Re: [Xen-devel] [PATCH v2 3/7] x86: detect and initialize Intel CAT feature

[Dario Faggioli <dario.faggioli@xxxxxxxxxx>]

On Thu, 2015-03-19 at 13:35 +0000, Jan Beulich wrote:
> >>> On 19.03.15 at 13:44, <dario.faggioli@xxxxxxxxxx> wrote:
> > On Thu, 2015-03-19 at 18:41 +0800, Chao Peng wrote:
> > 
> >> diff --git a/docs/misc/xen-command-line.markdown 
> > b/docs/misc/xen-command-line.markdown
> >> index 63871cb..768c55f 100644
> >> --- a/docs/misc/xen-command-line.markdown
> >> +++ b/docs/misc/xen-command-line.markdown
> >> @@ -1090,9 +1090,9 @@ This option can be specified more than once (up to 8 
> > times at present).
> >>  > `= <integer>`
> >>  
> >>  ### psr (Intel)
> >> -> `= List of ( cmt:<boolean> | rmid_max:<integer> )`
> >> +> `= List of ( cmt:<boolean> | rmid_max:<integer> | cat:<boolean> | 
> > num_sockets:<interger> )`
> >>
> > As Tim, I also don't like the 'num_sockets' part of the interface.
> > 
> > Actually, I'm not sure I understand what the idea behind this parameter
> > is. You say:
> >  
> >> +* Cache Alllocation Technology (Broadwell and later).  Information 
> >> regarding
> >> +  the cache allocation.
> >> +  * `cat` instructs Xen to enable/disable Cache Allocation Technology.
> >> +  * `num_sockets` indicates the number of available sockets for CAT 
> >> feature
> >> +    detection. All the sockets up to num_sockets will be checked for CAT
> >> +    feature. 
> >>
> > What it the use case of specifying, say, num_sockets=2, on a 4 socket
> > system? 
> > 
> > Is it for making it possible for the user to disable the feature on some
> > of the socket? If yes, I think it does make sense, but then why always
> > the last ones? shouldn't this be a list of socket  IDs?
> 
> I had asked about this before, and the answer was that this is to
> _extend_ the range of sockets, to cover the CPU hotplug case. 
>
Ah... yes, hotplug is mentioned in the text.

> But
> of course I could see ways to at least default the count to something
> reasonable even taking hotplug into account (like taking the boot
> time CPUs-per-socket value and dividing nr_cpu_ids by that value).
> 
Indeed. I'm not that much a user of hotplug, but AFAICT, what's cool of
it is that it is dynamic, and having to define this parameter upfront
and/or then being limited by it is certainly suboptimal.

Dario

Attachment: signature.asc
Description: This is a digitally signed message part

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel