Am 26.09.2010 19:19, Eddie Kohler wrote: > OK, thanks. I understand how you're relying on the current behavior. > > I'd rather not change all of QEMU and GDB in one step,
The first step is changing gdb anyway. > but I'd like to > address this. QEMU documentation implies, and new users expect, that > debugging uses virtual addresses, not the segmentation-specific "linear > addresses" that are actually used now. > > - How about a maintenance packet type that changed behavior to what I > would prefer (breakpoints and memory access use virtual addresses, not > linear addresses)? > > - We could add a "segment identifier" parameter to > cpu_get_phys_page_debug, ignored on all targets but i386 at first. Then > we could pass information through to cpu_get_phys_page_debug about what > kind of address is being translated. This change could be propagated to > cpu_memory_rw_debug (now or later). Would you object? These changes would establish a temporary interface for an incomplete workaround, and that even with impact on non-x86 code. I would prefer if you could invest your time on the gdb side instead. Anything improved there is not lost - in contrast to the modifications of qemu. I would try to support any effort in this direction. BTW, gdb has nice Python binding these days, so some extensions may also start their life as a helper script. Jan
signature.asc
Description: OpenPGP digital signature
