xref: /openbmc/linux/drivers/gpu/drm/vc4/vc4_hvs.c (revision 79e790ff)
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 drm_device *dev)
68 {
69 	struct vc4_dev *vc4 = to_vc4_dev(dev);
70 	struct drm_printer p = drm_info_printer(&vc4->hvs->pdev->dev);
71 	int i;
72 
73 	drm_print_regset32(&p, &vc4->hvs->regset);
74 
75 	DRM_INFO("HVS ctx:\n");
76 	for (i = 0; i < 64; i += 4) {
77 		DRM_INFO("0x%08x (%s): 0x%08x 0x%08x 0x%08x 0x%08x\n",
78 			 i * 4, i < HVS_BOOTLOADER_DLIST_END ? "B" : "D",
79 			 readl((u32 __iomem *)vc4->hvs->dlist + i + 0),
80 			 readl((u32 __iomem *)vc4->hvs->dlist + i + 1),
81 			 readl((u32 __iomem *)vc4->hvs->dlist + i + 2),
82 			 readl((u32 __iomem *)vc4->hvs->dlist + i + 3));
83 	}
84 }
85 
86 static int vc4_hvs_debugfs_underrun(struct seq_file *m, void *data)
87 {
88 	struct drm_info_node *node = m->private;
89 	struct drm_device *dev = node->minor->dev;
90 	struct vc4_dev *vc4 = to_vc4_dev(dev);
91 	struct drm_printer p = drm_seq_file_printer(m);
92 
93 	drm_printf(&p, "%d\n", atomic_read(&vc4->underrun));
94 
95 	return 0;
96 }
97 
98 /* The filter kernel is composed of dwords each containing 3 9-bit
99  * signed integers packed next to each other.
100  */
101 #define VC4_INT_TO_COEFF(coeff) (coeff & 0x1ff)
102 #define VC4_PPF_FILTER_WORD(c0, c1, c2)				\
103 	((((c0) & 0x1ff) << 0) |				\
104 	 (((c1) & 0x1ff) << 9) |				\
105 	 (((c2) & 0x1ff) << 18))
106 
107 /* The whole filter kernel is arranged as the coefficients 0-16 going
108  * up, then a pad, then 17-31 going down and reversed within the
109  * dwords.  This means that a linear phase kernel (where it's
110  * symmetrical at the boundary between 15 and 16) has the last 5
111  * dwords matching the first 5, but reversed.
112  */
113 #define VC4_LINEAR_PHASE_KERNEL(c0, c1, c2, c3, c4, c5, c6, c7, c8,	\
114 				c9, c10, c11, c12, c13, c14, c15)	\
115 	{VC4_PPF_FILTER_WORD(c0, c1, c2),				\
116 	 VC4_PPF_FILTER_WORD(c3, c4, c5),				\
117 	 VC4_PPF_FILTER_WORD(c6, c7, c8),				\
118 	 VC4_PPF_FILTER_WORD(c9, c10, c11),				\
119 	 VC4_PPF_FILTER_WORD(c12, c13, c14),				\
120 	 VC4_PPF_FILTER_WORD(c15, c15, 0)}
121 
122 #define VC4_LINEAR_PHASE_KERNEL_DWORDS 6
123 #define VC4_KERNEL_DWORDS (VC4_LINEAR_PHASE_KERNEL_DWORDS * 2 - 1)
124 
125 /* Recommended B=1/3, C=1/3 filter choice from Mitchell/Netravali.
126  * http://www.cs.utexas.edu/~fussell/courses/cs384g/lectures/mitchell/Mitchell.pdf
127  */
128 static const u32 mitchell_netravali_1_3_1_3_kernel[] =
129 	VC4_LINEAR_PHASE_KERNEL(0, -2, -6, -8, -10, -8, -3, 2, 18,
130 				50, 82, 119, 155, 187, 213, 227);
131 
132 static int vc4_hvs_upload_linear_kernel(struct vc4_hvs *hvs,
133 					struct drm_mm_node *space,
134 					const u32 *kernel)
135 {
136 	int ret, i;
137 	u32 __iomem *dst_kernel;
138 
139 	ret = drm_mm_insert_node(&hvs->dlist_mm, space, VC4_KERNEL_DWORDS);
140 	if (ret) {
141 		DRM_ERROR("Failed to allocate space for filter kernel: %d\n",
142 			  ret);
143 		return ret;
144 	}
145 
146 	dst_kernel = hvs->dlist + space->start;
147 
148 	for (i = 0; i < VC4_KERNEL_DWORDS; i++) {
149 		if (i < VC4_LINEAR_PHASE_KERNEL_DWORDS)
150 			writel(kernel[i], &dst_kernel[i]);
151 		else {
152 			writel(kernel[VC4_KERNEL_DWORDS - i - 1],
153 			       &dst_kernel[i]);
154 		}
155 	}
156 
157 	return 0;
158 }
159 
160 static void vc4_hvs_lut_load(struct drm_crtc *crtc)
161 {
162 	struct drm_device *dev = crtc->dev;
163 	struct vc4_dev *vc4 = to_vc4_dev(dev);
164 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
165 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
166 	u32 i;
167 
168 	/* The LUT memory is laid out with each HVS channel in order,
169 	 * each of which takes 256 writes for R, 256 for G, then 256
170 	 * for B.
