xref: /linux/drivers/gpu/drm/vc4/vc4_crtc.c (revision dd9a41bc61cc62d38306465ed62373b98df0049e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2015 Broadcom
4  */
5 
6 /**
7  * DOC: VC4 CRTC module
8  *
9  * In VC4, the Pixel Valve is what most closely corresponds to the
10  * DRM's concept of a CRTC.  The PV generates video timings from the
11  * encoder's clock plus its configuration.  It pulls scaled pixels from
12  * the HVS at that timing, and feeds it to the encoder.
13  *
14  * However, the DRM CRTC also collects the configuration of all the
15  * DRM planes attached to it.  As a result, the CRTC is also
16  * responsible for writing the display list for the HVS channel that
17  * the CRTC will use.
18  *
19  * The 2835 has 3 different pixel valves.  pv0 in the audio power
20  * domain feeds DSI0 or DPI, while pv1 feeds DS1 or SMI.  pv2 in the
21  * image domain can feed either HDMI or the SDTV controller.  The
22  * pixel valve chooses from the CPRMAN clocks (HSM for HDMI, VEC for
23  * SDTV, etc.) according to which output type is chosen in the mux.
24  *
25  * For power management, the pixel valve's registers are all clocked
26  * by the AXI clock, while the timings and FIFOs make use of the
27  * output-specific clock.  Since the encoders also directly consume
28  * the CPRMAN clocks, and know what timings they need, they are the
29  * ones that set the clock.
30  */
31 
32 #include <linux/clk.h>
33 #include <linux/component.h>
34 #include <linux/of_device.h>
35 
36 #include <drm/drm_atomic.h>
37 #include <drm/drm_atomic_helper.h>
38 #include <drm/drm_atomic_uapi.h>
39 #include <drm/drm_fb_cma_helper.h>
40 #include <drm/drm_print.h>
41 #include <drm/drm_probe_helper.h>
42 #include <drm/drm_vblank.h>
43 
44 #include "vc4_drv.h"
45 #include "vc4_regs.h"
46 
47 struct vc4_crtc_state {
48 	struct drm_crtc_state base;
49 	/* Dlist area for this CRTC configuration. */
50 	struct drm_mm_node mm;
51 	bool feed_txp;
52 	bool txp_armed;
53 
54 	struct {
55 		unsigned int left;
56 		unsigned int right;
57 		unsigned int top;
58 		unsigned int bottom;
59 	} margins;
60 };
61 
62 static inline struct vc4_crtc_state *
63 to_vc4_crtc_state(struct drm_crtc_state *crtc_state)
64 {
65 	return (struct vc4_crtc_state *)crtc_state;
66 }
67 
68 #define CRTC_WRITE(offset, val) writel(val, vc4_crtc->regs + (offset))
69 #define CRTC_READ(offset) readl(vc4_crtc->regs + (offset))
70 
71 static const struct debugfs_reg32 crtc_regs[] = {
72 	VC4_REG32(PV_CONTROL),
73 	VC4_REG32(PV_V_CONTROL),
74 	VC4_REG32(PV_VSYNCD_EVEN),
75 	VC4_REG32(PV_HORZA),
76 	VC4_REG32(PV_HORZB),
77 	VC4_REG32(PV_VERTA),
78 	VC4_REG32(PV_VERTB),
79 	VC4_REG32(PV_VERTA_EVEN),
80 	VC4_REG32(PV_VERTB_EVEN),
81 	VC4_REG32(PV_INTEN),
82 	VC4_REG32(PV_INTSTAT),
83 	VC4_REG32(PV_STAT),
84 	VC4_REG32(PV_HACT_ACT),
85 };
86 
87 static bool vc4_crtc_get_scanout_position(struct drm_crtc *crtc,
88 					  bool in_vblank_irq,
89 					  int *vpos, int *hpos,
90 					  ktime_t *stime, ktime_t *etime,
91 					  const struct drm_display_mode *mode)
92 {
93 	struct drm_device *dev = crtc->dev;
94 	struct vc4_dev *vc4 = to_vc4_dev(dev);
95 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
96 	u32 val;
97 	int fifo_lines;
98 	int vblank_lines;
99 	bool ret = false;
100 
101 	/* preempt_disable_rt() should go right here in PREEMPT_RT patchset. */
102 
103 	/* Get optional system timestamp before query. */
104 	if (stime)
105 		*stime = ktime_get();
106 
107 	/*
108 	 * Read vertical scanline which is currently composed for our
109 	 * pixelvalve by the HVS, and also the scaler status.
110 	 */
111 	val = HVS_READ(SCALER_DISPSTATX(vc4_crtc->channel));
112 
113 	/* Get optional system timestamp after query. */
114 	if (etime)
115 		*etime = ktime_get();
116 
117 	/* preempt_enable_rt() should go right here in PREEMPT_RT patchset. */
118 
119 	/* Vertical position of hvs composed scanline. */
120 	*vpos = VC4_GET_FIELD(val, SCALER_DISPSTATX_LINE);
121 	*hpos = 0;
122 
123 	if (mode->flags & DRM_MODE_FLAG_INTERLACE) {
124 		*vpos /= 2;
125 
126 		/* Use hpos to correct for field offset in interlaced mode. */
127 		if (VC4_GET_FIELD(val, SCALER_DISPSTATX_FRAME_COUNT) % 2)
128 			*hpos += mode->crtc_htotal / 2;
129 	}
130 
131 	/* This is the offset we need for translating hvs -> pv scanout pos. */
132 	fifo_lines = vc4_crtc->cob_size / mode->crtc_hdisplay;
133 
134 	if (fifo_lines > 0)
135 		ret = true;
136 
137 	/* HVS more than fifo_lines into frame for compositing? */
138 	if (*vpos > fifo_lines) {
139 		/*
140 		 * We are in active scanout and can get some meaningful results
141 		 * from HVS. The actual PV scanout can not trail behind more
142 		 * than fifo_lines as that is the fifo's capacity. Assume that
143 		 * in active scanout the HVS and PV work in lockstep wrt. HVS
144 		 * refilling the fifo and PV consuming from the fifo, ie.
145 		 * whenever the PV consumes and frees up a scanline in the
146 		 * fifo, the HVS will immediately refill it, therefore
147 		 * incrementing vpos. Therefore we choose HVS read position -
148 		 * fifo size in scanlines as a estimate of the real scanout
149 		 * position of the PV.
150 		 */
151 		*vpos -= fifo_lines + 1;
152 
153 		return ret;
154 	}
155 
156 	/*
157 	 * Less: This happens when we are in vblank and the HVS, after getting
158 	 * the VSTART restart signal from the PV, just started refilling its
159 	 * fifo with new lines from the top-most lines of the new framebuffers.
160 	 * The PV does not scan out in vblank, so does not remove lines from
161 	 * the fifo, so the fifo will be full quickly and the HVS has to pause.
162 	 * We can't get meaningful readings wrt. scanline position of the PV
163 	 * and need to make things up in a approximative but consistent way.
