On 07/25/2014 03:45 PM, Peter Maydell wrote:
On 25 July 2014 08:41, Richard Henderson <r...@twiddle.net> wrote:
On 07/24/2014 09:37 PM, Peter Maydell wrote:
Huh? We already don't emit goto_tb if single-stepping, surely?
(Well, I guess some of the backends might well be broken, but
in that case they probably don't get the other bits of singlestep
support right either...)
Indeed. I noticed this a month or so ago.
Almost all backends check the gdb env->single_step to prevent goto_tb, but
forget about the tcg debugging singlestep.
Oh, we have two flavours of singlestep? That's confusing...
IMHO, CPUState->singlestep_enabled is a cpu execute mode, for emulating
it, an exception should be raised.
But '-singlestep' from command line rules qemu how to generate TBs and
their generated codes. In this situation, a TB only contains one
instruction,
and should be unlinked.
Am I right?
(I'm currently working on the ARMv8 architectural singlestep,
which will make 3 for target-arm.)
thanks
-- PMM
