From mboxrd@z Thu Jan 1 00:00:00 1970
Return-Path: <dev-bounces@dpdk.org>
Received: from dpdk.org (dpdk.org [92.243.14.124])
by inbox.dpdk.org (Postfix) with ESMTP id DC6F2A0471
for <public@inbox.dpdk.org>; Fri, 16 Aug 2019 21:19:27 +0200 (CEST)
Received: from [92.243.14.124] (localhost [127.0.0.1])
by dpdk.org (Postfix) with ESMTP id ACB3F1BF21;
Fri, 16 Aug 2019 21:19:27 +0200 (CEST)
Received: from mga05.intel.com (mga05.intel.com [192.55.52.43])
by dpdk.org (Postfix) with ESMTP id 091861BF19
for <dev@dpdk.org>; Fri, 16 Aug 2019 21:19:25 +0200 (CEST)
X-Amp-Result: SKIPPED(no attachment in message)
X-Amp-File-Uploaded: False
Received: from fmsmga006.fm.intel.com ([10.253.24.20])
by fmsmga105.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384;
16 Aug 2019 12:19:25 -0700
X-ExtLoop1: 1
X-IronPort-AV: E=Sophos;i="5.64,394,1559545200"; d="scan'208";a="377512597"
Received: from jrharri1-skx.ch.intel.com (HELO [127.0.1.1]) ([143.182.137.73])
by fmsmga006.fm.intel.com with ESMTP; 16 Aug 2019 12:19:25 -0700
From: Jim Harris <james.r.harris@intel.com>
To: dev@dpdk.org, anatoly.burakov@intel.com
Date: Fri, 16 Aug 2019 05:14:18 -0700
Message-ID: <156595765801.19075.15759465891004687257.stgit@jrharri1-skx>
User-Agent: StGit/0.17.1-dirty
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 7bit
Subject: [dpdk-dev] [PATCH v3 1/2] timer: use rte_mp_msg to pass TSC hz to
secondary procs
X-BeenThere: dev@dpdk.org
X-Mailman-Version: 2.1.15
Precedence: list
List-Id: DPDK patches and discussions <dev.dpdk.org>
List-Unsubscribe: <https://mails.dpdk.org/options/dev>,
<mailto:dev-request@dpdk.org?subject=unsubscribe>
List-Archive: <http://mails.dpdk.org/archives/dev/>
List-Post: <mailto:dev@dpdk.org>
List-Help: <mailto:dev-request@dpdk.org?subject=help>
List-Subscribe: <https://mails.dpdk.org/listinfo/dev>,
<mailto:dev-request@dpdk.org?subject=subscribe>
Errors-To: dev-bounces@dpdk.org
Sender: "dev" <dev-bounces@dpdk.org>
rte_eal_init() is much faster in secondary processes since
hugepages don't need to be zeroed. But there's still
non-trivial delays in the timer subsystem initialization
due to the 100ms sleep used to calculate TSC hz. So use
the rte_mp_msg framework to allow secondary processes
to get the TSC hz from the primary process.
This cuts rte_eal_init() execution time in a secondary
process from 165ms to 66ms in my test program.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
---
lib/librte_eal/common/eal_common_timer.c | 70 +++++++++++++++++++++++++++++-
1 file changed, 68 insertions(+), 2 deletions(-)
diff --git a/lib/librte_eal/common/eal_common_timer.c b/lib/librte_eal/common/eal_common_timer.c
index 145543de7..3eb0309f5 100644
--- a/lib/librte_eal/common/eal_common_timer.c
+++ b/lib/librte_eal/common/eal_common_timer.c
@@ -15,9 +15,16 @@
#include <rte_log.h>
#include <rte_cycles.h>
#include <rte_pause.h>
+#include <rte_eal.h>
#include "eal_private.h"
+#define EAL_TIMER_MP "eal_timer_mp_sync"
+
+struct timer_mp_param {
+ uint64_t tsc;
+};
+
/* The frequency of the RDTSC timer resolution */
static uint64_t eal_tsc_resolution_hz;
@@ -74,8 +81,8 @@ estimate_tsc_freq(void)
return RTE_ALIGN_MUL_NEAR(rte_rdtsc() - start, CYC_PER_10MHZ);
}
-void
-set_tsc_freq(void)
+static void
+set_tsc_freq_primary(void)
{
uint64_t freq;
@@ -89,6 +96,65 @@ set_tsc_freq(void)
eal_tsc_resolution_hz = freq;
}
+static void
+set_tsc_freq_secondary(void)
+{
+ struct rte_mp_msg mp_req = {0};
+ struct rte_mp_reply mp_reply = {0};
+ struct timer_mp_param *r;
+ struct timespec ts = {.tv_sec = 1, .tv_nsec = 0};
+
+ strcpy(mp_req.name, EAL_TIMER_MP);
+ if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) || mp_reply.nb_received != 1) {
+ /* We weren't able to get the tsc hz from the primary process. So we will
+ * just calculate it here in the secondary process instead.
+ */
+ set_tsc_freq_primary();
+ free(mp_reply.msgs);
+ return;
+ }
+
+ r = (struct timer_mp_param *)mp_reply.msgs[0].param;
+ eal_tsc_resolution_hz = r->tsc;
+ free(mp_reply.msgs);
+}
+
+static int
+timer_mp_primary(__attribute__((unused)) const struct rte_mp_msg *msg, const void *peer)
+{
+ struct rte_mp_msg reply;
+ struct timer_mp_param *r = (struct timer_mp_param *)reply.param;
+
+ memset(&reply, 0, sizeof(reply));
+ r->tsc = eal_tsc_resolution_hz;
+ strcpy(reply.name, EAL_TIMER_MP);
+ reply.len_param = sizeof(*r);
+
+ return rte_mp_reply(&reply, peer);
+}
+
+void
+set_tsc_freq(void)
+{
+ int rc;
+
+ /* We use a 100ms timer to calculate the TSC hz. We can save this 100ms in
+ * secondary processes, by getting the TSC hz from the primary process.
+ * So register an mp_action callback in the primary process, which secondary
+ * processes will use to get the TSC hz.
+ */
+ if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
+ set_tsc_freq_primary();
+ rc = rte_mp_action_register(EAL_TIMER_MP, timer_mp_primary);
+ if (rc) {
+ RTE_LOG(WARNING, EAL, "Could not register mp_action - secondary "
+ " processes will calculate TSC independently.\n");
+ }
+ } else {
+ set_tsc_freq_secondary();
+ }
+}
+
void rte_delay_us_callback_register(void (*userfunc)(unsigned int))
{
rte_delay_us = userfunc;