[PATCH] drm/pl111: Use max memory bandwidth for resolution

[eric@anholt.net (Eric Anholt)]

Linus Walleij <linus.walleij@linaro.org> writes:

> We were previously selecting 1024x768 and 32BPP as the default
> set-up for the PL111 consumers.
>
> This does not work on elder systems: the device tree bindings
> support a property "max-memory-bandwidth" in bytes/second that
> states that if you exceed this the memory bus will saturate.
> The result is flickering and unstable images.
>
> Parse the "max-memory-bandwidth" and respect it when
> intializing the driver. On the RealView PB11MP we get a nice
> 1024x768 console with RGB565 as default with this code.
>
> If the device tree does not specify the maximum memory
> bandwidth we keep the old assumption that we can support
> 1024x768, 32BPP.
>
> Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
> ---
>  drivers/gpu/drm/pl111/pl111_drm.h |  1 +
>  drivers/gpu/drm/pl111/pl111_drv.c | 99 +++++++++++++++++++++++++++++++++++++--
>  2 files changed, 97 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/pl111/pl111_drm.h b/drivers/gpu/drm/pl111/pl111_drm.h
> index 440f53ebee8c..770ef5ce3645 100644
> --- a/drivers/gpu/drm/pl111/pl111_drm.h
> +++ b/drivers/gpu/drm/pl111/pl111_drm.h
> @@ -56,6 +56,7 @@ struct pl111_drm_dev_private {
>  	struct drm_fbdev_cma *fbdev;
>  
>  	void *regs;
> +	u32 memory_bw;
>  	u32 ienb;
>  	u32 ctrl;
>  	/* The pixel clock (a reference to our clock divider off of CLCDCLK). */
> diff --git a/drivers/gpu/drm/pl111/pl111_drv.c b/drivers/gpu/drm/pl111/pl111_drv.c
> index 101a9c7db6ff..1b47ee67097d 100644
> --- a/drivers/gpu/drm/pl111/pl111_drv.c
> +++ b/drivers/gpu/drm/pl111/pl111_drv.c
> @@ -58,6 +58,7 @@
>  #include <linux/dma-buf.h>
>  #include <linux/module.h>
>  #include <linux/slab.h>
> +#include <linux/of.h>
>  
>  #include <drm/drmP.h>
>  #include <drm/drm_atomic_helper.h>
> @@ -80,21 +81,107 @@ static const struct drm_mode_config_funcs mode_config_funcs = {
>  	.atomic_commit = drm_atomic_helper_commit,
>  };
>  
> +/**
> + * pl111_choose_max_resolution() - choose max resolution
> + * @dev: DRM device
> + * @memory_bw: the graphics maximum memory bandwidth in bytes/s
> + * @max_width: returns the maximum width
> + * @max_height: returns the maximum height
> + * @bpp: returns the maximum bips per pixed (32 or 16)
> + *
> + * This function attempts to cut down the maximum supported resolution
> + * to something the system memory bus can handle. The PL111 and pixel
> + * clocks may be able to support higher resolutions and color depths
> + * but as we go up the memory bus get saturated and becomes the
> + * bottleneck for the resolution. On several systems using e.g.
> + * 1024x768 with 32 BPP results in instable flickering images.
> + */
> +static void pl111_choose_max_resolution(struct drm_device *dev,
> +					u32 memory_bw,
> +					int *max_width,
> +					int *max_height,
> +					unsigned int *bpp)
> +{
> +	/* No limitations, this is the most aggressive */
> +	if (!memory_bw) {
> +		*max_width = 1024;
> +		*max_height = 768;
> +		*bpp = 32;
> +		return;
> +	}
> +
> +	/*
> +	 * 1024x768 with RGBX8888 requires a memory bandwidth of
> +	 * 65Mhz * 4 bytes = 260000000 bytes per second.
> +	 */
> +	if (memory_bw >= 260000000) {
> +		*max_width = 1024;
> +		*max_height = 768;
> +		*bpp = 32;
> +		return;
> +	}
> +
> +	/*
> +	 * 800x600 with RGB8888 requires a memory bandwidth of
> +	 * 36Mhz * 4 bytes = 144000000 bytes per second. But we
> +	 * assume the user prefer higher resolution over more
> +	 * color depth, so we do not add this mode here.
> +	 */
> +
> +	/*
> +	 * 1024x768 with RGB565 requires a memory bandwidth of
> +	 * 65Mhz * 2 bytes = 130000000 bytes per second.
> +	 */
> +	if (memory_bw >= 130000000) {
> +		*max_width = 1024;
> +		*max_height = 768;
> +		*bpp = 16;
> +		return;
> +	}
> +
> +	/*
> +	 * 800x600 with RGB565 requires a memory bandwidth of
> +	 * 36Mhz * 2 bytes = 72000000 bytes per second.
> +	 */
> +	if (memory_bw >= 72000000) {
> +		*max_width = 800;
> +		*max_height = 600;
> +		*bpp = 16;
> +		return;
> +	}
> +
> +	/*
> +	 * 640x480 with RGB565 requires a memory bandwidth of
> +	 * 25.175Mhz * 2 bytes = 50350000 bytes per second.
> +	 */
> +	if (memory_bw < 50350000)
> +		dev_err(dev->dev, "can't even do 640x480 VGA RGB565, proceed anyway\n");
> +
> +	*max_width = 640;
> +	*max_height = 480;
> +	*bpp = 16;
> +}
> +
>  static int pl111_modeset_init(struct drm_device *dev)
>  {
>  	struct drm_mode_config *mode_config;
>  	struct pl111_drm_dev_private *priv = dev->dev_private;
>  	struct drm_panel *panel;
>  	struct drm_bridge *bridge;
> +	unsigned int bpp; /* bits per pixel */
>  	int ret = 0;
>  
>  	drm_mode_config_init(dev);
>  	mode_config = &dev->mode_config;
>  	mode_config->funcs = &mode_config_funcs;
>  	mode_config->min_width = 1;
> -	mode_config->max_width = 1024;
>  	mode_config->min_height = 1;
> -	mode_config->max_height = 768;
> +
> +	pl111_choose_max_resolution(dev, priv->memory_bw,
> +				    &mode_config->max_width,
> +				    &mode_config->max_height, &bpp);
> +	dev_info(dev->dev, "cap resolution at %u x %u, %u BPP\n",
> +		 mode_config->max_width, mode_config->max_height, bpp);

I think this is the wrong place in the pipeline to be doing this, but I
don't have a complete solution so I'm not necessarily saying no.  Things
I think we should do for bandwidth limits:

A new pl111_mode_valid() rejects modes with width*height*2 > bandwidth
(if we can't scan it out with our smallest format, don't advertise it).

pl111_display_check() rejects modes with width*height*bpp > bandwidth
(if we can't scan out this particular configuration, let them know we
can't set the mode).

Ideally given those two things, fbdev and X11 would notice that the
preferred mode fails at 24bpp and fall back to 16bpp.  I don't think
either of those does so today, though.

Interested in tackling any of these?
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 832 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20180125/89f34bf6/attachment.sig>