164 	 */
165 	vblank_lines = mode->vtotal - mode->vdisplay;
166 
167 	if (in_vblank_irq) {
168 		/*
169 		 * Assume the irq handler got called close to first
170 		 * line of vblank, so PV has about a full vblank
171 		 * scanlines to go, and as a base timestamp use the
172 		 * one taken at entry into vblank irq handler, so it
173 		 * is not affected by random delays due to lock
174 		 * contention on event_lock or vblank_time lock in
175 		 * the core.
176 		 */
177 		*vpos = -vblank_lines;
178 
179 		if (stime)
180 			*stime = vc4_crtc->t_vblank;
181 		if (etime)
182 			*etime = vc4_crtc->t_vblank;
183 
184 		/*
185 		 * If the HVS fifo is not yet full then we know for certain
186 		 * we are at the very beginning of vblank, as the hvs just
187 		 * started refilling, and the stime and etime timestamps
188 		 * truly correspond to start of vblank.
189 		 *
190 		 * Unfortunately there's no way to report this to upper levels
191 		 * and make it more useful.
192 		 */
193 	} else {
194 		/*
195 		 * No clue where we are inside vblank. Return a vpos of zero,
196 		 * which will cause calling code to just return the etime
197 		 * timestamp uncorrected. At least this is no worse than the
198 		 * standard fallback.
199 		 */
200 		*vpos = 0;
201 	}
202 
203 	return ret;
204 }
205 
206 static void vc4_crtc_destroy(struct drm_crtc *crtc)
207 {
208 	drm_crtc_cleanup(crtc);
209 }
210 
211 static void
212 vc4_crtc_lut_load(struct drm_crtc *crtc)
213 {
214 	struct drm_device *dev = crtc->dev;
215 	struct vc4_dev *vc4 = to_vc4_dev(dev);
216 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
217 	u32 i;
218 
219 	/* The LUT memory is laid out with each HVS channel in order,
220 	 * each of which takes 256 writes for R, 256 for G, then 256
221 	 * for B.
222 	 */
223 	HVS_WRITE(SCALER_GAMADDR,
224 		  SCALER_GAMADDR_AUTOINC |
225 		  (vc4_crtc->channel * 3 * crtc->gamma_size));
226 
227 	for (i = 0; i < crtc->gamma_size; i++)
228 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_r[i]);
229 	for (i = 0; i < crtc->gamma_size; i++)
230 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_g[i]);
231 	for (i = 0; i < crtc->gamma_size; i++)
232 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_b[i]);
233 }
234 
235 static void
236 vc4_crtc_update_gamma_lut(struct drm_crtc *crtc)
237 {
238 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
239 	struct drm_color_lut *lut = crtc->state->gamma_lut->data;
240 	u32 length = drm_color_lut_size(crtc->state->gamma_lut);
241 	u32 i;
242 
243 	for (i = 0; i < length; i++) {
244 		vc4_crtc->lut_r[i] = drm_color_lut_extract(lut[i].red, 8);
245 		vc4_crtc->lut_g[i] = drm_color_lut_extract(lut[i].green, 8);
246 		vc4_crtc->lut_b[i] = drm_color_lut_extract(lut[i].blue, 8);
247 	}
248 
249 	vc4_crtc_lut_load(crtc);
250 }
251 
252 static u32 vc4_get_fifo_full_level(u32 format)
253 {
254 	static const u32 fifo_len_bytes = 64;
255 	static const u32 hvs_latency_pix = 6;
256 
257 	switch (format) {
258 	case PV_CONTROL_FORMAT_DSIV_16:
259 	case PV_CONTROL_FORMAT_DSIC_16:
260 		return fifo_len_bytes - 2 * hvs_latency_pix;
261 	case PV_CONTROL_FORMAT_DSIV_18:
262 		return fifo_len_bytes - 14;
263 	case PV_CONTROL_FORMAT_24:
264 	case PV_CONTROL_FORMAT_DSIV_24:
265 	default:
266 		return fifo_len_bytes - 3 * hvs_latency_pix;
267 	}
268 }
269 
270 /*
271  * Returns the encoder attached to the CRTC.
272  *
273  * VC4 can only scan out to one encoder at a time, while the DRM core
274  * allows drivers to push pixels to more than one encoder from the
275  * same CRTC.
276  */
277 static struct drm_encoder *vc4_get_crtc_encoder(struct drm_crtc *crtc)
278 {
279 	struct drm_connector *connector;
280 	struct drm_connector_list_iter conn_iter;
281 
282 	drm_connector_list_iter_begin(crtc->dev, &conn_iter);
283 	drm_for_each_connector_iter(connector, &conn_iter) {
284 		if (connector->state->crtc == crtc) {
285 			drm_connector_list_iter_end(&conn_iter);
286 			return connector->encoder;
287 		}
288 	}
289 	drm_connector_list_iter_end(&conn_iter);
290 
291 	return NULL;
292 }
293 
294 static void vc4_crtc_config_pv(struct drm_crtc *crtc)
295 {
296 	struct drm_encoder *encoder = vc4_get_crtc_encoder(crtc);
297 	struct vc4_encoder *vc4_encoder = to_vc4_encoder(encoder);
298 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
299 	struct drm_crtc_state *state = crtc->state;
300 	struct drm_display_mode *mode = &state->adjusted_mode;
301 	bool interlace = mode->flags & DRM_MODE_FLAG_INTERLACE;
302 	u32 pixel_rep = (mode->flags & DRM_MODE_FLAG_DBLCLK) ? 2 : 1;
303 	bool is_dsi = (vc4_encoder->type == VC4_ENCODER_TYPE_DSI0 ||
304 		       vc4_encoder->type == VC4_ENCODER_TYPE_DSI1);
305 	u32 format = is_dsi ? PV_CONTROL_FORMAT_DSIV_24 : PV_CONTROL_FORMAT_24;
306 
307 	/* Reset the PV fifo. */
308 	CRTC_WRITE(PV_CONTROL, 0);
309 	CRTC_WRITE(PV_CONTROL, PV_CONTROL_FIFO_CLR | PV_CONTROL_EN);
310 	CRTC_WRITE(PV_CONTROL, 0);
311 
312 	CRTC_WRITE(PV_HORZA,
313 		   VC4_SET_FIELD((mode->htotal -
314 				  mode->hsync_end) * pixel_rep,
315 				 PV_HORZA_HBP) |
316 		   VC4_SET_FIELD((mode->hsync_end -
317 				  mode->hsync_start) * pixel_rep,
318 				 PV_HORZA_HSYNC));
319 	CRTC_WRITE(PV_HORZB,
320 		   VC4_SET_FIELD((mode->hsync_start -
321 				  mode->hdisplay) * pixel_rep,
322 				 PV_HORZB_HFP) |
323 		   VC4_SET_FIELD(mode->hdisplay * pixel_rep, PV_HORZB_HACTIVE));
324 
325 	CRTC_WRITE(PV_VERTA,
326 		   VC4_SET_FIELD(mode->crtc_vtotal - mode->crtc_vsync_end,
327 				 PV_VERTA_VBP) |
328 		   VC4_SET_FIELD(mode->crtc_vsync_end - mode->crtc_vsync_start,
329 				 PV_VERTA_VSYNC));
330 	CRTC_WRITE(PV_VERTB,
331 		   VC4_SET_FIELD(mode->crtc_vsync_start - mode->crtc_vdisplay,
332 				 PV_VERTB_VFP) |
333 		   VC4_SET_FIELD(mode->crtc_vdisplay, PV_VERTB_VACTIVE));
334 
335 	if (interlace) {
336 		CRTC_WRITE(PV_VERTA_EVEN,
337 			   VC4_SET_FIELD(mode->crtc_vtotal -
338 					 mode->crtc_vsync_end - 1,
339 					 PV_VERTA_VBP) |
340 			   VC4_SET_FIELD(mode->crtc_vsync_end -
341 					 mode->crtc_vsync_start,
342 					 PV_VERTA_VSYNC));
343 		CRTC_WRITE(PV_VERTB_EVEN,
344 			   VC4_SET_FIELD(mode->crtc_vsync_start -
345 					 mode->crtc_vdisplay,
346 					 PV_VERTB_VFP) |
347 			   VC4_SET_FIELD(mode->crtc_vdisplay, PV_VERTB_VACTIVE));
348 
349 		/* We set up first field even mode for HDMI.  VEC's
350 		 * NTSC mode would want first field odd instead, once
351 		 * we support it (to do so, set ODD_FIRST and put the
352 		 * delay in VSYNCD_EVEN instead).
