. */
425 	WARN_ON_ONCE(HVS_READ(SCALER_DISPCTRLX(chan)) & SCALER_DISPCTRLX_RESET);
426 
427 	WARN_ON_ONCE(VC4_GET_FIELD(HVS_READ(SCALER_DISPSTATX(chan)),
428 				   SCALER_DISPSTATX_MODE) !=
429 		     SCALER_DISPSTATX_MODE_DISABLED);
430 
431 	WARN_ON_ONCE((HVS_READ(SCALER_DISPSTATX(chan)) &
432 		      (SCALER_DISPSTATX_FULL | SCALER_DISPSTATX_EMPTY)) !=
433 		     SCALER_DISPSTATX_EMPTY);
434 
435 out:
436 	drm_dev_exit(idx);
437 }
438 
439 int vc4_hvs_atomic_check(struct drm_crtc *crtc, struct drm_atomic_state *state)
440 {
441 	struct drm_crtc_state *crtc_state = drm_atomic_get_new_crtc_state(state, crtc);
442 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc_state);
443 	struct drm_device *dev = crtc->dev;
444 	struct vc4_dev *vc4 = to_vc4_dev(dev);
445 	struct drm_plane *plane;
446 	unsigned long flags;
447 	const struct drm_plane_state *plane_state;
448 	u32 dlist_count = 0;
449 	int ret;
450 
451 	/* The pixelvalve can only feed one encoder (and encoders are
452 	 * 1:1 with connectors.)
453 	 */
454 	if (hweight32(crtc_state->connector_mask) > 1)
455 		return -EINVAL;
456 
457 	drm_atomic_crtc_state_for_each_plane_state(plane, plane_state, crtc_state)
458 		dlist_count += vc4_plane_dlist_size(plane_state);
459 
460 	dlist_count++; /* Account for SCALER_CTL0_END. */
461 
462 	spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
463 	ret = drm_mm_insert_node(&vc4->hvs->dlist_mm, &vc4_state->mm,
464 				 dlist_count);
465 	spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
466 	if (ret)
467 		return ret;
468 
469 	return 0;
470 }
471 
472 static void vc4_hvs_install_dlist(struct drm_crtc *crtc)
473 {
474 	struct drm_device *dev = crtc->dev;
475 	struct vc4_dev *vc4 = to_vc4_dev(dev);
476 	struct vc4_hvs *hvs = vc4->hvs;
477 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
478 	int idx;
479 
480 	if (!drm_dev_enter(dev, &idx))
481 		return;
482 
483 	HVS_WRITE(SCALER_DISPLISTX(vc4_state->assigned_channel),
484 		  vc4_state->mm.start);
485 
486 	drm_dev_exit(idx);
487 }
488 
489 static void vc4_hvs_update_dlist(struct drm_crtc *crtc)
490 {
491 	struct drm_device *dev = crtc->dev;
492 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
493 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
494 	unsigned long flags;
495 
496 	if (crtc->state->event) {
497 		crtc->state->event->pipe = drm_crtc_index(crtc);
498 
499 		WARN_ON(drm_crtc_vblank_get(crtc) != 0);
500 
501 		spin_lock_irqsave(&dev->event_lock, flags);
502 
503 		if (!vc4_crtc->feeds_txp || vc4_state->txp_armed) {
504 			vc4_crtc->event = crtc->state->event;
505 			crtc->state->event = NULL;
506 		}
507 
508 		spin_unlock_irqrestore(&dev->event_lock, flags);
509 	}
510 
511 	spin_lock_irqsave(&vc4_crtc->irq_lock, flags);
512 	vc4_crtc->current_dlist = vc4_state->mm.start;
513 	spin_unlock_irqrestore(&vc4_crtc->irq_lock, flags);
514 }
515 
516 void vc4_hvs_atomic_begin(struct drm_crtc *crtc,
517 			  struct drm_atomic_state *state)
518 {
519 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
520 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
521 	unsigned long flags;
522 
523 	spin_lock_irqsave(&vc4_crtc->irq_lock, flags);
524 	vc4_crtc->current_hvs_channel = vc4_state->assigned_channel;
525 	spin_unlock_irqrestore(&vc4_crtc->irq_lock, flags);
