Re: [PATCH 3/3] DRM: Add KMS driver for the Ingenic JZ47xx SoCs

From: Paul Cercueil
Date: Fri Mar 01 2019 - 18:33:52 EST


Hi,

Le ven. 1 mars 2019 à 18:00, Sam Ravnborg <sam@xxxxxxxxxxxx> a écrit :
Hi Paul.

Driver looks good and is a very nice piece of work.
In the following a number of minor issues.

One area that jumped at me was framedesc and the use of dma_alloc_coherent()
I hope someone that knows the memory handling better can give some advice here.
To me it looks like something drm should be able to help you with.

Sam

--- a/drivers/gpu/drm/Makefile
+++ b/drivers/gpu/drm/Makefile
@@ -94,6 +94,7 @@ obj-$(CONFIG_DRM_TEGRA) += tegra/
obj-$(CONFIG_DRM_STM) += stm/
obj-$(CONFIG_DRM_STI) += sti/
obj-$(CONFIG_DRM_IMX) += imx/
+obj-y += ingenic/

To avoid visiting ingenic/ dir for every build use:
obj-$(CONFIG_DRM_INGENIC) += ingennic/

And accept that you need to do this also in ingenic/Makefile

OK.

obj-$(CONFIG_DRM_MEDIATEK) += mediatek/
obj-$(CONFIG_DRM_MESON) += meson/
obj-y += i2c/
+#include <drm/drm_gem_cma_helper.h>
+#include <drm/drm_fb_cma_helper.h>
+#include <drm/drm_fb_helper.h>
+#include <drm/drm_fourcc.h>
+#include <drm/drm_gem_framebuffer_helper.h>
+#include <drm/drm_irq.h>
+#include <drm/drm_panel.h>
+#include <drm/drm_plane.h>
+#include <drm/drm_plane_helper.h>
+#include <drm/drm_probe_helper.h>
+#include <drm/drm_vblank.h>
+
+#include <dt-bindings/display/ingenic,drm.h>
+
+#include "../drm_internal.h"
No other drivers needs drm_internal.h - what makes this driver special?
Or do we have something in drm_internal that should be moved to
include/drm/ ?

I needed to include it for drm_vblank_cleanup().
But I guess I don't need to call that function? It's not obvious to me
how the error handling and driver removal should be done.

+struct ingenic_framedesc {
+ uint32_t next;
+ uint32_t addr;
+ uint32_t id;
+ uint32_t cmd;
+} __packed;
For internel types u32 is the typical use.
uint32_t is usually used in uapi.

Consider to use dma_addr_t for addresses - we see this in other drivers.
If there are alignemnt constraints then add these.

So 'ingenic_framedesc' represents a DMA descriptor, in the format that
the hardware expects it. The fields must be 32-bit. Is a dma_addr_t
the best option in that case?

+
+struct jz_soc_info {
+ bool needs_dev_clk;
+};
+
+struct ingenic_drm {
+ struct device *dev;
+ void __iomem *base;
+ struct regmap *map;
+ struct clk *lcd_clk, *pix_clk;
+
+ u32 lcd_mode;
+
+ struct ingenic_framedesc *framedesc;
+ dma_addr_t framedesc_phys;
+
+ struct drm_device *drm;
+ struct drm_plane primary;
+ struct drm_crtc crtc;
+ struct drm_connector connector;
+ struct drm_encoder encoder;
+ struct drm_panel *panel;
+
+ struct device_node *panel_node;
panel_node is not used outside ingenic_drm_probe() so no need
to have it here.

Noted.


+
+ unsigned short ps_start, ps_end, cls_start,
+ cls_end, spl_start, spl_end, rev_start;
I do not see these used, they can all be dropped.

+};
+
+
+static const struct regmap_config ingenic_drm_regmap_config = {
+ .reg_bits = 32,
+ .val_bits = 32,
+ .reg_stride = 4,
+
+ .max_register = JZ_REG_LCD_CMD1,
+ .writeable_reg = ingenic_drm_writeable_reg,
+};

+1 for using regmap.

+
+static inline bool ingenic_drm_lcd_is_special_mode(u32 mode)
+{
+ switch (mode) {
+ case JZ_DRM_LCD_SPECIAL_TFT_1:
+ case JZ_DRM_LCD_SPECIAL_TFT_2:
+ case JZ_DRM_LCD_SPECIAL_TFT_3:
+ return true;
+ default:
+ return false;
+ }
+}
Does it make sense to support these modes today?
If it is not used in practice then no need to carry it in the driver.

Yes, I have a board that uses such a panel.

+
+static inline bool ingenic_drm_lcd_is_stn_mode(u32 mode)
+{
+ switch (mode) {
+ case JZ_DRM_LCD_SINGLE_COLOR_STN:
+ case JZ_DRM_LCD_SINGLE_MONOCHROME_STN:
+ case JZ_DRM_LCD_DUAL_COLOR_STN:
+ case JZ_DRM_LCD_DUAL_MONOCHROME_STN:
+ return true;
+ default:
+ return false;
+ }
+}
This function is not used and can be deleted.
stn display are not really worth it these days anyway.


