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 <drm/drm_atomic_helper.h>
23 #include <linux/component.h>
27 static const struct debugfs_reg32 hvs_regs[] = {
28 VC4_REG32(SCALER_DISPCTRL),
29 VC4_REG32(SCALER_DISPSTAT),
30 VC4_REG32(SCALER_DISPID),
31 VC4_REG32(SCALER_DISPECTRL),
32 VC4_REG32(SCALER_DISPPROF),
33 VC4_REG32(SCALER_DISPDITHER),
34 VC4_REG32(SCALER_DISPEOLN),
35 VC4_REG32(SCALER_DISPLIST0),
36 VC4_REG32(SCALER_DISPLIST1),
37 VC4_REG32(SCALER_DISPLIST2),
38 VC4_REG32(SCALER_DISPLSTAT),
39 VC4_REG32(SCALER_DISPLACT0),
40 VC4_REG32(SCALER_DISPLACT1),
41 VC4_REG32(SCALER_DISPLACT2),
42 VC4_REG32(SCALER_DISPCTRL0),
43 VC4_REG32(SCALER_DISPBKGND0),
44 VC4_REG32(SCALER_DISPSTAT0),
45 VC4_REG32(SCALER_DISPBASE0),
46 VC4_REG32(SCALER_DISPCTRL1),
47 VC4_REG32(SCALER_DISPBKGND1),
48 VC4_REG32(SCALER_DISPSTAT1),
49 VC4_REG32(SCALER_DISPBASE1),
50 VC4_REG32(SCALER_DISPCTRL2),
51 VC4_REG32(SCALER_DISPBKGND2),
52 VC4_REG32(SCALER_DISPSTAT2),
53 VC4_REG32(SCALER_DISPBASE2),
54 VC4_REG32(SCALER_DISPALPHA2),
55 VC4_REG32(SCALER_OLEDOFFS),
56 VC4_REG32(SCALER_OLEDCOEF0),
57 VC4_REG32(SCALER_OLEDCOEF1),
58 VC4_REG32(SCALER_OLEDCOEF2),
61 void vc4_hvs_dump_state(struct drm_device *dev)
63 struct vc4_dev *vc4 = to_vc4_dev(dev);
64 struct drm_printer p = drm_info_printer(&vc4->hvs->pdev->dev);
67 drm_print_regset32(&p, &vc4->hvs->regset);
69 DRM_INFO("HVS ctx:\n");
70 for (i = 0; i < 64; i += 4) {
71 DRM_INFO("0x%08x (%s): 0x%08x 0x%08x 0x%08x 0x%08x\n",
72 i * 4, i < HVS_BOOTLOADER_DLIST_END ? "B" : "D",
73 readl((u32 __iomem *)vc4->hvs->dlist + i + 0),
74 readl((u32 __iomem *)vc4->hvs->dlist + i + 1),
75 readl((u32 __iomem *)vc4->hvs->dlist + i + 2),
76 readl((u32 __iomem *)vc4->hvs->dlist + i + 3));
80 static int vc4_hvs_debugfs_underrun(struct seq_file *m, void *data)
82 struct drm_info_node *node = m->private;
83 struct drm_device *dev = node->minor->dev;
84 struct vc4_dev *vc4 = to_vc4_dev(dev);
85 struct drm_printer p = drm_seq_file_printer(m);
87 drm_printf(&p, "%d\n", atomic_read(&vc4->underrun));
92 /* The filter kernel is composed of dwords each containing 3 9-bit
93 * signed integers packed next to each other.
95 #define VC4_INT_TO_COEFF(coeff) (coeff & 0x1ff)
96 #define VC4_PPF_FILTER_WORD(c0, c1, c2) \
97 ((((c0) & 0x1ff) << 0) | \
98 (((c1) & 0x1ff) << 9) | \
99 (((c2) & 0x1ff) << 18))
101 /* The whole filter kernel is arranged as the coefficients 0-16 going
102 * up, then a pad, then 17-31 going down and reversed within the
103 * dwords. This means that a linear phase kernel (where it's
104 * symmetrical at the boundary between 15 and 16) has the last 5
105 * dwords matching the first 5, but reversed.
