summaryrefslogtreecommitdiffstats
path: root/src/spdk/dpdk/examples/performance-thread/common
diff options
context:
space:
mode:
Diffstat (limited to 'src/spdk/dpdk/examples/performance-thread/common')
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/arch/arm64/ctx.c62
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/arch/arm64/ctx.h55
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/arch/arm64/stack.h56
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/arch/x86/ctx.c37
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/arch/x86/ctx.h36
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/arch/x86/stack.h40
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/common.mk21
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread.c468
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread.h51
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_api.h784
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_cond.c184
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_cond.h30
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_diag.c293
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_diag.h112
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_diag_api.h304
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_int.h152
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_mutex.c224
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_mutex.h31
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_objcache.h136
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_pool.h277
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_queue.h247
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_sched.c544
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_sched.h104
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_timer.h68
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_tls.c222
-rw-r--r--src/spdk/dpdk/examples/performance-thread/common/lthread_tls.h35
26 files changed, 4573 insertions, 0 deletions
diff --git a/src/spdk/dpdk/examples/performance-thread/common/arch/arm64/ctx.c b/src/spdk/dpdk/examples/performance-thread/common/arch/arm64/ctx.c
new file mode 100644
index 000000000..7c5c91658
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/arch/arm64/ctx.c
@@ -0,0 +1,62 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2017 Cavium, Inc
+ */
+
+#include <rte_common.h>
+#include <ctx.h>
+
+void
+ctx_switch(struct ctx *new_ctx __rte_unused, struct ctx *curr_ctx __rte_unused)
+{
+ /* SAVE CURRENT CONTEXT */
+ asm volatile (
+ /* Save SP */
+ "mov x3, sp\n"
+ "str x3, [x1, #0]\n"
+
+ /* Save FP and LR */
+ "stp x29, x30, [x1, #8]\n"
+
+ /* Save Callee Saved Regs x19 - x28 */
+ "stp x19, x20, [x1, #24]\n"
+ "stp x21, x22, [x1, #40]\n"
+ "stp x23, x24, [x1, #56]\n"
+ "stp x25, x26, [x1, #72]\n"
+ "stp x27, x28, [x1, #88]\n"
+
+ /*
+ * Save bottom 64-bits of Callee Saved
+ * SIMD Regs v8 - v15
+ */
+ "stp d8, d9, [x1, #104]\n"
+ "stp d10, d11, [x1, #120]\n"
+ "stp d12, d13, [x1, #136]\n"
+ "stp d14, d15, [x1, #152]\n"
+ );
+
+ /* RESTORE NEW CONTEXT */
+ asm volatile (
+ /* Restore SP */
+ "ldr x3, [x0, #0]\n"
+ "mov sp, x3\n"
+
+ /* Restore FP and LR */
+ "ldp x29, x30, [x0, #8]\n"
+
+ /* Restore Callee Saved Regs x19 - x28 */
+ "ldp x19, x20, [x0, #24]\n"
+ "ldp x21, x22, [x0, #40]\n"
+ "ldp x23, x24, [x0, #56]\n"
+ "ldp x25, x26, [x0, #72]\n"
+ "ldp x27, x28, [x0, #88]\n"
+
+ /*
+ * Restore bottom 64-bits of Callee Saved
+ * SIMD Regs v8 - v15
+ */
+ "ldp d8, d9, [x0, #104]\n"
+ "ldp d10, d11, [x0, #120]\n"
+ "ldp d12, d13, [x0, #136]\n"
+ "ldp d14, d15, [x0, #152]\n"
+ );
+}
diff --git a/src/spdk/dpdk/examples/performance-thread/common/arch/arm64/ctx.h b/src/spdk/dpdk/examples/performance-thread/common/arch/arm64/ctx.h
new file mode 100644
index 000000000..74c2e7a73
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/arch/arm64/ctx.h
@@ -0,0 +1,55 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2017 Cavium, Inc
+ */
+
+#ifndef CTX_H
+#define CTX_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+ * CPU context registers
+ */
+struct ctx {
+ void *sp; /* 0 */
+ void *fp; /* 8 */
+ void *lr; /* 16 */
+
+ /* Callee Saved Generic Registers */
+ void *r19; /* 24 */
+ void *r20; /* 32 */
+ void *r21; /* 40 */
+ void *r22; /* 48 */
+ void *r23; /* 56 */
+ void *r24; /* 64 */
+ void *r25; /* 72 */
+ void *r26; /* 80 */
+ void *r27; /* 88 */
+ void *r28; /* 96 */
+
+ /*
+ * Callee Saved SIMD Registers. Only the bottom 64-bits
+ * of these registers needs to be saved.
+ */
+ void *v8; /* 104 */
+ void *v9; /* 112 */
+ void *v10; /* 120 */
+ void *v11; /* 128 */
+ void *v12; /* 136 */
+ void *v13; /* 144 */
+ void *v14; /* 152 */
+ void *v15; /* 160 */
+};
+
+
+void
+ctx_switch(struct ctx *new_ctx, struct ctx *curr_ctx);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* RTE_CTX_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/arch/arm64/stack.h b/src/spdk/dpdk/examples/performance-thread/common/arch/arm64/stack.h
new file mode 100644
index 000000000..722c47335
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/arch/arm64/stack.h
@@ -0,0 +1,56 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2017 Cavium, Inc
+ */
+
+#ifndef STACK_H
+#define STACK_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "lthread_int.h"
+
+/*
+ * Sets up the initial stack for the lthread.
+ */
+static inline void
+arch_set_stack(struct lthread *lt, void *func)
+{
+ void **stack_top = (void *)((char *)(lt->stack) + lt->stack_size);
+
+ /*
+ * Align stack_top to 16 bytes. Arm64 has the constraint that the
+ * stack pointer must always be quad-word aligned.
+ */
+ stack_top = (void **)(((unsigned long)(stack_top)) & ~0xfUL);
+
+ /*
+ * First Stack Frame
+ */
+ stack_top[0] = NULL;
+ stack_top[-1] = NULL;
+
+ /*
+ * Initialize the context
+ */
+ lt->ctx.fp = &stack_top[-1];
+ lt->ctx.sp = &stack_top[-2];
+
+ /*
+ * Here only the address of _lthread_exec is saved as the link
+ * register value. The argument to _lthread_exec i.e the address of
+ * the lthread struct is not saved. This is because the first
+ * argument to ctx_switch is the address of the new context,
+ * which also happens to be the address of required lthread struct.
+ * So while returning from ctx_switch into _thread_exec, parameter
+ * register x0 will always contain the required value.
+ */
+ lt->ctx.lr = func;
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* STACK_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/arch/x86/ctx.c b/src/spdk/dpdk/examples/performance-thread/common/arch/x86/ctx.c
new file mode 100644
index 000000000..d63fd9fc0
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/arch/x86/ctx.c
@@ -0,0 +1,37 @@
+/*
+ * SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2015 Intel Corporation.
+ * Copyright 2012 Hasan Alayli <halayli@gmail.com>
+ */
+
+#if defined(__x86_64__)
+__asm__ (
+".text\n"
+".p2align 4,,15\n"
+".globl ctx_switch\n"
+".globl _ctx_switch\n"
+"ctx_switch:\n"
+"_ctx_switch:\n"
+" movq %rsp, 0(%rsi) # save stack_pointer\n"
+" movq %rbp, 8(%rsi) # save frame_pointer\n"
+" movq (%rsp), %rax # save insn_pointer\n"
+" movq %rax, 16(%rsi)\n"
+" movq %rbx, 24(%rsi)\n # save rbx,r12-r15\n"
+" movq 24(%rdi), %rbx\n"
+" movq %r15, 56(%rsi)\n"
+" movq %r14, 48(%rsi)\n"
+" movq 48(%rdi), %r14\n"
+" movq 56(%rdi), %r15\n"
+" movq %r13, 40(%rsi)\n"
+" movq %r12, 32(%rsi)\n"
+" movq 32(%rdi), %r12\n"
+" movq 40(%rdi), %r13\n"
+" movq 0(%rdi), %rsp # restore stack_pointer\n"
+" movq 16(%rdi), %rax # restore insn_pointer\n"
+" movq 8(%rdi), %rbp # restore frame_pointer\n"
+" movq %rax, (%rsp)\n"
+" ret\n"
+ );
+#else
+#pragma GCC error "__x86_64__ is not defined"
+#endif
diff --git a/src/spdk/dpdk/examples/performance-thread/common/arch/x86/ctx.h b/src/spdk/dpdk/examples/performance-thread/common/arch/x86/ctx.h
new file mode 100644
index 000000000..c6a46c529
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/arch/x86/ctx.h
@@ -0,0 +1,36 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2015 Intel Corporation
+ */
+
+
+#ifndef CTX_H
+#define CTX_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+ * CPU context registers
+ */
+struct ctx {
+ void *rsp; /* 0 */
+ void *rbp; /* 8 */
+ void *rip; /* 16 */
+ void *rbx; /* 24 */
+ void *r12; /* 32 */
+ void *r13; /* 40 */
+ void *r14; /* 48 */
+ void *r15; /* 56 */
+};
+
+
+void
+ctx_switch(struct ctx *new_ctx, struct ctx *curr_ctx);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* RTE_CTX_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/arch/x86/stack.h b/src/spdk/dpdk/examples/performance-thread/common/arch/x86/stack.h
new file mode 100644
index 000000000..7cdd5c7ae
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/arch/x86/stack.h
@@ -0,0 +1,40 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2015 Intel Corporation.
+ * Copyright(c) Cavium, Inc. 2017.
+ * All rights reserved
+ * Copyright (C) 2012, Hasan Alayli <halayli@gmail.com>
+ * Portions derived from: https://github.com/halayli/lthread
+ * With permissions from Hasan Alayli to use them as BSD-3-Clause
+ */
+
+#ifndef STACK_H
+#define STACK_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "lthread_int.h"
+
+/*
+ * Sets up the initial stack for the lthread.
+ */
+static inline void
+arch_set_stack(struct lthread *lt, void *func)
+{
+ char *stack_top = (char *)(lt->stack) + lt->stack_size;
+ void **s = (void **)stack_top;
+
+ /* set initial context */
+ s[-3] = NULL;
+ s[-2] = (void *)lt;
+ lt->ctx.rsp = (void *)(stack_top - (4 * sizeof(void *)));
+ lt->ctx.rbp = (void *)(stack_top - (3 * sizeof(void *)));
+ lt->ctx.rip = func;
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* STACK_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/common.mk b/src/spdk/dpdk/examples/performance-thread/common/common.mk
new file mode 100644
index 000000000..5e2b18a9f
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/common.mk
@@ -0,0 +1,21 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright(c) 2015 Intel Corporation
+
+# list the C files belonging to the lthread subsystem, these are common to all
+# lthread apps. Any makefile including this should set VPATH to include this
+# directory path
+#
+
+MKFILE_PATH=$(abspath $(dir $(lastword $(MAKEFILE_LIST))))
+
+ifeq ($(CONFIG_RTE_ARCH_X86_64),y)
+ARCH_PATH += $(MKFILE_PATH)/arch/x86
+else ifeq ($(CONFIG_RTE_ARCH_ARM64),y)
+ARCH_PATH += $(MKFILE_PATH)/arch/arm64
+endif
+
+VPATH := $(MKFILE_PATH) $(ARCH_PATH)
+
+SRCS-y += lthread.c lthread_sched.c lthread_cond.c lthread_tls.c lthread_mutex.c lthread_diag.c ctx.c
+
+INCLUDES += -I$(MKFILE_PATH) -I$(ARCH_PATH)
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread.c b/src/spdk/dpdk/examples/performance-thread/common/lthread.c
new file mode 100644
index 000000000..3f1f48db4
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread.c
@@ -0,0 +1,468 @@
+/*
+ * SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2015 Intel Corporation.
+ * Copyright 2012 Hasan Alayli <halayli@gmail.com>
+ */
+
+#define RTE_MEM 1
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdint.h>
+#include <stddef.h>
+#include <limits.h>
+#include <inttypes.h>
+#include <unistd.h>
+#include <pthread.h>
+#include <fcntl.h>
+#include <sys/time.h>
+#include <sys/mman.h>
+
+#include <rte_log.h>
+#include <ctx.h>
+#include <stack.h>
+
+#include "lthread_api.h"
+#include "lthread.h"
+#include "lthread_timer.h"
+#include "lthread_tls.h"
+#include "lthread_objcache.h"
+#include "lthread_diag.h"
+
+
+/*
+ * This function gets called after an lthread function has returned.
+ */
+void _lthread_exit_handler(struct lthread *lt)
+{
+
+ lt->state |= BIT(ST_LT_EXITED);
+
+ if (!(lt->state & BIT(ST_LT_DETACH))) {
+ /* thread is this not explicitly detached
+ * it must be joinable, so we call lthread_exit().
+ */
+ lthread_exit(NULL);
+ }
+
+ /* if we get here the thread is detached so we can reschedule it,
+ * allowing the scheduler to free it
+ */
+ _reschedule();
+}
+
+
+/*
+ * Free resources allocated to an lthread
+ */
+void _lthread_free(struct lthread *lt)
+{
+
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_FREE, lt, 0);
+
+ /* invoke any user TLS destructor functions */
+ _lthread_tls_destroy(lt);
+
+ /* free memory allocated for TLS defined using RTE_PER_LTHREAD macros */
+ if (sizeof(void *) < (uint64_t)RTE_PER_LTHREAD_SECTION_SIZE)
+ _lthread_objcache_free(lt->tls->root_sched->per_lthread_cache,
+ lt->per_lthread_data);
+
+ /* free pthread style TLS memory */
+ _lthread_objcache_free(lt->tls->root_sched->tls_cache, lt->tls);
+
+ /* free the stack */
+ _lthread_objcache_free(lt->stack_container->root_sched->stack_cache,
+ lt->stack_container);
+
+ /* now free the thread */
+ _lthread_objcache_free(lt->root_sched->lthread_cache, lt);
+
+}
+
+/*
+ * Allocate a stack and maintain a cache of stacks
+ */
+struct lthread_stack *_stack_alloc(void)
+{
+ struct lthread_stack *s;
+
+ s = _lthread_objcache_alloc((THIS_SCHED)->stack_cache);
+ RTE_ASSERT(s != NULL);
+
+ s->root_sched = THIS_SCHED;
+ s->stack_size = LTHREAD_MAX_STACK_SIZE;
+ return s;
+}
+
+/*
+ * Execute a ctx by invoking the start function
+ * On return call an exit handler if the user has provided one
+ */
+static void _lthread_exec(void *arg)
+{
+ struct lthread *lt = (struct lthread *)arg;
+
+ /* invoke the contexts function */
+ lt->fun(lt->arg);
+ /* do exit handling */
+ if (lt->exit_handler != NULL)
+ lt->exit_handler(lt);
+}
+
+/*
+ * Initialize an lthread
+ * Set its function, args, and exit handler
+ */
+void
+_lthread_init(struct lthread *lt,
+ lthread_func_t fun, void *arg, lthread_exit_func exit_handler)
+{
+
+ /* set ctx func and args */
+ lt->fun = fun;
+ lt->arg = arg;
+ lt->exit_handler = exit_handler;
+
+ /* set initial state */
+ lt->birth = _sched_now();
+ lt->state = BIT(ST_LT_INIT);
+ lt->join = LT_JOIN_INITIAL;
+}
+
+/*
+ * set the lthread stack
+ */
+void _lthread_set_stack(struct lthread *lt, void *stack, size_t stack_size)
+{
+ /* set stack */
+ lt->stack = stack;
+ lt->stack_size = stack_size;
+
+ arch_set_stack(lt, _lthread_exec);
+}
+
+/*
+ * Create an lthread on the current scheduler
+ * If there is no current scheduler on this pthread then first create one
+ */
+int
+lthread_create(struct lthread **new_lt, int lcore_id,
+ lthread_func_t fun, void *arg)
+{
+ if ((new_lt == NULL) || (fun == NULL))
+ return POSIX_ERRNO(EINVAL);
+
+ if (lcore_id < 0)
+ lcore_id = rte_lcore_id();
+ else if (lcore_id > LTHREAD_MAX_LCORES)
+ return POSIX_ERRNO(EINVAL);
+
+ struct lthread *lt = NULL;
+
+ if (THIS_SCHED == NULL) {
+ THIS_SCHED = _lthread_sched_create(0);
+ if (THIS_SCHED == NULL) {
+ perror("Failed to create scheduler");
+ return POSIX_ERRNO(EAGAIN);
+ }
+ }
+
+ /* allocate a thread structure */
+ lt = _lthread_objcache_alloc((THIS_SCHED)->lthread_cache);
+ if (lt == NULL)
+ return POSIX_ERRNO(EAGAIN);
+
+ bzero(lt, sizeof(struct lthread));
+ lt->root_sched = THIS_SCHED;
+
+ /* set the function args and exit handlder */
+ _lthread_init(lt, fun, arg, _lthread_exit_handler);
+
+ /* put it in the ready queue */
+ *new_lt = lt;
+
+ if (lcore_id < 0)
+ lcore_id = rte_lcore_id();
+
+ DIAG_CREATE_EVENT(lt, LT_DIAG_LTHREAD_CREATE);
+
+ rte_wmb();
+ _ready_queue_insert(_lthread_sched_get(lcore_id), lt);
+ return 0;
+}
+
+/*
+ * Schedules lthread to sleep for `nsecs`
+ * setting the lthread state to LT_ST_SLEEPING.
+ * lthread state is cleared upon resumption or expiry.
+ */
+static inline void _lthread_sched_sleep(struct lthread *lt, uint64_t nsecs)
+{
+ uint64_t state = lt->state;
+ uint64_t clks = _ns_to_clks(nsecs);
+
+ if (clks) {
+ _timer_start(lt, clks);
+ lt->state = state | BIT(ST_LT_SLEEPING);
+ }
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_SLEEP, clks, 0);
+ _suspend();
+}
+
+
+
+/*
+ * Cancels any running timer.
+ * This can be called multiple times on the same lthread regardless if it was
+ * sleeping or not.
