Re: [Qemu-devel] [RFC 02/10] use a different translation block list for each cpu.

[Paolo Bonzini <pbonzini@redhat.com>]

On 03/02/2015 17:17, Richard Henderson wrote:
>> > @@ -759,7 +760,9 @@ static void page_flush_tb_1(int level, void **lp)
>> >          PageDesc *pd = *lp;
>> >  
>> >          for (i = 0; i < V_L2_SIZE; ++i) {
>> > -            pd[i].first_tb = NULL;
>> > +            for (j = 0; j < MAX_CPUS; j++) {
>> > +                pd[i].first_tb[j] = NULL;
>> > +            }
>> >              invalidate_page_bitmap(pd + i);
>> >          }
>> >      } else {
> Surely you've got to do some locking somewhere in order to be able to modify
> another thread's cpu tb list.

But that's probably not even necessary.  page_flush_tb_1 is called from
tb_flush, which in turn is only called in very special circumstances.

It should be possible to have something like the kernel's "stop_machine"
that does the following:

1) schedule a callback on all TCG CPU threads

2) wait for all CPUs to have reached that callback

3) do tb_flush on all CPUs, while it knows they are not holding any lock

4) release all TCG CPU threads

With one TCG thread, just use qemu_bh_new (hidden behind a suitable API
of course!).  Once you have multiple TCG CPU threads, loop on all CPUs
with the same run_on_cpu function that KVM uses.

Paolo