[PATCH v17 0/4] add support for self monitoring

[Thomas Monjalon]

06/06/2025 18:27, Tomasz Duszynski:
> Hi Thomas,
> 
> >Hello,
> >
> >17/01/2025 10:00, Tomasz Duszynski:
> >>  lib/pmu/meson.build                      |  22 +
> >>  lib/pmu/pmu_arm64.c                      |  94 ++++
> >>  lib/pmu/pmu_private.h                    |  32 ++
> >>  lib/pmu/rte_pmu.c                        | 537 +++++++++++++++++++++++
> >>  lib/pmu/rte_pmu.h                        | 251 +++++++++++
> >>  lib/pmu/rte_pmu_pmc_arm64.h              |  30 ++
> >>  lib/pmu/rte_pmu_pmc_x86_64.h             |  24 +
> >
> >Reading it again, I wonder why it is a separate library.
> >In general we give access to the hardware in EAL.
> >Why not having PMU in EAL?
> 
> I initially started PMU as something tightly coupled with EAL, but after
> feedback from other developers I decided to decouple it completely [1].
> 
> The Technical Board also discussed this and supported the idea of having
> PMU as a separate library [2].
> 
> The main goal is to keep EAL lean and focused only on core essentials,
> making the PMU code easier to manage separately. This separation aligns
> nicely with the broader effort to simplify EAL by moving optional
> features out.
> 
> [1] https://lore.kernel.org/dpdk-dev/20230111210547.GA2134@linuxonhyperv3.guj3yctzbm1etfxqx2vob5hsef.xx.internal.cloudapp.net/
> [2] https://mails.dpdk.org/archives/dev/2023-January/260035.html#:~:text=PMU%20Library%20,principle%20for%20inclusion%20in%20DPDK

OK I forgot this history.
Let's go with a separate library.

I'm trying to merge it.
I'm moving to a new release notes.
I did a bit of cleanup.
I've replaced __builtin_ffsll with rte_ffs64.
I've exported symbols with the new macro RTE_EXPORT_EXPERIMENTAL_SYMBOL.
Now I hit another kind of issue:
	error: "__rte_weak" is deprecated
I fix this one with using flags like RTE_PMU_SUPPORTED.
(note: it replaces RTE_LIB_PMU which was not defined)

And at the end I get this error for tracing:

rte_eal_trace.h:136:37: error: passing argument 1 of 'rte_pmu_read' makes integer from pointer without a cast [-Wint-conversion]
  136 |         uint64_t val = rte_pmu_read(index);
      |                                     ^~~~~
      |                                     |
      |                                     char * (*)(const char *, int)

I think integration in tracing is too much risky at this stage anyway,
so I'll postpone it.

I'll post my rebased version of this last patch for follow-up.