1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2015 Broadcom
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.
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
22 #include <linux/component.h>
23 #include <linux/platform_device.h>
25 #include <drm/drm_atomic_helper.h>
30 static const struct debugfs_reg32 hvs_regs
[] = {
31 VC4_REG32(SCALER_DISPCTRL
),
32 VC4_REG32(SCALER_DISPSTAT
),
33 VC4_REG32(SCALER_DISPID
),
34 VC4_REG32(SCALER_DISPECTRL
),
35 VC4_REG32(SCALER_DISPPROF
),
36 VC4_REG32(SCALER_DISPDITHER
),
37 VC4_REG32(SCALER_DISPEOLN
),
38 VC4_REG32(SCALER_DISPLIST0
),
39 VC4_REG32(SCALER_DISPLIST1
),
40 VC4_REG32(SCALER_DISPLIST2
),
41 VC4_REG32(SCALER_DISPLSTAT
),
42 VC4_REG32(SCALER_DISPLACT0
),
43 VC4_REG32(SCALER_DISPLACT1
),
44 VC4_REG32(SCALER_DISPLACT2
),
45 VC4_REG32(SCALER_DISPCTRL0
),
46 VC4_REG32(SCALER_DISPBKGND0
),
47 VC4_REG32(SCALER_DISPSTAT0
),
48 VC4_REG32(SCALER_DISPBASE0
),
49 VC4_REG32(SCALER_DISPCTRL1
),
50 VC4_REG32(SCALER_DISPBKGND1
),
51 VC4_REG32(SCALER_DISPSTAT1
),
52 VC4_REG32(SCALER_DISPBASE1
),
53 VC4_REG32(SCALER_DISPCTRL2
),
54 VC4_REG32(SCALER_DISPBKGND2
),
55 VC4_REG32(SCALER_DISPSTAT2
),
56 VC4_REG32(SCALER_DISPBASE2
),
57 VC4_REG32(SCALER_DISPALPHA2
),
58 VC4_REG32(SCALER_OLEDOFFS
),
59 VC4_REG32(SCALER_OLEDCOEF0
),
60 VC4_REG32(SCALER_OLEDCOEF1
),
61 VC4_REG32(SCALER_OLEDCOEF2
),
64 void vc4_hvs_dump_state(struct drm_device
*dev
)
66 struct vc4_dev
*vc4
= to_vc4_dev(dev
);
67 struct drm_printer p
= drm_info_printer(&vc4
->hvs
->pdev
->dev
);
70 drm_print_regset32(&p
, &vc4
->hvs
->regset
);
72 DRM_INFO("HVS ctx:\n");
73 for (i
= 0; i
< 64; i
+= 4) {
74 DRM_INFO("0x%08x (%s): 0x%08x 0x%08x 0x%08x 0x%08x\n",
75 i
* 4, i
< HVS_BOOTLOADER_DLIST_END
? "B" : "D",
76 readl((u32 __iomem
*)vc4
->hvs
->dlist
+ i
+ 0),
77 readl((u32 __iomem
*)vc4
->hvs
->dlist
+ i
+ 1),
78 readl((u32 __iomem
*)vc4
->hvs
->dlist
+ i
+ 2),
79 readl((u32 __iomem
*)vc4
->hvs
->dlist
+ i
+ 3));
83 static int vc4_hvs_debugfs_underrun(struct seq_file
*m
, void *data
)
85 struct drm_info_node
*node
= m
->private;
86 struct drm_device
*dev
= node
->minor
->dev
;
87 struct vc4_dev
*vc4
= to_vc4_dev(dev
);
88 struct drm_printer p
= drm_seq_file_printer(m
);
90 drm_printf(&p
, "%d\n", atomic_read(&vc4
->underrun
));
95 /* The filter kernel is composed of dwords each containing 3 9-bit
96 * signed integers packed next to each other.
98 #define VC4_INT_TO_COEFF(coeff) (coeff & 0x1ff)
99 #define VC4_PPF_FILTER_WORD(c0, c1, c2) \
100 ((((c0) & 0x1ff) << 0) | \
101 (((c1) & 0x1ff) << 9) | \
102 (((c2) & 0x1ff) << 18))
104 /* The whole filter kernel is arranged as the coefficients 0-16 going
105 * up, then a pad, then 17-31 going down and reversed within the
106 * dwords. This means that a linear phase kernel (where it's
107 * symmetrical at the boundary between 15 and 16) has the last 5
108 * dwords matching the first 5, but reversed.
