Hi Paolo,
On Thu, Jun 10, 2021 at 11:42 AM Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote:
>
> On 03/06/21 23:14, Jing Zhang wrote:
> > +struct _kvm_stats_header {
> > + __u32 name_size;
> > + __u32 count;
> > + __u32 desc_offset;
> > + __u32 data_offset;
> > +};
> > +
>
> Keeping this struct in sync with kvm_stats_header is a bit messy. If
> you move the id at the end of the header, however, you can use the same
> trick with the zero-sized array that you used for _kvm_stats_desc.
>
Good point. Will do.
> > +struct kvm_vm_stats_data {
> > + unsigned long value[0];
> > +};
> > +
>
> I posted the patch to switch the VM statistics to 64-bit; you can rebase
> on top of it.
Cool!
>
> > +#define KVM_GET_STATS_FD _IOR(KVMIO, 0xcc, struct kvm_stats_header)
>
> This should be _IO(KVMIO, 0xcc) since it does not have an argument.
>
Will correct it.
> > +#define STATS_DESC(stat, type, unit, scale, exp) \
> > + { \
> > + { \
> > + .flags = type | unit | scale, \
> > + .exponent = exp, \
> > + .size = 1 \
> > + }, \
> > + .name = stat, \
>
> Here you can use
>
> type | BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) |
> unit | BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) |
> scale | BUILD_BUG_ON_ZERO(scale & ~KVM_STATS_SCALE_MASK) |
>
> to get a little bit of type checking.
Sure, will do.
>
> Paolo
>
Thanks,
Jing
