1 files changed, 193 insertions, 37 deletions

diff --git a/baseline/source/extra.h b/baseline/source/extra.h
index ca92812..df8164f 100644
--- a/baseline/source/extra.h
+++ b/baseline/source/extra.h
@@ -1,3 +1,8 @@
+/**
+ * Copyright 2019 Sims Hill Osborne and Joshua Bakita
+ *
+ * This header provides facilities by which to separably run and time TACLeBench
+ **/
 #include <time.h>
 #include <sys/mman.h>
 #include <stdlib.h>
@@ -5,12 +10,47 @@
 #include <string.h>
 #include <signal.h>
 #include <limits.h>
+#include <fcntl.h>
 
-#define L3_CACHE_SIZE (11264*1024)
+// These constants correspond to the imx6q-sabredb platform
+#define LINE_SIZE 32
+#define L2_SIZE 16*2048*32
 
+#if __arm__
+#include <unistd.h>
+#include <sys/syscall.h>
+#endif
 
+#define LITMUS 1
+#define MC2 1
+#define MMDC_PROF 1
 
-#define SET_UP char *thisProgram=argv[1];\
+#if LITMUS
+#include <litmus.h>
+#endif
+
+#if MMDC_PROF
+#include  "/media/speedy/litmus/tools/mmdc/mmdc.h"
+#endif
+
+#if LITMUS
+#define SET_UP LOAD_PARAMS SETUP_LITMUS
+#else
+#define SET_UP LOAD_PARAMS
+#endif
+
+#if MMDC_PROF
+#define LOAD_PARAMS LOAD_PARAMS_ITRL SETUP_MMDC
+#else
+#define LOAD_PARAMS LOAD_PARAMS_ITRL
+#endif
+
+#define LOAD_PARAMS_ITRL \
+    if (argc < 9) { \
+        printf("Usage: %s <name> <loops> <my core> <unused> <unused> <runID> <\?\?\?>", argv[0]);\
+        exit(1);\
+    }\
+    char *thisProgram=argv[1];\
     int maxJobs=atoi(argv[2]);\
     char *thisCore=argv[3];\
     char *otherCore=argv[4];\
@@ -20,37 +60,152 @@
     pid_t killMe;\
     struct timespec start, end;\
     int jobsComplete;\
+    int jobs_complete = -1;\
     long progTime[maxJobs*output];\
+    memset(progTime, 0, sizeof(long)*maxJobs*output);\
     char fileName[50];\
     char *bigArray;\
     int wasteCount;\
+    unsigned long mmdc_read[maxJobs];\
+    unsigned long mmdc_write[maxJobs];\
+    memset(mmdc_read, 0, sizeof(unsigned long)*maxJobs);\
+    memset(mmdc_write, 0, sizeof(unsigned long)*maxJobs);\
     strcpy(fileName, runID);\
     strcat(fileName, ".txt");\
     mlockall(MCL_CURRENT || MCL_FUTURE);
 
