2 * Copyright (C) 2015 Broadcom
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
10 * DOC: VC4 HVS module.
12 * The Hardware Video Scaler (HVS) is the piece of hardware that does
13 * translation, scaling, colorspace conversion, and compositing of
14 * pixels stored in framebuffers into a FIFO of pixels going out to
15 * the Pixel Valve (CRTC). It operates at the system clock rate (the
16 * system audio clock gate, specifically), which is much higher than
17 * the pixel clock rate.
19 * There is a single global HVS, with multiple output FIFOs that can
20 * be consumed by the PVs. This file just manages the resources for
21 * the HVS, while the vc4_crtc.c code actually drives HVS setup for
25 #include <linux/component.h>
26 #include <linux/platform_device.h>
28 #include <drm/drm_atomic_helper.h>
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),
67 void vc4_hvs_dump_state(struct drm_device *dev)
69 struct vc4_dev *vc4 = to_vc4_dev(dev);
70 struct drm_printer p = drm_info_printer(&vc4->hvs->pdev->dev);
73 drm_print_regset32(&p, &vc4->hvs->regset);
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));
86 static int vc4_hvs_debugfs_underrun(struct seq_file *m, void *data)
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);
93 drm_printf(&p, "%d\n", atomic_read(&vc4->underrun));
98 /* The filter kernel is composed of dwords each containing 3 9-bit
99 * signed integers packed next to each other.
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))
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.
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)}
122 #define VC4_LINEAR_PHASE_KERNEL_DWORDS 6
123 #define VC4_KERNEL_DWORDS (VC4_LINEAR_PHASE_KERNEL_DWORDS * 2 - 1)
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
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);
132 static int vc4_hvs_upload_linear_kernel(struct vc4_hvs *hvs,
133 struct drm_mm_node *space,
137 u32 __iomem *dst_kernel;
139 ret = drm_mm_insert_node(&hvs->dlist_mm, space, VC4_KERNEL_DWORDS);
141 DRM_ERROR("Failed to allocate space for filter kernel: %d\n",
146 dst_kernel = hvs->dlist + space->start;
148 for (i = 0; i < VC4_KERNEL_DWORDS; i++) {
149 if (i < VC4_LINEAR_PHASE_KERNEL_DWORDS)
150 writel(kernel[i], &dst_kernel[i]);
152 writel(kernel[VC4_KERNEL_DWORDS - i - 1],
160 void vc4_hvs_mask_underrun(struct drm_device *dev, int channel)
162 struct vc4_dev *vc4 = to_vc4_dev(dev);
163 u32 dispctrl = HVS_READ(SCALER_DISPCTRL);
165 dispctrl &= ~SCALER_DISPCTRL_DSPEISLUR(channel);
167 HVS_WRITE(SCALER_DISPCTRL, dispctrl);
170 void vc4_hvs_unmask_underrun(struct drm_device *dev, int channel)
172 struct vc4_dev *vc4 = to_vc4_dev(dev);
173 u32 dispctrl = HVS_READ(SCALER_DISPCTRL);
175 dispctrl |= SCALER_DISPCTRL_DSPEISLUR(channel);
177 HVS_WRITE(SCALER_DISPSTAT,
178 SCALER_DISPSTAT_EUFLOW(channel));
179 HVS_WRITE(SCALER_DISPCTRL, dispctrl);
182 static void vc4_hvs_report_underrun(struct drm_device *dev)
184 struct vc4_dev *vc4 = to_vc4_dev(dev);
186 atomic_inc(&vc4->underrun);
187 DRM_DEV_ERROR(dev->dev, "HVS underrun\n");
190 static irqreturn_t vc4_hvs_irq_handler(int irq, void *data)
192 struct drm_device *dev = data;
193 struct vc4_dev *vc4 = to_vc4_dev(dev);
194 irqreturn_t irqret = IRQ_NONE;
199 status = HVS_READ(SCALER_DISPSTAT);
200 control = HVS_READ(SCALER_DISPCTRL);
202 for (channel = 0; channel < SCALER_CHANNELS_COUNT; channel++) {
203 /* Interrupt masking is not always honored, so check it here. */
204 if (status & SCALER_DISPSTAT_EUFLOW(channel) &&
205 control & SCALER_DISPCTRL_DSPEISLUR(channel)) {
206 vc4_hvs_mask_underrun(dev, channel);
207 vc4_hvs_report_underrun(dev);
209 irqret = IRQ_HANDLED;
213 /* Clear every per-channel interrupt flag. */
214 HVS_WRITE(SCALER_DISPSTAT, SCALER_DISPSTAT_IRQMASK(0) |
215 SCALER_DISPSTAT_IRQMASK(1) |
216 SCALER_DISPSTAT_IRQMASK(2));
221 static int vc4_hvs_bind(struct device *dev, struct device *master, void *data)
223 struct platform_device *pdev = to_platform_device(dev);
224 struct drm_device *drm = dev_get_drvdata(master);
225 struct vc4_dev *vc4 = drm->dev_private;
226 struct vc4_hvs *hvs = NULL;
230 hvs = devm_kzalloc(&pdev->dev, sizeof(*hvs), GFP_KERNEL);
236 hvs->regs = vc4_ioremap_regs(pdev, 0);
237 if (IS_ERR(hvs->regs))
238 return PTR_ERR(hvs->regs);
240 hvs->regset.base = hvs->regs;
241 hvs->regset.regs = hvs_regs;
242 hvs->regset.nregs = ARRAY_SIZE(hvs_regs);
244 hvs->dlist = hvs->regs + SCALER_DLIST_START;
246 spin_lock_init(&hvs->mm_lock);
248 /* Set up the HVS display list memory manager. We never
249 * overwrite the setup from the bootloader (just 128b out of
250 * our 16K), since we don't want to scramble the screen when
251 * transitioning from the firmware's boot setup to runtime.