+ */
+int _lthread_desched_sleep(struct lthread *lt)
+{
+ uint64_t state = lt->state;
+
+ if (state & BIT(ST_LT_SLEEPING)) {
+ _timer_stop(lt);
+ state &= (CLEARBIT(ST_LT_SLEEPING) & CLEARBIT(ST_LT_EXPIRED));
+ lt->state = state | BIT(ST_LT_READY);
+ return 1;
+ }
+ return 0;
+}
+
+/*
+ * set user data pointer in an lthread
+ */
+void lthread_set_data(void *data)
+{
+ if (sizeof(void *) == RTE_PER_LTHREAD_SECTION_SIZE)
+ THIS_LTHREAD->per_lthread_data = data;
+}
+
+/*
+ * Retrieve user data pointer from an lthread
+ */
+void *lthread_get_data(void)
+{
+ return THIS_LTHREAD->per_lthread_data;
+}
+
+/*
+ * Return the current lthread handle
+ */
+struct lthread *lthread_current(void)
+{
+ struct lthread_sched *sched = THIS_SCHED;
+
+ if (sched)
+ return sched->current_lthread;
+ return NULL;
+}
+
+
+
+/*
+ * Tasklet to cancel a thread
+ */
+static void *
+_cancel(void *arg)
+{
+ struct lthread *lt = (struct lthread *) arg;
+
+ lt->state |= BIT(ST_LT_CANCELLED);
+ lthread_detach();
+ return NULL;
+}
+
+
+/*
+ * Mark the specified as canceled
+ */
+int lthread_cancel(struct lthread *cancel_lt)
+{
+ struct lthread *lt;
+
+ if ((cancel_lt == NULL) || (cancel_lt == THIS_LTHREAD))
+ return POSIX_ERRNO(EINVAL);
+
+ DIAG_EVENT(cancel_lt, LT_DIAG_LTHREAD_CANCEL, cancel_lt, 0);
+
+ if (cancel_lt->sched != THIS_SCHED) {
+
+ /* spawn task-let to cancel the thread */
+ lthread_create(&lt,
+ cancel_lt->sched->lcore_id,
+ _cancel,
+ cancel_lt);
+ return 0;
+ }
+ cancel_lt->state |= BIT(ST_LT_CANCELLED);
+ return 0;
+}
+
+/*
+ * Suspend the current lthread for specified time
+ */
+void lthread_sleep(uint64_t nsecs)
+{
+ struct lthread *lt = THIS_LTHREAD;
+
+ _lthread_sched_sleep(lt, nsecs);
+
+}
+
+/*
+ * Suspend the current lthread for specified time
+ */
+void lthread_sleep_clks(uint64_t clks)
+{
+ struct lthread *lt = THIS_LTHREAD;
+ uint64_t state = lt->state;
+
+ if (clks) {
+ _timer_start(lt, clks);
+ lt->state = state | BIT(ST_LT_SLEEPING);
+ }
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_SLEEP, clks, 0);
+ _suspend();
+}
+
+/*
+ * Requeue the current thread to the back of the ready queue
+ */
+void lthread_yield(void)
+{
+ struct lthread *lt = THIS_LTHREAD;
+
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_YIELD, 0, 0);
+
+ _ready_queue_insert(THIS_SCHED, lt);
+ ctx_switch(&(THIS_SCHED)->ctx, &lt->ctx);
+}
+
+/*
+ * Exit the current lthread
+ * If a thread is joining pass the user pointer to it
+ */
+void lthread_exit(void *ptr)
+{
+ struct lthread *lt = THIS_LTHREAD;
+
+ /* if thread is detached (this is not valid) just exit */
+ if (lt->state & BIT(ST_LT_DETACH))
+ return;
+
+ /* There is a race between lthread_join() and lthread_exit()
+ * - if exit before join then we suspend and resume on join
+ * - if join before exit then we resume the joining thread
+ */
+ if ((lt->join == LT_JOIN_INITIAL)
+ && rte_atomic64_cmpset(&lt->join, LT_JOIN_INITIAL,
+ LT_JOIN_EXITING)) {
+
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_EXIT, 1, 0);
+ _suspend();
+ /* set the exit value */
+ if ((ptr != NULL) && (lt->lt_join->lt_exit_ptr != NULL))
+ *(lt->lt_join->lt_exit_ptr) = ptr;
+
+ /* let the joining thread know we have set the exit value */
+ lt->join = LT_JOIN_EXIT_VAL_SET;
+ } else {
+
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_EXIT, 0, 0);
+ /* set the exit value */
+ if ((ptr != NULL) && (lt->lt_join->lt_exit_ptr != NULL))
+ *(lt->lt_join->lt_exit_ptr) = ptr;
+ /* let the joining thread know we have set the exit value */
+ lt->join = LT_JOIN_EXIT_VAL_SET;
+ _ready_queue_insert(lt->lt_join->sched,
+ (struct lthread *)lt->lt_join);
+ }
+
+
+ /* wait until the joinging thread has collected the exit value */
+ while (lt->join != LT_JOIN_EXIT_VAL_READ)
+ _reschedule();
+
+ /* reset join state */
+ lt->join = LT_JOIN_INITIAL;
+
+ /* detach it so its resources can be released */
+ lt->state |= (BIT(ST_LT_DETACH) | BIT(ST_LT_EXITED));
+}
+
+/*
+ * Join an lthread
+ * Suspend until the joined thread returns
+ */
+int lthread_join(struct lthread *lt, void **ptr)
+{
+ if (lt == NULL)
+ return POSIX_ERRNO(EINVAL);
+
+ struct lthread *current = THIS_LTHREAD;
+ uint64_t lt_state = lt->state;
+
+ /* invalid to join a detached thread, or a thread that is joined */
+ if ((lt_state & BIT(ST_LT_DETACH)) || (lt->join == LT_JOIN_THREAD_SET))
+ return POSIX_ERRNO(EINVAL);
+ /* pointer to the joining thread and a poingter to return a value */
+ lt->lt_join = current;
+ current->lt_exit_ptr = ptr;
+ /* There is a race between lthread_join() and lthread_exit()
+ * - if join before exit we suspend and will resume when exit is called
+ * - if exit before join we resume the exiting thread
+ */
+ if ((lt->join == LT_JOIN_INITIAL)
+ && rte_atomic64_cmpset(&lt->join, LT_JOIN_INITIAL,
+ LT_JOIN_THREAD_SET)) {
+
+ DIAG_EVENT(current, LT_DIAG_LTHREAD_JOIN, lt, 1);
+ _suspend();
+ } else {
+ DIAG_EVENT(current, LT_DIAG_LTHREAD_JOIN, lt, 0);
+ _ready_queue_insert(lt->sched, lt);
+ }
+
+ /* wait for exiting thread to set return value */
+ while (lt->join != LT_JOIN_EXIT_VAL_SET)
+ _reschedule();
+
+ /* collect the return value */
+ if (ptr != NULL)
+ *ptr = *current->lt_exit_ptr;
+
+ /* let the exiting thread proceed to exit */
+ lt->join = LT_JOIN_EXIT_VAL_READ;
+ return 0;
+}
+
+
+/*
+ * Detach current lthread
+ * A detached thread cannot be joined
+ */
+void lthread_detach(void)
+{
+ struct lthread *lt = THIS_LTHREAD;
+
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_DETACH, 0, 0);
+
+ uint64_t state = lt->state;
+
+ lt->state = state | BIT(ST_LT_DETACH);
+}
+
+/*
+ * Set function name of an lthread
+ * this is a debug aid
+ */
+void lthread_set_funcname(const char *f)
+{
+ struct lthread *lt = THIS_LTHREAD;
+
+ strncpy(lt->funcname, f, sizeof(lt->funcname));
+ lt->funcname[sizeof(lt->funcname)-1] = 0;
+}
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread.h b/src/spdk/dpdk/examples/performance-thread/common/lthread.h
new file mode 100644
index 000000000..4c945cf76
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread.h
@@ -0,0 +1,51 @@
+/*
+ * SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2015 Intel Corporation.
+ * Copyright 2012 Hasan Alayli <halayli@gmail.com>
+ */
+#ifndef LTHREAD_H_
+#define LTHREAD_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <rte_per_lcore.h>
+
+#include "lthread_api.h"
+#include "lthread_diag.h"
+
+struct lthread;
+struct lthread_sched;
+
+/* function to be called when a context function returns */
+typedef void (*lthread_exit_func) (struct lthread *);
+
+void _lthread_exit_handler(struct lthread *lt);
+
+void lthread_set_funcname(const char *f);
+
+void _lthread_sched_busy_sleep(struct lthread *lt, uint64_t nsecs);
+
+int _lthread_desched_sleep(struct lthread *lt);
+
+void _lthread_free(struct lthread *lt);
+
+struct lthread_sched *_lthread_sched_get(unsigned int lcore_id);
+
+struct lthread_stack *_stack_alloc(void);
+
+struct
+lthread_sched *_lthread_sched_create(size_t stack_size);
+
+void
+_lthread_init(struct lthread *lt,
+ lthread_func_t fun, void *arg, lthread_exit_func exit_handler);
+
+void _lthread_set_stack(struct lthread *lt, void *stack, size_t stack_size);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* LTHREAD_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_api.h b/src/spdk/dpdk/examples/performance-thread/common/lthread_api.h
new file mode 100644
index 000000000..e6879ea5c
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_api.h
@@ -0,0 +1,784 @@
+/*
+ * SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2015 Intel Corporation.
+ * Copyright 2012 Hasan Alayli <halayli@gmail.com>
+ */
+/**
+ * @file lthread_api.h
+ *
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice
+ *
+ * This file contains the public API for the L-thread subsystem
+ *
+ * The L_thread subsystem provides a simple cooperative scheduler to
+ * enable arbitrary functions to run as cooperative threads within a
+ * single P-thread.
+ *
+ * The subsystem provides a P-thread like API that is intended to assist in
+ * reuse of legacy code written for POSIX p_threads.
+ *
+ * The L-thread subsystem relies on cooperative multitasking, as such
+ * an L-thread must possess frequent rescheduling points. Often these
+ * rescheduling points are provided transparently when the application
+ * invokes an L-thread API.
+ *
+ * In some applications it is possible that the program may enter a loop the
+ * exit condition for which depends on the action of another thread or a
+ * response from hardware. In such a case it is necessary to yield the thread
+ * periodically in the loop body, to allow other threads an opportunity to
+ * run. This can be done by inserting a call to lthread_yield() or
+ * lthread_sleep(n) in the body of the loop.
+ *
+ * If the application makes expensive / blocking system calls or does other
+ * work that would take an inordinate amount of time to complete, this will
+ * stall the cooperative scheduler resulting in very poor performance.
+ *
+ * In such cases an L-thread can be migrated temporarily to another scheduler
+ * running in a different P-thread on another core. When the expensive or
+ * blocking operation is completed it can be migrated back to the original
+ * scheduler. In this way other threads can continue to run on the original
+ * scheduler and will be completely unaffected by the blocking behaviour.
+ * To migrate an L-thread to another scheduler the API lthread_set_affinity()
+ * is provided.
+ *
+ * If L-threads that share data are running on the same core it is possible
+ * to design programs where mutual exclusion mechanisms to protect shared data
+ * can be avoided. This is due to the fact that the cooperative threads cannot
+ * preempt each other.
+ *
+ * There are two cases where mutual exclusion mechanisms are necessary.
+ *
+ * a) Where the L-threads sharing data are running on different cores.
+ * b) Where code must yield while updating data shared with another thread.
+ *
+ * The L-thread subsystem provides a set of mutex APIs to help with such
+ * scenarios, however excessive reliance on on these will impact performance
+ * and is best avoided if possible.
+ *
+ * L-threads can synchronise using a fast condition variable implementation
+ * that supports signal and broadcast. An L-thread running on any core can
+ * wait on a condition.
+ *
+ * L-threads can have L-thread local storage with an API modelled on either the
+ * P-thread get/set specific API or using PER_LTHREAD macros modelled on the
+ * RTE_PER_LCORE macros. Alternatively a simple user data pointer may be set
+ * and retrieved from a thread.
+ */
+#ifndef LTHREAD_H
+#define LTHREAD_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stdint.h>
+#include <sys/socket.h>
+#include <fcntl.h>
+#include <netinet/in.h>
+
+#include <rte_cycles.h>
+
+
+struct lthread;
+struct lthread_cond;
+struct lthread_mutex;
+
+struct lthread_condattr;
+struct lthread_mutexattr;
+
+typedef void *(*lthread_func_t) (void *);
+
+/*
+ * Define the size of stack for an lthread
+ * Then this is the size that will be allocated on lthread creation
+ * This is a fixed size and will not grow.
+ */
+#define LTHREAD_MAX_STACK_SIZE (1024*64)
+
+/**
+ * Define the maximum number of TLS keys that can be created
+ *
+ */
+#define LTHREAD_MAX_KEYS 1024
+
+/**
+ * Define the maximum number of attempts to destroy an lthread's
+ * TLS data on thread exit
+ */
+#define LTHREAD_DESTRUCTOR_ITERATIONS 4
+
+
+/**
+ * Define the maximum number of lcores that will support lthreads
+ */
+#define LTHREAD_MAX_LCORES RTE_MAX_LCORE
+
+/**
+ * How many lthread objects to pre-allocate as the system grows
+ * applies to lthreads + stacks, TLS, mutexs, cond vars.
+ *
+ * @see _lthread_alloc()
+ * @see _cond_alloc()
+ * @see _mutex_alloc()
+ *
+ */
+#define LTHREAD_PREALLOC 100
+
+/**
+ * Set the number of schedulers in the system.
+ *
+ * This function may optionally be called before starting schedulers.
+ *
+ * If the number of schedulers is not set, or set to 0 then each scheduler
+ * will begin scheduling lthreads immediately it is started.
+
+ * If the number of schedulers is set to greater than 0, then each scheduler
+ * will wait until all schedulers have started before beginning to schedule
+ * lthreads.
+ *
+ * If an application wishes to have threads migrate between cores using
+ * lthread_set_affinity(), or join threads running on other cores using
+ * lthread_join(), then it is prudent to set the number of schedulers to ensure
+ * that all schedulers are initialised beforehand.
+ *
+ * @param num
+ * the number of schedulers in the system
+ * @return
+ * the number of schedulers in the system
+ */
+int lthread_num_schedulers_set(int num);
+
+/**
+ * Return the number of schedulers currently running
+ * @return
+ * the number of schedulers in the system
+ */
+int lthread_active_schedulers(void);
+
+/**
+ * Shutdown the specified scheduler
+ *
+ * This function tells the specified scheduler to
+ * exit if/when there is no more work to do.
+ *
+ * Note that although the scheduler will stop
+ * resources are not freed.
+ *
+ * @param lcore
+ * The lcore of the scheduler to shutdown
+ *
+ * @return
+ * none
+ */
+void lthread_scheduler_shutdown(unsigned lcore);
+
+/**
+ * Shutdown all schedulers
+ *
+ * This function tells all schedulers including the current scheduler to
+ * exit if/when there is no more work to do.
+ *
+ * Note that although the schedulers will stop
+ * resources are not freed.
+ *
+ * @return
+ * none
+ */
+void lthread_scheduler_shutdown_all(void);
+
+/**
+ * Run the lthread scheduler
+ *
+ * Runs the lthread scheduler.
+ * This function returns only if/when all lthreads have exited.
+ * This function must be the main loop of an EAL thread.
+ *
+ * @return
+ * none
+ */
+
+void lthread_run(void);
+
+/**
+ * Create an lthread
+ *
+ * Creates an lthread and places it in the ready queue on a particular
+ * lcore.
+ *
+ * If no scheduler exists yet on the current lcore then one is created.
+ *
+ * @param new_lt
+ * Pointer to an lthread pointer that will be initialized
+ * @param lcore
+ * the lcore the thread should be started on or the current lcore
+ * -1 the current lcore
+ * 0 - LTHREAD_MAX_LCORES any other lcore
+ * @param lthread_func
+ * Pointer to the function the for the thread to run
+ * @param arg
+ * Pointer to args that will be passed to the thread
+ *
+ * @return
+ * 0 success
+ * EAGAIN no resources available
+ * EINVAL NULL thread or function pointer, or lcore_id out of range
+ */
+int
+lthread_create(struct lthread **new_lt,
+ int lcore, lthread_func_t func, void *arg);
+
+/**
+ * Cancel an lthread
+ *
+ * Cancels an lthread and causes it to be terminated
+ * If the lthread is detached it will be freed immediately
+ * otherwise its resources will not be released until it is joined.
+ *
+ * @param new_lt
+ * Pointer to an lthread that will be cancelled
+ *
+ * @return
+ * 0 success
+ * EINVAL thread was NULL
+ */
+int lthread_cancel(struct lthread *lt);
+
+/**
+ * Join an lthread
+ *
+ * Joins the current thread with the specified lthread, and waits for that
+ * thread to exit.
+ * Passes an optional pointer to collect returned data.
+ *
+ * @param lt
+ * Pointer to the lthread to be joined
+ * @param ptr
+ * Pointer to pointer to collect returned data
+ *
+0 * @return
+ * 0 success
+ * EINVAL lthread could not be joined.
+ */
+int lthread_join(struct lthread *lt, void **ptr);
+
+/**
+ * Detach an lthread
+ *
+ * Detaches the current thread
+ * On exit a detached lthread will be freed immediately and will not wait
+ * to be joined. The default state for a thread is not detached.
+ *
+ * @return
+ * none
+ */
+void lthread_detach(void);
+
+/**
+ * Exit an lthread
+ *
+ * Terminate the current thread, optionally return data.
+ * The data may be collected by lthread_join()
+ *
+ * After calling this function the lthread will be suspended until it is
+ * joined. After it is joined then its resources will be freed.
+ *
+ * @param ptr
+ * Pointer to pointer to data to be returned
+ *
+ * @return
+ * none
+ */
+void lthread_exit(void *val);
+
+/**
+ * Cause the current lthread to sleep for n nanoseconds
+ *
+ * The current thread will be suspended until the specified time has elapsed
+ * or has been exceeded.
+ *
+ * Execution will switch to the next lthread that is ready to run
+ *
+ * @param nsecs
+ * Number of nanoseconds to sleep
+ *
+ * @return
+ * none
+ */
+void lthread_sleep(uint64_t nsecs);
+
+/**
+ * Cause the current lthread to sleep for n cpu clock ticks
+ *
+ * The current thread will be suspended until the specified time has elapsed
+ * or has been exceeded.
