On Thu, Oct 7, 2021 at 7:27 PM Jens Axboe <axboe@xxxxxxxxx> wrote:
>
> On 10/7/21 12:31 AM, Ammar Faizi wrote:
> > Create `src/nolibc.c` to substitute libc functions like `memset()`,
> > `malloc()` and `free()`. Only build this file when we build liburing
> > without libc.
> >
> > Wrap libc functions with `static inline` functions, defined in
> > `src/lib.h`, currently we have:
> > 1) `get_page_size`
> > 2) `uring_memset`
> > 3) `uring_malloc`
> > 4) `uring_free`
> >
> > Add conditional preprocessor in `src/{syscall,lib}.h` to use arch
> > dependent and nolibc functions when we build liburing without libc.
> >
> > Extra notes for tests:
> > 1) Functions in `src/syscall.c` require libc to work.
> > 2) Tests require functions in `src/syscall.c`.
> >
> > So we build `src/syscall.c` manually from test's Makefile.
> >
> > The Makefile in `src/` dir still builds `src/syscall.c` when we
> > compile liburing with libc.
>
> Why are we jumping through hoops on the naming? You add a uring_memset()
> that the lib is supposed to use, then that calls memset() with libc or
> __uring_memset() if we're building without libc. Why not just have the
> nolibc build provide memset() and avoids this churn? Not just in terms
> of the immediate patch, that matters less. But longer term where
> inevitably a memset() or similar will be added to the existing code
> base.
>
Ah, good point. The syscall wrappers gave me this bad suggestion. I know
we have a rationale to wrap syscalls because we need to eliminate `errno`
being used in C files. But we don't actually need extra wrappers for
`memset()`, `malloc()` and `free()`.
I agree that we can just provide `memset()`, `malloc()` and `free()` for
the nolibc build without having extra wrappers like that.
I will address this and send the rest as a PATCHSET.
--
Ammar Faizi