171 	 */
172 	HVS_WRITE(SCALER_GAMADDR,
173 		  SCALER_GAMADDR_AUTOINC |
174 		  (vc4_state->assigned_channel * 3 * crtc->gamma_size));
175 
176 	for (i = 0; i < crtc->gamma_size; i++)
177 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_r[i]);
178 	for (i = 0; i < crtc->gamma_size; i++)
179 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_g[i]);
180 	for (i = 0; i < crtc->gamma_size; i++)
181 		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_b[i]);
182 }
183 
184 static void vc4_hvs_update_gamma_lut(struct drm_crtc *crtc)
185 {
186 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
187 	struct drm_color_lut *lut = crtc->state->gamma_lut->data;
188 	u32 length = drm_color_lut_size(crtc->state->gamma_lut);
189 	u32 i;
190 
191 	for (i = 0; i < length; i++) {
192 		vc4_crtc->lut_r[i] = drm_color_lut_extract(lut[i].red, 8);
193 		vc4_crtc->lut_g[i] = drm_color_lut_extract(lut[i].green, 8);
194 		vc4_crtc->lut_b[i] = drm_color_lut_extract(lut[i].blue, 8);
195 	}
196 
197 	vc4_hvs_lut_load(crtc);
198 }
199 
200 int vc4_hvs_get_fifo_from_output(struct drm_device *dev, unsigned int output)
201 {
202 	struct vc4_dev *vc4 = to_vc4_dev(dev);
203 	u32 reg;
204 	int ret;
205 
206 	if (!vc4->hvs->hvs5)
207 		return output;
208 
209 	switch (output) {
210 	case 0:
211 		return 0;
212 
213 	case 1:
214 		return 1;
215 
216 	case 2:
217 		reg = HVS_READ(SCALER_DISPECTRL);
218 		ret = FIELD_GET(SCALER_DISPECTRL_DSP2_MUX_MASK, reg);
219 		if (ret == 0)
220 			return 2;
221 
222 		return 0;
223 
224 	case 3:
225 		reg = HVS_READ(SCALER_DISPCTRL);
226 		ret = FIELD_GET(SCALER_DISPCTRL_DSP3_MUX_MASK, reg);
227 		if (ret == 3)
228 			return -EPIPE;
229 
230 		return ret;
231 
232 	case 4:
233 		reg = HVS_READ(SCALER_DISPEOLN);
234 		ret = FIELD_GET(SCALER_DISPEOLN_DSP4_MUX_MASK, reg);
235 		if (ret == 3)
236 			return -EPIPE;
237 
238 		return ret;
239 
240 	case 5:
241 		reg = HVS_READ(SCALER_DISPDITHER);
242 		ret = FIELD_GET(SCALER_DISPDITHER_DSP5_MUX_MASK, reg);
243 		if (ret == 3)
244 			return -EPIPE;
245 
246 		return ret;
247 
248 	default:
249 		return -EPIPE;
250 	}
251 }
252 
253 static int vc4_hvs_init_channel(struct vc4_dev *vc4, struct drm_crtc *crtc,
254 				struct drm_display_mode *mode, bool oneshot)
255 {
256 	struct vc4_crtc_state *vc4_crtc_state = to_vc4_crtc_state(crtc->state);
257 	unsigned int chan = vc4_crtc_state->assigned_channel;
258 	bool interlace = mode->flags & DRM_MODE_FLAG_INTERLACE;
259 	u32 dispbkgndx;
260 	u32 dispctrl;
261 
262 	HVS_WRITE(SCALER_DISPCTRLX(chan), 0);
263 	HVS_WRITE(SCALER_DISPCTRLX(chan), SCALER_DISPCTRLX_RESET);
264 	HVS_WRITE(SCALER_DISPCTRLX(chan), 0);
265 
266 	/* Turn on the scaler, which will wait for vstart to start
267 	 * compositing.