+ *
+ * Execution will switch to the next lthread that is ready to run
+ *
+ * @param clks
+ * Number of clock ticks to sleep
+ *
+ * @return
+ * none
+ */
+void lthread_sleep_clks(uint64_t clks);
+
+/**
+ * Yield the current lthread
+ *
+ * The current thread will yield and execution will switch to the
+ * next lthread that is ready to run
+ *
+ * @return
+ * none
+ */
+void lthread_yield(void);
+
+/**
+ * Migrate the current thread to another scheduler
+ *
+ * This function migrates the current thread to another scheduler.
+ * Execution will switch to the next lthread that is ready to run on the
+ * current scheduler. The current thread will be resumed on the new scheduler.
+ *
+ * @param lcore
+ * The lcore to migrate to
+ *
+ * @return
+ * 0 success we are now running on the specified core
+ * EINVAL the destination lcore was not valid
+ */
+int lthread_set_affinity(unsigned lcore);
+
+/**
+ * Return the current lthread
+ *
+ * Returns the current lthread
+ *
+ * @return
+ * pointer to the current lthread
+ */
+struct lthread
+*lthread_current(void);
+
+/**
+ * Associate user data with an lthread
+ *
+ * This function sets a user data pointer in the current lthread
+ * The pointer can be retrieved with lthread_get_data()
+ * It is the users responsibility to allocate and free any data referenced
+ * by the user pointer.
+ *
+ * @param data
+ * pointer to user data
+ *
+ * @return
+ * none
+ */
+void lthread_set_data(void *data);
+
+/**
+ * Get user data for the current lthread
+ *
+ * This function returns a user data pointer for the current lthread
+ * The pointer must first be set with lthread_set_data()
+ * It is the users responsibility to allocate and free any data referenced
+ * by the user pointer.
+ *
+ * @return
+ * pointer to user data
+ */
+void
+*lthread_get_data(void);
+
+struct lthread_key;
+typedef void (*tls_destructor_func) (void *);
+
+/**
+ * Create a key for lthread TLS
+ *
+ * This function is modelled on pthread_key_create
+ * It creates a thread-specific data key visible to all lthreads on the
+ * current scheduler.
+ *
+ * Key values may be used to locate thread-specific data.
+ * The same key value may be used by different threads, the values bound
+ * to the key by lthread_setspecific() are maintained on a per-thread
+ * basis and persist for the life of the calling thread.
+ *
+ * An optional destructor function may be associated with each key value.
+ * At thread exit, if a key value has a non-NULL destructor pointer, and the
+ * thread has a non-NULL value associated with the key, the function pointed
+ * to is called with the current associated value as its sole argument.
+ *
+ * @param key
+ * Pointer to the key to be created
+ * @param destructor
+ * Pointer to destructor function
+ *
+ * @return
+ * 0 success
+ * EINVAL the key ptr was NULL
+ * EAGAIN no resources available
+ */
+int lthread_key_create(unsigned int *key, tls_destructor_func destructor);
+
+/**
+ * Delete key for lthread TLS
+ *
+ * This function is modelled on pthread_key_delete().
+ * It deletes a thread-specific data key previously returned by
+ * lthread_key_create().
+ * The thread-specific data values associated with the key need not be NULL
+ * at the time that lthread_key_delete is called.
+ * It is the responsibility of the application to free any application
+ * storage or perform any cleanup actions for data structures related to the
+ * deleted key. This cleanup can be done either before or after
+ * lthread_key_delete is called.
+ *
+ * @param key
+ * The key to be deleted
+ *
+ * @return
+ * 0 Success
+ * EINVAL the key was invalid
+ */
+int lthread_key_delete(unsigned int key);
+
+/**
+ * Get lthread TLS
+ *
+ * This function is modelled on pthread_get_specific().
+ * It returns the value currently bound to the specified key on behalf of the
+ * calling thread. Calling lthread_getspecific() with a key value not
+ * obtained from lthread_key_create() or after key has been deleted with
+ * lthread_key_delete() will result in undefined behaviour.
+ * lthread_getspecific() may be called from a thread-specific data destructor
+ * function.
+ *
+ * @param key
+ * The key for which data is requested
+ *
+ * @return
+ * Pointer to the thread specific data associated with that key
+ * or NULL if no data has been set.
+ */
+void
+*lthread_getspecific(unsigned int key);
+
+/**
+ * Set lthread TLS
+ *
+ * This function is modelled on pthread_set_specific()
+ * It associates a thread-specific value with a key obtained via a previous
+ * call to lthread_key_create().
+ * Different threads may bind different values to the same key. These values
+ * are typically pointers to dynamically allocated memory that have been
+ * reserved by the calling thread. Calling lthread_setspecific with a key
+ * value not obtained from lthread_key_create or after the key has been
+ * deleted with lthread_key_delete will result in undefined behaviour.
+ *
+ * @param key
+ * The key for which data is to be set
+ * @param key
+ * Pointer to the user data
+ *
+ * @return
+ * 0 success
+ * EINVAL the key was invalid
+ */
+
+int lthread_setspecific(unsigned int key, const void *value);
+
+/**
+ * The macros below provide an alternative mechanism to access lthread local
+ * storage.
+ *
+ * The macros can be used to declare define and access per lthread local
+ * storage in a similar way to the RTE_PER_LCORE macros which control storage
+ * local to an lcore.
+ *
+ * Memory for per lthread variables declared in this way is allocated when the
+ * lthread is created and a pointer to this memory is stored in the lthread.
+ * The per lthread variables are accessed via the pointer + the offset of the
+ * particular variable.
+ *
+ * The total size of per lthread storage, and the variable offsets are found by
+ * defining the variables in a unique global memory section, the start and end
+ * of which is known. This global memory section is used only in the
+ * computation of the addresses of the lthread variables, and is never actually
+ * used to store any data.
+ *
+ * Due to the fact that variables declared this way may be scattered across
+ * many files, the start and end of the section and variable offsets are only
+ * known after linking, thus the computation of section size and variable
+ * addresses is performed at run time.
+ *
+ * These macros are primarily provided to aid porting of code that makes use
+ * of the existing RTE_PER_LCORE macros. In principle it would be more efficient
+ * to gather all lthread local variables into a single structure and
+ * set/retrieve a pointer to that struct using the alternative
+ * lthread_data_set/get APIs.
+ *
+ * These macros are mutually exclusive with the lthread_data_set/get APIs.
+ * If you define storage using these macros then the lthread_data_set/get APIs
+ * will not perform as expected, the lthread_data_set API does nothing, and the
+ * lthread_data_get API returns the start of global section.
+ *
+ */
+/* start and end of per lthread section */
+extern char __start_per_lt;
+extern char __stop_per_lt;
+
+
+#define RTE_DEFINE_PER_LTHREAD(type, name) \
+__typeof__(type)__attribute((section("per_lt"))) per_lt_##name
+
+/**
+ * Macro to declare an extern per lthread variable "var" of type "type"
+ */
+#define RTE_DECLARE_PER_LTHREAD(type, name) \
+extern __typeof__(type)__attribute((section("per_lt"))) per_lt_##name
+
+/**
+ * Read/write the per-lcore variable value
+ */
+#define RTE_PER_LTHREAD(name) ((typeof(per_lt_##name) *)\
+((char *)lthread_get_data() +\
+((char *) &per_lt_##name - &__start_per_lt)))
+
+/**
+ * Initialize a mutex
+ *
+ * This function provides a mutual exclusion device, the need for which
+ * can normally be avoided in a cooperative multitasking environment.
+ * It is provided to aid porting of legacy code originally written for
+ * preemptive multitasking environments such as pthreads.
+ *
+ * A mutex may be unlocked (not owned by any thread), or locked (owned by
+ * one thread).
+ *
+ * A mutex can never be owned by more than one thread simultaneously.
+ * A thread attempting to lock a mutex that is already locked by another
+ * thread is suspended until the owning thread unlocks the mutex.
+ *
+ * lthread_mutex_init() initializes the mutex object pointed to by mutex
+ * Optional mutex attributes specified in mutexattr, are reserved for future
+ * use and are currently ignored.
+ *
+ * If a thread calls lthread_mutex_lock() on the mutex, then if the mutex
+ * is currently unlocked, it becomes locked and owned by the calling
+ * thread, and lthread_mutex_lock returns immediately. If the mutex is
+ * already locked by another thread, lthread_mutex_lock suspends the calling
+ * thread until the mutex is unlocked.
+ *
+ * lthread_mutex_trylock behaves identically to rte_thread_mutex_lock, except
+ * that it does not block the calling thread if the mutex is already locked
+ * by another thread.
+ *
+ * lthread_mutex_unlock() unlocks the specified mutex. The mutex is assumed
+ * to be locked and owned by the calling thread.
+ *
+ * lthread_mutex_destroy() destroys a mutex object, freeing its resources.
+ * The mutex must be unlocked with nothing blocked on it before calling
+ * lthread_mutex_destroy.
+ *
+ * @param name
+ * Optional pointer to string describing the mutex
+ * @param mutex
+ * Pointer to pointer to the mutex to be initialized
+ * @param attribute
+ * Pointer to attribute - unused reserved
+ *
+ * @return
+ * 0 success
+ * EINVAL mutex was not a valid pointer
+ * EAGAIN insufficient resources
+ */
+
+int
+lthread_mutex_init(char *name, struct lthread_mutex **mutex,
+ const struct lthread_mutexattr *attr);
+
+/**
+ * Destroy a mutex
+ *
+ * This function destroys the specified mutex freeing its resources.
+ * The mutex must be unlocked before calling lthread_mutex_destroy.
+ *
+ * @see lthread_mutex_init()
+ *
+ * @param mutex
+ * Pointer to pointer to the mutex to be initialized
+ *
+ * @return
+ * 0 success
+ * EINVAL mutex was not an initialized mutex
+ * EBUSY mutex was still in use
+ */
+int lthread_mutex_destroy(struct lthread_mutex *mutex);
+
+/**
+ * Lock a mutex
+ *
+ * This function attempts to lock a mutex.
+ * If a thread calls lthread_mutex_lock() on the mutex, then if the mutex
+ * is currently unlocked, it becomes locked and owned by the calling
+ * thread, and lthread_mutex_lock returns immediately. If the mutex is
+ * already locked by another thread, lthread_mutex_lock suspends the calling
+ * thread until the mutex is unlocked.
+ *
+ * @see lthread_mutex_init()
+ *
+ * @param mutex
+ * Pointer to pointer to the mutex to be initialized
+ *
+ * @return
+ * 0 success
+ * EINVAL mutex was not an initialized mutex
+ * EDEADLOCK the mutex was already owned by the calling thread
+ */
+
+int lthread_mutex_lock(struct lthread_mutex *mutex);
+
+/**
+ * Try to lock a mutex
+ *
+ * This function attempts to lock a mutex.
+ * lthread_mutex_trylock behaves identically to rte_thread_mutex_lock, except
+ * that it does not block the calling thread if the mutex is already locked
+ * by another thread.
+ *
+ *
+ * @see lthread_mutex_init()
+ *
+ * @param mutex
+ * Pointer to pointer to the mutex to be initialized
+ *
+ * @return
+ * 0 success
+ * EINVAL mutex was not an initialized mutex
+ * EBUSY the mutex was already locked by another thread
+ */
+int lthread_mutex_trylock(struct lthread_mutex *mutex);
+
+/**
+ * Unlock a mutex
+ *
+ * This function attempts to unlock the specified mutex. The mutex is assumed
+ * to be locked and owned by the calling thread.
+ *
+ * The oldest of any threads blocked on the mutex is made ready and may
+ * compete with any other running thread to gain the mutex, it fails it will
+ * be blocked again.
+ *
+ * @param mutex
+ * Pointer to pointer to the mutex to be initialized
+ *
+ * @return
+ * 0 mutex was unlocked
+ * EINVAL mutex was not an initialized mutex
+ * EPERM the mutex was not owned by the calling thread
+ */
+
+int lthread_mutex_unlock(struct lthread_mutex *mutex);
+
+/**
+ * Initialize a condition variable
+ *
+ * This function initializes a condition variable.
+ *
+ * Condition variables can be used to communicate changes in the state of data
+ * shared between threads.
+ *
+ * @see lthread_cond_wait()
+ *
+ * @param name
+ * Pointer to optional string describing the condition variable
+ * @param c
+ * Pointer to pointer to the condition variable to be initialized
+ * @param attr
+ * Pointer to optional attribute reserved for future use, currently ignored
+ *
+ * @return
+ * 0 success
+ * EINVAL cond was not a valid pointer
+ * EAGAIN insufficient resources
+ */
+int
+lthread_cond_init(char *name, struct lthread_cond **c,
+ const struct lthread_condattr *attr);
+
+/**
+ * Destroy a condition variable
+ *
+ * This function destroys a condition variable that was created with
+ * lthread_cond_init() and releases its resources.
+ *
+ * @param cond
+ * Pointer to pointer to the condition variable to be destroyed
+ *
+ * @return
+ * 0 Success
+ * EBUSY condition variable was still in use
+ * EINVAL was not an initialised condition variable
+ */
+int lthread_cond_destroy(struct lthread_cond *cond);
+
+/**
+ * Wait on a condition variable
+ *
+ * The function blocks the current thread waiting on the condition variable
+ * specified by cond. The waiting thread unblocks only after another thread
+ * calls lthread_cond_signal, or lthread_cond_broadcast, specifying the
+ * same condition variable.
+ *
+ * @param cond
+ * Pointer to pointer to the condition variable to be waited on
+ *
+ * @param reserved
+ * reserved for future use
+ *
+ * @return
+ * 0 The condition was signalled ( Success )
+ * EINVAL was not a an initialised condition variable
+ */
+int lthread_cond_wait(struct lthread_cond *c, uint64_t reserved);
+
+/**
+ * Signal a condition variable
+ *
+ * The function unblocks one thread waiting for the condition variable cond.
+ * If no threads are waiting on cond, the rte_lthread_cond_signal() function
+ * has no effect.
+ *
+ * @param cond
+ * Pointer to pointer to the condition variable to be signalled
+ *
+ * @return
+ * 0 The condition was signalled ( Success )
+ * EINVAL was not a an initialised condition variable
+ */
+int lthread_cond_signal(struct lthread_cond *c);
+
+/**
+ * Broadcast a condition variable
+ *
+ * The function unblocks all threads waiting for the condition variable cond.
+ * If no threads are waiting on cond, the rte_lathed_cond_broadcast()
+ * function has no effect.
+ *
+ * @param cond
+ * Pointer to pointer to the condition variable to be signalled
+ *
+ * @return
+ * 0 The condition was signalled ( Success )
+ * EINVAL was not a an initialised condition variable
+ */
+int lthread_cond_broadcast(struct lthread_cond *c);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* LTHREAD_H */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_cond.c b/src/spdk/dpdk/examples/performance-thread/common/lthread_cond.c
new file mode 100644
index 000000000..cdcc7a7b5
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_cond.c
@@ -0,0 +1,184 @@
+/*
+ * SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2015 Intel Corporation.
+ * Copyright 2012 Hasan Alayli <halayli@gmail.com>
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdint.h>
+#include <stddef.h>
+#include <limits.h>
+#include <inttypes.h>
+#include <unistd.h>
+#include <pthread.h>
+#include <fcntl.h>
+#include <sys/time.h>
+#include <sys/mman.h>
+#include <errno.h>
+
+#include <rte_log.h>
+#include <rte_common.h>
+
+#include "lthread_api.h"
+#include "lthread_diag_api.h"
+#include "lthread_diag.h"
+#include "lthread_int.h"
+#include "lthread_sched.h"
+#include "lthread_queue.h"
+#include "lthread_objcache.h"
+#include "lthread_timer.h"
+#include "lthread_mutex.h"
+#include "lthread_cond.h"
+
+/*
+ * Create a condition variable
+ */
+int
+lthread_cond_init(char *name, struct lthread_cond **cond,
+ __rte_unused const struct lthread_condattr *attr)
+{
+ struct lthread_cond *c;
+
+ if (cond == NULL)
+ return POSIX_ERRNO(EINVAL);
+
+ /* allocate a condition variable from cache */
+ c = _lthread_objcache_alloc((THIS_SCHED)->cond_cache);
+
+ if (c == NULL)
+ return POSIX_ERRNO(EAGAIN);
+
+ c->blocked = _lthread_queue_create("blocked");
+ if (c->blocked == NULL) {
+ _lthread_objcache_free((THIS_SCHED)->cond_cache, (void *)c);
+ return POSIX_ERRNO(EAGAIN);
+ }
+
+ if (name == NULL)
+ strncpy(c->name, "no name", sizeof(c->name));
+ else
+ strncpy(c->name, name, sizeof(c->name));
+ c->name[sizeof(c->name)-1] = 0;
+
+ c->root_sched = THIS_SCHED;
+
+ (*cond) = c;
+ DIAG_CREATE_EVENT((*cond), LT_DIAG_COND_CREATE);
+ return 0;
+}
+
+/*
+ * Destroy a condition variable
+ */
+int lthread_cond_destroy(struct lthread_cond *c)
+{
+ if (c == NULL) {
+ DIAG_EVENT(c, LT_DIAG_COND_DESTROY, c, POSIX_ERRNO(EINVAL));
+ return POSIX_ERRNO(EINVAL);
+ }
+
+ /* try to free it */
+ if (_lthread_queue_destroy(c->blocked) < 0) {
+ /* queue in use */
+ DIAG_EVENT(c, LT_DIAG_COND_DESTROY, c, POSIX_ERRNO(EBUSY));
+ return POSIX_ERRNO(EBUSY);
+ }
+
+ /* okay free it */
+ _lthread_objcache_free(c->root_sched->cond_cache, c);
+ DIAG_EVENT(c, LT_DIAG_COND_DESTROY, c, 0);
+ return 0;
+}
+
+/*
+ * Wait on a condition variable
+ */
+int lthread_cond_wait(struct lthread_cond *c, __rte_unused uint64_t reserved)
+{
+ struct lthread *lt = THIS_LTHREAD;
+
+ if (c == NULL) {
+ DIAG_EVENT(c, LT_DIAG_COND_WAIT, c, POSIX_ERRNO(EINVAL));
+ return POSIX_ERRNO(EINVAL);
+ }
+
+
+ DIAG_EVENT(c, LT_DIAG_COND_WAIT, c, 0);
+
+ /* queue the current thread in the blocked queue
+ * this will be written when we return to the scheduler
+ * to ensure that the current thread context is saved
+ * before any signal could result in it being dequeued and
+ * resumed
+ */
+ lt->pending_wr_queue = c->blocked;
+ _suspend();
+
+ /* the condition happened */
+ return 0;
+}
+
+/*
+ * Signal a condition variable
+ * attempt to resume any blocked thread
+ */
+int lthread_cond_signal(struct lthread_cond *c)
+{
+ struct lthread *lt;
+
+ if (c == NULL) {
+ DIAG_EVENT(c, LT_DIAG_COND_SIGNAL, c, POSIX_ERRNO(EINVAL));
+ return POSIX_ERRNO(EINVAL);
+ }
+
+ lt = _lthread_queue_remove(c->blocked);
+
+ if (lt != NULL) {
+ /* okay wake up this thread */
+ DIAG_EVENT(c, LT_DIAG_COND_SIGNAL, c, lt);
+ _ready_queue_insert((struct lthread_sched *)lt->sched, lt);
+ }
+ return 0;
+}
+
+/*
+ * Broadcast a condition variable
+ */
+int lthread_cond_broadcast(struct lthread_cond *c)
+{
+ struct lthread *lt;
+
+ if (c == NULL) {
+ DIAG_EVENT(c, LT_DIAG_COND_BROADCAST, c, POSIX_ERRNO(EINVAL));
+ return POSIX_ERRNO(EINVAL);
+ }
+
+ DIAG_EVENT(c, LT_DIAG_COND_BROADCAST, c, 0);
+ do {
+ /* drain the queue waking everybody */
+ lt = _lthread_queue_remove(c->blocked);
+
+ if (lt != NULL) {
+ DIAG_EVENT(c, LT_DIAG_COND_BROADCAST, c, lt);
+ /* wake up */
+ _ready_queue_insert((struct lthread_sched *)lt->sched,
+ lt);
+ }
+ } while (!_lthread_queue_empty(c->blocked));
+ _reschedule();
+ DIAG_EVENT(c, LT_DIAG_COND_BROADCAST, c, 0);
+ return 0;
+}
+
+/*
+ * return the diagnostic ref val stored in a condition var
+ */
+uint64_t
+lthread_cond_diag_ref(struct lthread_cond *c)
+{
+ if (c == NULL)
+ return 0;
+ return c->diag_ref;
+}
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_cond.h b/src/spdk/dpdk/examples/performance-thread/common/lthread_cond.h
new file mode 100644
index 000000000..616a55c4d
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_cond.h
@@ -0,0 +1,30 @@
+/*
+ * SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2015 Intel Corporation.
