buffers: add page-size config

Type: feature

Add a `buffers {page-size}` parameter to specify page size
for buffers. This also fixes an issue with the parsing in
unformat_log2_page_size.

Change-Id: I7d7b1fa0bb7febaa7509cf2c625882f07eeafaad
Signed-off-by: Nathan Skrzypczak <nathan.skrzypczak@gmail.com>

5 files changed, 101 insertions, 72 deletions

diff --git a/docs/gettingstarted/users/configuring/startup.rst b/docs/gettingstarted/users/configuring/startup.rst
index 69d6ee5e3ea..e594b6a08a4 100644
--- a/docs/gettingstarted/users/configuring/startup.rst
+++ b/docs/gettingstarted/users/configuring/startup.rst
@@ -579,6 +579,7 @@ The buffers Section
    buffers {
       buffers-per-numa 128000
       default data-size 2048
+      page-size default-hugepage
    }
 
 buffers-per-numa number
@@ -601,6 +602,19 @@ Size of buffer data area, default is 2048
 
    default data-size 2048
 
+page-size number
+^^^^^^^^^^^^^^^^
+
+Set the page size for buffer allocation
+
+.. code-block:: console
+
+   page-size 4K
+   page-size 2M
+   page-size 1G
+   page-size default
+   page-size default-hugepage
+
 
 The dpdk Section
 ----------------
diff --git a/src/vlib/buffer.c b/src/vlib/buffer.c
index 17e7d24cb27..a30fca32d8b 100644
--- a/src/vlib/buffer.c
+++ b/src/vlib/buffer.c
@@ -685,72 +685,90 @@ vlib_buffer_worker_init (vlib_main_t * vm)
 VLIB_WORKER_INIT_FUNCTION (vlib_buffer_worker_init);
 
 static clib_error_t *
-vlib_buffer_main_init_numa_node (struct vlib_main_t *vm, u32 numa_node,
-				 u8 * index)
+vlib_buffer_main_init_numa_alloc (struct vlib_main_t *vm, u32 numa_node,
+				  u32 * physmem_map_index,
+				  clib_mem_page_sz_t log2_page_size,
+				  u8 unpriv)
 {
   vlib_buffer_main_t *bm = vm->buffer_main;
+  u32 buffers_per_numa = bm->buffers_per_numa;
   clib_error_t *error;
-  u32 physmem_map_index;
+  u32 buffer_size;
   uword n_pages, pagesize;
-  u32 buffers_per_numa;
-  u32 buffer_size = vlib_buffer_alloc_size (bm->ext_hdr_size,
-					    vlib_buffer_get_default_data_size
-					    (vm));
-  u8 *name;
-
-  pagesize = clib_mem_get_default_hugepage_size ();
-  name = format (0, "buffers-numa-%d%c", numa_node, 0);
+  u8 *name = 0;
 
-  buffers_per_numa = bm->buffers_per_numa ? bm->buffers_per_numa :
-    VLIB_BUFFER_DEFAULT_BUFFERS_PER_NUMA;
-
-retry:
+  ASSERT (log2_page_size != CLIB_MEM_PAGE_SZ_UNKNOWN);
 
+  pagesize = clib_mem_page_bytes (log2_page_size);
+  buffer_size = vlib_buffer_alloc_size (bm->ext_hdr_size,
+					vlib_buffer_get_default_data_size
+					(vm));
   if (buffer_size > pagesize)
-    {
-      error =
-	clib_error_return (0,
-			   "buffer size (%llu) is greater than page size (%llu)",
-			   buffer_size, pagesize);
-      goto done;
-    }
+    return clib_error_return (0, "buffer size (%llu) is greater than page "
+			      "size (%llu)", buffer_size, pagesize);
 
+  if (buffers_per_numa == 0)
+    buffers_per_numa = unpriv ? VLIB_BUFFER_DEFAULT_BUFFERS_PER_NUMA_UNPRIV :
+      VLIB_BUFFER_DEFAULT_BUFFERS_PER_NUMA;
+
+  name = format (0, "buffers-numa-%d%c", numa_node, 0);
   n_pages = (buffers_per_numa - 1) / (pagesize / buffer_size) + 1;
   error = vlib_physmem_shared_map_create (vm, (char *) name,
 					  n_pages * pagesize,
 					  min_log2 (pagesize), numa_node,
-					  &physmem_map_index);
+					  physmem_map_index);
+  vec_free (name);
+  return error;
+}
 
-  if (error && pagesize != clib_mem_get_page_size ())
+static clib_error_t *
+vlib_buffer_main_init_numa_node (struct vlib_main_t *vm, u32 numa_node,
+				 u8 * index)
+{
+  vlib_buffer_main_t *bm = vm->buffer_main;
+  u32 physmem_map_index;
+  clib_error_t *error;
+  u8 *name = 0;
+
+  if (bm->log2_page_size == CLIB_MEM_PAGE_SZ_UNKNOWN)
     {
-      vlib_log_warn (bm->log_default, "%U", format_clib_error, error);
+      error = vlib_buffer_main_init_numa_alloc (vm, numa_node,
+						&physmem_map_index,
+						CLIB_MEM_PAGE_SZ_DEFAULT_HUGE,
+						0 /* unpriv */ );
+      if (!error)
+	goto buffer_pool_create;
+
+      /* If alloc failed, retry without hugepages */
+      vlib_log_warn (bm->log_default,
+		     "numa[%u] falling back to non-hugepage backed "
+		     "buffer pool (%U)", numa_node, format_clib_error, error);
       clib_error_free (error);
-      vlib_log_warn (bm->log_default, "falling back to non-hugepage "
-		     "backed buffer pool");
-      pagesize = clib_mem_get_page_size ();
-      buffers_per_numa = bm->buffers_per_numa ? bm->buffers_per_numa :
-	VLIB_BUFFER_DEFAULT_BUFFERS_PER_NUMA_UNPRIV;
-      goto retry;
-    }
 