526 }
527 
528 void vc4_hvs_atomic_enable(struct drm_crtc *crtc,
529 			   struct drm_atomic_state *state)
530 {
531 	struct drm_device *dev = crtc->dev;
532 	struct vc4_dev *vc4 = to_vc4_dev(dev);
533 	struct drm_display_mode *mode = &crtc->state->adjusted_mode;
534 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
535 	bool oneshot = vc4_crtc->feeds_txp;
536 
537 	vc4_hvs_install_dlist(crtc);
538 	vc4_hvs_update_dlist(crtc);
539 	vc4_hvs_init_channel(vc4->hvs, crtc, mode, oneshot);
540 }
541 
542 void vc4_hvs_atomic_disable(struct drm_crtc *crtc,
543 			    struct drm_atomic_state *state)
544 {
545 	struct drm_device *dev = crtc->dev;
546 	struct vc4_dev *vc4 = to_vc4_dev(dev);
547 	struct drm_crtc_state *old_state = drm_atomic_get_old_crtc_state(state, crtc);
548 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(old_state);
549 	unsigned int chan = vc4_state->assigned_channel;
550 
551 	vc4_hvs_stop_channel(vc4->hvs, chan);
552 }
553 
554 void vc4_hvs_atomic_flush(struct drm_crtc *crtc,
555 			  struct drm_atomic_state *state)
556 {
557 	struct drm_crtc_state *old_state = drm_atomic_get_old_crtc_state(state,
558 									 crtc);
559 	struct drm_device *dev = crtc->dev;
560 	struct vc4_dev *vc4 = to_vc4_dev(dev);
561 	struct vc4_hvs *hvs = vc4->hvs;
562 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
563 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
564 	unsigned int channel = vc4_state->assigned_channel;
565 	struct drm_plane *plane;
566 	struct vc4_plane_state *vc4_plane_state;
567 	bool debug_dump_regs = false;
568 	bool enable_bg_fill = false;
569 	u32 __iomem *dlist_start = vc4->hvs->dlist + vc4_state->mm.start;
570 	u32 __iomem *dlist_next = dlist_start;
571 	int idx;
572 
573 	if (!drm_dev_enter(dev, &idx)) {
574 		vc4_crtc_send_vblank(crtc);
575 		return;
576 	}
577 
578 	if (debug_dump_regs) {
579 		DRM_INFO("CRTC %d HVS before:\n", drm_crtc_index(crtc));
580 		vc4_hvs_dump_state(hvs);
581 	}
582 
583 	/* Copy all the active planes' dlist contents to the hardware dlist. */
584 	drm_atomic_crtc_for_each_plane(plane, crtc) {
585 		/* Is this the first active plane? */
586 		if (dlist_next == dlist_start) {
587 			/* We need to enable background fill when a plane
588 			 * could be alpha blending from the background, i.e.
589 			 * where no other plane is underneath. It suffices to
590 			 * consider the first active plane here since we set
591 			 * needs_bg_fill such that either the first plane
592 			 * already needs it or all planes on top blend from
593 			 * the first or a lower plane.
594 			 */
595 			vc4_plane_state = to_vc4_plane_state(plane->state);
596 			enable_bg_fill = vc4_plane_state->needs_bg_fill;
597 		}
598 
599 		dlist_next += vc4_plane_write_dlist(plane, dlist_next);
600 	}
601 
602 	writel(SCALER_CTL0_END, dlist_next);
603 	dlist_next++;
604 
605 	WARN_ON_ONCE(dlist_next - dlist_start != vc4_state->mm.size);
606 
607 	if (enable_bg_fill)
608 		/* This sets a black background color fill, as is the case
609 		 * with other DRM drivers.
610 		 */
611 		HVS_WRITE(SCALER_DISPBKGNDX(channel),
612 			  HVS_READ(SCALER_DISPBKGNDX(channel)) |
613 			  SCALER_DISPBKGND_FILL);
614 
615 	/* Only update DISPLIST if the CRTC was already running and is not
616 	 * being disabled.
