RE: [PATCH 0/3] xenoprof: XSA-313 follow-up

[Paul Durrant <xadimgnik@xxxxxxxxx>]

> -----Original Message-----
> From: Julien Grall <julien@xxxxxxx>
> Sent: 20 April 2020 15:15
> To: paul@xxxxxxx; 'Jan Beulich' <jbeulich@xxxxxxxx>; 
> xen-devel@xxxxxxxxxxxxxxxxxxxx
> Cc: 'Andrew Cooper' <andrew.cooper3@xxxxxxxxxx>; 'George Dunlap' 
> <george.dunlap@xxxxxxxxxx>; 'Ian
> Jackson' <ian.jackson@xxxxxxxxxxxxx>; 'Stefano Stabellini' 
> <sstabellini@xxxxxxxxxx>; 'Wei Liu'
> <wl@xxxxxxx>
> Subject: Re: [PATCH 0/3] xenoprof: XSA-313 follow-up
> 
> Hi Paul,
> 
> On 15/04/2020 09:50, Paul Durrant wrote:
> >> -----Original Message-----
> >> From: Jan Beulich <jbeulich@xxxxxxxx>
> >> Sent: 15 April 2020 09:45
> >> To: xen-devel@xxxxxxxxxxxxxxxxxxxx
> >> Cc: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>; George Dunlap 
> >> <george.dunlap@xxxxxxxxxx>; Ian
> Jackson
> >> <ian.jackson@xxxxxxxxxxxxx>; Julien Grall <julien@xxxxxxx>; Stefano 
> >> Stabellini
> >> <sstabellini@xxxxxxxxxx>; Wei Liu <wl@xxxxxxx>; Paul Durrant <paul@xxxxxxx>
> >> Subject: [PATCH 0/3] xenoprof: XSA-313 follow-up
> >>
> >> Patch 1 was considered to become part of the XSA, but it was then
> >> decided against. The other two are a little bit of cleanup, albeit
> >> there's certainly far more room for tidying. Yet then again Paul,
> >> in his mail from Mar 13, was asking whether we shouldn't drop
> >> xenoprof altogether, at which point cleaning up the code would be
> >> wasted effort.
> >>
> >
> > That's still my opinion. This is a large chunk of (only passively 
> > maintained) code which I think is
> of very limited value (since it relates to an old tool, and it only works for 
> PV domains).
> 
> While there are no active user we are aware of, this is an example on
> how to implement a profiler backend with Xen. So I would agree with
> Andrew here.
> 
> IIRC, the reason behind your request is it makes difficult for your
> xenheap work. Am I correct? If so, do you have a thread explaining the
> issues?

After shared info and grant table, it is the only other occurrence of a xenheap 
page shared with a non-system domain. Also, it cannot be trivially replaced 
with an 'extra' domheap page because its assignment changes. Thus a whole bunch 
of cleanup work that I was hoping to do (largely in domain_relinquish_resources 
and free_domheap_pages) is either ruled out, or would have to special-case this 
type of page.
Also, I am unconvinced that PV guests are sufficiently common these days (apart 
from dom0) that profiling them is of any real use.

  Paul