[5/9] mm: vmalloc: Insert busy-VA per-cpu zone

Message ID 20230522110849.2921-6-urezki@gmail.com
State New
Headers
Series Mitigate a vmap lock contention |

Commit Message

Uladzislau Rezki May 22, 2023, 11:08 a.m. UTC
  Store busy-VA objects per a CPU zone. A va->va_start address is
converted into a correct zone where it is placed and resides. An
addr_to_cvz() function is used to do a proper address conversion.

Such approach balances VAs across CPUs. That is why an access
becomes scalable to number of CPUs in a system.

Please note:

Since a zone size is set to ULONG_MAX, i.e. everything is bound
thus accessed to the CPU_0 so far, this patch does not give any
difference comparing with a current behavior.

The global vmap_area_lock, vmap_area_root are removed as there
is no need in it anymore. The vmap_area_list is still kept and
is _empty_. It is exported for a kexec only.

The vmallocinfo and vread() have to be reworked to be able to
handle multiple zones. As a result of this patch it can handle
only one zone, i.e. when cache is disabled.

Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
---
 mm/vmalloc.c | 127 ++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 84 insertions(+), 43 deletions(-)
  

Comments

Christoph Hellwig May 23, 2023, 6:12 a.m. UTC | #1
>  /* Look up the first VA which satisfies addr < va_end, NULL if none. */
> -static struct vmap_area *find_vmap_area_exceed_addr(unsigned long addr)
> +static struct vmap_area *find_vmap_area_exceed_addr(unsigned long addr, struct rb_root *root)

Please avoid the overly long line.

> +	struct cpu_vmap_zone *z = addr_to_cvz(va->va_start);
> +
>  	/*
>  	 * Remove from the busy tree/list.
>  	 */
> -	spin_lock(&vmap_area_lock);
> -	unlink_va(va, &vmap_area_root);
> -	spin_unlock(&vmap_area_lock);
> +	fbl_lock(z, BUSY);
> +	unlink_va(va, &fbl_root(z, BUSY));
> +	fbl_unlock(z, BUSY);

I find the BUSY magic here very confusing, and would prefer to
just spell the actual lock reference out.
  
Uladzislau Rezki May 23, 2023, 3 p.m. UTC | #2
On Mon, May 22, 2023 at 11:12:08PM -0700, Christoph Hellwig wrote:
> >  /* Look up the first VA which satisfies addr < va_end, NULL if none. */
> > -static struct vmap_area *find_vmap_area_exceed_addr(unsigned long addr)
> > +static struct vmap_area *find_vmap_area_exceed_addr(unsigned long addr, struct rb_root *root)
> 
> Please avoid the overly long line.
> 
Will fix it.

> > +	struct cpu_vmap_zone *z = addr_to_cvz(va->va_start);
> > +
> >  	/*
> >  	 * Remove from the busy tree/list.
> >  	 */
> > -	spin_lock(&vmap_area_lock);
> > -	unlink_va(va, &vmap_area_root);
> > -	spin_unlock(&vmap_area_lock);
> > +	fbl_lock(z, BUSY);
> > +	unlink_va(va, &fbl_root(z, BUSY));
> > +	fbl_unlock(z, BUSY);
> 
> I find the BUSY magic here very confusing, and would prefer to
> just spell the actual lock reference out.
>
No problem. I can make it open-coded. What is about an access to tree/list?

>> unlink_va(va, &fbl_root(z, BUSY));
I mean this one.

--
Uladzislau Rezki
  

Patch

diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index f6da2590b4de..a9170fe19909 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -729,11 +729,9 @@  EXPORT_SYMBOL(vmalloc_to_pfn);
 #define DEBUG_AUGMENT_LOWEST_MATCH_CHECK 0
 
 
-static DEFINE_SPINLOCK(vmap_area_lock);
 static DEFINE_SPINLOCK(free_vmap_area_lock);
 /* Export for kexec only */
 LIST_HEAD(vmap_area_list);
-static struct rb_root vmap_area_root = RB_ROOT;
 static bool vmap_initialized __read_mostly;
 
 static struct rb_root purge_vmap_area_root = RB_ROOT;
@@ -823,7 +821,7 @@  addr_to_cpu(unsigned long addr)
 }
 
 static inline struct cpu_vmap_zone *
-__maybe_unused addr_to_cvz(unsigned long addr)
+addr_to_cvz(unsigned long addr)
 {
 	return &per_cpu(cpu_vmap_zone, addr_to_cpu(addr));
 }
@@ -859,10 +857,10 @@  unsigned long vmalloc_nr_pages(void)
 }
 
