Target Page Align

[akuhlken]

For my project I am attempting to use the uc_mem_map_ptr() function to map an address space created on the host with malloc into the emulator. For my purposes I would like the guest addresses to line up 1-1 with host addresses for the process. To do this I hoped to do something along the lines of:

*code_ptr = malloc(code_size);
uint64_t code_addr = (uint64_t)code_ptr;
uc_mem_map_ptr(*uc, code_addr, code_size, UC_PROT_ALL, *code_ptr);

However this throws an error in the check done by mem_map_check() specifically it fails the check here:
// address must be aligned to uc->target_page_size
if ((address & uc->target_page_align) != 0) {
return UC_ERR_ARG;
}

My question is why does Unicorn enforce this page align requirement for mapped sections and would it be possible to remove this requirement. I'm also curious if there are any other suggestions on how to accomplish this. My motivation is this: I would like to be able to pass a pointer to an object created by the host as an argument in one of the argument registers for a guest function without needing to hard code an offset from an arbitrary memory mapped region.

[wtdcode]

Unfortunately, that's the requirements from the target MMU. You could calculate the offset to the nearest aligned address I think.

[akuhlken]

I spent some more time with this issue and it seems as though simply commenting out that check for both mapping size and for address alignment allows me to map any arbitrary start address and size. I’m not confident that this is correct but for my own test cases as well as the built in unit tests everything seems to function correctly. Do you have any insight on where I might see problems crop up with this change?

[wtdcode]

I have no specific idea for this but many qemu internal implementation would assume page aligned memory regions.

[akuhlken]

Do you have any suggestions on how I can better verify correctness before continuing? Or know of any documentation that describes this issue?

[wtdcode]

Generally I would suggest you map the whole page and write your contents there. Why does that not work for you?

I think you will be mostly safe if you switch to virtual tlb mode but I have no guarantee.

[akuhlken]

I have opted to allocate page aligned sections on the host with posix_memalign and then as you suggested map the entire page in the guest. That way the guest and host addresses can still be mapped 1:1