xref: /openbmc/linux/drivers/gpu/drm/vc4/vc4_hvs.c (revision dd5e5554)
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 /* The filter kernel is composed of dwords each containing 3 9-bit
98  * signed integers packed next to each other.
99  */
100 #define VC4_INT_TO_COEFF(coeff) (coeff & 0x1ff)
101 #define VC4_PPF_FILTER_WORD(c0, c1, c2)				\
102 	((((c0) & 0x1ff) << 0) |				\
103 	 (((c1) & 0x1ff) << 9) |				\
104 	 (((c2) & 0x1ff) << 18))
105 
106 /* The whole filter kernel is arranged as the coefficients 0-16 going
107  * up, then a pad, then 17-31 going down and reversed within the
108  * dwords.  This means that a linear phase kernel (where it's
109  * symmetrical at the boundary between 15 and 16) has the last 5
110  * dwords matching the first 5, but reversed.
111  */
112 #define VC4_LINEAR_PHASE_KERNEL(c0, c1, c2, c3, c4, c5, c6, c7, c8,	\
113 				c9, c10, c11, c12, c13, c14, c15)	\
114 	{VC4_PPF_FILTER_WORD(c0, c1, c2),				\
115 	 VC4_PPF_FILTER_WORD(c3, c4, c5),				\
116 	 VC4_PPF_FILTER_WORD(c6, c7, c8),				\
117 	 VC4_PPF_FILTER_WORD(c9, c10, c11),				\
118 	 VC4_PPF_FILTER_WORD(c12, c13, c14),				\
119 	 VC4_PPF_FILTER_WORD(c15, c15, 0)}
120 
121 #define VC4_LINEAR_PHASE_KERNEL_DWORDS 6
122 #define VC4_KERNEL_DWORDS (VC4_LINEAR_PHASE_KERNEL_DWORDS * 2 - 1)
123 
124 /* Recommended B=1/3, C=1/3 filter choice from Mitchell/Netravali.
125  * http://www.cs.utexas.edu/~fussell/courses/cs384g/lectures/mitchell/Mitchell.pdf
126  */
127 static const u32 mitchell_netravali_1_3_1_3_kernel[] =
128 	VC4_LINEAR_PHASE_KERNEL(0, -2, -6, -8, -10, -8, -3, 2, 18,
129 				50, 82, 119, 155, 187, 213, 227);
130 
131 static int vc4_hvs_upload_linear_kernel(struct vc4_hvs *hvs,
132 					struct drm_mm_node *space,
133 					const u32 *kernel)
134 {
135 	int ret, i;
136 	u32 __iomem *dst_kernel;
137 
138 	ret = drm_mm_insert_node(&hvs->dlist_mm, space, VC4_KERNEL_DWORDS);
139 	if (ret) {
140 		DRM_ERROR("Failed to allocate space for filter kernel: %d\n",
141 			  ret);
142 		return ret;
143 	}
144 
145 	dst_kernel = hvs->dlist + space->start;
146 
147 	for (i = 0; i < VC4_KERNEL_DWORDS; i++) {
148 		if (i < VC4_LINEAR_PHASE_KERNEL_DWORDS)
149 			writel(kernel[i], &dst_kernel[i]);
150 		else {
151 			writel(kernel[VC4_KERNEL_DWORDS - i - 1],
152 			       &dst_kernel[i]);
153 		}
154 	}
155 
156 	return 0;
157 }
158 
159 static void vc4_hvs_lut_load(struct vc4_hvs *hvs,
160 			     struct vc4_crtc *vc4_crtc)
161 {
162 	struct drm_crtc *crtc = &vc4_crtc->base;
163 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
164 	u32 i;
165 
166 	/* The LUT memory is laid out with each HVS channel in order,
167 	 * each of which takes 256 writes for R, 256 for G, then 256
168 	 * for B.
