On 11/08/2015 11:21, Peter Maydell wrote: > > > I think it's definitely not sufficient. Is user-mode multithread still > > > working today? > > > > For some definition of "working", yes. It's not sufficient, but it's a > > good start. > > > > The main problem with user-mode multithreading is that there is no clear > > lock hierarchy between mmap_lock and tb_lock. But this is not a problem > > for softmmu. > > And also that we don't have a serious design for the locking at all. > I was hoping this would be something that would come out of the > multithreaded-TCG work...
Indeed. Fred made an effort to identify the translate-all.c functions that need tb_lock, and there are only a few that need mmap_lock. I'm now trying to document it so that the patches are actually reviewable... Paolo
