mm: teach core mm about pte markers

This patch still does not use pte marker in any way, however it teaches
the core mm about the pte marker idea.

For example, handle_pte_marker() is introduced that will parse and handle
all the pte marker faults.

Many of the places are more about commenting it up - so that we know
there's the possibility of pte marker showing up, and why we don't need
special code for the cases.

[[email protected]: userfaultfd.c needs swapops.h]
  Link: https://lkml.kernel.org/r/[email protected]
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Peter Xu <[email protected]>
Cc: Alistair Popple <[email protected]>
Cc: Andrea Arcangeli <[email protected]>
Cc: Axel Rasmussen <[email protected]>
Cc: David Hildenbrand <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Jerome Glisse <[email protected]>
Cc: "Kirill A . Shutemov" <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Mike Kravetz <[email protected]>
Cc: Mike Rapoport <[email protected]>
Cc: Nadav Amit <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

fs/userfaultfd.c

@@ -29,6 +29,7 @@
 #include <linux/ioctl.h>
 #include <linux/security.h>
 #include <linux/hugetlb.h>
+#include <linux/swapops.h>
 
 int sysctl_unprivileged_userfaultfd __read_mostly;
 
@@ -249,9 +250,10 @@ static inline bool userfaultfd_huge_must_wait(struct userfaultfd_ctx *ctx,
 
 	/*
 	 * Lockless access: we're in a wait_event so it's ok if it
-	 * changes under us.
+	 * changes under us.  PTE markers should be handled the same as none
+	 * ptes here.
 	 */
-	if (huge_pte_none(pte))
+	if (huge_pte_none_mostly(pte))
 		ret = true;
 	if (!huge_pte_write(pte) && (reason & VM_UFFD_WP))
 		ret = true;
@@ -330,9 +332,10 @@ static inline bool userfaultfd_must_wait(struct userfaultfd_ctx *ctx,
 	pte = pte_offset_map(pmd, address);
 	/*
 	 * Lockless access: we're in a wait_event so it's ok if it
-	 * changes under us.
+	 * changes under us.  PTE markers should be handled the same as none
+	 * ptes here.
 	 */
-	if (pte_none(*pte))
+	if (pte_none_mostly(*pte))
 		ret = true;
 	if (!pte_write(*pte) && (reason & VM_UFFD_WP))
 		ret = true;

mm/filemap.c

@@ -3376,6 +3376,11 @@ vm_fault_t filemap_map_pages(struct vm_fault *vmf,
 		vmf->pte += xas.xa_index - last_pgoff;
 		last_pgoff = xas.xa_index;
 
+		/*
+		 * NOTE: If there're PTE markers, we'll leave them to be
+		 * handled in the specific fault path, and it'll prohibit the
+		 * fault-around logic.
+		 */
 		if (!pte_none(*vmf->pte))
 			goto unlock;
 

mm/hmm.c

@@ -239,7 +239,7 @@ static int hmm_vma_handle_pte(struct mm_walk *walk, unsigned long addr,
 	pte_t pte = *ptep;
 	uint64_t pfn_req_flags = *hmm_pfn;
 
-	if (pte_none(pte)) {
+	if (pte_none_mostly(pte)) {
 		required_fault =
 			hmm_pte_need_fault(hmm_vma_walk, pfn_req_flags, 0);
 		if (required_fault)

mm/memcontrol.c

@@ -5640,10 +5640,14 @@ static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
 
 	if (pte_present(ptent))
 		page = mc_handle_present_pte(vma, addr, ptent);
+	else if (pte_none_mostly(ptent))
+		/*
+		 * PTE markers should be treated as a none pte here, separated
+		 * from other swap handling below.
+		 */
+		page = mc_handle_file_pte(vma, addr, ptent);
 	else if (is_swap_pte(ptent))
 		page = mc_handle_swap_pte(vma, ptent, &ent);
-	else if (pte_none(ptent))
-		page = mc_handle_file_pte(vma, addr, ptent);
 
 	if (!page && !ent.val)
 		return ret;

mm/memory.c

@@ -100,6 +100,8 @@ struct page *mem_map;
 EXPORT_SYMBOL(mem_map);
 #endif
 
+static vm_fault_t do_fault(struct vm_fault *vmf);
+
 /*
  * A number of key systems in x86 including ioremap() rely on the assumption
  * that high_memory defines the upper bound on direct map memory, then end
@@ -1415,6 +1417,8 @@ static unsigned long zap_pte_range(struct mmu_gather *tlb,
 			if (!should_zap_page(details, page))
 				continue;
 			rss[mm_counter(page)]--;
+		} else if (is_pte_marker_entry(entry)) {
+			/* By default, simply drop all pte markers when zap */
 		} else if (is_hwpoison_entry(entry)) {
 			if (!should_zap_cows(details))
 				continue;
@@ -3555,6 +3559,23 @@ static inline bool should_try_to_free_swap(struct page *page,
 		page_count(page) == 2;
 }
 
+static vm_fault_t handle_pte_marker(struct vm_fault *vmf)
+{
+	swp_entry_t entry = pte_to_swp_entry(vmf->orig_pte);
+	unsigned long marker = pte_marker_get(entry);
+
+	/*
+	 * PTE markers should always be with file-backed memories, and the
+	 * marker should never be empty.  If anything weird happened, the best
+	 * thing to do is to kill the process along with its mm.
+	 */
+	if (WARN_ON_ONCE(vma_is_anonymous(vmf->vma) || !marker))
+		return VM_FAULT_SIGBUS;
+
+	/* TODO: handle pte markers */
+	return 0;
+}
+
 /*
  * We enter with non-exclusive mmap_lock (to exclude vma changes,
  * but allow concurrent faults), and pte mapped but not yet locked.
@@ -3592,6 +3613,8 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			ret = vmf->page->pgmap->ops->migrate_to_ram(vmf);
 		} else if (is_hwpoison_entry(entry)) {
 			ret = VM_FAULT_HWPOISON;
+		} else if (is_pte_marker_entry(entry)) {
+			ret = handle_pte_marker(vmf);
 		} else {
 			print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL);
 			ret = VM_FAULT_SIGBUS;

mm/mincore.c

@@ -122,7 +122,8 @@ static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 	for (; addr != end; ptep++, addr += PAGE_SIZE) {
 		pte_t pte = *ptep;
 
-		if (pte_none(pte))
+		/* We need to do cache lookup too for pte markers */
+		if (pte_none_mostly(pte))
 			__mincore_unmapped_range(addr, addr + PAGE_SIZE,
 						 vma, vec);
 		else if (pte_present(pte))

mm/mprotect.c

@@ -193,6 +193,9 @@ static unsigned long change_pte_range(struct mmu_gather *tlb,
 					newpte = pte_swp_mksoft_dirty(newpte);
 				if (pte_swp_uffd_wp(oldpte))
 					newpte = pte_swp_mkuffd_wp(newpte);
+			} else if (is_pte_marker_entry(entry)) {
+				/* Skip it, the same as none pte */
+				continue;
 			} else {
 				newpte = oldpte;
 			}