+ * Copyright 2012 Hasan Alayli <halayli@gmail.com>
+ */
+
+#ifndef LTHREAD_COND_H_
+#define LTHREAD_COND_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "lthread_queue.h"
+
+#define MAX_COND_NAME_SIZE 64
+
+struct lthread_cond {
+ struct lthread_queue *blocked;
+ struct lthread_sched *root_sched;
+ int count;
+ char name[MAX_COND_NAME_SIZE];
+ uint64_t diag_ref; /* optional ref to user diag data */
+} __rte_cache_aligned;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* LTHREAD_COND_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_diag.c b/src/spdk/dpdk/examples/performance-thread/common/lthread_diag.c
new file mode 100644
index 000000000..57760a1e2
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_diag.c
@@ -0,0 +1,293 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2015 Intel Corporation
+ */
+
+#include <rte_log.h>
+#include <rte_common.h>
+
+#include "lthread_diag.h"
+#include "lthread_queue.h"
+#include "lthread_pool.h"
+#include "lthread_objcache.h"
+#include "lthread_sched.h"
+#include "lthread_diag_api.h"
+
+
+/* dummy ref value of default diagnostic callback */
+static uint64_t dummy_ref;
+
+#define DIAG_SCHED_STATS_FORMAT \
+"core %d\n%33s %12s %12s %12s %12s\n"
+
+#define DIAG_CACHE_STATS_FORMAT \
+"%20s %12lu %12lu %12lu %12lu %12lu\n"
+
+#define DIAG_QUEUE_STATS_FORMAT \
+"%20s %12lu %12lu %12lu\n"
+
+
+/*
+ * texts used in diagnostic events,
+ * corresponding diagnostic mask bit positions are given as comment
+ */
+const char *diag_event_text[] = {
+ "LTHREAD_CREATE ", /* 00 */
+ "LTHREAD_EXIT ", /* 01 */
+ "LTHREAD_JOIN ", /* 02 */
+ "LTHREAD_CANCEL ", /* 03 */
+ "LTHREAD_DETACH ", /* 04 */
+ "LTHREAD_FREE ", /* 05 */
+ "LTHREAD_SUSPENDED ", /* 06 */
+ "LTHREAD_YIELD ", /* 07 */
+ "LTHREAD_RESCHEDULED", /* 08 */
+ "LTHREAD_SLEEP ", /* 09 */
+ "LTHREAD_RESUMED ", /* 10 */
+ "LTHREAD_AFFINITY ", /* 11 */
+ "LTHREAD_TMR_START ", /* 12 */
+ "LTHREAD_TMR_DELETE ", /* 13 */
+ "LTHREAD_TMR_EXPIRED", /* 14 */
+ "COND_CREATE ", /* 15 */
+ "COND_DESTROY ", /* 16 */
+ "COND_WAIT ", /* 17 */
+ "COND_SIGNAL ", /* 18 */
+ "COND_BROADCAST ", /* 19 */
+ "MUTEX_CREATE ", /* 20 */
+ "MUTEX_DESTROY ", /* 21 */
+ "MUTEX_LOCK ", /* 22 */
+ "MUTEX_TRYLOCK ", /* 23 */
+ "MUTEX_BLOCKED ", /* 24 */
+ "MUTEX_UNLOCKED ", /* 25 */
+ "SCHED_CREATE ", /* 26 */
+ "SCHED_SHUTDOWN " /* 27 */
+};
+
+
+/*
+ * set diagnostic ,ask
+ */
+void lthread_diagnostic_set_mask(DIAG_USED uint64_t mask)
+{
+#if LTHREAD_DIAG
+ diag_mask = mask;
+#else
+ RTE_LOG(INFO, LTHREAD,
+ "LTHREAD_DIAG is not set, see lthread_diag_api.h\n");
+#endif
+}
+
+
+/*
+ * Check consistency of the scheduler stats
+ * Only sensible run after the schedulers are stopped
+ * Count the number of objects lying in caches and queues
+ * and available in the qnode pool.
+ * This should be equal to the total capacity of all
+ * qnode pools.
+ */
+void
+_sched_stats_consistency_check(void);
+void
+_sched_stats_consistency_check(void)
+{
+#if LTHREAD_DIAG
+ int i;
+ struct lthread_sched *sched;
+ uint64_t count = 0;
+ uint64_t capacity = 0;
+
+ for (i = 0; i < LTHREAD_MAX_LCORES; i++) {
+ sched = schedcore[i];
+ if (sched == NULL)
+ continue;
+
+ /* each of these queues consumes a stub node */
+ count += 8;
+ count += DIAG_COUNT(sched->ready, size);
+ count += DIAG_COUNT(sched->pready, size);
+ count += DIAG_COUNT(sched->lthread_cache, available);
+ count += DIAG_COUNT(sched->stack_cache, available);
+ count += DIAG_COUNT(sched->tls_cache, available);
+ count += DIAG_COUNT(sched->per_lthread_cache, available);
+ count += DIAG_COUNT(sched->cond_cache, available);
+ count += DIAG_COUNT(sched->mutex_cache, available);
+
+ /* the node pool does not consume a stub node */
+ if (sched->qnode_pool->fast_alloc != NULL)
+ count++;
+ count += DIAG_COUNT(sched->qnode_pool, available);
+
+ capacity += DIAG_COUNT(sched->qnode_pool, capacity);
+ }
+ if (count != capacity) {
+ RTE_LOG(CRIT, LTHREAD,
+ "Scheduler caches are inconsistent\n");
+ } else {
+ RTE_LOG(INFO, LTHREAD,
+ "Scheduler caches are ok\n");
+ }
+#endif
+}
+
+
+#if LTHREAD_DIAG
+/*
+ * Display node pool stats
+ */
+static inline void
+_qnode_pool_display(DIAG_USED struct qnode_pool *p)
+{
+
+ printf(DIAG_CACHE_STATS_FORMAT,
+ p->name,
+ DIAG_COUNT(p, rd),
+ DIAG_COUNT(p, wr),
+ DIAG_COUNT(p, available),
+ DIAG_COUNT(p, prealloc),
+ DIAG_COUNT(p, capacity));
+ fflush(stdout);
+}
+#endif
+
+
+#if LTHREAD_DIAG
+/*
+ * Display queue stats
+ */
+static inline void
+_lthread_queue_display(DIAG_USED struct lthread_queue *q)
+{
+#if DISPLAY_OBJCACHE_QUEUES
+ printf(DIAG_QUEUE_STATS_FORMAT,
+ q->name,
+ DIAG_COUNT(q, rd),
+ DIAG_COUNT(q, wr),
+ DIAG_COUNT(q, size));
+ fflush(stdout);
+#else
+ printf("%s: queue stats disabled\n",
+ q->name);
+
+#endif
+}
+#endif
+
+#if LTHREAD_DIAG
+/*
+ * Display objcache stats
+ */
+static inline void
+_objcache_display(DIAG_USED struct lthread_objcache *c)
+{
+
+ printf(DIAG_CACHE_STATS_FORMAT,
+ c->name,
+ DIAG_COUNT(c, rd),
+ DIAG_COUNT(c, wr),
+ DIAG_COUNT(c, available),
+ DIAG_COUNT(c, prealloc),
+ DIAG_COUNT(c, capacity));
+ _lthread_queue_display(c->q);
+ fflush(stdout);
+}
+#endif
+
+/*
+ * Display sched stats
+ */
+void
+lthread_sched_stats_display(void)
+{
+#if LTHREAD_DIAG
+ int i;
+ struct lthread_sched *sched;
+
+ for (i = 0; i < LTHREAD_MAX_LCORES; i++) {
+ sched = schedcore[i];
+ if (sched != NULL) {
+ printf(DIAG_SCHED_STATS_FORMAT,
+ sched->lcore_id,
+ "rd",
+ "wr",
+ "present",
+ "nb preallocs",
+ "capacity");
+ _lthread_queue_display(sched->ready);
+ _lthread_queue_display(sched->pready);
+ _qnode_pool_display(sched->qnode_pool);
+ _objcache_display(sched->lthread_cache);
+ _objcache_display(sched->stack_cache);
+ _objcache_display(sched->tls_cache);
+ _objcache_display(sched->per_lthread_cache);
+ _objcache_display(sched->cond_cache);
+ _objcache_display(sched->mutex_cache);
+ fflush(stdout);
+ }
+ }
+ _sched_stats_consistency_check();
+#else
+ RTE_LOG(INFO, LTHREAD,
+ "lthread diagnostics disabled\n"
+ "hint - set LTHREAD_DIAG in lthread_diag_api.h\n");
+#endif
+}
+
+/*
+ * Defafult diagnostic callback
+ */
+static uint64_t
+_lthread_diag_default_cb(uint64_t time, struct lthread *lt, int diag_event,
+ uint64_t diag_ref, const char *text, uint64_t p1, uint64_t p2)
+{
+ uint64_t _p2;
+ int lcore = (int) rte_lcore_id();
+
+ switch (diag_event) {
+ case LT_DIAG_LTHREAD_CREATE:
+ case LT_DIAG_MUTEX_CREATE:
+ case LT_DIAG_COND_CREATE:
+ _p2 = dummy_ref;
+ break;
+ default:
+ _p2 = p2;
+ break;
+ }
+
+ printf("%"PRIu64" %d %8.8lx %8.8lx %s %8.8lx %8.8lx\n",
+ time,
+ lcore,
+ (uint64_t) lt,
+ diag_ref,
+ text,
+ p1,
+ _p2);
+
+ return dummy_ref++;
+}
+
+/*
+ * plug in default diag callback with mask off
+ */
+RTE_INIT(_lthread_diag_ctor)
+{
+ diag_cb = _lthread_diag_default_cb;
+ diag_mask = 0;
+}
+
+
+/*
+ * enable diagnostics
+ */
+void lthread_diagnostic_enable(DIAG_USED diag_callback cb,
+ DIAG_USED uint64_t mask)
+{
+#if LTHREAD_DIAG
+ if (cb == NULL)
+ diag_cb = _lthread_diag_default_cb;
+ else
+ diag_cb = cb;
+ diag_mask = mask;
+#else
+ RTE_LOG(INFO, LTHREAD,
+ "LTHREAD_DIAG is not set, see lthread_diag_api.h\n");
+#endif
+}
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_diag.h b/src/spdk/dpdk/examples/performance-thread/common/lthread_diag.h
new file mode 100644
index 000000000..e876dda6d
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_diag.h
@@ -0,0 +1,112 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2015 Intel Corporation
+ */
+
+#ifndef LTHREAD_DIAG_H_
+#define LTHREAD_DIAG_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stdint.h>
+#include <inttypes.h>
+
+#include <rte_log.h>
+#include <rte_common.h>
+
+#include "lthread_api.h"
+#include "lthread_diag_api.h"
+
+extern diag_callback diag_cb;
+
+extern const char *diag_event_text[];
+extern uint64_t diag_mask;
+
+/* max size of name strings */
+#define LT_MAX_NAME_SIZE 64
+
+#if LTHREAD_DIAG
+#define DISPLAY_OBJCACHE_QUEUES 1
+
+/*
+ * Generate a diagnostic trace or event in the case where an object is created.
+ *
+ * The value returned by the callback is stored in the object.
+ *
+ * @ param obj
+ * pointer to the object that was created
+ * @ param ev
+ * the event code
+ *
+ */
+#define DIAG_CREATE_EVENT(obj, ev) do { \
+ struct lthread *ct = RTE_PER_LCORE(this_sched)->current_lthread;\
+ if ((BIT(ev) & diag_mask) && (ev < LT_DIAG_EVENT_MAX)) { \
+ (obj)->diag_ref = (diag_cb)(rte_rdtsc(), \
+ ct, \
+ (ev), \
+ 0, \
+ diag_event_text[(ev)], \
+ (uint64_t)obj, \
+ 0); \
+ } \
+} while (0)
+
+/*
+ * Generate a diagnostic trace event.
+ *
+ * @ param obj
+ * pointer to the lthread, cond or mutex object
+ * @ param ev
+ * the event code
+ * @ param p1
+ * object specific value ( see lthread_diag_api.h )
+ * @ param p2
+ * object specific value ( see lthread_diag_api.h )
+ */
+#define DIAG_EVENT(obj, ev, p1, p2) do { \
+ struct lthread *ct = RTE_PER_LCORE(this_sched)->current_lthread;\
+ if ((BIT(ev) & diag_mask) && (ev < LT_DIAG_EVENT_MAX)) { \
+ (diag_cb)(rte_rdtsc(), \
+ ct, \
+ ev, \
+ (obj)->diag_ref, \
+ diag_event_text[(ev)], \
+ (uint64_t)(p1), \
+ (uint64_t)(p2)); \
+ } \
+} while (0)
+
+#define DIAG_COUNT_DEFINE(x) rte_atomic64_t count_##x
+#define DIAG_COUNT_INIT(o, x) rte_atomic64_init(&((o)->count_##x))
+#define DIAG_COUNT_INC(o, x) rte_atomic64_inc(&((o)->count_##x))
+#define DIAG_COUNT_DEC(o, x) rte_atomic64_dec(&((o)->count_##x))
+#define DIAG_COUNT(o, x) rte_atomic64_read(&((o)->count_##x))
+
+#define DIAG_USED
+
+#else
+
+/* no diagnostics configured */
+
+#define DISPLAY_OBJCACHE_QUEUES 0
+
+#define DIAG_CREATE_EVENT(obj, ev)
+#define DIAG_EVENT(obj, ev, p1, p)
+
+#define DIAG_COUNT_DEFINE(x)
+#define DIAG_COUNT_INIT(o, x) do {} while (0)
+#define DIAG_COUNT_INC(o, x) do {} while (0)
+#define DIAG_COUNT_DEC(o, x) do {} while (0)
+#define DIAG_COUNT(o, x) 0
+
+#define DIAG_USED __rte_unused
+
+#endif /* LTHREAD_DIAG */
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* LTHREAD_DIAG_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_diag_api.h b/src/spdk/dpdk/examples/performance-thread/common/lthread_diag_api.h
new file mode 100644
index 000000000..d65f486ec
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_diag_api.h
@@ -0,0 +1,304 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2015 Intel Corporation
+ */
+#ifndef LTHREAD_DIAG_API_H_
+#define LTHREAD_DIAG_API_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stdint.h>
+#include <inttypes.h>
+
+/*
+ * Enable diagnostics
+ * 0 = conditionally compiled out
+ * 1 = compiled in and maskable at run time, see below for details
+ */
+#define LTHREAD_DIAG 0
+
+/**
+ *
+ * @file lthread_diag_api.h
+ *
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice
+ *
+ * lthread diagnostic interface
+ *
+ * If enabled via configuration file option ( tbd ) the lthread subsystem
+ * can generate selected trace information, either RTE_LOG (INFO) messages,
+ * or else invoke a user supplied callback function when any of the events
+ * listed below occur.
+ *
+ * Reporting of events can be selectively masked, the bit position in the
+ * mask is determined by the corresponding event identifier listed below.
+ *
+ * Diagnostics are enabled by registering the callback function and mask
+ * using the API lthread_diagnostic_enable().
+ *
+ * Various interesting parameters are passed to the callback, including the
+ * time in cpu clks, the lthread id, the diagnostic event id, a user ref value,
+ * event text string, object being traced, and two context dependent parameters
+ * (p1 and p2). The meaning of the two parameters p1 and p2 depends on
+ * the specific event.
+ *
+ * The events LT_DIAG_LTHREAD_CREATE, LT_DIAG_MUTEX_CREATE and
+ * LT_DIAG_COND_CREATE are implicitly enabled if the event mask includes any of
+ * the LT_DIAG_LTHREAD_XXX, LT_DIAG_MUTEX_XXX or LT_DIAG_COND_XXX events
+ * respectively.
+ *
+ * These create events may also be included in the mask discreetly if it is
+ * desired to monitor only create events.