268 	 * When feeding the transposer, we should operate in oneshot
269 	 * mode.
270 	 */
271 	dispctrl = SCALER_DISPCTRLX_ENABLE;
272 
273 	if (!vc4->hvs->hvs5)
274 		dispctrl |= VC4_SET_FIELD(mode->hdisplay,
275 					  SCALER_DISPCTRLX_WIDTH) |
276 			    VC4_SET_FIELD(mode->vdisplay,
277 					  SCALER_DISPCTRLX_HEIGHT) |
278 			    (oneshot ? SCALER_DISPCTRLX_ONESHOT : 0);
279 	else
280 		dispctrl |= VC4_SET_FIELD(mode->hdisplay,
281 					  SCALER5_DISPCTRLX_WIDTH) |
282 			    VC4_SET_FIELD(mode->vdisplay,
283 					  SCALER5_DISPCTRLX_HEIGHT) |
284 			    (oneshot ? SCALER5_DISPCTRLX_ONESHOT : 0);
285 
286 	HVS_WRITE(SCALER_DISPCTRLX(chan), dispctrl);
287 
288 	dispbkgndx = HVS_READ(SCALER_DISPBKGNDX(chan));
289 	dispbkgndx &= ~SCALER_DISPBKGND_GAMMA;
290 	dispbkgndx &= ~SCALER_DISPBKGND_INTERLACE;
291 
292 	HVS_WRITE(SCALER_DISPBKGNDX(chan), dispbkgndx |
293 		  SCALER_DISPBKGND_AUTOHS |
294 		  ((!vc4->hvs->hvs5) ? SCALER_DISPBKGND_GAMMA : 0) |
295 		  (interlace ? SCALER_DISPBKGND_INTERLACE : 0));
296 
297 	/* Reload the LUT, since the SRAMs would have been disabled if
298 	 * all CRTCs had SCALER_DISPBKGND_GAMMA unset at once.
299 	 */
300 	vc4_hvs_lut_load(crtc);
301 
302 	return 0;
303 }
304 
305 void vc4_hvs_stop_channel(struct drm_device *dev, unsigned int chan)
306 {
307 	struct vc4_dev *vc4 = to_vc4_dev(dev);
308 
309 	if (HVS_READ(SCALER_DISPCTRLX(chan)) & SCALER_DISPCTRLX_ENABLE)
310 		return;
311 
312 	HVS_WRITE(SCALER_DISPCTRLX(chan),
313 		  HVS_READ(SCALER_DISPCTRLX(chan)) | SCALER_DISPCTRLX_RESET);
314 	HVS_WRITE(SCALER_DISPCTRLX(chan),
315 		  HVS_READ(SCALER_DISPCTRLX(chan)) & ~SCALER_DISPCTRLX_ENABLE);
316 
317 	/* Once we leave, the scaler should be disabled and its fifo empty. */
318 	WARN_ON_ONCE(HVS_READ(SCALER_DISPCTRLX(chan)) & SCALER_DISPCTRLX_RESET);
319 
320 	WARN_ON_ONCE(VC4_GET_FIELD(HVS_READ(SCALER_DISPSTATX(chan)),
321 				   SCALER_DISPSTATX_MODE) !=
322 		     SCALER_DISPSTATX_MODE_DISABLED);
323 
324 	WARN_ON_ONCE((HVS_READ(SCALER_DISPSTATX(chan)) &
325 		      (SCALER_DISPSTATX_FULL | SCALER_DISPSTATX_EMPTY)) !=
326 		     SCALER_DISPSTATX_EMPTY);
327 }
328 
329 int vc4_hvs_atomic_check(struct drm_crtc *crtc, struct drm_atomic_state *state)
330 {
331 	struct drm_crtc_state *crtc_state = drm_atomic_get_new_crtc_state(state, crtc);
332 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc_state);
333 	struct drm_device *dev = crtc->dev;
334 	struct vc4_dev *vc4 = to_vc4_dev(dev);
335 	struct drm_plane *plane;
336 	unsigned long flags;
337 	const struct drm_plane_state *plane_state;
338 	u32 dlist_count = 0;
339 	int ret;
340 
341 	/* The pixelvalve can only feed one encoder (and encoders are
342 	 * 1:1 with connectors.)
