Re: [Qemu-devel] [PATCH 4/6] linux-user: Provide safe_syscall for aarch64

[Richard Henderson]

On 06/13/2016 03:04 PM, Peter Maydell wrote:
> > +        .global safe_syscall_base
> > +        .global safe_syscall_start
> > +        .global safe_syscall_end
> > +        .type   safe_syscall_base, #function
> > +        .type   safe_syscall_start, #function
> > +        .type   safe_syscall_end, #function
>
> _start and _end aren't function entry points, so is it OK
> to mark them as functions?

Yes.  Indeed, if you don't, the objdump will think these are data symbols and 
fail to disassemble the instructions that follow.


> (The 'as' manual doesn't document what setting .type does in much
> detail...)

It sets the Elf_Sym.st_type field.
But what that implies is very host specific.

> http://man7.org/linux/man-pages/man2/syscall.2.html
> doesn't mention a syscall argument in x7, just x0..x5.

I took glibc as definitive, since that's code that definitely works.

https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/aarch64/syscall.S;h=98c1b42ee96fe0fb39955849773ca16d095803e0;hb=HEAD

I'll also point you at the kernel's __sys_trace, which does in fact save and 
restore all 8 registers around the tracing call-outs.

> > +       mov     x9, x0          /* signal_pending pointer */
> > +       mov     w8, w1          /* syscall number */
>
> Seems a bit odd to use a 32-bit move for this when our input
> calling convention has it as 64 bits and the kernel's calling
> convention has it as 64 bits.

I got that from glibc too.



r~