xref: /openbmc/linux/drivers/gpu/drm/vc4/vc4_crtc.c (revision b830f94f)
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 <drm/drm_atomic.h>
33 #include <drm/drm_atomic_helper.h>
34 #include <drm/drm_atomic_uapi.h>
35 #include <drm/drm_print.h>
36 #include <drm/drm_probe_helper.h>
37 #include <linux/clk.h>
38 #include <drm/drm_fb_cma_helper.h>
39 #include <linux/component.h>
40 #include <linux/of_device.h>
41 #include "vc4_drv.h"
42 #include "vc4_regs.h"
43 
44 struct vc4_crtc_state {
45 	struct drm_crtc_state base;
46 	/* Dlist area for this CRTC configuration. */
47 	struct drm_mm_node mm;
48 	bool feed_txp;
49 	bool txp_armed;
50 
51 	struct {
52 		unsigned int left;
53 		unsigned int right;
54 		unsigned int top;
55 		unsigned int bottom;
56 	} margins;
57 };
58 
59 static inline struct vc4_crtc_state *
60 to_vc4_crtc_state(struct drm_crtc_state *crtc_state)
61 {
62 	return (struct vc4_crtc_state *)crtc_state;
63 }
64 
65 #define CRTC_WRITE(offset, val) writel(val, vc4_crtc->regs + (offset))
66 #define CRTC_READ(offset) readl(vc4_crtc->regs + (offset))
67 
68 static const struct debugfs_reg32 crtc_regs[] = {
69 	VC4_REG32(PV_CONTROL),
70 	VC4_REG32(PV_V_CONTROL),
71 	VC4_REG32(PV_VSYNCD_EVEN),
72 	VC4_REG32(PV_HORZA),
73 	VC4_REG32(PV_HORZB),
74 	VC4_REG32(PV_VERTA),
75 	VC4_REG32(PV_VERTB),
76 	VC4_REG32(PV_VERTA_EVEN),
77 	VC4_REG32(PV_VERTB_EVEN),
78 	VC4_REG32(PV_INTEN),
79 	VC4_REG32(PV_INTSTAT),
80 	VC4_REG32(PV_STAT),
81 	VC4_REG32(PV_HACT_ACT),
82 };
83 
84 bool vc4_crtc_get_scanoutpos(struct drm_device *dev, unsigned int crtc_id,
85 			     bool in_vblank_irq, int *vpos, int *hpos,
86 			     ktime_t *stime, ktime_t *etime,
87 			     const struct drm_display_mode *mode)
88 {
89 	struct vc4_dev *vc4 = to_vc4_dev(dev);
90 	struct drm_crtc *crtc = drm_crtc_from_index(dev, crtc_id);
91 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
92 	u32 val;
93 	int fifo_lines;
94 	int vblank_lines;
95 	bool ret = false;
96 
97 	/* preempt_disable_rt() should go right here in PREEMPT_RT patchset. */
98 
99 	/* Get optional system timestamp before query. */
100 	if (stime)
101 		*stime = ktime_get();
102 
103 	/*
104 	 * Read vertical scanline which is currently composed for our
105 	 * pixelvalve by the HVS, and also the scaler status.
106 	 */
107 	val = HVS_READ(SCALER_DISPSTATX(vc4_crtc->channel));
108 
109 	/* Get optional system timestamp after query. */
110 	if (etime)
111 		*etime = ktime_get();
112 
113 	/* preempt_enable_rt() should go right here in PREEMPT_RT patchset. */
114 
115 	/* Vertical position of hvs composed scanline. */
116 	*vpos = VC4_GET_FIELD(val, SCALER_DISPSTATX_LINE);
117 	*hpos = 0;
118 
119 	if (mode->flags & DRM_MODE_FLAG_INTERLACE) {
120 		*vpos /= 2;
121 
122 		/* Use hpos to correct for field offset in interlaced mode. */
123 		if (VC4_GET_FIELD(val, SCALER_DISPSTATX_FRAME_COUNT) % 2)
124 			*hpos += mode->crtc_htotal / 2;
125 	}
126 
127 	/* This is the offset we need for translating hvs -> pv scanout pos. */
128 	fifo_lines = vc4_crtc->cob_size / mode->crtc_hdisplay;
129 
130 	if (fifo_lines > 0)
131 		ret = true;
132 
133 	/* HVS more than fifo_lines into frame for compositing? */
134 	if (*vpos > fifo_lines) {
135 		/*
136 		 * We are in active scanout and can get some meaningful results
137 		 * from HVS. The actual PV scanout can not trail behind more
138 		 * than fifo_lines as that is the fifo's capacity. Assume that
139 		 * in active scanout the HVS and PV work in lockstep wrt. HVS
140 		 * refilling the fifo and PV consuming from the fifo, ie.
141 		 * whenever the PV consumes and frees up a scanline in the
142 		 * fifo, the HVS will immediately refill it, therefore
143 		 * incrementing vpos. Therefore we choose HVS read position -
144 		 * fifo size in scanlines as a estimate of the real scanout
145 		 * position of the PV.
146 		 */
147 		*vpos -= fifo_lines + 1;
148 
149 		return ret;
150 	}
151 
152 	/*
153 	 * Less: This happens when we are in vblank and the HVS, after getting
154 	 * the VSTART restart signal from the PV, just started refilling its
155 	 * fifo with new lines from the top-most lines of the new framebuffers.
156 	 * The PV does not scan out in vblank, so does not remove lines from
157 	 * the fifo, so the fifo will be full quickly and the HVS has to pause.
158 	 * We can't get meaningful readings wrt. scanline position of the PV
159 	 * and need to make things up in a approximative but consistent way.
160 	 */
161 	vblank_lines = mode->vtotal - mode->vdisplay;
162 
163 	if (in_vblank_irq) {
164 		/*
165 		 * Assume the irq handler got called close to first
166 		 * line of vblank, so PV has about a full vblank
167 		 * scanlines to go, and as a base timestamp use the
168 		 * one taken at entry into vblank irq handler, so it
169 		 * is not affected by random delays due to lock
170 		 * contention on event_lock or vblank_time lock in
171 		 * the core.
172 		 */
173 		*vpos = -vblank_lines;
174 
175 		if (stime)
176 			*stime = vc4_crtc->t_vblank;
177 		if (etime)
178 			*etime = vc4_crtc->t_vblank;
179 
180 		/*
181 		 * If the HVS fifo is not yet full then we know for certain
182 		 * we are at the very beginning of vblank, as the hvs just
183 		 * started refilling, and the stime and etime timestamps
184 		 * truly correspond to start of vblank.
185 		 *
186 		 * Unfortunately there's no way to report this to upper levels
187 		 * and make it more useful.
188 		 */
189 	} else {
190 		/*
191 		 * No clue where we are inside vblank. Return a vpos of zero,
192 		 * which will cause calling code to just return the etime
193 		 * timestamp uncorrected. At least this is no worse than the
194 		 * standard fallback.
