Add 11th lab

README.md

@@ -10,11 +10,14 @@
 - [Лабораторная работа 7](lab7/README.md)
 - [Лабораторная работа 9](lab9/README.md)
 - [Лабораторная работа 10](lab10/README.md)
+- [Лабораторная работа 11](lab11/README.md)
 
 ## Запуск
 
 Требуется CMake 3.16+ (более ранние версии не проверялись) и компилятор с поддержкой C11. 
 
+Для лабораторной №11 требуются make и gcc.
+
 ```bash
 ./run_lab.sh <lab_directory>
 ```

lab11/Makefile

@@ -0,0 +1,24 @@
+
+ALL = vector-deadlock vector-global-order vector-try-wait vector-avoid-hold-and-wait vector-nolock
+COMMON = vector-header.h main-common.c main-header.h
+
+all: $(ALL)
+
+clean:
+	rm -f $(ALL) *~
+
+vector-deadlock: vector-deadlock.c $(COMMON)
+	gcc -o vector-deadlock vector-deadlock.c -Wall -pthread -O
+
+vector-global-order: vector-global-order.c $(COMMON)
+	gcc -o vector-global-order vector-global-order.c -Wall -pthread -O
+
+vector-try-wait: vector-try-wait.c $(COMMON)
+	gcc -o vector-try-wait vector-try-wait.c -Wall -pthread -O
+
+vector-avoid-hold-and-wait: vector-avoid-hold-and-wait.c $(COMMON)
+	gcc -o vector-avoid-hold-and-wait vector-avoid-hold-and-wait.c -Wall -pthread -O
+
+vector-nolock: vector-nolock.c $(COMMON)
+	gcc -o vector-nolock vector-nolock.c -Wall -pthread -O
+

lab11/README.md

@@ -0,0 +1,65 @@
+# Лабораторная работа №11
+
+## Задание 1
+
+Программа печатает в каждом потоке начальные данные перед `vector_add` и результат выполнения данной функции.
+Результат может меняться от вызова к вызову, так как весь worker не покрыт мьютексами, но на практике на такой небольшой
+программе это маловероятно достижимо.
+
+При добавлении `-d` взаимная блокировка возникает не всегда, а только при попадании переключения потоков между мьютексами.
+
+В случае, если число потоков = 1, взаимной блокировки не возникнет.
+
+## Задание 2
+
+Программа избегает мёртвой блокировки за счёт упорядочивания по адресам, что позволяет постоянно сохранять порядок 
+блокировки.
+
+В случае если адреса совпадают, то это один мьютекс, и для корректной работы программы его надо блокировать 1 раз.
+
+В случае увеличения числа циклов и потоков, время выполнения растёт.
+
+В случае включения `-p` время уменьшается, так как разрешается параллелизм.
+
+## Задание 3
+
+Вызовы pthread_mutex_trylock необходимы для создания порядка блокировки, для того чтобы избежать дедлока.
+
+С увеличением числа потоков происходит рост повторных попыток, что является логичным, так как переключение между потоками 
+становится более частым.
+
+## Задание 4
+
+Данный подход защищает уязвимое место дедлока, созданием глобального мьютекса, но при этом не даёт различным векторам 
+выполняться параллельно.
+
+При использовании `-p` время уменьшается.
+
+## Задание 5
+
+Указав memory, мы дожидаемся завершения всех операцией с памятью, что своего рода позволяет заменить мьютексы. 
+(https://ru.wikipedia.org/wiki/GCC_Inline_Assembly)
+
+Сравним время выполнения следующих команд:
+
+```text
+./vector-nolock -t -n 2 -l 1000000 -d = 4.08
+./vector-nolock -t -n 2 -l 1000000 -d -p = 0.65
+```
+
+```text
+./vector-avoid-hold-and-wait -t -n 2 -l 1000000 -d = 2.98
+./vector-avoid-hold-and-wait -t -n 2 -l 1000000 -d -p = 0.45
+```
+
+```text
+./vector-try-wait -t -n 2 -l 1000000 -d = 1.30
+./vector-try-wait -t -n 2 -l 1000000 -d -p = 0.18
+```
+
+```text
+./vector-global-order -t -n 2 -l 1000000 -d = 0.69
+./vector-global-order -t -n 2 -l 1000000 -d -p = 0.19
+```
+
+Таким образом видно, что vector-nolock работает медленнее других в любом случае.