343 	 */
344 	if (hweight32(crtc_state->connector_mask) > 1)
345 		return -EINVAL;
346 
347 	drm_atomic_crtc_state_for_each_plane_state(plane, plane_state, crtc_state)
348 		dlist_count += vc4_plane_dlist_size(plane_state);
349 
350 	dlist_count++; /* Account for SCALER_CTL0_END. */
351 
352 	spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
353 	ret = drm_mm_insert_node(&vc4->hvs->dlist_mm, &vc4_state->mm,
354 				 dlist_count);
355 	spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
356 	if (ret)
357 		return ret;
358 
359 	return 0;
360 }
361 
362 static void vc4_hvs_update_dlist(struct drm_crtc *crtc)
363 {
364 	struct drm_device *dev = crtc->dev;
365 	struct vc4_dev *vc4 = to_vc4_dev(dev);
366 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
367 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
368 
369 	if (crtc->state->event) {
370 		unsigned long flags;
371 
372 		crtc->state->event->pipe = drm_crtc_index(crtc);
373 
374 		WARN_ON(drm_crtc_vblank_get(crtc) != 0);
375 
376 		spin_lock_irqsave(&dev->event_lock, flags);
377 
378 		if (!vc4_state->feed_txp || vc4_state->txp_armed) {
379 			vc4_crtc->event = crtc->state->event;
380 			crtc->state->event = NULL;
381 		}
382 
383 		HVS_WRITE(SCALER_DISPLISTX(vc4_state->assigned_channel),
384 			  vc4_state->mm.start);
385 
386 		spin_unlock_irqrestore(&dev->event_lock, flags);
387 	} else {
388 		HVS_WRITE(SCALER_DISPLISTX(vc4_state->assigned_channel),
389 			  vc4_state->mm.start);
390 	}
391 }
392 
393 void vc4_hvs_atomic_enable(struct drm_crtc *crtc,
394 			   struct drm_atomic_state *state)
395 {
396 	struct drm_device *dev = crtc->dev;
397 	struct vc4_dev *vc4 = to_vc4_dev(dev);
398 	struct drm_crtc_state *new_crtc_state = drm_atomic_get_new_crtc_state(state, crtc);
399 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(new_crtc_state);
400 	struct drm_display_mode *mode = &crtc->state->adjusted_mode;
401 	bool oneshot = vc4_state->feed_txp;
402 
403 	vc4_hvs_update_dlist(crtc);
404 	vc4_hvs_init_channel(vc4, crtc, mode, oneshot);
405 }
406 
407 void vc4_hvs_atomic_disable(struct drm_crtc *crtc,
408 			    struct drm_atomic_state *state)
409 {
410 	struct drm_device *dev = crtc->dev;
411 	struct drm_crtc_state *old_state = drm_atomic_get_old_crtc_state(state, crtc);
412 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(old_state);
413 	unsigned int chan = vc4_state->assigned_channel;
414 
415 	vc4_hvs_stop_channel(dev, chan);
416 }
417 
418 void vc4_hvs_atomic_flush(struct drm_crtc *crtc,
419 			  struct drm_atomic_state *state)
420 {
421 	struct drm_crtc_state *old_state = drm_atomic_get_old_crtc_state(state,
422 									 crtc);
423 	struct drm_device *dev = crtc->dev;
424 	struct vc4_dev *vc4 = to_vc4_dev(dev);
425 	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
426 	struct drm_plane *plane;
427 	struct vc4_plane_state *vc4_plane_state;
428 	bool debug_dump_regs = false;
429 	bool enable_bg_fill = false;
430 	u32 __iomem *dlist_start = vc4->hvs->dlist + vc4_state->mm.start;
431 	u32 __iomem *dlist_next = dlist_start;
432 
433 	if (debug_dump_regs) {
434 		DRM_INFO("CRTC %d HVS before:\n", drm_crtc_index(crtc));
435 		vc4_hvs_dump_state(dev);
436 	}
437 
438 	/* Copy all the active planes' dlist contents to the hardware dlist. */
439 	drm_atomic_crtc_for_each_plane(plane, crtc) {
440 		/* Is this the first active plane? */
441 		if (dlist_next == dlist_start) {
442 			/* We need to enable background fill when a plane
443 			 * could be alpha blending from the background, i.e.
