pmalloc: support for 4K pages

Change-Id: Iecceffe06a92660976ebb58cd3cbec4be8931db0
Signed-off-by: Damjan Marion <damarion@cisco.com>

4 files changed, 35 insertions, 12 deletions

diff --git a/src/vlib/buffer.c b/src/vlib/buffer.c
index 22cf7999be7..ff91d3b920c 100644
--- a/src/vlib/buffer.c
+++ b/src/vlib/buffer.c
@@ -55,6 +55,9 @@ static u32 vlib_buffer_physmem_sz = 14 << 20;
 
 vlib_buffer_main_t buffer_main;
 
+/* logging */
+static vlib_log_class_t buffer_log_default;
+
 uword
 vlib_buffer_length_in_chain_slow_path (vlib_main_t * vm,
 				       vlib_buffer_t * b_first)
@@ -963,6 +966,9 @@ vlib_buffer_main_init (struct vlib_main_t * vm)
   clib_error_t *error;
   u32 physmem_map_index;
   u8 pool_index;
+  int log2_page_size = 0;
+
+  buffer_log_default = vlib_log_register_class ("buffer", 0);
 
   if (vlib_buffer_callbacks)
     {
@@ -981,10 +987,24 @@ vlib_buffer_main_init (struct vlib_main_t * vm)
     &vlib_buffer_delete_free_list_internal;
   clib_spinlock_init (&bm->buffer_known_hash_lockp);
 
-  if ((error = vlib_physmem_shared_map_create (vm, "buffers",
-					       vlib_buffer_physmem_sz,
-					       CLIB_PMALLOC_NUMA_LOCAL,
-					       &physmem_map_index)))
+retry:
+  error = vlib_physmem_shared_map_create (vm, "buffers",
+					  vlib_buffer_physmem_sz,
+					  log2_page_size,
+					  CLIB_PMALLOC_NUMA_LOCAL,
+					  &physmem_map_index);
+
+  if (error && log2_page_size == 0)
+    {
+      vlib_log_warn (buffer_log_default, "%U", format_clib_error, error);
+      clib_error_free (error);
+      vlib_log_warn (buffer_log_default, "falling back to non-hugepage "
+		     "backed buffer pool");
+      log2_page_size = min_log2 (clib_mem_get_page_size ());
+      goto retry;
+    }
+
+  if (error)
     return error;
 
   pool_index = vlib_buffer_register_physmem_map (vm, physmem_map_index);
diff --git a/src/vlib/linux/vfio.c b/src/vlib/linux/vfio.c
index d300a683dd7..1ed99ceee55 100644
--- a/src/vlib/linux/vfio.c
+++ b/src/vlib/linux/vfio.c
@@ -40,7 +40,7 @@ vfio_map_physmem_page (vlib_main_t * vm, void *addr)
   vlib_physmem_main_t *vpm = &vm->physmem_main;
   linux_vfio_main_t *lvm = &vfio_main;
   struct vfio_iommu_type1_dma_map dm = { 0 };
-  uword log2_page_size = vpm->pmalloc_main->log2_page_sz;
+  uword log2_page_size = vpm->pmalloc_main->def_log2_page_sz;
   uword physmem_start = pointer_to_uword (vpm->pmalloc_main->base);
 
   if (lvm->container_fd == -1)
diff --git a/src/vlib/physmem.c b/src/vlib/physmem.c
index 2599bce944d..f881e0a8e71 100755
--- a/src/vlib/physmem.c
+++ b/src/vlib/physmem.c
@@ -31,7 +31,8 @@
 
 clib_error_t *
 vlib_physmem_shared_map_create (vlib_main_t * vm, char *name, uword size,
-				u32 numa_node, u32 * map_index)
+				u32 log2_page_sz, u32 numa_node,
+				u32 * map_index)
 {
   clib_pmalloc_main_t *pm = vm->physmem_main.pmalloc_main;
   vlib_physmem_main_t *vpm = &vm->physmem_main;
@@ -41,7 +42,8 @@ vlib_physmem_shared_map_create (vlib_main_t * vm, char *name, uword size,
   void *va;
   uword i;
 
-  va = clib_pmalloc_create_shared_arena (pm, name, size, numa_node);
+  va = clib_pmalloc_create_shared_arena (pm, name, size, log2_page_sz,
+					 numa_node);
 
   if (va == 0)
     return clib_error_return (0, "%U", format_clib_error,
@@ -53,12 +55,13 @@ vlib_physmem_shared_map_create (vlib_main_t * vm, char *name, uword size,
   *map_index = map->index = map - vpm->maps;
   map->base = va;
   map->fd = a->fd;
-  map->n_pages = a->n_pages;
-  map->log2_page_size = a->log2_page_sz;
+  map->n_pages = a->n_pages * a->subpages_per_page;
+  map->log2_page_size = a->log2_subpage_sz;
 
   for (i = 0; i < a->n_pages; i++)
     {
-      uword pa = clib_pmalloc_get_pa (pm, (u8 *) va + (i << a->log2_page_sz));
+      uword pa =
+	clib_pmalloc_get_pa (pm, (u8 *) va + (i << a->log2_subpage_sz));
 
       /* maybe iova */
       if (pa == 0)
diff --git a/src/vlib/physmem_funcs.h b/src/vlib/physmem_funcs.h
index 0082f85c70d..70fb8e74fc7 100644
--- a/src/vlib/physmem_funcs.h
+++ b/src/vlib/physmem_funcs.h
@@ -42,8 +42,8 @@
 
 clib_error_t *vlib_physmem_init (vlib_main_t * vm);
 clib_error_t *vlib_physmem_shared_map_create (vlib_main_t * vm, char *name,
-					      uword size, u32 numa_node,
-					      u32 * map_index);
+					      uword size, u32 log2_page_sz,
+					      u32 numa_node, u32 * map_index);
 
 vlib_physmem_map_t *vlib_physmem_get_map (vlib_main_t * vm, u32 index);