vlib physmem rework

This patch adds supprot support for multiple numa-aware physmem regions.

Change-Id: I5c69a6f4da33c8ee21bdb8604d52fd2886f2327e
Signed-off-by: Damjan Marion <damarion@cisco.com>

5 files changed, 35 insertions, 18 deletions

diff --git a/src/plugins/dpdk/api/dpdk_api.c b/src/plugins/dpdk/api/dpdk_api.c
index 08afdd700ef..97c4bc75721 100755
--- a/src/plugins/dpdk/api/dpdk_api.c
+++ b/src/plugins/dpdk/api/dpdk_api.c
@@ -20,7 +20,6 @@
 
 #include <vnet/ethernet/ethernet.h>
 #include <dpdk/device/dpdk.h>
-#include <vlib/unix/physmem.h>
 #include <vlib/pci/pci.h>
 
 #include <stdio.h>
diff --git a/src/plugins/dpdk/device/init.c b/src/plugins/dpdk/device/init.c
index a795ba0e88e..e23542f7554 100755
--- a/src/plugins/dpdk/device/init.c
+++ b/src/plugins/dpdk/device/init.c
@@ -17,10 +17,10 @@
 #include <vppinfra/error.h>
 #include <vppinfra/format.h>
 #include <vppinfra/bitmap.h>
+#include <vlib/unix/unix.h>
 
 #include <vnet/ethernet/ethernet.h>
 #include <dpdk/device/dpdk.h>
-#include <vlib/unix/physmem.h>
 #include <vlib/pci/pci.h>
 
 #include <stdio.h>
@@ -1026,21 +1026,28 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input)
       clib_bitmap_foreach (c, tm->cpu_socket_bitmap, (
         {
 	  int pages_avail, page_size, mem;
+	  clib_error_t  *e = 0;
 
 	  vec_validate(mem_by_socket, c);
 	  mem = mem_by_socket[c];
 
 	  page_size = 1024;
-	  pages_avail = vlib_sysfs_get_free_hugepages(c, page_size * 1024);
+	  e = vlib_sysfs_get_free_hugepages(c, page_size * 1024, &pages_avail);
 
-	  if (pages_avail < 0 || page_size * pages_avail < mem)
+	  if (e != 0 || pages_avail < 0 || page_size * pages_avail < mem)
 	    use_1g = 0;
 
+	  if (e)
+	   clib_error_free (e);
+
 	  page_size = 2;
-	  pages_avail = vlib_sysfs_get_free_hugepages(c, page_size * 1024);
+	  e = vlib_sysfs_get_free_hugepages(c, page_size * 1024, &pages_avail);
 
-	  if (pages_avail < 0 || page_size * pages_avail < mem)
+	  if (e != 0 || pages_avail < 0 || page_size * pages_avail < mem)
 	    use_2m = 0;
+
+	  if (e)
+	   clib_error_free (e);
       }));
       /* *INDENT-ON* */
 
diff --git a/src/plugins/dpdk/hqos/hqos.c b/src/plugins/dpdk/hqos/hqos.c
index 813eb91cf99..c9b85652863 100644
--- a/src/plugins/dpdk/hqos/hqos.c
+++ b/src/plugins/dpdk/hqos/hqos.c
@@ -29,7 +29,6 @@
 #include <vnet/ethernet/ethernet.h>
 #include <dpdk/device/dpdk.h>
 
-#include <vlib/unix/physmem.h>
 #include <vlib/pci/pci.h>
 #include <vlibmemory/api.h>
 #include <vlibmemory/vl_memory_msg_enum.h>	/* enumerate all vlib messages */
diff --git a/src/plugins/ixge/ixge.c b/src/plugins/ixge/ixge.c
index e0150f41a70..222c148c67d 100644
--- a/src/plugins/ixge/ixge.c
+++ b/src/plugins/ixge/ixge.c
@@ -2493,10 +2493,11 @@ ixge_dma_init (ixge_device_t * xd, vlib_rx_or_tx_t rt, u32 queue_index)
     round_pow2 (xm->n_descriptors[rt], xm->n_descriptors_per_cache_line);
   dq->head_index = dq->tail_index = 0;
 