+      error = vlib_buffer_main_init_numa_alloc (vm, numa_node,
+						&physmem_map_index,
+						CLIB_MEM_PAGE_SZ_DEFAULT,
+						1 /* unpriv */ );
+    }
+  else
+    error = vlib_buffer_main_init_numa_alloc (vm, numa_node,
+					      &physmem_map_index,
+					      bm->log2_page_size,
+					      0 /* unpriv */ );
   if (error)
-    goto done;
+    return error;
 
-  vec_reset_length (name);
+buffer_pool_create:
   name = format (name, "default-numa-%d%c", numa_node, 0);
-
   *index = vlib_buffer_pool_create (vm, (char *) name,
 				    vlib_buffer_get_default_data_size (vm),
 				    physmem_map_index);
 
   if (*index == (u8) ~ 0)
-    {
-      error = clib_error_return (0, "maximum number of buffer pools reached");
-      goto done;
-    }
-
-done:
+    error = clib_error_return (0, "maximum number of buffer pools reached");
   vec_free (name);
+
+
   return error;
 }
 
@@ -935,11 +953,15 @@ vlib_buffers_configure (vlib_main_t * vm, unformat_input_t * input)
   vlib_buffer_main_alloc (vm);
 
   bm = vm->buffer_main;
+  bm->log2_page_size = CLIB_MEM_PAGE_SZ_UNKNOWN;
 
   while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
     {
       if (unformat (input, "buffers-per-numa %u", &bm->buffers_per_numa))
 	;
+      else if (unformat (input, "page-size %U", unformat_log2_page_size,
+			 &bm->log2_page_size))
+	;
       else if (unformat (input, "default data-size %u",
 			 &bm->default_data_size))
 	;
diff --git a/src/vlib/buffer.h b/src/vlib/buffer.h
index d1a00bb3a45..297240d45f7 100644
--- a/src/vlib/buffer.h
+++ b/src/vlib/buffer.h
@@ -469,6 +469,7 @@ typedef struct
   u32 buffers_per_numa;
   u16 ext_hdr_size;
   u32 default_data_size;
+  clib_mem_page_sz_t log2_page_size;
 
   /* logging */
   vlib_log_class_t log_default;
diff --git a/src/vpp/conf/startup.conf b/src/vpp/conf/startup.conf
index 864de11251f..270160217a0 100644
--- a/src/vpp/conf/startup.conf
+++ b/src/vpp/conf/startup.conf
@@ -99,6 +99,11 @@ cpu {
 	## Size of buffer data area
 	## Default is 2048
 	# default data-size 2048
+
+	## Size of the memory pages allocated for buffer data
+	## Default will try 'default-hugepage' then 'default'
+	## you can also pass a size in K/M/G e.g. '8M'
+	# page-size default-hugepage
 # }
 
 # dpdk {
diff --git a/src/vppinfra/std-formats.c b/src/vppinfra/std-formats.c
index 1667c93fd88..1616001f9c5 100644
--- a/src/vppinfra/std-formats.c
+++ b/src/vppinfra/std-formats.c
@@ -299,40 +299,27 @@ format_log2_page_size (u8 * s, va_list * va)
 __clib_export uword
 unformat_log2_page_size (unformat_input_t * input, va_list * va)
 {
-  uword amount, shift, c;
+  uword amount;
   clib_mem_page_sz_t *result = va_arg (*va, clib_mem_page_sz_t *);
 
-  if (unformat (input, "default"))
-    return CLIB_MEM_PAGE_SZ_DEFAULT;
-
   if (unformat (input, "default-hugepage"))
-    return CLIB_MEM_PAGE_SZ_DEFAULT_HUGE;
-
-  if (!unformat (input, "%wd%_", &amount))
-    return CLIB_MEM_PAGE_SZ_UNKNOWN;
-
-  c = unformat_get_input (input);
-  switch (c)
-    {
-    case 'k':
-    case 'K':
-      shift = 10;
-      break;
-    case 'm':
-    case 'M':
-      shift = 20;
-      break;
-    case 'g':
-    case 'G':
-      shift = 30;
-      break;
-    default:
-      shift = 0;
-      unformat_put_input (input);
-      break;
-    }
-
-  *result = min_log2 (amount) + shift;
+    *result = CLIB_MEM_PAGE_SZ_DEFAULT_HUGE;
+  else if (unformat (input, "default"))
+    *result = CLIB_MEM_PAGE_SZ_DEFAULT;
+  else if (unformat (input, "%wdk", &amount))
+    *result = min_log2 (amount) + 10;
+  else if (unformat (input, "%wdK", &amount))
+    *result = min_log2 (amount) + 10;
+  else if (unformat (input, "%wdm", &amount))
+    *result = min_log2 (amount) + 20;
+  else if (unformat (input, "%wdM", &amount))
+    *result = min_log2 (amount) + 20;
+  else if (unformat (input, "%wdg", &amount))
+    *result = min_log2 (amount) + 30;
+  else if (unformat (input, "%wdG", &amount))
+    *result = min_log2 (amount) + 30;
+  else
+    return 0;
   return 1;
 }