195 		 */
196 		*vpos = 0;
197 	}
198 
199 	return ret;
200 }
201 
202 static void vc4_crtc_destroy(struct drm_crtc *crtc)
203 {
204 	drm_crtc_cleanup(crtc);
205 }
206 
207 static void
208 vc4_crtc_lut_load(struct drm_crtc *crtc)
209 {
210 	struct drm_device *dev = crtc->dev;
211 	struct vc4_dev *vc4 = to_vc4_dev(dev);
212 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
213 	u32 i;
214 
215 	/* The LUT memory is laid out with each HVS channel in order,
216 	 * each of which takes 256 writes for R, 256 for G, then 256
217 	 * for B.
218 	 */
219 	HVS_WRITE(SCALER_GAMADDR,
220 		  SCALER_GAMADDR_AUTOINC |
221 		  (vc4_crtc->channel * 3 * crtc->gamma_size));
222 
223 	for (i = 0; i < crtc->gamma_size; i++)
224 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_r[i]);
225 	for (i = 0; i < crtc->gamma_size; i++)
226 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_g[i]);
227 	for (i = 0; i < crtc->gamma_size; i++)
228 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_b[i]);
229 }
230 
231 static void
232 vc4_crtc_update_gamma_lut(struct drm_crtc *crtc)
233 {
234 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
235 	struct drm_color_lut *lut = crtc->state->gamma_lut->data;
236 	u32 length = drm_color_lut_size(crtc->state->gamma_lut);
237 	u32 i;
238 
239 	for (i = 0; i < length; i++) {
240 		vc4_crtc->lut_r[i] = drm_color_lut_extract(lut[i].red, 8);
241 		vc4_crtc->lut_g[i] = drm_color_lut_extract(lut[i].green, 8);
242 		vc4_crtc->lut_b[i] = drm_color_lut_extract(lut[i].blue, 8);
243 	}
244 
245 	vc4_crtc_lut_load(crtc);
246 }
247 
248 static u32 vc4_get_fifo_full_level(u32 format)
249 {
250 	static const u32 fifo_len_bytes = 64;
251 	static const u32 hvs_latency_pix = 6;
252 
253 	switch (format) {
254 	case PV_CONTROL_FORMAT_DSIV_16:
255 	case PV_CONTROL_FORMAT_DSIC_16:
256 		return fifo_len_bytes - 2 * hvs_latency_pix;
257 	case PV_CONTROL_FORMAT_DSIV_18:
258 		return fifo_len_bytes - 14;
259 	case PV_CONTROL_FORMAT_24:
260 	case PV_CONTROL_FORMAT_DSIV_24:
261 	default:
262 		return fifo_len_bytes - 3 * hvs_latency_pix;
263 	}
264 }
265 
266 /*
267  * Returns the encoder attached to the CRTC.
268  *
269  * VC4 can only scan out to one encoder at a time, while the DRM core
270  * allows drivers to push pixels to more than one encoder from the
271  * same CRTC.
272  */
273 static struct drm_encoder *vc4_get_crtc_encoder(struct drm_crtc *crtc)
274 {
275 	struct drm_connector *connector;
276 	struct drm_connector_list_iter conn_iter;
277 
278 	drm_connector_list_iter_begin(crtc->dev, &conn_iter);
279 	drm_for_each_connector_iter(connector, &conn_iter) {
280 		if (connector->state->crtc == crtc) {
281 			drm_connector_list_iter_end(&conn_iter);
282 			return connector->encoder;
283 		}
284 	}
285 	drm_connector_list_iter_end(&conn_iter);
286 
287 	return NULL;
288 }
289 
290 static void vc4_crtc_config_pv(struct drm_crtc *crtc)
291 {
292 	struct drm_encoder *encoder = vc4_get_crtc_encoder(crtc);
293 	struct vc4_encoder *vc4_encoder = to_vc4_encoder(encoder);
294 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
295 	struct drm_crtc_state *state = crtc->state;
296 	struct drm_display_mode *mode = &state->adjusted_mode;
297 	bool interlace = mode->flags & DRM_MODE_FLAG_INTERLACE;
298 	u32 pixel_rep = (mode->flags & DRM_MODE_FLAG_DBLCLK) ? 2 : 1;
299 	bool is_dsi = (vc4_encoder->type == VC4_ENCODER_TYPE_DSI0 ||
300 		       vc4_encoder->type == VC4_ENCODER_TYPE_DSI1);
301 	u32 format = is_dsi ? PV_CONTROL_FORMAT_DSIV_24 : PV_CONTROL_FORMAT_24;
302 
303 	/* Reset the PV fifo. */
304 	CRTC_WRITE(PV_CONTROL, 0);
305 	CRTC_WRITE(PV_CONTROL, PV_CONTROL_FIFO_CLR | PV_CONTROL_EN);
306 	CRTC_WRITE(PV_CONTROL, 0);
307 
308 	CRTC_WRITE(PV_HORZA,
309 		   VC4_SET_FIELD((mode->htotal -
310 				  mode->hsync_end) * pixel_rep,
311 				 PV_HORZA_HBP) |
312 		   VC4_SET_FIELD((mode->hsync_end -
313 				  mode->hsync_start) * pixel_rep,
314 				 PV_HORZA_HSYNC));
315 	CRTC_WRITE(PV_HORZB,
316 		   VC4_SET_FIELD((mode->hsync_start -
317 				  mode->hdisplay) * pixel_rep,
318 				 PV_HORZB_HFP) |
319 		   VC4_SET_FIELD(mode->hdisplay * pixel_rep, PV_HORZB_HACTIVE));
320 
321 	CRTC_WRITE(PV_VERTA,
322 		   VC4_SET_FIELD(mode->crtc_vtotal - mode->crtc_vsync_end,
323 				 PV_VERTA_VBP) |
324 		   VC4_SET_FIELD(mode->crtc_vsync_end - mode->crtc_vsync_start,
325 				 PV_VERTA_VSYNC));
326 	CRTC_WRITE(PV_VERTB,
327 		   VC4_SET_FIELD(mode->crtc_vsync_start - mode->crtc_vdisplay,
328 				 PV_VERTB_VFP) |
329 		   VC4_SET_FIELD(mode->crtc_vdisplay, PV_VERTB_VACTIVE));
330 
331 	if (interlace) {
332 		CRTC_WRITE(PV_VERTA_EVEN,
333 			   VC4_SET_FIELD(mode->crtc_vtotal -
334 					 mode->crtc_vsync_end - 1,
335 					 PV_VERTA_VBP) |
336 			   VC4_SET_FIELD(mode->crtc_vsync_end -
337 					 mode->crtc_vsync_start,
338 					 PV_VERTA_VSYNC));
339 		CRTC_WRITE(PV_VERTB_EVEN,
340 			   VC4_SET_FIELD(mode->crtc_vsync_start -
341 					 mode->crtc_vdisplay,
342 					 PV_VERTB_VFP) |
343 			   VC4_SET_FIELD(mode->crtc_vdisplay, PV_VERTB_VACTIVE));
344 
345 		/* We set up first field even mode for HDMI.  VEC's
346 		 * NTSC mode would want first field odd instead, once
347 		 * we support it (to do so, set ODD_FIRST and put the
348 		 * delay in VSYNCD_EVEN instead).
