How to give a user space program direct access to the frame buffer?

[ChocolateLoverRaj]

I can do this:

• Create a page table mapping that allows user space to to access and write to the frame buffer
• Give the user space program the pointer to the frame buffer and the frame buffer's metadata

But is the frame buffer guaranteed to have phys frames all to itself? I am worried about the same phys frames used by the frame buffer also containing other memory that shouldn't be exposed to user space.

[ChocolateLoverRaj]

I did some research and it looks like the frame buffer could have phys frames all to itself, but is not guaranteed to. I only tested it on qemu (with BIOS and UEFI) and in both the frame buffer was aligned to 4096 bytes and was a multiple of 4096 bytes so the frame buffer can be mapped to user accessible pages to share it with user space. I did not test it on any other computer though (but I will test it on robo360 running coreboot).

If the frame buffer is not aligned to 4096 bytes and not a multiple of 4096 bytes, then the only secure way is for the kernel to copy memory from user accessible memory to the frame buffer.

Update: I tested it on the Jinlon Chromebook running MrChromebox firmware and it also has a framebuffer which is a multiple of 4096 bytes.

[tsatke]

For more advanced stuff like that, I'd suggest not using "the framebuffer" that you get from the bootloader, but iterating the PCI device tree and getting your framebuffer from (e.g.) the VGA card directly (devos does it like that if you're interested). With that, you probably get more guarantees about size, alignment and writability, and you get more control about your mapping (WC, WT or whatever you want, although I'd assume the bootloader already maps it correctly).

[ChocolateLoverRaj]

Will the VGA card method also work on real modern computers?

[tsatke]

Probably.

https://wiki.osdev.org/VGA_Hardware