 /* Look up the first VA which satisfies addr < va_end, NULL if none. */
-static struct vmap_area *find_vmap_area_exceed_addr(unsigned long addr)
+static struct vmap_area *find_vmap_area_exceed_addr(unsigned long addr, struct rb_root *root)
 {
 	struct vmap_area *va = NULL;
-	struct rb_node *n = vmap_area_root.rb_node;
+	struct rb_node *n = root->rb_node;
 
 	addr = (unsigned long)kasan_reset_tag((void *)addr);
 
@@ -1608,12 +1606,14 @@  __alloc_vmap_area(struct rb_root *root, struct list_head *head,
  */
 static void free_vmap_area(struct vmap_area *va)
 {
+	struct cpu_vmap_zone *z = addr_to_cvz(va->va_start);
+
 	/*
 	 * Remove from the busy tree/list.
 	 */
-	spin_lock(&vmap_area_lock);
-	unlink_va(va, &vmap_area_root);
-	spin_unlock(&vmap_area_lock);
+	fbl_lock(z, BUSY);
+	unlink_va(va, &fbl_root(z, BUSY));
+	fbl_unlock(z, BUSY);
 
 	/*
 	 * Insert/Merge it back to the free tree/list.
@@ -1656,6 +1656,7 @@  static struct vmap_area *alloc_vmap_area(unsigned long size,
 				int node, gfp_t gfp_mask,
 				unsigned long va_flags)
 {
+	struct cpu_vmap_zone *z;
 	struct vmap_area *va;
 	unsigned long freed;
 	unsigned long addr;
@@ -1701,9 +1702,11 @@  static struct vmap_area *alloc_vmap_area(unsigned long size,
 	va->vm = NULL;
 	va->flags = va_flags;
 
-	spin_lock(&vmap_area_lock);
-	insert_vmap_area(va, &vmap_area_root, &vmap_area_list);
-	spin_unlock(&vmap_area_lock);
+	z = addr_to_cvz(va->va_start);
+
+	fbl_lock(z, BUSY);
+	insert_vmap_area(va, &fbl_root(z, BUSY), &fbl_head(z, BUSY));
+	fbl_unlock(z, BUSY);
 
 	BUG_ON(!IS_ALIGNED(va->va_start, align));
 	BUG_ON(va->va_start < vstart);
@@ -1926,24 +1929,26 @@  static void free_unmap_vmap_area(struct vmap_area *va)
 
 struct vmap_area *find_vmap_area(unsigned long addr)
 {
+	struct cpu_vmap_zone *z = addr_to_cvz(addr);
 	struct vmap_area *va;
 
-	spin_lock(&vmap_area_lock);
-	va = __find_vmap_area(addr, &vmap_area_root);
-	spin_unlock(&vmap_area_lock);
+	fbl_lock(z, BUSY);
+	va = __find_vmap_area(addr, &fbl_root(z, BUSY));
+	fbl_unlock(z, BUSY);
 
 	return va;
 }
 
 static struct vmap_area *find_unlink_vmap_area(unsigned long addr)
 {
+	struct cpu_vmap_zone *z = addr_to_cvz(addr);
 	struct vmap_area *va;
 
-	spin_lock(&vmap_area_lock);
-	va = __find_vmap_area(addr, &vmap_area_root);
+	fbl_lock(z, BUSY);
+	va = __find_vmap_area(addr, &fbl_root(z, BUSY));
 	if (va)
-		unlink_va(va, &vmap_area_root);
-	spin_unlock(&vmap_area_lock);
+		unlink_va(va, &fbl_root(z, BUSY));
+	fbl_unlock(z, BUSY);
 
 	return va;
 }
@@ -2095,14 +2100,17 @@  static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
 
 static void free_vmap_block(struct vmap_block *vb)
 {
+	struct cpu_vmap_zone *z;
 	struct vmap_block *tmp;
 
 	tmp = xa_erase(&vmap_blocks, addr_to_vb_idx(vb->va->va_start));
 	BUG_ON(tmp != vb);
 
