Re: [PATCH v5 22/22] loongarch: Fix Race conditions in pthread cancellation [BZ#12683]

[Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>]

On 11/04/23 09:55, Xi Ruoyao wrote:
> On Tue, 2023-04-11 at 20:00 +0800, caiyinyu wrote:
>>
>> 在 2023/4/11 下午2:19, Xi Ruoyao 写道:
>>> On Mon, 2023-04-10 at 17:46 -0300, Adhemerval Zanella via Libc-alpha
>>> wrote:
>>>
>>> /* snip */
>>>
>>>> +1:
>>>> +       addi.d  $r3,$r3,-16
>>>> +       cfi_def_cfa_offset (16)
>>>> +       st.d    $r1,$r3,8
>>>> +       cfi_offset (1, -8)
>>>> +       bl      __syscall_do_cancel
>>> Can we simply use "b __syscall_do_cancel" for 1: like AArch64?
>>
>> In LoongArch, the "b" instruction is equivalent to the "b" instruction
>> in AArch64, and similarly,
>>
>> the "bl" instruction in LoongArch is equivalent to the "bl"
>> instruction 
>> in AArch64.
>>
>>>> +
>>>> +END (__syscall_cancel_arch)
> 
> I've rewritten the syscall_cancel.S file as follow to exploit the tail
> call and improve readability.  It passed nptl/tst-cancel31:
> 
> #include <sysdep.h>
> #include <descr-const.h>
> 
> ENTRY (__syscall_cancel_arch)
> 
> 	.global __syscall_cancel_arch_start
> __syscall_cancel_arch_start:
> 
> 	/* if (*cancelhandling & CANCELED_BITMASK)
> 	     __syscall_do_cancel()  */
> 	ld.w	t0, a0, 0
> 	andi	t0, t0, TCB_CANCELED_BITMASK
> 	bnez	t0, 1f
> 
> 	/* Issue a 6 argument syscall.  */
> 	move	t1, a1
> 	move	a0, a2
> 	move	a1, a3
> 	move	a2, a4
> 	move	a3, a5
> 	move	a4, a6
> 	move	a5, a7
> 	move	a7, t1
> 	syscall 0
> 
> 	.global __syscall_cancel_arch_end
> __syscall_cancel_arch_end:
> 	jr	ra
> 1:
> 	b	__syscall_do_cancel
> 
> END (__syscall_cancel_arch)
> 

Thanks, I have merged it locally.  I will update my sourceware branch
with this change.