[v6,06/10] eal: register non-EAL threads as lcores
diff mbox series

Message ID 20200706205234.8040-7-david.marchand@redhat.com
State Accepted
Delegated to: Thomas Monjalon
Headers show
Series
  • Register non-EAL threads as lcore
Related show

Checks

Context Check Description
ci/Intel-compilation fail apply issues
ci/checkpatch success coding style OK

Commit Message

David Marchand July 6, 2020, 8:52 p.m. UTC
DPDK allows calling some part of its API from a non-EAL thread but this
has some limitations.
OVS (and other applications) has its own thread management but still
want to avoid such limitations by hacking RTE_PER_LCORE(_lcore_id) and
faking EAL threads potentially unknown of some DPDK component.

Introduce a new API to register non-EAL thread and associate them to a
free lcore with a new NON_EAL role.
This role denotes lcores that do not run DPDK mainloop and as such
prevents use of rte_eal_wait_lcore() and consorts.

Multiprocess is not supported as the need for cohabitation with this new
feature is unclear at the moment.

Signed-off-by: David Marchand <david.marchand@redhat.com>
Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
---
Changes since v5:
- added release notes,
- fixed windows build,

Changes since v4:
- returned an error when rte_thread_register fails + set rte_errno,
- prefixed unit tests logs with Error: when applicable,

Changes since v2:
- refused multiprocess init once rte_thread_register got called, and
  vice versa,
- added warning on multiprocess in rte_thread_register doxygen,

Changes since v1:
- moved cleanup on lcore role code in patch 5,
- added unit test,
- updated documentation,
- changed naming from "external thread" to "registered non-EAL thread"

---
 MAINTAINERS                                   |   1 +
 app/test/Makefile                             |   1 +
 app/test/autotest_data.py                     |   6 +
 app/test/meson.build                          |   2 +
 app/test/test_lcores.c                        | 143 ++++++++++++++++++
 doc/guides/howto/debug_troubleshoot.rst       |   5 +-
 .../prog_guide/env_abstraction_layer.rst      |  22 +--
 doc/guides/prog_guide/mempool_lib.rst         |   2 +-
 doc/guides/rel_notes/release_20_08.rst        |   6 +
 lib/librte_eal/common/eal_common_lcore.c      |  45 +++++-
 lib/librte_eal/common/eal_common_proc.c       |  42 +++++
 lib/librte_eal/common/eal_common_thread.c     |  49 +++++-
 lib/librte_eal/common/eal_memcfg.h            |   2 +
 lib/librte_eal/common/eal_private.h           |  28 ++++
 lib/librte_eal/freebsd/eal.c                  |   4 +
 lib/librte_eal/include/rte_lcore.h            |  28 +++-
 lib/librte_eal/linux/eal.c                    |   4 +
 lib/librte_eal/rte_eal_version.map            |   4 +
 lib/librte_eal/windows/eal.c                  |   6 +
 lib/librte_mempool/rte_mempool.h              |  11 +-
 20 files changed, 388 insertions(+), 23 deletions(-)
 create mode 100644 app/test/test_lcores.c

Patch
diff mbox series

diff --git a/MAINTAINERS b/MAINTAINERS
index 27a7be8e64..5a0e1b48b2 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -182,6 +182,7 @@  F: app/test/test_cycles.c
 F: app/test/test_debug.c
 F: app/test/test_eal*
 F: app/test/test_errno.c
+F: app/test/test_lcores.c
 F: app/test/test_logs.c
 F: app/test/test_memcpy*
 F: app/test/test_per_lcore.c
diff --git a/app/test/Makefile b/app/test/Makefile
index e5440774b1..f4065271e4 100644
--- a/app/test/Makefile
+++ b/app/test/Makefile
@@ -98,6 +98,7 @@  SRCS-$(CONFIG_RTE_LIBRTE_FLOW_CLASSIFY) += test_flow_classify.c
 endif
 