lab11/README_FROM_SRC

@@ -0,0 +1,89 @@
+
+This homework lets you play around with a number of ways to implement a small,
+deadlock-free vector object in C. The vector object is quite limited (e.g., it
+only has add() and init() functions) but is just used to illustrate different
+approaches to avoiding deadlock. 
+
+Some files that you should pay attention to are as follows. They, in
+particular, are used by all the variants in this homework.
+
+- mythreads.h
+  The usual wrappers around many different pthread (and other) library calls,
+  so as to ensure they are not failing silently
+
+- vector-header.h
+  A simple header for the vector routines, mostly defining a fixed vector size
+  and then a struct that is used per vector (vector_t)
+
+- main-header.h
+  A number of global variables common to each different program
+
+- main-common.c
+  Contains the main() routine (with arg parsing) that initializes two vectors,
+  starts some threads to access them (via a worker() routine), and then waits
+  for the many vector_add()'s to complete
+
+The variants of this homework are found in the following files. Each takes a
+different approach to dealing with concurrency inside a "vector addition"
+routine called vector_add(); examine the code in these files to get a sense of
+what is going on. They all use the files above to make a complete runnable
+program. 
+
+- vector-deadlock.c
+  This version blithely grabs the locks in a particular order (dst then
+  src). By doing so, it creates an "invitation to deadlock", as one thread
+  might call vector_add(v1, v2) while another concurrently calls
+  vector_add(v2, v1).
+
+- vector-global-order.c
+  This version of vector_add() grabs the locks in a total order, based on
+  address of the vector. 
+
+- vector-try-wait.c
+  This version of vector_add() uses pthread_mutex_trylock() to attempt to grab
+  locks; when the try fails, the code releases any locks it may hold and goes
+  back to the top and tries it all over again.
+
+- vector-avoid-hold-and-wait.c
+  This version of vector_add() ensures it can't get stuck in a hold and wait
+  pattern by using a single lock around lock acquisition.
+
+- vector-nolock.c
+  This version of vector_add() doesn't even use locks; rather, it uses an
+  atomic fetch-and-add to implement the vector_add() routine. Its semantics
+  (as a result) are slightly different.
+
+Type "make" (and read the Makefile) to build each of five executables. 
+
+  prompt> make
+
+Then you can run a program by simply typing its name:
+
+  prompt> ./vector-deadlock
+
+Each program takes the same set of arguments (see main-common.c for details):
+
+-d                 
+   This flag turns on the ability for threads to deadlock.
+   When you pass -d to the program, every other thread calls vector_add()
+   with the vectors in a different order, e.g., with two threads, and -d enabled, 
+   Thread 0 calls vector_add(v1, v2) and Thread 1 calls vector_add(v2, v1)
+
+-p
+   This flag gives each thread a different set of vectors to call add()
+   upon, instead of just two vectors. Use this to see how things perform
+   when there isn't contention for the same set of vectors.
+
+-n num_threads
+   Creates some number of threads; you need more than one to deadlock.
+
+-l loops
+   How many times should each thread call vector_add()?
+
+-v 
+   Verbose flag: prints out a little more about what is going on.
+
+-t
+   Turns on timing and shows how long everything took.
+
+