107 #define VC4_LINEAR_PHASE_KERNEL(c0, c1, c2, c3, c4, c5, c6, c7, c8, \
108 c9, c10, c11, c12, c13, c14, c15) \
109 {VC4_PPF_FILTER_WORD(c0, c1, c2), \
110 VC4_PPF_FILTER_WORD(c3, c4, c5), \
111 VC4_PPF_FILTER_WORD(c6, c7, c8), \
112 VC4_PPF_FILTER_WORD(c9, c10, c11), \
113 VC4_PPF_FILTER_WORD(c12, c13, c14), \
114 VC4_PPF_FILTER_WORD(c15, c15, 0)}
116 #define VC4_LINEAR_PHASE_KERNEL_DWORDS 6
117 #define VC4_KERNEL_DWORDS (VC4_LINEAR_PHASE_KERNEL_DWORDS * 2 - 1)
119 /* Recommended B=1/3, C=1/3 filter choice from Mitchell/Netravali.
120 * http://www.cs.utexas.edu/~fussell/courses/cs384g/lectures/mitchell/Mitchell.pdf
122 static const u32 mitchell_netravali_1_3_1_3_kernel[] =
123 VC4_LINEAR_PHASE_KERNEL(0, -2, -6, -8, -10, -8, -3, 2, 18,
124 50, 82, 119, 155, 187, 213, 227);
126 static int vc4_hvs_upload_linear_kernel(struct vc4_hvs *hvs,
127 struct drm_mm_node *space,
131 u32 __iomem *dst_kernel;
133 ret = drm_mm_insert_node(&hvs->dlist_mm, space, VC4_KERNEL_DWORDS);
135 DRM_ERROR("Failed to allocate space for filter kernel: %d\n",
140 dst_kernel = hvs->dlist + space->start;
142 for (i = 0; i < VC4_KERNEL_DWORDS; i++) {
143 if (i < VC4_LINEAR_PHASE_KERNEL_DWORDS)
144 writel(kernel[i], &dst_kernel[i]);
146 writel(kernel[VC4_KERNEL_DWORDS - i - 1],
154 void vc4_hvs_mask_underrun(struct drm_device *dev, int channel)
156 struct vc4_dev *vc4 = to_vc4_dev(dev);
157 u32 dispctrl = HVS_READ(SCALER_DISPCTRL);
159 dispctrl &= ~SCALER_DISPCTRL_DSPEISLUR(channel);
161 HVS_WRITE(SCALER_DISPCTRL, dispctrl);
164 void vc4_hvs_unmask_underrun(struct drm_device *dev, int channel)
166 struct vc4_dev *vc4 = to_vc4_dev(dev);
167 u32 dispctrl = HVS_READ(SCALER_DISPCTRL);
169 dispctrl |= SCALER_DISPCTRL_DSPEISLUR(channel);
171 HVS_WRITE(SCALER_DISPSTAT,
172 SCALER_DISPSTAT_EUFLOW(channel));
173 HVS_WRITE(SCALER_DISPCTRL, dispctrl);
176 static void vc4_hvs_report_underrun(struct drm_device *dev)
178 struct vc4_dev *vc4 = to_vc4_dev(dev);
180 atomic_inc(&vc4->underrun);
181 DRM_DEV_ERROR(dev->dev, "HVS underrun\n");
184 static irqreturn_t vc4_hvs_irq_handler(int irq, void *data)
186 struct drm_device *dev = data;
187 struct vc4_dev *vc4 = to_vc4_dev(dev);
188 irqreturn_t irqret = IRQ_NONE;
193 status = HVS_READ(SCALER_DISPSTAT);
194 control = HVS_READ(SCALER_DISPCTRL);
196 for (channel = 0; channel < SCALER_CHANNELS_COUNT; channel++) {
197 /* Interrupt masking is not always honored, so check it here. */
198 if (status & SCALER_DISPSTAT_EUFLOW(channel) &&
199 control & SCALER_DISPCTRL_DSPEISLUR(channel)) {
200 vc4_hvs_mask_underrun(dev, channel);
201 vc4_hvs_report_underrun(dev);
203 irqret = IRQ_HANDLED;
207 /* Clear every per-channel interrupt flag. */
208 HVS_WRITE(SCALER_DISPSTAT, SCALER_DISPSTAT_IRQMASK(0) |
209 SCALER_DISPSTAT_IRQMASK(1) |
210 SCALER_DISPSTAT_IRQMASK(2));
215 static int vc4_hvs_bind(struct device *dev, struct device *master, void *data)
217 struct platform_device *pdev = to_platform_device(dev);
218 struct drm_device *drm = dev_get_drvdata(master);
219 struct vc4_dev *vc4 = drm->dev_private;
220 struct vc4_hvs *hvs = NULL;
224 hvs = devm_kzalloc(&pdev->dev, sizeof(*hvs), GFP_KERNEL);
230 hvs->regs = vc4_ioremap_regs(pdev, 0);
231 if (IS_ERR(hvs->regs))
232 return PTR_ERR(hvs->regs);
234 hvs->regset.base = hvs->regs;
235 hvs->regset.regs = hvs_regs;
236 hvs->regset.nregs = ARRAY_SIZE(hvs_regs);
238 hvs->dlist = hvs->regs + SCALER_DLIST_START;
240 spin_lock_init(&hvs->mm_lock);
242 /* Set up the HVS display list memory manager. We never
243 * overwrite the setup from the bootloader (just 128b out of
244 * our 16K), since we don't want to scramble the screen when
245 * transitioning from the firmware's boot setup to runtime.