-	spin_lock(&vmap_area_lock);
-	unlink_va(vb->va, &vmap_area_root);
-	spin_unlock(&vmap_area_lock);
+	z = addr_to_cvz(vb->va->va_start);
+
+	fbl_lock(z, BUSY);
+	unlink_va(vb->va, &fbl_root(z, BUSY));
+	fbl_unlock(z, BUSY);
 
 	free_vmap_area_noflush(vb->va);
 	kfree_rcu(vb, rcu_head);
@@ -2484,9 +2492,11 @@  static inline void setup_vmalloc_vm_locked(struct vm_struct *vm,
 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
 			      unsigned long flags, const void *caller)
 {
-	spin_lock(&vmap_area_lock);
+	struct cpu_vmap_zone *z = addr_to_cvz(va->va_start);
+
+	fbl_lock(z, BUSY);
 	setup_vmalloc_vm_locked(vm, va, flags, caller);
-	spin_unlock(&vmap_area_lock);
+	fbl_unlock(z, BUSY);
 }
 
 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
@@ -3605,6 +3615,7 @@  static void vmap_ram_vread(char *buf, char *addr, int count, unsigned long flags
  */
 long vread(char *buf, char *addr, unsigned long count)
 {
+	struct cpu_vmap_zone *z;
 	struct vmap_area *va;
 	struct vm_struct *vm;
 	char *vaddr, *buf_start = buf;
@@ -3617,8 +3628,11 @@  long vread(char *buf, char *addr, unsigned long count)
 	if ((unsigned long) addr + count < count)
 		count = -(unsigned long) addr;
 
-	spin_lock(&vmap_area_lock);
-	va = find_vmap_area_exceed_addr((unsigned long)addr);
+	/* Hooked to CPU0 because a cache is not activated. */
+	z = &per_cpu(cpu_vmap_zone, 0);
+	fbl_lock(z, BUSY);
+
+	va = find_vmap_area_exceed_addr((unsigned long)addr, &fbl_root(z, BUSY));
 	if (!va)
 		goto finished;
 
@@ -3626,7 +3640,7 @@  long vread(char *buf, char *addr, unsigned long count)
 	if ((unsigned long)addr + count <= va->va_start)
 		goto finished;
 
-	list_for_each_entry_from(va, &vmap_area_list, list) {
+	list_for_each_entry_from(va, &fbl_head(z, BUSY), list) {
 		if (!count)
 			break;
 
@@ -3674,7 +3688,7 @@  long vread(char *buf, char *addr, unsigned long count)
 		count -= n;
 	}
 finished:
-	spin_unlock(&vmap_area_lock);
+	fbl_unlock(z, BUSY);
 
 	if (buf == buf_start)
 		return 0;
@@ -4014,14 +4028,15 @@  struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
 	}
 
 	/* insert all vm's */
-	spin_lock(&vmap_area_lock);
 	for (area = 0; area < nr_vms; area++) {
-		insert_vmap_area(vas[area], &vmap_area_root, &vmap_area_list);
+		struct cpu_vmap_zone *z = addr_to_cvz(vas[area]->va_start);
 
+		fbl_lock(z, BUSY);
+		insert_vmap_area(vas[area], &fbl_root(z, BUSY), &fbl_head(z, BUSY));
 		setup_vmalloc_vm_locked(vms[area], vas[area], VM_ALLOC,
 				 pcpu_get_vm_areas);
+		fbl_unlock(z, BUSY);
 	}
-	spin_unlock(&vmap_area_lock);
 
 	/*
 	 * Mark allocated areas as accessible. Do it now as a best-effort
@@ -4145,24 +4160,24 @@  bool vmalloc_dump_obj(void *object)
 #ifdef CONFIG_PROC_FS
 static void *s_start(struct seq_file *m, loff_t *pos)
 	__acquires(&vmap_purge_lock)
-	__acquires(&vmap_area_lock)
+	__acquires(&fbl(&per_cpu(cpu_vmap_zone, 0), BUSY, lock))
 {
 	mutex_lock(&vmap_purge_lock);
-	spin_lock(&vmap_area_lock);
+	fbl_lock((&per_cpu(cpu_vmap_zone, 0)), BUSY);
 
-	return seq_list_start(&vmap_area_list, *pos);
+	return seq_list_start(&fbl_head((&per_cpu(cpu_vmap_zone, 0)), BUSY), *pos);
 }
 
