[PATCH v1 00/19] Intel TMA 5.1 metrics and event updates

[Namhyung Kim]

On Sat, Jul 26, 2025 at 04:34:00PM -0700, Ian Rogers wrote:
> On Sat, Jul 26, 2025 at 03:34:09PM -0700, Namhyung Kim wrote:
> > On Tue, Jul 22, 2025 at 06:05:38PM -0700, Namhyung Kim wrote:
> > > On Tue, Jul 22, 2025 at 06:14:30PM +0000, Falcon, Thomas wrote:
> > > > On Fri, 2025-07-18 at 20:44 -0700, Ian Rogers wrote:
> > > > > Update events from:
> > > > > v1.31 -> v1.33 for alderlake
> > > > > v1.09 -> v1.12 for arrowlake
> > > > > v1.14 -> v1.16 for emeraldrapids
> > > > > v1.10 -> v1.12 for graniterpaids
> > > > > v1.14 -> v1.17 for lunarlake
> > > > > v1.14 -> v1.16 for meteorlake
> > > > > v1.28 -> v1.30 for sapphirerapids
> > > > > 
> > > > > Update TMA 5.0 to 5.1 removing the slots event workaround as the patch series:
> > > > > https://lore.kernel.org/lkml/20250719030517.1990983-1-irogers@google.com/
> > > > > is assumed.
> > > 
> > > So I'll merge this series after the above patchset.
> > 
> > Now I'm seeing this on my AMD machine.
> > 
> >   Found metric 'tma_info_system_cpu_utilization'                                  
> >   metric expr cpu_atom at CPU_CLK_UNHALTED.REF_TSC@ / msr at tsc\,cpu\=cpu_atom@ for tma_info_system_cpu_utilization
> >   metric expr tma_info_system_cpus_utilized / #num_cpus_online for tma_info_system_cpu_utilization
> >   metric expr cpu_core at CPU_CLK_UNHALTED.REF_TSC@ / msr at tsc\,cpu\=cpu_core@ for tma_info_system_cpus_utilized
> >   found event msr at tsc,cpu=cpu_core@                                               
> >   found event cpu_core at CPU_CLK_UNHALTED.REF_TSC@                                  
> >   Parsing metric events '{msr/tsc,cpu=cpu_core,metric-id=msr!3tsc!0cpu!2cpu_core!3/,cpu_core/CPU_CLK_UNHALTED.REF_TSC,metric-id=cpu_core!3CPU_CLK_UNHALTED.REF_TSC!3/}:W'
> >   event syntax error: '{msr/tsc,cpu=cpu_core,metric-id=msr!3tsc!0cpu!2cpu_core!..'
> >                         \___ Bad event or PMU                                     
> >                                                                                   
> >   Unable to find PMU or event on a PMU of 'msr'                                   
> >                                                                                   
> >   event syntax error: '{msr/tsc,cpu=cpu_core,metric-id=msr!3tsc!0cpu!2cpu_core!..'
> >                                     \___ not a valid PMU or CPU number            
> >                                                                                   
> >   event syntax error: '{msr/tsc,cpu=cpu_core,metric-id=msr!3tsc!0cpu!2cpu_core!..'
> >                                     \___ not a valid PMU or CPU number            
> >                                                                                   
> >   event syntax error: '{msr/tsc,cpu=cpu_core,metric-id=msr!3tsc!0cpu!2cpu_core!..'
> >                                     \___ not a valid PMU or CPU number            
> >                                                                                   
> >   event syntax error: '{msr/tsc,cpu=cpu_core,metric-id=msr!3tsc!0cpu!2cpu_core!..'
> >                                     \___ not a valid PMU or CPU number            
> >   Broken metric tma_info_system_cpu_utilization                                   
> >   ---- end(-1) ----                                                               
> >     9.3: Parsing of PMU event table metrics                            : FAILED! 
> >   
> >   parsing 'tma_info_mem_mix_memload_ratio': '1e3 * cpu_atom at MEM_UOPS_RETIRED.ALL_LOADS@ / cpu_atom at UOPS_RETIRED.ALL@'
> >   parsing 'tma_info_serialization_%_tpause_cycles': '100 * cpu_atom at SERIALIZATION.C01_MS_SCB@ / (5 * cpu_atom at CPU_CLK_UNHALTED.CORE@)'
> >   parsing 'tma_info_system_cpu_utilization': 'cpu_atom at CPU_CLK_UNHALTED.REF_TSC@ / msr at tsc\,cpu\=cpu_atom@'
> >   check_parse_fake failed                                                         
> >   ---- end(-1) ----                                                               
> >     9.4: Parsing of PMU event table metrics with fake PMUs             : FAILED!

> Sorry for this. The parse state says whether fake pmus are used.
> Parse events needs updating for things like get_config_cpu so that
> failing to find a pmu is ignored with fake pmus enabled. Sorry for
> this, it is exposed in the updated metrics but the issue was caused
> by the prior series adding the support in the term.

No problem.  I've removed this series from the tree and we can fix the
issue once you're back.  Then add the metrics later.

Thanks,
Namhyung