misc: add callback hooks and refactor pmc

Callbacks for monitoring and performance measurement:
- Add new callback list type, with context
- Add callbacks for API, CLI, and barrier sync
- Modify node dispatch callback to pass plugin-specific context
- Modify perfmon plugin to keep PMC samples local to the plugin
- Include process nodes in dispatch callback
- Pass dispatch function return value to callback

Type: refactor

Signed-off-by: Tom Seidenberg <tseidenb@cisco.com>
Change-Id: I28b06c58490611e08d76ff5b01b2347ba2109b22

9 files changed, 326 insertions, 164 deletions

diff --git a/src/plugins/mdata/mdata.c b/src/plugins/mdata/mdata.c
index fc5bbfbb571..f74564eb33c 100644
--- a/src/plugins/mdata/mdata.c
+++ b/src/plugins/mdata/mdata.c
@@ -21,6 +21,7 @@
 
 #include <vlibapi/api.h>
 #include <vlibmemory/api.h>
+#include <vppinfra/callback_data.h>
 #include <vpp/app/version.h>
 #include <stdbool.h>
 
@@ -42,9 +43,8 @@ static mdata_t mdata_none;
     before_or_after: 0 => before, 1=> after
 */
 static void
-mdata_trace_callback (vlib_main_t * vm, u64 * c0, u64 * c1,
-		      vlib_node_runtime_t * node,
-		      vlib_frame_t * frame, int before_or_after)
+mdata_trace_callback (vlib_node_runtime_perf_callback_data_t * data,
+		      vlib_node_runtime_perf_callback_args_t * args)
 {
   int i;
   mdata_main_t *mm = &mdata_main;
@@ -53,6 +53,12 @@ mdata_trace_callback (vlib_main_t * vm, u64 * c0, u64 * c1,
   u32 n_left_from;
   mdata_t *before, *modifies;
   u8 *after;
+  vlib_main_t *vm = args->vm;
+  vlib_frame_t *frame = args->frame;
+  vlib_node_runtime_t *node = args->node;
+
+  if (PREDICT_FALSE (args->call_type == VLIB_NODE_RUNTIME_PERF_RESET))
+    return;
 
   /* Input nodes don't have frames, etc. */
   if (frame == 0)
@@ -68,7 +74,7 @@ mdata_trace_callback (vlib_main_t * vm, u64 * c0, u64 * c1,
   vlib_get_buffers (vm, from, bufs, n_left_from);
   b = bufs;
 
-  if (before_or_after == 1 /* after */ )
+  if (args->call_type == VLIB_NODE_RUNTIME_PERF_AFTER)
     goto after_pass;
 
   /* Resize the per-thread "before" vector to cover the current frame */
@@ -152,11 +158,9 @@ mdata_enable_disable (mdata_main_t * mmp, int enable_disable)
       if (vlib_mains[i] == 0)
 	continue;
 
-      clib_callback_enable_disable
-	(vlib_mains[i]->vlib_node_runtime_perf_counter_cbs,
-	 vlib_mains[i]->vlib_node_runtime_perf_counter_cb_tmp,
-	 vlib_mains[i]->worker_thread_main_loop_callback_lock,
-	 (void *) mdata_trace_callback, enable_disable);
+      clib_callback_data_enable_disable
+	(&vlib_mains[i]->vlib_node_runtime_perf_callbacks,
+	 mdata_trace_callback, enable_disable);
     }
 
   return rv;
diff --git a/src/plugins/perfmon/CMakeLists.txt b/src/plugins/perfmon/CMakeLists.txt
index a3f045f75f3..69e225b4a3f 100644
--- a/src/plugins/perfmon/CMakeLists.txt
+++ b/src/plugins/perfmon/CMakeLists.txt
@@ -11,7 +11,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-add_vpp_plugin(perfmon
+add_vpp_library (perfcore
   SOURCES
   perfmon.c
   perfmon_periodic.c
@@ -32,6 +32,22 @@ add_vpp_plugin(perfmon
   perfmon_intel_wsm_ep_dp.c
   perfmon_intel_wsm_ep_sp.c
   perfmon_intel_wsm_ex.c
+
+  INSTALL_HEADERS
+  perfmon.h
+
+  LINK_LIBRARIES
+  vppinfra
+  vlib
+  vnet
+)
+
+add_vpp_plugin(perfmon
+  SOURCES
+  perfmon_plugin.c
+
+  LINK_LIBRARIES
+  perfcore
 )
 
 option(VPP_BUILD_MAPFILE_TOOL "Build perfmon mapfile utility." OFF)
diff --git a/src/plugins/perfmon/perfmon.c b/src/plugins/perfmon/perfmon.c
index 7e276c30810..525a864b584 100644
--- a/src/plugins/perfmon/perfmon.c
+++ b/src/plugins/perfmon/perfmon.c
@@ -16,7 +16,6 @@
  */
 
 #include <vnet/vnet.h>
-#include <vnet/plugin/plugin.h>
 #include <perfmon/perfmon.h>
 #include <perfmon/perfmon_intel.h>
 
@@ -98,6 +97,7 @@ perfmon_init (vlib_main_t * vm)
   u32 cpuid;
   u8 model, stepping;
   perfmon_intel_pmc_event_t *ev;
+  int i;
 
   pm->vlib_main = vm;
   pm->vnet_main = vnet_get_main ();
@@ -109,9 +109,17 @@ perfmon_init (vlib_main_t * vm)
 
