On Wed, Mar 15, 2023 at 6:47 PM Alexander Potapenko <glider@xxxxxxxxxx> wrote:
On Wed, Mar 15, 2023 at 6:22 PM Emanuel Attila Czirai <corre.a.buscar@xxxxxxxxx> wrote:On Wed, Mar 15, 2023 at 2:58 PM Alexander Potapenko <glider@xxxxxxxxxx> wrote:Hi Emmanuel,hello, thank you for looking into this.Can you please share the kernel config where the problem is reproducible, along with the full dmesg?attached bothOk, so according to your dmesg, heap initialization is disabled:mem auto-init: stack:all(zero), heap alloc:off, heap free:off, which causes the test failures.
Looks like I've set page_poison=1 which is why I've also set init_on_free=0 init_on_alloc=0 otherwise they're considered to be =0 due to the former being 1, as per:
[ 22.533779] mem auto-init: CONFIG_PAGE_POISONING is on, will take precedence over init_on_alloc and init_on_free
[ 22.535704] mem auto-init: stack:all(zero), heap alloc:off, heap free:off
^ that's dmesg when i try them all =1
However if I do enable heap initialization by /proc/cmdline having:
page_poison=0 init_on_free=1 init_on_alloc=1
then:
[ 25.467867] test_meminit: all 130 tests passed!
(attached dmesg for this case)
and the ref_tracker issue remains, but as you've said below, it's expected to give that output as if to show that it's able to "track" the reference properly, and it makes sense to me now. I'll do CONFIG_TEST_REF_TRACKER=n and make a note about it.
The warnings indeed originate from CONFIG_TEST_REF_TRACKER=y, and seem to be yielding the desired output for that test (ccing Eric who wrote it).Perhaps that test could print something about its expectations to dmesg, but anyway there's nothing to fix in the kernel.
Alright, thank you for taking a look at this. I didn't realize there were two unrelated issues here.
May I ask what is the purpose of the config in question?
it's what I use on my home desktop PC(but merged the laptop .config into it, didn't have the time to test it on laptop so far yet) and I've enabled all tests to catch things early in the hopes of avoiding some potential btrfs filesystem corruption(for a third time, so far) if nothing else... but otherwise it'd be just reassuring to know that things are performing within their expected parameters (with tests passing and all). I don't remember why I've tried to keep modules at a minimum though.
Best regards!
Attachment:
626_heap_init.config
Description: Binary data