+ *
+ * @param time
+ * The time in cpu clks at which the event occurred
+ *
+ * @param lthread
+ * The current lthread
+ *
+ * @param diag_event
+ * The diagnostic event id (bit position in the mask)
+ *
+ * @param diag_ref
+ *
+ * For LT_DIAG_LTHREAD_CREATE, LT_DIAG_MUTEX_CREATE or LT_DIAG_COND_CREATE
+ * this parameter is not used and set to 0.
+ * All other events diag_ref contains the user ref value returned by the
+ * callback function when lthread is created.
+ *
+ * The diag_ref values assigned to mutex and cond var can be retrieved
+ * using the APIs lthread_mutex_diag_ref(), and lthread_cond_diag_ref()
+ * respectively.
+ *
+ * @param p1
+ * see below
+ *
+ * @param p1
+ * see below
+ *
+ * @returns
+ * For LT_DIAG_LTHREAD_CREATE, LT_DIAG_MUTEX_CREATE or LT_DIAG_COND_CREATE
+ * expects a user diagnostic ref value that will be saved in the lthread, mutex
+ * or cond var.
+ *
+ * For all other events return value is ignored.
+ *
+ * LT_DIAG_SCHED_CREATE - Invoked when a scheduler is created
+ * p1 = the scheduler that was created
+ * p2 = not used
+ * return value will be ignored
+ *
+ * LT_DIAG_SCHED_SHUTDOWN - Invoked when a shutdown request is received
+ * p1 = the scheduler to be shutdown
+ * p2 = not used
+ * return value will be ignored
+ *
+ * LT_DIAG_LTHREAD_CREATE - Invoked when a thread is created
+ * p1 = the lthread that was created
+ * p2 = not used
+ * return value will be stored in the lthread
+ *
+ * LT_DIAG_LTHREAD_EXIT - Invoked when a lthread exits
+ * p2 = 0 if the thread was already joined
+ * p2 = 1 if the thread was not already joined
+ * return val ignored
+ *
+ * LT_DIAG_LTHREAD_JOIN - Invoked when a lthread exits
+ * p1 = the lthread that is being joined
+ * p2 = 0 if the thread was already exited
+ * p2 = 1 if the thread was not already exited
+ * return val ignored
+ *
+ * LT_DIAG_LTHREAD_CANCELLED - Invoked when an lthread is cancelled
+ * p1 = not used
+ * p2 = not used
+ * return val ignored
+ *
+ * LT_DIAG_LTHREAD_DETACH - Invoked when an lthread is detached
+ * p1 = not used
+ * p2 = not used
+ * return val ignored
+ *
+ * LT_DIAG_LTHREAD_FREE - Invoked when an lthread is freed
+ * p1 = not used
+ * p2 = not used
+ * return val ignored
+ *
+ * LT_DIAG_LTHREAD_SUSPENDED - Invoked when an lthread is suspended
+ * p1 = not used
+ * p2 = not used
+ * return val ignored
+ *
+ * LT_DIAG_LTHREAD_YIELD - Invoked when an lthread explicitly yields
+ * p1 = not used
+ * p2 = not used
+ * return val ignored
+ *
+ * LT_DIAG_LTHREAD_RESCHEDULED - Invoked when an lthread is rescheduled
+ * p1 = not used
+ * p2 = not used
+ * return val ignored
+ *
+ * LT_DIAG_LTHREAD_RESUMED - Invoked when an lthread is resumed
+ * p1 = not used
+ * p2 = not used
+ * return val ignored
+ *
+ * LT_DIAG_LTHREAD_AFFINITY - Invoked when an lthread is affinitised
+ * p1 = the destination lcore_id
+ * p2 = not used
+ * return val ignored
+ *
+ * LT_DIAG_LTHREAD_TMR_START - Invoked when an lthread starts a timer
+ * p1 = address of timer node
+ * p2 = the timeout value
+ * return val ignored
+ *
+ * LT_DIAG_LTHREAD_TMR_DELETE - Invoked when an lthread deletes a timer
+ * p1 = address of the timer node
+ * p2 = 0 the timer and the was successfully deleted
+ * p2 = not usee
+ * return val ignored
+ *
+ * LT_DIAG_LTHREAD_TMR_EXPIRED - Invoked when an lthread timer expires
+ * p1 = address of scheduler the timer expired on
+ * p2 = the thread associated with the timer
+ * return val ignored
+ *
+ * LT_DIAG_COND_CREATE - Invoked when a condition variable is created
+ * p1 = address of cond var that was created
+ * p2 = not used
+ * return diag ref value will be stored in the condition variable
+ *
+ * LT_DIAG_COND_DESTROY - Invoked when a condition variable is destroyed
+ * p1 = not used
+ * p2 = not used
+ * return val ignored
+ *
+ * LT_DIAG_COND_WAIT - Invoked when an lthread waits on a cond var
+ * p1 = the address of the condition variable
+ * p2 = not used
+ * return val ignored
+ *
+ * LT_DIAG_COND_SIGNAL - Invoked when an lthread signals a cond var
+ * p1 = the address of the cond var
+ * p2 = the lthread that was signalled, or error code
+ * return val ignored
+ *
+ * LT_DIAG_COND_BROADCAST - Invoked when an lthread broadcasts a cond var
+ * p1 = the address of the condition variable
+ * p2 = the lthread(s) that are signalled, or error code
+ *
+ * LT_DIAG_MUTEX_CREATE - Invoked when a mutex is created
+ * p1 = address of muex
+ * p2 = not used
+ * return diag ref value will be stored in the mutex variable
+ *
+ * LT_DIAG_MUTEX_DESTROY - Invoked when a mutex is destroyed
+ * p1 = address of mutex
+ * p2 = not used
+ * return val ignored
+ *
+ * LT_DIAG_MUTEX_LOCK - Invoked when a mutex lock is obtained
+ * p1 = address of mutex
+ * p2 = function return value
+ * return val ignored
+ *
+ * LT_DIAG_MUTEX_BLOCKED - Invoked when an lthread blocks on a mutex
+ * p1 = address of mutex
+ * p2 = function return value
+ * return val ignored
+ *
+ * LT_DIAG_MUTEX_TRYLOCK - Invoked when a mutex try lock is attempted
+ * p1 = address of mutex
+ * p2 = the function return value
+ * return val ignored
+ *
+ * LT_DIAG_MUTEX_UNLOCKED - Invoked when a mutex is unlocked
+ * p1 = address of mutex
+ * p2 = the thread that was unlocked, or error code
+ * return val ignored
+ */
+typedef uint64_t (*diag_callback) (uint64_t time, struct lthread *lt,
+ int diag_event, uint64_t diag_ref,
+ const char *text, uint64_t p1, uint64_t p2);
+
+/*
+ * Set user diagnostic callback and mask
+ * If the callback function pointer is NULL the default
+ * callback handler will be restored.
+ */
+void lthread_diagnostic_enable(diag_callback cb, uint64_t diag_mask);
+
+/*
+ * Set diagnostic mask
+ */
+void lthread_diagnostic_set_mask(uint64_t mask);
+
+/*
+ * lthread diagnostic callback
+ */
+enum lthread_diag_ev {
+ /* bits 0 - 14 lthread flag group */
+ LT_DIAG_LTHREAD_CREATE, /* 00 mask 0x00000001 */
+ LT_DIAG_LTHREAD_EXIT, /* 01 mask 0x00000002 */
+ LT_DIAG_LTHREAD_JOIN, /* 02 mask 0x00000004 */
+ LT_DIAG_LTHREAD_CANCEL, /* 03 mask 0x00000008 */
+ LT_DIAG_LTHREAD_DETACH, /* 04 mask 0x00000010 */
+ LT_DIAG_LTHREAD_FREE, /* 05 mask 0x00000020 */
+ LT_DIAG_LTHREAD_SUSPENDED, /* 06 mask 0x00000040 */
+ LT_DIAG_LTHREAD_YIELD, /* 07 mask 0x00000080 */
+ LT_DIAG_LTHREAD_RESCHEDULED, /* 08 mask 0x00000100 */
+ LT_DIAG_LTHREAD_SLEEP, /* 09 mask 0x00000200 */
+ LT_DIAG_LTHREAD_RESUMED, /* 10 mask 0x00000400 */
+ LT_DIAG_LTHREAD_AFFINITY, /* 11 mask 0x00000800 */
+ LT_DIAG_LTHREAD_TMR_START, /* 12 mask 0x00001000 */
+ LT_DIAG_LTHREAD_TMR_DELETE, /* 13 mask 0x00002000 */
+ LT_DIAG_LTHREAD_TMR_EXPIRED, /* 14 mask 0x00004000 */
+ /* bits 15 - 19 conditional variable flag group */
+ LT_DIAG_COND_CREATE, /* 15 mask 0x00008000 */
+ LT_DIAG_COND_DESTROY, /* 16 mask 0x00010000 */
+ LT_DIAG_COND_WAIT, /* 17 mask 0x00020000 */
+ LT_DIAG_COND_SIGNAL, /* 18 mask 0x00040000 */
+ LT_DIAG_COND_BROADCAST, /* 19 mask 0x00080000 */
+ /* bits 20 - 25 mutex flag group */
+ LT_DIAG_MUTEX_CREATE, /* 20 mask 0x00100000 */
+ LT_DIAG_MUTEX_DESTROY, /* 21 mask 0x00200000 */
+ LT_DIAG_MUTEX_LOCK, /* 22 mask 0x00400000 */
+ LT_DIAG_MUTEX_TRYLOCK, /* 23 mask 0x00800000 */
+ LT_DIAG_MUTEX_BLOCKED, /* 24 mask 0x01000000 */
+ LT_DIAG_MUTEX_UNLOCKED, /* 25 mask 0x02000000 */
+ /* bits 26 - 27 scheduler flag group - 8 bits */
+ LT_DIAG_SCHED_CREATE, /* 26 mask 0x04000000 */
+ LT_DIAG_SCHED_SHUTDOWN, /* 27 mask 0x08000000 */
+ LT_DIAG_EVENT_MAX
+};
+
+#define LT_DIAG_ALL 0xffffffffffffffff
+
+
+/*
+ * Display scheduler stats
+ */
+void
+lthread_sched_stats_display(void);
+
+/*
+ * return the diagnostic ref val stored in a condition var
+ */
+uint64_t
+lthread_cond_diag_ref(struct lthread_cond *c);
+
+/*
+ * return the diagnostic ref val stored in a mutex
+ */
+uint64_t
+lthread_mutex_diag_ref(struct lthread_mutex *m);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* LTHREAD_DIAG_API_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_int.h b/src/spdk/dpdk/examples/performance-thread/common/lthread_int.h
new file mode 100644
index 000000000..a352f13b7
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_int.h
@@ -0,0 +1,152 @@
+/*
+ * SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2015 Intel Corporation.
+ * Copyright 2012 Hasan Alayli <halayli@gmail.com>
+ */
+#ifndef LTHREAD_INT_H
+#define LTHREAD_INT_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stdint.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <errno.h>
+#include <pthread.h>
+#include <time.h>
+
+#include <rte_memory.h>
+#include <rte_cycles.h>
+#include <rte_per_lcore.h>
+#include <rte_timer.h>
+#include <rte_atomic_64.h>
+#include <rte_spinlock.h>
+#include <ctx.h>
+
+#include <lthread_api.h>
+#include "lthread.h"
+#include "lthread_diag.h"
+#include "lthread_tls.h"
+
+struct lthread;
+struct lthread_sched;
+struct lthread_cond;
+struct lthread_mutex;
+struct lthread_key;
+
+struct key_pool;
+struct qnode;
+struct qnode_pool;
+struct lthread_sched;
+struct lthread_tls;
+
+
+#define BIT(x) (1 << (x))
+#define CLEARBIT(x) ~(1 << (x))
+
+#define POSIX_ERRNO(x) (x)
+
+#define MAX_LTHREAD_NAME_SIZE 64
+
+#define RTE_LOGTYPE_LTHREAD RTE_LOGTYPE_USER1
+
+
+/* define some shorthand for current scheduler and current thread */
+#define THIS_SCHED RTE_PER_LCORE(this_sched)
+#define THIS_LTHREAD RTE_PER_LCORE(this_sched)->current_lthread
+
+/*
+ * Definition of an scheduler struct
+ */
+struct lthread_sched {
+ struct ctx ctx; /* cpu context */
+ uint64_t birth; /* time created */
+ struct lthread *current_lthread; /* running thread */
+ unsigned lcore_id; /* this sched lcore */
+ int run_flag; /* sched shutdown */
+ uint64_t nb_blocked_threads; /* blocked threads */
+ struct lthread_queue *ready; /* local ready queue */
+ struct lthread_queue *pready; /* peer ready queue */
+ struct lthread_objcache *lthread_cache; /* free lthreads */
+ struct lthread_objcache *stack_cache; /* free stacks */
+ struct lthread_objcache *per_lthread_cache; /* free per lthread */
+ struct lthread_objcache *tls_cache; /* free TLS */
+ struct lthread_objcache *cond_cache; /* free cond vars */
+ struct lthread_objcache *mutex_cache; /* free mutexes */
+ struct qnode_pool *qnode_pool; /* pool of queue nodes */
+ struct key_pool *key_pool; /* pool of free TLS keys */
+ size_t stack_size;
+ uint64_t diag_ref; /* diag ref */
+} __rte_cache_aligned;
+
+RTE_DECLARE_PER_LCORE(struct lthread_sched *, this_sched);
+
+
+/*
+ * State for an lthread
+ */
+enum lthread_st {
+ ST_LT_INIT, /* initial state */
+ ST_LT_READY, /* lthread is ready to run */
+ ST_LT_SLEEPING, /* lthread is sleeping */
+ ST_LT_EXPIRED, /* lthread timeout has expired */
+ ST_LT_EXITED, /* lthread has exited and needs cleanup */
+ ST_LT_DETACH, /* lthread frees on exit*/
+ ST_LT_CANCELLED, /* lthread has been cancelled */
+};
+
+/*
+ * lthread sub states for exit/join
+ */
+enum join_st {
+ LT_JOIN_INITIAL, /* initial state */
+ LT_JOIN_EXITING, /* thread is exiting */
+ LT_JOIN_THREAD_SET, /* joining thread has been set */
+ LT_JOIN_EXIT_VAL_SET, /* exiting thread has set ret val */
+ LT_JOIN_EXIT_VAL_READ, /* joining thread has collected ret val */
+};
+
+/* defnition of an lthread stack object */
+struct lthread_stack {
+ uint8_t stack[LTHREAD_MAX_STACK_SIZE];
+ size_t stack_size;
+ struct lthread_sched *root_sched;
+} __rte_cache_aligned;
+
+/*
+ * Definition of an lthread
+ */
+struct lthread {
+ struct ctx ctx; /* cpu context */
+
+ uint64_t state; /* current lthread state */
+
+ struct lthread_sched *sched; /* current scheduler */
+ void *stack; /* ptr to actual stack */
+ size_t stack_size; /* current stack_size */
+ size_t last_stack_size; /* last yield stack_size */
+ lthread_func_t fun; /* func ctx is running */
+ void *arg; /* func args passed to func */
+ void *per_lthread_data; /* per lthread user data */
+ lthread_exit_func exit_handler; /* called when thread exits */
+ uint64_t birth; /* time lthread was born */
+ struct lthread_queue *pending_wr_queue; /* deferred queue to write */
+ struct lthread *lt_join; /* lthread to join on */
+ uint64_t join; /* state for joining */
+ void **lt_exit_ptr; /* exit ptr for lthread_join */
+ struct lthread_sched *root_sched; /* thread was created here*/
+ struct queue_node *qnode; /* node when in a queue */
+ struct rte_timer tim; /* sleep timer */
+ struct lthread_tls *tls; /* keys in use by the thread */
+ struct lthread_stack *stack_container; /* stack */
+ char funcname[MAX_LTHREAD_NAME_SIZE]; /* thread func name */
+ uint64_t diag_ref; /* ref to user diag data */
+} __rte_cache_aligned;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* LTHREAD_INT_H */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_mutex.c b/src/spdk/dpdk/examples/performance-thread/common/lthread_mutex.c
new file mode 100644
index 000000000..01da6cad4
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_mutex.c
@@ -0,0 +1,224 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2015 Intel Corporation
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdint.h>
+#include <stddef.h>
+#include <limits.h>
+#include <inttypes.h>
+#include <unistd.h>
+#include <pthread.h>
+#include <fcntl.h>
+#include <sys/time.h>
+#include <sys/mman.h>
+
+#include <rte_per_lcore.h>
+#include <rte_log.h>
+#include <rte_spinlock.h>
+#include <rte_common.h>
+
+#include "lthread_api.h"
+#include "lthread_int.h"
+#include "lthread_mutex.h"
+#include "lthread_sched.h"
+#include "lthread_queue.h"
+#include "lthread_objcache.h"
+#include "lthread_diag.h"
+
+/*
+ * Create a mutex
+ */
+int
+lthread_mutex_init(char *name, struct lthread_mutex **mutex,
+ __rte_unused const struct lthread_mutexattr *attr)
+{
+ struct lthread_mutex *m;
+
+ if (mutex == NULL)
+ return POSIX_ERRNO(EINVAL);
+
+
+ m = _lthread_objcache_alloc((THIS_SCHED)->mutex_cache);
+ if (m == NULL)
+ return POSIX_ERRNO(EAGAIN);
+
+ m->blocked = _lthread_queue_create("blocked queue");
+ if (m->blocked == NULL) {
+ _lthread_objcache_free((THIS_SCHED)->mutex_cache, m);
+ return POSIX_ERRNO(EAGAIN);
+ }
+
+ if (name == NULL)
+ strncpy(m->name, "no name", sizeof(m->name));
+ else
+ strncpy(m->name, name, sizeof(m->name));
+ m->name[sizeof(m->name)-1] = 0;
+
+ m->root_sched = THIS_SCHED;
+ m->owner = NULL;
+
+ rte_atomic64_init(&m->count);
+
+ DIAG_CREATE_EVENT(m, LT_DIAG_MUTEX_CREATE);
+ /* success */