   /* Default data collection interval */
   pm->timeout_interval = 2.0;	/* seconds */
-  vec_validate (pm->pm_fds, 1);
-  vec_validate (pm->perf_event_pages, 1);
-  vec_validate (pm->rdpmc_indices, 1);
+
+  vec_validate (pm->threads, vlib_get_thread_main ()->n_vlib_mains - 1);
+  for (i = 0; i < vec_len (pm->threads); i++)
+    {
+      perfmon_thread_t *pt = clib_mem_alloc_aligned
+	(sizeof (perfmon_thread_t), CLIB_CACHE_LINE_BYTES);
+      clib_memset (pt, 0, sizeof (*pt));
+      pm->threads[i] = pt;
+      pt->pm_fds[0] = -1;
+      pt->pm_fds[1] = -1;
+    }
   pm->page_size = getpagesize ();
 
   pm->perfmon_table = 0;
@@ -147,18 +155,7 @@ perfmon_init (vlib_main_t * vm)
 
 VLIB_INIT_FUNCTION (perfmon_init);
 
-/* *INDENT-OFF* */
-VLIB_PLUGIN_REGISTER () =
-{
-  .version = VPP_BUILD_VER,
-  .description = "Performance Monitor",
-#if !defined(__x86_64__)
-  .default_disabled = 1,
-#endif
-};
-/* *INDENT-ON* */
-
-static uword
+uword
 unformat_processor_event (unformat_input_t * input, va_list * args)
 {
   perfmon_main_t *pm = va_arg (*args, perfmon_main_t *);
@@ -185,6 +182,10 @@ unformat_processor_event (unformat_input_t * input, va_list * args)
 
   pe_config |= pm->perfmon_table[idx].event_code[0];
   pe_config |= pm->perfmon_table[idx].umask << 8;
+  pe_config |= pm->perfmon_table[idx].edge << 18;
+  pe_config |= pm->perfmon_table[idx].anyt << 21;
+  pe_config |= pm->perfmon_table[idx].inv << 23;
+  pe_config |= pm->perfmon_table[idx].cmask << 24;
 
   ep->name = (char *) hp->key;
   ep->pe_type = PERF_TYPE_RAW;
diff --git a/src/plugins/perfmon/perfmon.h b/src/plugins/perfmon/perfmon.h
index 000e3c2849c..c8782023597 100644
--- a/src/plugins/perfmon/perfmon.h
+++ b/src/plugins/perfmon/perfmon.h
@@ -78,6 +78,32 @@ typedef struct
 
 typedef struct
 {
+  u64 ticks[2];
+  u64 vectors;
+} perfmon_counters_t;
+
+typedef struct
+{
+  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
+
+  /* Current counters */
+  u64 c[2];
+
+  /* Current perf_event file descriptors, per thread */
+  int pm_fds[2];
+
+  /* mmap base of mapped struct perf_event_mmap_page */
+  u8 *perf_event_pages[2];
+
+  u32 rdpmc_indices[2];
+
+  /* vector of counters by node index */
+  perfmon_counters_t *counters;
+
+} perfmon_thread_t;
+
+typedef struct
+{
   /* API message ID base */
   u16 msg_id_base;
 
@@ -112,17 +138,15 @@ typedef struct
   /* Current event (index) being collected */
   u32 current_event;
   int n_active;
-  u32 **rdpmc_indices;
-  /* mmap base / size of (mapped) struct perf_event_mmap_page */
-  u8 ***perf_event_pages;
+  /* mmap size of (mapped) struct perf_event_mmap_page */
   u32 page_size;
 
-  /* Current perf_event file descriptors, per thread */
-  int **pm_fds;
-
   /* thread bitmap */
   uword *thread_bitmap;
 
+  /* per-thread data */
+  perfmon_thread_t **threads;
+
   /* Logging */
   vlib_log_class_t log_class;
 
@@ -137,6 +161,8 @@ extern perfmon_main_t perfmon_main;
 extern vlib_node_registration_t perfmon_periodic_node;
 uword *perfmon_parse_table (perfmon_main_t * pm, char *path, char *filename);
 
+uword unformat_processor_event (unformat_input_t * input, va_list * args);
+
 /* Periodic function events */
 #define PERFMON_START 1
 
diff --git a/src/plugins/perfmon/perfmon_intel.h b/src/plugins/perfmon/perfmon_intel.h
index 6bb849244d5..475309124ea 100644
--- a/src/plugins/perfmon/perfmon_intel.h
+++ b/src/plugins/perfmon/perfmon_intel.h
@@ -25,6 +25,10 @@ typedef struct
 {
   u8 event_code[2];
   u8 umask;
+  u8 cmask;
+  u8 inv;
+  u8 anyt;
+  u8 edge;
   char *event_name;
 } perfmon_intel_pmc_event_t;
 
