I am also doing some tests on THP migration and discover that there are some corner cases not handled in this patchset. For example, in handle_mm_fault, without taking pmd_lock, the kernel may see pmd_none(*pmd) during THP migrations, which leads to handle_pte_fault or even deeper in the code path. At that moment, pmd_trans_unstable() will treat a pmd_migration_entry as pmd_bad and clear it. This leads to application crashing and page table leaks, since a deposited PTE page is not released when the application crashes. Even after I add is_pmd_migration_entry() into pmd_trans_unstable(), I still see application data corruptions. I hope someone can shed some light on how to debug this. Should I also look into pmd_trans_huge() call sites where pmd_migration_entry should be handled differently? Thanks. Anshuman Khandual wrote: > On 11/08/2016 05:01 AM, Naoya Horiguchi wrote: >> Hi everyone, >> >> I've updated thp migration patches for v4.9-rc2-mmotm-2016-10-27-18-27 >> with feedbacks for ver.1. > > Hello Noaya, > > I have been working with Zi Yan on the parallel huge page migration series > (https://lkml.org/lkml/2016/11/22/457) and planning to post them on top of > this THP migration enhancement series. Hence we were wondering if you have > plans to post a new version of this series in near future ? > > Regards > Anshuman > -- Best Regards, Yan Zi
Attachment:
signature.asc
Description: OpenPGP digital signature