+ (*mutex) = m;
+ return 0;
+}
+
+/*
+ * Destroy a mutex
+ */
+int lthread_mutex_destroy(struct lthread_mutex *m)
+{
+ if ((m == NULL) || (m->blocked == NULL)) {
+ DIAG_EVENT(m, LT_DIAG_MUTEX_DESTROY, m, POSIX_ERRNO(EINVAL));
+ return POSIX_ERRNO(EINVAL);
+ }
+
+ if (m->owner == NULL) {
+ /* try to delete the blocked queue */
+ if (_lthread_queue_destroy(m->blocked) < 0) {
+ DIAG_EVENT(m, LT_DIAG_MUTEX_DESTROY,
+ m, POSIX_ERRNO(EBUSY));
+ return POSIX_ERRNO(EBUSY);
+ }
+
+ /* free the mutex to cache */
+ _lthread_objcache_free(m->root_sched->mutex_cache, m);
+ DIAG_EVENT(m, LT_DIAG_MUTEX_DESTROY, m, 0);
+ return 0;
+ }
+ /* can't do its still in use */
+ DIAG_EVENT(m, LT_DIAG_MUTEX_DESTROY, m, POSIX_ERRNO(EBUSY));
+ return POSIX_ERRNO(EBUSY);
+}
+
+/*
+ * Try to obtain a mutex
+ */
+int lthread_mutex_lock(struct lthread_mutex *m)
+{
+ struct lthread *lt = THIS_LTHREAD;
+
+ if ((m == NULL) || (m->blocked == NULL)) {
+ DIAG_EVENT(m, LT_DIAG_MUTEX_LOCK, m, POSIX_ERRNO(EINVAL));
+ return POSIX_ERRNO(EINVAL);
+ }
+
+ /* allow no recursion */
+ if (m->owner == lt) {
+ DIAG_EVENT(m, LT_DIAG_MUTEX_LOCK, m, POSIX_ERRNO(EDEADLK));
+ return POSIX_ERRNO(EDEADLK);
+ }
+
+ for (;;) {
+ rte_atomic64_inc(&m->count);
+ do {
+ if (rte_atomic64_cmpset
+ ((uint64_t *) &m->owner, 0, (uint64_t) lt)) {
+ /* happy days, we got the lock */
+ DIAG_EVENT(m, LT_DIAG_MUTEX_LOCK, m, 0);
+ return 0;
+ }
+ /* spin due to race with unlock when
+ * nothing was blocked
+ */
+ } while ((rte_atomic64_read(&m->count) == 1) &&
+ (m->owner == NULL));
+
+ /* queue the current thread in the blocked queue
+ * we defer this to after we return to the scheduler
+ * to ensure that the current thread context is saved
+ * before unlock could result in it being dequeued and
+ * resumed
+ */
+ DIAG_EVENT(m, LT_DIAG_MUTEX_BLOCKED, m, lt);
+ lt->pending_wr_queue = m->blocked;
+ /* now relinquish cpu */
+ _suspend();
+ /* resumed, must loop and compete for the lock again */
+ }
+ return 0;
+}
+
+/* try to lock a mutex but don't block */
+int lthread_mutex_trylock(struct lthread_mutex *m)
+{
+ struct lthread *lt = THIS_LTHREAD;
+
+ if ((m == NULL) || (m->blocked == NULL)) {
+ DIAG_EVENT(m, LT_DIAG_MUTEX_TRYLOCK, m, POSIX_ERRNO(EINVAL));
+ return POSIX_ERRNO(EINVAL);
+ }
+
+ if (m->owner == lt) {
+ /* no recursion */
+ DIAG_EVENT(m, LT_DIAG_MUTEX_TRYLOCK, m, POSIX_ERRNO(EDEADLK));
+ return POSIX_ERRNO(EDEADLK);
+ }
+
+ rte_atomic64_inc(&m->count);
+ if (rte_atomic64_cmpset
+ ((uint64_t *) &m->owner, (uint64_t) NULL, (uint64_t) lt)) {
+ /* got the lock */
+ DIAG_EVENT(m, LT_DIAG_MUTEX_TRYLOCK, m, 0);
+ return 0;
+ }
+
+ /* failed so return busy */
+ rte_atomic64_dec(&m->count);
+ DIAG_EVENT(m, LT_DIAG_MUTEX_TRYLOCK, m, POSIX_ERRNO(EBUSY));
+ return POSIX_ERRNO(EBUSY);
+}
+
+/*
+ * Unlock a mutex
+ */
+int lthread_mutex_unlock(struct lthread_mutex *m)
+{
+ struct lthread *lt = THIS_LTHREAD;
+ struct lthread *unblocked;
+
+ if ((m == NULL) || (m->blocked == NULL)) {
+ DIAG_EVENT(m, LT_DIAG_MUTEX_UNLOCKED, m, POSIX_ERRNO(EINVAL));
+ return POSIX_ERRNO(EINVAL);
+ }
+
+ /* fail if its owned */
+ if (m->owner != lt || m->owner == NULL) {
+ DIAG_EVENT(m, LT_DIAG_MUTEX_UNLOCKED, m, POSIX_ERRNO(EPERM));
+ return POSIX_ERRNO(EPERM);
+ }
+
+ rte_atomic64_dec(&m->count);
+ /* if there are blocked threads then make one ready */
+ while (rte_atomic64_read(&m->count) > 0) {
+ unblocked = _lthread_queue_remove(m->blocked);
+
+ if (unblocked != NULL) {
+ rte_atomic64_dec(&m->count);
+ DIAG_EVENT(m, LT_DIAG_MUTEX_UNLOCKED, m, unblocked);
+ RTE_ASSERT(unblocked->sched != NULL);
+ _ready_queue_insert((struct lthread_sched *)
+ unblocked->sched, unblocked);
+ break;
+ }
+ }
+ /* release the lock */
+ m->owner = NULL;
+ return 0;
+}
+
+/*
+ * return the diagnostic ref val stored in a mutex
+ */
+uint64_t
+lthread_mutex_diag_ref(struct lthread_mutex *m)
+{
+ if (m == NULL)
+ return 0;
+ return m->diag_ref;
+}
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_mutex.h b/src/spdk/dpdk/examples/performance-thread/common/lthread_mutex.h
new file mode 100644
index 000000000..cd866f87b
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_mutex.h
@@ -0,0 +1,31 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2015 Intel Corporation
+ */
+
+
+#ifndef LTHREAD_MUTEX_H_
+#define LTHREAD_MUTEX_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "lthread_queue.h"
+
+
+#define MAX_MUTEX_NAME_SIZE 64
+
+struct lthread_mutex {
+ struct lthread *owner;
+ rte_atomic64_t count;
+ struct lthread_queue *blocked __rte_cache_aligned;
+ struct lthread_sched *root_sched;
+ char name[MAX_MUTEX_NAME_SIZE];
+ uint64_t diag_ref; /* optional ref to user diag data */
+} __rte_cache_aligned;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* LTHREAD_MUTEX_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_objcache.h b/src/spdk/dpdk/examples/performance-thread/common/lthread_objcache.h
new file mode 100644
index 000000000..777a1945b
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_objcache.h
@@ -0,0 +1,136 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2015 Intel Corporation
+ */
+#ifndef LTHREAD_OBJCACHE_H_
+#define LTHREAD_OBJCACHE_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <string.h>
+
+#include <rte_per_lcore.h>
+#include <rte_malloc.h>
+#include <rte_memory.h>
+
+#include "lthread_int.h"
+#include "lthread_diag.h"
+#include "lthread_queue.h"
+
+
+RTE_DECLARE_PER_LCORE(struct lthread_sched *, this_sched);
+
+struct lthread_objcache {
+ struct lthread_queue *q;
+ size_t obj_size;
+ int prealloc_size;
+ char name[LT_MAX_NAME_SIZE];
+
+ DIAG_COUNT_DEFINE(rd);
+ DIAG_COUNT_DEFINE(wr);
+ DIAG_COUNT_DEFINE(prealloc);
+ DIAG_COUNT_DEFINE(capacity);
+ DIAG_COUNT_DEFINE(available);
+};
+
+/*
+ * Create a cache
+ */
+static inline struct
+lthread_objcache *_lthread_objcache_create(const char *name,
+ size_t obj_size,
+ int prealloc_size)
+{
+ struct lthread_objcache *c =
+ rte_malloc_socket(NULL, sizeof(struct lthread_objcache),
+ RTE_CACHE_LINE_SIZE,
+ rte_socket_id());
+ if (c == NULL)
+ return NULL;
+
+ c->q = _lthread_queue_create("cache queue");
+ if (c->q == NULL) {
+ rte_free(c);
+ return NULL;
+ }
+ c->obj_size = obj_size;
+ c->prealloc_size = prealloc_size;
+
+ if (name != NULL)
+ strncpy(c->name, name, LT_MAX_NAME_SIZE);
+ c->name[sizeof(c->name)-1] = 0;
+
+ DIAG_COUNT_INIT(c, rd);
+ DIAG_COUNT_INIT(c, wr);
+ DIAG_COUNT_INIT(c, prealloc);
+ DIAG_COUNT_INIT(c, capacity);
+ DIAG_COUNT_INIT(c, available);
+ return c;
+}
+
+/*
+ * Destroy an objcache
+ */
+static inline int
+_lthread_objcache_destroy(struct lthread_objcache *c)
+{
+ if (_lthread_queue_destroy(c->q) == 0) {
+ rte_free(c);
+ return 0;
+ }
+ return -1;
+}
+
+/*
+ * Allocate an object from an object cache
+ */
+static inline void *
+_lthread_objcache_alloc(struct lthread_objcache *c)
+{
+ int i;
+ void *data;
+ struct lthread_queue *q = c->q;
+ size_t obj_size = c->obj_size;
+ int prealloc_size = c->prealloc_size;
+
+ data = _lthread_queue_remove(q);
+
+ if (data == NULL) {
+ DIAG_COUNT_INC(c, prealloc);
+ for (i = 0; i < prealloc_size; i++) {
+ data =
+ rte_zmalloc_socket(NULL, obj_size,
+ RTE_CACHE_LINE_SIZE,
+ rte_socket_id());
+ if (data == NULL)
+ return NULL;
+
+ DIAG_COUNT_INC(c, available);
+ DIAG_COUNT_INC(c, capacity);
+ _lthread_queue_insert_mp(q, data);
+ }
+ data = _lthread_queue_remove(q);
+ }
+ DIAG_COUNT_INC(c, rd);
+ DIAG_COUNT_DEC(c, available);
+ return data;
+}
+
+/*
+ * free an object to a cache
+ */
+static inline void
+_lthread_objcache_free(struct lthread_objcache *c, void *obj)
+{
+ DIAG_COUNT_INC(c, wr);
+ DIAG_COUNT_INC(c, available);
+ _lthread_queue_insert_mp(c->q, obj);
+}
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* LTHREAD_OBJCACHE_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_pool.h b/src/spdk/dpdk/examples/performance-thread/common/lthread_pool.h
new file mode 100644
index 000000000..6f93775fb
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_pool.h
@@ -0,0 +1,277 @@
+/*
+ * SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2015 Intel Corporation.
+ * Copyright 2010-2011 Dmitry Vyukov
+ */
+
+#ifndef LTHREAD_POOL_H_
+#define LTHREAD_POOL_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <rte_malloc.h>
+#include <rte_per_lcore.h>
+#include <rte_log.h>
+
+#include "lthread_int.h"
+#include "lthread_diag.h"
+
+/*
+ * This file implements pool of queue nodes used by the queue implemented
+ * in lthread_queue.h.
+ *
+ * The pool is an intrusive lock free MPSC queue.
+ *
+ * The pool is created empty and populated lazily, i.e. on first attempt to
+ * allocate a the pool.
+ *
+ * Whenever the pool is empty more nodes are added to the pool
+ * The number of nodes preallocated in this way is a parameter of
+ * _qnode_pool_create. Freeing an object returns it to the pool.
+ *
+ * Each lthread scheduler maintains its own pool of nodes. L-threads must always
+ * allocate from this local pool ( because it is a single consumer queue ).
+ * L-threads can free nodes to any pool (because it is a multi producer queue)
+ * This enables threads that have affined to a different scheduler to free
+ * nodes safely.
+ */
+
+struct qnode;
+struct qnode_cache;
+
+/*
+ * define intermediate node
+ */
+struct qnode {
+ struct qnode *next;
+ void *data;
+ struct qnode_pool *pool;
+} __rte_cache_aligned;
+
+/*
+ * a pool structure
+ */
+struct qnode_pool {
+ struct qnode *head;
+ struct qnode *stub;
+ struct qnode *fast_alloc;
+ struct qnode *tail __rte_cache_aligned;
+ int pre_alloc;
+ char name[LT_MAX_NAME_SIZE];
+
+ DIAG_COUNT_DEFINE(rd);
+ DIAG_COUNT_DEFINE(wr);
+ DIAG_COUNT_DEFINE(available);
+ DIAG_COUNT_DEFINE(prealloc);
+ DIAG_COUNT_DEFINE(capacity);
+} __rte_cache_aligned;
+
+/*
+ * Create a pool of qnodes
+ */
+
+static inline struct qnode_pool *
+_qnode_pool_create(const char *name, int prealloc_size) {
+
+ struct qnode_pool *p = rte_malloc_socket(NULL,
+ sizeof(struct qnode_pool),
+ RTE_CACHE_LINE_SIZE,
+ rte_socket_id());
+
+ RTE_ASSERT(p);
+
+ p->stub = rte_malloc_socket(NULL,
+ sizeof(struct qnode),
+ RTE_CACHE_LINE_SIZE,
+ rte_socket_id());
+
+ RTE_ASSERT(p->stub);
+
+ if (name != NULL)
+ strncpy(p->name, name, LT_MAX_NAME_SIZE);
+ p->name[sizeof(p->name)-1] = 0;
+
+ p->stub->pool = p;
+ p->stub->next = NULL;
+ p->tail = p->stub;
+ p->head = p->stub;
+ p->pre_alloc = prealloc_size;
+
+ DIAG_COUNT_INIT(p, rd);
+ DIAG_COUNT_INIT(p, wr);
+ DIAG_COUNT_INIT(p, available);
+ DIAG_COUNT_INIT(p, prealloc);
+ DIAG_COUNT_INIT(p, capacity);
+
+ return p;
+}
+
+
+/*
+ * Insert a node into the pool
+ */
+static __rte_always_inline void
+_qnode_pool_insert(struct qnode_pool *p, struct qnode *n)
+{
+ n->next = NULL;
+ struct qnode *prev = n;
+ /* We insert at the head */
+ prev = (struct qnode *) __sync_lock_test_and_set((uint64_t *)&p->head,
+ (uint64_t) prev);
+ /* there is a window of inconsistency until prev next is set */
+ /* which is why remove must retry */
+ prev->next = (n);
+}
+
+/*
+ * Remove a node from the pool
+ *
+ * There is a race with _qnode_pool_insert() whereby the queue could appear
+ * empty during a concurrent insert, this is handled by retrying
+ *
+ * The queue uses a stub node, which must be swung as the queue becomes
+ * empty, this requires an insert of the stub, which means that removing the
+ * last item from the queue incurs the penalty of an atomic exchange. Since the
+ * pool is maintained with a bulk pre-allocation the cost of this is amortised.
+ */
+static __rte_always_inline struct qnode *
+_pool_remove(struct qnode_pool *p)
+{
+ struct qnode *head;
+ struct qnode *tail = p->tail;
+ struct qnode *next = tail->next;
+
+ /* we remove from the tail */
+ if (tail == p->stub) {
+ if (next == NULL)
+ return NULL;
+ /* advance the tail */
+ p->tail = next;
+ tail = next;
+ next = next->next;
+ }
+ if (likely(next != NULL)) {
+ p->tail = next;
+ return tail;
+ }
+
+ head = p->head;
+ if (tail == head)
+ return NULL;
+
+ /* swing stub node */
+ _qnode_pool_insert(p, p->stub);
+
+ next = tail->next;
+ if (next) {
+ p->tail = next;
+ return tail;
+ }
+ return NULL;
+}
+
+
+/*
+ * This adds a retry to the _pool_remove function
+ * defined above
+ */
+static __rte_always_inline struct qnode *
+_qnode_pool_remove(struct qnode_pool *p)
+{
+ struct qnode *n;
+
+ do {
+ n = _pool_remove(p);
+ if (likely(n != NULL))
+ return n;
+
+ rte_compiler_barrier();
+ } while ((p->head != p->tail) &&
+ (p->tail != p->stub));
+ return NULL;
+}
+
+/*
+ * Allocate a node from the pool
+ * If the pool is empty add mode nodes
+ */
+static __rte_always_inline struct qnode *
+_qnode_alloc(void)
+{
+ struct qnode_pool *p = (THIS_SCHED)->qnode_pool;
+ int prealloc_size = p->pre_alloc;
+ struct qnode *n;
+ int i;
+
+ if (likely(p->fast_alloc != NULL)) {
+ n = p->fast_alloc;
+ p->fast_alloc = NULL;
+ return n;
+ }
+
+ n = _qnode_pool_remove(p);
+
+ if (unlikely(n == NULL)) {
+ DIAG_COUNT_INC(p, prealloc);
+ for (i = 0; i < prealloc_size; i++) {
+ n = rte_malloc_socket(NULL,
+ sizeof(struct qnode),
+ RTE_CACHE_LINE_SIZE,
+ rte_socket_id());
+ if (n == NULL)
+ return NULL;
+
+ DIAG_COUNT_INC(p, available);
+ DIAG_COUNT_INC(p, capacity);
+
+ n->pool = p;
+ _qnode_pool_insert(p, n);
+ }
+ n = _qnode_pool_remove(p);
+ }
+ n->pool = p;
+ DIAG_COUNT_INC(p, rd);
+ DIAG_COUNT_DEC(p, available);
+ return n;
+}
+
+
+
+/*
+* free a queue node to the per scheduler pool from which it came
+*/
+static __rte_always_inline void
+_qnode_free(struct qnode *n)
+{
+ struct qnode_pool *p = n->pool;
+
+
+ if (unlikely(p->fast_alloc != NULL) ||
+ unlikely(n->pool != (THIS_SCHED)->qnode_pool)) {
+ DIAG_COUNT_INC(p, wr);
+ DIAG_COUNT_INC(p, available);
+ _qnode_pool_insert(p, n);
+ return;
+ }
+ p->fast_alloc = n;
+}
+
+/*
+ * Destroy an qnode pool
+ * queue must be empty when this is called
+ */
+static inline int
+_qnode_pool_destroy(struct qnode_pool *p)
+{
+ rte_free(p->stub);
+ rte_free(p);
+ return 0;
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* LTHREAD_POOL_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_queue.h b/src/spdk/dpdk/examples/performance-thread/common/lthread_queue.h
new file mode 100644
index 000000000..5b63ba220
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_queue.h
@@ -0,0 +1,247 @@
+/*
+ * SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2015 Intel Corporation.