169 	 */
170 	HVS_WRITE(SCALER_GAMADDR,
171 		  SCALER_GAMADDR_AUTOINC |
172 		  (vc4_state->assigned_channel * 3 * crtc->gamma_size));
173 
174 	for (i = 0; i < crtc->gamma_size; i++)
175 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_r[i]);
176 	for (i = 0; i < crtc->gamma_size; i++)
177 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_g[i]);
178 	for (i = 0; i < crtc->gamma_size; i++)
179 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_b[i]);
180 }
181 
182 static void vc4_hvs_update_gamma_lut(struct vc4_hvs *hvs,
183 				     struct vc4_crtc *vc4_crtc)
184 {
185 	struct drm_crtc_state *crtc_state = vc4_crtc->base.state;
186 	struct drm_color_lut *lut = crtc_state->gamma_lut->data;
187 	u32 length = drm_color_lut_size(crtc_state->gamma_lut);
188 	u32 i;
189 
190 	for (i = 0; i < length; i++) {
191 		vc4_crtc->lut_r[i] = drm_color_lut_extract(lut[i].red, 8);
192 		vc4_crtc->lut_g[i] = drm_color_lut_extract(lut[i].green, 8);
193 		vc4_crtc->lut_b[i] = drm_color_lut_extract(lut[i].blue, 8);
194 	}
195 
196 	vc4_hvs_lut_load(hvs, vc4_crtc);
197 }
198 
199 u8 vc4_hvs_get_fifo_frame_count(struct vc4_hvs *hvs, unsigned int fifo)
200 {
201 	u8 field = 0;
202 
203 	switch (fifo) {
204 	case 0:
205 		field = VC4_GET_FIELD(HVS_READ(SCALER_DISPSTAT1),
206 				      SCALER_DISPSTAT1_FRCNT0);
207 		break;
208 	case 1:
209 		field = VC4_GET_FIELD(HVS_READ(SCALER_DISPSTAT1),
210 				      SCALER_DISPSTAT1_FRCNT1);
211 		break;
212 	case 2:
213 		field = VC4_GET_FIELD(HVS_READ(SCALER_DISPSTAT2),
214 				      SCALER_DISPSTAT2_FRCNT2);
215 		break;
216 	}
217 
218 	return field;
219 }
220 
221 int vc4_hvs_get_fifo_from_output(struct vc4_hvs *hvs, unsigned int output)
222 {
223 	struct vc4_dev *vc4 = hvs->vc4;
224 	u32 reg;
225 	int ret;
226 
227 	if (!vc4->is_vc5)
228 		return output;
229 
230 	switch (output) {
231 	case 0:
232 		return 0;
233 
234 	case 1:
235 		return 1;
236 
237 	case 2:
238 		reg = HVS_READ(SCALER_DISPECTRL);
239 		ret = FIELD_GET(SCALER_DISPECTRL_DSP2_MUX_MASK, reg);
240 		if (ret == 0)
241 			return 2;
242 
243 		return 0;
244 
245 	case 3:
246 		reg = HVS_READ(SCALER_DISPCTRL);
247 		ret = FIELD_GET(SCALER_DISPCTRL_DSP3_MUX_MASK, reg);
248 		if (ret == 3)
249 			return -EPIPE;
250 
251 		return ret;
252 
253 	case 4:
254 		reg = HVS_READ(SCALER_DISPEOLN);
255 		ret = FIELD_GET(SCALER_DISPEOLN_DSP4_MUX_MASK, reg);
256 		if (ret == 3)
257 			return -EPIPE;
258 
259 		return ret;
260 
261 	case 5:
262 		reg = HVS_READ(SCALER_DISPDITHER);
263 		ret = FIELD_GET(SCALER_DISPDITHER_DSP5_MUX_MASK, reg);
264 		if (ret == 3)
265 			return -EPIPE;
266 
267 		return ret;
268 
269 	default:
270 		return -EPIPE;
271 	}
272 }
273 
274 static int vc4_hvs_init_channel(struct vc4_hvs *hvs, struct drm_crtc *crtc,
275 				struct drm_display_mode *mode, bool oneshot)
276 {
277 	struct vc4_dev *vc4 = hvs->vc4;
278 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
279 	struct vc4_crtc_state *vc4_crtc_state = to_vc4_crtc_state(crtc->state);
280 	unsigned int chan = vc4_crtc_state->assigned_channel;
281 	bool interlace = mode->flags & DRM_MODE_FLAG_INTERLACE;
282 	u32 dispbkgndx;
283 	u32 dispctrl;
284 
285 	HVS_WRITE(SCALER_DISPCTRLX(chan), 0);
286 	HVS_WRITE(SCALER_DISPCTRLX(chan), SCALER_DISPCTRLX_RESET);
287 	HVS_WRITE(SCALER_DISPCTRLX(chan), 0);
288 
289 	/* Turn on the scaler, which will wait for vstart to start
290 	 * compositing.
