On 07/07/2015 14:34, Paolo Bonzini wrote:
On 07/07/2015 13:48, Frederic Konrad wrote:
this eventually ends up doing a tb_lock on the find_slow path which IIRC
is when might end up doing the actual code generation.
Up to this point, system emulation is using the BQL for everything. I
guess things change later.
Actually we use tb_lock to protect all the tb related structure such as
TBContext etc.. Is it better to use the global lock for this?
No, on the contrary. But using the BQL is the status as of patch 2, so
it's okay to keep the #ifdefs. Thanks for confirming that it changes
later in the patch.
Paolo
In fact I changed nothing in patch 2 except abstracting out the #ifdef from
spinlock_t and using qemu_mutex (pthread_t on linux) instead of spinlock_t.
The only reason for that is only to use tb_lock for both user and system
mode.
And yes as it's the first patch tb_lock is not used in this step except
in the user
code.
Thanks,
Fred