444 			 * where no other plane is underneath. It suffices to
445 			 * consider the first active plane here since we set
446 			 * needs_bg_fill such that either the first plane
447 			 * already needs it or all planes on top blend from
448 			 * the first or a lower plane.
449 			 */
450 			vc4_plane_state = to_vc4_plane_state(plane->state);
451 			enable_bg_fill = vc4_plane_state->needs_bg_fill;
452 		}
453 
454 		dlist_next += vc4_plane_write_dlist(plane, dlist_next);
455 	}
456 
457 	writel(SCALER_CTL0_END, dlist_next);
458 	dlist_next++;
459 
460 	WARN_ON_ONCE(dlist_next - dlist_start != vc4_state->mm.size);
461 
462 	if (enable_bg_fill)
463 		/* This sets a black background color fill, as is the case
464 		 * with other DRM drivers.
465 		 */
466 		HVS_WRITE(SCALER_DISPBKGNDX(vc4_state->assigned_channel),
467 			  HVS_READ(SCALER_DISPBKGNDX(vc4_state->assigned_channel)) |
468 			  SCALER_DISPBKGND_FILL);
469 
470 	/* Only update DISPLIST if the CRTC was already running and is not
471 	 * being disabled.
472 	 * vc4_crtc_enable() takes care of updating the dlist just after
473 	 * re-enabling VBLANK interrupts and before enabling the engine.
474 	 * If the CRTC is being disabled, there's no point in updating this
475 	 * information.
476 	 */
477 	if (crtc->state->active && old_state->active)
478 		vc4_hvs_update_dlist(crtc);
479 
480 	if (crtc->state->color_mgmt_changed) {
481 		u32 dispbkgndx = HVS_READ(SCALER_DISPBKGNDX(vc4_state->assigned_channel));
482 
483 		if (crtc->state->gamma_lut) {
484 			vc4_hvs_update_gamma_lut(crtc);
485 			dispbkgndx |= SCALER_DISPBKGND_GAMMA;
486 		} else {
487 			/* Unsetting DISPBKGND_GAMMA skips the gamma lut step
488 			 * in hardware, which is the same as a linear lut that
489 			 * DRM expects us to use in absence of a user lut.
490 			 */
491 			dispbkgndx &= ~SCALER_DISPBKGND_GAMMA;
492 		}
493 		HVS_WRITE(SCALER_DISPBKGNDX(vc4_state->assigned_channel), dispbkgndx);
494 	}
495 
496 	if (debug_dump_regs) {
497 		DRM_INFO("CRTC %d HVS after:\n", drm_crtc_index(crtc));
498 		vc4_hvs_dump_state(dev);
499 	}
500 }
501 
502 void vc4_hvs_mask_underrun(struct drm_device *dev, int channel)
503 {
504 	struct vc4_dev *vc4 = to_vc4_dev(dev);
505 	u32 dispctrl = HVS_READ(SCALER_DISPCTRL);
506 
507 	dispctrl &= ~SCALER_DISPCTRL_DSPEISLUR(channel);
508 
509 	HVS_WRITE(SCALER_DISPCTRL, dispctrl);
510 }
511 
512 void vc4_hvs_unmask_underrun(struct drm_device *dev, int channel)
513 {
514 	struct vc4_dev *vc4 = to_vc4_dev(dev);
515 	u32 dispctrl = HVS_READ(SCALER_DISPCTRL);
516 
517 	dispctrl |= SCALER_DISPCTRL_DSPEISLUR(channel);
518 
519 	HVS_WRITE(SCALER_DISPSTAT,
520 		  SCALER_DISPSTAT_EUFLOW(channel));
521 	HVS_WRITE(SCALER_DISPCTRL, dispctrl);
522 }
523 
524 static void vc4_hvs_report_underrun(struct drm_device *dev)
525 {
526 	struct vc4_dev *vc4 = to_vc4_dev(dev);
527 
528 	atomic_inc(&vc4->underrun);
529 	DRM_DEV_ERROR(dev->dev, "HVS underrun\n");
530 }
531 
532 static irqreturn_t vc4_hvs_irq_handler(int irq, void *data)
533 {
534 	struct drm_device *dev = data;
535 	struct vc4_dev *vc4 = to_vc4_dev(dev);
536 	irqreturn_t irqret = IRQ_NONE;