+#define SETUP_MMDC \
+    MMDC_PROFILE_RES_t mmdc_res; \
+    memset(&mmdc_res, 0, sizeof(MMDC_PROFILE_RES_t));\
+    int fd = open("/dev/mem", O_RDWR, 0);\
+    if (fd < 0) {\
+        perror("Unable to open /dev/mem");\
+        exit(1);\
+    }\
+    pMMDC_t mmdc = mmap(NULL, 0x4000, PROT_READ | PROT_WRITE, MAP_SHARED, fd, MMDC_P0_IPS_BASE_ADDR);\
+    if (mmdc == MAP_FAILED) {\
+        perror("Unable to map MMDC address space");\
+        exit(1);\
+    }\
+    mmdc->madpcr1 = axi_arm1;\
+    msync(&(mmdc->madpcr1),4,MS_SYNC);
+
+#define SETUP_LITMUS \
+    unsigned int cpu = atoi(thisCore); \
+    unsigned int wait = 0; \
+    if (be_migrate_to_domain(cpu) < 0) { \
+        perror("Unable to migrate to specified CPU"); \
+        exit(1); \
+    } \
+    struct reservation_config res; \
+    res.id = gettid(); \
+    res.cpu = cpu; \
+    res.priority = LITMUS_HIGHEST_PRIORITY; \
+    /* we take over half the CPU time (these are ns) */ \
+    res.polling_params.budget = ms2ns(999); \
+    res.polling_params.period = ms2ns(1000); \
+    res.polling_params.offset = 0; \
+    res.polling_params.relative_deadline = ms2ns(999); \
+    /* Not 100% sure that we should use periodic polling */ \
+    if (reservation_create(PERIODIC_POLLING, &res) < 0) { \
+        perror("Unable to create reservation"); \
+        exit(1); \
+    } \
+    struct rt_task rt_param; \
+    init_rt_task_param(&rt_param); \
+    /* Supposedly the next two parameters are irrelevant when reservations are enabled, but I'm leaving them anyway... */ \
+    rt_param.exec_cost = ms2ns(999); \
+    rt_param.period = ms2ns(1000); \
+    rt_param.priority = LITMUS_HIGHEST_PRIORITY; \
+    rt_param.cls = RT_CLASS_HARD; \
+    rt_param.release_policy = TASK_PERIODIC; \
+    rt_param.budget_policy = NO_ENFORCEMENT; \
+    rt_param.cpu = cpu; \
+    if (set_rt_task_param(gettid(), &rt_param) < 0) { \
+        perror("Unable to set real-time parameters"); \
+        exit(1); \
+    } \
+    if (init_litmus() != 0) { \
+        perror("init_litmus failed"); \
+        exit(1); \
+    } \
+    MC2_SETUP \
+    if (task_mode(LITMUS_RT_TASK) != 0) { \
+        perror("Unable to become real-time task"); \
+        exit(1); \
+    } \
+    if (wait && wait_for_ts_release() != 0) { \
+        perror("Unable to wait for taskset release"); \
+        exit(1); \
+    }
+
+#if MC2
+#define MC2_SETUP \
+    struct mc2_task mc2_param; \
+    mc2_param.res_id = gettid(); \
+    mc2_param.crit = CRIT_LEVEL_A; \
+    if (set_mc2_task_param(gettid(), &mc2_param) < 0) { \
+        perror("Unable to set MC^2 task params"); \
+        exit(1); \
+    } \
+    set_page_color(rt_param.cpu);
+#else
+#define MC2_SETUP
+#endif
 
-//if output==0, endless loop
-//avoids int overflow error with large numbers for background loops
+#define CLEANUP_LITMUS \
+    if (task_mode(BACKGROUND_TASK) != 0) { \
+        perror("Unable to become a real-time task"); \
+        exit(1); \
+    } \
+    reservation_destroy(gettid(), rt_param.cpu);
 
-#define SAVE_RESULTS if(jobsComplete>-1) progTime[jobsComplete]=(end.tv_nsec-start.tv_nsec)+(1000000000*(end.tv_sec-start.tv_sec));
+#if MMDC_PROF
+#define SAVE_RESULTS \
+    if(jobs_complete>-1) {\
+        progTime[jobs_complete]=(end.tv_nsec-start.tv_nsec)+(1000000000*(end.tv_sec-start.tv_sec));\
+        mmdc_read[jobs_complete] = mmdc_res.read_bytes;\
+        mmdc_write[jobs_complete] = mmdc_res.write_bytes;\
+    }
+#else
+#define SAVE_RESULTS \
+    if(jobs_complete>-1)\
+        progTime[jobs_complete]=(end.tv_nsec-start.tv_nsec)+(1000000000*(end.tv_sec-start.tv_sec));
+#endif
 
-#define WRITE_TO_FILE if (output){\
+
+#if LITMUS
+#define WRITE_TO_FILE WRITE_TO_FILE_ITRNL CLEANUP_LITMUS
+#else
+#define WRITE_TO_FILE WRITE_TO_FILE_ITRNL
+#endif
+
+#define WRITE_TO_FILE_ITRNL if (output){\
     munlockall();\
     FILE *fp=fopen(fileName, "a");\
     if (fp == NULL) {\
         perror("Error opening file. \n");\
         exit(1);\
     }\
-    for(jobsComplete=0; jobsComplete<maxJobs; jobsComplete++){\
-        fprintf(fp, "%s %s %s %s %d %ld %s %d \n",\
+    for(int i = 0; i <= jobs_complete; i++){\
+        fprintf(fp, "%s %s %s %s %d %ld %s %d %lu %lu \n",\
         thisProgram, otherProgram, thisCore, otherCore, maxJobs,\
-        progTime[jobsComplete], runID, jobsComplete);\
+        progTime[i], runID, i, mmdc_read[i], mmdc_write[i]);\
     }\
     fclose(fp);\
 }
 
