MFC r259232:

Create own free list for each of the first 32 possible allocation sizes.
In case of 4K allocation quantum that means for allocations up to 128K.

With growth of memory fragmentation these lists may grow to quite a large
sizes (tenths and hundreds of thousands items).  Having in one list items
of different sizes in worst case may require full linear list traversal,
that may be very expensive.  Having lists for items of single size means
that unless user specify some alignment or border requirements (that are
very rare cases) first item found on the list should satisfy the request.

While running SPEC NFS benchmark on top of ZFS on 24-core machine with
84GB RAM this change reduces CPU time spent in vmem_xalloc() from 8%
and lock congestion spinning around it from 20% to invisible levels.
And that all is by the cost of just 26 more pointers per vmem instance.

If at some point our kernel will start to actively use KVA allocations
with odd sizes above 128K, something may need to be done to bigger lists
also.

1 files changed, 17 insertions, 9 deletions

diff --git a/sys/kern/subr_vmem.c b/sys/kern/subr_vmem.c
index f3f3eec..3b70738 100644
--- a/sys/kern/subr_vmem.c
+++ b/sys/kern/subr_vmem.c
@@ -70,7 +70,10 @@ __FBSDID("$FreeBSD$");
 #include <vm/vm_param.h>
 #include <vm/vm_pageout.h>
 
-#define	VMEM_MAXORDER		(sizeof(vmem_size_t) * NBBY)
+#define	VMEM_OPTORDER		5
+#define	VMEM_OPTVALUE		(1 << VMEM_OPTORDER)
+#define	VMEM_MAXORDER						\
+    (VMEM_OPTVALUE - 1 + sizeof(vmem_size_t) * NBBY - VMEM_OPTORDER)
 
 #define	VMEM_HASHSIZE_MIN	16
 #define	VMEM_HASHSIZE_MAX	131072
@@ -200,8 +203,10 @@ static LIST_HEAD(, vmem) vmem_list = LIST_HEAD_INITIALIZER(vmem_list);
 #define	VMEM_CROSS_P(addr1, addr2, boundary) \
 	((((addr1) ^ (addr2)) & -(boundary)) != 0)
 
-#define	ORDER2SIZE(order)	((vmem_size_t)1 << (order))
-#define	SIZE2ORDER(size)	((int)flsl(size) - 1)
+#define	ORDER2SIZE(order)	((order) < VMEM_OPTVALUE ? ((order) + 1) : \
+    (vmem_size_t)1 << ((order) - (VMEM_OPTVALUE - VMEM_OPTORDER - 1)))
+#define	SIZE2ORDER(size)	((size) <= VMEM_OPTVALUE ? ((size) - 1) : \
+    (flsl(size) + (VMEM_OPTVALUE - VMEM_OPTORDER - 2)))
 
 /*
  * Maximum number of boundary tags that may be required to satisfy an
@@ -334,11 +339,14 @@ bt_free(vmem_t *vm, bt_t *bt)
 
 /*
  * freelist[0] ... [1, 1]
- * freelist[1] ... [2, 3]
- * freelist[2] ... [4, 7]
- * freelist[3] ... [8, 15]
+ * freelist[1] ... [2, 2]
  *  :
- * freelist[n] ... [(1 << n), (1 << (n + 1)) - 1]
+ * freelist[29] ... [30, 30]
+ * freelist[30] ... [31, 31]
+ * freelist[31] ... [32, 63]
+ * freelist[33] ... [64, 127]
+ *  :
+ * freelist[n] ... [(1 << (n - 26)), (1 << (n - 25)) - 1]
  *  :
  */
 
@@ -979,6 +987,7 @@ vmem_init(vmem_t *vm, const char *name, vmem_addr_t base, vmem_size_t size,
 	int i;
 
 	MPASS(quantum > 0);
+	MPASS((quantum & (quantum - 1)) == 0);
 
 	bzero(vm, sizeof(*vm));
 
@@ -988,8 +997,7 @@ vmem_init(vmem_t *vm, const char *name, vmem_addr_t base, vmem_size_t size,
 	LIST_INIT(&vm->vm_freetags);
 	strlcpy(vm->vm_name, name, sizeof(vm->vm_name));
 	vm->vm_quantum_mask = quantum - 1;
-	vm->vm_quantum_shift = SIZE2ORDER(quantum);
-	MPASS(ORDER2SIZE(vm->vm_quantum_shift) == quantum);
+	vm->vm_quantum_shift = flsl(quantum) - 1;
 	vm->vm_nbusytag = 0;
 	vm->vm_size = 0;
 	vm->vm_inuse = 0;