 SRCS-y += test_rwlock.c
+SRCS-y += test_lcores.c
 
 SRCS-$(CONFIG_RTE_LIBRTE_STACK) += test_stack.c
 SRCS-$(CONFIG_RTE_LIBRTE_STACK) += test_stack_perf.c
diff --git a/app/test/autotest_data.py b/app/test/autotest_data.py
index 238ab631b4..4b7da45e09 100644
--- a/app/test/autotest_data.py
+++ b/app/test/autotest_data.py
@@ -62,6 +62,12 @@ 
         "Func":    rwlock_autotest,
         "Report":  None,
     },
+    {
+        "Name":    "Lcores autotest",
+        "Command": "lcores_autotest",
+        "Func":    default_autotest,
+        "Report":  None,
+    },
     {
         "Name":    "Logs autotest",
         "Command": "logs_autotest",
diff --git a/app/test/meson.build b/app/test/meson.build
index 56591db4e0..86b4d4d490 100644
--- a/app/test/meson.build
+++ b/app/test/meson.build
@@ -67,6 +67,7 @@  test_sources = files('commands.c',
 	'test_ipsec_perf.c',
 	'test_kni.c',
 	'test_kvargs.c',
+	'test_lcores.c',
 	'test_logs.c',
 	'test_lpm.c',
 	'test_lpm6.c',
@@ -207,6 +208,7 @@  fast_tests = [
         ['hash_autotest', true],
         ['interrupt_autotest', true],
         ['ipfrag_autotest', false],
+        ['lcores_autotest', true],
         ['logs_autotest', true],
         ['lpm_autotest', true],
         ['lpm6_autotest', true],
diff --git a/app/test/test_lcores.c b/app/test/test_lcores.c
new file mode 100644
index 0000000000..afb9cdd444
--- /dev/null
+++ b/app/test/test_lcores.c
@@ -0,0 +1,143 @@ 
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright (c) 2020 Red Hat, Inc.
+ */
+
+#include <pthread.h>
+#include <string.h>
+
+#include <rte_errno.h>
+#include <rte_lcore.h>
+
+#include "test.h"
+
+struct thread_context {
+	enum { INIT, ERROR, DONE } state;
+	bool lcore_id_any;
+	pthread_t id;
+	unsigned int *registered_count;
+};
+
+static void *thread_loop(void *arg)
+{
+	struct thread_context *t = arg;
+	unsigned int lcore_id;
+
+	lcore_id = rte_lcore_id();
+	if (lcore_id != LCORE_ID_ANY) {
+		printf("Error: incorrect lcore id for new thread %u\n", lcore_id);
+		t->state = ERROR;
+	}
+	if (rte_thread_register() < 0)
+		printf("Warning: could not register new thread (this might be expected during this test), reason %s\n",
+			rte_strerror(rte_errno));
+	lcore_id = rte_lcore_id();
+	if ((t->lcore_id_any && lcore_id != LCORE_ID_ANY) ||
+			(!t->lcore_id_any && lcore_id == LCORE_ID_ANY)) {
+		printf("Error: could not register new thread, got %u while %sexpecting %u\n",
+			lcore_id, t->lcore_id_any ? "" : "not ", LCORE_ID_ANY);
+		t->state = ERROR;
+	}
+	/* Report register happened to the control thread. */
+	__atomic_add_fetch(t->registered_count, 1, __ATOMIC_RELEASE);
+
+	/* Wait for release from the control thread. */
+	while (__atomic_load_n(t->registered_count, __ATOMIC_ACQUIRE) != 0)
+		;
+	rte_thread_unregister();
+	lcore_id = rte_lcore_id();
+	if (lcore_id != LCORE_ID_ANY) {
