This is the mail archive of the mailing list for the glibc project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH 4/6] aarch64: Improve sysdep-cancel.h

On 21 May 2014 16:10, Richard Henderson <> wrote:
> On 05/20/2014 01:56 PM, Richard Henderson wrote:
>> -    mov x16, x0;     /* put mask in safe place.  */                  \
>> -    UNDOCARGS_##args;        /* restore syscall args.  */                    \
>> -    mov x8, SYS_ify (syscall_name);  /* do the call.  */             \
>> -    svc      0;                                                              \
>> -    str x0, [sp, -16]!;      /* save syscall return value.  */               \
>> -    cfi_adjust_cfa_offset (16);                                              \
>> -    mov x0, x16;      /* get mask back.  */                          \
> Oh, I should have mentioned, that there's either a bug or inconsistency here.
> We're saving a value in x16 around the syscall.  Except that on the C side, for
> the inline syscalls, we mark x16 as clobbered.
> My patch assumes that we want to assume "normal-ish" calling conventions for
> the kernel and puts the value in the normal call saved register x19.
> If there are in fact a set of calling-convention call-clobbered registers that
> are not clobbered by syscalls, then we should consider adjusting the inline
> syscalls to match.

When this was discussed previously it turns out that everything apart
from the return value is preserved:

I guess it would be better to avoid clobbering so many registers in
the inline case.

Will Newton
Toolchain Working Group, Linaro

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]