Re: [Tinycc-devel] Incorrect function call code on ARMv6

[Balazs Kezes]

On 2016-08-11 01:17 +0200, Daniel Glöckner wrote:
> On Wed, Aug 10, 2016 at 06:59:46AM +0100, Balazs Kezes wrote:
> >   90:   e5912000        ldr     r2, [r1]  ; <-- This is the y argument.
> >   94:   e50b2018        str     r2, [fp, #-24]
> >   98:   e5901000        ldr     r1, [r0]
> >   9c:   e2800004        add     r0, r0, #4
> >   a0:   e5902000        ldr     r2, [r0]  ; <-- The clobbering happens here.
> >   a4:   e50b101c        str     r1, [fp, #-28]
> >   a8:   e1a01002        mov     r1, r2
> >   ac:   e51b001c        ldr     r0, [fp, #-28]
> >   b0:   ebfffffe        bl      0 <foo>
>
> I think the error is caused by the "vtop--;" lines in copy_params().
> The former gfunc_call() implementation for ARM rotated the elements
> on the parameter stack to let get_reg() know that the registers are
> still in use. The current implementation drops each element directly
> after it has been moved to the final register.

Thanks for the pointer but looking through the code I'm not so sure
about that. At 0x94 you can see that it actually saves the register to
the stack in order to free up r2. I can see this in the stacktrace when
I put a breakpoint to the output function like this:

(gdb) b o if ind == 0x94
Breakpoint 1 at 0x331b0: file arm-gen.c, line 205.
(gdb) r
Starting program: /home/rlblaster/proj/tcc/tinycc/tcc -c callbug.c

Breakpoint 1, o (i=3842711576) at arm-gen.c:205
205       ind1 = ind + 4;
(gdb) bt
#0  o (i=3842711576) at arm-gen.c:205
#1  0x00034d28 in store (r=2, sv=0xbefff340) at arm-gen.c:704
#2  0x0001def8 in save_reg (r=2) at tccgen.c:526
#3  0x0001e2f0 in save_regs (n=1) at tccgen.c:630
#4  0x0001e8bc in gv (rc=1) at tccgen.c:825
#5  0x0001eca4 in lexpand () at tccgen.c:951
#6  0x0001ee60 in lexpand_nr () at tccgen.c:980
#7  0x00035d64 in copy_params (nb_args=2, plan=0xbefff4a4, todo=0) at 
arm-gen.c:1098
#8  0x000360f4 in gfunc_call (nb_args=2) at arm-gen.c:1159
...

I think it's just that that copy_params() never restores the spilled
registers. Maybe it needs some extra code at the end to see if any
parameters have been spilled to stack and then restore them?

-- 
Balazs