lab11/main-common.c

@@ -0,0 +1,139 @@
+
+vector_t v[2 * MAX_THREADS];
+
+// used to ensure print outs make sense
+pthread_mutex_t print_lock = PTHREAD_MUTEX_INITIALIZER;
+
+void usage(char *prog) {
+    fprintf(stderr, "usage: %s [-d (turn on deadlocking behavior)] [-l loops] [-n num_threads] [-t (do timing)] [-v (for verbose)]\n", prog);
+    exit(1);
+}
+
+// only called by one thread (not thread-safe)
+void vector_init(vector_t *v, int value) {
+    int i;
+    for (i = 0; i < VECTOR_SIZE; i++) {
+	v->values[i] = value;
+    }
+    Pthread_mutex_init(&v->lock, NULL); 
+}
+
+// only called by one thread (not thread-safe)
+void vector_print(vector_t *v, char *str) {
+    int i;
+    for (i = 0; i < VECTOR_SIZE; i++) {
+	printf("%s[%d] %d\n", str, i, v->values[i]);
+    }
+}
+
+void print_info(int call_return, int thread_id, int v0, int v1) {
+    if (verbose == 0)
+	return;
+    Pthread_mutex_lock(&print_lock);
+    int j;
+    for (j = 0; j < thread_id; j++) printf("              ");
+    if (call_return) 
+	printf("<-add(%d, %d)\n", v0, v1);
+    else
+	printf("->add(%d, %d)\n", v0, v1);
+    Pthread_mutex_unlock(&print_lock);
+}
+
+typedef struct __thread_arg_t {
+    int tid;
+    int vector_add_order;
+    int vector_0;
+    int vector_1;
+} thread_arg_t;
+
+void *worker(void *arg) {
+    thread_arg_t *args = (thread_arg_t *) arg;
+    int i, v0, v1;
+    for (i = 0; i < loops; i++) {
+	if (args->vector_add_order == 0) { 
+	    v0 = args->vector_0;
+	    v1 = args->vector_1;
+	} else {
+	    v0 = args->vector_1;
+	    v1 = args->vector_0;
+	}
+
+	print_info(0, args->tid, v0, v1);
+
+	vector_add(&v[v0], &v[v1]);
+
+	print_info(1, args->tid, v0, v1);
+    }
+    return NULL;
+}
+
+int main(int argc, char *argv[]) {
+    opterr = 0;
+    int c;
+    while ((c = getopt (argc, argv, "l:n:vtdp")) != -1) {
+	switch (c) {
+	case 'l':
+	    loops = atoi(optarg);
+	    break;
+	case 'n':
+	    num_threads = atoi(optarg);
+	    break;
+	case 'v':
+	    verbose = 1;
+	    break;
+	case 't':
+	    do_timing = 1;
+	    break;
+	case 'd':
+	    cause_deadlock = 1;
+	    break;
+	case 'p':
+	    enable_parallelism = 1;
+	    break;
+	default:
+	    usage(argv[0]);
+	}
+    }
+
+    assert(num_threads < MAX_THREADS);
+
+    pthread_t pid[MAX_THREADS];
+    thread_arg_t args[MAX_THREADS];
+    int i;
+    for (i = 0; i < num_threads; i++) {
+	args[i].tid = i;
+	if (enable_parallelism == 0) {
+	    args[i].vector_0 = 0;
+	    args[i].vector_1 = 1;
+	} else {
+	    args[i].vector_0 = i * 2;
+	    args[i].vector_1 = i * 2 + 1;
+	}
+
+	if (cause_deadlock && i % 2 == 1)
+	    args[i].vector_add_order = 1;
+	else
+	    args[i].vector_add_order = 0;
+    }
+
+    for (i = 0; i < 2 * MAX_THREADS; i++) 
+	vector_init(&v[i], i);
+
+    double t1 = Time_GetSeconds();
+
+    for (i = 0; i < num_threads; i++) 
+	Pthread_create(&pid[i], NULL, worker, (void *) &args[i]);
+    for (i = 0; i < num_threads; i++) 
+	Pthread_join(pid[i], NULL); 
+
+    double t2 = Time_GetSeconds();
+
+    fini();
+    if (do_timing) {
+	printf("Time: %.2f seconds\n", t2 - t1);
+    }
+
+    //vector_print(&v[0], "v1");
+    //vector_print(&v[1], "v2");
+    return 0;
+}

lab11/main-header.h

@@ -0,0 +1,13 @@
+#ifndef __main_header_h__
+#define __main_header_h__
+
+#define MAX_THREADS (100)
+
+int loops = 1;
+int verbose = 0;
+int num_threads = 2;
+int do_timing = 0;
+int cause_deadlock = 0;
+int enable_parallelism = 0;
+
+#endif // __main_header_h__

lab11/mythreads.h

@@ -0,0 +1,52 @@
+#ifndef __MYTHREADS_h__
+#define __MYTHREADS_h__
+
+#include <pthread.h>
+#include <assert.h>
+#include <stdlib.h>
+#include <sys/time.h>
+
+void *Malloc(size_t size) {
+    void *rc = malloc(size);
+    assert(rc != NULL);
+    return rc;
+}
+
+double Time_GetSeconds() {
+    struct timeval t;
+    int rc = gettimeofday(&t, NULL);
+    assert(rc == 0);
+    return (double) ((double)t.tv_sec + (double)t.tv_usec / 1e6);
+}
+
+void Pthread_mutex_init(pthread_mutex_t *mutex,
+			const pthread_mutexattr_t *attr) {
+    int rc = pthread_mutex_init(mutex, attr);
+    assert(rc == 0);
+}
+
+
+void Pthread_mutex_lock(pthread_mutex_t *m) {
+    int rc = pthread_mutex_lock(m);
+    assert(rc == 0);
+}
+                                                                                
+void Pthread_mutex_unlock(pthread_mutex_t *m) {
+    int rc = pthread_mutex_unlock(m);
+    assert(rc == 0);
+}
+                                                                                
+void Pthread_create(pthread_t *thread, const pthread_attr_t *attr, 	 
+		    void *(*start_routine)(void*), void *arg) {
+    int rc = pthread_create(thread, attr, start_routine, arg);
+    assert(rc == 0);
+}
+
+void Pthread_join(pthread_t thread, void **value_ptr) {
+    int rc = pthread_join(thread, value_ptr);
+    assert(rc == 0);
+}
+
+
+
+#endif // __MYTHREADS_h__

