On Wed, Feb 5, 2020 at 4:42 AM Sandipan Das sandipan@linux.ibm.com wrote:
Hi,
On 05/02/20 4:03 am, Mina Almasry wrote:
On Tue, Feb 4, 2020 at 12:36 PM Mina Almasry almasrymina@google.com wrote:
So the problem in this log seems to be that this log line is missing: echo Waiting for hugetlb memory to reach size $size.
The way the test works is that it starts a process that writes the hugetlb memory, then it *should* wait until the memory is written, then it should record the cgroup accounting and kill the process. It seems from your log that the wait doesn't happen, so the test continues before the background process has had time to write the memory properly. Essentially wait_for_hugetlb_memory_to_get_written() never gets called in your log.
Can you try this additional attached diff on top of your changes? I attached the diff and pasted the same here, hopefully one works for you: ...
I got my hands on a machine with 16MB default hugepage size and charge_reserved_hugetlb.sh passes now after my changes. Please let me know if you still run into issues.
With your updates, the tests are passing. Ran the tests on a ppc64 system that uses radix MMU (2MB hugepages) and everything passed there as well.
Thanks, please consider reviewing the next iteration of the patch then.
- Sandipan