rtm test

artificial_evaluation/7-rtm/test_rtm.sh

@@ -0,0 +1,37 @@
+#!/bin/bash
+
+source ../config.sh
+logdir=$logbasedir/rtm/
+loop=(0)
+threads=(32)
+modes=(non log rtmlog)
+size_types=(0 1)
+
+mkdir -p $logdir
+
+function run {
+  cd $basedir
+  ./quick-build.sh
+  for thread in ${threads[@]}
+  do
+      for size_type in ${size_types[@]}
+      do
+          # baseline
+          mkdir -p $logdir/$thread/$1
+          $appdir/rtm.exp $size_type 2>&1 | tee $logdir/$thread/$1/$size_type.out
+          sleep 30
+      done
+  done
+}
+
+run rtmlog
+
+sed -i 's/^\(#define USE_TSX\)/\/\/ \1/' $rtmconfig
+run log
+
+sed -i 's/^\(#define USE_LOG\)/\/\/ \1/' $rtmconfig
+run non
+
+sed -i '/^\/\/ #define USE_TSX$/s/^\/\/ //' $rtmconfig
+sed -i '/^\/\/ #define USE_LOG$/s/^\/\/ //' $rtmconfig
+

artificial_evaluation/applications/rtm.exp

@@ -0,0 +1,29 @@
+#!/usr/bin/expect  
+
+set timeout 30
+set size_type [lindex $argv 0]
+
+# start
+spawn ./build/simulate.sh
+expect "Welcome to ChCore shell!"
+# run anything you like here
+# interact
+send -- "rtm_test.bin $size_type &\r"
+
+while {1} {
+  expect {
+    "rtm test finished" {
+      puts "success\n"
+      exit
+    }
+    timeout {
+      puts "failed\n"
+      exit
+    }
+  }
+}
+
+expect ""
+
+interact
+

artificial_evaluation/config.sh

@@ -4,8 +4,9 @@
 # test_mode="IPMI"
 test_mode="QEMU"
 
-basedir="/home/xxx/treesls"
+basedir="/home/yjs/treesls"
 aedir="$basedir/artificial_evaluation"
 logbasedir="$aedir/logs/$test_mode"
 appdir="$aedir/applications"
 kconfig="$basedir/kernel/sls_config.cmake"
+rtmconfig="$basedir/kernel/include/arch/x86_64/plat/intel/immintrin.h"

kernel/arch/x86_64/boot/CMakeLists.txt

@@ -33,5 +33,5 @@ chcore_generate_emulate_sh(
     "--enable-kvm -machine pc,nvdimm=on -m 16G,slots=2,maxmem=64G \
     -object memory-backend-file,id=mem1,share=on,mem-path=\$nvm_backend_file,size=16G \
     -device nvdimm,id=nvdimm1,memdev=mem1 \
-    -cpu host -smp 20 -serial mon:stdio -nographic -cdrom \$basedir/chcore.iso"
+    -cpu host -smp 64 -serial mon:stdio -nographic -cdrom \$basedir/chcore.iso"
 )

kernel/include/arch/x86_64/plat/intel/immintrin.h

@@ -1,8 +1,8 @@
 #pragma once
 
-#define USE_TSX
+// #define USE_TSX
+// #define USE_LOG
 // #define PRE_TOUCH
-// #define PRINT_ABORT_COUNT
 
 #ifndef __RTM__
 #pragma GCC push_options

kernel/include/arch/x86_64/plat/intel/machine.h

@@ -1,7 +1,7 @@
 #pragma once
 
 /* Should be set at boot time */
-#define PLAT_CPU_NUM    10
+#define PLAT_CPU_NUM    32
 