617 	 * vc4_crtc_enable() takes care of updating the dlist just after
618 	 * re-enabling VBLANK interrupts and before enabling the engine.
619 	 * If the CRTC is being disabled, there's no point in updating this
620 	 * information.
621 	 */
622 	if (crtc->state->active && old_state->active) {
623 		vc4_hvs_install_dlist(crtc);
624 		vc4_hvs_update_dlist(crtc);
625 	}
626 
627 	if (crtc->state->color_mgmt_changed) {
628 		u32 dispbkgndx = HVS_READ(SCALER_DISPBKGNDX(channel));
629 
630 		if (crtc->state->gamma_lut) {
631 			vc4_hvs_update_gamma_lut(hvs, vc4_crtc);
632 			dispbkgndx |= SCALER_DISPBKGND_GAMMA;
633 		} else {
634 			/* Unsetting DISPBKGND_GAMMA skips the gamma lut step
635 			 * in hardware, which is the same as a linear lut that
636 			 * DRM expects us to use in absence of a user lut.
637 			 */
638 			dispbkgndx &= ~SCALER_DISPBKGND_GAMMA;
639 		}
640 		HVS_WRITE(SCALER_DISPBKGNDX(channel), dispbkgndx);
641 	}
642 
643 	if (debug_dump_regs) {
644 		DRM_INFO("CRTC %d HVS after:\n", drm_crtc_index(crtc));
645 		vc4_hvs_dump_state(hvs);
646 	}
647 
648 	drm_dev_exit(idx);
649 }
650 
651 void vc4_hvs_mask_underrun(struct vc4_hvs *hvs, int channel)
652 {
653 	struct drm_device *drm = &hvs->vc4->base;
654 	u32 dispctrl;
655 	int idx;
656 
657 	if (!drm_dev_enter(drm, &idx))
658 		return;
659 
660 	dispctrl = HVS_READ(SCALER_DISPCTRL);
661 	dispctrl &= ~SCALER_DISPCTRL_DSPEISLUR(channel);
662 
663 	HVS_WRITE(SCALER_DISPCTRL, dispctrl);
664 
665 	drm_dev_exit(idx);
666 }
667 
668 void vc4_hvs_unmask_underrun(struct vc4_hvs *hvs, int channel)
669 {
670 	struct drm_device *drm = &hvs->vc4->base;
671 	u32 dispctrl;
672 	int idx;
673 
674 	if (!drm_dev_enter(drm, &idx))
675 		return;
676 
677 	dispctrl = HVS_READ(SCALER_DISPCTRL);
678 	dispctrl |= SCALER_DISPCTRL_DSPEISLUR(channel);
679 
680 	HVS_WRITE(SCALER_DISPSTAT,
681 		  SCALER_DISPSTAT_EUFLOW(channel));
682 	HVS_WRITE(SCALER_DISPCTRL, dispctrl);
683 
684 	drm_dev_exit(idx);
685 }
686 
687 static void vc4_hvs_report_underrun(struct drm_device *dev)
688 {
689 	struct vc4_dev *vc4 = to_vc4_dev(dev);
690 
691 	atomic_inc(&vc4->underrun);
692 	DRM_DEV_ERROR(dev->dev, "HVS underrun\n");
693 }
694 
695 static irqreturn_t vc4_hvs_irq_handler(int irq, void *data)
696 {
697 	struct drm_device *dev = data;
698 	struct vc4_dev *vc4 = to_vc4_dev(dev);
699 	struct vc4_hvs *hvs = vc4->hvs;
700 	irqreturn_t irqret = IRQ_NONE;
701 	int channel;
702 	u32 control;
703 	u32 status;
704 
705 	/*
706 	 * NOTE: We don't need to protect the register access using
707 	 * drm_dev_enter() there because the interrupt handler lifetime
708 	 * is tied to the device itself, and not to the DRM device.
