Fix Linux cgroup2 initialization

[janvrany]

This commit fixes a subtle bug in cgroup2 initialization, namely in populateCgroupEntryListV2().

Consider a system with only memory control group enabled. Then the cgroup.controllers for the process's cgroup2 would contain string memory followed by newline (\n). See Linux kernel source, file kernel/cgroup/cgroup.c, function cgroup_print_ss_mask() 1.

The code that iterates over cgroups listed only considers space as a separator:

separator = strchr(cursor, ' ');
if (NULL != separator) {
    *separator = '\0';
}

So the last (an in our example also the first) cgroup name (pointed to by cursor) contains the \n as the end, therefore the strcmp comparison fails:

...
   && (0 == strcmp(cursor, supportedSubsystems[i].name))
...

(since we're comparing 'memory\n' with just 'memory'). In other words, the last controller in controllers file is always "missed" by OMR.

This went unnoticed probably because most linux systems used have more (all?) controller enabled, so likely the last one was some that OMR is not interested in (such as pids, for example). However, if there's only one controller enabled, it is "missed" and OMR falsely reports CGroups2 are unavailable. This actually happened on RISC-V board running custom, stripped down kernel.

This commit fixes the problem by stripping the newline.

[babsingh]

Functionally, LGTM.

Minor formatting nitpicks. The unprocessed markdown syntax in the commit message should be removed.

[janvrany]

Thanks @babsingh! b544b75 should address your comments (hope I did not miss anything),

[babsingh]

jenkins build all

[babsingh]

LGTM. Thanks for fixing this issue.

Just saw your message on OMR slack; was on vacation earlier; otherwise, would have helped.

[janvrany]

Thanks for reviewing!
Just to let you know, RISC-V build will not finish as the build node is dead (we're working on it, but...) I have tested the change locally for RISC-V both using QEMU (as cross job does) and on real hardware.