Re: [MIPS PATCH RFA] fix serious problem w/ MIPS .cprestore.

[Thiemo Seufer <ica2_ts at csv dot ica dot uni-stuttgart dot de>]

cgd@broadcom.com wrote:
[snip]
> i think the .cpreturn handling is similarly broken, but it's never
> bit me, I don't know how to use it, and there don't seem to be
> existing tests for .cpreturn in the testsuite so i don't have anything
> to copy!!!

Yes, .cpsetup/.cpreturn has the same problem. I'll fix it when you
have committed this patch.

[snip]
> diff -N testsuite/gas/mips/mips-abi32-pic2.s
> --- /dev/null	1 Jan 1970 00:00:00 -0000
> +++ testsuite/gas/mips/mips-abi32-pic2.s	14 Sep 2002 06:22:13 -0000
> @@ -0,0 +1,89 @@
> +
> +	.text
> +	.ent	func1
> +func1:
> +	.frame $sp,0,$31
> +	.set noreorder
> +	.cpload $25		# 0000 lui	gp,hi(_gp_disp)
> +				# 0004 addiu	gp,gp,lo(_gp_disp)
> +				# 0008 addu	gp,gp,t9
> +	.set reorder
> +	.cprestore 8		# 000c sw	gp,8(sp)
> +
> +	jal	end		# 0048 lw	t9,got(.text)(gp)
> +				# 004c nop
> +				# 0050 addiu	t9,t9,lo(end)
> +				# 0054 jalr	t9
> +				# 0058 nop
> +				# 005c lw	gp,8(sp)
> +
> +	# avoid confusion: without something here, the lw above might be
> +	# put into the delay slot for the jalr below!
> +	nop
> +
> +	jal	end		# 0054 jalr	t9
> +				# 0058 nop
> +				# 005c lw	gp,8(sp)
> +	.end	func1

To avoid confusion: What about adding a proper function epilogue
at the function's end?

> +
> +
> +	.text
> +	.ent	func2
> +func2:
> +	.frame $sp,0,$31
> +	.set noreorder
> +	.cpload $25		# 0000 lui	gp,hi(_gp_disp)
> +				# 0004 addiu	gp,gp,lo(_gp_disp)
> +				# 0008 addu	gp,gp,t9
> +	.set reorder
> +	.cprestore 32768	# 000c sw	gp,8(sp)
                   ^^^^^                           ^
Which one to believe? :-)

> +
> +	jal	end		# 0048 lw	t9,got(.text)(gp)
> +				# 004c nop
> +				# 0050 addiu	t9,t9,lo(end)
> +				# 0054 jalr	t9
> +				# 0058 nop
> +				# 005c lw	gp,8(sp)
                                                   ^
Likewise for the rest below.


Thiemo