diff mbox series

[ovs-dev,v9,12/16] dpdk: Cache result of CPU ISA checks.

Message ID 20210212171718.2189798-13-harry.van.haaren@intel.com
State New
Headers show
Series DPIF Framework + Optimizations | expand

Commit Message

Harry van Haaren Feb. 12, 2021, 5:17 p.m. UTC
As a small optimization, this patch caches the result of a CPU ISA
check from DPDK. Particularly in the case of running the DPCLS
autovalidator (which repeatedly probes subtables) this reduces
the amount of CPU ISA lookups from the DPDK level.

By caching them at the OVS/dpdk.c level, the ISA checks remain
runtime for the CPU where they are executed, but subsequent checks
for the same ISA feature become much cheaper.

Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Co-authored-by: Cian Ferriter <cian.ferriter@intel.com>
Signed-off-by: Cian Ferriter <cian.ferriter@intel.com>

---

v8: Add NEWS entry.
---
 NEWS       |  1 +
 lib/dpdk.c | 28 ++++++++++++++++++++++++----
 2 files changed, 25 insertions(+), 4 deletions(-)
diff mbox series

Patch

diff --git a/NEWS b/NEWS
index a5bb16da2..0a093e582 100644
--- a/NEWS
+++ b/NEWS
@@ -33,6 +33,7 @@  v2.15.0 - xx xxx xxxx
    - DPDK:
      * Removed support for vhost-user dequeue zero-copy.
      * Add support for DPDK 20.11.
+     * Cache results for CPU ISA checks, reduces overhead on repeated lookups.
    - Userspace datapath:
      * Add the 'pmd' option to "ovs-appctl dpctl/dump-flows", which
        restricts a flow dump to a single PMD thread if set.
diff --git a/lib/dpdk.c b/lib/dpdk.c
index 319540394..c883a4b8b 100644
--- a/lib/dpdk.c
+++ b/lib/dpdk.c
@@ -614,13 +614,33 @@  print_dpdk_version(void)
     puts(rte_version());
 }
 
+/* Avoid calling rte_cpu_get_flag_enabled() excessively, by caching the
+ * result of the call for each CPU flag in a static variable. To avoid
+ * allocating large numbers of static variables, use a uint8 as a bitfield.
+ * Note the macro must only return if the ISA check is done and available.
+ */
+#define ISA_CHECK_DONE_BIT (1 << 0)
+#define ISA_AVAILABLE_BIT  (1 << 1)
+
 #define CHECK_CPU_FEATURE(feature, name_str, RTE_CPUFLAG)               \
     do {                                                                \
         if (strncmp(feature, name_str, strlen(name_str)) == 0) {        \
-            int has_isa = rte_cpu_get_flag_enabled(RTE_CPUFLAG);        \
-            VLOG_DBG("CPU flag %s, available %s\n", name_str,           \
-                      has_isa ? "yes" : "no");                          \
-            return true;                                                \
+            static uint8_t isa_check_##RTE_CPUFLAG;                     \
+            int check = isa_check_##RTE_CPUFLAG & ISA_CHECK_DONE_BIT;   \
+            if (OVS_UNLIKELY(!check)) {                                 \
+                int has_isa = rte_cpu_get_flag_enabled(RTE_CPUFLAG);    \
+                VLOG_DBG("CPU flag %s, available %s\n",                 \
+                         name_str, has_isa ? "yes" : "no");             \
+                isa_check_##RTE_CPUFLAG = ISA_CHECK_DONE_BIT;           \
+                if (has_isa) {                                          \
+                    isa_check_##RTE_CPUFLAG |= ISA_AVAILABLE_BIT;       \
+                }                                                       \
+            }                                                           \
+            if (isa_check_##RTE_CPUFLAG & ISA_AVAILABLE_BIT) {          \
+                return true;                                            \
+            } else {                                                    \
+                return false;                                           \
+            }                                                           \
         }                                                               \
     } while (0)