349 		 */
350 		CRTC_WRITE(PV_V_CONTROL,
351 			   PV_VCONTROL_CONTINUOUS |
352 			   (is_dsi ? PV_VCONTROL_DSI : 0) |
353 			   PV_VCONTROL_INTERLACE |
354 			   VC4_SET_FIELD(mode->htotal * pixel_rep / 2,
355 					 PV_VCONTROL_ODD_DELAY));
356 		CRTC_WRITE(PV_VSYNCD_EVEN, 0);
357 	} else {
358 		CRTC_WRITE(PV_V_CONTROL,
359 			   PV_VCONTROL_CONTINUOUS |
360 			   (is_dsi ? PV_VCONTROL_DSI : 0));
361 	}
362 
363 	CRTC_WRITE(PV_HACT_ACT, mode->hdisplay * pixel_rep);
364 
365 	CRTC_WRITE(PV_CONTROL,
366 		   VC4_SET_FIELD(format, PV_CONTROL_FORMAT) |
367 		   VC4_SET_FIELD(vc4_get_fifo_full_level(format),
368 				 PV_CONTROL_FIFO_LEVEL) |
369 		   VC4_SET_FIELD(pixel_rep - 1, PV_CONTROL_PIXEL_REP) |
370 		   PV_CONTROL_CLR_AT_START |
371 		   PV_CONTROL_TRIGGER_UNDERFLOW |
372 		   PV_CONTROL_WAIT_HSTART |
373 		   VC4_SET_FIELD(vc4_encoder->clock_select,
374 				 PV_CONTROL_CLK_SELECT) |
375 		   PV_CONTROL_FIFO_CLR |
376 		   PV_CONTROL_EN);
377 }
378 
379 static void vc4_crtc_mode_set_nofb(struct drm_crtc *crtc)
380 {
381 	struct drm_device *dev = crtc->dev;
382 	struct vc4_dev *vc4 = to_vc4_dev(dev);
383 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
384 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
385 	struct drm_display_mode *mode = &crtc->state->adjusted_mode;
386 	bool interlace = mode->flags & DRM_MODE_FLAG_INTERLACE;
387 	bool debug_dump_regs = false;
388 
389 	if (debug_dump_regs) {
390 		struct drm_printer p = drm_info_printer(&vc4_crtc->pdev->dev);
391 		dev_info(&vc4_crtc->pdev->dev, "CRTC %d regs before:\n",
392 			 drm_crtc_index(crtc));
393 		drm_print_regset32(&p, &vc4_crtc->regset);
394 	}
395 
396 	if (vc4_crtc->channel == 2) {
397 		u32 dispctrl;
398 		u32 dsp3_mux;
399 
400 		/*
401 		 * SCALER_DISPCTRL_DSP3 = X, where X < 2 means 'connect DSP3 to
402 		 * FIFO X'.
403 		 * SCALER_DISPCTRL_DSP3 = 3 means 'disable DSP 3'.
404 		 *
405 		 * DSP3 is connected to FIFO2 unless the transposer is
406 		 * enabled. In this case, FIFO 2 is directly accessed by the
407 		 * TXP IP, and we need to disable the FIFO2 -> pixelvalve1
408 		 * route.
409 		 */
410 		if (vc4_state->feed_txp)
411 			dsp3_mux = VC4_SET_FIELD(3, SCALER_DISPCTRL_DSP3_MUX);
412 		else
413 			dsp3_mux = VC4_SET_FIELD(2, SCALER_DISPCTRL_DSP3_MUX);
414 
415 		dispctrl = HVS_READ(SCALER_DISPCTRL) &
416 			   ~SCALER_DISPCTRL_DSP3_MUX_MASK;
417 		HVS_WRITE(SCALER_DISPCTRL, dispctrl | dsp3_mux);
418 	}
419 
420 	if (!vc4_state->feed_txp)
421 		vc4_crtc_config_pv(crtc);
422 
423 	HVS_WRITE(SCALER_DISPBKGNDX(vc4_crtc->channel),
424 		  SCALER_DISPBKGND_AUTOHS |
425 		  SCALER_DISPBKGND_GAMMA |
426 		  (interlace ? SCALER_DISPBKGND_INTERLACE : 0));
427 
428 	/* Reload the LUT, since the SRAMs would have been disabled if
429 	 * all CRTCs had SCALER_DISPBKGND_GAMMA unset at once.
430 	 */
431 	vc4_crtc_lut_load(crtc);
432 
433 	if (debug_dump_regs) {
434 		struct drm_printer p = drm_info_printer(&vc4_crtc->pdev->dev);
435 		dev_info(&vc4_crtc->pdev->dev, "CRTC %d regs after:\n",
436 			 drm_crtc_index(crtc));
437 		drm_print_regset32(&p, &vc4_crtc->regset);
438 	}
439 }
440 
441 static void require_hvs_enabled(struct drm_device *dev)
442 {
443 	struct vc4_dev *vc4 = to_vc4_dev(dev);
444 
445 	WARN_ON_ONCE((HVS_READ(SCALER_DISPCTRL) & SCALER_DISPCTRL_ENABLE) !=
446 		     SCALER_DISPCTRL_ENABLE);
447 }
448 
449 static void vc4_crtc_atomic_disable(struct drm_crtc *crtc,
450 				    struct drm_crtc_state *old_state)
451 {
452 	struct drm_device *dev = crtc->dev;
453 	struct vc4_dev *vc4 = to_vc4_dev(dev);
454 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
455 	u32 chan = vc4_crtc->channel;
456 	int ret;
457 	require_hvs_enabled(dev);
458 
459 	/* Disable vblank irq handling before crtc is disabled. */
460 	drm_crtc_vblank_off(crtc);
461 
462 	CRTC_WRITE(PV_V_CONTROL,
463 		   CRTC_READ(PV_V_CONTROL) & ~PV_VCONTROL_VIDEN);
464 	ret = wait_for(!(CRTC_READ(PV_V_CONTROL) & PV_VCONTROL_VIDEN), 1);
465 	WARN_ONCE(ret, "Timeout waiting for !PV_VCONTROL_VIDEN\n");
466 
467 	if (HVS_READ(SCALER_DISPCTRLX(chan)) &
468 	    SCALER_DISPCTRLX_ENABLE) {
469 		HVS_WRITE(SCALER_DISPCTRLX(chan),
470 			  SCALER_DISPCTRLX_RESET);
471 
472 		/* While the docs say that reset is self-clearing, it
473 		 * seems it doesn't actually.
