Hi York,
On 7 August 2015 at 17:12, York Sun yorksun@freescale.com wrote:
On 08/07/2015 03:47 PM, Simon Glass wrote:
Hi York,
On 7 August 2015 at 14:48, York Sun yorksun@freescale.com wrote:
Simon,
I was doing an experiment to put the load address and entry address of Linux to higher than 32-bit address. I found it is broken to process more than 32-bit addresses. When I attempted to fix it, I was troubled by those code used for both host and target, like common/image-fit.c. For example, to process 64-bit address, the function
int fit_image_get_load(const void *fit, int noffset, ulong *load)
should be converted to
int fit_image_get_load(const void *fit, int noffset, uint64_t *load)
ulong is 64-bit for 64-bit target such as ARMv8, but it can be 32-bit on host. If I use uint64_t, all related code in bootm and others need to change. Before I go too far, I'd like to check if anyone has tried to enable this in FIT image.
#address-cells = <2>;
I can try to use uint64_t in place of ulong for all related code if that's right. That will be a lot of change.
Perhaps I misunderstand something, but I think ulong should be OK on the host. I just needs to hold a machine address. On a 32-bit host this cannot be 64-bit. Can you explain the problem a bit more?
I have not need #address-cells in a FIT.
It would be better to use ulong for addresses in U-Boot I think. It is safe and efficient on both 32- and 64-bit machines.
Simon,
Considering this situation, building FIT image on an ubuntu 12.04 32-bit host, for ARMv8 target. I believe ulong is OK on 64-bit target. But it is not for the host mkimage. To proper deal with address cell = 2, the function fit_image_get_load() and fit_image_get_entry() need to make 64-bit address from FIT image. mkimage runs on 32-bit host doesn't take ulong as 64-bit, does it? So I can generate a FIT image with the address cell and load/entry address I need, but I cannot display it correctly with mkimage. I think I can use the image on 64-bit target after fixing some parsing code as I mentioned.
The background for this experiment is I am trying to shift DDR to a continuous space. For ARMv8, DDR can have three regions, with only 2GB under 32-bit address space.
We really don't want to end up with everything being 64-bit just because the tools need to work correctly. I wonder if phys_addr_t might help here? We could define that to be 64-bit for the host, perhaps?
Regards, Simon