353 		 */
354 		CRTC_WRITE(PV_V_CONTROL,
355 			   PV_VCONTROL_CONTINUOUS |
356 			   (is_dsi ? PV_VCONTROL_DSI : 0) |
357 			   PV_VCONTROL_INTERLACE |
358 			   VC4_SET_FIELD(mode->htotal * pixel_rep / 2,
359 					 PV_VCONTROL_ODD_DELAY));
360 		CRTC_WRITE(PV_VSYNCD_EVEN, 0);
361 	} else {
362 		CRTC_WRITE(PV_V_CONTROL,
363 			   PV_VCONTROL_CONTINUOUS |
364 			   (is_dsi ? PV_VCONTROL_DSI : 0));
365 	}
366 
367 	CRTC_WRITE(PV_HACT_ACT, mode->hdisplay * pixel_rep);
368 
369 	CRTC_WRITE(PV_CONTROL,
370 		   VC4_SET_FIELD(format, PV_CONTROL_FORMAT) |
371 		   VC4_SET_FIELD(vc4_get_fifo_full_level(format),
372 				 PV_CONTROL_FIFO_LEVEL) |
373 		   VC4_SET_FIELD(pixel_rep - 1, PV_CONTROL_PIXEL_REP) |
374 		   PV_CONTROL_CLR_AT_START |
375 		   PV_CONTROL_TRIGGER_UNDERFLOW |
376 		   PV_CONTROL_WAIT_HSTART |
377 		   VC4_SET_FIELD(vc4_encoder->clock_select,
378 				 PV_CONTROL_CLK_SELECT) |
379 		   PV_CONTROL_FIFO_CLR |
380 		   PV_CONTROL_EN);
381 }
382 
383 static void vc4_crtc_mode_set_nofb(struct drm_crtc *crtc)
384 {
385 	struct drm_device *dev = crtc->dev;
386 	struct vc4_dev *vc4 = to_vc4_dev(dev);
387 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
388 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
389 	struct drm_display_mode *mode = &crtc->state->adjusted_mode;
390 	bool interlace = mode->flags & DRM_MODE_FLAG_INTERLACE;
391 	bool debug_dump_regs = false;
392 
393 	if (debug_dump_regs) {
394 		struct drm_printer p = drm_info_printer(&vc4_crtc->pdev->dev);
395 		dev_info(&vc4_crtc->pdev->dev, "CRTC %d regs before:\n",
396 			 drm_crtc_index(crtc));
397 		drm_print_regset32(&p, &vc4_crtc->regset);
398 	}
399 
400 	if (vc4_crtc->channel == 2) {
401 		u32 dispctrl;
402 		u32 dsp3_mux;
403 
404 		/*
405 		 * SCALER_DISPCTRL_DSP3 = X, where X < 2 means 'connect DSP3 to
406 		 * FIFO X'.
407 		 * SCALER_DISPCTRL_DSP3 = 3 means 'disable DSP 3'.
408 		 *
409 		 * DSP3 is connected to FIFO2 unless the transposer is
410 		 * enabled. In this case, FIFO 2 is directly accessed by the
411 		 * TXP IP, and we need to disable the FIFO2 -> pixelvalve1
412 		 * route.
413 		 */
414 		if (vc4_state->feed_txp)
415 			dsp3_mux = VC4_SET_FIELD(3, SCALER_DISPCTRL_DSP3_MUX);
416 		else
417 			dsp3_mux = VC4_SET_FIELD(2, SCALER_DISPCTRL_DSP3_MUX);
418 
419 		dispctrl = HVS_READ(SCALER_DISPCTRL) &
420 			   ~SCALER_DISPCTRL_DSP3_MUX_MASK;
421 		HVS_WRITE(SCALER_DISPCTRL, dispctrl | dsp3_mux);
422 	}
423 
424 	if (!vc4_state->feed_txp)
425 		vc4_crtc_config_pv(crtc);
426 
427 	HVS_WRITE(SCALER_DISPBKGNDX(vc4_crtc->channel),
428 		  SCALER_DISPBKGND_AUTOHS |
429 		  SCALER_DISPBKGND_GAMMA |
430 		  (interlace ? SCALER_DISPBKGND_INTERLACE : 0));
431 
432 	/* Reload the LUT, since the SRAMs would have been disabled if
433 	 * all CRTCs had SCALER_DISPBKGND_GAMMA unset at once.
434 	 */
435 	vc4_crtc_lut_load(crtc);
436 
437 	if (debug_dump_regs) {
438 		struct drm_printer p = drm_info_printer(&vc4_crtc->pdev->dev);
439 		dev_info(&vc4_crtc->pdev->dev, "CRTC %d regs after:\n",
440 			 drm_crtc_index(crtc));
441 		drm_print_regset32(&p, &vc4_crtc->regset);
442 	}
443 }
444 
445 static void require_hvs_enabled(struct drm_device *dev)
446 {
447 	struct vc4_dev *vc4 = to_vc4_dev(dev);
448 
449 	WARN_ON_ONCE((HVS_READ(SCALER_DISPCTRL) & SCALER_DISPCTRL_ENABLE) !=
450 		     SCALER_DISPCTRL_ENABLE);
451 }
452 
453 static void vc4_crtc_atomic_disable(struct drm_crtc *crtc,
454 				    struct drm_crtc_state *old_state)
455 {
456 	struct drm_device *dev = crtc->dev;
457 	struct vc4_dev *vc4 = to_vc4_dev(dev);
458 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
459 	u32 chan = vc4_crtc->channel;
460 	int ret;
461 	require_hvs_enabled(dev);
462 
463 	/* Disable vblank irq handling before crtc is disabled. */
464 	drm_crtc_vblank_off(crtc);
465 
466 	CRTC_WRITE(PV_V_CONTROL,
467 		   CRTC_READ(PV_V_CONTROL) & ~PV_VCONTROL_VIDEN);
468 	ret = wait_for(!(CRTC_READ(PV_V_CONTROL) & PV_VCONTROL_VIDEN), 1);
469 	WARN_ONCE(ret, "Timeout waiting for !PV_VCONTROL_VIDEN\n");
470 
471 	if (HVS_READ(SCALER_DISPCTRLX(chan)) &
472 	    SCALER_DISPCTRLX_ENABLE) {
473 		HVS_WRITE(SCALER_DISPCTRLX(chan),
474 			  SCALER_DISPCTRLX_RESET);
475 
476 		/* While the docs say that reset is self-clearing, it
477 		 * seems it doesn't actually.
