+Aneesh
On Fri, Dec 14, 2018 at 02:27:40PM +0800, Huang Ying wrote:
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index bd2543e10938..49df3e7c96c7 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> +int split_swap_cluster_map(swp_entry_t entry)
...
> + VM_BUG_ON(!IS_ALIGNED(offset, SWAPFILE_CLUSTER));
Hi Ying, I crashed on this in v6 as reported and it still dies on me now.
> @@ -2150,6 +2185,9 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,
...
> + if (IS_ENABLED(CONFIG_THP_SWAP) && is_swap_pmd(old_pmd))
> + return __split_huge_swap_pmd(vma, haddr, pmd);
Problem is 'pmd' is passed here, which has been pmdp_invalidate()ed under the
assumption that it is not a swap entry. pmd's pfn bits get inverted for L1TF,
so the swap entry gets corrupted and this BUG is the first place that notices.
I don't see a reason to invalidate so soon, so what about just moving the
invalidation down, past the migration/swap checks?
