malc wrote:
> Revision: 5620
>           http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=5620
> Author:   malc
> Date:     2008-11-04 14:18:13 +0000 (Tue, 04 Nov 2008)
> 
> Log Message:
> -----------
> Add safety net against potential infinite loop
> 
> cpu_interrupt might be called while translating the TB, but before it
> is linked into a potentially infinite loop and becomes env->current_tb.
> 
> Currently this can (and does) cause huge problems only when using
> dyntick clock, with other (periodic) clocks host_alarm_handler will
> eventually be executed resulting in a call to cpu_interrupt which will
> reset the recursion of running TB and the damage is "only" latency.
> 
> Modified Paths:
> --------------
>     trunk/cpu-exec.c
> 
> Modified: trunk/cpu-exec.c
> ===================================================================
> --- trunk/cpu-exec.c	2008-11-04 13:17:17 UTC (rev 5619)
> +++ trunk/cpu-exec.c	2008-11-04 14:18:13 UTC (rev 5620)
> @@ -623,6 +623,14 @@
>                  }
>                  spin_unlock(&tb_lock);
>                  env->current_tb = tb;
> +
> +                /* cpu_interrupt might be called while translating the
> +                   TB, but before it is linked into a potentially
> +                   infinite loop and becomes env->current_tb. Avoid
> +                   starting execution if there is a pending interrupt. */
> +                if (unlikely (env->interrupt_request & CPU_INTERRUPT_EXIT))
> +                    env->current_tb = NULL;
> +
>                  while (env->current_tb) {
>                      tc_ptr = tb->tc_ptr;
>                  /* execute the generated code */

This somehow breaks single-stepping via gdbstub. Unless you have an
instant idea (or even fix), I will dig for the reasons.

Jan