On Fri, Nov 19, 2021 at 05:57:01PM -0600, Samuel Holland wrote:

Hi Maxime,

On 11/19/21 2:28 AM, Maxime Ripard wrote:

quoted

Hi,

On Thu, Nov 18, 2021 at 11:40:43PM -0600, Samuel Holland wrote:

quoted

+static const struct sun50i_r329_ledc_timing sun50i_r329_ledc_default_timing = {
+	.t0h_ns = 336,
+	.t0l_ns = 840,
+	.t1h_ns = 882,
+	.t1l_ns = 294,
+	.treset_ns = 300000,
+};

This should be mentioned in the binding as well (using the default keyword)

Ok, I'll do this for v4.

quoted

quoted

+static int sun50i_r329_ledc_parse_timing(const struct device_node *np,
+					 struct sun50i_r329_ledc *priv)
+{
+	struct sun50i_r329_ledc_timing *timing = &priv->timing;
+
+	*timing = sun50i_r329_ledc_default_timing;
+
+	of_property_read_u32(np, "allwinner,t0h-ns", &timing->t0h_ns);
+	of_property_read_u32(np, "allwinner,t0l-ns", &timing->t0l_ns);
+	of_property_read_u32(np, "allwinner,t1h-ns", &timing->t1h_ns);
+	of_property_read_u32(np, "allwinner,t1l-ns", &timing->t1l_ns);
+	of_property_read_u32(np, "allwinner,treset-ns", &timing->treset_ns);
+
+	return 0;
+}
+
+static void sun50i_r329_ledc_set_timing(struct sun50i_r329_ledc *priv)
+{
+	const struct sun50i_r329_ledc_timing *timing = &priv->timing;
+	unsigned long mod_freq = clk_get_rate(priv->mod_clk);
+	u32 cycle_ns = NSEC_PER_SEC / mod_freq;
+	u32 val;
+
+	val = (timing->t1h_ns / cycle_ns) << 21 |
+	      (timing->t1l_ns / cycle_ns) << 16 |
+	      (timing->t0h_ns / cycle_ns) <<  6 |
+	      (timing->t0l_ns / cycle_ns);
+	writel(val, priv->base + LEDC_T01_TIMING_CTRL_REG);
+
+	val = (timing->treset_ns / cycle_ns) << 16 |
+	      (priv->num_leds - 1);
+	writel(val, priv->base + LEDC_RESET_TIMING_CTRL_REG);
+}
+
+static int sun50i_r329_ledc_resume(struct device *dev)
+{
+	struct sun50i_r329_ledc *priv = dev_get_drvdata(dev);
+	u32 val;
+	int ret;
+
+	ret = reset_control_deassert(priv->reset);
+	if (ret)
+		return ret;
+
+	ret = clk_prepare_enable(priv->bus_clk);
+	if (ret)
+		goto err_assert_reset;
+
+	ret = clk_prepare_enable(priv->mod_clk);
+	if (ret)
+		goto err_disable_bus_clk;
+
+	sun50i_r329_ledc_set_format(priv);
+	sun50i_r329_ledc_set_timing(priv);
+
+	/* The trigger level must be at least the burst length. */
+	val = readl(priv->base + LEDC_DMA_CTRL_REG);
+	val &= ~LEDC_DMA_CTRL_REG_FIFO_TRIG_LEVEL;
+	val |= LEDC_FIFO_DEPTH / 2;
+	writel(val, priv->base + LEDC_DMA_CTRL_REG);
+
+	val = LEDC_INT_CTRL_REG_GLOBAL_INT_EN |
+	      LEDC_INT_CTRL_REG_TRANS_FINISH_INT_EN;
+	writel(val, priv->base + LEDC_INT_CTRL_REG);
+
+	return 0;
+
+err_disable_bus_clk:
+	clk_disable_unprepare(priv->bus_clk);
+err_assert_reset:
+	reset_control_assert(priv->reset);
+
+	return ret;
+}
+
+static int sun50i_r329_ledc_suspend(struct device *dev)
+{
+	struct sun50i_r329_ledc *priv = dev_get_drvdata(dev);
+
+	clk_disable_unprepare(priv->mod_clk);
+	clk_disable_unprepare(priv->bus_clk);
+	reset_control_assert(priv->reset);
+
+	return 0;
+}
+
+static void sun50i_r329_ledc_dma_cleanup(void *data)
+{
+	struct sun50i_r329_ledc *priv = data;
+	struct device *dma_dev = dmaengine_get_dma_device(priv->dma_chan);
+
+	if (priv->buffer)
+		dma_free_wc(dma_dev, LEDS_TO_BYTES(priv->num_leds),