709 	 *
710 	 * So when the device will be gone, one of the first thing we
711 	 * will be doing will be to unregister the interrupt handler,
712 	 * and then unregister the DRM device. drm_dev_enter() would
713 	 * thus always succeed if we are here.
714 	 */
715 
716 	status = HVS_READ(SCALER_DISPSTAT);
717 	control = HVS_READ(SCALER_DISPCTRL);
718 
719 	for (channel = 0; channel < SCALER_CHANNELS_COUNT; channel++) {
720 		/* Interrupt masking is not always honored, so check it here. */
721 		if (status & SCALER_DISPSTAT_EUFLOW(channel) &&
722 		    control & SCALER_DISPCTRL_DSPEISLUR(channel)) {
723 			vc4_hvs_mask_underrun(hvs, channel);
724 			vc4_hvs_report_underrun(dev);
725 
726 			irqret = IRQ_HANDLED;
727 		}
728 	}
729 
730 	/* Clear every per-channel interrupt flag. */
731 	HVS_WRITE(SCALER_DISPSTAT, SCALER_DISPSTAT_IRQMASK(0) |
732 				   SCALER_DISPSTAT_IRQMASK(1) |
733 				   SCALER_DISPSTAT_IRQMASK(2));
734 
735 	return irqret;
736 }
737 
738 int vc4_hvs_debugfs_init(struct drm_minor *minor)
739 {
740 	struct drm_device *drm = minor->dev;
741 	struct vc4_dev *vc4 = to_vc4_dev(drm);
742 	struct vc4_hvs *hvs = vc4->hvs;
743 	int ret;
744 
745 	if (!vc4->hvs)
746 		return -ENODEV;
747 
748 	if (!vc4->is_vc5)
749 		debugfs_create_bool("hvs_load_tracker", S_IRUGO | S_IWUSR,
750 				    minor->debugfs_root,
751 				    &vc4->load_tracker_enabled);
752 
753 	ret = vc4_debugfs_add_file(minor, "hvs_dlists",
754 				   vc4_hvs_debugfs_dlist, NULL);
755 	if (ret)
756 		return ret;
757 
758 	ret = vc4_debugfs_add_file(minor, "hvs_underrun",
759 				   vc4_hvs_debugfs_underrun, NULL);
760 	if (ret)
761 		return ret;
762 
763 	ret = vc4_debugfs_add_regset32(minor, "hvs_regs",
764 				       &hvs->regset);
765 	if (ret)
766 		return ret;
767 
768 	return 0;
769 }
770 
771 static int vc4_hvs_bind(struct device *dev, struct device *master, void *data)
772 {
773 	struct platform_device *pdev = to_platform_device(dev);
774 	struct drm_device *drm = dev_get_drvdata(master);
775 	struct vc4_dev *vc4 = to_vc4_dev(drm);
776 	struct vc4_hvs *hvs = NULL;
777 	int ret;
778 	u32 dispctrl;
779 	u32 reg;
780 
781 	hvs = drmm_kzalloc(drm, sizeof(*hvs), GFP_KERNEL);
782 	if (!hvs)
783 		return -ENOMEM;
784 	hvs->vc4 = vc4;
785 	hvs->pdev = pdev;
786 
787 	hvs->regs = vc4_ioremap_regs(pdev, 0);
788 	if (IS_ERR(hvs->regs))
789 		return PTR_ERR(hvs->regs);
790 
791 	hvs->regset.base = hvs->regs;
792 	hvs->regset.regs = hvs_regs;
793 	hvs->regset.nregs = ARRAY_SIZE(hvs_regs);
794 
795 	if (vc4->is_vc5) {
796 		struct rpi_firmware *firmware;
797 		struct device_node *node;
798 		unsigned int max_rate;
799 
800 		node = rpi_firmware_find_node();
801 		if (!node)
802 			return -EINVAL;
803 
804 		firmware = rpi_firmware_get(node);
