[Qemu-devel] s390 tcg breakage

[Qemu-devel] s390 tcg breakage

[Alexander Graf]

Hi Richard,

I've finally gotten around to debug why TCG on s390 hosts breaks for me. The reason turned out to be quite simple. The broken TB ends with a goto_tb instruction:

   exit_tb $0x3fffb258010

which gets translated into:

   0x90000040:	lgfi	%r2,-81428464
   0x90000046:	iihl	%r2,1023
   0x9000004a:	jg	0x91fffc10

which again leaves r2 as:

  0xffff03fffb258010

This is because lgfi leaves the register negative, but iihl doesn't replace the upper 16 bits, only [48..32].

The patch below fixes it for me, but I'm sure you can come up with something cleaner, better and working on non-ext-imm machines :).


Alex

diff --git a/tcg/s390/tcg-target.c b/tcg/s390/tcg-target.c
index e12a152..6b72ff4 100644
--- a/tcg/s390/tcg-target.c
+++ b/tcg/s390/tcg-target.c
@@ -775,13 +775,7 @@ static void tcg_out_movi(TCGContext *s, TCGType type,
     /* Insert data into the high 32-bits.  */
     uval = uval >> 31 >> 1;
     if (facilities & FACILITY_EXT_IMM) {
-        if (uval < 0x10000) {
-            tcg_out_insn(s, RI, IIHL, ret, uval);
-        } else if ((uval & 0xffff) == 0) {
-            tcg_out_insn(s, RI, IIHH, ret, uval >> 16);
-        } else {
-            tcg_out_insn(s, RIL, IIHF, ret, uval);
-        }
+        tcg_out_insn(s, RIL, IIHF, ret, uval);
     } else {
         if (uval & 0xffff) {
             tcg_out_insn(s, RI, IIHL, ret, uval);

Re: [Qemu-devel] s390 tcg breakage

[Richard Henderson]

On 2013-03-25 06:19, Alexander Graf wrote:
> This is because lgfi leaves the register negative, but iihl doesn't replace the upper 16 bits, only [48..32].
>
> The patch below fixes it for me, but I'm sure you can come up with something cleaner, better and working on non-ext-imm machines :).

Oops, yes.  I've posted a patch that addresses the problem for real.


r~