291 	 * When feeding the transposer, we should operate in oneshot
292 	 * mode.
293 	 */
294 	dispctrl = SCALER_DISPCTRLX_ENABLE;
295 
296 	if (!vc4->is_vc5)
297 		dispctrl |= VC4_SET_FIELD(mode->hdisplay,
298 					  SCALER_DISPCTRLX_WIDTH) |
299 			    VC4_SET_FIELD(mode->vdisplay,
300 					  SCALER_DISPCTRLX_HEIGHT) |
301 			    (oneshot ? SCALER_DISPCTRLX_ONESHOT : 0);
302 	else
303 		dispctrl |= VC4_SET_FIELD(mode->hdisplay,
304 					  SCALER5_DISPCTRLX_WIDTH) |
305 			    VC4_SET_FIELD(mode->vdisplay,
306 					  SCALER5_DISPCTRLX_HEIGHT) |
307 			    (oneshot ? SCALER5_DISPCTRLX_ONESHOT : 0);
308 
309 	HVS_WRITE(SCALER_DISPCTRLX(chan), dispctrl);
310 
311 	dispbkgndx = HVS_READ(SCALER_DISPBKGNDX(chan));
312 	dispbkgndx &= ~SCALER_DISPBKGND_GAMMA;
313 	dispbkgndx &= ~SCALER_DISPBKGND_INTERLACE;
314 
315 	HVS_WRITE(SCALER_DISPBKGNDX(chan), dispbkgndx |
316 		  SCALER_DISPBKGND_AUTOHS |
317 		  ((!vc4->is_vc5) ? SCALER_DISPBKGND_GAMMA : 0) |
318 		  (interlace ? SCALER_DISPBKGND_INTERLACE : 0));
319 
320 	/* Reload the LUT, since the SRAMs would have been disabled if
321 	 * all CRTCs had SCALER_DISPBKGND_GAMMA unset at once.
322 	 */
323 	vc4_hvs_lut_load(hvs, vc4_crtc);
324 
325 	return 0;
326 }
327 
328 void vc4_hvs_stop_channel(struct vc4_hvs *hvs, unsigned int chan)
329 {
330 	if (HVS_READ(SCALER_DISPCTRLX(chan)) & SCALER_DISPCTRLX_ENABLE)
331 		return;
332 
333 	HVS_WRITE(SCALER_DISPCTRLX(chan),
334 		  HVS_READ(SCALER_DISPCTRLX(chan)) | SCALER_DISPCTRLX_RESET);
335 	HVS_WRITE(SCALER_DISPCTRLX(chan),
336 		  HVS_READ(SCALER_DISPCTRLX(chan)) & ~SCALER_DISPCTRLX_ENABLE);
337 
338 	/* Once we leave, the scaler should be disabled and its fifo empty. */
339 	WARN_ON_ONCE(HVS_READ(SCALER_DISPCTRLX(chan)) & SCALER_DISPCTRLX_RESET);
340 
341 	WARN_ON_ONCE(VC4_GET_FIELD(HVS_READ(SCALER_DISPSTATX(chan)),
342 				   SCALER_DISPSTATX_MODE) !=
343 		     SCALER_DISPSTATX_MODE_DISABLED);
344 
345 	WARN_ON_ONCE((HVS_READ(SCALER_DISPSTATX(chan)) &
346 		      (SCALER_DISPSTATX_FULL | SCALER_DISPSTATX_EMPTY)) !=
347 		     SCALER_DISPSTATX_EMPTY);
348 }
349 
350 int vc4_hvs_atomic_check(struct drm_crtc *crtc, struct drm_atomic_state *state)
351 {
352 	struct drm_crtc_state *crtc_state = drm_atomic_get_new_crtc_state(state, crtc);
353 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc_state);
354 	struct drm_device *dev = crtc->dev;
355 	struct vc4_dev *vc4 = to_vc4_dev(dev);
356 	struct drm_plane *plane;
357 	unsigned long flags;
358 	const struct drm_plane_state *plane_state;
359 	u32 dlist_count = 0;
360 	int ret;
361 
362 	/* The pixelvalve can only feed one encoder (and encoders are
363 	 * 1:1 with connectors.)