diff --git a/src/plugins/perfmon/perfmon_intel_skl.c b/src/plugins/perfmon/perfmon_intel_skl.c
index 726dbb4dd8c..b1c03140651 100644
--- a/src/plugins/perfmon/perfmon_intel_skl.c
+++ b/src/plugins/perfmon/perfmon_intel_skl.c
@@ -88,6 +88,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x0D},
    .umask = 0x01,
+   .anyt = 1,
    .event_name = "int_misc.recovery_cycles_any",
    },
   {
@@ -103,6 +104,8 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x0E},
    .umask = 0x01,
+   .cmask = 1,
+   .inv = 1,
    .event_name = "uops_issued.stall_cycles",
    },
   {
@@ -233,6 +236,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x3C},
    .umask = 0x00,
+   .anyt = 1,
    .event_name = "cpu_clk_unhalted.thread_p_any",
    },
   {
@@ -248,6 +252,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x3C},
    .umask = 0x01,
+   .anyt = 1,
    .event_name = "cpu_clk_thread_unhalted.ref_xclk_any",
    },
   {
@@ -268,6 +273,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x48},
    .umask = 0x01,
+   .cmask = 1,
    .event_name = "l1d_pend_miss.pending",
    },
   {
@@ -308,6 +314,12 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x49},
    .umask = 0x10,
+   .cmask = 1,
+   .event_name = "dtlb_store_misses.walk_active",
+   },
+  {
+   .event_code = {0x49},
+   .umask = 0x10,
    .event_name = "dtlb_store_misses.walk_pending",
    },
   {
@@ -403,6 +415,8 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x5E},
    .umask = 0x01,
+   .cmask = 1,
+   .inv = 1,
    .event_name = "rs_events.empty_end",
    },
   {
@@ -413,6 +427,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x60},
    .umask = 0x01,
+   .cmask = 1,
    .event_name = "offcore_requests_outstanding.cycles_with_demand_data_rd",
    },
   {
@@ -423,6 +438,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x60},
    .umask = 0x02,
+   .cmask = 1,
    .event_name = "offcore_requests_outstanding.cycles_with_demand_code_rd",
    },
   {
@@ -433,6 +449,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x60},
    .umask = 0x04,
+   .cmask = 1,
    .event_name = "offcore_requests_outstanding.cycles_with_demand_rfo",
    },
   {
@@ -443,6 +460,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x60},
    .umask = 0x08,
+   .cmask = 1,
    .event_name = "offcore_requests_outstanding.cycles_with_data_rd",
    },
   {
@@ -458,6 +476,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x79},
    .umask = 0x04,
+   .cmask = 1,
    .event_name = "idq.mite_cycles",
    },
   {
@@ -468,6 +487,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x79},
    .umask = 0x08,
+   .cmask = 1,
    .event_name = "idq.dsb_cycles",
    },
   {
@@ -478,11 +498,13 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x79},
    .umask = 0x18,
+   .cmask = 4,
    .event_name = "idq.all_dsb_cycles_4_uops",
    },
   {
    .event_code = {0x79},
    .umask = 0x18,
+   .cmask = 1,
    .event_name = "idq.all_dsb_cycles_any_uops",
    },
   {
@@ -503,11 +525,13 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x79},
    .umask = 0x30,
+   .cmask = 1,
    .event_name = "idq.ms_cycles",
    },
   {
    .event_code = {0x79},
    .umask = 0x30,
+   .edge = 1,
    .event_name = "idq.ms_switches",
    },
   {
@@ -588,26 +612,32 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x9C},
    .umask = 0x01,
+   .cmask = 4,
    .event_name = "idq_uops_not_delivered.cycles_0_uops_deliv.core",
    },
   {
    .event_code = {0x9C},
    .umask = 0x01,
+   .cmask = 3,
    .event_name = "idq_uops_not_delivered.cycles_le_1_uop_deliv.core",
    },
   {
    .event_code = {0x9C},
    .umask = 0x01,
+   .cmask = 4,
    .event_name = "idq_uops_not_delivered.cycles_le_2_uop_deliv.core",
    },
   {
    .event_code = {0x9C},
    .umask = 0x01,
+   .cmask = 1,
    .event_name = "idq_uops_not_delivered.cycles_le_3_uop_deliv.core",
    },
   {
    .event_code = {0x9C},
    .umask = 0x01,
+   .cmask = 1,
+   .inv = 1,
    .event_name = "idq_uops_not_delivered.cycles_fe_was_ok",
    },
   {
@@ -663,36 +693,43 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xA3},
    .umask = 0x01,
+   .cmask = 1,
    .event_name = "cycle_activity.cycles_l2_miss",
    },
   {
    .event_code = {0xA3},
    .umask = 0x04,
+   .cmask = 4,
    .event_name = "cycle_activity.stalls_total",
    },
   {
    .event_code = {0xA3},
    .umask = 0x05,
+   .cmask = 5,
    .event_name = "cycle_activity.stalls_l2_miss",
    },
   {
    .event_code = {0xA3},
    .umask = 0x08,
+   .cmask = 8,
    .event_name = "cycle_activity.cycles_l1d_miss",
    },
   {
    .event_code = {0xA3},
    .umask = 0x0C,
+   .cmask = 12,
    .event_name = "cycle_activity.stalls_l1d_miss",
    },
   {
    .event_code = {0xA3},
    .umask = 0x10,
+   .cmask = 16,
    .event_name = "cycle_activity.cycles_mem_any",
    },
   {
    .event_code = {0xA3},
    .umask = 0x14,
+   .cmask = 20,
    .event_name = "cycle_activity.stalls_mem_any",
    },
   {
@@ -733,11 +770,13 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xA8},
    .umask = 0x01,
+   .cmask = 1,