110 #define VC4_LINEAR_PHASE_KERNEL(c0, c1, c2, c3, c4, c5, c6, c7, c8, \
111 c9, c10, c11, c12, c13, c14, c15) \
112 {VC4_PPF_FILTER_WORD(c0, c1, c2), \
113 VC4_PPF_FILTER_WORD(c3, c4, c5), \
114 VC4_PPF_FILTER_WORD(c6, c7, c8), \
115 VC4_PPF_FILTER_WORD(c9, c10, c11), \
116 VC4_PPF_FILTER_WORD(c12, c13, c14), \
117 VC4_PPF_FILTER_WORD(c15, c15, 0)}
119 #define VC4_LINEAR_PHASE_KERNEL_DWORDS 6
120 #define VC4_KERNEL_DWORDS (VC4_LINEAR_PHASE_KERNEL_DWORDS * 2 - 1)
122 /* Recommended B=1/3, C=1/3 filter choice from Mitchell/Netravali.
123 * http://www.cs.utexas.edu/~fussell/courses/cs384g/lectures/mitchell/Mitchell.pdf
125 static const u32 mitchell_netravali_1_3_1_3_kernel
[] =
126 VC4_LINEAR_PHASE_KERNEL(0, -2, -6, -8, -10, -8, -3, 2, 18,
127 50, 82, 119, 155, 187, 213, 227);
129 static int vc4_hvs_upload_linear_kernel(struct vc4_hvs
*hvs
,
130 struct drm_mm_node
*space
,
134 u32 __iomem
*dst_kernel
;
136 ret
= drm_mm_insert_node(&hvs
->dlist_mm
, space
, VC4_KERNEL_DWORDS
);
138 DRM_ERROR("Failed to allocate space for filter kernel: %d\n",
143 dst_kernel
= hvs
->dlist
+ space
->start
;
145 for (i
= 0; i
< VC4_KERNEL_DWORDS
; i
++) {
146 if (i
< VC4_LINEAR_PHASE_KERNEL_DWORDS
)
147 writel(kernel
[i
], &dst_kernel
[i
]);
149 writel(kernel
[VC4_KERNEL_DWORDS
- i
- 1],
157 void vc4_hvs_mask_underrun(struct drm_device
*dev
, int channel
)
159 struct vc4_dev
*vc4
= to_vc4_dev(dev
);
160 u32 dispctrl
= HVS_READ(SCALER_DISPCTRL
);
162 dispctrl
&= ~SCALER_DISPCTRL_DSPEISLUR(channel
);
164 HVS_WRITE(SCALER_DISPCTRL
, dispctrl
);
167 void vc4_hvs_unmask_underrun(struct drm_device
*dev
, int channel
)
169 struct vc4_dev
*vc4
= to_vc4_dev(dev
);
170 u32 dispctrl
= HVS_READ(SCALER_DISPCTRL
);
172 dispctrl
|= SCALER_DISPCTRL_DSPEISLUR(channel
);
174 HVS_WRITE(SCALER_DISPSTAT
,
175 SCALER_DISPSTAT_EUFLOW(channel
));
176 HVS_WRITE(SCALER_DISPCTRL
, dispctrl
);
179 static void vc4_hvs_report_underrun(struct drm_device
*dev
)
181 struct vc4_dev
*vc4
= to_vc4_dev(dev
);
183 atomic_inc(&vc4
->underrun
);
184 DRM_DEV_ERROR(dev
->dev
, "HVS underrun\n");
187 static irqreturn_t
vc4_hvs_irq_handler(int irq
, void *data
)
189 struct drm_device
*dev
= data
;
190 struct vc4_dev
*vc4
= to_vc4_dev(dev
);
191 irqreturn_t irqret
= IRQ_NONE
;
196 status
= HVS_READ(SCALER_DISPSTAT
);
197 control
= HVS_READ(SCALER_DISPCTRL
);
199 for (channel
= 0; channel
< SCALER_CHANNELS_COUNT
; channel
++) {
200 /* Interrupt masking is not always honored, so check it here. */
201 if (status
& SCALER_DISPSTAT_EUFLOW(channel
) &&
202 control
& SCALER_DISPCTRL_DSPEISLUR(channel
)) {
203 vc4_hvs_mask_underrun(dev
, channel
);
204 vc4_hvs_report_underrun(dev
);
206 irqret
= IRQ_HANDLED
;
210 /* Clear every per-channel interrupt flag. */
211 HVS_WRITE(SCALER_DISPSTAT
, SCALER_DISPSTAT_IRQMASK(0) |
212 SCALER_DISPSTAT_IRQMASK(1) |
213 SCALER_DISPSTAT_IRQMASK(2));
218 static int vc4_hvs_bind(struct device
*dev
, struct device
*master
, void *data
)
220 struct platform_device
*pdev
= to_platform_device(dev
);
221 struct drm_device
*drm
= dev_get_drvdata(master
);
222 struct vc4_dev
*vc4
= drm
->dev_private
;
223 struct vc4_hvs
*hvs
= NULL
;
227 hvs
= devm_kzalloc(&pdev
->dev
, sizeof(*hvs
), GFP_KERNEL
);
233 hvs
->regs
= vc4_ioremap_regs(pdev
, 0);
234 if (IS_ERR(hvs
->regs
))
235 return PTR_ERR(hvs
->regs
);
237 hvs
->regset
.base
= hvs
->regs
;
238 hvs
->regset
.regs
= hvs_regs
;
239 hvs
->regset
.nregs
= ARRAY_SIZE(hvs_regs
);
241 hvs
->dlist
= hvs
->regs
+ SCALER_DLIST_START
;
243 spin_lock_init(&hvs
->mm_lock
);
245 /* Set up the HVS display list memory manager. We never
246 * overwrite the setup from the bootloader (just 128b out of
247 * our 16K), since we don't want to scramble the screen when
248 * transitioning from the firmware's boot setup to runtime.