364 	 */
365 	if (hweight32(crtc_state->connector_mask) > 1)
366 		return -EINVAL;
367 
368 	drm_atomic_crtc_state_for_each_plane_state(plane, plane_state, crtc_state)
369 		dlist_count += vc4_plane_dlist_size(plane_state);
370 
371 	dlist_count++; /* Account for SCALER_CTL0_END. */
372 
373 	spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
374 	ret = drm_mm_insert_node(&vc4->hvs->dlist_mm, &vc4_state->mm,
375 				 dlist_count);
376 	spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
377 	if (ret)
378 		return ret;
379 
380 	return 0;
381 }
382 
383 static void vc4_hvs_install_dlist(struct drm_crtc *crtc)
384 {
385 	struct drm_device *dev = crtc->dev;
386 	struct vc4_dev *vc4 = to_vc4_dev(dev);
387 	struct vc4_hvs *hvs = vc4->hvs;
388 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
389 
390 	HVS_WRITE(SCALER_DISPLISTX(vc4_state->assigned_channel),
391 		  vc4_state->mm.start);
392 }
393 
394 static void vc4_hvs_update_dlist(struct drm_crtc *crtc)
395 {
396 	struct drm_device *dev = crtc->dev;
397 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
398 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
399 	unsigned long flags;
400 
401 	if (crtc->state->event) {
402 		crtc->state->event->pipe = drm_crtc_index(crtc);
403 
404 		WARN_ON(drm_crtc_vblank_get(crtc) != 0);
405 
406 		spin_lock_irqsave(&dev->event_lock, flags);
407 
408 		if (!vc4_crtc->feeds_txp || vc4_state->txp_armed) {
409 			vc4_crtc->event = crtc->state->event;
410 			crtc->state->event = NULL;
411 		}
412 
413 		spin_unlock_irqrestore(&dev->event_lock, flags);
414 	}
415 
416 	spin_lock_irqsave(&vc4_crtc->irq_lock, flags);
417 	vc4_crtc->current_dlist = vc4_state->mm.start;
418 	spin_unlock_irqrestore(&vc4_crtc->irq_lock, flags);
419 }
420 
421 void vc4_hvs_atomic_begin(struct drm_crtc *crtc,
422 			  struct drm_atomic_state *state)
423 {
424 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
425 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
426 	unsigned long flags;
427 
428 	spin_lock_irqsave(&vc4_crtc->irq_lock, flags);
429 	vc4_crtc->current_hvs_channel = vc4_state->assigned_channel;
430 	spin_unlock_irqrestore(&vc4_crtc->irq_lock, flags);
431 }
432 
433 void vc4_hvs_atomic_enable(struct drm_crtc *crtc,
434 			   struct drm_atomic_state *state)
435 {
436 	struct drm_device *dev = crtc->dev;
437 	struct vc4_dev *vc4 = to_vc4_dev(dev);
438 	struct drm_display_mode *mode = &crtc->state->adjusted_mode;
439 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
440 	bool oneshot = vc4_crtc->feeds_txp;
441 
442 	vc4_hvs_install_dlist(crtc);
443 	vc4_hvs_update_dlist(crtc);
444 	vc4_hvs_init_channel(vc4->hvs, crtc, mode, oneshot);
445 }
446 
447 void vc4_hvs_atomic_disable(struct drm_crtc *crtc,
448 			    struct drm_atomic_state *state)
449 {
450 	struct drm_device *dev = crtc->dev;
451 	struct vc4_dev *vc4 = to_vc4_dev(dev);
452 	struct drm_crtc_state *old_state = drm_atomic_get_old_crtc_state(state, crtc);
453 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(old_state);
454 	unsigned int chan = vc4_state->assigned_channel;