    .event_name = "lsd.cycles_active",
    },
   {
    .event_code = {0xA8},
    .umask = 0x01,
+   .cmask = 4,
    .event_name = "lsd.cycles_4_uops",
    },
   {
@@ -788,26 +827,32 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xB1},
    .umask = 0x01,
+   .cmask = 1,
+   .inv = 1,
    .event_name = "uops_executed.stall_cycles",
    },
   {
    .event_code = {0xB1},
    .umask = 0x01,
+   .cmask = 1,
    .event_name = "uops_executed.cycles_ge_1_uop_exec",
    },
   {
    .event_code = {0xB1},
    .umask = 0x01,
+   .cmask = 2,
    .event_name = "uops_executed.cycles_ge_2_uops_exec",
    },
   {
    .event_code = {0xB1},
    .umask = 0x01,
+   .cmask = 3,
    .event_name = "uops_executed.cycles_ge_3_uops_exec",
    },
   {
    .event_code = {0xB1},
    .umask = 0x01,
+   .cmask = 4,
    .event_name = "uops_executed.cycles_ge_4_uops_exec",
    },
   {
@@ -818,26 +863,32 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xB1},
    .umask = 0x02,
+   .cmask = 1,
    .event_name = "uops_executed.core_cycles_ge_1",
    },
   {
    .event_code = {0xB1},
    .umask = 0x02,
+   .cmask = 2,
    .event_name = "uops_executed.core_cycles_ge_2",
    },
   {
    .event_code = {0xB1},
    .umask = 0x02,
+   .cmask = 3,
    .event_name = "uops_executed.core_cycles_ge_3",
    },
   {
    .event_code = {0xB1},
    .umask = 0x02,
+   .cmask = 4,
    .event_name = "uops_executed.core_cycles_ge_4",
    },
   {
    .event_code = {0xB1},
    .umask = 0x02,
+   .cmask = 1,
+   .inv = 1,
    .event_name = "uops_executed.core_cycles_none",
    },
   {
@@ -873,6 +924,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xC0},
    .umask = 0x01,
+   .cmask = 10,
    .event_name = "inst_retired.total_cycles_ps",
    },
   {
@@ -883,16 +935,22 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xC2},
    .umask = 0x02,
+   .cmask = 1,
+   .inv = 1,
    .event_name = "uops_retired.stall_cycles",
    },
   {
    .event_code = {0xC2},
    .umask = 0x02,
+   .cmask = 10,
+   .inv = 1,
    .event_name = "uops_retired.total_cycles",
    },
   {
    .event_code = {0xC3},
    .umask = 0x01,
+   .cmask = 1,
+   .edge = 1,
    .event_name = "machine_clears.count",
    },
   {
@@ -1083,6 +1141,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xCA},
    .umask = 0x1E,
+   .cmask = 1,
    .event_name = "fp_assist.any",
    },
   {
diff --git a/src/plugins/perfmon/perfmon_intel_skx.c b/src/plugins/perfmon/perfmon_intel_skx.c
index 399174477ac..9de202d22a3 100644
--- a/src/plugins/perfmon/perfmon_intel_skx.c
+++ b/src/plugins/perfmon/perfmon_intel_skx.c
@@ -88,6 +88,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x0D},
    .umask = 0x01,
+   .anyt = 1,
    .event_name = "int_misc.recovery_cycles_any",
    },
   {
@@ -98,6 +99,8 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x0E},
    .umask = 0x01,
+   .cmask = 1,
+   .inv = 1,
    .event_name = "uops_issued.stall_cycles",
    },
   {
@@ -253,6 +256,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x3C},
    .umask = 0x00,
+   .anyt = 1,
    .event_name = "cpu_clk_unhalted.thread_p_any",
    },
   {
@@ -268,6 +272,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x3C},
    .umask = 0x01,
+   .anyt = 1,
    .event_name = "cpu_clk_thread_unhalted.ref_xclk_any",
    },
   {
@@ -288,6 +293,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x48},
    .umask = 0x01,
+   .cmask = 1,
    .event_name = "l1d_pend_miss.pending_cycles",
    },
   {
@@ -328,6 +334,12 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x49},
    .umask = 0x10,
+   .cmask = 1,
+   .event_name = "dtlb_store_misses.walk_active",
+   },
+  {
+   .event_code = {0x49},
+   .umask = 0x10,
    .event_name = "dtlb_store_misses.walk_pending",
    },
   {
@@ -418,6 +430,8 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x5E},
    .umask = 0x01,
+   .cmask = 1,
+   .inv = 1,
    .event_name = "rs_events.empty_end",
    },
   {
@@ -428,6 +442,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x60},
    .umask = 0x01,
+   .cmask = 1,
    .event_name = "offcore_requests_outstanding.cycles_with_demand_data_rd",
    },
   {
@@ -443,6 +458,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x60},
    .umask = 0x02,
+   .cmask = 1,
    .event_name = "offcore_requests_outstanding.cycles_with_demand_code_rd",
    },
   {
@@ -453,11 +469,13 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x60},
    .umask = 0x04,
+   .cmask = 1,
    .event_name = "offcore_requests_outstanding.cycles_with_demand_rfo",
    },
   {
    .event_code = {0x60},
    .umask = 0x08,
+   .cmask = 1,
    .event_name = "offcore_requests_outstanding.cycles_with_data_rd",
    },
   {
@@ -473,6 +491,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x79},
    .umask = 0x04,
+   .cmask = 1,
    .event_name = "idq.mite_cycles",
    },
   {
@@ -483,6 +502,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x79},
    .umask = 0x08,
+   .cmask = 1,
    .event_name = "idq.dsb_cycles",
    },
   {
@@ -498,11 +518,13 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x79},
