Re: [PATCH v11 17/31] x86/resctrl: Find and enable usable telemetry events

["Luck, Tony" <tony.luck@intel.com>]

On Fri, Oct 03, 2025 at 04:52:01PM -0700, Reinette Chatre wrote:
> Hi Tony,
> 
> On 9/25/25 1:03 PM, Tony Luck wrote:
> > The INTEL_PMT_TELEMETRY driver provides telemetry region structures of the
> > types requested by resctrl.
> > 
> > Scan these structures to discover which pass sanity checks to derive
> > a list of valid regions:
> 
> The "to derive a list of valid regions" does not align with the
> "At least one region passes the above checks" requirement. If this is about
> valid (usable?) regions then I think (4) should be dropped. If this is instead about
> valid events then above should be reworded to say that instead.

Will drop "4".
> 
> > 
> > 1) They have guid known to resctrl.
> > 2) They have a valid package ID.
> > 3) The enumerated size of the MMIO region matches the expected
> >    value from the XML description file.
> > 4) At least one region passes the above checks.
> > 
> 
> Everything below is clear by looking at the patch. It can also be seen from patch
> that enabling is done only once if there is *any* valid region instead of "for each
> valid region". One thing that may be useful to add is "why" all events
> can be enabled. If I understand correctly it can be something like:
> 
> 	Enable events that usable telemetry regions are responsible for.

Looks better. I will use this.

> 
> > For each valid region enable all the events in the associated
> > event_group::evts[].
> > 
> > Pass a pointer to the pmt_event structure of the event within the struct
> > event_group that resctrl stores in mon_evt::arch_priv. resctrl passes
> > this pointer back when asking to read the event data which enables the
> > data to be found in MMIO.
> > 
> > Signed-off-by: Tony Luck <tony.luck@intel.com>
> > ---
> >  arch/x86/kernel/cpu/resctrl/intel_aet.c | 38 +++++++++++++++++++++++--
> >  1 file changed, 36 insertions(+), 2 deletions(-)
> > 
> > diff --git a/arch/x86/kernel/cpu/resctrl/intel_aet.c b/arch/x86/kernel/cpu/resctrl/intel_aet.c
> > index f9b5f6cd08f8..98ba9ba05ee5 100644
> > --- a/arch/x86/kernel/cpu/resctrl/intel_aet.c
> > +++ b/arch/x86/kernel/cpu/resctrl/intel_aet.c
> > @@ -20,9 +20,11 @@
> >  #include <linux/intel_pmt_features.h>
> >  #include <linux/intel_vsec.h>
> >  #include <linux/overflow.h>
> > +#include <linux/printk.h>
> >  #include <linux/resctrl.h>
> >  #include <linux/resctrl_types.h>
> >  #include <linux/stddef.h>
> > +#include <linux/topology.h>
> >  #include <linux/types.h>
> >  
> >  #include "internal.h"
> > @@ -114,12 +116,44 @@ static struct event_group *known_perf_event_groups[] = {
> >  	for (_peg = (_grp); _peg < &_grp[ARRAY_SIZE(_grp)]; _peg++)	\
> >  		if ((*_peg)->pfg)
> >  
> > -/* Stub for now */
> > -static bool enable_events(struct event_group *e, struct pmt_feature_group *p)
> > +static bool skip_telem_region(struct telemetry_region *tr, struct event_group *e)
> >  {
> > +	if (tr->guid != e->guid)
> > +		return true;
> > +	if (tr->plat_info.package_id >= topology_max_packages()) {
> > +		pr_warn("Bad package %u in guid 0x%x\n", tr->plat_info.package_id,
> > +			tr->guid);
> > +		return true;
> > +	}
> 
> I have not encountered any mention of the possibility that packages may differ
> in which telemetry region types they support. For example, could it be possible for package
> A to have usable regions of the PERF type but package B doesn't? From what I can tell
> INTEL_PMT_TELEMETRY supports layouts where this can be possible. If I understand correctly
> this implementation will create event files for these domains but when the user attempts to
> read the data it will fail. Can this work add some snippet about possibility of this
> scenario and if/how it is supported?

Yes, this is architecturally possible. But I do not expect that systems will
be built that do this. You are right that such a system will create files that
always return "Unavailable" when read.

Is it sufficient to document this in the commit message?

I don't feel that it would be worthwhile to suppress creation of these files for
a "can't happen" situation. I'm not sure that doing so would be significantly
better from a user interface perspective. Users would get slightly more notice
(-ENOENT when trying to open the file). But the code would require
architecture calls from file system code to check which files need to be created
separately for each domain.

> 
> > +	if (tr->size != e->mmio_size) {
> > +		pr_warn("MMIO space wrong size (%zu bytes) for guid 0x%x. Expected %zu bytes.\n",
> > +			tr->size, e->guid, e->mmio_size);
> > +		return true;
> > +	}
> > +
> >  	return false;
> >  }
> >  
> > +static bool enable_events(struct event_group *e, struct pmt_feature_group *p)
> > +{
> > +	bool usable_events = false;
> > +
> > +	for (int i = 0; i < p->count; i++) {
> > +		if (skip_telem_region(&p->regions[i], e))
> > +			continue;
> > +		usable_events = true;
> 
> A previous concern [1] was why this loop does not break out at this point. I think it will 
> help to make this clear if marking a telemetry region as unusable (mark_telem_region_unusable())
> is done in this patch. Doing so makes the "usable" and "unusable" distinction in one
> patch while making clear that the loop needs to complete.

Ok. I'll pull mark_telem_region_unusable() into this patch.

> > +	}
> > +
> > +	if (!usable_events)
> > +		return false;
> > +
> > +	for (int j = 0; j < e->num_events; j++)
> > +		resctrl_enable_mon_event(e->evts[j].id, true,
> > +					 e->evts[j].bin_bits, &e->evts[j]);
> > +
> > +	return true;
> > +}
> > +
> >  DEFINE_FREE(intel_pmt_put_feature_group, struct pmt_feature_group *,
> >  		if (!IS_ERR_OR_NULL(_T))
> >  			intel_pmt_put_feature_group(_T))
> 
> Reinette
> 
> [1] https://lore.kernel.org/lkml/9ac43e78-8955-db5d-61be-e08008e41f0d@linux.intel.com/

-Tony