805 		of_node_put(node);
806 		if (!firmware)
807 			return -EPROBE_DEFER;
808 
809 		hvs->core_clk = devm_clk_get(&pdev->dev, NULL);
810 		if (IS_ERR(hvs->core_clk)) {
811 			dev_err(&pdev->dev, "Couldn't get core clock\n");
812 			return PTR_ERR(hvs->core_clk);
813 		}
814 
815 		max_rate = rpi_firmware_clk_get_max_rate(firmware,
816 							 RPI_FIRMWARE_CORE_CLK_ID);
817 		rpi_firmware_put(firmware);
818 		if (max_rate >= 550000000)
819 			hvs->vc5_hdmi_enable_hdmi_20 = true;
820 
821 		if (max_rate >= 600000000)
822 			hvs->vc5_hdmi_enable_4096by2160 = true;
823 
824 		hvs->max_core_rate = max_rate;
825 
826 		ret = clk_prepare_enable(hvs->core_clk);
827 		if (ret) {
828 			dev_err(&pdev->dev, "Couldn't enable the core clock\n");
829 			return ret;
830 		}
831 	}
832 
833 	if (!vc4->is_vc5)
834 		hvs->dlist = hvs->regs + SCALER_DLIST_START;
835 	else
836 		hvs->dlist = hvs->regs + SCALER5_DLIST_START;
837 
838 	spin_lock_init(&hvs->mm_lock);
839 
840 	/* Set up the HVS display list memory manager.  We never
841 	 * overwrite the setup from the bootloader (just 128b out of
842 	 * our 16K), since we don't want to scramble the screen when
843 	 * transitioning from the firmware's boot setup to runtime.
844 	 */
845 	drm_mm_init(&hvs->dlist_mm,
846 		    HVS_BOOTLOADER_DLIST_END,
847 		    (SCALER_DLIST_SIZE >> 2) - HVS_BOOTLOADER_DLIST_END);
848 
849 	/* Set up the HVS LBM memory manager.  We could have some more
850 	 * complicated data structure that allowed reuse of LBM areas
851 	 * between planes when they don't overlap on the screen, but
852 	 * for now we just allocate globally.
853 	 */
854 	if (!vc4->is_vc5)
855 		/* 48k words of 2x12-bit pixels */
856 		drm_mm_init(&hvs->lbm_mm, 0, 48 * 1024);
857 	else
858 		/* 60k words of 4x12-bit pixels */
859 		drm_mm_init(&hvs->lbm_mm, 0, 60 * 1024);
860 
861 	/* Upload filter kernels.  We only have the one for now, so we
862 	 * keep it around for the lifetime of the driver.
863 	 */
864 	ret = vc4_hvs_upload_linear_kernel(hvs,
865 					   &hvs->mitchell_netravali_filter,
866 					   mitchell_netravali_1_3_1_3_kernel);
867 	if (ret)
868 		return ret;
869 
870 	vc4->hvs = hvs;
871 
872 	reg = HVS_READ(SCALER_DISPECTRL);
873 	reg &= ~SCALER_DISPECTRL_DSP2_MUX_MASK;
874 	HVS_WRITE(SCALER_DISPECTRL,
875 		  reg | VC4_SET_FIELD(0, SCALER_DISPECTRL_DSP2_MUX));
876 
877 	reg = HVS_READ(SCALER_DISPCTRL);
878 	reg &= ~SCALER_DISPCTRL_DSP3_MUX_MASK;
879 	HVS_WRITE(SCALER_DISPCTRL,
880 		  reg | VC4_SET_FIELD(3, SCALER_DISPCTRL_DSP3_MUX));
881 
882 	reg = HVS_READ(SCALER_DISPEOLN);
883 	reg &= ~SCALER_DISPEOLN_DSP4_MUX_MASK;
884 	HVS_WRITE(SCALER_DISPEOLN,
885 		  reg | VC4_SET_FIELD(3, SCALER_DISPEOLN_DSP4_MUX));
886 