474 		 */
475 		HVS_WRITE(SCALER_DISPCTRLX(chan), 0);
476 	}
477 
478 	/* Once we leave, the scaler should be disabled and its fifo empty. */
479 
480 	WARN_ON_ONCE(HVS_READ(SCALER_DISPCTRLX(chan)) & SCALER_DISPCTRLX_RESET);
481 
482 	WARN_ON_ONCE(VC4_GET_FIELD(HVS_READ(SCALER_DISPSTATX(chan)),
483 				   SCALER_DISPSTATX_MODE) !=
484 		     SCALER_DISPSTATX_MODE_DISABLED);
485 
486 	WARN_ON_ONCE((HVS_READ(SCALER_DISPSTATX(chan)) &
487 		      (SCALER_DISPSTATX_FULL | SCALER_DISPSTATX_EMPTY)) !=
488 		     SCALER_DISPSTATX_EMPTY);
489 
490 	/*
491 	 * Make sure we issue a vblank event after disabling the CRTC if
492 	 * someone was waiting it.
493 	 */
494 	if (crtc->state->event) {
495 		unsigned long flags;
496 
497 		spin_lock_irqsave(&dev->event_lock, flags);
498 		drm_crtc_send_vblank_event(crtc, crtc->state->event);
499 		crtc->state->event = NULL;
500 		spin_unlock_irqrestore(&dev->event_lock, flags);
501 	}
502 }
503 
504 void vc4_crtc_txp_armed(struct drm_crtc_state *state)
505 {
506 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(state);
507 
508 	vc4_state->txp_armed = true;
509 }
510 
511 static void vc4_crtc_update_dlist(struct drm_crtc *crtc)
512 {
513 	struct drm_device *dev = crtc->dev;
514 	struct vc4_dev *vc4 = to_vc4_dev(dev);
515 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
516 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
517 
518 	if (crtc->state->event) {
519 		unsigned long flags;
520 
521 		crtc->state->event->pipe = drm_crtc_index(crtc);
522 
523 		WARN_ON(drm_crtc_vblank_get(crtc) != 0);
524 
525 		spin_lock_irqsave(&dev->event_lock, flags);
526 
527 		if (!vc4_state->feed_txp || vc4_state->txp_armed) {
528 			vc4_crtc->event = crtc->state->event;
529 			crtc->state->event = NULL;
530 		}
531 
532 		HVS_WRITE(SCALER_DISPLISTX(vc4_crtc->channel),
533 			  vc4_state->mm.start);
534 
535 		spin_unlock_irqrestore(&dev->event_lock, flags);
536 	} else {
537 		HVS_WRITE(SCALER_DISPLISTX(vc4_crtc->channel),
538 			  vc4_state->mm.start);
539 	}
540 }
541 
542 static void vc4_crtc_atomic_enable(struct drm_crtc *crtc,
543 				   struct drm_crtc_state *old_state)
544 {
545 	struct drm_device *dev = crtc->dev;
546 	struct vc4_dev *vc4 = to_vc4_dev(dev);
547 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
548 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
549 	struct drm_display_mode *mode = &crtc->state->adjusted_mode;
550 
551 	require_hvs_enabled(dev);
552 
553 	/* Enable vblank irq handling before crtc is started otherwise
554 	 * drm_crtc_get_vblank() fails in vc4_crtc_update_dlist().
555 	 */
556 	drm_crtc_vblank_on(crtc);
557 	vc4_crtc_update_dlist(crtc);
558 
559 	/* Turn on the scaler, which will wait for vstart to start
560 	 * compositing.
561 	 * When feeding the transposer, we should operate in oneshot
562 	 * mode.
563 	 */
564 	HVS_WRITE(SCALER_DISPCTRLX(vc4_crtc->channel),
565 		  VC4_SET_FIELD(mode->hdisplay, SCALER_DISPCTRLX_WIDTH) |
566 		  VC4_SET_FIELD(mode->vdisplay, SCALER_DISPCTRLX_HEIGHT) |
567 		  SCALER_DISPCTRLX_ENABLE |
568 		  (vc4_state->feed_txp ? SCALER_DISPCTRLX_ONESHOT : 0));
569 
570 	/* When feeding the transposer block the pixelvalve is unneeded and
571 	 * should not be enabled.
572 	 */
573 	if (!vc4_state->feed_txp)
574 		CRTC_WRITE(PV_V_CONTROL,
575 			   CRTC_READ(PV_V_CONTROL) | PV_VCONTROL_VIDEN);
576 }
577 
578 static enum drm_mode_status vc4_crtc_mode_valid(struct drm_crtc *crtc,
579 						const struct drm_display_mode *mode)
580 {
581 	/* Do not allow doublescan modes from user space */
582 	if (mode->flags & DRM_MODE_FLAG_DBLSCAN) {
583 		DRM_DEBUG_KMS("[CRTC:%d] Doublescan mode rejected.\n",
584 			      crtc->base.id);
585 		return MODE_NO_DBLESCAN;
586 	}
587 
588 	return MODE_OK;
589 }
590 
591 void vc4_crtc_get_margins(struct drm_crtc_state *state,
592 			  unsigned int *left, unsigned int *right,
593 			  unsigned int *top, unsigned int *bottom)
594 {
595 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(state);
596 	struct drm_connector_state *conn_state;
597 	struct drm_connector *conn;
598 	int i;
599 
600 	*left = vc4_state->margins.left;
601 	*right = vc4_state->margins.right;
602 	*top = vc4_state->margins.top;
603 	*bottom = vc4_state->margins.bottom;
604 
605 	/* We have to interate over all new connector states because
606 	 * vc4_crtc_get_margins() might be called before
607 	 * vc4_crtc_atomic_check() which means margins info in vc4_crtc_state
608 	 * might be outdated.
609 	 */
610 	for_each_new_connector_in_state(state->state, conn, conn_state, i) {
611 		if (conn_state->crtc != state->crtc)
612 			continue;
613 
614 		*left = conn_state->tv.margins.left;
615 		*right = conn_state->tv.margins.right;
616 		*top = conn_state->tv.margins.top;
617 		*bottom = conn_state->tv.margins.bottom;
618 		break;
619 	}
620 }
621 
622 static int vc4_crtc_atomic_check(struct drm_crtc *crtc,
623 				 struct drm_crtc_state *state)
624 {
625 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(state);
626 	struct drm_device *dev = crtc->dev;
627 	struct vc4_dev *vc4 = to_vc4_dev(dev);
628 	struct drm_plane *plane;
629 	unsigned long flags;
630 	const struct drm_plane_state *plane_state;
631 	struct drm_connector *conn;
632 	struct drm_connector_state *conn_state;
633 	u32 dlist_count = 0;
634 	int ret, i;
635 
636 	/* The pixelvalve can only feed one encoder (and encoders are
637 	 * 1:1 with connectors.)
