support user set remote mmap vma address

[duguhaotian]

1. os auto assignment vma addr maybe conflict with vma in gpu living migrate scene;
2. so, we should give choice to user;
3. default, user_set_remote_map is 0L, so we donot change default behave;

user set vma usecase:

// we use compel/test/infect/spy.c to show example

static int do_infection(int pid)
{
       ... ....
	ictx = compel_infect_ctx(ctl);
	ictx->log_fd = STDERR_FILENO;

       // TODO: use should set assignment vma addr at here
      ictx->user_set_remote_map = xxxxL;

	parasite_setup_c_header(ctl);
       .... .....
}

[rst0git]

os auto assignment vma addr maybe conflict with vma in gpu living migrate scene;

There are several different methods for checkpoint/restore of GPU state (e.g., cuda-checkpoint, amdgpu plugin, cricket, cedana, a device-proxy mechanism from Microsoft, and a very similar solution from Tencent).

@duguhaotian Would you be able to provide more information about this use-case and why this change is necessary?

[duguhaotian]

os auto assignment vma addr maybe conflict with vma in gpu living migrate scene;

There are several different methods for checkpoint/restore of GPU state (e.g., cuda-checkpoint, amdgpu plugin, cricket, cedana, a device-proxy mechanism from Microsoft, and a very similar solution from Tencent).

@duguhaotian Would you be able to provide more information about this use-case and why this change is necessary?

similar solution from Microsoft, log and reply.

1. But, i encountered a log: mmap vma address and do somethings.
2. So, i should reply this mmap vma and do somethings.
3. And then, probabilistic failure if criu vma address same to log vma address;

[duguhaotian]

@rst0git

[duguhaotian]

@avagin @rst0git please check this pr, thks.

[avagin]

LGTM pls resolve all items and we will merge it.

btw: how and where do you use this?

[duguhaotian]

LGTM pls resolve all items and we will merge it.

btw: how and where do you use this?

greate, thks!! @avagin

1. pr comments show how to use it;
2. log/reply to restore state of xpu, We encountered a mmap address conflict, causing the parasitic code execution to get stuck