On Thu, Nov 03, 2022, Philippe Mathieu-Daudé wrote:
> Hi Sean,
>
> On 3/11/22 00:18, Sean Christopherson wrote:
> > Move arm/arch specific initialization directly in arm's module_init(),
> > now called kvm_arm_init(), instead of bouncing through kvm_init() to
> > reach kvm_arch_init(). Invoking kvm_arch_init() is the very first action
> > performed by kvm_init(), i.e. this is a glorified nop.
> >
> > Making kvm_arch_init() a nop will allow dropping it entirely once all
> > other architectures follow suit.
> >
> > No functional change intended.
> >
> > Signed-off-by: Sean Christopherson <seanjc@xxxxxxxxxx>
> > ---
> > arch/arm64/kvm/arm.c | 25 ++++++++++++++++---------
> > 1 file changed, 16 insertions(+), 9 deletions(-)
>
> > /* NOP: Compiling as a module not supported */
> > void kvm_arch_exit(void)
> > {
> > - kvm_unregister_perf_callbacks();
>
> Doesn't this belong to the previous patch?
No, but the above changelog is a lie, there is very much a functional change here.
The goal of the previous patch is to fix the error paths in kvm_arch_init(), a.k.a.
kvm_arm_init(). After fixing kvm_arch_init(), there are still bugs in the sequence
as a whole because kvm_arch_exit() doesn't unwind other state, e.g. kvm_arch_exit()
should really look something like:
void kvm_arch_exit(void)
{
teardown_subsystems();
if (!is_kernel_in_hyp_mode())
teardown_hyp_mode();
kvm_arm_vmid_alloc_free();
if (is_protected_kvm_enabled())
???
}
Becuase although the comment "NOP: Compiling as a module not supported" is correct
about KVM ARM always having to be built into the kernel, kvm_arch_exit() can still
be called if a later stage of kvm_init() fails.
But rather than add a patch to fix kvm_arch_exit(), I chose to fix the bug by
moving code out of kvm_arch_init() so that the unwind sequence established in the
previous patch could be reused.
Except I managed to forget those details when writing the changelog. The changelog
should instead be:
KVM: arm64: Do arm/arch initialization without bouncing through kvm_init()
Do arm/arch specific initialization directly in arm's module_init(), now
called kvm_arm_init(), instead of bouncing through kvm_init() to reach
kvm_arch_init(). Invoking kvm_arch_init() is the very first action
performed by kvm_init(), so from a initialization perspective this is a
glorified nop.
Avoiding kvm_arch_init() also fixes a mostly benign bug as kvm_arch_exit()
doesn't properly unwind if a later stage of kvm_init() fails. While the
soon-to-be-deleted comment about compiling as a module being unsupported
is correct, kvm_arch_exit() can still be called by kvm_init() if any step
after the call to kvm_arch_init() succeeds.
Add a FIXME to call out that pKVM initialization isn't unwound if
kvm_init() fails, which is a pre-existing problem inherited from
kvm_arch_exit().
Making kvm_arch_init() a nop will also allow dropping kvm_arch_init() and
kvm_arch_exit() entirely once all other architectures follow suit.
