[PATCH v3 03/14] KVM: selftests: Return a value from vcpu_get_reg() instead of using an out-param

[Sean Christopherson]

On Fri, Nov 01, 2024, Oliver Upton wrote:
> Hey,
> 
> On Fri, Nov 01, 2024 at 07:48:00AM -0700, Sean Christopherson wrote:
> > On Fri, Nov 01, 2024, Mark Brown wrote:
> > > On Wed, Oct 09, 2024 at 08:49:42AM -0700, Sean Christopherson wrote:
> > > > Return a uint64_t from vcpu_get_reg() instead of having the caller provide
> > > > a pointer to storage, as none of the vcpu_get_reg() usage in KVM selftests
> > > > accesses a register larger than 64 bits, and vcpu_set_reg() only accepts a
> > > > 64-bit value.  If a use case comes along that needs to get a register that
> > > > is larger than 64 bits, then a utility can be added to assert success and
> > > > take a void pointer, but until then, forcing an out param yields ugly code
> > > > and prevents feeding the output of vcpu_get_reg() into vcpu_set_reg().
> > > 
> > > This commit, which is in today's -next as 5c6c7b71a45c9c, breaks the
> > > build on arm64:
> > > 
> > > aarch64/psci_test.c: In function ‘host_test_system_off2’:
> > > aarch64/psci_test.c:247:9: error: too many arguments to function ‘vcpu_get_reg’
> > >   247 |         vcpu_get_reg(target, KVM_REG_ARM_PSCI_VERSION, &psci_version);
> > >       |         ^~~~~~~~~~~~
> > > In file included from aarch64/psci_test.c:18:
> > > include/kvm_util.h:705:24: note: declared here
> > >   705 | static inline uint64_t vcpu_get_reg(struct kvm_vcpu *vcpu, uint64_t id)
> > >       |                        ^~~~~~~~~~~~
> > > At top level:
> > > cc1: note: unrecognized command-line option ‘-Wno-gnu-variable-sized-type-not-at
> > > -end’ may have been intended to silence earlier diagnostics
> > > 
> > > since the updates done to that file did not take account of 72be5aa6be4
> > > ("KVM: selftests: Add test for PSCI SYSTEM_OFF2") which has been merged
> > > in the kvm-arm64 tree.
> > 
> > Bugger.  In hindsight, it's obvious that of course arch selftests would add usage
> > of vcpu_get_reg().
> > 
> > Unless someone has a better idea, I'll drop the series from kvm-x86, post a new
> > version that applies on linux-next, and then re-apply the series just before the
> > v6.13 merge window (rinse and repeat as needed if more vcpu_get_reg() users come
> > along).
> 
> Can you instead just push out a topic branch and let the affected
> maintainers deal with it? This is the usual way we handle conflicts
> between trees...

That'd work too, but as you note below, doing that now throws a wrench in things
because essentially all arch maintainers would need merge that topic branch,
otherwise linux-next would end up in the same state.

> > That would be a good oppurtunity to do the $(ARCH) directory switch[*] too, e.g.
> > have a "selftests_late" or whatever topic branch.
> 
> The right time to do KVM-wide changes (even selftests) is *early* in the
> development cycle, not last minute. It gives us plenty of time to iron out
> the wrinkles.

Yeah, that was the original plan, then the stupid strict aliasing bug happened,
and I honestly forgot the vcpu_get_reg() changes would need to be consumed by
other architectures.

Other than letting me forget about this mess a few weeks earlier, there's no
good reason to force this into 6.13.  So, I'll drop the series from 6.13, post
new versions of the this and the $(ARCH) series just before the merge window,
and then either send a pull request to Paolo for 6.14 as soon as the 6.13 merge
window closes, or ask/bribe Paolo to apply everything directly.