RR crash in PerfCounters::read_ticks

In a multithreaded program, rr record crashes randomly (and passes sometimes too) with following stack trace
Please let me know for any ideas/suggestions?

rr(rr::dump_rr_stack()+0x3b)[0x5f229b]
rr(rr::GdbServer::emergency_debug(rr::Task*)+0x325)[0x504d95]
rr[0x513c64]
rr(rr::PerfCounters::read_ticks(rr::Task*)+0x11d)[0x52f8fd]
rr(rr::Task::did_waitpid(rr::WaitStatus)+0x19f)[0x5d26bf]
rr(rr::Task::wait(double)+0x48f)[0x5d341f]
rr(rr::Scheduler::reschedule(rr::Switchable)+0x14f)[0x5b436f]
rr(rr::RecordSession::record_step()+0xdc)[0x5400ec]
rr(rr::RecordCommand::run(std::vector<std::__cxx11::basic_string<char, std::char_traits, std::allocator >, std::allocator<std::__cxx11::basic_string<char, std::char_traits, std::allocator > > >&)+0x1011)[0x535e21]
rr(main+0x353)[0x4ac743]
/lib64/libc.so.6(__libc_start_main+0xf5)[0x7f4ee24ee6e5]

uname -a:
Linux lxserv01 4.4.73-5-default #1 SMP Tue Jul 4 15:33:39 UTC 2017 (b7ce4e4) x86_64 x86_64 x86_64 GNU/Linux

rr --version
rr version 5.3.0

Start with PerfCounters::read_ticks and its caller Task::did_waitpid, then inspect the recording path through Scheduler::reschedule and RecordSession::record_step. Reproduce the multithreaded crash on the reported Linux 4.4.73 environment and determine a fix that lets rr record complete reliably without the emergency stack dump.