455 
456 	vc4_hvs_stop_channel(vc4->hvs, chan);
457 }
458 
459 void vc4_hvs_atomic_flush(struct drm_crtc *crtc,
460 			  struct drm_atomic_state *state)
461 {
462 	struct drm_crtc_state *old_state = drm_atomic_get_old_crtc_state(state,
463 									 crtc);
464 	struct drm_device *dev = crtc->dev;
465 	struct vc4_dev *vc4 = to_vc4_dev(dev);
466 	struct vc4_hvs *hvs = vc4->hvs;
467 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
468 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
469 	unsigned int channel = vc4_state->assigned_channel;
470 	struct drm_plane *plane;
471 	struct vc4_plane_state *vc4_plane_state;
472 	bool debug_dump_regs = false;
473 	bool enable_bg_fill = false;
474 	u32 __iomem *dlist_start = vc4->hvs->dlist + vc4_state->mm.start;
475 	u32 __iomem *dlist_next = dlist_start;
476 
477 	if (debug_dump_regs) {
478 		DRM_INFO("CRTC %d HVS before:\n", drm_crtc_index(crtc));
479 		vc4_hvs_dump_state(hvs);
480 	}
481 
482 	/* Copy all the active planes' dlist contents to the hardware dlist. */
483 	drm_atomic_crtc_for_each_plane(plane, crtc) {
484 		/* Is this the first active plane? */
485 		if (dlist_next == dlist_start) {
486 			/* We need to enable background fill when a plane
487 			 * could be alpha blending from the background, i.e.
488 			 * where no other plane is underneath. It suffices to
489 			 * consider the first active plane here since we set
490 			 * needs_bg_fill such that either the first plane
491 			 * already needs it or all planes on top blend from
492 			 * the first or a lower plane.
493 			 */
494 			vc4_plane_state = to_vc4_plane_state(plane->state);
495 			enable_bg_fill = vc4_plane_state->needs_bg_fill;
496 		}
497 
498 		dlist_next += vc4_plane_write_dlist(plane, dlist_next);
499 	}
500 
501 	writel(SCALER_CTL0_END, dlist_next);
502 	dlist_next++;
503 
504 	WARN_ON_ONCE(dlist_next - dlist_start != vc4_state->mm.size);
505 
506 	if (enable_bg_fill)
507 		/* This sets a black background color fill, as is the case
508 		 * with other DRM drivers.
509 		 */
510 		HVS_WRITE(SCALER_DISPBKGNDX(channel),
511 			  HVS_READ(SCALER_DISPBKGNDX(channel)) |
512 			  SCALER_DISPBKGND_FILL);
513 
514 	/* Only update DISPLIST if the CRTC was already running and is not
515 	 * being disabled.
516 	 * vc4_crtc_enable() takes care of updating the dlist just after
517 	 * re-enabling VBLANK interrupts and before enabling the engine.
518 	 * If the CRTC is being disabled, there's no point in updating this
519 	 * information.
520 	 */
521 	if (crtc->state->active && old_state->active) {
522 		vc4_hvs_install_dlist(crtc);
523 		vc4_hvs_update_dlist(crtc);
524 	}
525 
526 	if (crtc->state->color_mgmt_changed) {
527 		u32 dispbkgndx = HVS_READ(SCALER_DISPBKGNDX(channel));
528 
529 		if (crtc->state->gamma_lut) {
530 			vc4_hvs_update_gamma_lut(hvs, vc4_crtc);
531 			dispbkgndx |= SCALER_DISPBKGND_GAMMA;
532 		} else {
533 			/* Unsetting DISPBKGND_GAMMA skips the gamma lut step
534 			 * in hardware, which is the same as a linear lut that
535 			 * DRM expects us to use in absence of a user lut.
