| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <201808131256.7IR4O74v%fengguang.wu@intel.com>
Date: Mon, 13 Aug 2018 12:17:33 +0800
From: kbuild test robot <lkp@...el.com>
To: Mike Kravetz <mike.kravetz@...cle.com>
Cc: kbuild-all@...org, linux-mm@...ck.org,
linux-kernel@...r.kernel.org,
"Kirill A . Shutemov" <kirill.shutemov@...ux.intel.com>,
Jérôme Glisse <jglisse@...hat.com>,
Vlastimil Babka <vbabka@...e.cz>,
Naoya Horiguchi <n-horiguchi@...jp.nec.com>,
Davidlohr Bueso <dave@...olabs.net>,
Michal Hocko <mhocko@...nel.org>,
Andrew Morton <akpm@...ux-foundation.org>,
Mike Kravetz <mike.kravetz@...cle.com>
Subject: Re: [PATCH] mm: migration: fix migration of huge PMD shared pages
Hi Mike,
I love your patch! Yet something to improve:
[auto build test ERROR on linus/master]
[also build test ERROR on v4.18 next-20180810]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]
url: https://github.com/0day-ci/linux/commits/Mike-Kravetz/mm-migration-fix-migration-of-huge-PMD-shared-pages/20180813-114549
config: i386-tinyconfig (attached as .config)
compiler: gcc-7 (Debian 7.3.0-16) 7.3.0
reproduce:
# save the attached .config to linux build tree
make ARCH=i386
All errors (new ones prefixed by >>):
mm/rmap.c: In function 'try_to_unmap_one':
>> mm/rmap.c:1425:7: error: implicit declaration of function 'huge_pmd_unshare'; did you mean 'do_huge_pmd_wp_page'? [-Werror=implicit-function-declaration]
huge_pmd_unshare(mm, &address, pvmw.pte)) {
^~~~~~~~~~~~~~~~
do_huge_pmd_wp_page
cc1: some warnings being treated as errors
vim +1425 mm/rmap.c
1382
1383 /*
1384 * If the page is mlock()d, we cannot swap it out.
1385 * If it's recently referenced (perhaps page_referenced
1386 * skipped over this mm) then we should reactivate it.
1387 */
1388 if (!(flags & TTU_IGNORE_MLOCK)) {
1389 if (vma->vm_flags & VM_LOCKED) {
1390 /* PTE-mapped THP are never mlocked */
1391 if (!PageTransCompound(page)) {
1392 /*
1393 * Holding pte lock, we do *not* need
1394 * mmap_sem here
1395 */
1396 mlock_vma_page(page);
1397 }
1398 ret = false;
1399 page_vma_mapped_walk_done(&pvmw);
1400 break;
1401 }
1402 if (flags & TTU_MUNLOCK)
1403 continue;
1404 }
1405
1406 /* Unexpected PMD-mapped THP? */
1407 VM_BUG_ON_PAGE(!pvmw.pte, page);
1408
1409 subpage = page - page_to_pfn(page) + pte_pfn(*pvmw.pte);
1410 address = pvmw.address;
1411
1412 /*
1413 * PMDs for hugetlbfs pages could be shared. In this case,
1414 * pages with shared PMDs will have a mapcount of 1 no matter
1415 * how many times it is actually mapped. Map counting for
1416 * PMD sharing is mostly done via the reference count on the
1417 * PMD page itself. If the page we are trying to unmap is a
1418 * hugetlbfs page, attempt to 'unshare' at the PMD level.
1419 * huge_pmd_unshare takes care of clearing the PUD and
1420 * reference counting on the PMD page which effectively unmaps
1421 * the page. Take care of flushing cache and TLB for page in
1422 * this specific mapping here.
1423 */
1424 if (PageHuge(page) &&
> 1425 huge_pmd_unshare(mm, &address, pvmw.pte)) {
1426 unsigned long end_add = address + vma_mmu_pagesize(vma);
1427
1428 flush_cache_range(vma, address, end_add);
1429 flush_tlb_range(vma, address, end_add);
1430 mmu_notifier_invalidate_range(mm, address, end_add);
1431 continue;
1432 }
1433
1434 if (IS_ENABLED(CONFIG_MIGRATION) &&
1435 (flags & TTU_MIGRATION) &&
1436 is_zone_device_page(page)) {
1437 swp_entry_t entry;
1438 pte_t swp_pte;
1439
1440 pteval = ptep_get_and_clear(mm, pvmw.address, pvmw.pte);
1441
1442 /*
1443 * Store the pfn of the page in a special migration
1444 * pte. do_swap_page() will wait until the migration
1445 * pte is removed and then restart fault handling.