+		printf("Error: could not unregister new thread, %u still assigned\n",
+			lcore_id);
+		t->state = ERROR;
+	}
+
+	if (t->state != ERROR)
+		t->state = DONE;
+
+	return NULL;
+}
+
+static int
+test_non_eal_lcores(unsigned int eal_threads_count)
+{
+	struct thread_context thread_contexts[RTE_MAX_LCORE];
+	unsigned int non_eal_threads_count;
+	unsigned int registered_count;
+	struct thread_context *t;
+	unsigned int i;
+	int ret;
+
+	non_eal_threads_count = 0;
+	registered_count = 0;
+
+	/* Try to create as many threads as possible. */
+	for (i = 0; i < RTE_MAX_LCORE - eal_threads_count; i++) {
+		t = &thread_contexts[i];
+		t->state = INIT;
+		t->registered_count = &registered_count;
+		t->lcore_id_any = false;
+		if (pthread_create(&t->id, NULL, thread_loop, t) != 0)
+			break;
+		non_eal_threads_count++;
+	}
+	printf("non-EAL threads count: %u\n", non_eal_threads_count);
+	/* Wait all non-EAL threads to register. */
+	while (__atomic_load_n(&registered_count, __ATOMIC_ACQUIRE) !=
+			non_eal_threads_count)
+		;
+
+	/* We managed to create the max number of threads, let's try to create
+	 * one more. This will allow one more check.
+	 */
+	if (eal_threads_count + non_eal_threads_count < RTE_MAX_LCORE)
+		goto skip_lcore_any;
+	t = &thread_contexts[non_eal_threads_count];
+	t->state = INIT;
+	t->registered_count = &registered_count;
+	t->lcore_id_any = true;
+	if (pthread_create(&t->id, NULL, thread_loop, t) == 0) {
+		non_eal_threads_count++;
+		printf("non-EAL threads count: %u\n", non_eal_threads_count);
+		while (__atomic_load_n(&registered_count, __ATOMIC_ACQUIRE) !=
+				non_eal_threads_count)
+			;
+	}
+
+skip_lcore_any:
+	/* Release all threads, and check their states. */
+	__atomic_store_n(&registered_count, 0, __ATOMIC_RELEASE);
+	ret = 0;
+	for (i = 0; i < non_eal_threads_count; i++) {
+		t = &thread_contexts[i];
+		pthread_join(t->id, NULL);
+		if (t->state != DONE)
+			ret = -1;
+	}
+
+	return ret;
+}
+
+static int
+test_lcores(void)
+{
+	unsigned int eal_threads_count = 0;
+	unsigned int i;
+
+	for (i = 0; i < RTE_MAX_LCORE; i++) {
+		if (!rte_lcore_has_role(i, ROLE_OFF))
+			eal_threads_count++;
+	}
+	if (eal_threads_count == 0) {
+		printf("Error: something is broken, no EAL thread detected.\n");
+		return TEST_FAILED;
+	}
+	printf("EAL threads count: %u, RTE_MAX_LCORE=%u\n", eal_threads_count,
+		RTE_MAX_LCORE);
+
+	if (test_non_eal_lcores(eal_threads_count) < 0)
+		return TEST_FAILED;
+
+	return TEST_SUCCESS;
+}
+
+REGISTER_TEST_COMMAND(lcores_autotest, test_lcores);
diff --git a/doc/guides/howto/debug_troubleshoot.rst b/doc/guides/howto/debug_troubleshoot.rst
index cef016b2fe..5a46f5fba3 100644
--- a/doc/guides/howto/debug_troubleshoot.rst
+++ b/doc/guides/howto/debug_troubleshoot.rst
@@ -307,8 +307,9 @@  Custom worker function :numref:`dtg_distributor_worker`.
 