638 	 */
639 	if (hweight32(state->connector_mask) > 1)
640 		return -EINVAL;
641 
642 	drm_atomic_crtc_state_for_each_plane_state(plane, plane_state, state)
643 		dlist_count += vc4_plane_dlist_size(plane_state);
644 
645 	dlist_count++; /* Account for SCALER_CTL0_END. */
646 
647 	spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
648 	ret = drm_mm_insert_node(&vc4->hvs->dlist_mm, &vc4_state->mm,
649 				 dlist_count);
650 	spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
651 	if (ret)
652 		return ret;
653 
654 	for_each_new_connector_in_state(state->state, conn, conn_state, i) {
655 		if (conn_state->crtc != crtc)
656 			continue;
657 
658 		/* The writeback connector is implemented using the transposer
659 		 * block which is directly taking its data from the HVS FIFO.
660 		 */
661 		if (conn->connector_type == DRM_MODE_CONNECTOR_WRITEBACK) {
662 			state->no_vblank = true;
663 			vc4_state->feed_txp = true;
664 		} else {
665 			state->no_vblank = false;
666 			vc4_state->feed_txp = false;
667 		}
668 
669 		vc4_state->margins.left = conn_state->tv.margins.left;
670 		vc4_state->margins.right = conn_state->tv.margins.right;
671 		vc4_state->margins.top = conn_state->tv.margins.top;
672 		vc4_state->margins.bottom = conn_state->tv.margins.bottom;
673 		break;
674 	}
675 
676 	return 0;
677 }
678 
679 static void vc4_crtc_atomic_flush(struct drm_crtc *crtc,
680 				  struct drm_crtc_state *old_state)
681 {
682 	struct drm_device *dev = crtc->dev;
683 	struct vc4_dev *vc4 = to_vc4_dev(dev);
684 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
685 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
686 	struct drm_plane *plane;
687 	struct vc4_plane_state *vc4_plane_state;
688 	bool debug_dump_regs = false;
689 	bool enable_bg_fill = false;
690 	u32 __iomem *dlist_start = vc4->hvs->dlist + vc4_state->mm.start;
691 	u32 __iomem *dlist_next = dlist_start;
692 
693 	if (debug_dump_regs) {
694 		DRM_INFO("CRTC %d HVS before:\n", drm_crtc_index(crtc));
695 		vc4_hvs_dump_state(dev);
696 	}
697 
698 	/* Copy all the active planes' dlist contents to the hardware dlist. */
699 	drm_atomic_crtc_for_each_plane(plane, crtc) {
700 		/* Is this the first active plane? */
701 		if (dlist_next == dlist_start) {
702 			/* We need to enable background fill when a plane
703 			 * could be alpha blending from the background, i.e.
704 			 * where no other plane is underneath. It suffices to
705 			 * consider the first active plane here since we set
706 			 * needs_bg_fill such that either the first plane
707 			 * already needs it or all planes on top blend from
708 			 * the first or a lower plane.
709 			 */
710 			vc4_plane_state = to_vc4_plane_state(plane->state);
711 			enable_bg_fill = vc4_plane_state->needs_bg_fill;
712 		}
713 
714 		dlist_next += vc4_plane_write_dlist(plane, dlist_next);
715 	}
716 
717 	writel(SCALER_CTL0_END, dlist_next);
718 	dlist_next++;
719 
720 	WARN_ON_ONCE(dlist_next - dlist_start != vc4_state->mm.size);
721 
722 	if (enable_bg_fill)
723 		/* This sets a black background color fill, as is the case
724 		 * with other DRM drivers.
725 		 */
726 		HVS_WRITE(SCALER_DISPBKGNDX(vc4_crtc->channel),
727 			  HVS_READ(SCALER_DISPBKGNDX(vc4_crtc->channel)) |
728 			  SCALER_DISPBKGND_FILL);
729 
730 	/* Only update DISPLIST if the CRTC was already running and is not
731 	 * being disabled.
732 	 * vc4_crtc_enable() takes care of updating the dlist just after
733 	 * re-enabling VBLANK interrupts and before enabling the engine.
734 	 * If the CRTC is being disabled, there's no point in updating this
735 	 * information.
736 	 */
737 	if (crtc->state->active && old_state->active)
738 		vc4_crtc_update_dlist(crtc);
739 
740 	if (crtc->state->color_mgmt_changed) {
741 		u32 dispbkgndx = HVS_READ(SCALER_DISPBKGNDX(vc4_crtc->channel));
742 
743 		if (crtc->state->gamma_lut) {
744 			vc4_crtc_update_gamma_lut(crtc);
745 			dispbkgndx |= SCALER_DISPBKGND_GAMMA;
746 		} else {
747 			/* Unsetting DISPBKGND_GAMMA skips the gamma lut step
748 			 * in hardware, which is the same as a linear lut that
749 			 * DRM expects us to use in absence of a user lut.
750 			 */
751 			dispbkgndx &= ~SCALER_DISPBKGND_GAMMA;
752 		}
753 		HVS_WRITE(SCALER_DISPBKGNDX(vc4_crtc->channel), dispbkgndx);
754 	}
755 
756 	if (debug_dump_regs) {
757 		DRM_INFO("CRTC %d HVS after:\n", drm_crtc_index(crtc));
758 		vc4_hvs_dump_state(dev);
759 	}
760 }
761 
762 static int vc4_enable_vblank(struct drm_crtc *crtc)
763 {
764 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
765 
766 	CRTC_WRITE(PV_INTEN, PV_INT_VFP_START);
767 
768 	return 0;
769 }
770 
771 static void vc4_disable_vblank(struct drm_crtc *crtc)
772 {
773 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
774 
775 	CRTC_WRITE(PV_INTEN, 0);
776 }
777 
778 static void vc4_crtc_handle_page_flip(struct vc4_crtc *vc4_crtc)
779 {
780 	struct drm_crtc *crtc = &vc4_crtc->base;
781 	struct drm_device *dev = crtc->dev;
782 	struct vc4_dev *vc4 = to_vc4_dev(dev);
783 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
784 	u32 chan = vc4_crtc->channel;
785 	unsigned long flags;
786 
787 	spin_lock_irqsave(&dev->event_lock, flags);
788 	if (vc4_crtc->event &&
789 	    (vc4_state->mm.start == HVS_READ(SCALER_DISPLACTX(chan)) ||
790 	     vc4_state->feed_txp)) {
791 		drm_crtc_send_vblank_event(crtc, vc4_crtc->event);
792 		vc4_crtc->event = NULL;
793 		drm_crtc_vblank_put(crtc);
794 
795 		/* Wait for the page flip to unmask the underrun to ensure that
796 		 * the display list was updated by the hardware. Before that
797 		 * happens, the HVS will be using the previous display list with
798 		 * the CRTC and encoder already reconfigured, leading to
799 		 * underruns. This can be seen when reconfiguring the CRTC.