478 		 */
479 		HVS_WRITE(SCALER_DISPCTRLX(chan), 0);
480 	}
481 
482 	/* Once we leave, the scaler should be disabled and its fifo empty. */
483 
484 	WARN_ON_ONCE(HVS_READ(SCALER_DISPCTRLX(chan)) & SCALER_DISPCTRLX_RESET);
485 
486 	WARN_ON_ONCE(VC4_GET_FIELD(HVS_READ(SCALER_DISPSTATX(chan)),
487 				   SCALER_DISPSTATX_MODE) !=
488 		     SCALER_DISPSTATX_MODE_DISABLED);
489 
490 	WARN_ON_ONCE((HVS_READ(SCALER_DISPSTATX(chan)) &
491 		      (SCALER_DISPSTATX_FULL | SCALER_DISPSTATX_EMPTY)) !=
492 		     SCALER_DISPSTATX_EMPTY);
493 
494 	/*
495 	 * Make sure we issue a vblank event after disabling the CRTC if
496 	 * someone was waiting it.
497 	 */
498 	if (crtc->state->event) {
499 		unsigned long flags;
500 
501 		spin_lock_irqsave(&dev->event_lock, flags);
502 		drm_crtc_send_vblank_event(crtc, crtc->state->event);
503 		crtc->state->event = NULL;
504 		spin_unlock_irqrestore(&dev->event_lock, flags);
505 	}
506 }
507 
508 void vc4_crtc_txp_armed(struct drm_crtc_state *state)
509 {
510 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(state);
511 
512 	vc4_state->txp_armed = true;
513 }
514 
515 static void vc4_crtc_update_dlist(struct drm_crtc *crtc)
516 {
517 	struct drm_device *dev = crtc->dev;
518 	struct vc4_dev *vc4 = to_vc4_dev(dev);
519 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
520 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
521 
522 	if (crtc->state->event) {
523 		unsigned long flags;
524 
525 		crtc->state->event->pipe = drm_crtc_index(crtc);
526 
527 		WARN_ON(drm_crtc_vblank_get(crtc) != 0);
528 
529 		spin_lock_irqsave(&dev->event_lock, flags);
530 
531 		if (!vc4_state->feed_txp || vc4_state->txp_armed) {
532 			vc4_crtc->event = crtc->state->event;
533 			crtc->state->event = NULL;
534 		}
535 
536 		HVS_WRITE(SCALER_DISPLISTX(vc4_crtc->channel),
537 			  vc4_state->mm.start);
538 
539 		spin_unlock_irqrestore(&dev->event_lock, flags);
540 	} else {
541 		HVS_WRITE(SCALER_DISPLISTX(vc4_crtc->channel),
542 			  vc4_state->mm.start);
543 	}
544 }
545 
546 static void vc4_crtc_atomic_enable(struct drm_crtc *crtc,
547 				   struct drm_crtc_state *old_state)
548 {
549 	struct drm_device *dev = crtc->dev;
550 	struct vc4_dev *vc4 = to_vc4_dev(dev);
551 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
552 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
553 	struct drm_display_mode *mode = &crtc->state->adjusted_mode;
554 
555 	require_hvs_enabled(dev);
556 
557 	/* Enable vblank irq handling before crtc is started otherwise
558 	 * drm_crtc_get_vblank() fails in vc4_crtc_update_dlist().
559 	 */
560 	drm_crtc_vblank_on(crtc);
561 	vc4_crtc_update_dlist(crtc);
562 
563 	/* Turn on the scaler, which will wait for vstart to start
564 	 * compositing.
565 	 * When feeding the transposer, we should operate in oneshot
566 	 * mode.
567 	 */
568 	HVS_WRITE(SCALER_DISPCTRLX(vc4_crtc->channel),
569 		  VC4_SET_FIELD(mode->hdisplay, SCALER_DISPCTRLX_WIDTH) |
570 		  VC4_SET_FIELD(mode->vdisplay, SCALER_DISPCTRLX_HEIGHT) |
571 		  SCALER_DISPCTRLX_ENABLE |
572 		  (vc4_state->feed_txp ? SCALER_DISPCTRLX_ONESHOT : 0));
573 
574 	/* When feeding the transposer block the pixelvalve is unneeded and
575 	 * should not be enabled.
576 	 */
577 	if (!vc4_state->feed_txp)
578 		CRTC_WRITE(PV_V_CONTROL,
579 			   CRTC_READ(PV_V_CONTROL) | PV_VCONTROL_VIDEN);
580 }
581 
582 static enum drm_mode_status vc4_crtc_mode_valid(struct drm_crtc *crtc,
583 						const struct drm_display_mode *mode)
584 {
585 	/* Do not allow doublescan modes from user space */
586 	if (mode->flags & DRM_MODE_FLAG_DBLSCAN) {
587 		DRM_DEBUG_KMS("[CRTC:%d] Doublescan mode rejected.\n",
588 			      crtc->base.id);
589 		return MODE_NO_DBLESCAN;
590 	}
591 
592 	return MODE_OK;
593 }
594 
595 void vc4_crtc_get_margins(struct drm_crtc_state *state,
596 			  unsigned int *left, unsigned int *right,
597 			  unsigned int *top, unsigned int *bottom)
598 {
599 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(state);
600 	struct drm_connector_state *conn_state;
601 	struct drm_connector *conn;
602 	int i;
603 
604 	*left = vc4_state->margins.left;
605 	*right = vc4_state->margins.right;
606 	*top = vc4_state->margins.top;
607 	*bottom = vc4_state->margins.bottom;
608 
609 	/* We have to interate over all new connector states because
610 	 * vc4_crtc_get_margins() might be called before
611 	 * vc4_crtc_atomic_check() which means margins info in vc4_crtc_state
612 	 * might be outdated.
613 	 */
614 	for_each_new_connector_in_state(state->state, conn, conn_state, i) {
615 		if (conn_state->crtc != state->crtc)
616 			continue;
617 
618 		*left = conn_state->tv.margins.left;
619 		*right = conn_state->tv.margins.right;
620 		*top = conn_state->tv.margins.top;
621 		*bottom = conn_state->tv.margins.bottom;
622 		break;
623 	}
624 }
625 
626 static int vc4_crtc_atomic_check(struct drm_crtc *crtc,
627 				 struct drm_crtc_state *state)
628 {
629 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(state);
630 	struct drm_device *dev = crtc->dev;
631 	struct vc4_dev *vc4 = to_vc4_dev(dev);
632 	struct drm_plane *plane;
633 	unsigned long flags;
634 	const struct drm_plane_state *plane_state;
635 	struct drm_connector *conn;
636 	struct drm_connector_state *conn_state;
637 	u32 dlist_count = 0;
638 	int ret, i;
639 
640 	/* The pixelvalve can only feed one encoder (and encoders are
641 	 * 1:1 with connectors.)