253 drm_mm_init(&hvs->dlist_mm,
254 HVS_BOOTLOADER_DLIST_END,
255 (SCALER_DLIST_SIZE >> 2) - HVS_BOOTLOADER_DLIST_END);
257 /* Set up the HVS LBM memory manager. We could have some more
258 * complicated data structure that allowed reuse of LBM areas
259 * between planes when they don't overlap on the screen, but
260 * for now we just allocate globally.
262 drm_mm_init(&hvs->lbm_mm, 0, 96 * 1024);
264 /* Upload filter kernels. We only have the one for now, so we
265 * keep it around for the lifetime of the driver.
267 ret = vc4_hvs_upload_linear_kernel(hvs,
268 &hvs->mitchell_netravali_filter,
269 mitchell_netravali_1_3_1_3_kernel);
275 dispctrl = HVS_READ(SCALER_DISPCTRL);
277 dispctrl |= SCALER_DISPCTRL_ENABLE;
278 dispctrl |= SCALER_DISPCTRL_DISPEIRQ(0) |
279 SCALER_DISPCTRL_DISPEIRQ(1) |
280 SCALER_DISPCTRL_DISPEIRQ(2);
282 /* Set DSP3 (PV1) to use HVS channel 2, which would otherwise
285 dispctrl &= ~SCALER_DISPCTRL_DSP3_MUX_MASK;
286 dispctrl &= ~(SCALER_DISPCTRL_DMAEIRQ |
287 SCALER_DISPCTRL_SLVWREIRQ |
288 SCALER_DISPCTRL_SLVRDEIRQ |
289 SCALER_DISPCTRL_DSPEIEOF(0) |
290 SCALER_DISPCTRL_DSPEIEOF(1) |
291 SCALER_DISPCTRL_DSPEIEOF(2) |
292 SCALER_DISPCTRL_DSPEIEOLN(0) |
293 SCALER_DISPCTRL_DSPEIEOLN(1) |
294 SCALER_DISPCTRL_DSPEIEOLN(2) |
295 SCALER_DISPCTRL_DSPEISLUR(0) |
296 SCALER_DISPCTRL_DSPEISLUR(1) |
297 SCALER_DISPCTRL_DSPEISLUR(2) |
298 SCALER_DISPCTRL_SCLEIRQ);
299 dispctrl |= VC4_SET_FIELD(2, SCALER_DISPCTRL_DSP3_MUX);
301 HVS_WRITE(SCALER_DISPCTRL, dispctrl);
303 ret = devm_request_irq(dev, platform_get_irq(pdev, 0),
304 vc4_hvs_irq_handler, 0, "vc4 hvs", drm);
308 vc4_debugfs_add_regset32(drm, "hvs_regs", &hvs->regset);
309 vc4_debugfs_add_file(drm, "hvs_underrun", vc4_hvs_debugfs_underrun,
315 static void vc4_hvs_unbind(struct device *dev, struct device *master,
318 struct drm_device *drm = dev_get_drvdata(master);
319 struct vc4_dev *vc4 = drm->dev_private;
321 if (vc4->hvs->mitchell_netravali_filter.allocated)
322 drm_mm_remove_node(&vc4->hvs->mitchell_netravali_filter);
324 drm_mm_takedown(&vc4->hvs->dlist_mm);
325 drm_mm_takedown(&vc4->hvs->lbm_mm);
330 static const struct component_ops vc4_hvs_ops = {
331 .bind = vc4_hvs_bind,
332 .unbind = vc4_hvs_unbind,
335 static int vc4_hvs_dev_probe(struct platform_device *pdev)
337 return component_add(&pdev->dev, &vc4_hvs_ops);
340 static int vc4_hvs_dev_remove(struct platform_device *pdev)
342 component_del(&pdev->dev, &vc4_hvs_ops);
346 static const struct of_device_id vc4_hvs_dt_match[] = {
347 { .compatible = "brcm,bcm2835-hvs" },
351 struct platform_driver vc4_hvs_driver = {
352 .probe = vc4_hvs_dev_probe,
353 .remove = vc4_hvs_dev_remove,
356 .of_match_table = vc4_hvs_dt_match,