Re: [PATCH] x86: Add trace points to (nearly) all vectors

[Andi Kleen]

On Thu, Mar 12, 2020 at 04:39:04PM -0700, Andy Lutomirski wrote:
> 
> > On Mar 12, 2020, at 4:19 PM, Andi Kleen <andi@xxxxxxxxxxxxxx> wrote:
> > 
> > ïFrom: Andi Kleen <ak@xxxxxxxxxxxxxxx>
> > 
> > In some scenarios it can be useful to count or trace every kernel
> > entry.
> 
> Can you elaborate?  What problem does this solve?

So that we know how often or where kernel entries happen,
how long every kernel execution is, and also can do an accurate break
down.

> 
> > Most entry paths are covered by trace points already,
> > but some of the more obscure entry points do not have
> > trace points.
> > 
> > The most common uncovered one was KVM async page fault.
> 
> NAK.  This path is going away. 

Okay. Can you provide details? When will it go away?

Anyways KVM async is just one case, my patch covers lots of other cases
too. Even if KVM async goes away it still makes sense
for the other entry points and can be easily rebased (just drop
that hunk)

-Andi