    .umask = 0x18,
+   .cmask = 1,
    .event_name = "idq.all_dsb_cycles_any_uops",
    },
   {
    .event_code = {0x79},
    .umask = 0x18,
+   .cmask = 4,
    .event_name = "idq.all_dsb_cycles_4_uops",
    },
   {
@@ -523,6 +545,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x79},
    .umask = 0x30,
+   .cmask = 1,
    .event_name = "idq.ms_cycles",
    },
   {
@@ -533,6 +556,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x79},
    .umask = 0x30,
+   .edge = 1,
    .event_name = "idq.ms_switches",
    },
   {
@@ -603,26 +627,32 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0x9C},
    .umask = 0x01,
+   .cmask = 1,
+   .inv = 1,
    .event_name = "idq_uops_not_delivered.cycles_fe_was_ok",
    },
   {
    .event_code = {0x9C},
    .umask = 0x01,
+   .cmask = 1,
    .event_name = "idq_uops_not_delivered.cycles_le_3_uop_deliv.core",
    },
   {
    .event_code = {0x9C},
    .umask = 0x01,
+   .cmask = 2,
    .event_name = "idq_uops_not_delivered.cycles_le_2_uop_deliv.core",
    },
   {
    .event_code = {0x9C},
    .umask = 0x01,
+   .cmask = 3,
    .event_name = "idq_uops_not_delivered.cycles_le_1_uop_deliv.core",
    },
   {
    .event_code = {0x9C},
    .umask = 0x01,
+   .cmask = 4,
    .event_name = "idq_uops_not_delivered.cycles_0_uops_deliv.core",
    },
   {
@@ -683,36 +713,43 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xA3},
    .umask = 0x01,
+   .cmask = 1,
    .event_name = "cycle_activity.cycles_l2_miss",
    },
   {
    .event_code = {0xA3},
    .umask = 0x04,
+   .cmask = 4,
    .event_name = "cycle_activity.stalls_total",
    },
   {
    .event_code = {0xA3},
    .umask = 0x05,
+   .cmask = 5,
    .event_name = "cycle_activity.stalls_l2_miss",
    },
   {
    .event_code = {0xA3},
    .umask = 0x08,
+   .cmask = 8,
    .event_name = "cycle_activity.cycles_l1d_miss",
    },
   {
    .event_code = {0xA3},
    .umask = 0x0C,
+   .cmask = 12,
    .event_name = "cycle_activity.stalls_l1d_miss",
    },
   {
    .event_code = {0xA3},
    .umask = 0x10,
+   .cmask = 16,
    .event_name = "cycle_activity.cycles_mem_any",
    },
   {
    .event_code = {0xA3},
    .umask = 0x14,
+   .cmask = 20,
    .event_name = "cycle_activity.stalls_mem_any",
    },
   {
@@ -753,11 +790,13 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xA8},
    .umask = 0x01,
+   .cmask = 4,
    .event_name = "lsd.cycles_4_uops",
    },
   {
    .event_code = {0xA8},
    .umask = 0x01,
+   .cmask = 1,
    .event_name = "lsd.cycles_active",
    },
   {
@@ -803,26 +842,32 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xB1},
    .umask = 0x01,
+   .cmask = 4,
    .event_name = "uops_executed.cycles_ge_4_uops_exec",
    },
   {
    .event_code = {0xB1},
    .umask = 0x01,
+   .cmask = 3,
    .event_name = "uops_executed.cycles_ge_3_uops_exec",
    },
   {
    .event_code = {0xB1},
    .umask = 0x01,
+   .cmask = 2,
    .event_name = "uops_executed.cycles_ge_2_uops_exec",
    },
   {
    .event_code = {0xB1},
    .umask = 0x01,
+   .cmask = 1,
    .event_name = "uops_executed.cycles_ge_1_uop_exec",
    },
   {
    .event_code = {0xB1},
    .umask = 0x01,
+   .cmask = 1,
+   .inv = 1,
    .event_name = "uops_executed.stall_cycles",
    },
   {
@@ -838,26 +883,32 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xB1},
    .umask = 0x02,
+   .cmask = 1,
+   .inv = 1,
    .event_name = "uops_executed.core_cycles_none",
    },
   {
    .event_code = {0xB1},
    .umask = 0x02,
+   .cmask = 4,
    .event_name = "uops_executed.core_cycles_ge_4",
    },
   {
    .event_code = {0xB1},
    .umask = 0x02,
+   .cmask = 3,
    .event_name = "uops_executed.core_cycles_ge_3",
    },
   {
    .event_code = {0xB1},
    .umask = 0x02,
+   .cmask = 2,
    .event_name = "uops_executed.core_cycles_ge_2",
    },
   {
    .event_code = {0xB1},
    .umask = 0x02,
+   .cmask = 1,
    .event_name = "uops_executed.core_cycles_ge_1",
    },
   {
@@ -898,16 +949,21 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xC0},
    .umask = 0x01,
+   .cmask = 10,
    .event_name = "inst_retired.total_cycles_ps",
    },
   {
    .event_code = {0xC2},
    .umask = 0x02,
+   .cmask = 10,
+   .inv = 1,
    .event_name = "uops_retired.total_cycles",
    },
   {
    .event_code = {0xC2},
    .umask = 0x02,
+   .cmask = 1,
+   .inv = 1,
    .event_name = "uops_retired.stall_cycles",
    },
   {
@@ -918,6 +974,8 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xC3},
    .umask = 0x01,
+   .cmask = 1,
+   .edge = 1,
    .event_name = "machine_clears.count",
    },
   {
@@ -1118,6 +1176,7 @@ static perfmon_intel_pmc_event_t event_table[] = {
   {
    .event_code = {0xCA},
    .umask = 0x1E,
+   .cmask = 1,
    .event_name = "fp_assist.any",
    },
   {
diff --git a/src/plugins/perfmon/perfmon_periodic.c b/src/plugins/perfmon/perfmon_periodic.c
index 37d669b8d13..de31221f6f4 100644
--- a/src/plugins/perfmon/perfmon_periodic.c
+++ b/src/plugins/perfmon/perfmon_periodic.c
@@ -33,52 +33,65 @@ perf_event_open (struct perf_event_attr *hw_event, pid_t pid, int cpu,
 }
 
