CC: [email protected] CC: "GNU/Weeb Mailing List" <[email protected]> CC: [email protected] TO: Vinayak Menon <[email protected]> CC: Todd Kjos <[email protected]>
tree: https://github.com/ammarfaizi2/linux-block google/android/kernel/common/android13-5.10 head: a5bf62aaed069a2582e25af1bda15343cd119219 commit: 35eacb5c87b9b5e9683e25a24282cde1a2d4a1d5 [684/9999] ANDROID: mm: allow vmas with vm_ops to be speculatively handled :::::: branch date: 11 hours ago :::::: commit date: 9 months ago config: i386-randconfig-m021-20211118 (https://download.01.org/0day-ci/archive/20220125/[email protected]/config) compiler: gcc-9 (Debian 9.3.0-22) 9.3.0 If you fix the issue, kindly add following tag as appropriate Reported-by: kernel test robot <[email protected]> Reported-by: Dan Carpenter <[email protected]> smatch warnings: mm/memory.c:4691 handle_pte_fault() warn: inconsistent returns 'vmf->ptl'. vim +4691 mm/memory.c a00cc7d9dd93d6 Matthew Wilcox 2017-02-24 4562 ^1da177e4c3f41 Linus Torvalds 2005-04-16 4563 /* ^1da177e4c3f41 Linus Torvalds 2005-04-16 4564 * These routines also need to handle stuff like marking pages dirty ^1da177e4c3f41 Linus Torvalds 2005-04-16 4565 * and/or accessed for architectures that don't do it in hardware (most ^1da177e4c3f41 Linus Torvalds 2005-04-16 4566 * RISC architectures). The early dirtying is also good on the i386. ^1da177e4c3f41 Linus Torvalds 2005-04-16 4567 * ^1da177e4c3f41 Linus Torvalds 2005-04-16 4568 * There is also a hook called "update_mmu_cache()" that architectures ^1da177e4c3f41 Linus Torvalds 2005-04-16 4569 * with external mmu caches can use to update those (ie the Sparc or ^1da177e4c3f41 Linus Torvalds 2005-04-16 4570 * PowerPC hashed page tables that act as extended TLBs). ^1da177e4c3f41 Linus Torvalds 2005-04-16 4571 * c1e8d7c6a7a682 Michel Lespinasse 2020-06-08 4572 * We enter with non-exclusive mmap_lock (to exclude vma changes, but allow 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4573 * concurrent faults). 9a95f3cf7b33d6 Paul Cassella 2014-08-06 4574 * c1e8d7c6a7a682 Michel Lespinasse 2020-06-08 4575 * The mmap_lock may have been released depending on flags and our return value. 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4576 * See filemap_fault() and __lock_page_or_retry(). ^1da177e4c3f41 Linus Torvalds 2005-04-16 4577 */ 2b7403035459c7 Souptick Joarder 2018-08-23 4578 static vm_fault_t handle_pte_fault(struct vm_fault *vmf) ^1da177e4c3f41 Linus Torvalds 2005-04-16 4579 { ^1da177e4c3f41 Linus Torvalds 2005-04-16 4580 pte_t entry; 35eacb5c87b9b5 Vinayak Menon 2021-03-18 4581 vm_fault_t ret = 0; ^1da177e4c3f41 Linus Torvalds 2005-04-16 4582 82b0f8c39a3869 Jan Kara 2016-12-14 4583 if (unlikely(pmd_none(*vmf->pmd))) { 1c5371744061fc Peter Zijlstra 2018-04-17 4584 /* 1c5371744061fc Peter Zijlstra 2018-04-17 4585 * In the case of the speculative page fault handler we abort 1c5371744061fc Peter Zijlstra 2018-04-17 4586 * the speculative path immediately as the pmd is probably 1c5371744061fc Peter Zijlstra 2018-04-17 4587 * in the way to be converted in a huge one. We will try 1c5371744061fc Peter Zijlstra 2018-04-17 4588 * again holding the mmap_sem (which implies that the collapse 1c5371744061fc Peter Zijlstra 2018-04-17 4589 * operation is done). 1c5371744061fc Peter Zijlstra 2018-04-17 4590 */ 1c5371744061fc Peter Zijlstra 2018-04-17 4591 if (vmf->flags & FAULT_FLAG_SPECULATIVE) 1c5371744061fc Peter Zijlstra 2018-04-17 4592 return VM_FAULT_RETRY; e37c6982706333 Christian Borntraeger 2014-12-07 4593 /* 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4594 * Leave __pte_alloc() until later: because vm_ops->fault may 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4595 * want to allocate huge page, and if we expose page table 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4596 * for an instant, it will be difficult to retract from 