Re: [PATCH liburing 0/3] Add x86 32-bit support for the nolibc build

[Jens Axboe <axboe@kernel.dk>]

On 4/14/22 4:41 PM, Ammar Faizi wrote:
> Hi,
> 
> This series adds nolibc support for x86 32-bit. There are 3 patches in
> this series:
> 
> 1) Use `__NR_mmap2` instead of `__NR_mmap` for x86 32-bit.
> 2) Provide `get_page_size()` function for x86 32-bit.
> 3) Add x86 32-bit native syscall support.
> 
> The most noticeable changes is the patch 3. Unlike x86-64, only
> use native syscall from the __do_syscall macros when CONFIG_NOLIBC is
> enabled for 32-bit build. The reason is because the libc syscall
> wrapper can do better in 32-bit. The libc syscall wrapper can dispatch
> the best syscall instruction that the environment is supported, there
> are at least two variants syscall instruction for x86 32-bit, they are:
> `int $0x80` and `sysenter`. The `int $0x80` instruction is always
> available, but `sysenter` is not, it relies on VDSO. liburing always
> uses `int $0x80` for syscall if it's compiled with CONFIG_NOLIBC,
> otherwise it uses whatever the libc provides.
> 
> Extra notes for __do_syscall6() macro:
> On i386, the 6th argument of syscall goes in %ebp. However, both Clang
> and GCC cannot use %ebp in the clobber list and in the "r" constraint
> without using -fomit-frame-pointer. To make it always available for
> any kind of compilation, the below workaround is implemented:
> 
>   1) Push the 6-th argument.
>   2) Push %ebp.
>   3) Load the 6-th argument from 4(%esp) to %ebp.
>   4) Do the syscall (int $0x80).
>   5) Pop %ebp (restore the old value of %ebp).
>   6) Add %esp by 4 (undo the stack pointer).
> 
> WARNING:
>   Don't use register variables for __do_syscall6(), there is a known
>   GCC bug that results in an endless loop.
> 
> BugLink: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105032
> 
> 
> ===== How is this tested? =====
> 
> This has been tested on x86-64 Linux (compiled with 32-bit bin support)
> with the following commands:
> 
>   sudo apt-get install gcc-i686-linux-gnu g++-i686-linux-gnu -y;
>   ./configure --cc=i686-linux-gnu-gcc --cxx=i686-linux-gnu-g++ --nolibc;
>   sudo make -j8 runtests;

Looks reasonable to me, even with the warts. I keep threatening to do a
2.2 release, and I do want to do that soon, so question is if we defer
this patchset until after that has happened?

I'm looking for a gauge of confidence on the series.

-- 
Jens Axboe