Tap: Fix the indirect buffers allocation VPP-1660

Indirect buffers are used to store indirect descriptors
to xmit big packets.
This patch moves the indirect buffer allocation from
interface creation to device node. Now it allocates
or deallocates buffers during tx for chained buffers.

Change-Id: I55cec208a2a7432e12fe9254a7f8ef84a9302bd5
Signed-off-by: Mohsin Kazmi <sykazmi@cisco.com>
(cherry picked from commit 55203e745f5e3f1f6c4dbe99d6eab8dee4d13ea6)

4 files changed, 61 insertions, 48 deletions

diff --git a/src/vnet/devices/virtio/device.c b/src/vnet/devices/virtio/device.c
index 04cf9ec49c8..e560c29e5e7 100644
--- a/src/vnet/devices/virtio/device.c
+++ b/src/vnet/devices/virtio/device.c
@@ -90,9 +90,8 @@ format_virtio_tx_trace (u8 * s, va_list * args)
   return s;
 }
 
-#ifndef CLIB_MARCH_VARIANT
-inline void
-virtio_free_used_desc (vlib_main_t * vm, virtio_vring_t * vring)
+static_always_inline void
+virtio_free_used_device_desc (vlib_main_t * vm, virtio_vring_t * vring)
 {
   u16 used = vring->desc_in_use;
   u16 sz = vring->size;
@@ -106,17 +105,35 @@ virtio_free_used_desc (vlib_main_t * vm, virtio_vring_t * vring)
   while (n_left)
     {
       struct vring_used_elem *e = &vring->used->ring[last & mask];
-      u16 slot = e->id;
+      u16 slot, n_buffers;
+      slot = n_buffers = e->id;
 
-      vlib_buffer_free (vm, &vring->buffers[slot], 1);
-      used--;
-      last++;
-      n_left--;
+      while (e->id == n_buffers)
+	{
+	  n_left--;
+	  last++;
+	  n_buffers++;
+	  if (n_left == 0)
+	    break;
+	  e = &vring->used->ring[last & mask];
+	}
+      vlib_buffer_free_from_ring (vm, vring->buffers, slot,
+				  sz, (n_buffers - slot));
+      used -= (n_buffers - slot);
+
+      if (n_left > 0)
+	{
+	  slot = e->id;
+
+	  vlib_buffer_free (vm, &vring->buffers[slot], 1);
+	  used--;
+	  last++;
+	  n_left--;
+	}
     }
   vring->desc_in_use = used;
   vring->last_used_idx = last;
 }
-#endif /* CLIB_MARCH_VARIANT */
 
 static_always_inline u16
 add_buffer_to_slot (vlib_main_t * vm, virtio_if_t * vif,
@@ -170,9 +187,15 @@ add_buffer_to_slot (vlib_main_t * vm, virtio_if_t * vif,
        * It can easily support 65535 bytes of Jumbo frames with
        * each data buffer size of 512 bytes minimum.
        */
-      vlib_buffer_t *indirect_desc =
-	vlib_get_buffer (vm, vring->indirect_buffers[next]);
+      u32 indirect_buffer = 0;
+      if (PREDICT_FALSE (vlib_buffer_alloc (vm, &indirect_buffer, 1) == 0))
+	return n_added;
+
+      vlib_buffer_t *indirect_desc = vlib_get_buffer (vm, indirect_buffer);
       indirect_desc->current_data = 0;
+      indirect_desc->flags |= VLIB_BUFFER_NEXT_PRESENT;
+      indirect_desc->next_buffer = bi;
+      bi = indirect_buffer;
 
       struct vring_desc *id =
 	(struct vring_desc *) vlib_buffer_get_current (indirect_desc);
@@ -261,7 +284,7 @@ virtio_interface_tx_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
     virtio_kick (vm, vring, vif);
 
   /* free consumed buffers */
-  virtio_free_used_desc (vm, vring);
+  virtio_free_used_device_desc (vm, vring);
 
   used = vring->desc_in_use;
   next = vring->desc_next;
diff --git a/src/vnet/devices/virtio/pci.c b/src/vnet/devices/virtio/pci.c
index 825cba1bb60..3736225b25c 100644
--- a/src/vnet/devices/virtio/pci.c
+++ b/src/vnet/devices/virtio/pci.c
@@ -694,20 +694,8 @@ virtio_pci_vring_init (vlib_main_t * vm, virtio_if_t * vif, u16 queue_num)
 