1446 */
1447 entry = make_migration_entry(page, 0);
1448 swp_pte = swp_entry_to_pte(entry);
1449 if (pte_soft_dirty(pteval))
1450 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1451 set_pte_at(mm, pvmw.address, pvmw.pte, swp_pte);
1452 /*
1453 * No need to invalidate here it will synchronize on
1454 * against the special swap migration pte.
1455 */
1456 goto discard;
1457 }
1458
1459 if (!(flags & TTU_IGNORE_ACCESS)) {
1460 if (ptep_clear_flush_young_notify(vma, address,
1461 pvmw.pte)) {
1462 ret = false;
1463 page_vma_mapped_walk_done(&pvmw);
1464 break;
1465 }
1466 }
1467
1468 /* Nuke the page table entry. */
1469 flush_cache_page(vma, address, pte_pfn(*pvmw.pte));
1470 if (should_defer_flush(mm, flags)) {
1471 /*
1472 * We clear the PTE but do not flush so potentially
1473 * a remote CPU could still be writing to the page.
1474 * If the entry was previously clean then the
1475 * architecture must guarantee that a clear->dirty
1476 * transition on a cached TLB entry is written through
1477 * and traps if the PTE is unmapped.
1478 */
1479 pteval = ptep_get_and_clear(mm, address, pvmw.pte);
1480
1481 set_tlb_ubc_flush_pending(mm, pte_dirty(pteval));
1482 } else {
1483 pteval = ptep_clear_flush(vma, address, pvmw.pte);
1484 }
1485
1486 /* Move the dirty bit to the page. Now the pte is gone. */
1487 if (pte_dirty(pteval))
1488 set_page_dirty(page);
1489
1490 /* Update high watermark before we lower rss */
1491 update_hiwater_rss(mm);
1492
1493 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
1494 pteval = swp_entry_to_pte(make_hwpoison_entry(subpage));
1495 if (PageHuge(page)) {
1496 int nr = 1 << compound_order(page);
1497 hugetlb_count_sub(nr, mm);
1498 set_huge_swap_pte_at(mm, address,
1499 pvmw.pte, pteval,
1500 vma_mmu_pagesize(vma));
1501 } else {
1502 dec_mm_counter(mm, mm_counter(page));
1503 set_pte_at(mm, address, pvmw.pte, pteval);
1504 }
1505
1506 } else if (pte_unused(pteval) && !userfaultfd_armed(vma)) {
1507 /*
1508 * The guest indicated that the page content is of no
1509 * interest anymore. Simply discard the pte, vmscan
1510 * will take care of the rest.
1511 * A future reference will then fault in a new zero
1512 * page. When userfaultfd is active, we must not drop
1513 * this page though, as its main user (postcopy
1514 * migration) will not expect userfaults on already
1515 * copied pages.
1516 */
1517 dec_mm_counter(mm, mm_counter(page));
1518 /* We have to invalidate as we cleared the pte */
1519 mmu_notifier_invalidate_range(mm, address,
1520 address + PAGE_SIZE);
1521 } else if (IS_ENABLED(CONFIG_MIGRATION) &&
1522 (flags & (TTU_MIGRATION|TTU_SPLIT_FREEZE))) {
1523 swp_entry_t entry;
1524 pte_t swp_pte;
1525
1526 if (arch_unmap_one(mm, vma, address, pteval) < 0) {
1527 set_pte_at(mm, address, pvmw.pte, pteval);
1528 ret = false;
1529 page_vma_mapped_walk_done(&pvmw);
1530 break;
1531 }
1532
1533 /*
1534 * Store the pfn of the page in a special migration
1535 * pte. do_swap_page() will wait until the migration
1536 * pte is removed and then restart fault handling.