800 		 */
801 		vc4_hvs_unmask_underrun(dev, vc4_crtc->channel);
802 	}
803 	spin_unlock_irqrestore(&dev->event_lock, flags);
804 }
805 
806 void vc4_crtc_handle_vblank(struct vc4_crtc *crtc)
807 {
808 	crtc->t_vblank = ktime_get();
809 	drm_crtc_handle_vblank(&crtc->base);
810 	vc4_crtc_handle_page_flip(crtc);
811 }
812 
813 static irqreturn_t vc4_crtc_irq_handler(int irq, void *data)
814 {
815 	struct vc4_crtc *vc4_crtc = data;
816 	u32 stat = CRTC_READ(PV_INTSTAT);
817 	irqreturn_t ret = IRQ_NONE;
818 
819 	if (stat & PV_INT_VFP_START) {
820 		CRTC_WRITE(PV_INTSTAT, PV_INT_VFP_START);
821 		vc4_crtc_handle_vblank(vc4_crtc);
822 		ret = IRQ_HANDLED;
823 	}
824 
825 	return ret;
826 }
827 
828 struct vc4_async_flip_state {
829 	struct drm_crtc *crtc;
830 	struct drm_framebuffer *fb;
831 	struct drm_framebuffer *old_fb;
832 	struct drm_pending_vblank_event *event;
833 
834 	struct vc4_seqno_cb cb;
835 };
836 
837 /* Called when the V3D execution for the BO being flipped to is done, so that
838  * we can actually update the plane's address to point to it.
839  */
840 static void
841 vc4_async_page_flip_complete(struct vc4_seqno_cb *cb)
842 {
843 	struct vc4_async_flip_state *flip_state =
844 		container_of(cb, struct vc4_async_flip_state, cb);
845 	struct drm_crtc *crtc = flip_state->crtc;
846 	struct drm_device *dev = crtc->dev;
847 	struct vc4_dev *vc4 = to_vc4_dev(dev);
848 	struct drm_plane *plane = crtc->primary;
849 
850 	vc4_plane_async_set_fb(plane, flip_state->fb);
851 	if (flip_state->event) {
852 		unsigned long flags;
853 
854 		spin_lock_irqsave(&dev->event_lock, flags);
855 		drm_crtc_send_vblank_event(crtc, flip_state->event);
856 		spin_unlock_irqrestore(&dev->event_lock, flags);
857 	}
858 
859 	drm_crtc_vblank_put(crtc);
860 	drm_framebuffer_put(flip_state->fb);
861 
862 	/* Decrement the BO usecnt in order to keep the inc/dec calls balanced
863 	 * when the planes are updated through the async update path.
864 	 * FIXME: we should move to generic async-page-flip when it's
865 	 * available, so that we can get rid of this hand-made cleanup_fb()
866 	 * logic.
867 	 */
868 	if (flip_state->old_fb) {
869 		struct drm_gem_cma_object *cma_bo;
870 		struct vc4_bo *bo;
871 
872 		cma_bo = drm_fb_cma_get_gem_obj(flip_state->old_fb, 0);
873 		bo = to_vc4_bo(&cma_bo->base);
874 		vc4_bo_dec_usecnt(bo);
875 		drm_framebuffer_put(flip_state->old_fb);
876 	}
877 
878 	kfree(flip_state);
879 
880 	up(&vc4->async_modeset);
881 }
882 
883 /* Implements async (non-vblank-synced) page flips.
884  *
885  * The page flip ioctl needs to return immediately, so we grab the
886  * modeset semaphore on the pipe, and queue the address update for
887  * when V3D is done with the BO being flipped to.
888  */
889 static int vc4_async_page_flip(struct drm_crtc *crtc,
890 			       struct drm_framebuffer *fb,
891 			       struct drm_pending_vblank_event *event,
892 			       uint32_t flags)
893 {
894 	struct drm_device *dev = crtc->dev;
895 	struct vc4_dev *vc4 = to_vc4_dev(dev);
896 	struct drm_plane *plane = crtc->primary;
897 	int ret = 0;
898 	struct vc4_async_flip_state *flip_state;
899 	struct drm_gem_cma_object *cma_bo = drm_fb_cma_get_gem_obj(fb, 0);
900 	struct vc4_bo *bo = to_vc4_bo(&cma_bo->base);
901 
902 	/* Increment the BO usecnt here, so that we never end up with an
903 	 * unbalanced number of vc4_bo_{dec,inc}_usecnt() calls when the
904 	 * plane is later updated through the non-async path.
905 	 * FIXME: we should move to generic async-page-flip when it's
906 	 * available, so that we can get rid of this hand-made prepare_fb()
907 	 * logic.
908 	 */
909 	ret = vc4_bo_inc_usecnt(bo);
910 	if (ret)
911 		return ret;
912 
913 	flip_state = kzalloc(sizeof(*flip_state), GFP_KERNEL);
914 	if (!flip_state) {
915 		vc4_bo_dec_usecnt(bo);
916 		return -ENOMEM;
917 	}
918 
919 	drm_framebuffer_get(fb);
920 	flip_state->fb = fb;
921 	flip_state->crtc = crtc;
922 	flip_state->event = event;
923 
924 	/* Make sure all other async modesetes have landed. */
925 	ret = down_interruptible(&vc4->async_modeset);
926 	if (ret) {
927 		drm_framebuffer_put(fb);
928 		vc4_bo_dec_usecnt(bo);
929 		kfree(flip_state);
930 		return ret;
931 	}
932 
933 	/* Save the current FB before it's replaced by the new one in
934 	 * drm_atomic_set_fb_for_plane(). We'll need the old FB in
935 	 * vc4_async_page_flip_complete() to decrement the BO usecnt and keep
936 	 * it consistent.
937 	 * FIXME: we should move to generic async-page-flip when it's
938 	 * available, so that we can get rid of this hand-made cleanup_fb()
939 	 * logic.
940 	 */
941 	flip_state->old_fb = plane->state->fb;
942 	if (flip_state->old_fb)
943 		drm_framebuffer_get(flip_state->old_fb);
944 
945 	WARN_ON(drm_crtc_vblank_get(crtc) != 0);
946 
947 	/* Immediately update the plane's legacy fb pointer, so that later
948 	 * modeset prep sees the state that will be present when the semaphore
949 	 * is released.