+ * Copyright 2010-2011 Dmitry Vyukov
+ */
+
+#ifndef LTHREAD_QUEUE_H_
+#define LTHREAD_QUEUE_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <string.h>
+
+#include <rte_prefetch.h>
+#include <rte_per_lcore.h>
+
+#include "lthread_int.h"
+#include "lthread.h"
+#include "lthread_diag.h"
+#include "lthread_pool.h"
+
+struct lthread_queue;
+
+/*
+ * This file implements an unbounded FIFO queue based on a lock free
+ * linked list.
+ *
+ * The queue is non-intrusive in that it uses intermediate nodes, and does
+ * not require these nodes to be inserted into the object being placed
+ * in the queue.
+ *
+ * This is slightly more efficient than the very similar queue in lthread_pool
+ * in that it does not have to swing a stub node as the queue becomes empty.
+ *
+ * The queue access functions allocate and free intermediate node
+ * transparently from/to a per scheduler pool ( see lthread_pool.h ).
+ *
+ * The queue provides both MPSC and SPSC insert methods
+ */
+
+/*
+ * define a queue of lthread nodes
+ */
+struct lthread_queue {
+ struct qnode *head;
+ struct qnode *tail __rte_cache_aligned;
+ struct lthread_queue *p;
+ char name[LT_MAX_NAME_SIZE];
+
+ DIAG_COUNT_DEFINE(rd);
+ DIAG_COUNT_DEFINE(wr);
+ DIAG_COUNT_DEFINE(size);
+
+} __rte_cache_aligned;
+
+
+
+static inline struct lthread_queue *
+_lthread_queue_create(const char *name)
+{
+ struct qnode *stub;
+ struct lthread_queue *new_queue;
+
+ new_queue = rte_malloc_socket(NULL, sizeof(struct lthread_queue),
+ RTE_CACHE_LINE_SIZE,
+ rte_socket_id());
+ if (new_queue == NULL)
+ return NULL;
+
+ /* allocated stub node */
+ stub = _qnode_alloc();
+ RTE_ASSERT(stub);
+
+ if (name != NULL)
+ strncpy(new_queue->name, name, sizeof(new_queue->name));
+ new_queue->name[sizeof(new_queue->name)-1] = 0;
+
+ /* initialize queue as empty */
+ stub->next = NULL;
+ new_queue->head = stub;
+ new_queue->tail = stub;
+
+ DIAG_COUNT_INIT(new_queue, rd);
+ DIAG_COUNT_INIT(new_queue, wr);
+ DIAG_COUNT_INIT(new_queue, size);
+
+ return new_queue;
+}
+
+/**
+ * Return true if the queue is empty
+ */
+static __rte_always_inline int
+_lthread_queue_empty(struct lthread_queue *q)
+{
+ return q->tail == q->head;
+}
+
+
+
+/**
+ * Destroy a queue
+ * fail if queue is not empty
+ */
+static inline int _lthread_queue_destroy(struct lthread_queue *q)
+{
+ if (q == NULL)
+ return -1;
+
+ if (!_lthread_queue_empty(q))
+ return -1;
+
+ _qnode_free(q->head);
+ rte_free(q);
+ return 0;
+}
+
+RTE_DECLARE_PER_LCORE(struct lthread_sched *, this_sched);
+
+/*
+ * Insert a node into a queue
+ * this implementation is multi producer safe
+ */
+static __rte_always_inline struct qnode *
+_lthread_queue_insert_mp(struct lthread_queue
+ *q, void *data)
+{
+ struct qnode *prev;
+ struct qnode *n = _qnode_alloc();
+
+ if (n == NULL)
+ return NULL;
+
+ /* set object in node */
+ n->data = data;
+ n->next = NULL;
+
+ /* this is an MPSC method, perform a locked update */
+ prev = n;
+ prev =
+ (struct qnode *)__sync_lock_test_and_set((uint64_t *) &(q)->head,
+ (uint64_t) prev);
+ /* there is a window of inconsistency until prev next is set,
+ * which is why remove must retry
+ */
+ prev->next = n;
+
+ DIAG_COUNT_INC(q, wr);
+ DIAG_COUNT_INC(q, size);
+
+ return n;
+}
+
+/*
+ * Insert an node into a queue in single producer mode
+ * this implementation is NOT mult producer safe
+ */
+static __rte_always_inline struct qnode *
+_lthread_queue_insert_sp(struct lthread_queue
+ *q, void *data)
+{
+ /* allocate a queue node */
+ struct qnode *prev;
+ struct qnode *n = _qnode_alloc();
+
+ if (n == NULL)
+ return NULL;
+
+ /* set data in node */
+ n->data = data;
+ n->next = NULL;
+
+ /* this is an SPSC method, no need for locked exchange operation */
+ prev = q->head;
+ prev->next = q->head = n;
+
+ DIAG_COUNT_INC(q, wr);
+ DIAG_COUNT_INC(q, size);
+
+ return n;
+}
+
+/*
+ * Remove a node from a queue
+ */
+static __rte_always_inline void *
+_lthread_queue_poll(struct lthread_queue *q)
+{
+ void *data = NULL;
+ struct qnode *tail = q->tail;
+ struct qnode *next = (struct qnode *)tail->next;
+ /*
+ * There is a small window of inconsistency between producer and
+ * consumer whereby the queue may appear empty if consumer and
+ * producer access it at the same time.
+ * The consumer must handle this by retrying
+ */
+
+ if (likely(next != NULL)) {
+ q->tail = next;
+ tail->data = next->data;
+ data = tail->data;
+
+ /* free the node */
+ _qnode_free(tail);
+
+ DIAG_COUNT_INC(q, rd);
+ DIAG_COUNT_DEC(q, size);
+ return data;
+ }
+ return NULL;
+}
+
+/*
+ * Remove a node from a queue
+ */
+static __rte_always_inline void *
+_lthread_queue_remove(struct lthread_queue *q)
+{
+ void *data = NULL;
+
+ /*
+ * There is a small window of inconsistency between producer and
+ * consumer whereby the queue may appear empty if consumer and
+ * producer access it at the same time. We handle this by retrying
+ */
+ do {
+ data = _lthread_queue_poll(q);
+
+ if (likely(data != NULL)) {
+
+ DIAG_COUNT_INC(q, rd);
+ DIAG_COUNT_DEC(q, size);
+ return data;
+ }
+ rte_compiler_barrier();
+ } while (unlikely(!_lthread_queue_empty(q)));
+ return NULL;
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* LTHREAD_QUEUE_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_sched.c b/src/spdk/dpdk/examples/performance-thread/common/lthread_sched.c
new file mode 100644
index 000000000..38ca0c45c
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_sched.c
@@ -0,0 +1,544 @@
+/*
+ * SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2015 Intel Corporation.
+ * Copyright 2012 Hasan Alayli <halayli@gmail.com>
+ */
+
+#define RTE_MEM 1
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdint.h>
+#include <stddef.h>
+#include <limits.h>
+#include <inttypes.h>
+#include <unistd.h>
+#include <pthread.h>
+#include <fcntl.h>
+#include <sys/time.h>
+#include <sys/mman.h>
+#include <sched.h>
+
+#include <rte_prefetch.h>
+#include <rte_per_lcore.h>
+#include <rte_atomic.h>
+#include <rte_atomic_64.h>
+#include <rte_log.h>
+#include <rte_common.h>
+#include <rte_branch_prediction.h>
+
+#include "lthread_api.h"
+#include "lthread_int.h"
+#include "lthread_sched.h"
+#include "lthread_objcache.h"
+#include "lthread_timer.h"
+#include "lthread_mutex.h"
+#include "lthread_cond.h"
+#include "lthread_tls.h"
+#include "lthread_diag.h"
+
+/*
+ * This file implements the lthread scheduler
+ * The scheduler is the function lthread_run()
+ * This must be run as the main loop of an EAL thread.
+ *
+ * Currently once a scheduler is created it cannot be destroyed
+ * When a scheduler shuts down it is assumed that the application is terminating
+ */
+
+static rte_atomic16_t num_schedulers;
+static rte_atomic16_t active_schedulers;
+
+/* one scheduler per lcore */
+RTE_DEFINE_PER_LCORE(struct lthread_sched *, this_sched) = NULL;
+
+struct lthread_sched *schedcore[LTHREAD_MAX_LCORES];
+
+diag_callback diag_cb;
+
+uint64_t diag_mask;
+
+
+/* constructor */
+RTE_INIT(lthread_sched_ctor)
+{
+ memset(schedcore, 0, sizeof(schedcore));
+ rte_atomic16_init(&num_schedulers);
+ rte_atomic16_set(&num_schedulers, 1);
+ rte_atomic16_init(&active_schedulers);
+ rte_atomic16_set(&active_schedulers, 0);
+ diag_cb = NULL;
+}
+
+
+enum sched_alloc_phase {
+ SCHED_ALLOC_OK,
+ SCHED_ALLOC_QNODE_POOL,
+ SCHED_ALLOC_READY_QUEUE,
+ SCHED_ALLOC_PREADY_QUEUE,
+ SCHED_ALLOC_LTHREAD_CACHE,
+ SCHED_ALLOC_STACK_CACHE,
+ SCHED_ALLOC_PERLT_CACHE,
+ SCHED_ALLOC_TLS_CACHE,
+ SCHED_ALLOC_COND_CACHE,
+ SCHED_ALLOC_MUTEX_CACHE,
+};
+
+static int
+_lthread_sched_alloc_resources(struct lthread_sched *new_sched)
+{
+ int alloc_status;
+
+ do {
+ /* Initialize per scheduler queue node pool */
+ alloc_status = SCHED_ALLOC_QNODE_POOL;
+ new_sched->qnode_pool =
+ _qnode_pool_create("qnode pool", LTHREAD_PREALLOC);
+ if (new_sched->qnode_pool == NULL)
+ break;
+
+ /* Initialize per scheduler local ready queue */
+ alloc_status = SCHED_ALLOC_READY_QUEUE;
+ new_sched->ready = _lthread_queue_create("ready queue");
+ if (new_sched->ready == NULL)
+ break;
+
+ /* Initialize per scheduler local peer ready queue */
+ alloc_status = SCHED_ALLOC_PREADY_QUEUE;
+ new_sched->pready = _lthread_queue_create("pready queue");
+ if (new_sched->pready == NULL)
+ break;
+
+ /* Initialize per scheduler local free lthread cache */
+ alloc_status = SCHED_ALLOC_LTHREAD_CACHE;
+ new_sched->lthread_cache =
+ _lthread_objcache_create("lthread cache",
+ sizeof(struct lthread),
+ LTHREAD_PREALLOC);
+ if (new_sched->lthread_cache == NULL)
+ break;
+
+ /* Initialize per scheduler local free stack cache */
+ alloc_status = SCHED_ALLOC_STACK_CACHE;
+ new_sched->stack_cache =
+ _lthread_objcache_create("stack_cache",
+ sizeof(struct lthread_stack),
+ LTHREAD_PREALLOC);
+ if (new_sched->stack_cache == NULL)
+ break;
+
+ /* Initialize per scheduler local free per lthread data cache */
+ alloc_status = SCHED_ALLOC_PERLT_CACHE;
+ new_sched->per_lthread_cache =
+ _lthread_objcache_create("per_lt cache",
+ RTE_PER_LTHREAD_SECTION_SIZE,
+ LTHREAD_PREALLOC);
+ if (new_sched->per_lthread_cache == NULL)
+ break;
+
+ /* Initialize per scheduler local free tls cache */
+ alloc_status = SCHED_ALLOC_TLS_CACHE;
+ new_sched->tls_cache =
+ _lthread_objcache_create("TLS cache",
+ sizeof(struct lthread_tls),
+ LTHREAD_PREALLOC);
+ if (new_sched->tls_cache == NULL)
+ break;
+
+ /* Initialize per scheduler local free cond var cache */
+ alloc_status = SCHED_ALLOC_COND_CACHE;
+ new_sched->cond_cache =
+ _lthread_objcache_create("cond cache",
+ sizeof(struct lthread_cond),
+ LTHREAD_PREALLOC);
+ if (new_sched->cond_cache == NULL)
+ break;
+
+ /* Initialize per scheduler local free mutex cache */
+ alloc_status = SCHED_ALLOC_MUTEX_CACHE;
+ new_sched->mutex_cache =
+ _lthread_objcache_create("mutex cache",
+ sizeof(struct lthread_mutex),
+ LTHREAD_PREALLOC);
+ if (new_sched->mutex_cache == NULL)
+ break;
+
+ alloc_status = SCHED_ALLOC_OK;
+ } while (0);
+
+ /* roll back on any failure */
+ switch (alloc_status) {
+ case SCHED_ALLOC_MUTEX_CACHE:
+ _lthread_objcache_destroy(new_sched->cond_cache);
+ /* fall through */
+ case SCHED_ALLOC_COND_CACHE:
+ _lthread_objcache_destroy(new_sched->tls_cache);
+ /* fall through */
+ case SCHED_ALLOC_TLS_CACHE:
+ _lthread_objcache_destroy(new_sched->per_lthread_cache);
+ /* fall through */
+ case SCHED_ALLOC_PERLT_CACHE:
+ _lthread_objcache_destroy(new_sched->stack_cache);
+ /* fall through */
+ case SCHED_ALLOC_STACK_CACHE:
+ _lthread_objcache_destroy(new_sched->lthread_cache);
+ /* fall through */
+ case SCHED_ALLOC_LTHREAD_CACHE:
+ _lthread_queue_destroy(new_sched->pready);
+ /* fall through */
+ case SCHED_ALLOC_PREADY_QUEUE:
+ _lthread_queue_destroy(new_sched->ready);
+ /* fall through */
+ case SCHED_ALLOC_READY_QUEUE:
+ _qnode_pool_destroy(new_sched->qnode_pool);
+ /* fall through */
+ case SCHED_ALLOC_QNODE_POOL:
+ /* fall through */
+ case SCHED_ALLOC_OK:
+ break;
+ }
+ return alloc_status;
+}
+
+
+/*
+ * Create a scheduler on the current lcore
+ */
+struct lthread_sched *_lthread_sched_create(size_t stack_size)
+{
+ int status;
+ struct lthread_sched *new_sched;
+ unsigned lcoreid = rte_lcore_id();
+
+ RTE_ASSERT(stack_size <= LTHREAD_MAX_STACK_SIZE);
+
+ if (stack_size == 0)
+ stack_size = LTHREAD_MAX_STACK_SIZE;
+
+ new_sched =
+ rte_calloc_socket(NULL, 1, sizeof(struct lthread_sched),
+ RTE_CACHE_LINE_SIZE,
+ rte_socket_id());
+ if (new_sched == NULL) {
+ RTE_LOG(CRIT, LTHREAD,
+ "Failed to allocate memory for scheduler\n");
+ return NULL;
+ }
+
+ _lthread_key_pool_init();
+
+ new_sched->stack_size = stack_size;
+ new_sched->birth = rte_rdtsc();
+ THIS_SCHED = new_sched;
+
+ status = _lthread_sched_alloc_resources(new_sched);
+ if (status != SCHED_ALLOC_OK) {
+ RTE_LOG(CRIT, LTHREAD,
+ "Failed to allocate resources for scheduler code = %d\n",
+ status);
+ rte_free(new_sched);
+ return NULL;
+ }
+
+ bzero(&new_sched->ctx, sizeof(struct ctx));
+
+ new_sched->lcore_id = lcoreid;
+
+ schedcore[lcoreid] = new_sched;
+
+ new_sched->run_flag = 1;
+
+ DIAG_EVENT(new_sched, LT_DIAG_SCHED_CREATE, rte_lcore_id(), 0);
+
+ rte_wmb();
+ return new_sched;
+}
+
+/*
+ * Set the number of schedulers in the system
+ */
+int lthread_num_schedulers_set(int num)
+{
+ rte_atomic16_set(&num_schedulers, num);
+ return (int)rte_atomic16_read(&num_schedulers);
+}
+
+/*
+ * Return the number of schedulers active
+ */
+int lthread_active_schedulers(void)
+{
+ return (int)rte_atomic16_read(&active_schedulers);
+}
+
+
+/**
+ * shutdown the scheduler running on the specified lcore
+ */
+void lthread_scheduler_shutdown(unsigned lcoreid)
+{
+ uint64_t coreid = (uint64_t) lcoreid;
+
+ if (coreid < LTHREAD_MAX_LCORES) {
+ if (schedcore[coreid] != NULL)
+ schedcore[coreid]->run_flag = 0;
+ }
+}
+
+/**
+ * shutdown all schedulers
+ */
+void lthread_scheduler_shutdown_all(void)
+{
+ uint64_t i;
+
+ /*
+ * give time for all schedulers to have started
+ * Note we use sched_yield() rather than pthread_yield() to allow
+ * for the possibility of a pthread wrapper on lthread_yield(),
+ * something that is not possible unless the scheduler is running.