642 	 */
643 	if (hweight32(state->connector_mask) > 1)
644 		return -EINVAL;
645 
646 	drm_atomic_crtc_state_for_each_plane_state(plane, plane_state, state)
647 		dlist_count += vc4_plane_dlist_size(plane_state);
648 
649 	dlist_count++; /* Account for SCALER_CTL0_END. */
650 
651 	spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
652 	ret = drm_mm_insert_node(&vc4->hvs->dlist_mm, &vc4_state->mm,
653 				 dlist_count);
654 	spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
655 	if (ret)
656 		return ret;
657 
658 	for_each_new_connector_in_state(state->state, conn, conn_state, i) {
659 		if (conn_state->crtc != crtc)
660 			continue;
661 
662 		/* The writeback connector is implemented using the transposer
663 		 * block which is directly taking its data from the HVS FIFO.
664 		 */
665 		if (conn->connector_type == DRM_MODE_CONNECTOR_WRITEBACK) {
666 			state->no_vblank = true;
667 			vc4_state->feed_txp = true;
668 		} else {
669 			state->no_vblank = false;
670 			vc4_state->feed_txp = false;
671 		}
672 
673 		vc4_state->margins.left = conn_state->tv.margins.left;
674 		vc4_state->margins.right = conn_state->tv.margins.right;
675 		vc4_state->margins.top = conn_state->tv.margins.top;
676 		vc4_state->margins.bottom = conn_state->tv.margins.bottom;
677 		break;
678 	}
679 
680 	return 0;
681 }
682 
683 static void vc4_crtc_atomic_flush(struct drm_crtc *crtc,
684 				  struct drm_crtc_state *old_state)
685 {
686 	struct drm_device *dev = crtc->dev;
687 	struct vc4_dev *vc4 = to_vc4_dev(dev);
688 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
689 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
690 	struct drm_plane *plane;
691 	struct vc4_plane_state *vc4_plane_state;
692 	bool debug_dump_regs = false;
693 	bool enable_bg_fill = false;
694 	u32 __iomem *dlist_start = vc4->hvs->dlist + vc4_state->mm.start;
695 	u32 __iomem *dlist_next = dlist_start;
696 
697 	if (debug_dump_regs) {
698 		DRM_INFO("CRTC %d HVS before:\n", drm_crtc_index(crtc));
699 		vc4_hvs_dump_state(dev);
700 	}
701 
702 	/* Copy all the active planes' dlist contents to the hardware dlist. */
703 	drm_atomic_crtc_for_each_plane(plane, crtc) {
704 		/* Is this the first active plane? */
705 		if (dlist_next == dlist_start) {
706 			/* We need to enable background fill when a plane
707 			 * could be alpha blending from the background, i.e.
708 			 * where no other plane is underneath. It suffices to
709 			 * consider the first active plane here since we set
710 			 * needs_bg_fill such that either the first plane
711 			 * already needs it or all planes on top blend from
712 			 * the first or a lower plane.
713 			 */
714 			vc4_plane_state = to_vc4_plane_state(plane->state);
715 			enable_bg_fill = vc4_plane_state->needs_bg_fill;
716 		}
717 
718 		dlist_next += vc4_plane_write_dlist(plane, dlist_next);
719 	}
720 
721 	writel(SCALER_CTL0_END, dlist_next);
722 	dlist_next++;
723 
724 	WARN_ON_ONCE(dlist_next - dlist_start != vc4_state->mm.size);
725 
726 	if (enable_bg_fill)
727 		/* This sets a black background color fill, as is the case
728 		 * with other DRM drivers.
729 		 */
730 		HVS_WRITE(SCALER_DISPBKGNDX(vc4_crtc->channel),
731 			  HVS_READ(SCALER_DISPBKGNDX(vc4_crtc->channel)) |
732 			  SCALER_DISPBKGND_FILL);
733 
734 	/* Only update DISPLIST if the CRTC was already running and is not
735 	 * being disabled.
736 	 * vc4_crtc_enable() takes care of updating the dlist just after
737 	 * re-enabling VBLANK interrupts and before enabling the engine.
738 	 * If the CRTC is being disabled, there's no point in updating this
739 	 * information.
740 	 */
741 	if (crtc->state->active && old_state->active)
742 		vc4_crtc_update_dlist(crtc);
743 
744 	if (crtc->state->color_mgmt_changed) {
745 		u32 dispbkgndx = HVS_READ(SCALER_DISPBKGNDX(vc4_crtc->channel));
746 
747 		if (crtc->state->gamma_lut) {
748 			vc4_crtc_update_gamma_lut(crtc);
749 			dispbkgndx |= SCALER_DISPBKGND_GAMMA;
750 		} else {
751 			/* Unsetting DISPBKGND_GAMMA skips the gamma lut step
752 			 * in hardware, which is the same as a linear lut that
753 			 * DRM expects us to use in absence of a user lut.
754 			 */
755 			dispbkgndx &= ~SCALER_DISPBKGND_GAMMA;
756 		}
757 		HVS_WRITE(SCALER_DISPBKGNDX(vc4_crtc->channel), dispbkgndx);
758 	}
759 
760 	if (debug_dump_regs) {
761 		DRM_INFO("CRTC %d HVS after:\n", drm_crtc_index(crtc));
762 		vc4_hvs_dump_state(dev);
763 	}
764 }
765 
766 static int vc4_enable_vblank(struct drm_crtc *crtc)
767 {
768 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
769 
770 	CRTC_WRITE(PV_INTEN, PV_INT_VFP_START);
771 
772 	return 0;
773 }
774 
775 static void vc4_disable_vblank(struct drm_crtc *crtc)
776 {
777 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
778 
779 	CRTC_WRITE(PV_INTEN, 0);
780 }
781 
782 static void vc4_crtc_handle_page_flip(struct vc4_crtc *vc4_crtc)
783 {
784 	struct drm_crtc *crtc = &vc4_crtc->base;
785 	struct drm_device *dev = crtc->dev;
786 	struct vc4_dev *vc4 = to_vc4_dev(dev);
787 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
788 	u32 chan = vc4_crtc->channel;
789 	unsigned long flags;
790 
791 	spin_lock_irqsave(&dev->event_lock, flags);
792 	if (vc4_crtc->event &&
793 	    (vc4_state->mm.start == HVS_READ(SCALER_DISPLACTX(chan)) ||
794 	     vc4_state->feed_txp)) {
795 		drm_crtc_send_vblank_event(crtc, vc4_crtc->event);
796 		vc4_crtc->event = NULL;
797 		drm_crtc_vblank_put(crtc);
798 
799 		/* Wait for the page flip to unmask the underrun to ensure that
800 		 * the display list was updated by the hardware. Before that
801 		 * happens, the HVS will be using the previous display list with
802 		 * the CRTC and encoder already reconfigured, leading to
803 		 * underruns. This can be seen when reconfiguring the CRTC.