247 drm_mm_init(&hvs->dlist_mm,
248 HVS_BOOTLOADER_DLIST_END,
249 (SCALER_DLIST_SIZE >> 2) - HVS_BOOTLOADER_DLIST_END);
251 /* Set up the HVS LBM memory manager. We could have some more
252 * complicated data structure that allowed reuse of LBM areas
253 * between planes when they don't overlap on the screen, but
254 * for now we just allocate globally.
256 drm_mm_init(&hvs->lbm_mm, 0, 96 * 1024);
258 /* Upload filter kernels. We only have the one for now, so we
259 * keep it around for the lifetime of the driver.
261 ret = vc4_hvs_upload_linear_kernel(hvs,
262 &hvs->mitchell_netravali_filter,
263 mitchell_netravali_1_3_1_3_kernel);
269 dispctrl = HVS_READ(SCALER_DISPCTRL);
271 dispctrl |= SCALER_DISPCTRL_ENABLE;
272 dispctrl |= SCALER_DISPCTRL_DISPEIRQ(0) |
273 SCALER_DISPCTRL_DISPEIRQ(1) |
274 SCALER_DISPCTRL_DISPEIRQ(2);
276 /* Set DSP3 (PV1) to use HVS channel 2, which would otherwise
279 dispctrl &= ~SCALER_DISPCTRL_DSP3_MUX_MASK;
280 dispctrl &= ~(SCALER_DISPCTRL_DMAEIRQ |
281 SCALER_DISPCTRL_SLVWREIRQ |
282 SCALER_DISPCTRL_SLVRDEIRQ |
283 SCALER_DISPCTRL_DSPEIEOF(0) |
284 SCALER_DISPCTRL_DSPEIEOF(1) |
285 SCALER_DISPCTRL_DSPEIEOF(2) |
286 SCALER_DISPCTRL_DSPEIEOLN(0) |
287 SCALER_DISPCTRL_DSPEIEOLN(1) |
288 SCALER_DISPCTRL_DSPEIEOLN(2) |
289 SCALER_DISPCTRL_DSPEISLUR(0) |
290 SCALER_DISPCTRL_DSPEISLUR(1) |
291 SCALER_DISPCTRL_DSPEISLUR(2) |
292 SCALER_DISPCTRL_SCLEIRQ);
293 dispctrl |= VC4_SET_FIELD(2, SCALER_DISPCTRL_DSP3_MUX);
295 HVS_WRITE(SCALER_DISPCTRL, dispctrl);
297 ret = devm_request_irq(dev, platform_get_irq(pdev, 0),
298 vc4_hvs_irq_handler, 0, "vc4 hvs", drm);
302 vc4_debugfs_add_regset32(drm, "hvs_regs", &hvs->regset);
303 vc4_debugfs_add_file(drm, "hvs_underrun", vc4_hvs_debugfs_underrun,
309 static void vc4_hvs_unbind(struct device *dev, struct device *master,
312 struct drm_device *drm = dev_get_drvdata(master);
313 struct vc4_dev *vc4 = drm->dev_private;
315 if (vc4->hvs->mitchell_netravali_filter.allocated)
316 drm_mm_remove_node(&vc4->hvs->mitchell_netravali_filter);
318 drm_mm_takedown(&vc4->hvs->dlist_mm);
319 drm_mm_takedown(&vc4->hvs->lbm_mm);
324 static const struct component_ops vc4_hvs_ops = {
325 .bind = vc4_hvs_bind,
326 .unbind = vc4_hvs_unbind,
329 static int vc4_hvs_dev_probe(struct platform_device *pdev)
331 return component_add(&pdev->dev, &vc4_hvs_ops);
334 static int vc4_hvs_dev_remove(struct platform_device *pdev)
336 component_del(&pdev->dev, &vc4_hvs_ops);
340 static const struct of_device_id vc4_hvs_dt_match[] = {
341 { .compatible = "brcm,bcm2835-hvs" },
345 struct platform_driver vc4_hvs_driver = {
346 .probe = vc4_hvs_dev_probe,
347 .remove = vc4_hvs_dev_remove,
350 .of_match_table = vc4_hvs_dt_match,