1537 */
1538 entry = make_migration_entry(subpage,
1539 pte_write(pteval));
1540 swp_pte = swp_entry_to_pte(entry);
1541 if (pte_soft_dirty(pteval))
1542 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1543 set_pte_at(mm, address, pvmw.pte, swp_pte);
1544 /*
1545 * No need to invalidate here it will synchronize on
1546 * against the special swap migration pte.
1547 */
1548 } else if (PageAnon(page)) {
1549 swp_entry_t entry = { .val = page_private(subpage) };
1550 pte_t swp_pte;
1551 /*
1552 * Store the swap location in the pte.
1553 * See handle_pte_fault() ...
1554 */
1555 if (unlikely(PageSwapBacked(page) != PageSwapCache(page))) {
1556 WARN_ON_ONCE(1);
1557 ret = false;
1558 /* We have to invalidate as we cleared the pte */
1559 mmu_notifier_invalidate_range(mm, address,
1560 address + PAGE_SIZE);
1561 page_vma_mapped_walk_done(&pvmw);
1562 break;
1563 }
1564
1565 /* MADV_FREE page check */
1566 if (!PageSwapBacked(page)) {
1567 if (!PageDirty(page)) {
1568 /* Invalidate as we cleared the pte */
1569 mmu_notifier_invalidate_range(mm,
1570 address, address + PAGE_SIZE);
1571 dec_mm_counter(mm, MM_ANONPAGES);
1572 goto discard;
1573 }
1574
1575 /*
1576 * If the page was redirtied, it cannot be
1577 * discarded. Remap the page to page table.
1578 */
1579 set_pte_at(mm, address, pvmw.pte, pteval);
1580 SetPageSwapBacked(page);
1581 ret = false;
1582 page_vma_mapped_walk_done(&pvmw);
1583 break;
1584 }
1585
1586 if (swap_duplicate(entry) < 0) {
1587 set_pte_at(mm, address, pvmw.pte, pteval);
1588 ret = false;
1589 page_vma_mapped_walk_done(&pvmw);
1590 break;
1591 }
1592 if (arch_unmap_one(mm, vma, address, pteval) < 0) {
1593 set_pte_at(mm, address, pvmw.pte, pteval);
1594 ret = false;
1595 page_vma_mapped_walk_done(&pvmw);
1596 break;
1597 }
1598 if (list_empty(&mm->mmlist)) {
1599 spin_lock(&mmlist_lock);
1600 if (list_empty(&mm->mmlist))
1601 list_add(&mm->mmlist, &init_mm.mmlist);
1602 spin_unlock(&mmlist_lock);
1603 }
1604 dec_mm_counter(mm, MM_ANONPAGES);
1605 inc_mm_counter(mm, MM_SWAPENTS);
1606 swp_pte = swp_entry_to_pte(entry);
1607 if (pte_soft_dirty(pteval))
1608 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1609 set_pte_at(mm, address, pvmw.pte, swp_pte);
1610 /* Invalidate as we cleared the pte */
1611 mmu_notifier_invalidate_range(mm, address,
1612 address + PAGE_SIZE);
1613 } else {
1614 /*
1615 * We should not need to notify here as we reach this
1616 * case only from freeze_page() itself only call from
1617 * split_huge_page_to_list() so everything below must
1618 * be true:
1619 * - page is not anonymous
1620 * - page is locked
1621 *
1622 * So as it is a locked file back page thus it can not
1623 * be remove from the page cache and replace by a new
1624 * page before mmu_notifier_invalidate_range_end so no
1625 * concurrent thread might update its page table to
1626 * point at new page while a device still is using this
1627 * page.
1628 *
1629 * See Documentation/vm/mmu_notifier.rst
1630 */
1631 dec_mm_counter(mm, mm_counter_file(page));
1632 }
1633 discard:
1634 /*
1635 * No need to call mmu_notifier_invalidate_range() it has be
1636 * done above for all cases requiring it to happen under page
1637 * table lock before mmu_notifier_invalidate_range_end()
1638 *
1639 * See Documentation/vm/mmu_notifier.rst
1640 */
1641 page_remove_rmap(subpage, PageHuge(page));
1642 put_page(page);
1643 }
1644
1645 mmu_notifier_invalidate_range_end(vma->vm_mm, start, end);
1646
1647 return ret;
1648 }
1649
---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
Download attachment ".config.gz" of type "application/gzip" (6367 bytes)
Powered by blists - more mailing lists