   ASSERT (vring->buffers == 0);
   vec_validate_aligned (vring->buffers, queue_size, CLIB_CACHE_LINE_BYTES);
-  ASSERT (vring->indirect_buffers == 0);
-  vec_validate_aligned (vring->indirect_buffers, queue_size,
-			CLIB_CACHE_LINE_BYTES);
   if (queue_num % 2)
     {
-      u32 n_alloc = 0;
-      do
-	{
-	  if (n_alloc < queue_size)
-	    n_alloc +=
-	      vlib_buffer_alloc (vm, vring->indirect_buffers + n_alloc,
-				 queue_size - n_alloc);
-	}
-      while (n_alloc != queue_size);
       virtio_log_debug (vim, vif, "tx-queue: number %u, size %u", queue_num,
 			queue_size);
     }
@@ -1246,7 +1234,6 @@ virtio_pci_delete_if (vlib_main_t * vm, virtio_if_t * vif)
 	virtio_free_rx_buffers (vm, vring);
       }
     vec_free (vring->buffers);
-    vec_free (vring->indirect_buffers);
     vlib_physmem_free (vm, vring->desc);
   }
 
@@ -1259,12 +1246,7 @@ virtio_pci_delete_if (vlib_main_t * vm, virtio_if_t * vif)
       {
 	virtio_free_used_desc (vm, vring);
       }
-    if (vring->queue_id % 2)
-      {
-	vlib_buffer_free_no_next (vm, vring->indirect_buffers, vring->size);
-      }
     vec_free (vring->buffers);
-    vec_free (vring->indirect_buffers);
     vlib_physmem_free (vm, vring->desc);
   }
 
diff --git a/src/vnet/devices/virtio/virtio.c b/src/vnet/devices/virtio/virtio.c
index 9fbbc1bf508..90acb75fa6d 100644
--- a/src/vnet/devices/virtio/virtio.c
+++ b/src/vnet/devices/virtio/virtio.c
@@ -113,20 +113,6 @@ virtio_vring_init (vlib_main_t * vm, virtio_if_t * vif, u16 idx, u16 sz)
   vring->queue_id = idx;
   ASSERT (vring->buffers == 0);
   vec_validate_aligned (vring->buffers, sz, CLIB_CACHE_LINE_BYTES);
-  ASSERT (vring->indirect_buffers == 0);
-  vec_validate_aligned (vring->indirect_buffers, sz, CLIB_CACHE_LINE_BYTES);
-  if (idx % 2)
-    {
-      u32 n_alloc = 0;
-      do
-	{
-	  if (n_alloc < sz)
-	    n_alloc +=
-	      vlib_buffer_alloc (vm, vring->indirect_buffers + n_alloc,
-				 sz - n_alloc);
-	}
-      while (n_alloc != sz);
-    }
 
   vring->size = sz;
   vring->call_fd = eventfd (0, EFD_NONBLOCK | EFD_CLOEXEC);
@@ -196,10 +182,35 @@ virtio_vring_free_rx (vlib_main_t * vm, virtio_if_t * vif, u32 idx)
   if (vring->avail)
     clib_mem_free (vring->avail);
   vec_free (vring->buffers);
-  vec_free (vring->indirect_buffers);
   return 0;
 }
 
+inline void
+virtio_free_used_desc (vlib_main_t * vm, virtio_vring_t * vring)
+{
+  u16 used = vring->desc_in_use;
+  u16 sz = vring->size;
+  u16 mask = sz - 1;
+  u16 last = vring->last_used_idx;
+  u16 n_left = vring->used->idx - last;
+
+  if (n_left == 0)
+    return;
+
+  while (n_left)
+    {
+      struct vring_used_elem *e = &vring->used->ring[last & mask];
+      u16 slot = e->id;
+
+      vlib_buffer_free (vm, &vring->buffers[slot], 1);
+      used--;
+      last++;
+      n_left--;
+    }
+  vring->desc_in_use = used;
+  vring->last_used_idx = last;
+}
+
 clib_error_t *
 virtio_vring_free_tx (vlib_main_t * vm, virtio_if_t * vif, u32 idx)
 {
@@ -218,9 +229,7 @@ virtio_vring_free_tx (vlib_main_t * vm, virtio_if_t * vif, u32 idx)
     clib_mem_free (vring->desc);
   if (vring->avail)
     clib_mem_free (vring->avail);
-  vlib_buffer_free_no_next (vm, vring->indirect_buffers, vring->size);
   vec_free (vring->buffers);
-  vec_free (vring->indirect_buffers);
   return 0;
 }
 
diff --git a/src/vnet/devices/virtio/virtio.h b/src/vnet/devices/virtio/virtio.h
index 5177dbabce4..57bb8d6c167 100644
--- a/src/vnet/devices/virtio/virtio.h
+++ b/src/vnet/devices/virtio/virtio.h
@@ -115,7 +115,6 @@ typedef struct
   u16 flags;
   u32 call_file_index;
   u32 *buffers;
-  u32 *indirect_buffers;
   u16 last_used_idx;
   u16 last_kick_avail_idx;
 } virtio_vring_t;