libk  Diff

Differences From Artifact [3fa83c18de]:

• File mod/kcore/syscall.fn.x86.lin.64.s — part of check-in [481509e134] at 2019-10-21 04:28:46 on branch trunk — don't clobber callee-saved registers (user: lexi, size: 2716) [annotate] [blame] [check-ins using]

To Artifact [a731f1dad0]:

• File mod/kcore/syscall.fn.x86.lin.64.s — part of check-in [6022ec08ab] at 2020-01-30 07:13:19 on branch trunk — Push r15 in linux x64 syscall This was causing the static, release, build of kcli.testbin to subtly segfault shortly after writing the first buffer. Still not sure why kcore.testbin seemingly wasn't affected, nor why the shared build still breaks worse. (user: glowpelt, size: 2740) [annotate] [blame] [check-ins using]

    34     34   	; arg 1 = errno ptr
    35     35   	; arg 2 = syscall num
    36     36   	; arg 3 = valency
    37     37   	; arg 4 = args ptr
    38     38   
    39         -	; we use two registers that are supposed
           39  +	; we use three registers that are supposed
    40     40   	; to be callee-saved, so we need to
    41     41   	; push them to the stack and then pop
    42     42   	; them back off just before the fn rets.
    43     43   	push rbx
    44     44   	push r12
           45  +	push r15
    45     46   
    46     47   	; store the locals in registers that
    47     48   	; are guaranteed not to be clobbered,
    48     49   	; saving us some cycles pushing to
    49     50   	; and popping back from the stack
................................................................................
   101    102   	.error: neg sys.reg.ret
   102    103   	        mov qword [rbx], -1
   103    104   			mov       [r12], sys.reg.ret
   104    105   			jmp .return
   105    106   
   106         -	.return: pop r12
          107  +	.return: pop r15 
          108  +		 pop r12
   107    109   	         pop rbx
   108    110   			 ret