250 drm_mm_init(&hvs
->dlist_mm
,
251 HVS_BOOTLOADER_DLIST_END
,
252 (SCALER_DLIST_SIZE
>> 2) - HVS_BOOTLOADER_DLIST_END
);
254 /* Set up the HVS LBM memory manager. We could have some more
255 * complicated data structure that allowed reuse of LBM areas
256 * between planes when they don't overlap on the screen, but
257 * for now we just allocate globally.
259 drm_mm_init(&hvs
->lbm_mm
, 0, 96 * 1024);
261 /* Upload filter kernels. We only have the one for now, so we
262 * keep it around for the lifetime of the driver.
264 ret
= vc4_hvs_upload_linear_kernel(hvs
,
265 &hvs
->mitchell_netravali_filter
,
266 mitchell_netravali_1_3_1_3_kernel
);
272 dispctrl
= HVS_READ(SCALER_DISPCTRL
);
274 dispctrl
|= SCALER_DISPCTRL_ENABLE
;
275 dispctrl
|= SCALER_DISPCTRL_DISPEIRQ(0) |
276 SCALER_DISPCTRL_DISPEIRQ(1) |
277 SCALER_DISPCTRL_DISPEIRQ(2);
279 /* Set DSP3 (PV1) to use HVS channel 2, which would otherwise
282 dispctrl
&= ~SCALER_DISPCTRL_DSP3_MUX_MASK
;
283 dispctrl
&= ~(SCALER_DISPCTRL_DMAEIRQ
|
284 SCALER_DISPCTRL_SLVWREIRQ
|
285 SCALER_DISPCTRL_SLVRDEIRQ
|
286 SCALER_DISPCTRL_DSPEIEOF(0) |
287 SCALER_DISPCTRL_DSPEIEOF(1) |
288 SCALER_DISPCTRL_DSPEIEOF(2) |
289 SCALER_DISPCTRL_DSPEIEOLN(0) |
290 SCALER_DISPCTRL_DSPEIEOLN(1) |
291 SCALER_DISPCTRL_DSPEIEOLN(2) |
292 SCALER_DISPCTRL_DSPEISLUR(0) |
293 SCALER_DISPCTRL_DSPEISLUR(1) |
294 SCALER_DISPCTRL_DSPEISLUR(2) |
295 SCALER_DISPCTRL_SCLEIRQ
);
296 dispctrl
|= VC4_SET_FIELD(2, SCALER_DISPCTRL_DSP3_MUX
);
298 HVS_WRITE(SCALER_DISPCTRL
, dispctrl
);
300 ret
= devm_request_irq(dev
, platform_get_irq(pdev
, 0),
301 vc4_hvs_irq_handler
, 0, "vc4 hvs", drm
);
305 vc4_debugfs_add_regset32(drm
, "hvs_regs", &hvs
->regset
);
306 vc4_debugfs_add_file(drm
, "hvs_underrun", vc4_hvs_debugfs_underrun
,
312 static void vc4_hvs_unbind(struct device
*dev
, struct device
*master
,
315 struct drm_device
*drm
= dev_get_drvdata(master
);
316 struct vc4_dev
*vc4
= drm
->dev_private
;
318 if (drm_mm_node_allocated(&vc4
->hvs
->mitchell_netravali_filter
))
319 drm_mm_remove_node(&vc4
->hvs
->mitchell_netravali_filter
);
321 drm_mm_takedown(&vc4
->hvs
->dlist_mm
);
322 drm_mm_takedown(&vc4
->hvs
->lbm_mm
);
327 static const struct component_ops vc4_hvs_ops
= {
328 .bind
= vc4_hvs_bind
,
329 .unbind
= vc4_hvs_unbind
,
332 static int vc4_hvs_dev_probe(struct platform_device
*pdev
)
334 return component_add(&pdev
->dev
, &vc4_hvs_ops
);
337 static int vc4_hvs_dev_remove(struct platform_device
*pdev
)
339 component_del(&pdev
->dev
, &vc4_hvs_ops
);
343 static const struct of_device_id vc4_hvs_dt_match
[] = {
344 { .compatible
= "brcm,bcm2835-hvs" },
348 struct platform_driver vc4_hvs_driver
= {
349 .probe
= vc4_hvs_dev_probe
,
350 .remove
= vc4_hvs_dev_remove
,
353 .of_match_table
= vc4_hvs_dt_match
,