 static void
-read_current_perf_counters (vlib_main_t * vm, u64 * c0, u64 * c1,
-			    vlib_node_runtime_t * node,
-			    vlib_frame_t * frame, int before_or_after)
+read_current_perf_counters (vlib_node_runtime_perf_callback_data_t * data,
+			    vlib_node_runtime_perf_callback_args_t * args)
 {
   int i;
-  u64 *cc;
   perfmon_main_t *pm = &perfmon_main;
-  uword my_thread_index = vm->thread_index;
+  perfmon_thread_t *pt = data->u[0].v;
+  u64 c[2] = { 0, 0 };
+  u64 *cc;
 
-  *c0 = *c1 = 0;
+  if (PREDICT_FALSE (args->call_type == VLIB_NODE_RUNTIME_PERF_RESET))
+    return;
+
+  if (args->call_type == VLIB_NODE_RUNTIME_PERF_BEFORE)
+    cc = pt->c;
+  else
+    cc = c;
 
   for (i = 0; i < pm->n_active; i++)
     {
-      cc = (i == 0) ? c0 : c1;
-      if (pm->rdpmc_indices[i][my_thread_index] != ~0)
-	*cc = clib_rdpmc ((int) pm->rdpmc_indices[i][my_thread_index]);
+      if (pt->rdpmc_indices[i] != ~0)
+	cc[i] = clib_rdpmc ((int) pt->rdpmc_indices[i]);
       else
 	{
 	  u64 sw_value;
 	  int read_result;
-	  if ((read_result = read (pm->pm_fds[i][my_thread_index], &sw_value,
-				   sizeof (sw_value)) != sizeof (sw_value)))
+	  if ((read_result = read (pt->pm_fds[i], &sw_value,
+				   sizeof (sw_value))) != sizeof (sw_value))
 	    {
 	      clib_unix_warning
 		("counter read returned %d, expected %d",
 		 read_result, sizeof (sw_value));
-	      clib_callback_enable_disable
-		(vm->vlib_node_runtime_perf_counter_cbs,
-		 vm->vlib_node_runtime_perf_counter_cb_tmp,
-		 vm->worker_thread_main_loop_callback_lock,
+	      clib_callback_data_enable_disable
+		(&args->vm->vlib_node_runtime_perf_callbacks,
 		 read_current_perf_counters, 0 /* enable */ );
 	      return;
 	    }
-	  *cc = sw_value;
+	  cc[i] = sw_value;
 	}
     }
+
+  if (args->call_type == VLIB_NODE_RUNTIME_PERF_AFTER)
+    {
+      u32 node_index = args->node->node_index;
+      vec_validate (pt->counters, node_index);
+      pt->counters[node_index].ticks[0] += c[0] - pt->c[0];
+      pt->counters[node_index].ticks[1] += c[1] - pt->c[1];
+      pt->counters[node_index].vectors += args->packets;
+    }
 }
 
 static void
 clear_counters (perfmon_main_t * pm)
 {
-  int i, j;
+  int j;
   vlib_main_t *vm = pm->vlib_main;
   vlib_main_t *stat_vm;
-  vlib_node_main_t *nm;
-  vlib_node_t *n;
+  perfmon_thread_t *pt;
+  u32 len;
+
 
   vlib_worker_thread_barrier_sync (vm);
 
@@ -88,26 +101,12 @@ clear_counters (perfmon_main_t * pm)
       if (stat_vm == 0)
 	continue;
 
-      nm = &stat_vm->node_main;
-
-      /* Clear the node runtime perfmon counters */
-      for (i = 0; i < vec_len (nm->nodes); i++)
-	{
-	  n = nm->nodes[i];
-	  vlib_node_sync_stats (stat_vm, n);
-	}
+      pt = pm->threads[j];
+      len = vec_len (pt->counters);
+      if (!len)
+	continue;
 
-      /* And clear the node perfmon counters */
-      for (i = 0; i < vec_len (nm->nodes); i++)
-	{
-	  n = nm->nodes[i];
-	  n->stats_total.perf_counter0_ticks = 0;
-	  n->stats_total.perf_counter1_ticks = 0;
-	  n->stats_total.perf_counter_vectors = 0;
-	  n->stats_last_clear.perf_counter0_ticks = 0;
-	  n->stats_last_clear.perf_counter1_ticks = 0;
-	  n->stats_last_clear.perf_counter_vectors = 0;
-	}
+      clib_memset (pt->counters, 0, len * sizeof (pt->counters[0]));
     }
   vlib_worker_thread_barrier_release (vm);
 }
@@ -121,19 +120,20 @@ enable_current_events (perfmon_main_t * pm)
   perfmon_event_config_t *c;
   vlib_main_t *vm = vlib_get_main ();
   u32 my_thread_index = vm->thread_index;
+  perfmon_thread_t *pt = pm->threads[my_thread_index];
   u32 index;
   int i, limit = 1;
   int cpu;
+  vlib_node_runtime_perf_callback_data_t cbdata = { 0 };
+  cbdata.fp = read_current_perf_counters;
+  cbdata.u[0].v = pt;
+  cbdata.u[1].v = vm;
 
   if ((pm->current_event + 1) < vec_len (pm->single_events_to_collect))
     limit = 2;
 
   for (i = 0; i < limit; i++)
     {
-      vec_validate (pm->pm_fds[i], vec_len (vlib_mains) - 1);
-      vec_validate (pm->perf_event_pages[i], vec_len (vlib_mains) - 1);
-      vec_validate (pm->rdpmc_indices[i], vec_len (vlib_mains) - 1);
-
       c = vec_elt_at_index (pm->single_events_to_collect,
 			    pm->current_event + i);
 
@@ -184,8 +184,8 @@ enable_current_events (perfmon_main_t * pm)
       if (ioctl (fd, PERF_EVENT_IOC_ENABLE, 0) < 0)
 	clib_unix_warning ("enable ioctl");
 
-      pm->perf_event_pages[i][my_thread_index] = (void *) p;
-      pm->pm_fds[i][my_thread_index] = fd;
+      pt->perf_event_pages[i] = (void *) p;
+      pt->pm_fds[i] = fd;
     }
 
