Re: [RFC PATCH v1 3/6] tools/nolibc: i386: Implement syscall with 6 arguments
From: Ammar Faizi
Date: Sun Mar 20 2022 - 11:07:56 EST
On 3/20/22 8:10 PM, David Laight wrote:
From: Ammar Faizi
Sent: 20 March 2022 09:38
In 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.
For clang (the Assembly statement can't clobber %ebp):
1) Save the %ebp value to the redzone area -4(%esp).
i386 doesn't have a redzone.
If you get a signal it will trash -4(%sp)
OK, I missed that one. Thanks for reviewing this.
For GCC, fortunately it has a #pragma that can force a specific function
to be compiled with -fomit-frame-pointer, so it can always use "r"(var)
where `var` is a variable bound to %ebp.
How is that going to work for an inlined functon?
It works, but obviously the inline modifier here is just a hint for the
compiler, not a requirement. I just took a look at the GCC generated code.
It doesn't inline the ____do_syscall6() despite it's marked as inline.
I think this one shouldn't be marked as inline. I will remove the inline
in the next version.
And using xchg is slow - it is always locked.
One possibility might be to do:
push arg6
push %ebp
mov %ebp, 4(%sp)
Did you mean `mov 4(%esp), %ebp`?
int 0x80
pop %ebp
add %esp,4
I think your solution is better than the xchg approach (with the 3rd line
fixed). Will take this in for the next version.
Although I'm not sure you really want to allocate 4k pages
for every malloc() call.
Probably better to write a mini 'libc' that uses sbrk()
and a best fit scan of a linear free list.
^ This part is addressed by Willy's response. I will still go with mmap()
in the next version.
And yes, we do waste space for small allocations here, but we don't hit
the scale that the waste space will cause problem.
--
Ammar Faizi