+static void ingenic_drm_crtc_atomic_flush(struct drm_crtc *crtc,
+ struct drm_crtc_state *oldstate)
+{
+ struct ingenic_drm *priv = drm_crtc_get_priv(crtc);
+ struct drm_crtc_state *state = crtc->state;
+ struct drm_pending_vblank_event *event = state->event;
+ struct drm_framebuffer *drm_fb = crtc->primary->state->fb;
+ const struct drm_format_info *finfo;
+ unsigned int width, height;
+
+ if (drm_atomic_crtc_needs_modeset(state)) {
+ finfo = drm_format_info(drm_fb->format->format);
+ width = state->adjusted_mode.hdisplay;
+ height = state->adjusted_mode.vdisplay;

width and height are unused and can be dropped.

+
+ ingenic_drm_crtc_update_timings(priv, &state->mode);
+
+ ingenic_drm_crtc_update_ctrl(priv, finfo->depth);
+ ingenic_drm_crtc_update_cfg(priv, &state->adjusted_mode);
+
+ clk_set_rate(priv->pix_clk, state->adjusted_mode.clock * 1000);
+
+ regmap_write(priv->map, JZ_REG_LCD_DA0, priv->framedesc->next);
+ }
+
+ if (event) {
+ state->event = NULL;
+
+ spin_lock_irq(&crtc->dev->event_lock);
+ if (drm_crtc_vblank_get(crtc) == 0)
+ drm_crtc_arm_vblank_event(crtc, event);
+ else
+ drm_crtc_send_vblank_event(crtc, event);
+ spin_unlock_irq(&crtc->dev->event_lock);
+ }
+}
+
+static void ingenic_drm_plane_atomic_update(struct drm_plane *plane,
+ struct drm_plane_state *oldstate)
+{
+ struct ingenic_drm *priv = drm_plane_get_priv(plane);
+ struct drm_plane_state *state = plane->state;
+ const struct drm_format_info *finfo;
+ unsigned int width, height;
+
+ finfo = drm_format_info(state->fb->format->format);
+ width = state->crtc->state->adjusted_mode.hdisplay;
+ height = state->crtc->state->adjusted_mode.vdisplay;
adjusted_mode->{h,v}display are both ints. So there is a hidden conversion to unsigned int here.
and framedesc->cmd is uint32_t then this is maybe fine.
Noticed so added a comment about it.

+
+ priv->framedesc->addr = drm_fb_cma_get_gem_addr(state->fb, state, 0);
+
+ priv->framedesc->cmd = width * height * ((finfo->depth + 7) / 8) / 4;
+ priv->framedesc->cmd |= JZ_LCD_CMD_EOF_IRQ;
+}
+
+static struct drm_driver ingenic_drm_driver_data = {
+ .driver_features = DRIVER_MODESET | DRIVER_GEM | DRIVER_PRIME
+ | DRIVER_ATOMIC | DRIVER_HAVE_IRQ,

DRIVER_HAVE_IRQ are only for legacy drivers these days.
You can drop it.

+ .name = "ingenic-drm",
+ .desc = "DRM module for Ingenic SoCs",
+ .date = "20190228",
+ .major = 1,
+ .minor = 0,
+ .patchlevel = 0,
+
+ .fops = &ingenic_drm_fops,
+
+ .dumb_create = drm_gem_cma_dumb_create,
+ .gem_free_object_unlocked = drm_gem_cma_free_object,
+ .gem_vm_ops = &drm_gem_cma_vm_ops,
+
+ .prime_handle_to_fd = drm_gem_prime_handle_to_fd,
+ .prime_fd_to_handle = drm_gem_prime_fd_to_handle,

+ .gem_prime_import = drm_gem_prime_import,
+ .gem_prime_export = drm_gem_prime_export,
The two assignments above are not needed. This is the default behaviour. See drm_drv.h
(from drm-misc-next)

