On 9 September 2013 17:39, Michael Walle <mich...@walle.cc> wrote: > Am 2013-09-04 11:04, schrieb Andreas Färber: >> On part 11, rth said for alpha that using CPUAlphaState in >> DisasContext was bad >> and that the relevant fields should be copied into DisasContext >> instead; a few >> targets still have an env field though and use it for register access and >> cpu_abort(), so I have just changed the type for now and call upon >> maintainers >> to review whether they want to do different changes. > > > Is there a preferred way to abort translation, if for example, the > translator finds an unknown opcode? I guess there should be some kind of > exception. But speaking of lm32, the real CPU, does not handle such a > situation, instead the behavior is just undefined.
If for the real CPU it's genuinely undefined we can do what we like, but killing QEMU is probably a bit harsh. If the CPU supports some kind of plausible exception then we can use it. Otherwise translate as no-op, maybe. In either case doing a qemu_log_mask(LOG_GUEST_ERROR, ...) would be a good plan. For lm32 you might consider raising EXCP_BREAKPOINT; then if you have a gdb connected to the debug stub the developer will get dropped into the debugger for the offending insn and can look at it. -- PMM