536 			 */
537 			dispbkgndx &= ~SCALER_DISPBKGND_GAMMA;
538 		}
539 		HVS_WRITE(SCALER_DISPBKGNDX(channel), dispbkgndx);
540 	}
541 
542 	if (debug_dump_regs) {
543 		DRM_INFO("CRTC %d HVS after:\n", drm_crtc_index(crtc));
544 		vc4_hvs_dump_state(hvs);
545 	}
546 }
547 
548 void vc4_hvs_mask_underrun(struct vc4_hvs *hvs, int channel)
549 {
550 	u32 dispctrl = HVS_READ(SCALER_DISPCTRL);
551 
552 	dispctrl &= ~SCALER_DISPCTRL_DSPEISLUR(channel);
553 
554 	HVS_WRITE(SCALER_DISPCTRL, dispctrl);
555 }
556 
557 void vc4_hvs_unmask_underrun(struct vc4_hvs *hvs, int channel)
558 {
559 	u32 dispctrl = HVS_READ(SCALER_DISPCTRL);
560 
561 	dispctrl |= SCALER_DISPCTRL_DSPEISLUR(channel);
562 
563 	HVS_WRITE(SCALER_DISPSTAT,
564 		  SCALER_DISPSTAT_EUFLOW(channel));
565 	HVS_WRITE(SCALER_DISPCTRL, dispctrl);
566 }
567 
568 static void vc4_hvs_report_underrun(struct drm_device *dev)
569 {
570 	struct vc4_dev *vc4 = to_vc4_dev(dev);
571 
572 	atomic_inc(&vc4->underrun);
573 	DRM_DEV_ERROR(dev->dev, "HVS underrun\n");
574 }
575 
576 static irqreturn_t vc4_hvs_irq_handler(int irq, void *data)
577 {
578 	struct drm_device *dev = data;
579 	struct vc4_dev *vc4 = to_vc4_dev(dev);
580 	struct vc4_hvs *hvs = vc4->hvs;
581 	irqreturn_t irqret = IRQ_NONE;
582 	int channel;
583 	u32 control;
584 	u32 status;
585 
586 	status = HVS_READ(SCALER_DISPSTAT);
587 	control = HVS_READ(SCALER_DISPCTRL);
588 
589 	for (channel = 0; channel < SCALER_CHANNELS_COUNT; channel++) {
590 		/* Interrupt masking is not always honored, so check it here. */
591 		if (status & SCALER_DISPSTAT_EUFLOW(channel) &&
592 		    control & SCALER_DISPCTRL_DSPEISLUR(channel)) {
593 			vc4_hvs_mask_underrun(hvs, channel);
594 			vc4_hvs_report_underrun(dev);
595 
596 			irqret = IRQ_HANDLED;
597 		}
598 	}
599 
600 	/* Clear every per-channel interrupt flag. */
601 	HVS_WRITE(SCALER_DISPSTAT, SCALER_DISPSTAT_IRQMASK(0) |
602 				   SCALER_DISPSTAT_IRQMASK(1) |
603 				   SCALER_DISPSTAT_IRQMASK(2));
604 
605 	return irqret;
606 }
607 
608 static int vc4_hvs_bind(struct device *dev, struct device *master, void *data)
609 {
610 	struct platform_device *pdev = to_platform_device(dev);
611 	struct drm_device *drm = dev_get_drvdata(master);
612 	struct vc4_dev *vc4 = to_vc4_dev(drm);
613 	struct vc4_hvs *hvs = NULL;
614 	int ret;
615 	u32 dispctrl;
616 	u32 reg;
617 
618 	hvs = devm_kzalloc(&pdev->dev, sizeof(*hvs), GFP_KERNEL);
619 	if (!hvs)
620 		return -ENOMEM;
621 
622 	hvs->vc4 = vc4;
623 	hvs->pdev = pdev;
624 
625 	hvs->regs = vc4_ioremap_regs(pdev, 0);
626 	if (IS_ERR(hvs->regs))