+static int ingenic_drm_probe(struct platform_device *pdev)
+{
+ const struct jz_soc_info *soc_info;
+ struct device *dev = &pdev->dev;
+ struct ingenic_drm *priv;
+ struct clk *parent_clk;
+ struct drm_device *drm;
+ struct resource *mem;
+ void __iomem *base;
+ long parent_rate;
+ int ret, irq;
+
+ soc_info = device_get_match_data(dev);
+ if (!soc_info)
+ return -EINVAL;
+
+ priv = devm_kzalloc(dev, sizeof(*priv), GFP_KERNEL);
+ if (!priv)
+ return -ENOMEM;
+
+ priv->dev = dev;
+
+ mem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+ priv->base = base = devm_ioremap_resource(dev, mem);
+ if (IS_ERR(base))
+ return PTR_ERR(base);
+
+ irq = platform_get_irq(pdev, 0);
+ if (irq < 0) {
+ dev_err(dev, "Failed to get platform irq\n");
+ return -ENOENT;
+ }
+
+ priv->map = devm_regmap_init_mmio(dev, base,
+ &ingenic_drm_regmap_config);
+ if (IS_ERR(priv->map)) {
+ dev_err(dev, "Failed to create regmap\n");
+ return PTR_ERR(priv->map);
+ }
+
+ if (soc_info->needs_dev_clk) {
+ priv->lcd_clk = devm_clk_get(dev, "lcd");
+ if (IS_ERR(priv->lcd_clk)) {
+ dev_err(dev, "Failed to get lcd clock\n");
+ return PTR_ERR(priv->lcd_clk);
+ }
+ }
+
+ priv->pix_clk = devm_clk_get(dev, "lcd_pclk");
+ if (IS_ERR(priv->pix_clk)) {
+ dev_err(dev, "Failed to get pixel clock\n");
+ return PTR_ERR(priv->pix_clk);
+ }
+
+ priv->panel_node = of_parse_phandle(dev->of_node, "ingenic,panel", 0);
+ if (!priv->panel_node) {
+ DRM_INFO("No panel found");
+ } else {
+ priv->panel = of_drm_find_panel(priv->panel_node);
+ of_node_put(priv->panel_node);
+
+ if (IS_ERR(priv->panel)) {
+ if (PTR_ERR(priv->panel) == -EPROBE_DEFER)
+ return -EPROBE_DEFER;
+
+ priv->panel = NULL;
+ } else {
+ ret = drm_panel_prepare(priv->panel);
+ if (ret && ret != -ENOSYS)
+ return ret;
+ }
+ }
+
+ if (priv->panel) {
+ ret = device_property_read_u32(dev, "ingenic,lcd-mode",
+ &priv->lcd_mode);
+ if (ret) {
+ dev_err(dev, "Unable to read ingenic,lcd-mode property\n");
+ return ret;
+ }
+ }
+
+ priv->framedesc = dma_alloc_coherent(dev, sizeof(*priv->framedesc),
+ &priv->framedesc_phys, GFP_KERNEL);
+ if (!priv->framedesc)
+ return -ENOMEM;
+
+ priv->framedesc->next = priv->framedesc_phys;
+ priv->framedesc->id = 0xdeafbead;
I did not really follow this code.
But you have framedesc that include a next pointer which is uint32_t,
and you assign to it framedesc_phys which is dmaaddr_t.
This looks fishy to me, but maybe it is fine.

As explained above, the framedesc is the hardware DMA descriptor, its fields
must be 32-bit, that's why it's done this way.

When browsing other drivers I see uses of for example {dma,dmam}_pool_create()

When lookign at the origianl fbdev code it is obvious where this comes from.
Notice that in the fbdev code JZ_REG_LCD_DA0 register is updated in the enable function.
This is done different in this driver.
Just an observation, what is right I dunno.

JZ_REG_LCD_DA0 should contain the address of the DMA descriptor. In this driver,
I create a single DMA descriptor that commands the transfer of a full frame.
It only has to be set once, so it could even be written in the probe function,
provided the clocks are enabled.

+static int ingenic_drm_remove(struct platform_device *pdev)
+{
+ struct drm_device *drm = platform_get_drvdata(pdev);
+ struct ingenic_drm *priv = drm->dev_private;
+
+ drm_dev_unregister(drm);
+ drm_mode_config_cleanup(drm);
+
+ if (priv->lcd_clk)
+ clk_disable_unprepare(priv->lcd_clk);
+ clk_disable_unprepare(priv->pix_clk);
+
+ drm_vblank_cleanup(drm);
+ drm_irq_uninstall(drm);
+
+ drm_encoder_cleanup(&priv->encoder);
+ drm_connector_cleanup(&priv->connector);
+ drm_crtc_cleanup(&priv->crtc);
+ drm_plane_cleanup(&priv->primary);
+
+ drm_dev_put(drm);
+
+ dma_free_coherent(&pdev->dev, sizeof(*priv->framedesc),
+ priv->framedesc, priv->framedesc_phys);
+
+ return 0;
+}
+
+static const struct jz_soc_info jz4740_soc_info = {
+ .needs_dev_clk = true,
+};
+
+static const struct jz_soc_info jz4725b_soc_info = {
+ .needs_dev_clk = false,
+};
+
+static const struct of_device_id ingenic_drm_of_match[] = {
+ { .compatible = "ingenic,jz4740-drm", .data = &jz4740_soc_info },
+ { .compatible = "ingenic,jz4725b-drm", .data = &jz4725b_soc_info },
+ {},
Use /* sentinel */ like in most drivers.

For this and other embedded comments I didn't directly reply: duly noted.

Thanks for the review.

Greetings,
-Paul