Revert "ulimit_stack_small: test multiple times for 64k pages"
This reverts commit 72e44713abe3872468b82199882e9d4c19536106.
seq will segfault in some cases, leading to an unexpected exit code. Shell
arithmetic could be used instead to avoid that, but this test needs to be
revisited entirely, as Steve Beattied pointed out.
In the meantime, revert this such that we don't regress on old cases where
this was succeeding.
Signed-off-by: Thadeu Lima de Souza Cascardo <email address hidden>
ulimit_stack_small: test multiple times for 64k pages
On systems with 64k pages, setting the stack limit to a really small value
may still succeed. If run multiple times, it will eventually fail. Run it
up to 10 times and exit as soon as it fails to identify such situations.
This has been tested on an ARM64 system with 64k pages. While at it, stop
skipping the test on ppc64el, where the page size is also 64k.
On 64k ARM, mmap_min_addr is 32768 by default, which is smaller than page
size. When running the low-mmap test, it won't even try to mmap unless the
testsize is at least page size.
It should, in fact, be aligned to pagesize so that the last page in the low
range is tested.
test-kernel-security: mmap-min-addr misreported on arm 64k page kernels
The kernel is built with a default mmap_min_addr of 32768 (for legacy
reasons, arm userspace have been broken in the past with the 64k value,
though one could argue whether that should be the default for 64K pages
configured kernels).
However, QRT commit d56f3347a1f69c06be638ca6b6917083143ce5a4 tried to deal
with that trying to expect the value to be reported as 65536 and the test
pass as such.
Instead, the sysctl value should still be expected to be 32768, while the
test should run with 65536.
