Thiemo Seufer wrote:
> Hello All,
> this patch changes (hopefully) all instances of instruction
> reading/writing in C code to a type of "unsigned int". Currently it
> is a mixture of "unsigned int" and "unsigned long", seemingly for
> historical reasons.
> For mips, it also adds handling for trap_AfterBreakpoint to
> arch_do_displaced_inst, which gives it at least a chance to work.
> Without that, handle-breakpoint loses the breakpoint set, because
> the trap_AfterBreakpoint case in sigtrap is never triggered. Other
> architectures which seem to have the same bug are ppc and hppa.
> For sparc, it adds in sigill_handler a missing os_flush_icache call
> to the trap_AfterBreakpoint case.
> Tested only on mips, please test this patch also on other architectures,
> I may have missed some bits. If it holds up intend to commit it shortly
> after the next release.
I committed a slightly updated version of this patch, some mips-related
parts went already in before 0.9.5.