 #. Configuration issue isolation
 
-   * Identify core role using ``rte_eal_lcore_role`` to identify RTE, OFF and
-     SERVICE. Check performance functions are mapped to run on the cores.
+   * Identify core role using ``rte_eal_lcore_role`` to identify RTE, OFF,
+     SERVICE and NON_EAL. Check performance functions are mapped to run on the
+     cores.
 
    * For high-performance execution logic ensure running it on correct NUMA
      and non-master core.
diff --git a/doc/guides/prog_guide/env_abstraction_layer.rst b/doc/guides/prog_guide/env_abstraction_layer.rst
index 48a2fec066..f64ae953d1 100644
--- a/doc/guides/prog_guide/env_abstraction_layer.rst
+++ b/doc/guides/prog_guide/env_abstraction_layer.rst
@@ -564,9 +564,13 @@  It's also compatible with the pattern of corelist('-l') option.
 non-EAL pthread support
 ~~~~~~~~~~~~~~~~~~~~~~~
 
-It is possible to use the DPDK execution context with any user pthread (aka. Non-EAL pthreads).
-In a non-EAL pthread, the *_lcore_id* is always LCORE_ID_ANY which identifies that it is not an EAL thread with a valid, unique, *_lcore_id*.
-Some libraries will use an alternative unique ID (e.g. TID), some will not be impacted at all, and some will work but with limitations (e.g. timer and mempool libraries).
+It is possible to use the DPDK execution context with any user pthread (aka. non-EAL pthreads).
+There are two kinds of non-EAL pthreads:
+
+- a registered non-EAL pthread with a valid *_lcore_id* that was successfully assigned by calling ``rte_thread_register()``,
+- a non registered non-EAL pthread with a LCORE_ID_ANY,
+
+For non registered non-EAL pthread (with a LCORE_ID_ANY *_lcore_id*), some libraries will use an alternative unique ID (e.g. TID), some will not be impacted at all, and some will work but with limitations (e.g. timer and mempool libraries).
 
 All these impacts are mentioned in :ref:`known_issue_label` section.
 
@@ -613,9 +617,9 @@  Known Issues
 + rte_mempool
 
   The rte_mempool uses a per-lcore cache inside the mempool.
-  For non-EAL pthreads, ``rte_lcore_id()`` will not return a valid number.
-  So for now, when rte_mempool is used with non-EAL pthreads, the put/get operations will bypass the default mempool cache and there is a performance penalty because of this bypass.
-  Only user-owned external caches can be used in a non-EAL context in conjunction with ``rte_mempool_generic_put()`` and ``rte_mempool_generic_get()`` that accept an explicit cache parameter.
+  For unregistered non-EAL pthreads, ``rte_lcore_id()`` will not return a valid number.
+  So for now, when rte_mempool is used with unregistered non-EAL pthreads, the put/get operations will bypass the default mempool cache and there is a performance penalty because of this bypass.
+  Only user-owned external caches can be used in an unregistered non-EAL context in conjunction with ``rte_mempool_generic_put()`` and ``rte_mempool_generic_get()`` that accept an explicit cache parameter.
 
 + rte_ring
 
@@ -660,15 +664,15 @@  Known Issues
 
 + rte_timer
 
-  Running  ``rte_timer_manage()`` on a non-EAL pthread is not allowed. However, resetting/stopping the timer from a non-EAL pthread is allowed.
+  Running  ``rte_timer_manage()`` on an unregistered non-EAL pthread is not allowed. However, resetting/stopping the timer from a non-EAL pthread is allowed.
 
 + rte_log
 
-  In non-EAL pthreads, there is no per thread loglevel and logtype, global loglevels are used.
+  In unregistered non-EAL pthreads, there is no per thread loglevel and logtype, global loglevels are used.
 
 + misc
 
-  The debug statistics of rte_ring, rte_mempool and rte_timer are not supported in a non-EAL pthread.
+  The debug statistics of rte_ring, rte_mempool and rte_timer are not supported in an unregistered non-EAL pthread.
 
 cgroup control
 ~~~~~~~~~~~~~~
diff --git a/doc/guides/prog_guide/mempool_lib.rst b/doc/guides/prog_guide/mempool_lib.rst
index f8b430d656..e3e1f940be 100644
--- a/doc/guides/prog_guide/mempool_lib.rst
+++ b/doc/guides/prog_guide/mempool_lib.rst
@@ -103,7 +103,7 @@  The maximum size of the cache is static and is defined at compilation time (CONF
 Alternatively to the internal default per-lcore local cache, an application can create and manage external caches through the ``rte_mempool_cache_create()``, ``rte_mempool_cache_free()`` and ``rte_mempool_cache_flush()`` calls.
 These user-owned caches can be explicitly passed to ``rte_mempool_generic_put()`` and ``rte_mempool_generic_get()``.
 The ``rte_mempool_default_cache()`` call returns the default internal cache if any.
-In contrast to the default caches, user-owned caches can be used by non-EAL threads too.
+In contrast to the default caches, user-owned caches can be used by unregistered non-EAL threads too.
 