887 	reg = HVS_READ(SCALER_DISPDITHER);
888 	reg &= ~SCALER_DISPDITHER_DSP5_MUX_MASK;
889 	HVS_WRITE(SCALER_DISPDITHER,
890 		  reg | VC4_SET_FIELD(3, SCALER_DISPDITHER_DSP5_MUX));
891 
892 	dispctrl = HVS_READ(SCALER_DISPCTRL);
893 
894 	dispctrl |= SCALER_DISPCTRL_ENABLE;
895 	dispctrl |= SCALER_DISPCTRL_DISPEIRQ(0) |
896 		    SCALER_DISPCTRL_DISPEIRQ(1) |
897 		    SCALER_DISPCTRL_DISPEIRQ(2);
898 
899 	dispctrl &= ~(SCALER_DISPCTRL_DMAEIRQ |
900 		      SCALER_DISPCTRL_SLVWREIRQ |
901 		      SCALER_DISPCTRL_SLVRDEIRQ |
902 		      SCALER_DISPCTRL_DSPEIEOF(0) |
903 		      SCALER_DISPCTRL_DSPEIEOF(1) |
904 		      SCALER_DISPCTRL_DSPEIEOF(2) |
905 		      SCALER_DISPCTRL_DSPEIEOLN(0) |
906 		      SCALER_DISPCTRL_DSPEIEOLN(1) |
907 		      SCALER_DISPCTRL_DSPEIEOLN(2) |
908 		      SCALER_DISPCTRL_DSPEISLUR(0) |
909 		      SCALER_DISPCTRL_DSPEISLUR(1) |
910 		      SCALER_DISPCTRL_DSPEISLUR(2) |
911 		      SCALER_DISPCTRL_SCLEIRQ);
912 
913 	HVS_WRITE(SCALER_DISPCTRL, dispctrl);
914 
915 	ret = devm_request_irq(dev, platform_get_irq(pdev, 0),
916 			       vc4_hvs_irq_handler, 0, "vc4 hvs", drm);
917 	if (ret)
918 		return ret;
919 
920 	return 0;
921 }
922 
923 static void vc4_hvs_unbind(struct device *dev, struct device *master,
924 			   void *data)
925 {
926 	struct drm_device *drm = dev_get_drvdata(master);
927 	struct vc4_dev *vc4 = to_vc4_dev(drm);
928 	struct vc4_hvs *hvs = vc4->hvs;
929 	struct drm_mm_node *node, *next;
930 
931 	if (drm_mm_node_allocated(&vc4->hvs->mitchell_netravali_filter))
932 		drm_mm_remove_node(&vc4->hvs->mitchell_netravali_filter);
933 
934 	drm_mm_for_each_node_safe(node, next, &vc4->hvs->dlist_mm)
935 		drm_mm_remove_node(node);
936 
937 	drm_mm_takedown(&vc4->hvs->dlist_mm);
938 
939 	drm_mm_for_each_node_safe(node, next, &vc4->hvs->lbm_mm)
940 		drm_mm_remove_node(node);
941 	drm_mm_takedown(&vc4->hvs->lbm_mm);
942 
943 	clk_disable_unprepare(hvs->core_clk);
944 
945 	vc4->hvs = NULL;
946 }
947 
948 static const struct component_ops vc4_hvs_ops = {
949 	.bind   = vc4_hvs_bind,
950 	.unbind = vc4_hvs_unbind,
951 };
952 
953 static int vc4_hvs_dev_probe(struct platform_device *pdev)
954 {
955 	return component_add(&pdev->dev, &vc4_hvs_ops);
956 }
957 
958 static int vc4_hvs_dev_remove(struct platform_device *pdev)
959 {
960 	component_del(&pdev->dev, &vc4_hvs_ops);
961 	return 0;
962 }
963 
964 static const struct of_device_id vc4_hvs_dt_match[] = {
965 	{ .compatible = "brcm,bcm2711-hvs" },
966 	{ .compatible = "brcm,bcm2835-hvs" },
967 	{}
968 };
969 
970 struct platform_driver vc4_hvs_driver = {
971 	.probe = vc4_hvs_dev_probe,
972 	.remove = vc4_hvs_dev_remove,
973 	.driver = {
974 		.name = "vc4_hvs",
975 		.of_match_table = vc4_hvs_dt_match,
976 	},
977 };
978