octeon: add support for Marvell Octeon9 SoC

Type: feature

Signed-off-by: Monendra Singh Kushwaha <kmonendra@marvell.com>
Change-Id: I5db58b8ec41b45596bc03b4a336a184c17871294
(cherry picked from commit a93c85a5793852b6edda20bc1100fa9fabd0eb29)

6 files changed, 106 insertions, 4 deletions

diff --git a/build/external/packages/octeon-roc.mk b/build/external/packages/octeon-roc.mk
index 62f5d823bdf..f96cd4c6e3d 100644
--- a/build/external/packages/octeon-roc.mk
+++ b/build/external/packages/octeon-roc.mk
@@ -2,9 +2,9 @@
 # SPDX-License-Identifier: Apache-2.0
 # https://spdx.org/licenses/Apache-2.0.html
 
-octeon-roc_version             := 0.3
+octeon-roc_version             := 0.4
 octeon-roc_tarball             := octeon-roc-v$(octeon-roc_version).tar.gz
-octeon-roc_tarball_md5sum      := e4a16beb76a6c63af1600dd4d1d752b8
+octeon-roc_tarball_md5sum      := f3772f6e05d71cf2709c00ff1a1929ca
 
 octeon-roc_tarball_strip_dirs  := 1
 octeon-roc_url                 := https://github.com/MarvellEmbeddedProcessors/marvell-vpp/archive/refs/tags/$(octeon-roc_tarball)
diff --git a/src/cmake/platform/octeon9.cmake b/src/cmake/platform/octeon9.cmake
new file mode 100644
index 00000000000..46ca7dfa64a
--- /dev/null
+++ b/src/cmake/platform/octeon9.cmake
@@ -0,0 +1,4 @@
+
+set(VPP_PLATFORM_CACHE_LINE_SIZE 128)
+set(VPP_PLATFORM_MARCH_FLAGS -march=armv8.2-a+crc+crypto)
+set(VPP_PLATFORM_BUFFER_ALIGN 128)
diff --git a/src/plugins/dev_octeon/CMakeLists.txt b/src/plugins/dev_octeon/CMakeLists.txt
index e8abf1a3389..0f6b32bbecd 100644
--- a/src/plugins/dev_octeon/CMakeLists.txt
+++ b/src/plugins/dev_octeon/CMakeLists.txt
@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # Copyright(c) 2022 Cisco Systems, Inc.
 
-if (NOT VPP_PLATFORM_NAME STREQUAL "octeon10")
+if (NOT VPP_PLATFORM_NAME STREQUAL "octeon10" AND NOT VPP_PLATFORM_NAME STREQUAL "octeon9")
   return()
 endif()
 
@@ -21,6 +21,10 @@ endif()
 
 include_directories (${OCTEON_ROC_DIR}/)
 
+if (VPP_PLATFORM_NAME STREQUAL "octeon9")
+   add_compile_definitions(PLATFORM_OCTEON9)
+endif()
+
 add_vpp_plugin(dev_octeon
   SOURCES
   init.c
diff --git a/src/plugins/dev_octeon/roc_helper.c b/src/plugins/dev_octeon/roc_helper.c
index f10c2cb578b..16e0a871a9d 100644
--- a/src/plugins/dev_octeon/roc_helper.c
+++ b/src/plugins/dev_octeon/roc_helper.c
@@ -49,6 +49,12 @@ oct_plt_get_thread_index (void)
   return __os_thread_index;
 }
 
+static u64
+oct_plt_get_cache_line_size (void)
+{
+  return CLIB_CACHE_LINE_BYTES;
+}
+
 static void
 oct_drv_physmem_free (vlib_main_t *vm, void *mem)
 {
@@ -178,4 +184,5 @@ oct_plt_init_param_t oct_plt_init_param = {
   .oct_plt_spinlock_unlock = oct_plt_spinlock_unlock,
   .oct_plt_spinlock_trylock = oct_plt_spinlock_trylock,
   .oct_plt_get_thread_index = oct_plt_get_thread_index,
+  .oct_plt_get_cache_line_size = oct_plt_get_cache_line_size,
 };
diff --git a/src/plugins/dev_octeon/rx_node.c b/src/plugins/dev_octeon/rx_node.c
index 997f1356199..1f8d5d93fa3 100644
--- a/src/plugins/dev_octeon/rx_node.c
+++ b/src/plugins/dev_octeon/rx_node.c
@@ -165,6 +165,38 @@ oct_rx_batch (vlib_main_t *vm, oct_rx_node_ctx_t *ctx,
   return n;
 }
 