 Mempool Handlers
 ------------------------
diff --git a/doc/guides/rel_notes/release_20_08.rst b/doc/guides/rel_notes/release_20_08.rst
index c0f81d8d40..1cb44be626 100644
--- a/doc/guides/rel_notes/release_20_08.rst
+++ b/doc/guides/rel_notes/release_20_08.rst
@@ -56,6 +56,12 @@  New Features
      Also, make sure to start the actual text at the margin.
      =========================================================
 
+* **Added non-EAL threads registration API.**
+
+  Added a new API to register non-EAL threads as lcores. This can be used by
+  applications to have its threads known of DPDK without suffering from the
+  non-EAL previous limitations in terms of performance.
+
 * **Updated PCAP driver.**
 
   Updated PCAP driver with new features and improvements, including:
diff --git a/lib/librte_eal/common/eal_common_lcore.c b/lib/librte_eal/common/eal_common_lcore.c
index 86d32a3dd7..2b7d262372 100644
--- a/lib/librte_eal/common/eal_common_lcore.c
+++ b/lib/librte_eal/common/eal_common_lcore.c
@@ -6,13 +6,15 @@ 
 #include <limits.h>
 #include <string.h>
 
-#include <rte_errno.h>
-#include <rte_log.h>
-#include <rte_eal.h>
-#include <rte_lcore.h>
 #include <rte_common.h>
 #include <rte_debug.h>
+#include <rte_eal.h>
+#include <rte_errno.h>
+#include <rte_lcore.h>
+#include <rte_log.h>
+#include <rte_spinlock.h>
 
+#include "eal_memcfg.h"
 #include "eal_private.h"
 #include "eal_thread.h"
 
@@ -220,3 +222,38 @@  rte_socket_id_by_idx(unsigned int idx)
 	}
 	return config->numa_nodes[idx];
 }
+
+static rte_spinlock_t lcore_lock = RTE_SPINLOCK_INITIALIZER;
+
+unsigned int
+eal_lcore_non_eal_allocate(void)
+{
+	struct rte_config *cfg = rte_eal_get_configuration();
+	unsigned int lcore_id;
+
+	rte_spinlock_lock(&lcore_lock);
+	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
+		if (cfg->lcore_role[lcore_id] != ROLE_OFF)
+			continue;
+		cfg->lcore_role[lcore_id] = ROLE_NON_EAL;
+		cfg->lcore_count++;
+		break;
+	}
+	if (lcore_id == RTE_MAX_LCORE)
+		RTE_LOG(DEBUG, EAL, "No lcore available.\n");
+	rte_spinlock_unlock(&lcore_lock);
+	return lcore_id;
+}
+
+void
+eal_lcore_non_eal_release(unsigned int lcore_id)
+{
+	struct rte_config *cfg = rte_eal_get_configuration();
+
+	rte_spinlock_lock(&lcore_lock);
+	if (cfg->lcore_role[lcore_id] == ROLE_NON_EAL) {
+		cfg->lcore_role[lcore_id] = ROLE_OFF;
+		cfg->lcore_count--;
+	}
+	rte_spinlock_unlock(&lcore_lock);
+}
diff --git a/lib/librte_eal/common/eal_common_proc.c b/lib/librte_eal/common/eal_common_proc.c
index c649789a52..10142683f8 100644
--- a/lib/librte_eal/common/eal_common_proc.c
+++ b/lib/librte_eal/common/eal_common_proc.c
@@ -29,6 +29,7 @@ 
 #include <rte_log.h>
 #include <rte_tailq.h>
 
+#include "eal_memcfg.h"
 #include "eal_private.h"
 #include "eal_filesystem.h"
 #include "eal_internal_cfg.h"
@@ -1232,3 +1233,44 @@  rte_mp_reply(struct rte_mp_msg *msg, const char *peer)
 
