2018-12-26 09:54+0800, Wei Yang: > In generic code kvm_setup_async_pf() checks the async_pf size by: > > if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU) > > While in x86, we define the array as: > > gfn_t gfns[roundup_pow_of_two(ASYNC_PF_PER_VCPU)]; > > And each time we compare the index like this: > > i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) > > The behavior is correct now, while not consistent with generic code. > > As for now ASYNC_PF_PER_VCPU is defined as 64, which is already a power of > 2. This patch introduces ASYNC_PF_PER_VCPU_ORDER and rewrite the definition > as: > > #define ASYNC_PF_PER_VCPU_ORDER 6 > #define ASYNC_PF_PER_VCPU (1 << ASYNC_PF_PER_VCPU_ORDER) > > By doing so, we can remove the roundup_pow_of_two()/order_base_2() helpers > when access async_pf on x86. And this make code consistent between > generic and x86. > > Signed-off-by: Wei Yang <richardw.yang@xxxxxxxxxxxxxxx> > > --- > v2: Radim suggest to change platform code instead of generic code. > --- Queued, thanks.
