On 03/09/14 11:27, Alexander Graf wrote: > > > On 02.09.14 09:07, Christian Borntraeger wrote: >> On 02/09/14 00:39, Alexander Graf wrote: >>> >>> >>> On 29.08.14 15:52, Jens Freimann wrote: >>>> From: David Hildenbrand <d...@linux.vnet.ibm.com> >>>> >>>> When reading/writing the psw mask, the condition code may only be touched >>>> if >>>> running on tcg. >>> >>> Why? Shouldn't we be able to set CC from gdb as well? >>> >> >> You can. What this patch does (and the patch description is a bit vague >> here) is to not modify the PSW it gets from KVM when passing it to gdb: >> The qemu core gets the PSW from KVM. Without this patch, we use cc_op to >> calculate the current CC of the PSW (No idea of TCG, I guess its evaluated >> lazy - at least this is how valgrind works). This is wrong for the KVM case, >> as cc_op does not contain any useful data for the KVM case. With this patch >> we simply pass the psw from KVM to gdb and back. >> >> The symptom was that the cc was always shown as zero. > > Ah, I see. I agree with the patch, but the patch description does not > actually describe what the patch does. Please rework it.
The patch was already pulled...Well the description is misleaded but with some interpretion still correct ;-). But your point is well taken. We will try to make sure that all patch description are descriptive and not ambiguous. Your question is a good indication that this was not the case in this patch. Christian > I also wouldn't mind if instead of hard coding this logic in the > gdbstub, we'd extract it as helper functions to read and write the > PSW.MASK in cpu.h.
