Why would system() use 100% CPU inside the fork() syscall?

I have a case where I create an ffmpeg ... command line and execute it with a simple system() call. I use that in my tests and I use that call within a require to make sure that it returns 0 as expected, I also print the command so I can easily see what ran and failed:

std::cerr << "--- command: " << cmd << std::endl;
CATCH_REQUIRE(system(cmd) == 0);  // <-- get stuck inside there

Once in a while, that process locks up, using 100% of one CPU core. Looking at where that happens, I found out that it was in the do_system() function¹ of the C library at the point where it does a syscall. Comparing to the source code, that would be the FORK () call.

#ifdef FORK
  pid = FORK ();
#else
  pid = __fork ();
#endif

The following is the corresponding assembly found with gdb, i.e.

ps -ef | grep process-name
sudo gdb -p <pid>

First I used where to see the stack. The last call was to do_system().

Then I typed disassemble to get this output:

   0x00007ffb9274b129 <+361>:   lea    0x1c(%rsp),%rdx
   0x00007ffb9274b12e <+366>:   xor    %esi,%esi
   0x00007ffb9274b130 <+368>:   mov    $0x100011,%edi
   0x00007ffb9274b135 <+373>:   mov    $0x38,%eax
   0x00007ffb9274b13a <+378>:   syscall 
=> 0x00007ffb9274b13c <+380>:   cmp    $0xfffffffffffff000,%rax
   0x00007ffb9274b142 <+386>:   ja     0x7ffb9274b2d0 <do_system+784>
   0x00007ffb9274b148 <+392>:   test   %eax,%eax
   0x00007ffb9274b14a <+394>:   mov    %eax,%edi
   0x00007ffb9274b14c <+396>:   mov    %eax,0x1c(%rsp)
   0x00007ffb9274b150 <+400>:   je     0x7ffb9274b3c7 <do_system+1031>

First, when I saw that I tried to continue to see what happens. I typed the following:

gdb> stepi

and it went back to 100% CPU usage for one core and never stopped again. Since I was running as root, I'm pretty confident that the stepi would function even inside the C library code.

I could understand some form of lock up in the C library, especially since I use threads (although at the point only the main thread is running, the other threads are all exiting before I call system()), but I have a hard time understanding why it would block inside the kernel fork() syscall.

Anyone encountered such a lock up before who would know why it occurs?

¹ see under sysdeps/posix/system.c (libc version 2.27 under Ubuntu 18.04)