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4597 * concurrent faults and from rmap lookups. e37c6982706333 Christian Borntraeger 2014-12-07 4598 */ 82b0f8c39a3869 Jan Kara 2016-12-14 4599 vmf->pte = NULL; 1c5371744061fc Peter Zijlstra 2018-04-17 4600 } else if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) { 0aa300a25229b9 Kirill A. Shutemov 2020-12-19 4601 /* 0aa300a25229b9 Kirill A. Shutemov 2020-12-19 4602 * If a huge pmd materialized under us just retry later. Use 0aa300a25229b9 Kirill A. Shutemov 2020-12-19 4603 * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead 0aa300a25229b9 Kirill A. Shutemov 2020-12-19 4604 * of pmd_trans_huge() to ensure the pmd didn't become 0aa300a25229b9 Kirill A. Shutemov 2020-12-19 4605 * pmd_trans_huge under us and then back to pmd_none, as a 0aa300a25229b9 Kirill A. Shutemov 2020-12-19 4606 * result of MADV_DONTNEED running immediately after a huge pmd 0aa300a25229b9 Kirill A. Shutemov 2020-12-19 4607 * fault in a different thread of this mm, in turn leading to a 0aa300a25229b9 Kirill A. Shutemov 2020-12-19 4608 * misleading pmd_trans_huge() retval. All we have to ensure is 0aa300a25229b9 Kirill A. Shutemov 2020-12-19 4609 * that it is a regular pmd that we can walk with 0aa300a25229b9 Kirill A. Shutemov 2020-12-19 4610 * pte_offset_map() and we can do that through an atomic read 0aa300a25229b9 Kirill A. Shutemov 2020-12-19 4611 * in C, which is what pmd_trans_unstable() provides. 0aa300a25229b9 Kirill A. Shutemov 2020-12-19 4612 */ d0f0931de936a0 Ross Zwisler 2017-06-02 4613 if (pmd_devmap_trans_unstable(vmf->pmd)) 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4614 return 0; 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4615 /* 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4616 * A regular pmd is established and it can't morph into a huge 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4617 * pmd from under us anymore at this point because we hold the c1e8d7c6a7a682 Michel Lespinasse 2020-06-08 4618 * mmap_lock read mode and khugepaged takes it in write mode. 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4619 * So now it's safe to run pte_offset_map(). 1c5371744061fc Peter Zijlstra 2018-04-17 4620 * This is not applicable to the speculative page fault handler 1c5371744061fc Peter Zijlstra 2018-04-17 4621 * but in that case, the pte is fetched earlier in 1c5371744061fc Peter Zijlstra 2018-04-17 4622 * handle_speculative_fault(). 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4623 */ 82b0f8c39a3869 Jan Kara 2016-12-14 4624 vmf->pte = pte_offset_map(vmf->pmd, vmf->address); 2994302bc8a171 Jan Kara 2016-12-14 4625 vmf->orig_pte = *vmf->pte; 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4626 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4627 /* 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4628 * some architectures can have larger ptes than wordsize, 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4629 * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and b03a0fe0c5e4b4 Paul E. McKenney 2017-10-23 4630 * CONFIG_32BIT=y, so READ_ONCE cannot guarantee atomic b03a0fe0c5e4b4 Paul E. McKenney 2017-10-23 4631 * accesses. The code below just needs a consistent view b03a0fe0c5e4b4 Paul E. McKenney 2017-10-23 4632 * for the ifs and we later double check anyway with the 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4633 * ptl lock held. So here a barrier will do. 