 	return mp_send(msg, peer, MP_REP);
 }
+
+/* Internally, the status of the mp feature is represented as a three-state:
+ * - "unknown" as long as no secondary process attached to a primary process
+ *   and there was no call to eal_disable_multiprocess yet,
+ * - "enabled" as soon as a secondary process attaches to a primary process,
+ * - "disabled" when a primary process successfully called
+ *   eal_disable_multiprocess,
+ */
+enum mp_status {
+	MP_STATUS_UNKNOWN,
+	MP_STATUS_DISABLED,
+	MP_STATUS_ENABLED,
+};
+
+static bool
+set_mp_status(enum mp_status status)
+{
+	struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
+	uint8_t expected;
+	uint8_t desired;
+
+	expected = MP_STATUS_UNKNOWN;
+	desired = status;
+	if (__atomic_compare_exchange_n(&mcfg->mp_status, &expected, desired,
+			false, __ATOMIC_RELAXED, __ATOMIC_RELAXED))
+		return true;
+
+	return __atomic_load_n(&mcfg->mp_status, __ATOMIC_RELAXED) == desired;
+}
+
+bool
+eal_disable_multiprocess(void)
+{
+	return set_mp_status(MP_STATUS_DISABLED);
+}
+
+bool
+eal_enable_multiprocess(void)
+{
+	return set_mp_status(MP_STATUS_ENABLED);
+}
diff --git a/lib/librte_eal/common/eal_common_thread.c b/lib/librte_eal/common/eal_common_thread.c
index b1b69ea58c..cead8ea5cf 100644
--- a/lib/librte_eal/common/eal_common_thread.c
+++ b/lib/librte_eal/common/eal_common_thread.c
@@ -12,9 +12,10 @@ 
 #include <assert.h>
 #include <string.h>
 
+#include <rte_errno.h>
 #include <rte_lcore.h>
-#include <rte_memory.h>
 #include <rte_log.h>
+#include <rte_memory.h>
 #include <rte_trace_point.h>
 
 #include "eal_internal_cfg.h"
@@ -241,3 +242,49 @@  rte_ctrl_thread_create(pthread_t *thread, const char *name,
 	pthread_join(*thread, NULL);
 	return -ret;
 }
+
+int
+rte_thread_register(void)
+{
+	unsigned int lcore_id;
+	rte_cpuset_t cpuset;
+
+	/* EAL init flushes all lcores, we can't register before. */
+	if (eal_get_internal_configuration()->init_complete != 1) {
+		RTE_LOG(DEBUG, EAL, "Called %s before EAL init.\n", __func__);
+		rte_errno = EINVAL;
+		return -1;
+	}
+	if (!eal_disable_multiprocess()) {
+		RTE_LOG(ERR, EAL, "Multiprocess in use, registering non-EAL threads is not supported.\n");
+		rte_errno = EINVAL;
+		return -1;
+	}
+	if (pthread_getaffinity_np(pthread_self(), sizeof(cpuset),
+			&cpuset) != 0)
+		CPU_ZERO(&cpuset);
+	lcore_id = eal_lcore_non_eal_allocate();
+	if (lcore_id >= RTE_MAX_LCORE)
+		lcore_id = LCORE_ID_ANY;
+	__rte_thread_init(lcore_id, &cpuset);
+	if (lcore_id == LCORE_ID_ANY) {
+		rte_errno = ENOMEM;
+		return -1;
+	}
+	RTE_LOG(DEBUG, EAL, "Registered non-EAL thread as lcore %u.\n",
+		lcore_id);
+	return 0;
+}
+
+void
+rte_thread_unregister(void)
+{
+	unsigned int lcore_id = rte_lcore_id();
+
+	if (lcore_id != LCORE_ID_ANY)
+		eal_lcore_non_eal_release(lcore_id);
+	__rte_thread_uninit();
+	if (lcore_id != LCORE_ID_ANY)
+		RTE_LOG(DEBUG, EAL, "Unregistered non-EAL thread (was lcore %u).\n",
+			lcore_id);
+}
diff --git a/lib/librte_eal/common/eal_memcfg.h b/lib/librte_eal/common/eal_memcfg.h
index 583fcb5953..ea013a9daf 100644
--- a/lib/librte_eal/common/eal_memcfg.h
+++ b/lib/librte_eal/common/eal_memcfg.h
@@ -41,6 +41,8 @@  struct rte_mem_config {
 	rte_rwlock_t memory_hotplug_lock;
 	/**< Indicates whether memory hotplug request is in progress. */
 