537 	int channel;
538 	u32 control;
539 	u32 status;
540 
541 	status = HVS_READ(SCALER_DISPSTAT);
542 	control = HVS_READ(SCALER_DISPCTRL);
543 
544 	for (channel = 0; channel < SCALER_CHANNELS_COUNT; channel++) {
545 		/* Interrupt masking is not always honored, so check it here. */
546 		if (status & SCALER_DISPSTAT_EUFLOW(channel) &&
547 		    control & SCALER_DISPCTRL_DSPEISLUR(channel)) {
548 			vc4_hvs_mask_underrun(dev, channel);
549 			vc4_hvs_report_underrun(dev);
550 
551 			irqret = IRQ_HANDLED;
552 		}
553 	}
554 
555 	/* Clear every per-channel interrupt flag. */
556 	HVS_WRITE(SCALER_DISPSTAT, SCALER_DISPSTAT_IRQMASK(0) |
557 				   SCALER_DISPSTAT_IRQMASK(1) |
558 				   SCALER_DISPSTAT_IRQMASK(2));
559 
560 	return irqret;
561 }
562 
563 static int vc4_hvs_bind(struct device *dev, struct device *master, void *data)
564 {
565 	struct platform_device *pdev = to_platform_device(dev);
566 	struct drm_device *drm = dev_get_drvdata(master);
567 	struct vc4_dev *vc4 = to_vc4_dev(drm);
568 	struct vc4_hvs *hvs = NULL;
569 	int ret;
570 	u32 dispctrl;
571 
572 	hvs = devm_kzalloc(&pdev->dev, sizeof(*hvs), GFP_KERNEL);
573 	if (!hvs)
574 		return -ENOMEM;
575 
576 	hvs->pdev = pdev;
577 
578 	if (of_device_is_compatible(pdev->dev.of_node, "brcm,bcm2711-hvs"))
579 		hvs->hvs5 = true;
580 
581 	hvs->regs = vc4_ioremap_regs(pdev, 0);
582 	if (IS_ERR(hvs->regs))
583 		return PTR_ERR(hvs->regs);
584 
585 	hvs->regset.base = hvs->regs;
586 	hvs->regset.regs = hvs_regs;
587 	hvs->regset.nregs = ARRAY_SIZE(hvs_regs);
588 
589 	if (hvs->hvs5) {
590 		hvs->core_clk = devm_clk_get(&pdev->dev, NULL);
591 		if (IS_ERR(hvs->core_clk)) {
592 			dev_err(&pdev->dev, "Couldn't get core clock\n");
593 			return PTR_ERR(hvs->core_clk);
594 		}
595 
596 		ret = clk_prepare_enable(hvs->core_clk);
597 		if (ret) {
598 			dev_err(&pdev->dev, "Couldn't enable the core clock\n");
599 			return ret;
600 		}
601 	}
602 
603 	if (!hvs->hvs5)
604 		hvs->dlist = hvs->regs + SCALER_DLIST_START;
605 	else
606 		hvs->dlist = hvs->regs + SCALER5_DLIST_START;
607 
608 	spin_lock_init(&hvs->mm_lock);
609 
610 	/* Set up the HVS display list memory manager.  We never
611 	 * overwrite the setup from the bootloader (just 128b out of
612 	 * our 16K), since we don't want to scramble the screen when
613 	 * transitioning from the firmware's boot setup to runtime.
614 	 */
615 	drm_mm_init(&hvs->dlist_mm,
616 		    HVS_BOOTLOADER_DLIST_END,
617 		    (SCALER_DLIST_SIZE >> 2) - HVS_BOOTLOADER_DLIST_END);
618 
619 	/* Set up the HVS LBM memory manager.  We could have some more
620 	 * complicated data structure that allowed reuse of LBM areas
621 	 * between planes when they don't overlap on the screen, but
622 	 * for now we just allocate globally.
623 	 */
624 	if (!hvs->hvs5)
625 		/* 48k words of 2x12-bit pixels */
626 		drm_mm_init(&hvs->lbm_mm, 0, 48 * 1024);
627 	else
628 		/* 60k words of 4x12-bit pixels */
629 		drm_mm_init(&hvs->lbm_mm, 0, 60 * 1024);
630 
631 	/* Upload filter kernels.  We only have the one for now, so we
632 	 * keep it around for the lifetime of the driver.