+#ifdef PLATFORM_OCTEON9
+static_always_inline u32
+oct_rxq_refill (vlib_main_t *vm, vnet_dev_rx_queue_t *rxq, u16 n_refill)
+{
+  u32 n_alloc, n_free;
+  u32 buffer_indices[n_refill];
+  vlib_buffer_t *buffers[n_refill];
+  u8 bpi = vnet_dev_get_rx_queue_buffer_pool_index (rxq);
+  oct_rxq_t *crq = vnet_dev_get_rx_queue_data (rxq);
+  u64 aura = roc_npa_aura_handle_to_aura (crq->aura_handle);
+  const uint64_t addr =
+    roc_npa_aura_handle_to_base (crq->aura_handle) + NPA_LF_AURA_OP_FREE0;
+
+  if (n_refill < 256)
+    return 0;
+
+  n_alloc = vlib_buffer_alloc (vm, buffer_indices, n_refill);
+  if (PREDICT_FALSE (n_alloc < n_refill))
+    goto alloc_fail;
+
+  vlib_get_buffers (vm, buffer_indices, (vlib_buffer_t **) buffers, n_alloc);
+
+  for (n_free = 0; n_free < n_alloc; n_free++)
+    roc_store_pair ((u64) buffers[n_free], aura, addr);
+
+  return n_alloc;
+
+alloc_fail:
+  vlib_buffer_unalloc_to_pool (vm, buffer_indices, n_alloc, bpi);
+  return 0;
+}
+#else
 static_always_inline void
 oct_rxq_refill_batch (vlib_main_t *vm, u64 lmt_id, u64 addr,
 		      oct_npa_lf_aura_batch_free_line_t *lines, u32 *bi,
@@ -260,6 +292,7 @@ oct_rxq_refill (vlib_main_t *vm, vnet_dev_rx_queue_t *rxq, u16 n_refill)
 
   return n_enq;
 }
+#endif
 
 static_always_inline void
 oct_rx_trace (vlib_main_t *vm, vlib_node_runtime_t *node,
diff --git a/src/plugins/dev_octeon/tx_node.c b/src/plugins/dev_octeon/tx_node.c
index a2e4b07de8a..0907493814d 100644
--- a/src/plugins/dev_octeon/tx_node.c
+++ b/src/plugins/dev_octeon/tx_node.c
@@ -32,6 +32,44 @@ typedef struct
   lmt_line_t *lmt_lines;
 } oct_tx_ctx_t;
 
+#ifdef PLATFORM_OCTEON9
+static_always_inline u32
+oct_batch_free (vlib_main_t *vm, oct_tx_ctx_t *ctx, vnet_dev_tx_queue_t *txq)
+{
+  oct_txq_t *ctq = vnet_dev_get_tx_queue_data (txq);
+  u16 off = ctq->hdr_off;
+  u64 ah = ctq->aura_handle;
+  u32 n_freed = 0, n;
+
+  ah = ctq->aura_handle;
+
+  if ((n = roc_npa_aura_op_available (ah)) >= 32)
+    {
+      u64 buffers[n];
+      u32 bi[n];
+
+      n_freed = roc_npa_aura_op_bulk_alloc (ah, buffers, n, 0, 1);
+      vlib_get_buffer_indices_with_offset (vm, (void **) &buffers, bi, n_freed,
+					   off);
+      vlib_buffer_free_no_next (vm, bi, n_freed);
+    }
+
+  return n_freed;
+}
+
+static_always_inline void
+oct_lmt_copy (void *lmt_addr, u64 io_addr, void *desc, u64 dwords)
+{
+  u64 lmt_status;
+
+  do
+    {
+      roc_lmt_mov_seg (lmt_addr, desc, dwords);
+      lmt_status = roc_lmt_submit_ldeor (io_addr);
+    }
+  while (lmt_status == 0);
+}
+#else
 static_always_inline u32
 oct_batch_free (vlib_main_t *vm, oct_tx_ctx_t *ctx, vnet_dev_tx_queue_t *txq)
 {
@@ -133,6 +171,7 @@ oct_batch_free (vlib_main_t *vm, oct_tx_ctx_t *ctx, vnet_dev_tx_queue_t *txq)
 