+	uint8_t mp_status; /**< Multiprocess status. */
+
 	/* memory segments and zones */
 	struct rte_fbarray memzones; /**< Memzone descriptors. */
 
diff --git a/lib/librte_eal/common/eal_private.h b/lib/librte_eal/common/eal_private.h
index a77ac7a963..e82fb80aa0 100644
--- a/lib/librte_eal/common/eal_private.h
+++ b/lib/librte_eal/common/eal_private.h
@@ -396,6 +396,24 @@  uint64_t get_tsc_freq(void);
  */
 uint64_t get_tsc_freq_arch(void);
 
+/**
+ * Allocate a free lcore to associate to a non-EAL thread.
+ *
+ * @return
+ *   - the id of a lcore with role ROLE_NON_EAL on success.
+ *   - RTE_MAX_LCORE if none was available.
+ */
+unsigned int eal_lcore_non_eal_allocate(void);
+
+/**
+ * Release the lcore used by a non-EAL thread.
+ * Counterpart of eal_lcore_non_eal_allocate().
+ *
+ * @param lcore_id
+ *   The lcore with role ROLE_NON_EAL to release.
+ */
+void eal_lcore_non_eal_release(unsigned int lcore_id);
+
 /**
  * Prepare physical memory mapping
  * i.e. hugepages on Linux and
@@ -714,4 +732,14 @@  void __rte_thread_init(unsigned int lcore_id, rte_cpuset_t *cpuset);
  */
 void __rte_thread_uninit(void);
 
+/**
+ * Mark primary process as not supporting multi-process.
+ */
+bool eal_disable_multiprocess(void);
+
+/**
+ * Instruct primary process that a secondary process wants to attach.
+ */
+bool eal_enable_multiprocess(void);
+
 #endif /* _EAL_PRIVATE_H_ */
diff --git a/lib/librte_eal/freebsd/eal.c b/lib/librte_eal/freebsd/eal.c
index fd577daf44..7e5001a864 100644
--- a/lib/librte_eal/freebsd/eal.c
+++ b/lib/librte_eal/freebsd/eal.c
@@ -400,6 +400,10 @@  rte_config_init(void)
 		}
 		if (rte_eal_config_reattach() < 0)
 			return -1;
+		if (!eal_enable_multiprocess()) {
+			RTE_LOG(ERR, EAL, "Primary process refused secondary attachment\n");
+			return -1;
+		}
 		eal_mcfg_update_internal();
 		break;
 	case RTE_PROC_AUTO:
diff --git a/lib/librte_eal/include/rte_lcore.h b/lib/librte_eal/include/rte_lcore.h
index 3968c40693..2fd1a03275 100644
--- a/lib/librte_eal/include/rte_lcore.h
+++ b/lib/librte_eal/include/rte_lcore.h
@@ -31,6 +31,7 @@  enum rte_lcore_role_t {
 	ROLE_RTE,
 	ROLE_OFF,
 	ROLE_SERVICE,
+	ROLE_NON_EAL,
 };
 
 /**
@@ -67,7 +68,8 @@  rte_lcore_has_role(unsigned int lcore_id, enum rte_lcore_role_t role);
  *   to run threads with lcore IDs 0, 1, 2 and 3 on physical core 10..
  *
  * @return
- *  Logical core ID (in EAL thread) or LCORE_ID_ANY (in non-EAL thread)
+ *  Logical core ID (in EAL thread or registered non-EAL thread) or
+ *  LCORE_ID_ANY (in unregistered non-EAL thread)
  */
 static inline unsigned
 rte_lcore_id(void)
@@ -279,6 +281,30 @@  int rte_thread_setname(pthread_t id, const char *name);
 __rte_experimental
 int rte_thread_getname(pthread_t id, char *name, size_t len);
 