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4634 */ e37c6982706333 Christian Borntraeger 2014-12-07 4635 barrier(); 2994302bc8a171 Jan Kara 2016-12-14 4636 if (pte_none(vmf->orig_pte)) { 82b0f8c39a3869 Jan Kara 2016-12-14 4637 pte_unmap(vmf->pte); 82b0f8c39a3869 Jan Kara 2016-12-14 4638 vmf->pte = NULL; 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4639 } 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4640 } 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4641 82b0f8c39a3869 Jan Kara 2016-12-14 4642 if (!vmf->pte) { 82b0f8c39a3869 Jan Kara 2016-12-14 4643 if (vma_is_anonymous(vmf->vma)) 82b0f8c39a3869 Jan Kara 2016-12-14 4644 return do_anonymous_page(vmf); 35eacb5c87b9b5 Vinayak Menon 2021-03-18 4645 else if ((vmf->flags & FAULT_FLAG_SPECULATIVE) && 35eacb5c87b9b5 Vinayak Menon 2021-03-18 4646 !vmf_allows_speculation(vmf)) 1c5371744061fc Peter Zijlstra 2018-04-17 4647 return VM_FAULT_RETRY; b5330628546616 Oleg Nesterov 2015-09-08 4648 else 82b0f8c39a3869 Jan Kara 2016-12-14 4649 return do_fault(vmf); 65500d234e74fc Hugh Dickins 2005-10-29 4650 } 7267ec008b5cd8 Kirill A. Shutemov 2016-07-26 4651 2994302bc8a171 Jan Kara 2016-12-14 4652 if (!pte_present(vmf->orig_pte)) 2994302bc8a171 Jan Kara 2016-12-14 4653 return do_swap_page(vmf); ^1da177e4c3f41 Linus Torvalds 2005-04-16 4654 2994302bc8a171 Jan Kara 2016-12-14 4655 if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) 2994302bc8a171 Jan Kara 2016-12-14 4656 return do_numa_page(vmf); d10e63f29488b0 Mel Gorman 2012-10-25 4657 b23ffc113b308e Laurent Dufour 2018-04-17 4658 if (!pte_spinlock(vmf)) b23ffc113b308e Laurent Dufour 2018-04-17 4659 return VM_FAULT_RETRY; 2994302bc8a171 Jan Kara 2016-12-14 4660 entry = vmf->orig_pte; 7df676974359f9 Bibo Mao 2020-05-27 4661 if (unlikely(!pte_same(*vmf->pte, entry))) { 7df676974359f9 Bibo Mao 2020-05-27 4662 update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); 8f4e2101fd7df9 Hugh Dickins 2005-10-29 4663 goto unlock; 7df676974359f9 Bibo Mao 2020-05-27 4664 } 82b0f8c39a3869 Jan Kara 2016-12-14 4665 if (vmf->flags & FAULT_FLAG_WRITE) { f6f3732162b5ae Linus Torvalds 2017-12-15 4666 if (!pte_write(entry)) 2994302bc8a171 Jan Kara 2016-12-14 4667 return do_wp_page(vmf); ^1da177e4c3f41 Linus Torvalds 2005-04-16 4668 entry = pte_mkdirty(entry); ^1da177e4c3f41 Linus Torvalds 2005-04-16 4669 } ^1da177e4c3f41 Linus Torvalds 2005-04-16 4670 entry = pte_mkyoung(entry); 82b0f8c39a3869 Jan Kara 2016-12-14 4671 if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, 82b0f8c39a3869 Jan Kara 2016-12-14 4672 vmf->flags & FAULT_FLAG_WRITE)) { 82b0f8c39a3869 Jan Kara 2016-12-14 4673 update_mmu_cache(vmf->vma, vmf->address, vmf->pte); 1a44e149084d77 Andrea Arcangeli 2005-10-29 4674 } else { b7333b58f358f3 Yang Shi 2020-08-14 4675 /* Skip spurious TLB flush for retried page fault */ b7333b58f358f3 Yang Shi 2020-08-14 4676 if (vmf->flags & FAULT_FLAG_TRIED) b7333b58f358f3 Yang Shi 2020-08-14 4677 goto unlock; 35eacb5c87b9b5 Vinayak Menon 2021-03-18 4678 if (vmf->flags & FAULT_FLAG_SPECULATIVE) 35eacb5c87b9b5 Vinayak Menon 2021-03-18 4679 ret = VM_FAULT_RETRY; 1a44e149084d77 Andrea Arcangeli 2005-10-29 4680 /* 1a44e149084d77 Andrea Arcangeli 2005-10-29 4681 * This is needed only for protection faults but the arch code 1a44e149084d77 Andrea Arcangeli 2005-10-29 4682 * is not yet telling us if this is a protection fault or not. 1a44e149084d77 Andrea Arcangeli 2005-10-29 4683 * This still avoids useless tlb flushes for .text page faults 1a44e149084d77 Andrea Arcangeli 2005-10-29 4684 * with threads. 1a44e149084d77 Andrea Arcangeli 2005-10-29 4685 */ 82b0f8c39a3869 Jan Kara 2016-12-14 4686 if (vmf->flags & FAULT_FLAG_WRITE) 82b0f8c39a3869 Jan Kara 2016-12-14 4687 flush_tlb_fix_spurious_fault(vmf->vma, vmf->address); 1a44e149084d77 Andrea Arcangeli 2005-10-29 4688 } 8f4e2101fd7df9 Hugh Dickins 2005-10-29 4689 unlock: 82b0f8c39a3869 Jan Kara 2016-12-14 4690 pte_unmap_unlock(vmf->pte, vmf->ptl); 35eacb5c87b9b5 Vinayak Menon 2021-03-18 @4691 return ret; ^1da177e4c3f41 Linus Torvalds 2005-04-16 4692 } ^1da177e4c3f41 Linus Torvalds 2005-04-16 4693 --- 0-DAY CI Kernel Test Service, Intel Corporation https://lists.01.org/hyperkitty/list/[email protected] _______________________________________________ kbuild mailing list -- [email protected] To unsubscribe send an email to [email protected]
