Re: [PATCH 1/2] gdbstub: Set current_cpu for memory read write

[Philippe Mathieu-Daudé]

+Thomas

On 22/3/22 16:56, Peter Maydell wrote:
> On Tue, 22 Mar 2022 at 15:43, Bin Meng <bmeng.cn@gmail.com> wrote:
> > When accessing the per-CPU register bank of some devices (e.g.: GIC)
> > from the GDB stub context, a segfault occurs. This is due to current_cpu
> > is not set, as the contect is not a guest CPU.
> >
> > Let's set current_cpu before doing the acutal memory read write.
> >
> > Resolves: https://gitlab.com/qemu-project/qemu/-/issues/124
> > Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
> > ---
>
> This works, but I worry a bit that it might have unexpected
> side effects, and setting globals (even if thread-local) to
> cause side-effects elsewhere isn't ideal...

Yeah, gdbstub is like a JTAG probe, CPU accessors/views shouldn't be
involved. Having current_cpu==NULL seems the correct behavior.

There was a thread few years ago about an issue similar to this one.
IIRC it was about how to have qtest commands select a different address
space instead of the 'current cpu' one.

I wonder why target_memory_rw_debug() involves CPU at all. Maybe it is
simply not using the correct API?