633 	 */
634 	ret = vc4_hvs_upload_linear_kernel(hvs,
635 					   &hvs->mitchell_netravali_filter,
636 					   mitchell_netravali_1_3_1_3_kernel);
637 	if (ret)
638 		return ret;
639 
640 	vc4->hvs = hvs;
641 
642 	dispctrl = HVS_READ(SCALER_DISPCTRL);
643 
644 	dispctrl |= SCALER_DISPCTRL_ENABLE;
645 	dispctrl |= SCALER_DISPCTRL_DISPEIRQ(0) |
646 		    SCALER_DISPCTRL_DISPEIRQ(1) |
647 		    SCALER_DISPCTRL_DISPEIRQ(2);
648 
649 	/* Set DSP3 (PV1) to use HVS channel 2, which would otherwise
650 	 * be unused.
651 	 */
652 	dispctrl &= ~SCALER_DISPCTRL_DSP3_MUX_MASK;
653 	dispctrl &= ~(SCALER_DISPCTRL_DMAEIRQ |
654 		      SCALER_DISPCTRL_SLVWREIRQ |
655 		      SCALER_DISPCTRL_SLVRDEIRQ |
656 		      SCALER_DISPCTRL_DSPEIEOF(0) |
657 		      SCALER_DISPCTRL_DSPEIEOF(1) |
658 		      SCALER_DISPCTRL_DSPEIEOF(2) |
659 		      SCALER_DISPCTRL_DSPEIEOLN(0) |
660 		      SCALER_DISPCTRL_DSPEIEOLN(1) |
661 		      SCALER_DISPCTRL_DSPEIEOLN(2) |
662 		      SCALER_DISPCTRL_DSPEISLUR(0) |
663 		      SCALER_DISPCTRL_DSPEISLUR(1) |
664 		      SCALER_DISPCTRL_DSPEISLUR(2) |
665 		      SCALER_DISPCTRL_SCLEIRQ);
666 	dispctrl |= VC4_SET_FIELD(2, SCALER_DISPCTRL_DSP3_MUX);
667 
668 	HVS_WRITE(SCALER_DISPCTRL, dispctrl);
669 
670 	ret = devm_request_irq(dev, platform_get_irq(pdev, 0),
671 			       vc4_hvs_irq_handler, 0, "vc4 hvs", drm);
672 	if (ret)
673 		return ret;
674 
675 	vc4_debugfs_add_regset32(drm, "hvs_regs", &hvs->regset);
676 	vc4_debugfs_add_file(drm, "hvs_underrun", vc4_hvs_debugfs_underrun,
677 			     NULL);
678 
679 	return 0;
680 }
681 
682 static void vc4_hvs_unbind(struct device *dev, struct device *master,
683 			   void *data)
684 {
685 	struct drm_device *drm = dev_get_drvdata(master);
686 	struct vc4_dev *vc4 = to_vc4_dev(drm);
687 	struct vc4_hvs *hvs = vc4->hvs;
688 
689 	if (drm_mm_node_allocated(&vc4->hvs->mitchell_netravali_filter))
690 		drm_mm_remove_node(&vc4->hvs->mitchell_netravali_filter);
691 
692 	drm_mm_takedown(&vc4->hvs->dlist_mm);
693 	drm_mm_takedown(&vc4->hvs->lbm_mm);
694 
695 	clk_disable_unprepare(hvs->core_clk);
696 
697 	vc4->hvs = NULL;
698 }
699 
700 static const struct component_ops vc4_hvs_ops = {
701 	.bind   = vc4_hvs_bind,
702 	.unbind = vc4_hvs_unbind,
703 };
704 
705 static int vc4_hvs_dev_probe(struct platform_device *pdev)
706 {
707 	return component_add(&pdev->dev, &vc4_hvs_ops);
708 }
709 
710 static int vc4_hvs_dev_remove(struct platform_device *pdev)
711 {
712 	component_del(&pdev->dev, &vc4_hvs_ops);
713 	return 0;
714 }
715 
716 static const struct of_device_id vc4_hvs_dt_match[] = {
717 	{ .compatible = "brcm,bcm2711-hvs" },
718 	{ .compatible = "brcm,bcm2835-hvs" },
719 	{}
720 };
721 
722 struct platform_driver vc4_hvs_driver = {
723 	.probe = vc4_hvs_dev_probe,
724 	.remove = vc4_hvs_dev_remove,
725 	.driver = {
726 		.name = "vc4_hvs",
727 		.of_match_table = vc4_hvs_dt_match,
728 	},
729 };
730