+ */
+ while (rte_atomic16_read(&active_schedulers) <
+ rte_atomic16_read(&num_schedulers))
+ sched_yield();
+
+ for (i = 0; i < LTHREAD_MAX_LCORES; i++) {
+ if (schedcore[i] != NULL)
+ schedcore[i]->run_flag = 0;
+ }
+}
+
+/*
+ * Resume a suspended lthread
+ */
+static __rte_always_inline void
+_lthread_resume(struct lthread *lt);
+static inline void _lthread_resume(struct lthread *lt)
+{
+ struct lthread_sched *sched = THIS_SCHED;
+ struct lthread_stack *s;
+ uint64_t state = lt->state;
+#if LTHREAD_DIAG
+ int init = 0;
+#endif
+
+ sched->current_lthread = lt;
+
+ if (state & (BIT(ST_LT_CANCELLED) | BIT(ST_LT_EXITED))) {
+ /* if detached we can free the thread now */
+ if (state & BIT(ST_LT_DETACH)) {
+ _lthread_free(lt);
+ sched->current_lthread = NULL;
+ return;
+ }
+ }
+
+ if (state & BIT(ST_LT_INIT)) {
+ /* first time this thread has been run */
+ /* assign thread to this scheduler */
+ lt->sched = THIS_SCHED;
+
+ /* allocate stack */
+ s = _stack_alloc();
+
+ lt->stack_container = s;
+ _lthread_set_stack(lt, s->stack, s->stack_size);
+
+ /* allocate memory for TLS used by this thread */
+ _lthread_tls_alloc(lt);
+
+ lt->state = BIT(ST_LT_READY);
+#if LTHREAD_DIAG
+ init = 1;
+#endif
+ }
+
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_RESUMED, init, lt);
+
+ /* switch to the new thread */
+ ctx_switch(&lt->ctx, &sched->ctx);
+
+ /* If posting to a queue that could be read by another lcore
+ * we defer the queue write till now to ensure the context has been
+ * saved before the other core tries to resume it
+ * This applies to blocking on mutex, cond, and to set_affinity
+ */
+ if (lt->pending_wr_queue != NULL) {
+ struct lthread_queue *dest = lt->pending_wr_queue;
+
+ lt->pending_wr_queue = NULL;
+
+ /* queue the current thread to the specified queue */
+ _lthread_queue_insert_mp(dest, lt);
+ }
+
+ sched->current_lthread = NULL;
+}
+
+/*
+ * Handle sleep timer expiry
+*/
+void
+_sched_timer_cb(struct rte_timer *tim, void *arg)
+{
+ struct lthread *lt = (struct lthread *) arg;
+ uint64_t state = lt->state;
+
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_TMR_EXPIRED, &lt->tim, 0);
+
+ rte_timer_stop(tim);
+
+ if (lt->state & BIT(ST_LT_CANCELLED))
+ (THIS_SCHED)->nb_blocked_threads--;
+
+ lt->state = state | BIT(ST_LT_EXPIRED);
+ _lthread_resume(lt);
+ lt->state = state & CLEARBIT(ST_LT_EXPIRED);
+}
+
+
+
+/*
+ * Returns 0 if there is a pending job in scheduler or 1 if done and can exit.
+ */
+static inline int _lthread_sched_isdone(struct lthread_sched *sched)
+{
+ return (sched->run_flag == 0) &&
+ (_lthread_queue_empty(sched->ready)) &&
+ (_lthread_queue_empty(sched->pready)) &&
+ (sched->nb_blocked_threads == 0);
+}
+
+/*
+ * Wait for all schedulers to start
+ */
+static inline void _lthread_schedulers_sync_start(void)
+{
+ rte_atomic16_inc(&active_schedulers);
+
+ /* wait for lthread schedulers
+ * Note we use sched_yield() rather than pthread_yield() to allow
+ * for the possibility of a pthread wrapper on lthread_yield(),
+ * something that is not possible unless the scheduler is running.
+ */
+ while (rte_atomic16_read(&active_schedulers) <
+ rte_atomic16_read(&num_schedulers))
+ sched_yield();
+
+}
+
+/*
+ * Wait for all schedulers to stop
+ */
+static inline void _lthread_schedulers_sync_stop(void)
+{
+ rte_atomic16_dec(&active_schedulers);
+ rte_atomic16_dec(&num_schedulers);
+
+ /* wait for schedulers
+ * Note we use sched_yield() rather than pthread_yield() to allow
+ * for the possibility of a pthread wrapper on lthread_yield(),
+ * something that is not possible unless the scheduler is running.
+ */
+ while (rte_atomic16_read(&active_schedulers) > 0)
+ sched_yield();
+
+}
+
+
+/*
+ * Run the lthread scheduler
+ * This loop is the heart of the system
+ */
+void lthread_run(void)
+{
+
+ struct lthread_sched *sched = THIS_SCHED;
+ struct lthread *lt = NULL;
+
+ RTE_LOG(INFO, LTHREAD,
+ "starting scheduler %p on lcore %u phys core %u\n",
+ sched, rte_lcore_id(),
+ rte_lcore_index(rte_lcore_id()));
+
+ /* if more than one, wait for all schedulers to start */
+ _lthread_schedulers_sync_start();
+
+
+ /*
+ * This is the main scheduling loop
+ * So long as there are tasks in existence we run this loop.
+ * We check for:-
+ * expired timers,
+ * the local ready queue,
+ * and the peer ready queue,
+ *
+ * and resume lthreads ad infinitum.
+ */
+ while (!_lthread_sched_isdone(sched)) {
+
+ rte_timer_manage();
+
+ lt = _lthread_queue_poll(sched->ready);
+ if (lt != NULL)
+ _lthread_resume(lt);
+ lt = _lthread_queue_poll(sched->pready);
+ if (lt != NULL)
+ _lthread_resume(lt);
+ }
+
+
+ /* if more than one wait for all schedulers to stop */
+ _lthread_schedulers_sync_stop();
+
+ (THIS_SCHED) = NULL;
+
+ RTE_LOG(INFO, LTHREAD,
+ "stopping scheduler %p on lcore %u phys core %u\n",
+ sched, rte_lcore_id(),
+ rte_lcore_index(rte_lcore_id()));
+ fflush(stdout);
+}
+
+/*
+ * Return the scheduler for this lcore
+ *
+ */
+struct lthread_sched *_lthread_sched_get(unsigned int lcore_id)
+{
+ struct lthread_sched *res = NULL;
+
+ if (lcore_id < LTHREAD_MAX_LCORES)
+ res = schedcore[lcore_id];
+
+ return res;
+}
+
+/*
+ * migrate the current thread to another scheduler running
+ * on the specified lcore.
+ */
+int lthread_set_affinity(unsigned lcoreid)
+{
+ struct lthread *lt = THIS_LTHREAD;
+ struct lthread_sched *dest_sched;
+
+ if (unlikely(lcoreid >= LTHREAD_MAX_LCORES))
+ return POSIX_ERRNO(EINVAL);
+
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_AFFINITY, lcoreid, 0);
+
+ dest_sched = schedcore[lcoreid];
+
+ if (unlikely(dest_sched == NULL))
+ return POSIX_ERRNO(EINVAL);
+
+ if (likely(dest_sched != THIS_SCHED)) {
+ lt->sched = dest_sched;
+ lt->pending_wr_queue = dest_sched->pready;
+ _affinitize();
+ return 0;
+ }
+ return 0;
+}
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_sched.h b/src/spdk/dpdk/examples/performance-thread/common/lthread_sched.h
new file mode 100644
index 000000000..d14bec1c8
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_sched.h
@@ -0,0 +1,104 @@
+/*
+ * SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2015 Intel Corporation.
+ * Copyright 2012 Hasan Alayli <halayli@gmail.com>
+ */
+
+#ifndef LTHREAD_SCHED_H_
+#define LTHREAD_SCHED_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "lthread_int.h"
+#include "lthread_queue.h"
+#include "lthread_objcache.h"
+#include "lthread_diag.h"
+#include "ctx.h"
+
+/*
+ * insert an lthread into a queue
+ */
+static inline void
+_ready_queue_insert(struct lthread_sched *sched, struct lthread *lt)
+{
+ if (sched == THIS_SCHED)
+ _lthread_queue_insert_sp((THIS_SCHED)->ready, lt);
+ else
+ _lthread_queue_insert_mp(sched->pready, lt);
+}
+
+/*
+ * remove an lthread from a queue
+ */
+static inline struct lthread *_ready_queue_remove(struct lthread_queue *q)
+{
+ return _lthread_queue_remove(q);
+}
+
+/**
+ * Return true if the ready queue is empty
+ */
+static inline int _ready_queue_empty(struct lthread_queue *q)
+{
+ return _lthread_queue_empty(q);
+}
+
+static inline uint64_t _sched_now(void)
+{
+ uint64_t now = rte_rdtsc();
+
+ if (now > (THIS_SCHED)->birth)
+ return now - (THIS_SCHED)->birth;
+ if (now < (THIS_SCHED)->birth)
+ return (THIS_SCHED)->birth - now;
+ /* never return 0 because this means sleep forever */
+ return 1;
+}
+
+static __rte_always_inline void
+_affinitize(void);
+static inline void
+_affinitize(void)
+{
+ struct lthread *lt = THIS_LTHREAD;
+
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_SUSPENDED, 0, 0);
+ ctx_switch(&(THIS_SCHED)->ctx, &lt->ctx);
+}
+
+static __rte_always_inline void
+_suspend(void);
+static inline void
+_suspend(void)
+{
+ struct lthread *lt = THIS_LTHREAD;
+
+ (THIS_SCHED)->nb_blocked_threads++;
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_SUSPENDED, 0, 0);
+ ctx_switch(&(THIS_SCHED)->ctx, &lt->ctx);
+ (THIS_SCHED)->nb_blocked_threads--;
+}
+
+static __rte_always_inline void
+_reschedule(void);
+static inline void
+_reschedule(void)
+{
+ struct lthread *lt = THIS_LTHREAD;
+
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_RESCHEDULED, 0, 0);
+ _ready_queue_insert(THIS_SCHED, lt);
+ ctx_switch(&(THIS_SCHED)->ctx, &lt->ctx);
+}
+
+extern struct lthread_sched *schedcore[];
+void _sched_timer_cb(struct rte_timer *tim, void *arg);
+void _sched_shutdown(__rte_unused void *arg);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* LTHREAD_SCHED_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_timer.h b/src/spdk/dpdk/examples/performance-thread/common/lthread_timer.h
new file mode 100644
index 000000000..f2d8671a4
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_timer.h
@@ -0,0 +1,68 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2015 Intel Corporation
+ */
+
+
+#ifndef LTHREAD_TIMER_H_
+#define LTHREAD_TIMER_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "lthread_int.h"
+#include "lthread_sched.h"
+
+
+static inline uint64_t
+_ns_to_clks(uint64_t ns)
+{
+ /*
+ * clkns needs to be divided by 1E9 to get ns clocks. However,
+ * dividing by this first would lose a lot of accuracy.
+ * Dividing after a multiply by ns, could cause overflow of
+ * uint64_t if ns is about 5 seconds [if we assume a max tsc
+ * rate of 4GHz]. Therefore we first divide by 1E4, then
+ * multiply and finally divide by 1E5. This allows ns to be
+ * values many hours long, without overflow, while still keeping
+ * reasonable accuracy.
+ */
+ uint64_t clkns = rte_get_tsc_hz() / 1e4;
+
+ clkns *= ns;
+ clkns /= 1e5;
+
+ return clkns;
+}
+
+
+static inline void
+_timer_start(struct lthread *lt, uint64_t clks)
+{
+ if (clks > 0) {
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_TMR_START, &lt->tim, clks);
+ rte_timer_init(&lt->tim);
+ rte_timer_reset(&lt->tim,
+ clks,
+ SINGLE,
+ rte_lcore_id(),
+ _sched_timer_cb,
+ (void *)lt);
+ }
+}
+
+
+static inline void
+_timer_stop(struct lthread *lt)
+{
+ if (lt != NULL) {
+ DIAG_EVENT(lt, LT_DIAG_LTHREAD_TMR_DELETE, &lt->tim, 0);
+ rte_timer_stop(&lt->tim);
+ }
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* LTHREAD_TIMER_H_ */
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_tls.c b/src/spdk/dpdk/examples/performance-thread/common/lthread_tls.c
new file mode 100644
index 000000000..07de6cafa
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_tls.c
@@ -0,0 +1,222 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2015 Intel Corporation
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdint.h>
+#include <limits.h>
+#include <inttypes.h>
+#include <unistd.h>
+#include <pthread.h>
+#include <fcntl.h>
+#include <sys/time.h>
+#include <sys/mman.h>
+#include <sched.h>
+
+#include <rte_malloc.h>
+#include <rte_log.h>
+#include <rte_ring.h>
+#include <rte_atomic_64.h>
+
+#include "lthread_tls.h"
+#include "lthread_queue.h"
+#include "lthread_objcache.h"
+#include "lthread_sched.h"
+
+static struct rte_ring *key_pool;
+static uint64_t key_pool_init;
+
+/* needed to cause section start and end to be defined */
+RTE_DEFINE_PER_LTHREAD(void *, dummy);
+
+static struct lthread_key key_table[LTHREAD_MAX_KEYS];
+
+RTE_INIT(thread_tls_ctor)
+{
+ key_pool = NULL;
+ key_pool_init = 0;
+}
+
+/*
+ * Initialize a pool of keys
+ * These are unique tokens that can be obtained by threads
+ * calling lthread_key_create()
+ */
+void _lthread_key_pool_init(void)
+{
+ static struct rte_ring *pool;
+ struct lthread_key *new_key;
+ char name[MAX_LTHREAD_NAME_SIZE];
+
+ bzero(key_table, sizeof(key_table));
+
+ /* only one lcore should do this */
+ if (rte_atomic64_cmpset(&key_pool_init, 0, 1)) {
+
+ snprintf(name,
+ MAX_LTHREAD_NAME_SIZE,
+ "lthread_key_pool_%d",
+ getpid());
+
+ pool = rte_ring_create(name,
+ LTHREAD_MAX_KEYS, 0, 0);
+ RTE_ASSERT(pool);
+
+ int i;
+
+ for (i = 1; i < LTHREAD_MAX_KEYS; i++) {
+ new_key = &key_table[i];
+ rte_ring_mp_enqueue((struct rte_ring *)pool,
+ (void *)new_key);
+ }
+ key_pool = pool;
+ }
+ /* other lcores wait here till done */
+ while (key_pool == NULL) {
+ rte_compiler_barrier();
+ sched_yield();
+ };
+}
+
+/*
+ * Create a key
+ * this means getting a key from the pool
+ */
+int lthread_key_create(unsigned int *key, tls_destructor_func destructor)
+{
+ if (key == NULL)
+ return POSIX_ERRNO(EINVAL);
+
+ struct lthread_key *new_key;
+
+ if (rte_ring_mc_dequeue((struct rte_ring *)key_pool, (void **)&new_key)
+ == 0) {
+ new_key->destructor = destructor;
+ *key = (new_key - key_table);
+
+ return 0;
+ }
+ return POSIX_ERRNO(EAGAIN);
+}
+
+
+/*
+ * Delete a key
+ */
+int lthread_key_delete(unsigned int k)
+{
+ struct lthread_key *key;
+
+ key = (struct lthread_key *) &key_table[k];
+
+ if (k > LTHREAD_MAX_KEYS)
+ return POSIX_ERRNO(EINVAL);
+
+ key->destructor = NULL;
+ rte_ring_mp_enqueue((struct rte_ring *)key_pool,
+ (void *)key);
+ return 0;
+}
+
+
+
+/*
+ * Break association for all keys in use by this thread
+ * invoke the destructor if available.
+ * Since a destructor can create keys we could enter an infinite loop
+ * therefore we give up after LTHREAD_DESTRUCTOR_ITERATIONS
+ * the behavior is modelled on pthread
+ */
+void _lthread_tls_destroy(struct lthread *lt)
+{
+ int i, k;
+ int nb_keys;
+ void *data;
+
+ for (i = 0; i < LTHREAD_DESTRUCTOR_ITERATIONS; i++) {
+
+ for (k = 1; k < LTHREAD_MAX_KEYS; k++) {
+
+ /* no keys in use ? */
+ nb_keys = lt->tls->nb_keys_inuse;
+ if (nb_keys == 0)
+ return;
+
+ /* this key not in use ? */
+ if (lt->tls->data[k] == NULL)
+ continue;
+
+ /* remove this key */
+ data = lt->tls->data[k];
+ lt->tls->data[k] = NULL;
+ lt->tls->nb_keys_inuse = nb_keys-1;
+
+ /* invoke destructor */
+ if (key_table[k].destructor != NULL)
+ key_table[k].destructor(data);
+ }
+ }
+}
+
+/*
+ * Return the pointer associated with a key
+ * If the key is no longer valid return NULL
+ */
+void
+*lthread_getspecific(unsigned int k)
+{
+ void *res = NULL;
+
+ if (k < LTHREAD_MAX_KEYS)
+ res = THIS_LTHREAD->tls->data[k];
+
+ return res;
+}
+
+/*
+ * Set a value against a key
+ * If the key is no longer valid return an error
+ * when storing value
+ */
+int lthread_setspecific(unsigned int k, const void *data)
+{
+ if (k >= LTHREAD_MAX_KEYS)
+ return POSIX_ERRNO(EINVAL);
+
+ int n = THIS_LTHREAD->tls->nb_keys_inuse;
+
+ /* discard const qualifier */
+ char *p = (char *) (uintptr_t) data;
+
+
+ if (data != NULL) {
+ if (THIS_LTHREAD->tls->data[k] == NULL)
+ THIS_LTHREAD->tls->nb_keys_inuse = n+1;
+ }
+
+ THIS_LTHREAD->tls->data[k] = (void *) p;
+ return 0;
+}
+
+/*
+ * Allocate data for TLS cache
+*/
+void _lthread_tls_alloc(struct lthread *lt)
+{
+ struct lthread_tls *tls;
+
+ tls = _lthread_objcache_alloc((THIS_SCHED)->tls_cache);
+
+ RTE_ASSERT(tls != NULL);
+
+ tls->root_sched = (THIS_SCHED);
+ lt->tls = tls;
+
+ /* allocate data for TLS varaiables using RTE_PER_LTHREAD macros */
+ if (sizeof(void *) < (uint64_t)RTE_PER_LTHREAD_SECTION_SIZE) {
+ lt->per_lthread_data =
+ _lthread_objcache_alloc((THIS_SCHED)->per_lthread_cache);
+ }
+}
diff --git a/src/spdk/dpdk/examples/performance-thread/common/lthread_tls.h b/src/spdk/dpdk/examples/performance-thread/common/lthread_tls.h
new file mode 100644
index 000000000..4c262e98b
--- /dev/null
+++ b/src/spdk/dpdk/examples/performance-thread/common/lthread_tls.h
@@ -0,0 +1,35 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2015 Intel Corporation
+ */
+
+#ifndef LTHREAD_TLS_H_
+#define LTHREAD_TLS_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "lthread_api.h"
+
+#define RTE_PER_LTHREAD_SECTION_SIZE \
+(&__stop_per_lt - &__start_per_lt)
+
+struct lthread_key {
+ tls_destructor_func destructor;
+};
+
+struct lthread_tls {
+ void *data[LTHREAD_MAX_KEYS];
+ int nb_keys_inuse;
+ struct lthread_sched *root_sched;
+};
+
+void _lthread_tls_destroy(struct lthread *lt);
+void _lthread_key_pool_init(void);
+void _lthread_tls_alloc(struct lthread *lt);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* LTHREAD_TLS_H_ */