 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
 {
-	return seq_list_next(p, &vmap_area_list, pos);
+	return seq_list_next(p, &fbl_head((&per_cpu(cpu_vmap_zone, 0)), BUSY), pos);
 }
 
 static void s_stop(struct seq_file *m, void *p)
-	__releases(&vmap_area_lock)
+	__releases(&fbl(&per_cpu(cpu_vmap_zone, 0), BUSY, lock))
 	__releases(&vmap_purge_lock)
 {
-	spin_unlock(&vmap_area_lock);
+	fbl_unlock((&per_cpu(cpu_vmap_zone, 0)), BUSY);
 	mutex_unlock(&vmap_purge_lock);
 }
 
@@ -4258,7 +4273,7 @@  static int s_show(struct seq_file *m, void *p)
 	 * As a final step, dump "unpurged" areas.
 	 */
 final:
-	if (list_is_last(&va->list, &vmap_area_list))
+	if (list_is_last(&va->list, &fbl_head((&per_cpu(cpu_vmap_zone, 0)), BUSY)))
 		show_purge_info(m);
 
 	return 0;
@@ -4289,7 +4304,8 @@  static void vmap_init_free_space(void)
 {
 	unsigned long vmap_start = 1;
 	const unsigned long vmap_end = ULONG_MAX;
-	struct vmap_area *busy, *free;
+	struct vmap_area *free;
+	struct vm_struct *busy;
 
 	/*
 	 *     B     F     B     B     B     F
@@ -4297,12 +4313,12 @@  static void vmap_init_free_space(void)
 	 *  |           The KVA space           |
 	 *  |<--------------------------------->|
 	 */
-	list_for_each_entry(busy, &vmap_area_list, list) {
-		if (busy->va_start - vmap_start > 0) {
+	for (busy = vmlist; busy; busy = busy->next) {
+		if (busy->addr - vmap_start > 0) {
 			free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT);
 			if (!WARN_ON_ONCE(!free)) {
 				free->va_start = vmap_start;
-				free->va_end = busy->va_start;
+				free->va_end = (unsigned long) busy->addr;
 
 				insert_vmap_area_augment(free, NULL,
 					&free_vmap_area_root,
@@ -4310,7 +4326,7 @@  static void vmap_init_free_space(void)
 			}
 		}
 
-		vmap_start = busy->va_end;
+		vmap_start = (unsigned long) busy->addr + busy->size;
 	}
 
 	if (vmap_end - vmap_start > 0) {
@@ -4326,6 +4342,22 @@  static void vmap_init_free_space(void)
 	}
 }
 
+static void vmap_init_pcpu_zones(void)
+{
+	struct cpu_vmap_zone *z;
+	int i, j;
+
+	for_each_possible_cpu(i) {
+		z = per_cpu_ptr(&cpu_vmap_zone, i);
+
+		for (j = 0; j < ARRAY_SIZE(z->fbl); j++) {
+			INIT_LIST_HEAD(&z->fbl[j].head);
+			z->fbl[j].root = RB_ROOT;
+			spin_lock_init(&z->fbl[j].lock);
+		}
+	}
+}
+
 void __init vmalloc_init(void)
 {
 	struct vmap_area *va;
@@ -4349,8 +4381,15 @@  void __init vmalloc_init(void)
 		INIT_WORK(&p->wq, delayed_vfree_work);
 	}
 
+	/*
+	 * Setup per-cpu data before importing vmlist.
+	 */
+	vmap_init_pcpu_zones();
+
 	/* Import existing vmlist entries. */
 	for (tmp = vmlist; tmp; tmp = tmp->next) {
+		struct cpu_vmap_zone *z;
+
 		va = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT);
 		if (WARN_ON_ONCE(!va))
 			continue;
@@ -4358,7 +4397,9 @@  void __init vmalloc_init(void)
 		va->va_start = (unsigned long)tmp->addr;
 		va->va_end = va->va_start + tmp->size;
 		va->vm = tmp;
-		insert_vmap_area(va, &vmap_area_root, &vmap_area_list);
+
+		z = addr_to_cvz(va->va_start);
+		insert_vmap_area(va, &fbl_root(z, BUSY), &fbl_head(z, BUSY));
 	}
 
 	/*