950 	 */
951 	drm_atomic_set_fb_for_plane(plane->state, fb);
952 
953 	vc4_queue_seqno_cb(dev, &flip_state->cb, bo->seqno,
954 			   vc4_async_page_flip_complete);
955 
956 	/* Driver takes ownership of state on successful async commit. */
957 	return 0;
958 }
959 
960 static int vc4_page_flip(struct drm_crtc *crtc,
961 			 struct drm_framebuffer *fb,
962 			 struct drm_pending_vblank_event *event,
963 			 uint32_t flags,
964 			 struct drm_modeset_acquire_ctx *ctx)
965 {
966 	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
967 		return vc4_async_page_flip(crtc, fb, event, flags);
968 	else
969 		return drm_atomic_helper_page_flip(crtc, fb, event, flags, ctx);
970 }
971 
972 static struct drm_crtc_state *vc4_crtc_duplicate_state(struct drm_crtc *crtc)
973 {
974 	struct vc4_crtc_state *vc4_state, *old_vc4_state;
975 
976 	vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
977 	if (!vc4_state)
978 		return NULL;
979 
980 	old_vc4_state = to_vc4_crtc_state(crtc->state);
981 	vc4_state->feed_txp = old_vc4_state->feed_txp;
982 	vc4_state->margins = old_vc4_state->margins;
983 
984 	__drm_atomic_helper_crtc_duplicate_state(crtc, &vc4_state->base);
985 	return &vc4_state->base;
986 }
987 
988 static void vc4_crtc_destroy_state(struct drm_crtc *crtc,
989 				   struct drm_crtc_state *state)
990 {
991 	struct vc4_dev *vc4 = to_vc4_dev(crtc->dev);
992 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(state);
993 
994 	if (vc4_state->mm.allocated) {
995 		unsigned long flags;
996 
997 		spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
998 		drm_mm_remove_node(&vc4_state->mm);
999 		spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
1000 
1001 	}
1002 
1003 	drm_atomic_helper_crtc_destroy_state(crtc, state);
1004 }
1005 
1006 static void
1007 vc4_crtc_reset(struct drm_crtc *crtc)
1008 {
1009 	if (crtc->state)
1010 		vc4_crtc_destroy_state(crtc, crtc->state);
1011 
1012 	crtc->state = kzalloc(sizeof(struct vc4_crtc_state), GFP_KERNEL);
1013 	if (crtc->state)
1014 		crtc->state->crtc = crtc;
1015 }
1016 
1017 static const struct drm_crtc_funcs vc4_crtc_funcs = {
1018 	.set_config = drm_atomic_helper_set_config,
1019 	.destroy = vc4_crtc_destroy,
1020 	.page_flip = vc4_page_flip,
1021 	.set_property = NULL,
1022 	.cursor_set = NULL, /* handled by drm_mode_cursor_universal */
1023 	.cursor_move = NULL, /* handled by drm_mode_cursor_universal */
1024 	.reset = vc4_crtc_reset,
1025 	.atomic_duplicate_state = vc4_crtc_duplicate_state,
1026 	.atomic_destroy_state = vc4_crtc_destroy_state,
1027 	.gamma_set = drm_atomic_helper_legacy_gamma_set,
1028 	.enable_vblank = vc4_enable_vblank,
1029 	.disable_vblank = vc4_disable_vblank,
1030 };
1031 
1032 static const struct drm_crtc_helper_funcs vc4_crtc_helper_funcs = {
1033 	.mode_set_nofb = vc4_crtc_mode_set_nofb,
1034 	.mode_valid = vc4_crtc_mode_valid,
1035 	.atomic_check = vc4_crtc_atomic_check,
1036 	.atomic_flush = vc4_crtc_atomic_flush,
1037 	.atomic_enable = vc4_crtc_atomic_enable,
1038 	.atomic_disable = vc4_crtc_atomic_disable,
1039 };
1040 
1041 static const struct vc4_crtc_data pv0_data = {
1042 	.hvs_channel = 0,
1043 	.debugfs_name = "crtc0_regs",
1044 	.encoder_types = {
1045 		[PV_CONTROL_CLK_SELECT_DSI] = VC4_ENCODER_TYPE_DSI0,
1046 		[PV_CONTROL_CLK_SELECT_DPI_SMI_HDMI] = VC4_ENCODER_TYPE_DPI,
1047 	},
1048 };
1049 
1050 static const struct vc4_crtc_data pv1_data = {
1051 	.hvs_channel = 2,
1052 	.debugfs_name = "crtc1_regs",
1053 	.encoder_types = {
1054 		[PV_CONTROL_CLK_SELECT_DSI] = VC4_ENCODER_TYPE_DSI1,
1055 		[PV_CONTROL_CLK_SELECT_DPI_SMI_HDMI] = VC4_ENCODER_TYPE_SMI,
1056 	},
1057 };
1058 
1059 static const struct vc4_crtc_data pv2_data = {
1060 	.hvs_channel = 1,
1061 	.debugfs_name = "crtc2_regs",
1062 	.encoder_types = {
1063 		[PV_CONTROL_CLK_SELECT_DPI_SMI_HDMI] = VC4_ENCODER_TYPE_HDMI,
1064 		[PV_CONTROL_CLK_SELECT_VEC] = VC4_ENCODER_TYPE_VEC,
1065 	},
1066 };
1067 
1068 static const struct of_device_id vc4_crtc_dt_match[] = {
1069 	{ .compatible = "brcm,bcm2835-pixelvalve0", .data = &pv0_data },
1070 	{ .compatible = "brcm,bcm2835-pixelvalve1", .data = &pv1_data },
1071 	{ .compatible = "brcm,bcm2835-pixelvalve2", .data = &pv2_data },
1072 	{}
1073 };
1074 
1075 static void vc4_set_crtc_possible_masks(struct drm_device *drm,
1076 					struct drm_crtc *crtc)
1077 {
1078 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
1079 	const struct vc4_crtc_data *crtc_data = vc4_crtc->data;
1080 	const enum vc4_encoder_type *encoder_types = crtc_data->encoder_types;
1081 	struct drm_encoder *encoder;
1082 
1083 	drm_for_each_encoder(encoder, drm) {
1084 		struct vc4_encoder *vc4_encoder;
1085 		int i;
1086 
1087 		/* HVS FIFO2 can feed the TXP IP. */
1088 		if (crtc_data->hvs_channel == 2 &&
1089 		    encoder->encoder_type == DRM_MODE_ENCODER_VIRTUAL) {
1090 			encoder->possible_crtcs |= drm_crtc_mask(crtc);
1091 			continue;
1092 		}
1093 
1094 		vc4_encoder = to_vc4_encoder(encoder);
1095 		for (i = 0; i < ARRAY_SIZE(crtc_data->encoder_types); i++) {
1096 			if (vc4_encoder->type == encoder_types[i]) {
1097 				vc4_encoder->clock_select = i;
1098 				encoder->possible_crtcs |= drm_crtc_mask(crtc);
1099 				break;
1100 			}
1101 		}
1102 	}
1103 }
1104 
1105 static void
1106 vc4_crtc_get_cob_allocation(struct vc4_crtc *vc4_crtc)
1107 {
1108 	struct drm_device *drm = vc4_crtc->base.dev;
1109 	struct vc4_dev *vc4 = to_vc4_dev(drm);
1110 	u32 dispbase = HVS_READ(SCALER_DISPBASEX(vc4_crtc->channel));
1111 	/* Top/base are supposed to be 4-pixel aligned, but the
1112 	 * Raspberry Pi firmware fills the low bits (which are
1113 	 * presumably ignored).
