Am 30.01.2013 23:48, schrieb Andreas Färber: > As a reminder here's a link to one of my original discussions of the new > types: > https://lists.nongnu.org/archive/html/qemu-devel/2012-05/msg01286.html > > That is, for any non-TCG functions (TCG does not support CPUState yet) an > S390CPU argument should be preferred over CPUS390XState since it allows cheap > access to its own fields, CPUState's via CPU() and to CPUS390XState via ->env. > Doing this consistently avoids costs of casting back and forth unnecessarily. > > s390 code should use s390_env_get_cpu() where needed, not ENV_GET_CPU(). > > As a rule of thumb, any field in include/exec/cpu-defs.h:CPU_COMMON can be > expected to end up in CPUState (or accessible from there) sooner or later.
> Per-target functions can be expected to change to CPUState soon. Maybe too brief: This was referring to functions like kvm_arch_*() that each target implements, knowing its CPU type. In particular do_interrupt() is one of my next candidates. Andreas > > New fields that do not need to be accessed via TCGv or from a hot TCG helper > function should be added to S390CPU, not to CPUS390XState. -- SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg
