Re: [Lightning] [BUG] [AMD64 ABI] %r12 not preserved and restored in trampoline function prolog/epilog

[Marc Nieper-Wißkirchen]

P.S.: Here is another version of the patch (diff vs the current GIT
version), which makes R12 into the new JIT_V4.

diff --git a/include/lightning/jit_x86.h b/include/lightning/jit_x86.h
index 3350021..a28e375 100644
--- a/include/lightning/jit_x86.h
+++ b/include/lightning/jit_x86.h
@@ -117,21 +117,21 @@ typedef enum {
 #    define jit_sse_reg_p(reg) ((reg) >= _XMM4 && (reg) <= _XMM0)
 #  else
 #    define jit_r(i)           (_RAX + (i))
-#    define jit_r_num()                4
+#    define jit_r_num()                3
 #    define jit_v(i)           (_RBX + (i))
-#    define jit_v_num()                4
+#    define jit_v_num()                5
 #    define jit_f(index)       (_XMM8 + (index))
 #    define jit_f_num()                8
 #    define JIT_R0             _RAX
 #    define JIT_R1             _R10
 #    define JIT_R2             _R11
-#    define JIT_R3             _R12
-    _RAX,      _R10,   _R11,   _R12,
+    _RAX,      _R10,   _R11,
 #    define JIT_V0             _RBX
 #    define JIT_V1             _R13
 #    define JIT_V2             _R14
 #    define JIT_V3             _R15
-    _RBX,      _R13,   _R14,   _R15,
+#    define JIT_V4             _R12
+    _RBX,      _R13,   _R14,   _R15,   _R12,
     _R9,       _R8,    _RCX,   _RDX,   _RSI,   _RDI,
     _RSP,      _RBP,
 #    define JIT_F0             _XMM8
diff --git a/lib/jit_x86.c b/lib/jit_x86.c
index aae796b..592610f 100644
--- a/lib/jit_x86.c
+++ b/lib/jit_x86.c
@@ -185,11 +185,11 @@ jit_register_t            _rvs[] = {
     { rc(arg) | rc(gpr) | rc(rg8) | 0, "%rax" },
     { rc(gpr) | rc(rg8) | 10,          "%r10" },
     { rc(gpr) | rc(rg8) | 11,          "%r11" },
-    { rc(gpr) | rc(rg8) | 12,          "%r12" },
     { rc(sav) | rc(rg8) | rc(gpr) | 3, "%rbx" },
     { rc(sav) | rc(rg8) | rc(gpr) | 13,        "%r13" },
     { rc(sav) | rc(rg8) | rc(gpr) | 14,        "%r14" },
     { rc(sav) | rc(rg8) | rc(gpr) | 15,        "%r15" },
+    { rc(sav) | rc(rg8) | rc(gpr) | 12,        "%r12" },
     { rc(arg) | rc(rg8) | rc(gpr) | 9, "%r9" },
     { rc(arg) | rc(rg8) | rc(gpr) | 8, "%r8" },
     { rc(arg) | rc(rg8) | rc(gpr) | 1, "%rcx" },

Am Do., 12. Sept. 2019 um 08:38 Uhr schrieb Marc Nieper-Wißkirchen
<address@hidden>:
>
> Hi Paulo,
>
> when creating a function prolog and epilogue for a trampoline function
> as this one
>
> prolog
> frame (...)
> ...
> epilog
>
> GNU lightning saves all callee-save registers in the prolog and
> restores them in the epilog. On the Unix port of x86_64, however, it
> fails to save and restore %r12, which is a callee-saved register.
>
> (This error has led to subtle bugs when I compiled my code with "-O2"
> and gcc began to use %r12 and expected it to be preserved over a call
> into a GNU lightning trampoline.)
>
> The error maybe related to the fact that on GNU lightning, %r12 is
> mapped to JIT_R3, which appears as a caller-save register to the user.
> See also this discussion here:
> https://lists.gnu.org/archive/html/lightning/2017-09/msg00006.html
>
> I'd suggest to apply the patch of this discussion so that %r12 is
> treated like %13, %r14, %15 and will be automatically saved in the
> prolog and restored in the epilog. The side effect will be that JIT_R3
> goes away and JIT_V4 appears.
>
> If you don't want this, the bug has to be fixed differently by
> patching the _prolog and _epilog procedures.
>
> In any case, a fix of this bug is urgent because it is a critical one
> (as non-ABI-conforming code is produced).
>
> Thanks,
>
> Marc