Re: [PATCH RFC 03/15] KVM: Add build-time error check on kvm_run size

[Sean Christopherson]

On Mon, Dec 02, 2019 at 05:40:34PM -0500, Peter Xu wrote:
> On Mon, Dec 02, 2019 at 02:19:49PM -0800, Sean Christopherson wrote:
> > On Mon, Dec 02, 2019 at 03:53:15PM -0500, Peter Xu wrote:
> > > On Mon, Dec 02, 2019 at 11:30:27AM -0800, Sean Christopherson wrote:
> > > > On Fri, Nov 29, 2019 at 04:34:53PM -0500, Peter Xu wrote:
> > > > > It's already going to reach 2400 Bytes (which is over half of page
> > > > > size on 4K page archs), so maybe it's good to have this build-time
> > > > > check in case it overflows when adding new fields.
> > > > 
> > > > Please explain why exceeding PAGE_SIZE is a bad thing.  I realize it's
> > > > almost absurdly obvious when looking at the code, but a) the patch itself
> > > > does not provide that context and b) the changelog should hold up on its
> > > > own,
> > > 
> > > Right, I'll enhance the commit message.
> > > 
> > > > e.g. in a mostly hypothetical case where the allocation of vcpu->run
> > > > were changed to something else.
> > > 
> > > And that's why I added BUILD_BUG_ON right beneath that allocation. :)
> > 
> > My point is that if the allocation were changed to no longer be a
> > straightforward alloc_page() then someone reading the combined code would
> > have no idea why the BUILD_BUG_ON() exists.  It's a bit ridiculous for
> > this case because the specific constraints of vcpu->run make it highly
> > unlikely to use anything else, but that's beside the point.
> > 
> > > It's just a helper for developers when adding new kvm_run fields, not
> > > a risk for anyone who wants to start allocating more pages for it.
> > 
> > But by adding a BUILD_BUG_ON without explaining *why*, you're placing an
> > extra burden on someone that wants to increase the size of kvm->run, e.g.
> > it's not at all obvious from the changelog whether this patch is adding
> > the BUILD_BUG_ON purely because the code allocates memory for vcpu->run
> > via alloc_page(), or if there is some fundamental aspect of vcpu->run that
> > requires it to never span multiple pages.
> 
> How about I add a comment above it?
> 
>   /*
>    * Currently kvm_run only uses one physical page.  Warn the develper
>    * if kvm_run accidentaly grows more than that.
>    */
>   BUILD_BUG_ON(...);

No need for a comment, adding a blurb in the changelog is sufficient.

The lengthy response was just trying to explain why it's helpful to
explicitly justify a change that may seem obvious in the current codebase.
Apologies if it only confused things.