Re: [PATCH] LoongArch: Add Syscall Assembly Implementation

[caiyinyu <caiyinyu@loongson.cn>]

在 2023/3/23 下午5:25, Xi Ruoyao 写道:
> General question: is there a notable benefit optimizing syscall with
> assembly?  AFAIK nobody will put syscall on a hot path, and the cycles
> saved by using the assembly implementation should be negligible
> comparing with all the cost of context switch etc.

Yes.

New patch:

https://sourceware.org/pipermail/libc-alpha/2023-March/146588.html

Without this patch(objdump -d libc.so...):

00000000000dd45c <syscall>:
    dd45c:       02fec063        addi.d          $sp, $sp, -80(0xfb0)
    dd460:       02c0606c        addi.d          $t0, $sp, 24(0x18)
    dd464:       29c06065        st.d            $a1, $sp, 24(0x18)
    dd468:       29c08066        st.d            $a2, $sp, 32(0x20)
    dd46c:       29c0a067        st.d            $a3, $sp, 40(0x28)
    dd470:       29c0c068        st.d            $a4, $sp, 48(0x30)
    dd474:       29c0e069        st.d            $a5, $sp, 56(0x38)
    dd478:       29c1206b        st.d            $a7, $sp, 72(0x48)
    dd47c:       29c1006a        st.d            $a6, $sp, 64(0x40)
    dd480:       0015008b        move            $a7, $a0
    dd484:       29c0206c        st.d            $t0, $sp, 8(0x8)
    dd488:       001500a4        move            $a0, $a1
    dd48c:       001500c5        move            $a1, $a2
    dd490:       001500e6        move            $a2, $a3
    dd494:       00150107        move            $a3, $a4
    dd498:       00150128        move            $a4, $a5
    dd49c:       00150149        move            $a5, $a6
    dd4a0:       002b0000        syscall         0x0
    dd4a4:       15ffffec        lu12i.w         $t0, -1(0xfffff)
    dd4a8:       68000d84        bltu            $t0, $a0, 12(0xc)       
# dd4b4 <syscall+0x58>
    dd4ac:       02c14063        addi.d          $sp, $sp, 80(0x50)
    dd4b0:       4c000020        jirl            $zero, $ra, 0
    dd4b4:       1a00128c        pcalau12i       $t0, 148(0x94)
    dd4b8:       28d9e18c        ld.d            $t0, $t0, 1656(0x678)
    dd4bc:       0011100d        sub.w           $t1, $zero, $a0
    dd4c0:       02bffc04        addi.w          $a0, $zero, -1(0xfff)
    dd4c4:       3818098d        stx.w           $t1, $t0, $tp
    dd4c8:       02c14063        addi.d          $sp, $sp, 80(0x50)
    dd4cc:       4c000020        jirl            $zero, $ra, 0


>
> On Thu, 2023-03-23 at 16:40 +0800, caiyinyu wrote:
>> + ENTRY (syscall)
>> +        move	t0, a7
>> +        move	a7, a0		/* Syscall number -> a0.  */
>> +        move	a0, a1		/* shift arg1 - arg6.  */
>> +        move	a1, a2
>> +        move	a2, a3
>> +        move	a3, a4
>> +        move	a4, a5
>> +        move	a5, a6
>> +        move	a6, t0		/* arg7 is saved in t0.  */
>> +        syscall 0		/* Do the system call.  */
>> +       lu12i.w t0, -1
> "li.w t0, -4096" will do the same thing, and be more readable.
>
> And this line seems indented with a tab, while other lines are indented
> with 8 spaces.

Fixed


>> +        bltu   t0, a0, L (error)
>> +        ret                     /* Return to caller.  */
> "ret" is not recognized by GNU assembler <= 2.39, it's better to use
> the old-style "jr      ra" for backward compatibility.

Fixed.



>
>> +
>> +L (error):
>> +        b      __syscall_error