804 		 */
805 		vc4_hvs_unmask_underrun(dev, vc4_crtc->channel);
806 	}
807 	spin_unlock_irqrestore(&dev->event_lock, flags);
808 }
809 
810 void vc4_crtc_handle_vblank(struct vc4_crtc *crtc)
811 {
812 	crtc->t_vblank = ktime_get();
813 	drm_crtc_handle_vblank(&crtc->base);
814 	vc4_crtc_handle_page_flip(crtc);
815 }
816 
817 static irqreturn_t vc4_crtc_irq_handler(int irq, void *data)
818 {
819 	struct vc4_crtc *vc4_crtc = data;
820 	u32 stat = CRTC_READ(PV_INTSTAT);
821 	irqreturn_t ret = IRQ_NONE;
822 
823 	if (stat & PV_INT_VFP_START) {
824 		CRTC_WRITE(PV_INTSTAT, PV_INT_VFP_START);
825 		vc4_crtc_handle_vblank(vc4_crtc);
826 		ret = IRQ_HANDLED;
827 	}
828 
829 	return ret;
830 }
831 
832 struct vc4_async_flip_state {
833 	struct drm_crtc *crtc;
834 	struct drm_framebuffer *fb;
835 	struct drm_framebuffer *old_fb;
836 	struct drm_pending_vblank_event *event;
837 
838 	struct vc4_seqno_cb cb;
839 };
840 
841 /* Called when the V3D execution for the BO being flipped to is done, so that
842  * we can actually update the plane's address to point to it.
843  */
844 static void
845 vc4_async_page_flip_complete(struct vc4_seqno_cb *cb)
846 {
847 	struct vc4_async_flip_state *flip_state =
848 		container_of(cb, struct vc4_async_flip_state, cb);
849 	struct drm_crtc *crtc = flip_state->crtc;
850 	struct drm_device *dev = crtc->dev;
851 	struct vc4_dev *vc4 = to_vc4_dev(dev);
852 	struct drm_plane *plane = crtc->primary;
853 
854 	vc4_plane_async_set_fb(plane, flip_state->fb);
855 	if (flip_state->event) {
856 		unsigned long flags;
857 
858 		spin_lock_irqsave(&dev->event_lock, flags);
859 		drm_crtc_send_vblank_event(crtc, flip_state->event);
860 		spin_unlock_irqrestore(&dev->event_lock, flags);
861 	}
862 
863 	drm_crtc_vblank_put(crtc);
864 	drm_framebuffer_put(flip_state->fb);
865 
866 	/* Decrement the BO usecnt in order to keep the inc/dec calls balanced
867 	 * when the planes are updated through the async update path.
868 	 * FIXME: we should move to generic async-page-flip when it's
869 	 * available, so that we can get rid of this hand-made cleanup_fb()
870 	 * logic.
871 	 */
872 	if (flip_state->old_fb) {
873 		struct drm_gem_cma_object *cma_bo;
874 		struct vc4_bo *bo;
875 
876 		cma_bo = drm_fb_cma_get_gem_obj(flip_state->old_fb, 0);
877 		bo = to_vc4_bo(&cma_bo->base);
878 		vc4_bo_dec_usecnt(bo);
879 		drm_framebuffer_put(flip_state->old_fb);
880 	}
881 
882 	kfree(flip_state);
883 
884 	up(&vc4->async_modeset);
885 }
886 
887 /* Implements async (non-vblank-synced) page flips.
888  *
889  * The page flip ioctl needs to return immediately, so we grab the
890  * modeset semaphore on the pipe, and queue the address update for
891  * when V3D is done with the BO being flipped to.
892  */
893 static int vc4_async_page_flip(struct drm_crtc *crtc,
894 			       struct drm_framebuffer *fb,
895 			       struct drm_pending_vblank_event *event,
896 			       uint32_t flags)
897 {
898 	struct drm_device *dev = crtc->dev;
899 	struct vc4_dev *vc4 = to_vc4_dev(dev);
900 	struct drm_plane *plane = crtc->primary;
901 	int ret = 0;
902 	struct vc4_async_flip_state *flip_state;
903 	struct drm_gem_cma_object *cma_bo = drm_fb_cma_get_gem_obj(fb, 0);
904 	struct vc4_bo *bo = to_vc4_bo(&cma_bo->base);
905 
906 	/* Increment the BO usecnt here, so that we never end up with an
907 	 * unbalanced number of vc4_bo_{dec,inc}_usecnt() calls when the
908 	 * plane is later updated through the non-async path.
909 	 * FIXME: we should move to generic async-page-flip when it's
910 	 * available, so that we can get rid of this hand-made prepare_fb()
911 	 * logic.
912 	 */
913 	ret = vc4_bo_inc_usecnt(bo);
914 	if (ret)
915 		return ret;
916 
917 	flip_state = kzalloc(sizeof(*flip_state), GFP_KERNEL);
918 	if (!flip_state) {
919 		vc4_bo_dec_usecnt(bo);
920 		return -ENOMEM;
921 	}
922 
923 	drm_framebuffer_get(fb);
924 	flip_state->fb = fb;
925 	flip_state->crtc = crtc;
926 	flip_state->event = event;
927 
928 	/* Make sure all other async modesetes have landed. */
929 	ret = down_interruptible(&vc4->async_modeset);
930 	if (ret) {
931 		drm_framebuffer_put(fb);
932 		vc4_bo_dec_usecnt(bo);
933 		kfree(flip_state);
934 		return ret;
935 	}
936 
937 	/* Save the current FB before it's replaced by the new one in
938 	 * drm_atomic_set_fb_for_plane(). We'll need the old FB in
939 	 * vc4_async_page_flip_complete() to decrement the BO usecnt and keep
940 	 * it consistent.
941 	 * FIXME: we should move to generic async-page-flip when it's
942 	 * available, so that we can get rid of this hand-made cleanup_fb()
943 	 * logic.
944 	 */
945 	flip_state->old_fb = plane->state->fb;
946 	if (flip_state->old_fb)
947 		drm_framebuffer_get(flip_state->old_fb);
948 
949 	WARN_ON(drm_crtc_vblank_get(crtc) != 0);
950 
951 	/* Immediately update the plane's legacy fb pointer, so that later
952 	 * modeset prep sees the state that will be present when the semaphore
953 	 * is released.
