mm: get rid of vmacache_flush_all() entirely

Jann Horn points out that the vmacache_flush_all() function is not only potentially expensive, it's buggy too. It also happens to be entirely unnecessary, because the sequence number overflow case can be avoided by simply making the sequence number be 64-bit. That doesn't even grow the data structures in question, because the other adjacent fields are already 64-bit. So simplify the whole thing by just making the sequence number overflow case go away entirely, which gets rid of all the complications and makes the code faster too. Win-win. [ Oleg Nesterov points out that the VMACACHE_FULL_FLUSHES statistics also just goes away entirely with this ] Reported-by: Jann Horn <[email protected]> Suggested-by: Will Deacon <[email protected]> Acked-by: Davidlohr Bueso <[email protected]> Cc: Oleg Nesterov <[email protected]> Cc: [email protected] Signed-off-by: Linus Torvalds <[email protected]>
heikomat · Sep 14, 2018 · 7a9cdeb · 7a9cdeb
1 parent 54eda9d
commit 7a9cdeb
Show file tree

Hide file tree

Showing 6 changed files with 4 additions and 48 deletions.
diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
@@ -341,7 +341,7 @@ struct mm_struct {
 	struct {
 		struct vm_area_struct *mmap;		/* list of VMAs */
 		struct rb_root mm_rb;
-		u32 vmacache_seqnum;                   /* per-thread vmacache */
+		u64 vmacache_seqnum;                   /* per-thread vmacache */
 #ifdef CONFIG_MMU
 		unsigned long (*get_unmapped_area) (struct file *filp,
 				unsigned long addr, unsigned long len,

diff --git a/include/linux/mm_types_task.h b/include/linux/mm_types_task.h
@@ -32,7 +32,7 @@
 #define VMACACHE_MASK (VMACACHE_SIZE - 1)
 
 struct vmacache {
-	u32 seqnum;
+	u64 seqnum;
 	struct vm_area_struct *vmas[VMACACHE_SIZE];
 };
 

diff --git a/include/linux/vm_event_item.h b/include/linux/vm_event_item.h
@@ -105,7 +105,6 @@ enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
 #ifdef CONFIG_DEBUG_VM_VMACACHE
 		VMACACHE_FIND_CALLS,
 		VMACACHE_FIND_HITS,
-		VMACACHE_FULL_FLUSHES,
 #endif
 #ifdef CONFIG_SWAP
 		SWAP_RA,

diff --git a/include/linux/vmacache.h b/include/linux/vmacache.h
@@ -10,7 +10,6 @@ static inline void vmacache_flush(struct task_struct *tsk)
 	memset(tsk->vmacache.vmas, 0, sizeof(tsk->vmacache.vmas));
 }
 
-extern void vmacache_flush_all(struct mm_struct *mm);
 extern void vmacache_update(unsigned long addr, struct vm_area_struct *newvma);
 extern struct vm_area_struct *vmacache_find(struct mm_struct *mm,
 						    unsigned long addr);
@@ -24,10 +23,6 @@ extern struct vm_area_struct *vmacache_find_exact(struct mm_struct *mm,
 static inline void vmacache_invalidate(struct mm_struct *mm)
 {
 	mm->vmacache_seqnum++;
-
-	/* deal with overflows */
-	if (unlikely(mm->vmacache_seqnum == 0))
-		vmacache_flush_all(mm);
 }
 
 #endif /* __LINUX_VMACACHE_H */
diff --git a/mm/debug.c b/mm/debug.c
@@ -114,7 +114,7 @@ EXPORT_SYMBOL(dump_vma);
 
 void dump_mm(const struct mm_struct *mm)
 {
-	pr_emerg("mm %px mmap %px seqnum %d task_size %lu\n"
+	pr_emerg("mm %px mmap %px seqnum %llu task_size %lu\n"
 #ifdef CONFIG_MMU
 		"get_unmapped_area %px\n"
 #endif
@@ -142,7 +142,7 @@ void dump_mm(const struct mm_struct *mm)
 		"tlb_flush_pending %d\n"
 		"def_flags: %#lx(%pGv)\n",
 
-		mm, mm->mmap, mm->vmacache_seqnum, mm->task_size,
+		mm, mm->mmap, (long long) mm->vmacache_seqnum, mm->task_size,
 #ifdef CONFIG_MMU
 		mm->get_unmapped_area,
 #endif

diff --git a/mm/vmacache.c b/mm/vmacache.c
@@ -19,44 +19,6 @@
 #endif
 #define VMACACHE_HASH(addr) ((addr >> VMACACHE_SHIFT) & VMACACHE_MASK)
 
-/*
- * Flush vma caches for threads that share a given mm.
- *
- * The operation is safe because the caller holds the mmap_sem
- * exclusively and other threads accessing the vma cache will
- * have mmap_sem held at least for read, so no extra locking
- * is required to maintain the vma cache.
- */
-void vmacache_flush_all(struct mm_struct *mm)
-{
-	struct task_struct *g, *p;
-
-	count_vm_vmacache_event(VMACACHE_FULL_FLUSHES);
-
-	/*
-	 * Single threaded tasks need not iterate the entire
-	 * list of process. We can avoid the flushing as well
-	 * since the mm's seqnum was increased and don't have
-	 * to worry about other threads' seqnum. Current's
-	 * flush will occur upon the next lookup.
-	 */
-	if (atomic_read(&mm->mm_users) == 1)
-		return;
-
-	rcu_read_lock();
-	for_each_process_thread(g, p) {
-		/*
-		 * Only flush the vmacache pointers as the
-		 * mm seqnum is already set and curr's will
-		 * be set upon invalidation when the next
-		 * lookup is done.
-		 */
-		if (mm == p->mm)
-			vmacache_flush(p);
-	}
-	rcu_read_unlock();
-}
-
 /*
  * This task may be accessing a foreign mm via (for example)
  * get_user_pages()->find_vma().  The vmacache is task-local and this