-// Call the wbinvld instruction (it's in a kernel module due to it being ring-0)
-#define FLUSH_CACHES FILE *fp = fopen("/proc/wbinvd", "r");\
+#if __arm__
+// On ARM, manually flush the cache
+#define FLUSH_CACHES \
+    volatile uint8_t buffer[L2_SIZE * 4]; \
+    for (uint32_t j = 0; j < 4; j++) \
+        for (uint32_t i = 0; i < L2_SIZE * 4; i += LINE_SIZE) \
+            buffer[i]++;
+#else
+// On x86 call the wbinvld instruction (it's in a kernel module due to it being ring-0)
+#define FLUSH_CACHES \
+    FILE *fp = fopen("/proc/wbinvd", "r");\
     if (fp == NULL) {\
         perror("Cache flush module interface cannot be opened");\
         exit(1);\
@@ -61,40 +216,41 @@
         exit(1);\
     }\
     fclose(fp);
+#endif
 
-//invoke start timer twice, stop timer to make sure timer and vars are in cache
-//#define START_TIMER clock_gettime(CLOCK_MONOTONIC, &start);\
-        clock_gettime(CLOCK_MONOTONIC, &end);\
-    clock_gettime(CLOCK_MONOTONIC, &start);\
-
+#if MMDC_PROF
+#define START_TIMER \
+    /* This disables profiling, resets the counters, clears the overflow bit, and enables profiling */ \
+    start_mmdc_profiling(mmdc); \
+    /*nanosleep(&(struct timespec){0, ms2ns(999)}, NULL);*/ \
+    clock_gettime(CLOCK_MONOTONIC, &start);
+#else
 #define START_TIMER clock_gettime(CLOCK_MONOTONIC, &start);
+#endif
 
+#if MMDC_PROF
+#define STOP_TIMER \
+    clock_gettime(CLOCK_MONOTONIC, &end); \
+    /* This freezes the profiling and makes results available */ \
+    pause_mmdc_profiling(mmdc); \
+    get_mmdc_profiling_results(mmdc, &mmdc_res);
+#else
 #define STOP_TIMER clock_gettime(CLOCK_MONOTONIC, &end);
-
-//waste a millisecond
-
-#define WASTE_TIME clock_gettime(CLOCK_MONOTONIC, &start);\
-do{clock_gettime(CLOCK_MONOTONIC, &end);}\
-while( (end.tv_sec*1000000000+end.tv_nsec)-(start.tv_sec*1000000000+start.tv_nsec) < 1000000);
-
+#endif
 
 #define SLEEP nanosleep((const struct timespec[]){{0, 1000000}}, NULL);
 
-//#define LOOP_WASTE for(wasteCount=0; wasteCount<1000000; wasteCount++){}
-
-//at beginning of loop clear cache, waste some time
-//wasting time allows interfering process to build up some cache presence
-//using sleep instead of waste time loop causes problems
-//both sleeping and spinning give advantage to threaded task
-//#define START_LOOP if (output) {KILL_CACHE  START_TIMER}
-
-//#define START_LOOP if (output){START_TIMER}
-
-//#define STOP_LOOP if (output) {STOP_TIMER SAVE_RESULTS}
-//if (!output) {jobsComplete--;}
-
+#if LITMUS
+#define START_LOOP \
+    if (sleep_next_period() != 0) { \
+        perror("Unable to sleep for next period"); \
+    } \
+    FLUSH_CACHES START_TIMER
+#else
 #define START_LOOP FLUSH_CACHES START_TIMER
-#define STOP_LOOP STOP_TIMER SAVE_RESULTS
+#endif
+
+#define STOP_LOOP STOP_TIMER SAVE_RESULTS jobs_complete++;
 
 
 /*