+			    priv->buffer, priv->dma_handle);
+	dma_release_channel(priv->dma_chan);
+}
+
+static int sun50i_r329_ledc_probe(struct platform_device *pdev)
+{
+	const struct device_node *np = pdev->dev.of_node;
+	struct dma_slave_config dma_cfg = {};
+	struct led_init_data init_data = {};
+	struct device *dev = &pdev->dev;
+	struct device_node *child;
+	struct sun50i_r329_ledc *priv;
+	struct resource *mem;
+	int count, irq, ret;
+
+	count = of_get_available_child_count(np);
+	if (!count)
+		return -ENODEV;
+	if (count > LEDC_MAX_LEDS) {
+		dev_err(dev, "Too many LEDs! (max is %d)\n", LEDC_MAX_LEDS);
+		return -EINVAL;
+	}
+
+	priv = devm_kzalloc(dev, struct_size(priv, leds, count), GFP_KERNEL);
+	if (!priv)
+		return -ENOMEM;
+
+	priv->dev = dev;
+	priv->num_leds = count;
+	spin_lock_init(&priv->lock);
+	dev_set_drvdata(dev, priv);
+
+	ret = sun50i_r329_ledc_parse_format(np, priv);
+	if (ret)
+		return ret;
+
+	ret = sun50i_r329_ledc_parse_timing(np, priv);
+	if (ret)
+		return ret;
+
+	priv->base = devm_platform_get_and_ioremap_resource(pdev, 0, &mem);
+	if (IS_ERR(priv->base))
+		return PTR_ERR(priv->base);
+
+	priv->bus_clk = devm_clk_get(dev, "bus");
+	if (IS_ERR(priv->bus_clk))
+		return PTR_ERR(priv->bus_clk);
+
+	priv->mod_clk = devm_clk_get(dev, "mod");
+	if (IS_ERR(priv->mod_clk))
+		return PTR_ERR(priv->mod_clk);
+
+	priv->reset = devm_reset_control_get_exclusive(dev, NULL);
+	if (IS_ERR(priv->reset))
+		return PTR_ERR(priv->reset);
+
+	priv->dma_chan = dma_request_chan(dev, "tx");
+	if (IS_ERR(priv->dma_chan))
+		return PTR_ERR(priv->dma_chan);
+
+	ret = devm_add_action_or_reset(dev, sun50i_r329_ledc_dma_cleanup, priv);
+	if (ret)
+		return ret;
+
+	dma_cfg.dst_addr	= mem->start + LEDC_DATA_REG;
+	dma_cfg.dst_addr_width	= DMA_SLAVE_BUSWIDTH_4_BYTES;
+	dma_cfg.dst_maxburst	= LEDC_FIFO_DEPTH / 2;
+	ret = dmaengine_slave_config(priv->dma_chan, &dma_cfg);
+	if (ret)
+		return ret;
+
+	priv->buffer = dma_alloc_wc(dmaengine_get_dma_device(priv->dma_chan),
+				    LEDS_TO_BYTES(priv->num_leds),
+				    &priv->dma_handle, GFP_KERNEL);
+	if (!priv->buffer)
+		return -ENOMEM;
+
+	irq = platform_get_irq(pdev, 0);
+	if (irq < 0)
+		return irq;
+
+	ret = devm_request_irq(dev, irq, sun50i_r329_ledc_irq,
+			       0, dev_name(dev), priv);
+	if (ret)
+		return ret;
+
+	ret = sun50i_r329_ledc_resume(dev);
+	if (ret)
+		return ret;

You seem to fill the runtime_pm hooks, but only call them directly and
never enable runtime_pm on that device, is that intentional?

Yes. I did not want to delay the initial version by adding runtime PM
(and debugging the refcounts) when the driver already works now.
However, I had runtime/system PM in mind while writing the driver.

If you think it is too confusing, I could rename the functions to
something like sun50i_r329_ledc_hw_init / sun50i_r329_ledc_hw_exit.

It's not really the functions themselves that are confusing but rather
that you set them as runtime_pm hooks.

If you plan on submitting it later on, I guess the easiest would just be
to remove the sun50i_r329_ledc_pm structure, and then you'll just have
to introduce it back with a call to pm_runtime_enable when it's ready

Maxime