 /* MSR Registers */
 #define IA32_APIC_BASE		0x0000001b

kernel/mm/buddy.c

@@ -198,10 +198,12 @@ struct page *buddy_get_pages(struct phys_mem_pool *pool, int order)
                 if (!list_empty(free_list)) {
                         /* Get a free memory chunck from the free list */
                         page = list_entry(free_list->next, struct page, node);
+                        #ifdef USE_LOG
                         prepare_latest_log(pool,
                                            ADD_PAGES,
                                            (u64)page,
                                            order, cur_order);
+                        #endif
                         list_del(&page->node);
                         pool->free_lists[cur_order].nr_free -= 1;
                         break;
@@ -249,10 +251,12 @@ struct page *buddy_get_pages(struct phys_mem_pool *pool, int order)
                         goto redo;
                 }
         } else {
+                #ifdef USE_LOG
                 prepare_latest_log(pool,
                                 ADD_PAGES,
                                 (u64)page,
                                 order, cur_order);
+                #endif
                 list_del(&page->node);
                 pool->free_lists[cur_order].nr_free -= 1;
                 page = split_chunk(pool, order, page);
@@ -267,7 +271,9 @@ struct page *buddy_get_pages(struct phys_mem_pool *pool, int order)
         }
 
 out:
+        #ifdef USE_LOG
         commit_latest_log(pool);
+        #endif
         unlock(&pool->buddy_lock);
         return page;
 }

kernel/mm/slab.c

@@ -25,6 +25,7 @@ struct slab_log_entry {
 	int order;
 };
 
+#ifdef USE_LOG
 static void prepare_slab_log(struct slab_log_entry *log, void *free_list_head, int current_free_count, void *next_slot, int order) {
 	log->commited = false;
 	log->free_list_head = free_list_head;
@@ -35,6 +36,7 @@ static void prepare_slab_log(struct slab_log_entry *log, void *free_list_head, i
 static void commit_slab_log(struct slab_log_entry *log) {
 	log->commited = true;
 }
+#endif
 
 /*
 static inline int order_to_index(int order)
@@ -162,13 +164,17 @@ static void *alloc_in_slab_impl(int order)
 	struct slab_header *current_slab;
 	struct slab_slot_list* free_list;
 	void* next_slot;
+	#ifdef USE_LOG
 	struct slab_log_entry log;
+	#endif
 
 	lock(&slabs_locks[order]);
 
 	current_slab = slab_pool[order].current_slab;
+	#ifdef USE_LOG
 	log.from_init = false;
 	log.current_slab = current_slab;
+	#endif
 	/* When serving the first allocation request. */
 	
 	if (unlikely(current_slab == NULL)) {
@@ -177,17 +183,19 @@ static void *alloc_in_slab_impl(int order)
 			unlock(&slabs_locks[order]);
 			return NULL;
 		}
+		#ifdef USE_LOG
 		log.from_init = true;
 		log.current_slab = current_slab;
+		#endif
 		slab_pool[order].current_slab = current_slab;
 	}
 	
 	free_list = (struct slab_slot_list*)current_slab->free_list_head;
 	BUG_ON(free_list == NULL);
 	next_slot = free_list->next_free;
-
+	#ifdef USE_LOG
 	prepare_slab_log(&log, current_slab->free_list_head, current_slab->current_free_cnt, next_slot, order);
-
+	#endif
 	#ifdef USE_TSX
 	if (start_rtm == true) {
 		int status;
@@ -219,8 +227,9 @@ static void *alloc_in_slab_impl(int order)
 	/* When current_slab is full, choose a new slab as the current one. */
 	if (unlikely(current_slab->current_free_cnt == 0))
 		choose_new_current_slab(&slab_pool[order], order);
-
+	#ifdef USE_LOG
 	commit_slab_log(&log);
+	#endif
 	unlock(&slabs_locks[order]);
 
 	return (void *)free_list;

kernel/sls_config.cmake

@@ -1,5 +1,5 @@
 # SLS Basic Configruations
-set(SLS_RESTORE ON)
+set(SLS_RESTORE OFF)
 set(SLS_EXT_SYNC OFF)
 set(SLS_HYBRID_MEM ON)
 

user/sample-apps/apps/CMakeLists.txt

@@ -24,7 +24,7 @@ target_compile_options(assert.bin PRIVATE -g3 -ggdb)
 add_executable(udp_echo_server.bin udp_echo_server.c)
 add_executable(tcp_echo_server.bin tcp_echo_server.c)
 
-add_executable(tsx_test.bin tsx_test.c)
+add_executable(rtm_test.bin rtm_test.c)
 