+/**
+ * Register current non-EAL thread as a lcore.
+ *
+ * @note This API is not compatible with the multi-process feature:
+ * - if a primary process registers a non-EAL thread, then no secondary process
+ *   will initialise.
+ * - if a secondary process initialises successfully, trying to register a
+ *   non-EAL thread from either primary or secondary processes will always end
+ *   up with the thread getting LCORE_ID_ANY as lcore.
+ *
+ * @return
+ *   On success, return 0; otherwise return -1 with rte_errno set.
+ */
+__rte_experimental
+int
+rte_thread_register(void);
+
+/**
+ * Unregister current thread and release lcore if one was associated.
+ */
+__rte_experimental
+void
+rte_thread_unregister(void);
+
 /**
  * Create a control thread.
  *
diff --git a/lib/librte_eal/linux/eal.c b/lib/librte_eal/linux/eal.c
index bd089cdd44..f4daf46c4e 100644
--- a/lib/librte_eal/linux/eal.c
+++ b/lib/librte_eal/linux/eal.c
@@ -492,6 +492,10 @@  rte_config_init(void)
 		}
 		if (rte_eal_config_reattach() < 0)
 			return -1;
+		if (!eal_enable_multiprocess()) {
+			RTE_LOG(ERR, EAL, "Primary process refused secondary attachment\n");
+			return -1;
+		}
 		eal_mcfg_update_internal();
 		break;
 	case RTE_PROC_AUTO:
diff --git a/lib/librte_eal/rte_eal_version.map b/lib/librte_eal/rte_eal_version.map
index 0d42d44ce9..5503dd7620 100644
--- a/lib/librte_eal/rte_eal_version.map
+++ b/lib/librte_eal/rte_eal_version.map
@@ -393,6 +393,10 @@  EXPERIMENTAL {
 	rte_trace_point_lookup;
 	rte_trace_regexp;
 	rte_trace_save;
+
+	# added in 20.08
+	rte_thread_register;
+	rte_thread_unregister;
 };
 
 INTERNAL {
diff --git a/lib/librte_eal/windows/eal.c b/lib/librte_eal/windows/eal.c
index addac62ae5..9fc831fe6f 100644
--- a/lib/librte_eal/windows/eal.c
+++ b/lib/librte_eal/windows/eal.c
@@ -66,6 +66,12 @@  eal_proc_type_detect(void)
 	return ptype;
 }
 
+bool
+eal_disable_multiprocess(void)
+{
+	return true;
+}
+
 /* display usage */
 static void
 eal_usage(const char *prgname)
diff --git a/lib/librte_mempool/rte_mempool.h b/lib/librte_mempool/rte_mempool.h
index 652d19f9f1..9e0ee052b3 100644
--- a/lib/librte_mempool/rte_mempool.h
+++ b/lib/librte_mempool/rte_mempool.h
@@ -28,9 +28,9 @@ 
  * rte_mempool_get() or rte_mempool_put() are designed to be called from an EAL
  * thread due to the internal per-lcore cache. Due to the lack of caching,
  * rte_mempool_get() or rte_mempool_put() performance will suffer when called
- * by non-EAL threads. Instead, non-EAL threads should call
- * rte_mempool_generic_get() or rte_mempool_generic_put() with a user cache
- * created with rte_mempool_cache_create().
+ * by unregistered non-EAL threads. Instead, unregistered non-EAL threads
+ * should call rte_mempool_generic_get() or rte_mempool_generic_put() with a
+ * user cache created with rte_mempool_cache_create().
  */
 
 #include <stdio.h>
@@ -1233,7 +1233,7 @@  void rte_mempool_dump(FILE *f, struct rte_mempool *mp);
 /**
  * Create a user-owned mempool cache.
  *
- * This can be used by non-EAL threads to enable caching when they
+ * This can be used by unregistered non-EAL threads to enable caching when they
  * interact with a mempool.
  *
  * @param size
@@ -1264,7 +1264,8 @@  rte_mempool_cache_free(struct rte_mempool_cache *cache);
  * @param lcore_id
  *   The logical core id.
  * @return
- *   A pointer to the mempool cache or NULL if disabled or non-EAL thread.
+ *   A pointer to the mempool cache or NULL if disabled or unregistered non-EAL
+ *   thread.
  */
 static __rte_always_inline struct rte_mempool_cache *
 rte_mempool_default_cache(struct rte_mempool *mp, unsigned lcore_id)