   /*
@@ -194,9 +194,7 @@ enable_current_events (perfmon_main_t * pm)
    */
   for (i = 0; i < limit; i++)
     {
-      p =
-	(struct perf_event_mmap_page *)
-	pm->perf_event_pages[i][my_thread_index];
+      p = (struct perf_event_mmap_page *) pt->perf_event_pages[i];
 
       /*
        * Software event counters - and others not capable of being
@@ -208,16 +206,12 @@ enable_current_events (perfmon_main_t * pm)
       else
 	index = p->index - 1;
 
-      pm->rdpmc_indices[i][my_thread_index] = index;
+      pt->rdpmc_indices[i] = index;
     }
 
   pm->n_active = i;
   /* Enable the main loop counter snapshot mechanism */
-  clib_callback_enable_disable
-    (vm->vlib_node_runtime_perf_counter_cbs,
-     vm->vlib_node_runtime_perf_counter_cb_tmp,
-     vm->worker_thread_main_loop_callback_lock,
-     read_current_perf_counters, 1 /* enable */ );
+  clib_callback_data_add (&vm->vlib_node_runtime_perf_callbacks, cbdata);
 }
 
 static void
@@ -225,35 +219,30 @@ disable_events (perfmon_main_t * pm)
 {
   vlib_main_t *vm = vlib_get_main ();
   u32 my_thread_index = vm->thread_index;
+  perfmon_thread_t *pt = pm->threads[my_thread_index];
   int i;
 
   /* Stop main loop collection */
-  clib_callback_enable_disable
-    (vm->vlib_node_runtime_perf_counter_cbs,
-     vm->vlib_node_runtime_perf_counter_cb_tmp,
-     vm->worker_thread_main_loop_callback_lock,
-     read_current_perf_counters, 0 /* enable */ );
+  clib_callback_data_remove (&vm->vlib_node_runtime_perf_callbacks,
+			     read_current_perf_counters);
 
   for (i = 0; i < pm->n_active; i++)
     {
-      if (pm->pm_fds[i][my_thread_index] == 0)
+      if (pt->pm_fds[i] == 0)
 	continue;
 
-      if (ioctl (pm->pm_fds[i][my_thread_index], PERF_EVENT_IOC_DISABLE, 0) <
-	  0)
+      if (ioctl (pt->pm_fds[i], PERF_EVENT_IOC_DISABLE, 0) < 0)
 	clib_unix_warning ("disable ioctl");
 
-      if (pm->perf_event_pages[i][my_thread_index])
+      if (pt->perf_event_pages[i])
 	{
-	  if (munmap (pm->perf_event_pages[i][my_thread_index],
-		      pm->page_size) < 0)
+	  if (munmap (pt->perf_event_pages[i], pm->page_size) < 0)
 	    clib_unix_warning ("munmap");
-	  pm->perf_event_pages[i][my_thread_index] = 0;
+	  pt->perf_event_pages[i] = 0;
 	}
 
-      (void) close (pm->pm_fds[i][my_thread_index]);
-      pm->pm_fds[i][my_thread_index] = 0;
-
+      (void) close (pt->pm_fds[i]);
+      pt->pm_fds[i] = 0;
     }
 }
 
@@ -265,7 +254,7 @@ worker_thread_start_event (vlib_main_t * vm)
   clib_callback_enable_disable (vm->worker_thread_main_loop_callbacks,
 				vm->worker_thread_main_loop_callback_tmp,
 				vm->worker_thread_main_loop_callback_lock,
-				worker_thread_start_event, 0 /* enable */ );
+				worker_thread_start_event, 0 /* disable */ );
   enable_current_events (pm);
 }
 
@@ -276,7 +265,7 @@ worker_thread_stop_event (vlib_main_t * vm)
   clib_callback_enable_disable (vm->worker_thread_main_loop_callbacks,
 				vm->worker_thread_main_loop_callback_tmp,
 				vm->worker_thread_main_loop_callback_lock,
-				worker_thread_stop_event, 0 /* enable */ );
+				worker_thread_stop_event, 0 /* disable */ );
   disable_events (pm);
 }
 
@@ -329,14 +318,15 @@ scrape_and_clear_counters (perfmon_main_t * pm)
   vlib_main_t *vm = pm->vlib_main;
   vlib_main_t *stat_vm;
   vlib_node_main_t *nm;
-  vlib_node_t ***node_dups = 0;
-  vlib_node_t **nodes;
-  vlib_node_t *n;
+  perfmon_counters_t *ctr;
+  perfmon_counters_t *ctrs;
+  perfmon_counters_t **ctr_dups = 0;
+  perfmon_thread_t *pt;
   perfmon_capture_t *c;
   perfmon_event_config_t *current_event;
   uword *p;
   u8 *counter_name;
-  u64 vectors_this_counter;
+  u32 len;
 
   /* snapshoot the nodes, including pm counters */
   vlib_worker_thread_barrier_sync (vm);
@@ -347,31 +337,16 @@ scrape_and_clear_counters (perfmon_main_t * pm)
       if (stat_vm == 0)
 	continue;
 