 # TreeSLS
 add_subdirectory(treesls)

user/sample-apps/apps/rtm_test.c

@@ -0,0 +1,90 @@
+#include <malloc.h>
+#include <stdio.h>
+#include <chcore/pmu.h>
+#include <pthread.h>
+#include <chcore/syscall.h>
+#include <string.h>
+
+#pragma GCC push_options
+#pragma GCC optimize ("O0")
+#define num_threads 32
+s64 time_list[num_threads];
+int fixed_size = 128;
+int varying_size_times_max = 12;
+int total_malloc_times = 100000;
+
+enum test_size_t {
+  FIXED,
+  VARYING
+} test_size_type;
+// non log rtm+log
+
+void *work(void *arg) {
+  int i = *(int *)arg;
+  usys_set_affinity(-1, i);
+  usys_yield();
+  s64 start = 0, end = 0, period;
+
+  int count = total_malloc_times / num_threads;
+  int size;
+  pmu_clear_cnt();
+  start = pmu_read_real_cycle();
+   
+  for (int i = 0; i < count; i++) {
+    if (test_size_type == FIXED) {
+      size = fixed_size;
+    } else {
+      size = (1 << ((i % varying_size_times_max)));
+    }
+    void *m = malloc(size);
+    (void)m;
+  }
+
+  end = pmu_read_real_cycle();
+  period = end - start;
+  time_list[i] = period;
+  return NULL;
+}
+
+bool parse_args(int argc, char *argv[]) {
+  if (argc != 2) {
+    return false;
+  }
+  if (strcmp(argv[1], "0")) {
+    test_size_type = FIXED;
+  } else {
+    test_size_type = VARYING;
+  }
+  return true;
+}
+
+int main(int argc, char *argv[])
+{
+  if (parse_args(argc, argv) == false) {
+    return -1;
+  }
+  pthread_t threads[num_threads];
+  int thread_index[num_threads];
+  for (int i = 0; i < num_threads; i++) {
+    thread_index[i] = i;
+    if (pthread_create(&threads[i], NULL, work, thread_index + i) != 0) {
+      perror("Failed to create thread");
+      return 1;
+    }
+  }
+
+  for (int i = 0; i < num_threads; i++) {
+    pthread_join(threads[i], NULL);
+  }
+
+  s64 avg = 0;
+  for (int i = 0; i < num_threads; i++) {
+    avg += time_list[i] / num_threads;
+  }
+
+  printf("result: %ld\n", avg);
+  printf("rtm test finished\n");
+
+  return 0;
+}
+#pragma GCC pop_options

user/sample-apps/apps/tsx_test.c

@@ -1,49 +0,0 @@
-#include <malloc.h>
-#include <stdio.h>
-#include <chcore/pmu.h>
-#include <pthread.h>
-#include <chcore/syscall.h>
-
-#pragma GCC push_options
-#pragma GCC optimize ("O0")
-const int num_threads = 10;
-
-void *work(void *arg) {
-  int i = *(int *)arg;
-  usys_set_affinity(-1, i);
-  usys_yield();
-  s64 start = 0, end = 0;
-  int size = 4096; 
-  int count = 1000000 / num_threads;
-  pmu_clear_cnt();
-  start = pmu_read_real_cycle();
-  for (int i = 0; i < count; i++)
-  {
-    void *m = malloc(size);
-    (void)m;
-  }
-
-  end = pmu_read_real_cycle();
-  printf("tsx_test total %ld\n", end - start);
-  return NULL;
-}
-
-int main(int argc, char *argv[])
-{
-  pthread_t threads[num_threads];
-  int thread_index[num_threads];
-  for (int i = 0; i < num_threads; i++) {
-    thread_index[i] = i;
-    if (pthread_create(&threads[i], NULL, work, thread_index + i) != 0) {
-      perror("Failed to create thread");
-      return 1;
-    }
-  }
-
-  for (int i = 0; i < num_threads; i++) {
-    pthread_join(threads[i], NULL);
-  }
-
-  return 0;
-}
-#pragma GCC pop_options