On 2/23/21 3:58 PM, Andrew Morton wrote:

On Tue, 23 Feb 2021 10:06:12 -0800 Mike Kravetz [off-list ref] wrote:

quoted

On 2/23/21 6:57 AM, Gerald Schaefer wrote:

quoted

Hi,

LTP triggered a panic on s390 in hugepage_subpool_put_pages() with
linux-next 5.12.0-20210222, see below.

It crashes on the spin_lock(&spool->lock) at the beginning, because the
passed-in *spool points to 0000004e00000000, which is not addressable
memory. It rather looks like some flags and not a proper address. I suspect
some relation to the recent rework in that area, e.g. commit f1280272ae4d
("hugetlb: use page.private for hugetlb specific page flags").

__free_huge_page() calls hugepage_subpool_put_pages() and takes *spool from
hugetlb_page_subpool(page), which was changed by that commit to use
page[1]->private now.

Thanks Gerald,

Yes, I believe f1280272ae4d is the root cause of this issue.  In that
commit, the subpool pointer was moved from page->private of the head
page to page->private of the first subpage.  The page allocator will
initialize (zero) the private field of the head page, but not that of
subpages.  So, that bad subpool pointer is likely an old page->private
value for the page.

That strange call path from set_max_huge_pages to __free_huge_page is
actually how the code puts newly allocated pages on it's interfal free
list.  

I will do a bit more verification and put together a patch (it should
be simple).

There's also Michel's documentation request:
https://lkml.kernel.org/r/20210127102645.GH827@dhcp22.suse.cz

Thanks Andrew, I forgot about that.

It looks like the patch which added synchronization documentation requested
by Michal may not have be picked up.
https://lore.kernel.org/linux-mm/9183032f-3d77-d9e3-9cc8-fbaf3e892022@oracle.com/ (local)

If you still need to add that patch, I could redo and add the page[1]->private
documentation request mentioned here.  Just let me know what is the easiest for
you.
-- 
Mike Kravetz