This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH 1/3] aarch64: Clean up _dl_runtime_resolve
On 08/02/2018 12:03 PM, Szabolcs Nagy wrote:
> On 01/08/18 23:23, rth@twiddle.net wrote:
>> From: Richard Henderson <richard.henderson@linaro.org>
>>
>> * sysdeps/aarch64/dl-trampoline.S (_dl_runtime_resolve):
>> Do not record unwind info for arguments; this is unneeded;
>> do not save x9 just to have a register to pair with x8;
>> properly include the 16 bytes of PLT stack into the unwind;
>> create a frame pointer with the spare stack slot;
>> rearrange the exit to only adjust the stack once.
>
> i thought the cfi annotations were needed for all registers
> in case the debugger wants to investigate register content
> across a _dl_runtime_resolve frame (possibly several frames
> up in the call stack),
However that's typically for the call-saved registers, where the compiler might
save data in that register across the call. These are not call-saved
registers. They are argument registers. There will not be any debug info that
refers to them.
> this may not be a common use case though
> and i don't know what's the convention in glibc asm, the compiler
> seems to emit annotation for all spilled registers with -g.
Sure, because the compiler is spilling call-saved registers.
The others it just clobbers with no annotation.
>> - cfi_startproc
>> .align 2
>> _dl_runtime_resolve:
>> /* AArch64 we get called with:
>> @@ -41,46 +40,24 @@ _dl_runtime_resolve:
>> [sp, #8] lr
>> [sp, #0] &PLTGOT[n]
>> */
>> -
>> + cfi_startproc
>
> is there a problem keeping it at its original place above?
> the tlsdesc asm has cfi_startproc at the same place.
It could stay where it is, but I thought it clearer to place the following two
annotations immediately adjacent (because it's state incoming, not anything we
are doing here, and should not be separated from the start). Further, to place
all of the annotations immediately after the comment that describes why.
r~