-  dq->descriptors = vlib_physmem_alloc_aligned (vm, &error,
-						dq->n_descriptors *
-						sizeof (dq->descriptors[0]),
-						128 /* per chip spec */ );
+  dq->descriptors =
+    vlib_physmem_alloc_aligned (vm, xm->physmem_region, &error,
+				dq->n_descriptors *
+				sizeof (dq->descriptors[0]),
+				128 /* per chip spec */ );
   if (error)
     return error;
 
@@ -2518,7 +2519,8 @@ ixge_dma_init (ixge_device_t * xd, vlib_rx_or_tx_t rt, u32 queue_index)
 	  vlib_buffer_t *b =
 	    vlib_get_buffer (vm, dq->descriptor_buffer_indices[i]);
 	  dq->descriptors[i].rx_to_hw.tail_address =
-	    vlib_physmem_virtual_to_physical (vm, b->data);
+	    vlib_physmem_virtual_to_physical (vm, xm->physmem_region,
+					      b->data);
 	}
     }
   else
@@ -2526,7 +2528,8 @@ ixge_dma_init (ixge_device_t * xd, vlib_rx_or_tx_t rt, u32 queue_index)
       u32 i;
 
       dq->tx.head_index_write_back =
-	vlib_physmem_alloc (vm, &error, CLIB_CACHE_LINE_BYTES);
+	vlib_physmem_alloc (vm, vm->buffer_main->physmem_region, &error,
+			    CLIB_CACHE_LINE_BYTES);
 
       for (i = 0; i < dq->n_descriptors; i++)
 	dq->descriptors[i].tx = xm->tx_descriptor_template;
@@ -2538,7 +2541,9 @@ ixge_dma_init (ixge_device_t * xd, vlib_rx_or_tx_t rt, u32 queue_index)
     ixge_dma_regs_t *dr = get_dma_regs (xd, rt, queue_index);
     u64 a;
 
-    a = vlib_physmem_virtual_to_physical (vm, dq->descriptors);
+    a =
+      vlib_physmem_virtual_to_physical (vm, vm->buffer_main->physmem_region,
+					dq->descriptors);
     dr->descriptor_address[0] = a & 0xFFFFFFFF;
     dr->descriptor_address[1] = a >> (u64) 32;
     dr->n_descriptor_bytes = dq->n_descriptors * sizeof (dq->descriptors[0]);
@@ -2564,7 +2569,9 @@ ixge_dma_init (ixge_device_t * xd, vlib_rx_or_tx_t rt, u32 queue_index)
 	dq->tx.head_index_write_back[0] = dq->head_index;
 
 	a =
-	  vlib_physmem_virtual_to_physical (vm, dq->tx.head_index_write_back);
+	  vlib_physmem_virtual_to_physical (vm,
+					    vm->buffer_main->physmem_region,
+					    dq->tx.head_index_write_back);
 	dr->tx.head_index_write_back_address[0] = /* enable bit */ 1 | a;
 	dr->tx.head_index_write_back_address[1] = (u64) a >> (u64) 32;
       }
@@ -2850,9 +2857,12 @@ ixge_pci_init (vlib_main_t * vm, vlib_pci_device_t * dev)
   void *r;
   ixge_device_t *xd;
 
-  /* Device found: make sure we have dma memory. */
-  if (unix_physmem_is_fake (vm))
-    return clib_error_return (0, "no physical memory available");
+  /* Allocate physmem region for DMA buffers */
+  error = vlib_physmem_region_alloc (vm, "ixge decriptors", 2 << 20, 0,
+				     VLIB_PHYSMEM_F_INIT_MHEAP,
+				     &xm->physmem_region);
+  if (error)
+    return error;
 
   error = vlib_pci_map_resource (dev, 0, &r);
   if (error)
diff --git a/src/plugins/ixge/ixge.h b/src/plugins/ixge/ixge.h
index 779603b391b..42c1bfa5c9b 100644
--- a/src/plugins/ixge/ixge.h
+++ b/src/plugins/ixge/ixge.h
@@ -1266,6 +1266,8 @@ typedef struct
   u32 *rx_buffers_to_add;
 
   f64 time_last_stats_update;
+
+  vlib_physmem_region_index_t physmem_region;
 } ixge_main_t;
 
 ixge_main_t ixge_main;