Sergey Fedorov <serge.f...@gmail.com> writes: > On 31/03/16 16:37, Alex Bennée wrote: >> Sergey Fedorov <serge.f...@gmail.com> writes: >>> Looks like no matter which approach we use, it's ultimately necessary to >>> ensure all CPUs have exited from translated code before the translation >>> buffer may be safely flushed. >> One approach would be to have multiple translation contexts with their >> own buffers and then you can safely flush TBs if no vCPUs are currently >> executing in those regions. But I suspect that is a much more complex >> future optimisation. > > Yes, this is much more complex and its performance impact should be > investigated. > >> Having said that is it safe to flush TBs from a given page if we know >> no vCPUs are currently executing in that page? As the execution loop has >> to exit the chained TBs as we cross page boundaries we could just keep >> account of which vCPUs are currently in which page. > > It should be safe to invalidate a TB while some other CPU is executing > its translated code. But it should be guaranteed that no CPU execute any > old TB after tb_flush() since we're going to start reusing those TBs. > > I see how TB cannot be patched if it spans two pages, is there any on > when TCG goto_tb can be generated?
Do you mean tcg_gen_goto_tb? AFAIUI all blocks end with goto_tb post-ambles but they should only directly jump to another TB if they are in the same page. -- Alex Bennée
