Re: [PATCH 0/5] kvm: fix latent guest entry/exit bugs

[Mark Rutland]

On Tue, Jan 18, 2022 at 05:50:51PM +0000, Mark Rutland wrote:
> On Tue, Jan 18, 2022 at 05:09:25PM +0100, Sven Schnelle wrote:
> > I wonder whether the code in irqentry_enter() should call a function
> > is_eqs() instead of is_idle_task(). The default implementation would
> > be just a
> > 
> > #ifndef is_eqs
> > #define is_eqs is_idle_task
> > #endif
> > 
> > and if an architecture has special requirements, it could just define
> > is_eqs() and do the required checks there. This way the architecture
> > could define whether it's a percpu bit, a cpu flag or something else.
> 
> I had come to almost the same approach: I've added an arch_in_rcu_eqs()
> which is checked in addition to the existing is_idle_thread() check.
> 
> In the case of checking is_idle_thread() and checking for PF_VCPU, I'm
> assuming the compiler can merge the loads of current->flags, and there's
> little gain by making this entirely architecture specific, but we can
> always check that and/or reconsider in future.

FWIW, I've pushed out my WIP to:

 https://git.kernel.org/pub/scm/linux/kernel/git/mark/linux.git/log/?h=kvm/entry-rework

... and I intend to clean that up and get it out on the list tomorrow.

The new entry/exit helpers are:

  https://git.kernel.org/pub/scm/linux/kernel/git/mark/linux.git/commit/?h=kvm/entry-rework&id=df292ecabba50145849d8c8888cec9153267b31d

The arch_in_rcu_eqs() bit is:

  https://git.kernel.org/pub/scm/linux/kernel/git/mark/linux.git/commit/?h=kvm/entry-rework&id=6e24c5ed7558ee7a4c95dfe62891dfdc51e6c6c4

The s390 changes are:

  https://git.kernel.org/pub/scm/linux/kernel/git/mark/linux.git/commit/?h=kvm/entry-rework&id=ca8daba1809b6e4f1be425ca93f6373a2ea0af6b

I need to clean up the commit messages (including typos, TODOs, and
deleting some stale gunk), and there are some comments to write, but by
and large I think the structure is about right.

Thanks,
Mark.