627 		return PTR_ERR(hvs->regs);
628 
629 	hvs->regset.base = hvs->regs;
630 	hvs->regset.regs = hvs_regs;
631 	hvs->regset.nregs = ARRAY_SIZE(hvs_regs);
632 
633 	if (vc4->is_vc5) {
634 		hvs->core_clk = devm_clk_get(&pdev->dev, NULL);
635 		if (IS_ERR(hvs->core_clk)) {
636 			dev_err(&pdev->dev, "Couldn't get core clock\n");
637 			return PTR_ERR(hvs->core_clk);
638 		}
639 
640 		ret = clk_prepare_enable(hvs->core_clk);
641 		if (ret) {
642 			dev_err(&pdev->dev, "Couldn't enable the core clock\n");
643 			return ret;
644 		}
645 	}
646 
647 	if (!vc4->is_vc5)
648 		hvs->dlist = hvs->regs + SCALER_DLIST_START;
649 	else
650 		hvs->dlist = hvs->regs + SCALER5_DLIST_START;
651 
652 	spin_lock_init(&hvs->mm_lock);
653 
654 	/* Set up the HVS display list memory manager.  We never
655 	 * overwrite the setup from the bootloader (just 128b out of
656 	 * our 16K), since we don't want to scramble the screen when
657 	 * transitioning from the firmware's boot setup to runtime.
658 	 */
659 	drm_mm_init(&hvs->dlist_mm,
660 		    HVS_BOOTLOADER_DLIST_END,
661 		    (SCALER_DLIST_SIZE >> 2) - HVS_BOOTLOADER_DLIST_END);
662 
663 	/* Set up the HVS LBM memory manager.  We could have some more
664 	 * complicated data structure that allowed reuse of LBM areas
665 	 * between planes when they don't overlap on the screen, but
666 	 * for now we just allocate globally.
667 	 */
668 	if (!vc4->is_vc5)
669 		/* 48k words of 2x12-bit pixels */
670 		drm_mm_init(&hvs->lbm_mm, 0, 48 * 1024);
671 	else
672 		/* 60k words of 4x12-bit pixels */
673 		drm_mm_init(&hvs->lbm_mm, 0, 60 * 1024);
674 
675 	/* Upload filter kernels.  We only have the one for now, so we
676 	 * keep it around for the lifetime of the driver.
677 	 */
678 	ret = vc4_hvs_upload_linear_kernel(hvs,
679 					   &hvs->mitchell_netravali_filter,
680 					   mitchell_netravali_1_3_1_3_kernel);
681 	if (ret)
682 		return ret;
683 
684 	vc4->hvs = hvs;
685 
686 	reg = HVS_READ(SCALER_DISPECTRL);
687 	reg &= ~SCALER_DISPECTRL_DSP2_MUX_MASK;
688 	HVS_WRITE(SCALER_DISPECTRL,
689 		  reg | VC4_SET_FIELD(0, SCALER_DISPECTRL_DSP2_MUX));
690 
691 	reg = HVS_READ(SCALER_DISPCTRL);
692 	reg &= ~SCALER_DISPCTRL_DSP3_MUX_MASK;
693 	HVS_WRITE(SCALER_DISPCTRL,
694 		  reg | VC4_SET_FIELD(3, SCALER_DISPCTRL_DSP3_MUX));
695 
696 	reg = HVS_READ(SCALER_DISPEOLN);
697 	reg &= ~SCALER_DISPEOLN_DSP4_MUX_MASK;
698 	HVS_WRITE(SCALER_DISPEOLN,
699 		  reg | VC4_SET_FIELD(3, SCALER_DISPEOLN_DSP4_MUX));
700 
701 	reg = HVS_READ(SCALER_DISPDITHER);
702 	reg &= ~SCALER_DISPDITHER_DSP5_MUX_MASK;
