Re: Valgrind port to Hurd

[Samuel Thibault]

Hello,

Athul M A, le dim. 10 mars 2019 14:35:41 +0530, a ecrit:
> My observation is that we need to handle mach traps in a way similar
> to what the other operating systems do theirs

Yes, very probably.

> and RPCs using a new specific handler. Could you guide me
> through the process and exactly how these fit into valgrind?

Well, I don't know how exactly how it fits into valgrind. What I know
is that valgrind has its own very light libc layer for its own uses
(allocating memory and reading files, basically), and that's what needs
to be ported on top of mach traps and hurd RPCs.

> Since we use RPCs in GNU Hurd, Will it be okay if we call them directly after
> we do the preprocessing valgrind does? Or should we make the call in a way it
> is implemented(using mach_msg_traps)? Reason why I felt this doubt is because
> in the implementation for other distributions, system calls are done using
> assembly code and not using glibc equivalents.

I guess they'll need to be implemented with traps. I don't know exactly
why valgrind reimplements system calls, but I guess it's to clearly
separate out the valgrind pieces (not to be tracked by valgrind) from
the application pieces (to be tracked by valgrind).

Samuel