On Sat, Mar 4, 2023 at 2:27 AM kernel test robot <lkp@xxxxxxxxx> wrote: > > tree: https://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm.git mm-unstable > head: df3ae4347aff9be1e9763ffa3b1015fca348bfbd > commit: 92e3612279f925881e96dcc89acfb6bf96a2bb2a [83/143] mm/khugepaged: fix vm_lock/i_mmap_rwsem inversion in retract_page_tables > config: riscv-randconfig-r004-20230303 (https://download.01.org/0day-ci/archive/20230304/202303041807.a3nYQrom-lkp@xxxxxxxxx/config) > compiler: clang version 17.0.0 (https://github.com/llvm/llvm-project 67409911353323ca5edf2049ef0df54132fa1ca7) > reproduce (this is a W=1 build): > wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross > chmod +x ~/bin/make.cross > # install riscv cross compiling tool for clang build > # apt-get install binutils-riscv64-linux-gnu > # https://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm.git/commit/?id=92e3612279f925881e96dcc89acfb6bf96a2bb2a > git remote add akpm-mm https://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm.git > git fetch --no-tags akpm-mm mm-unstable > git checkout 92e3612279f925881e96dcc89acfb6bf96a2bb2a > # save the config file > mkdir build_dir && cp config build_dir/.config > COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=riscv olddefconfig > COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=riscv SHELL=/bin/bash > > If you fix the issue, kindly add following tag where applicable > | Reported-by: kernel test robot <lkp@xxxxxxxxx> > | Link: https://lore.kernel.org/oe-kbuild-all/202303041807.a3nYQrom-lkp@xxxxxxxxx/ > > All errors (new ones prefixed by >>): > > >> mm/khugepaged.c:1704:9: error: call to undeclared function 'vma_try_start_write'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration] > if (!vma_try_start_write(vma)) > ^ > mm/khugepaged.c:1704:9: note: did you mean 'vma_start_write'? > include/linux/mm.h:719:20: note: 'vma_start_write' declared here > static inline void vma_start_write(struct vm_area_struct *vma) {} > ^ > 1 error generated. Should be fixed once https://lore.kernel.org/all/20230304232856.DD36BC433D2@xxxxxxxxxxxxxxx/ change is merged. > > > vim +/vma_try_start_write +1704 mm/khugepaged.c > > 1641 > 1642 static int retract_page_tables(struct address_space *mapping, pgoff_t pgoff, > 1643 struct mm_struct *target_mm, > 1644 unsigned long target_addr, struct page *hpage, > 1645 struct collapse_control *cc) > 1646 { > 1647 struct vm_area_struct *vma; > 1648 int target_result = SCAN_FAIL; > 1649 > 1650 i_mmap_lock_write(mapping); > 1651 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { > 1652 int result = SCAN_FAIL; > 1653 struct mm_struct *mm = NULL; > 1654 unsigned long addr = 0; > 1655 pmd_t *pmd; > 1656 bool is_target = false; > 1657 > 1658 /* > 1659 * Check vma->anon_vma to exclude MAP_PRIVATE mappings that > 1660 * got written to. These VMAs are likely not worth investing > 1661 * mmap_write_lock(mm) as PMD-mapping is likely to be split > 1662 * later. > 1663 * > 1664 * Note that vma->anon_vma check is racy: it can be set up after > 1665 * the check but before we took mmap_lock by the fault path. > 1666 * But page lock would prevent establishing any new ptes of the > 1667 * page, so we are safe. > 1668 * > 1669 * An alternative would be drop the check, but check that page > 1670 * table is clear before calling pmdp_collapse_flush() under > 1671 * ptl. It has higher chance to recover THP for the VMA, but > 1672 * has higher cost too. It would also probably require locking > 1673 * the anon_vma. > 1674 */ > 1675 if (READ_ONCE(vma->anon_vma)) { > 1676 result = SCAN_PAGE_ANON; > 1677 goto next; > 1678 } > 1679 addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); > 1680 if (addr & ~HPAGE_PMD_MASK || > 1681 vma->vm_end < addr + HPAGE_PMD_SIZE) { > 1682 result = SCAN_VMA_CHECK; > 1683 goto next; > 1684 } > 1685 mm = vma->vm_mm; > 1686 is_target = mm == target_mm && addr == target_addr; > 1687 result = find_pmd_or_thp_or_none(mm, addr, &pmd); > 1688 if (result != SCAN_SUCCEED) > 1689 goto next; > 1690 /* > 1691 * We need exclusive mmap_lock to retract page table. > 1692 * > 1693 * We use trylock due to lock inversion: we need to acquire > 1694 * mmap_lock while holding page lock. Fault path does it in > 1695 * reverse order. Trylock is a way to avoid deadlock. > 1696 * > 1697 * Also, it's not MADV_COLLAPSE's job to collapse other > 1698 * mappings - let khugepaged take care of them later. > 1699 */ > 1700 result = SCAN_PTE_MAPPED_HUGEPAGE; > 1701 if ((cc->is_khugepaged || is_target) && > 1702 mmap_write_trylock(mm)) { > 1703 /* trylock for the same lock inversion as above */ > > 1704 if (!vma_try_start_write(vma)) > 1705 goto unlock_next; > 1706 > 1707 /* > 1708 * Re-check whether we have an ->anon_vma, because > 1709 * collapse_and_free_pmd() requires that either no > 1710 * ->anon_vma exists or the anon_vma is locked. > 1711 * We already checked ->anon_vma above, but that check > 1712 * is racy because ->anon_vma can be populated under the > 1713 * mmap lock in read mode. > 1714 */ > 1715 if (vma->anon_vma) { > 1716 result = SCAN_PAGE_ANON; > 1717 goto unlock_next; > 1718 } > 1719 /* > 1720 * When a vma is registered with uffd-wp, we can't > 1721 * recycle the pmd pgtable because there can be pte > 1722 * markers installed. Skip it only, so the rest mm/vma > 1723 * can still have the same file mapped hugely, however > 1724 * it'll always mapped in small page size for uffd-wp > 1725 * registered ranges. > 1726 */ > 1727 if (hpage_collapse_test_exit(mm)) { > 1728 result = SCAN_ANY_PROCESS; > 1729 goto unlock_next; > 1730 } > 1731 if (userfaultfd_wp(vma)) { > 1732 result = SCAN_PTE_UFFD_WP; > 1733 goto unlock_next; > 1734 } > 1735 collapse_and_free_pmd(mm, vma, addr, pmd); > 1736 if (!cc->is_khugepaged && is_target) > 1737 result = set_huge_pmd(vma, addr, pmd, hpage); > 1738 else > 1739 result = SCAN_SUCCEED; > 1740 > 1741 unlock_next: > 1742 mmap_write_unlock(mm); > 1743 goto next; > 1744 } > 1745 /* > 1746 * Calling context will handle target mm/addr. Otherwise, let > 1747 * khugepaged try again later. > 1748 */ > 1749 if (!is_target) { > 1750 khugepaged_add_pte_mapped_thp(mm, addr); > 1751 continue; > 1752 } > 1753 next: > 1754 if (is_target) > 1755 target_result = result; > 1756 } > 1757 i_mmap_unlock_write(mapping); > 1758 return target_result; > 1759 } > 1760 > > -- > 0-DAY CI Kernel Test Service > https://github.com/intel/lkp-tests