   return n_freed;
 }
+#endif
 
 static_always_inline u8
 oct_tx_enq1 (vlib_main_t *vm, oct_tx_ctx_t *ctx, vlib_buffer_t *b,
@@ -158,6 +197,11 @@ oct_tx_enq1 (vlib_main_t *vm, oct_tx_ctx_t *ctx, vlib_buffer_t *b,
       return 0;
     }
 
+#ifdef PLATFORM_OCTEON9
+  /* Override line for Octeon9 */
+  line = ctx->lmt_lines;
+#endif
+
   if (!simple && flags & VLIB_BUFFER_NEXT_PRESENT)
     {
       u8 n_tail_segs = 0;
@@ -238,8 +282,12 @@ oct_tx_enq1 (vlib_main_t *vm, oct_tx_ctx_t *ctx, vlib_buffer_t *b,
       t->sw_if_index = vnet_buffer (b)->sw_if_index[VLIB_TX];
     }
 
+#ifdef PLATFORM_OCTEON9
+  oct_lmt_copy (line, ctx->lmt_ioaddr, &d, n_dwords);
+#else
   for (u32 i = 0; i < n_dwords; i++)
     line->dwords[i] = d.as_u128[i];
+#endif
 
   *dpl = n_dwords;
   *n = *n + 1;
@@ -252,7 +300,7 @@ oct_tx_enq16 (vlib_main_t *vm, oct_tx_ctx_t *ctx, vnet_dev_tx_queue_t *txq,
 	      vlib_buffer_t **b, u32 n_pkts, int trace)
 {
   u8 dwords_per_line[16], *dpl = dwords_per_line;
-  u64 lmt_arg, ioaddr, n_lines;
+  u64 __attribute__ ((unused)) lmt_arg, ioaddr, n_lines;
   u32 n_left, or_flags_16 = 0, n = 0;
   const u32 not_simple_flags =
     VLIB_BUFFER_NEXT_PRESENT | VNET_BUFFER_F_OFFLOAD;
@@ -331,6 +379,7 @@ oct_tx_enq16 (vlib_main_t *vm, oct_tx_ctx_t *ctx, vnet_dev_tx_queue_t *txq,
   if (PREDICT_FALSE (!n_lines))
     return n_pkts;
 
+#ifndef PLATFORM_OCTEON9
   if (PREDICT_FALSE (or_flags_16 & VLIB_BUFFER_NEXT_PRESENT))
     {
       dpl = dwords_per_line;
@@ -359,6 +408,7 @@ oct_tx_enq16 (vlib_main_t *vm, oct_tx_ctx_t *ctx, vnet_dev_tx_queue_t *txq,
     }
 
   roc_lmt_submit_steorl (lmt_arg, ioaddr);
+#endif
 
   return n_pkts;
 }
@@ -375,7 +425,11 @@ VNET_DEV_NODE_FN (oct_tx_node)
   u32 *from = vlib_frame_vector_args (frame);
   u32 n, n_enq, n_left, n_pkts = frame->n_vectors;
   vlib_buffer_t *buffers[VLIB_FRAME_SIZE + 8], **b = buffers;
+#ifdef PLATFORM_OCTEON9
+  u64 lmt_id = 0;
+#else
   u64 lmt_id = vm->thread_index << ROC_LMT_LINES_PER_CORE_LOG2;
+#endif
 
   oct_tx_ctx_t ctx = {
     .node = node,