954 	 */
955 	drm_atomic_set_fb_for_plane(plane->state, fb);
956 
957 	vc4_queue_seqno_cb(dev, &flip_state->cb, bo->seqno,
958 			   vc4_async_page_flip_complete);
959 
960 	/* Driver takes ownership of state on successful async commit. */
961 	return 0;
962 }
963 
964 static int vc4_page_flip(struct drm_crtc *crtc,
965 			 struct drm_framebuffer *fb,
966 			 struct drm_pending_vblank_event *event,
967 			 uint32_t flags,
968 			 struct drm_modeset_acquire_ctx *ctx)
969 {
970 	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
971 		return vc4_async_page_flip(crtc, fb, event, flags);
972 	else
973 		return drm_atomic_helper_page_flip(crtc, fb, event, flags, ctx);
974 }
975 
976 static struct drm_crtc_state *vc4_crtc_duplicate_state(struct drm_crtc *crtc)
977 {
978 	struct vc4_crtc_state *vc4_state, *old_vc4_state;
979 
980 	vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
981 	if (!vc4_state)
982 		return NULL;
983 
984 	old_vc4_state = to_vc4_crtc_state(crtc->state);
985 	vc4_state->feed_txp = old_vc4_state->feed_txp;
986 	vc4_state->margins = old_vc4_state->margins;
987 
988 	__drm_atomic_helper_crtc_duplicate_state(crtc, &vc4_state->base);
989 	return &vc4_state->base;
990 }
991 
992 static void vc4_crtc_destroy_state(struct drm_crtc *crtc,
993 				   struct drm_crtc_state *state)
994 {
995 	struct vc4_dev *vc4 = to_vc4_dev(crtc->dev);
996 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(state);
997 
998 	if (drm_mm_node_allocated(&vc4_state->mm)) {
999 		unsigned long flags;
1000 
1001 		spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
1002 		drm_mm_remove_node(&vc4_state->mm);
1003 		spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
1004 
1005 	}
1006 
1007 	drm_atomic_helper_crtc_destroy_state(crtc, state);
1008 }
1009 
1010 static void
1011 vc4_crtc_reset(struct drm_crtc *crtc)
1012 {
1013 	if (crtc->state)
1014 		vc4_crtc_destroy_state(crtc, crtc->state);
1015 
1016 	crtc->state = kzalloc(sizeof(struct vc4_crtc_state), GFP_KERNEL);
1017 	if (crtc->state)
1018 		crtc->state->crtc = crtc;
1019 }
1020 
1021 static const struct drm_crtc_funcs vc4_crtc_funcs = {
1022 	.set_config = drm_atomic_helper_set_config,
1023 	.destroy = vc4_crtc_destroy,
1024 	.page_flip = vc4_page_flip,
1025 	.set_property = NULL,
1026 	.cursor_set = NULL, /* handled by drm_mode_cursor_universal */
1027 	.cursor_move = NULL, /* handled by drm_mode_cursor_universal */
1028 	.reset = vc4_crtc_reset,
1029 	.atomic_duplicate_state = vc4_crtc_duplicate_state,
1030 	.atomic_destroy_state = vc4_crtc_destroy_state,
1031 	.gamma_set = drm_atomic_helper_legacy_gamma_set,
1032 	.enable_vblank = vc4_enable_vblank,
1033 	.disable_vblank = vc4_disable_vblank,
1034 	.get_vblank_timestamp = drm_crtc_vblank_helper_get_vblank_timestamp,
1035 };
1036 
1037 static const struct drm_crtc_helper_funcs vc4_crtc_helper_funcs = {
1038 	.mode_set_nofb = vc4_crtc_mode_set_nofb,
1039 	.mode_valid = vc4_crtc_mode_valid,
1040 	.atomic_check = vc4_crtc_atomic_check,
1041 	.atomic_flush = vc4_crtc_atomic_flush,
1042 	.atomic_enable = vc4_crtc_atomic_enable,
1043 	.atomic_disable = vc4_crtc_atomic_disable,
1044 	.get_scanout_position = vc4_crtc_get_scanout_position,
1045 };
1046 
1047 static const struct vc4_crtc_data pv0_data = {
1048 	.hvs_channel = 0,
1049 	.debugfs_name = "crtc0_regs",
1050 	.encoder_types = {
1051 		[PV_CONTROL_CLK_SELECT_DSI] = VC4_ENCODER_TYPE_DSI0,
1052 		[PV_CONTROL_CLK_SELECT_DPI_SMI_HDMI] = VC4_ENCODER_TYPE_DPI,
1053 	},
1054 };
1055 
1056 static const struct vc4_crtc_data pv1_data = {
1057 	.hvs_channel = 2,
1058 	.debugfs_name = "crtc1_regs",
1059 	.encoder_types = {
1060 		[PV_CONTROL_CLK_SELECT_DSI] = VC4_ENCODER_TYPE_DSI1,
1061 		[PV_CONTROL_CLK_SELECT_DPI_SMI_HDMI] = VC4_ENCODER_TYPE_SMI,
1062 	},
1063 };
1064 
1065 static const struct vc4_crtc_data pv2_data = {
1066 	.hvs_channel = 1,
1067 	.debugfs_name = "crtc2_regs",
1068 	.encoder_types = {
1069 		[PV_CONTROL_CLK_SELECT_DPI_SMI_HDMI] = VC4_ENCODER_TYPE_HDMI,
1070 		[PV_CONTROL_CLK_SELECT_VEC] = VC4_ENCODER_TYPE_VEC,
1071 	},
1072 };
1073 
1074 static const struct of_device_id vc4_crtc_dt_match[] = {
1075 	{ .compatible = "brcm,bcm2835-pixelvalve0", .data = &pv0_data },
1076 	{ .compatible = "brcm,bcm2835-pixelvalve1", .data = &pv1_data },
1077 	{ .compatible = "brcm,bcm2835-pixelvalve2", .data = &pv2_data },
1078 	{}
1079 };
1080 
1081 static void vc4_set_crtc_possible_masks(struct drm_device *drm,
1082 					struct drm_crtc *crtc)
1083 {
1084 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
1085 	const struct vc4_crtc_data *crtc_data = vc4_crtc->data;
1086 	const enum vc4_encoder_type *encoder_types = crtc_data->encoder_types;
1087 	struct drm_encoder *encoder;
1088 
1089 	drm_for_each_encoder(encoder, drm) {
1090 		struct vc4_encoder *vc4_encoder;
1091 		int i;
1092 
1093 		/* HVS FIFO2 can feed the TXP IP. */
1094 		if (crtc_data->hvs_channel == 2 &&
1095 		    encoder->encoder_type == DRM_MODE_ENCODER_VIRTUAL) {
1096 			encoder->possible_crtcs |= drm_crtc_mask(crtc);
1097 			continue;
1098 		}
1099 
1100 		vc4_encoder = to_vc4_encoder(encoder);
1101 		for (i = 0; i < ARRAY_SIZE(crtc_data->encoder_types); i++) {
1102 			if (vc4_encoder->type == encoder_types[i]) {
1103 				vc4_encoder->clock_select = i;
1104 				encoder->possible_crtcs |= drm_crtc_mask(crtc);
1105 				break;
1106 			}
1107 		}
1108 	}
1109 }
1110 
1111 static void
1112 vc4_crtc_get_cob_allocation(struct vc4_crtc *vc4_crtc)
1113 {
1114 	struct drm_device *drm = vc4_crtc->base.dev;
1115 	struct vc4_dev *vc4 = to_vc4_dev(drm);
1116 	u32 dispbase = HVS_READ(SCALER_DISPBASEX(vc4_crtc->channel));
1117 	/* Top/base are supposed to be 4-pixel aligned, but the
1118 	 * Raspberry Pi firmware fills the low bits (which are
1119 	 * presumably ignored).
