Handle gracefully TPMI initilization fail (with missing MCFG tables)

[ppalucki]

In enviornments without access to MCFG tables like unprivileged docker without mapping directories or when running inside VM, MCFG tables are missing and I would expect pcm to collect limited set of metrics but not fail.

Instead afetr adding SRF support (TPMI) with this commit we get this:

Initialization of TPMI that happens in initUncoreObjects,
with following code:

• if (TPMIHandle::getNumInstances() == (size_t)num_sockets)
  calls implicitly:
• PFSInstances::get() to get singleton,
• processprocessDVSEC() -> forAllIntelDevices() to do discovery and then...
• getMCFGRecords() -> PciHandleMM::getMCFGRecords() -> readMCfg()
• and finally openMcfgTable
  which throws "anonymous" exception, when MCFG files aren't not available (files cannot be open):

 if (mcfg_handle < 0)
    {
        throw std::exception();
    }

above exception is not handled anywhere and propagtes up to main routine, resulting with fatal error like this:

docker run -ti --cap-add SYS_ADMIN --cap-add SYS_RAWIO -e PCM_NO_MSR=1 -e PCM_NO_PERF=0 -e PCM_USE_UNCORE_PERF=1 ghcr.io/intel/pcm

output is:

Package thermal spec power: 0 Watt; Package minimum power: 0 Watt; Package maximum power: 0 Watt;

Can't open MCFG table. Check permission of /sys/firmware/acpi/tables/MCFG
Can't open MCFG table. Check permission of /sys/firmware/acpi/tables/MCFG1
Can't open MCFG table. Check permission of /pcm/sys/firmware/acpi/tables/MCFG
Can't open MCFG table. Check permission of /pcm/sys/firmware/acpi/tables/MCFG1
WARNING: enumeration of devices in UncorePMUDiscovery failed
INFO: Linux perf interface to program uncore PMUs is present
INFO: using Linux perf interface to program uncore PMUs because env variable PCM_USE_UNCORE_PERF=1
INFO: Secure Boot detected. Using Linux perf for uncore PMU programming.
Socket 0: 2 memory controllers detected with total number of 6 channels. 3 UPI ports detected. 2 M2M (mesh to memory)/B2CMI blocks detected. 0 HBM M2M blocks detected. 0 EDC/HBM channels detected. 0 Home Agents detected. 3 M3UPI/B2UPI blocks detected.
Socket 1: 2 memory controllers detected with total number of 6 channels. 3 UPI ports detected. 2 M2M (mesh to memory)/B2CMI blocks detected. 0 HBM M2M blocks detected. 0 EDC/HBM channels detected. 0 Home Agents detected. 3 M3UPI/B2UPI blocks detected.
Can't open MCFG table. Check permission of /sys/firmware/acpi/tables/MCFG
Can't open MCFG table. Check permission of /sys/firmware/acpi/tables/MCFG1
Can't open MCFG table. Check permission of /pcm/sys/firmware/acpi/tables/MCFG
Can't open MCFG table. Check permission of /pcm/sys/firmware/acpi/tables/MCFG1
terminate called without an active exception...

and pcm-sensor-server stops.

Last message from output: "terminate called without an active exception..." is not very informative and missleading, because missing access to MCFG tables is not a blocker/crtiical, as like in previously during Uncore discovery (see WARNING: enumeration of devices in UncorePMUDiscovery failed above) - just causing some metrics to be unavailable.

In other words, I would expect pcm/pcm-sensor-server not to fail resulting with TPMI metrics missing. I assume we need catch exception one level upper in initUncoreObjects as proposed in this pull request.

Additionally this pull requests replaces "anonymous" std::exception() with runtime_error() so we will get more detailed warning like this:

When running this PR

# 1) build
docker build . -t pcm-local
# 2) run 
docker run -ti --cap-add SYS_ADMIN --cap-add SYS_RAWIO -e PCM_NO_MSR=1 -e PCM_NO_PERF=0 -e PCM_USE_UNCORE_PERF=1 pcm-local

output is more friendly and doesn't block collection of other metircs:

...
Can't open MCFG table. Check permission of /sys/firmware/acpi/tables/MCFG
Can't open MCFG table. Check permission of /sys/firmware/acpi/tables/MCFG1
Can't open MCFG table. Check permission of /pcm/sys/firmware/acpi/tables/MCFG
Can't open MCFG table. Check permission of /pcm/sys/firmware/acpi/tables/MCFG1
ERROR: Could not initialize TPMI. Uncore frequency metrics will be unavailable. Exception details: cannot open /[pcm]/sys/firmware/acpi/tables/MCFG* files
...
Starting plain HTTP server on http://localhost:9738/

Generally, IMO it would be much helpful in future debuging if we replace "plain" std::exception with std::runtime_error so we when we run with PCM_NO_MAIN_EXCEPTION_HANDLER=1 and then if we forgot to handle exception then we can easily identify source of exception (instead of debugging all places with std::exception).

[rdementi]

merged via #752