[8u] RFR: 8149548: slowdebug build crashes if -XX:+TraceBytecodes flag is used

[Edward Nevill]

Hi Sergey,

I was trying to forward port this patch to the jdk9 tree, but I get SEGVs on the slowdebug build even when just running 'java' on its own.

The problem seems to be the following part of the patch

     // Load *adr into c_rarg1, may fault.
     *fault_pc = __ pc();
+    __ mov(c_rarg2, c_rarg0);
     switch (size) {
       case 4:

The resultant code after patching looks like

    // Load *adr into c_rarg1, may fault.
    *fault_pc = __ pc();
    __ mov(c_rarg2, c_rarg0);
    switch (size) {
      case 4:
        // int32_t
        __ ldr(c_rarg0, Address(c_rarg2, 0));
        break;

The problem seems to be that fault_pc is pointing to the mov c_rarg2, c_rarg0 rather than the actual instruction causing the fault.

What I don't quite understand is why we don't see the fault on the jdk8u tree, only the jdk9 tree. Maybe jdk8u isn't calling safefetch?

Anyway, if you agree that the correct fix is just to swap the lines I will push a patch.

    *fault_pc = __ pc();
    __ mov(c_rarg2, c_rarg0);

to

    __ mov(c_rarg2, c_rarg0);
    *fault_pc = __ pc();

Regards,
Ed.

On Wed, 2016-02-10 at 15:59 +0000, Sergey Nazarkin wrote:
> Hi!
> 
> Please review fix for JDK8 interpreter.  The changeset covers several SIGILL and NPE generated by signal handler
> 
> To reproduce call “java -XX:+TraceBytecodes -version”
> 
> http://cr.openjdk.java.net/~snazarki/8149548/
> 
> Sergey Nazarkin
> 
> 
> 
>