Andrey Wagin wrote:
> Hi Kirill,
>
> I have a test server, which executes CRIU tests. It crashed today. I
> don't know how to reproduce this bug. If these information will be not
> enough, I will try to get more.
...
> [174344.225025] Call Trace:
> [174344.225025] [<ffffffff8119427f>] free_pgd_range+0x2bf/0x410
> [174344.225025] [<ffffffff8119449e>] free_pgtables+0xce/0x120
> [174344.225025] [<ffffffff8119b900>] unmap_region+0xe0/0x120
> [174344.225025] [<ffffffff811a0036>] ? move_page_tables+0x526/0x6b0
> [174344.225025] [<ffffffff8119d6a9>] do_munmap+0x249/0x360
> [174344.225025] [<ffffffff811a0304>] move_vma+0x144/0x270
> [174344.225025] [<ffffffff811a07e9>] SyS_mremap+0x3b9/0x510
> [174344.225025] [<ffffffff8172d512>] system_call_fastpath+0x16/0x1b
> [174344.225025] Code: 83 7c 24 20 00 75 24 4c 89 e7 e8 bd b7 14 00 4c
> 89 e6 48 89 df e8 82 b9 14 00 85 c0 75 08 48 89 df e8 36 c9 14 00 5b
> 41 5c c9 c3 <0f> 0b eb fe 90 90 90 90 90 90 90 90 90 90 90 90 90 55 48
> 89 e5
> [174344.225025] RIP [<ffffffff81046f7f>] ___pmd_free_tlb+0x6f/0x80
> [174344.225025] RSP <ffff88008f267c28>
I see. We need to move page->pmd_huge_pte to new struct page.
Could you test the patch below?
I only build-tested it [from my vacation].
It suppose to work on x86-64, but it will require more work to get it
right for sparc and other archs with custom pgtable_trans_huge_deposit()
and pgtable_trans_huge_withdraw(). I'll prepare this a bit later.
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index bccd5a628ea6..546c30193235 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1481,8 +1481,22 @@ int move_huge_pmd(struct vm_area_struct *vma, struct vm_area_struct *new_vma,
pmd = pmdp_get_and_clear(mm, old_addr, old_pmd);
VM_BUG_ON(!pmd_none(*new_pmd));
set_pmd_at(mm, new_addr, new_pmd, pmd_mksoft_dirty(pmd));
- if (new_ptl != old_ptl)
+ if (new_ptl != old_ptl) {
+ pgtable_t old_pte = pmd_huge_pte(mm, old_pmd);
+ pgtable_t new_pte = pmd_huge_pte(mm, new_pmd);
+
+ /*
+ * Move page->pmd_huge_pmd if new_pmd is on different
+ * page table.
+ */
+ if (new_pte)
+ list_splice(&old_pte->lru, &new_pte->lru);
+ else
+ pmd_huge_pte(mm, new_pmd) = old_pte;
+ pmd_huge_pte(mm, old_pmd) = NULL;
+
spin_unlock(new_ptl);
+ }
spin_unlock(old_ptl);
}
out:
--
Kirill A. Shutemov
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>