703 	HVS_WRITE(SCALER_DISPDITHER,
704 		  reg | VC4_SET_FIELD(3, SCALER_DISPDITHER_DSP5_MUX));
705 
706 	dispctrl = HVS_READ(SCALER_DISPCTRL);
707 
708 	dispctrl |= SCALER_DISPCTRL_ENABLE;
709 	dispctrl |= SCALER_DISPCTRL_DISPEIRQ(0) |
710 		    SCALER_DISPCTRL_DISPEIRQ(1) |
711 		    SCALER_DISPCTRL_DISPEIRQ(2);
712 
713 	dispctrl &= ~(SCALER_DISPCTRL_DMAEIRQ |
714 		      SCALER_DISPCTRL_SLVWREIRQ |
715 		      SCALER_DISPCTRL_SLVRDEIRQ |
716 		      SCALER_DISPCTRL_DSPEIEOF(0) |
717 		      SCALER_DISPCTRL_DSPEIEOF(1) |
718 		      SCALER_DISPCTRL_DSPEIEOF(2) |
719 		      SCALER_DISPCTRL_DSPEIEOLN(0) |
720 		      SCALER_DISPCTRL_DSPEIEOLN(1) |
721 		      SCALER_DISPCTRL_DSPEIEOLN(2) |
722 		      SCALER_DISPCTRL_DSPEISLUR(0) |
723 		      SCALER_DISPCTRL_DSPEISLUR(1) |
724 		      SCALER_DISPCTRL_DSPEISLUR(2) |
725 		      SCALER_DISPCTRL_SCLEIRQ);
726 
727 	HVS_WRITE(SCALER_DISPCTRL, dispctrl);
728 
729 	ret = devm_request_irq(dev, platform_get_irq(pdev, 0),
730 			       vc4_hvs_irq_handler, 0, "vc4 hvs", drm);
731 	if (ret)
732 		return ret;
733 
734 	vc4_debugfs_add_regset32(drm, "hvs_regs", &hvs->regset);
735 	vc4_debugfs_add_file(drm, "hvs_underrun", vc4_hvs_debugfs_underrun,
736 			     NULL);
737 
738 	return 0;
739 }
740 
741 static void vc4_hvs_unbind(struct device *dev, struct device *master,
742 			   void *data)
743 {
744 	struct drm_device *drm = dev_get_drvdata(master);
745 	struct vc4_dev *vc4 = to_vc4_dev(drm);
746 	struct vc4_hvs *hvs = vc4->hvs;
747 
748 	if (drm_mm_node_allocated(&vc4->hvs->mitchell_netravali_filter))
749 		drm_mm_remove_node(&vc4->hvs->mitchell_netravali_filter);
750 
751 	drm_mm_takedown(&vc4->hvs->dlist_mm);
752 	drm_mm_takedown(&vc4->hvs->lbm_mm);
753 
754 	clk_disable_unprepare(hvs->core_clk);
755 
756 	vc4->hvs = NULL;
757 }
758 
759 static const struct component_ops vc4_hvs_ops = {
760 	.bind   = vc4_hvs_bind,
761 	.unbind = vc4_hvs_unbind,
762 };
763 
764 static int vc4_hvs_dev_probe(struct platform_device *pdev)
765 {
766 	return component_add(&pdev->dev, &vc4_hvs_ops);
767 }
768 
769 static int vc4_hvs_dev_remove(struct platform_device *pdev)
770 {
771 	component_del(&pdev->dev, &vc4_hvs_ops);
772 	return 0;
773 }
774 
775 static const struct of_device_id vc4_hvs_dt_match[] = {
776 	{ .compatible = "brcm,bcm2711-hvs" },
777 	{ .compatible = "brcm,bcm2835-hvs" },
778 	{}
779 };
780 
781 struct platform_driver vc4_hvs_driver = {
782 	.probe = vc4_hvs_dev_probe,
783 	.remove = vc4_hvs_dev_remove,
784 	.driver = {
785 		.name = "vc4_hvs",
786 		.of_match_table = vc4_hvs_dt_match,
787 	},
788 };
789