1114 	 */
1115 	u32 top = VC4_GET_FIELD(dispbase, SCALER_DISPBASEX_TOP) & ~3;
1116 	u32 base = VC4_GET_FIELD(dispbase, SCALER_DISPBASEX_BASE) & ~3;
1117 
1118 	vc4_crtc->cob_size = top - base + 4;
1119 }
1120 
1121 static int vc4_crtc_bind(struct device *dev, struct device *master, void *data)
1122 {
1123 	struct platform_device *pdev = to_platform_device(dev);
1124 	struct drm_device *drm = dev_get_drvdata(master);
1125 	struct vc4_crtc *vc4_crtc;
1126 	struct drm_crtc *crtc;
1127 	struct drm_plane *primary_plane, *cursor_plane, *destroy_plane, *temp;
1128 	const struct of_device_id *match;
1129 	int ret, i;
1130 
1131 	vc4_crtc = devm_kzalloc(dev, sizeof(*vc4_crtc), GFP_KERNEL);
1132 	if (!vc4_crtc)
1133 		return -ENOMEM;
1134 	crtc = &vc4_crtc->base;
1135 
1136 	match = of_match_device(vc4_crtc_dt_match, dev);
1137 	if (!match)
1138 		return -ENODEV;
1139 	vc4_crtc->data = match->data;
1140 	vc4_crtc->pdev = pdev;
1141 
1142 	vc4_crtc->regs = vc4_ioremap_regs(pdev, 0);
1143 	if (IS_ERR(vc4_crtc->regs))
1144 		return PTR_ERR(vc4_crtc->regs);
1145 
1146 	vc4_crtc->regset.base = vc4_crtc->regs;
1147 	vc4_crtc->regset.regs = crtc_regs;
1148 	vc4_crtc->regset.nregs = ARRAY_SIZE(crtc_regs);
1149 
1150 	/* For now, we create just the primary and the legacy cursor
1151 	 * planes.  We should be able to stack more planes on easily,
1152 	 * but to do that we would need to compute the bandwidth
1153 	 * requirement of the plane configuration, and reject ones
1154 	 * that will take too much.
1155 	 */
1156 	primary_plane = vc4_plane_init(drm, DRM_PLANE_TYPE_PRIMARY);
1157 	if (IS_ERR(primary_plane)) {
1158 		dev_err(dev, "failed to construct primary plane\n");
1159 		ret = PTR_ERR(primary_plane);
1160 		goto err;
1161 	}
1162 
1163 	drm_crtc_init_with_planes(drm, crtc, primary_plane, NULL,
1164 				  &vc4_crtc_funcs, NULL);
1165 	drm_crtc_helper_add(crtc, &vc4_crtc_helper_funcs);
1166 	vc4_crtc->channel = vc4_crtc->data->hvs_channel;
1167 	drm_mode_crtc_set_gamma_size(crtc, ARRAY_SIZE(vc4_crtc->lut_r));
1168 	drm_crtc_enable_color_mgmt(crtc, 0, false, crtc->gamma_size);
1169 
1170 	/* We support CTM, but only for one CRTC at a time. It's therefore
1171 	 * implemented as private driver state in vc4_kms, not here.
1172 	 */
1173 	drm_crtc_enable_color_mgmt(crtc, 0, true, crtc->gamma_size);
1174 
1175 	/* Set up some arbitrary number of planes.  We're not limited
1176 	 * by a set number of physical registers, just the space in
1177 	 * the HVS (16k) and how small an plane can be (28 bytes).
1178 	 * However, each plane we set up takes up some memory, and
1179 	 * increases the cost of looping over planes, which atomic
1180 	 * modesetting does quite a bit.  As a result, we pick a
1181 	 * modest number of planes to expose, that should hopefully
1182 	 * still cover any sane usecase.
1183 	 */
1184 	for (i = 0; i < 8; i++) {
1185 		struct drm_plane *plane =
1186 			vc4_plane_init(drm, DRM_PLANE_TYPE_OVERLAY);
1187 
1188 		if (IS_ERR(plane))
1189 			continue;
1190 
1191 		plane->possible_crtcs = drm_crtc_mask(crtc);
1192 	}
1193 
1194 	/* Set up the legacy cursor after overlay initialization,
1195 	 * since we overlay planes on the CRTC in the order they were
1196 	 * initialized.
1197 	 */
1198 	cursor_plane = vc4_plane_init(drm, DRM_PLANE_TYPE_CURSOR);
1199 	if (!IS_ERR(cursor_plane)) {
1200 		cursor_plane->possible_crtcs = drm_crtc_mask(crtc);
1201 		crtc->cursor = cursor_plane;
1202 	}
1203 
1204 	vc4_crtc_get_cob_allocation(vc4_crtc);
1205 
1206 	CRTC_WRITE(PV_INTEN, 0);
1207 	CRTC_WRITE(PV_INTSTAT, PV_INT_VFP_START);
1208 	ret = devm_request_irq(dev, platform_get_irq(pdev, 0),
1209 			       vc4_crtc_irq_handler, 0, "vc4 crtc", vc4_crtc);
1210 	if (ret)
1211 		goto err_destroy_planes;
1212 
1213 	vc4_set_crtc_possible_masks(drm, crtc);
1214 
1215 	for (i = 0; i < crtc->gamma_size; i++) {
1216 		vc4_crtc->lut_r[i] = i;
1217 		vc4_crtc->lut_g[i] = i;
1218 		vc4_crtc->lut_b[i] = i;
1219 	}
1220 
1221 	platform_set_drvdata(pdev, vc4_crtc);
1222 
1223 	vc4_debugfs_add_regset32(drm, vc4_crtc->data->debugfs_name,
1224 				 &vc4_crtc->regset);
1225 
1226 	return 0;
1227 
1228 err_destroy_planes:
1229 	list_for_each_entry_safe(destroy_plane, temp,
1230 				 &drm->mode_config.plane_list, head) {
1231 		if (destroy_plane->possible_crtcs == drm_crtc_mask(crtc))
1232 		    destroy_plane->funcs->destroy(destroy_plane);
1233 	}
1234 err:
1235 	return ret;
1236 }
1237 
1238 static void vc4_crtc_unbind(struct device *dev, struct device *master,
1239 			    void *data)
1240 {
1241 	struct platform_device *pdev = to_platform_device(dev);
1242 	struct vc4_crtc *vc4_crtc = dev_get_drvdata(dev);
1243 
1244 	vc4_crtc_destroy(&vc4_crtc->base);
1245 
1246 	CRTC_WRITE(PV_INTEN, 0);
1247 
1248 	platform_set_drvdata(pdev, NULL);
1249 }
1250 
1251 static const struct component_ops vc4_crtc_ops = {
1252 	.bind   = vc4_crtc_bind,
1253 	.unbind = vc4_crtc_unbind,
1254 };
1255 
1256 static int vc4_crtc_dev_probe(struct platform_device *pdev)
1257 {
1258 	return component_add(&pdev->dev, &vc4_crtc_ops);
1259 }
1260 
1261 static int vc4_crtc_dev_remove(struct platform_device *pdev)
1262 {
1263 	component_del(&pdev->dev, &vc4_crtc_ops);
1264 	return 0;
1265 }
1266 
1267 struct platform_driver vc4_crtc_driver = {
1268 	.probe = vc4_crtc_dev_probe,
1269 	.remove = vc4_crtc_dev_remove,
1270 	.driver = {
1271 		.name = "vc4_crtc",
1272 		.of_match_table = vc4_crtc_dt_match,
1273 	},
1274 };
1275