Quoting lib/igt_core.c:
 * - Code blocks with magic control flow are implemented with setjmp() and
 *   longjmp(). This applies to #igt_fixture and #igt_subtest blocks and all the
 *   three variants to finish test: igt_success(), igt_skip() and igt_fail().
 *   Mostly this is of no concern, except when such a control block changes
 *   stack variables defined in the same function as the control block resides.
 *   Any store/load behaviour after a longjmp() is ill-defined for these
 *   variables. Avoid such code.
 *
 *   Quoting the man page for longjmp():
 *
 *   "The values of automatic variables are unspecified after a call to
 *   longjmp() if they meet all the following criteria:"
 *    - "they are local to the function that made the corresponding setjmp() call;
 *    - "their values are changed between the calls to setjmp() and longjmp(); and
 *    - "they are not declared as volatile."

We have quite a few igt_subtest_group that define local variables ...