Re: [dpdk-dev] [PATCH 18.05 v4] eal: add function to return number of detected sockets

[Burakov, Anatoly]

On 21-Mar-18 4:59 AM, gowrishankar muthukrishnan wrote:
On Wednesday 07 February 2018 03:28 PM, Anatoly Burakov wrote:
During lcore scan, find maximum socket ID and store it. This will
break the ABI, so bump ABI version.
Signed-off-by: Anatoly Burakov <anatoly.bura...@intel.com>
---

Notes:
     v4:
     - Remove backwards ABI compatibility, bump ABI instead
     v3:
     - Added ABI compatibility
     v2:
     - checkpatch changes
     - check socket before deciding if the core is not to be used

  lib/librte_eal/bsdapp/eal/Makefile        |  2 +-
  lib/librte_eal/common/eal_common_lcore.c  | 37 
+++++++++++++++++++++----------
  lib/librte_eal/common/include/rte_eal.h   |  1 +
  lib/librte_eal/common/include/rte_lcore.h |  8 +++++++
  lib/librte_eal/linuxapp/eal/Makefile      |  2 +-
  lib/librte_eal/rte_eal_version.map        |  9 +++++++-
  6 files changed, 44 insertions(+), 15 deletions(-)

diff --git a/lib/librte_eal/bsdapp/eal/Makefile 
b/lib/librte_eal/bsdapp/eal/Makefile
index dd455e6..ed1d17b 100644
--- a/lib/librte_eal/bsdapp/eal/Makefile
+++ b/lib/librte_eal/bsdapp/eal/Makefile
@@ -21,7 +21,7 @@ LDLIBS += -lgcc_s

  EXPORT_MAP := ../../rte_eal_version.map

-LIBABIVER := 6
+LIBABIVER := 7

  # specific to bsdapp exec-env
  SRCS-$(CONFIG_RTE_EXEC_ENV_BSDAPP) := eal.c
diff --git a/lib/librte_eal/common/eal_common_lcore.c 
b/lib/librte_eal/common/eal_common_lcore.c
index 7724fa4..827ddeb 100644
--- a/lib/librte_eal/common/eal_common_lcore.c
+++ b/lib/librte_eal/common/eal_common_lcore.c
@@ -28,6 +28,7 @@ rte_eal_cpu_init(void)
      struct rte_config *config = rte_eal_get_configuration();
      unsigned lcore_id;
      unsigned count = 0;
+    unsigned int socket_id, max_socket_id = 0;

      /*
       * Parse the maximum set of logical cores, detect the subset of 
running
@@ -39,6 +40,19 @@ rte_eal_cpu_init(void)
          /* init cpuset for per lcore config */
          CPU_ZERO(&lcore_config[lcore_id].cpuset);

+        /* find socket first */
+        socket_id = eal_cpu_socket_id(lcore_id);
+        if (socket_id >= RTE_MAX_NUMA_NODES) {
+#ifdef RTE_EAL_ALLOW_INV_SOCKET_ID
+            socket_id = 0;
+#else
+            RTE_LOG(ERR, EAL, "Socket ID (%u) is greater than 
RTE_MAX_NUMA_NODES (%d)\n",
+                    socket_id, RTE_MAX_NUMA_NODES);
+            return -1;
+#endif
+        }
+        max_socket_id = RTE_MAX(max_socket_id, socket_id);
+
          /* in 1:1 mapping, record related cpu detected state */
          lcore_config[lcore_id].detected = eal_cpu_detected(lcore_id);
          if (lcore_config[lcore_id].detected == 0) {
@@ -54,18 +68,7 @@ rte_eal_cpu_init(void)
          config->lcore_role[lcore_id] = ROLE_RTE;
          lcore_config[lcore_id].core_role = ROLE_RTE;
          lcore_config[lcore_id].core_id = eal_cpu_core_id(lcore_id);
-        lcore_config[lcore_id].socket_id = eal_cpu_socket_id(lcore_id);
-        if (lcore_config[lcore_id].socket_id >= RTE_MAX_NUMA_NODES) {
-#ifdef RTE_EAL_ALLOW_INV_SOCKET_ID
-            lcore_config[lcore_id].socket_id = 0;
-#else
-            RTE_LOG(ERR, EAL, "Socket ID (%u) is greater than "
-                "RTE_MAX_NUMA_NODES (%d)\n",
-                lcore_config[lcore_id].socket_id,
-                RTE_MAX_NUMA_NODES);
-            return -1;
-#endif
-        }
+        lcore_config[lcore_id].socket_id = socket_id;
          RTE_LOG(DEBUG, EAL, "Detected lcore %u as "
                  "core %u on socket %u\n",
                  lcore_id, lcore_config[lcore_id].core_id,
@@ -79,5 +82,15 @@ rte_eal_cpu_init(void)
          RTE_MAX_LCORE);
      RTE_LOG(INFO, EAL, "Detected %u lcore(s)\n", config->lcore_count);

+    config->numa_node_count = max_socket_id + 1;

In some IBM servers, socket ID number does not seem to be in sequence. 
For an instance, 0 and 8 for a 2 node server.

In this case, numa_node_count would mislead users if wrongly understood 
by its variable name IMO (see below)
+    RTE_LOG(INFO, EAL, "Detected %u NUMA nodes\n", 
config->numa_node_count);

For an instance, reading above message would tell 'EAL detected 8 nodes' 
in my server, but actually there are only two nodes.

Could its name better be 'numa_node_id_max' ?. Also, we store in actual 
count of numa nodes in _count variable.

Also, there could be a case when there is no local memory available to a 
numa node too.

Thanks,
Gowrishankar

The point of this patchset is to (pre)allocate memory only on existing 
sockets.

If we don't know how many sockets there are, we are forced to 
preallocate VA space per each *possible* NUMA node - that is, reserve 
e.g. 8x128G of memory, 6 of which will go unused on a 2-socket system. 
We can't know if there is no memory on socket in advance, but we can at 
least avoid preallocating VA space for sockets that don't exist in the 
first place.

How about we store all possible socket id's instead? e.g. something like:

static int numa_node_ids[MAX_NUMA_NODES];
<...>
int rte_eal_cpu_init() {
        int sockets[RTE_MAX_LCORE];
        <...>
        for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
                core_to_socket[lcore_id] = socket;
        }
        <...>
        qsort(sockets);
        <...>
        // store all unique sockets in numa_node_ids in ascending order
}
<...>

on a 2 socket system we then get:

rte_num_sockets() => return 2
rte_get_socket_id(int idx) => return numa_node_ids[idx]

Would that be suitable?

--
Thanks,
Anatoly