1120 	 */
1121 	u32 top = VC4_GET_FIELD(dispbase, SCALER_DISPBASEX_TOP) & ~3;
1122 	u32 base = VC4_GET_FIELD(dispbase, SCALER_DISPBASEX_BASE) & ~3;
1123 
1124 	vc4_crtc->cob_size = top - base + 4;
1125 }
1126 
1127 static int vc4_crtc_bind(struct device *dev, struct device *master, void *data)
1128 {
1129 	struct platform_device *pdev = to_platform_device(dev);
1130 	struct drm_device *drm = dev_get_drvdata(master);
1131 	struct vc4_crtc *vc4_crtc;
1132 	struct drm_crtc *crtc;
1133 	struct drm_plane *primary_plane, *cursor_plane, *destroy_plane, *temp;
1134 	const struct of_device_id *match;
1135 	int ret, i;
1136 
1137 	vc4_crtc = devm_kzalloc(dev, sizeof(*vc4_crtc), GFP_KERNEL);
1138 	if (!vc4_crtc)
1139 		return -ENOMEM;
1140 	crtc = &vc4_crtc->base;
1141 
1142 	match = of_match_device(vc4_crtc_dt_match, dev);
1143 	if (!match)
1144 		return -ENODEV;
1145 	vc4_crtc->data = match->data;
1146 	vc4_crtc->pdev = pdev;
1147 
1148 	vc4_crtc->regs = vc4_ioremap_regs(pdev, 0);
1149 	if (IS_ERR(vc4_crtc->regs))
1150 		return PTR_ERR(vc4_crtc->regs);
1151 
1152 	vc4_crtc->regset.base = vc4_crtc->regs;
1153 	vc4_crtc->regset.regs = crtc_regs;
1154 	vc4_crtc->regset.nregs = ARRAY_SIZE(crtc_regs);
1155 
1156 	/* For now, we create just the primary and the legacy cursor
1157 	 * planes.  We should be able to stack more planes on easily,
1158 	 * but to do that we would need to compute the bandwidth
1159 	 * requirement of the plane configuration, and reject ones
1160 	 * that will take too much.
1161 	 */
1162 	primary_plane = vc4_plane_init(drm, DRM_PLANE_TYPE_PRIMARY);
1163 	if (IS_ERR(primary_plane)) {
1164 		dev_err(dev, "failed to construct primary plane\n");
1165 		ret = PTR_ERR(primary_plane);
1166 		goto err;
1167 	}
1168 
1169 	drm_crtc_init_with_planes(drm, crtc, primary_plane, NULL,
1170 				  &vc4_crtc_funcs, NULL);
1171 	drm_crtc_helper_add(crtc, &vc4_crtc_helper_funcs);
1172 	vc4_crtc->channel = vc4_crtc->data->hvs_channel;
1173 	drm_mode_crtc_set_gamma_size(crtc, ARRAY_SIZE(vc4_crtc->lut_r));
1174 	drm_crtc_enable_color_mgmt(crtc, 0, false, crtc->gamma_size);
1175 
1176 	/* We support CTM, but only for one CRTC at a time. It's therefore
1177 	 * implemented as private driver state in vc4_kms, not here.
1178 	 */
1179 	drm_crtc_enable_color_mgmt(crtc, 0, true, crtc->gamma_size);
1180 
1181 	/* Set up some arbitrary number of planes.  We're not limited
1182 	 * by a set number of physical registers, just the space in
1183 	 * the HVS (16k) and how small an plane can be (28 bytes).
1184 	 * However, each plane we set up takes up some memory, and
1185 	 * increases the cost of looping over planes, which atomic
1186 	 * modesetting does quite a bit.  As a result, we pick a
1187 	 * modest number of planes to expose, that should hopefully
1188 	 * still cover any sane usecase.
1189 	 */
1190 	for (i = 0; i < 8; i++) {
1191 		struct drm_plane *plane =
1192 			vc4_plane_init(drm, DRM_PLANE_TYPE_OVERLAY);
1193 
1194 		if (IS_ERR(plane))
1195 			continue;
1196 
1197 		plane->possible_crtcs = drm_crtc_mask(crtc);
1198 	}
1199 
1200 	/* Set up the legacy cursor after overlay initialization,
1201 	 * since we overlay planes on the CRTC in the order they were
1202 	 * initialized.
1203 	 */
1204 	cursor_plane = vc4_plane_init(drm, DRM_PLANE_TYPE_CURSOR);
1205 	if (!IS_ERR(cursor_plane)) {
1206 		cursor_plane->possible_crtcs = drm_crtc_mask(crtc);
1207 		crtc->cursor = cursor_plane;
1208 	}
1209 
1210 	vc4_crtc_get_cob_allocation(vc4_crtc);
1211 
1212 	CRTC_WRITE(PV_INTEN, 0);
1213 	CRTC_WRITE(PV_INTSTAT, PV_INT_VFP_START);
1214 	ret = devm_request_irq(dev, platform_get_irq(pdev, 0),
1215 			       vc4_crtc_irq_handler, 0, "vc4 crtc", vc4_crtc);
1216 	if (ret)
1217 		goto err_destroy_planes;
1218 
1219 	vc4_set_crtc_possible_masks(drm, crtc);
1220 
1221 	for (i = 0; i < crtc->gamma_size; i++) {
1222 		vc4_crtc->lut_r[i] = i;
1223 		vc4_crtc->lut_g[i] = i;
1224 		vc4_crtc->lut_b[i] = i;
1225 	}
1226 
1227 	platform_set_drvdata(pdev, vc4_crtc);
1228 
1229 	vc4_debugfs_add_regset32(drm, vc4_crtc->data->debugfs_name,
1230 				 &vc4_crtc->regset);
1231 
1232 	return 0;
1233 
1234 err_destroy_planes:
1235 	list_for_each_entry_safe(destroy_plane, temp,
1236 				 &drm->mode_config.plane_list, head) {
1237 		if (destroy_plane->possible_crtcs == drm_crtc_mask(crtc))
1238 		    destroy_plane->funcs->destroy(destroy_plane);
1239 	}
1240 err:
1241 	return ret;
1242 }
1243 
1244 static void vc4_crtc_unbind(struct device *dev, struct device *master,
1245 			    void *data)
1246 {
1247 	struct platform_device *pdev = to_platform_device(dev);
1248 	struct vc4_crtc *vc4_crtc = dev_get_drvdata(dev);
1249 
1250 	vc4_crtc_destroy(&vc4_crtc->base);
1251 
1252 	CRTC_WRITE(PV_INTEN, 0);
1253 
1254 	platform_set_drvdata(pdev, NULL);
1255 }
1256 
1257 static const struct component_ops vc4_crtc_ops = {
1258 	.bind   = vc4_crtc_bind,
1259 	.unbind = vc4_crtc_unbind,
1260 };
1261 
1262 static int vc4_crtc_dev_probe(struct platform_device *pdev)
1263 {
1264 	return component_add(&pdev->dev, &vc4_crtc_ops);
1265 }
1266 
1267 static int vc4_crtc_dev_remove(struct platform_device *pdev)
1268 {
1269 	component_del(&pdev->dev, &vc4_crtc_ops);
1270 	return 0;
1271 }
1272 
1273 struct platform_driver vc4_crtc_driver = {
1274 	.probe = vc4_crtc_dev_probe,
1275 	.remove = vc4_crtc_dev_remove,
1276 	.driver = {
1277 		.name = "vc4_crtc",
1278 		.of_match_table = vc4_crtc_dt_match,
1279 	},
1280 };
1281