-      nm = &stat_vm->node_main;
-
-      for (i = 0; i < vec_len (nm->nodes); i++)
-	{
-	  n = nm->nodes[i];
-	  vlib_node_sync_stats (stat_vm, n);
-	}
-
-      nodes = 0;
-      vec_validate (nodes, vec_len (nm->nodes) - 1);
-      vec_add1 (node_dups, nodes);
-
-      /* Snapshoot and clear the per-node perfmon counters */
-      for (i = 0; i < vec_len (nm->nodes); i++)
+      pt = pm->threads[j];
+      len = vec_len (pt->counters);
+      ctrs = 0;
+      if (len)
 	{
-	  n = nm->nodes[i];
-	  nodes[i] = clib_mem_alloc (sizeof (*n));
-	  clib_memcpy_fast (nodes[i], n, sizeof (*n));
-	  n->stats_total.perf_counter0_ticks = 0;
-	  n->stats_total.perf_counter1_ticks = 0;
-	  n->stats_total.perf_counter_vectors = 0;
-	  n->stats_last_clear.perf_counter0_ticks = 0;
-	  n->stats_last_clear.perf_counter1_ticks = 0;
-	  n->stats_last_clear.perf_counter_vectors = 0;
+	  vec_validate (ctrs, len - 1);
+	  clib_memcpy (ctrs, pt->counters, len * sizeof (pt->counters[0]));
+	  clib_memset (pt->counters, 0, len * sizeof (pt->counters[0]));
 	}
+      vec_add1 (ctr_dups, ctrs);
     }
 
   vlib_worker_thread_barrier_release (vm);
@@ -382,22 +357,21 @@ scrape_and_clear_counters (perfmon_main_t * pm)
       if (stat_vm == 0)
 	continue;
 
-      nodes = node_dups[j];
+      pt = pm->threads[j];
+      ctrs = ctr_dups[j];
 
-      for (i = 0; i < vec_len (nodes); i++)
+      for (i = 0; i < vec_len (ctrs); i++)
 	{
 	  u8 *capture_name;
 
-	  n = nodes[i];
+	  ctr = &ctrs[i];
+	  nm = &stat_vm->node_main;
 
-	  if (n->stats_total.perf_counter0_ticks == 0 &&
-	      n->stats_total.perf_counter1_ticks == 0)
-	    goto skip_this_node;
+	  if (ctr->ticks[0] == 0 && ctr->ticks[1] == 0)
+	    continue;
 
 	  for (k = 0; k < 2; k++)
 	    {
-	      u64 counter_value, counter_last_clear;
-
 	      /*
 	       * We collect 2 counters at once, except for the
 	       * last counter when the user asks for an odd number of
@@ -407,20 +381,7 @@ scrape_and_clear_counters (perfmon_main_t * pm)
 		  >= vec_len (pm->single_events_to_collect))
 		break;
 
-	      if (k == 0)
-		{
-		  counter_value = n->stats_total.perf_counter0_ticks;
-		  counter_last_clear =
-		    n->stats_last_clear.perf_counter0_ticks;
-		}
-	      else
-		{
-		  counter_value = n->stats_total.perf_counter1_ticks;
-		  counter_last_clear =
-		    n->stats_last_clear.perf_counter1_ticks;
-		}
-
-	      capture_name = format (0, "t%d-%v%c", j, n->name, 0);
+	      capture_name = format (0, "t%d-%v%c", j, nm->nodes[i]->name, 0);
 
 	      p = hash_get_mem (pm->capture_by_thread_and_node_name,
 				capture_name);
@@ -443,20 +404,15 @@ scrape_and_clear_counters (perfmon_main_t * pm)
 	      current_event = pm->single_events_to_collect
 		+ pm->current_event + k;
 	      counter_name = (u8 *) current_event->name;
-	      vectors_this_counter = n->stats_total.perf_counter_vectors -
-		n->stats_last_clear.perf_counter_vectors;
 
 	      vec_add1 (c->counter_names, counter_name);
-	      vec_add1 (c->counter_values,
-			counter_value - counter_last_clear);
-	      vec_add1 (c->vectors_this_counter, vectors_this_counter);
+	      vec_add1 (c->counter_values, ctr->ticks[k]);
+	      vec_add1 (c->vectors_this_counter, ctr->vectors);
 	    }
-	skip_this_node:
-	  clib_mem_free (n);
 	}
-      vec_free (nodes);
+      vec_free (ctrs);
     }
-  vec_free (node_dups);
+  vec_free (ctr_dups);
 }
 
 static void
@@ -492,9 +448,8 @@ handle_timeout (vlib_main_t * vm, perfmon_main_t * pm, f64 now)
       for (i = 1; i < vec_len (vlib_mains); i++)
 	{
 	  /* Has the worker actually stopped collecting data? */
-	  while (clib_callback_is_set
-		 (vlib_mains[i]->worker_thread_main_loop_callbacks,
-		  vlib_mains[i]->worker_thread_main_loop_callback_lock,
+	  while (clib_callback_data_is_set
+		 (&vm->vlib_node_runtime_perf_callbacks,
 		  read_current_perf_counters))
 	    {
 	      if (vlib_time_now (vm) > deadman)
@@ -528,7 +483,7 @@ handle_timeout (vlib_main_t * vm, perfmon_main_t * pm, f64 now)
 	  (vlib_mains[i]->worker_thread_main_loop_callbacks,
 	   vlib_mains[i]->worker_thread_main_loop_callback_tmp,
 	   vlib_mains[i]->worker_thread_main_loop_callback_lock,
-	   worker_thread_start_event, 1 /* enable */ );
+	   worker_thread_start_event, 0 /* disable */ );
     }
 }
 
diff --git a/src/plugins/perfmon/perfmon_plugin.c b/src/plugins/perfmon/perfmon_plugin.c
new file mode 100644
index 00000000000..1d56573abd5
--- /dev/null
+++ b/src/plugins/perfmon/perfmon_plugin.c
@@ -0,0 +1,38 @@
+/*
+ * perfmon_plugin.c - perf monitor plugin
+ *
+ * Copyright (c) <current-year> <your-organization>
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <vnet/plugin/plugin.h>
+#include <vpp/app/version.h>
+
+/* *INDENT-OFF* */
+VLIB_PLUGIN_REGISTER () =
+{
+  .version = VPP_BUILD_VER,
+  .description = "Performance Monitor",
+#if !defined(__x86_64__)
+  .default_disabled = 1,
+#endif
+};
+/* *INDENT-ON* */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */