On Sat, Apr 16, 2016 at 2:07 AM, Vineet Gupta Vineet.Gupta1@synopsys.com wrote:

On Friday 15 April 2016 09:18 PM, Alexey Brodkin wrote:

...

And now the question is how to force DRM subsystem or just that driver to use whatever predefined (say via device tree) location in memory for data buffer allocation.

It seems this is pretty easy to do with DT reserved-memory binding.

You need to partition memory into @memory and @reserved-memory. Later can be subdivided into more granular regions and your driver can refer to one of the regions.

jfyi, it might be useful to look at msm_init_vram() which has support to wrap vram carveout as gem buffer, for exact same purpose..

BR, -R

Something like below (untested)

•   memory {
  

•           device_type = "memory";
  

•           reg = <0x0 0x80000000 0x0 0xA0000000>;
  

•           #address-cells = <2>;
  

•           #size-cells = <2>;
  

•   };
  

•   reserved-memory {
  

•           #address-cells = <2>;
  

•           #size-cells = <2>;
  

•           ranges;
  

•           /* This memory bypasses IOC port */
  

•           fb_reserved@A0000000 {
  

•                   reg = <0x0 0xA0000000 0x0 0xAF000000>;
  

•                   #address-cells = <2>;
  

•                   #size-cells = <2>;
  

•                   /* no-map;   */
  

•           };
  

•   };
  

•   fb0: video@12300000 {
  

•           memory-region = <&fb_reserved>;
  

•           /* ... */
  

•   };
  
  

This might also need a DT helper in ARC mm init code.

•   early_init_fdt_scan_reserved_mem();
  
  

HTH, -Vineet