lab11/vector-avoid-hold-and-wait.c

@@ -0,0 +1,32 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <unistd.h>
+
+#include "mythreads.h"
+
+#include "main-header.h"
+#include "vector-header.h"
+
+// use this to make lock acquisition ATOMIC
+pthread_mutex_t global = PTHREAD_MUTEX_INITIALIZER; 
+
+void vector_add(vector_t *v_dst, vector_t *v_src) {
+    // put GLOBAL lock around all lock acquisition...
+    Pthread_mutex_lock(&global);
+    Pthread_mutex_lock(&v_dst->lock);
+    Pthread_mutex_lock(&v_src->lock);
+    Pthread_mutex_unlock(&global);
+    int i;
+    for (i = 0; i < VECTOR_SIZE; i++) {
+	v_dst->values[i] = v_dst->values[i] + v_src->values[i];
+    }
+    Pthread_mutex_unlock(&v_dst->lock);
+    Pthread_mutex_unlock(&v_src->lock);
+}
+
+void fini() {}
+
+#include "main-common.c"
+

lab11/vector-deadlock.c

@@ -0,0 +1,24 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#include "mythreads.h"
+
+#include "main-header.h"
+#include "vector-header.h"
+
+void vector_add(vector_t *v_dst, vector_t *v_src) {
+    Pthread_mutex_lock(&v_dst->lock);
+    Pthread_mutex_lock(&v_src->lock);
+    int i;
+    for (i = 0; i < VECTOR_SIZE; i++) {
+	v_dst->values[i] = v_dst->values[i] + v_src->values[i];
+    }
+    Pthread_mutex_unlock(&v_dst->lock);
+    Pthread_mutex_unlock(&v_src->lock);
+}
+
+void fini() {}
+
+#include "main-common.c"

lab11/vector-global-order.c

@@ -0,0 +1,35 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#include "mythreads.h"
+
+#include "main-header.h"
+#include "vector-header.h"
+
+void vector_add(vector_t *v_dst, vector_t *v_src) {
+    if (v_dst < v_src) {
+	Pthread_mutex_lock(&v_dst->lock);
+	Pthread_mutex_lock(&v_src->lock);
+    } else if (v_dst > v_src) {
+	Pthread_mutex_lock(&v_src->lock);
+	Pthread_mutex_lock(&v_dst->lock);
+    } else {
+	// special case: src and dst are the same
+	Pthread_mutex_lock(&v_src->lock);
+    }
+    int i;
+    for (i = 0; i < VECTOR_SIZE; i++) {
+	v_dst->values[i] = v_dst->values[i] + v_src->values[i];
+    }
+    Pthread_mutex_unlock(&v_src->lock);
+    if (v_dst != v_src) 
+	Pthread_mutex_unlock(&v_dst->lock);
+}
+
+void fini() {}
+
+
+#include "main-common.c"
+

lab11/vector-header.h

@@ -0,0 +1,13 @@
+#ifndef __vector_header_h__
+#define __vector_header_h__
+
+#define VECTOR_SIZE (100)
+
+typedef struct __vector {
+    pthread_mutex_t lock;
+    int values[VECTOR_SIZE];
+} vector_t;
+
+
+#endif // __vector_header_h__
+

lab11/vector-nolock.c

@@ -0,0 +1,31 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#include "mythreads.h"
+
+#include "main-header.h"
+#include "vector-header.h"
+
+// taken from https://en.wikipedia.org/wiki/Fetch-and-add
+int fetch_and_add(int * variable, int value) {
+    asm volatile("lock; xaddl %%eax, %2;"
+		 :"=a" (value)                  
+		 :"a" (value), "m" (*variable)  
+		 :"memory");
+    return value;
+}
+
+void vector_add(vector_t *v_dst, vector_t *v_src) {
+    int i;
+    for (i = 0; i < VECTOR_SIZE; i++) {
+	fetch_and_add(&v_dst->values[i], v_src->values[i]);
+    }
+}
+
+void fini() {}
+
+
+#include "main-common.c"
+

lab11/vector-try-wait.c

@@ -0,0 +1,36 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#include "mythreads.h"
+
+#include "main-header.h"
+#include "vector-header.h"
+
+int retry = 0;
+
+void vector_add(vector_t *v_dst, vector_t *v_src) {
+  top:
+    if (pthread_mutex_trylock(&v_dst->lock) != 0) {
+	goto top;
+    }
+    if (pthread_mutex_trylock(&v_src->lock) != 0) {
+	retry++;
+	Pthread_mutex_unlock(&v_dst->lock);
+	goto top;
+    }
+    int i;
+    for (i = 0; i < VECTOR_SIZE; i++) {
+	v_dst->values[i] = v_dst->values[i] + v_src->values[i];
+    }
+    Pthread_mutex_unlock(&v_dst->lock);
+    Pthread_mutex_unlock(&v_src->lock);
+}
+
+void fini() {
+    printf("Retries: %d\n", retry);
+}
+
+#include "main-common.c"
+