-/* Run this as: for i in $(seq 0 5); do prem-test $i & done */
-
#define _GNU_SOURCE /* See feature_test_macros(7) */
#include <err.h>
#include <sched.h>
#include <stdlib.h>
#include <unistd.h>
#include <time.h>
+#include <inttypes.h>
+#include <sys/mman.h>
+#include <stdbool.h>
+#include <pthread.h>
+#include <stdlib.h>
+#include <error.h>
-#ifndef SYS_prem_memguard_check
-#define SYS_prem_memguard_check 793
+#ifndef SYS_memguard
+#define SYS_memguard 793
#endif
-#define RANDOM_ARRAY_SIZE 16777216 // 16M * 8 = 128 MB array
+/* Return value:
+ * 0-31 - Sum of memory events
+ * 32-55 - Total time
+ * 61 - Memory budget overrun
+ * 62 - Timeout
+ * 63 - Error */
+#define MGRET_MEM_POS 0
+#define MGRET_TIM_POS 32
+#define MGRET_OVER_MEM_POS 61
+#define MGRET_OVER_TIM_POS 62
+#define MGRET_ERROR_POS 63
+
+#define MGRET_TIM_MASK (0x00FFFFFFul << MGRET_TIM_POS)
+#define MGRET_MEM_MASK (0xFFFFFFFFul << MGRET_MEM_POS)
+#define MGRET_OVER_MEM_MASK (1ul << MGRET_OVER_MEM_POS)
+#define MGRET_OVER_TIM_MASK (1ul << MGRET_OVER_TIM_POS)
+#define MGRET_ERROR_MASK (1ul << MGRET_ERROR_POS)
+
+#define MAX_CORES 6
-uint64_t random_values[RANDOM_ARRAY_SIZE];
-volatile uint64_t sum;
+char memory[128*10204*1024];
-enum prem_phase {
- PREM_COMPATIBLE = 0,
- PREM_MEMORY = 1,
- PREM_COMPUTE = 2,
+static pthread_barrier_t barrier;
+
+struct mg_ret {
+ uint32_t time;
+ uint32_t mem;
+ bool time_ovf;
+ bool mem_ovf;
};
-void compute_kernel(int time_ms){
- struct timespec ts;
- uint64_t current_us, end_us;
- clock_gettime(CLOCK_MONOTONIC, &ts);
- end_us = ts.tv_sec * 1000000 + ts.tv_nsec / 1000 + time_ms * 1000;
- do {
+static struct mg_ret mgret(uint64_t retval)
+{
+ struct mg_ret mgr = {};
+ mgr.time = (retval & MGRET_TIM_MASK) >> MGRET_TIM_POS;
+ mgr.mem = (retval & MGRET_MEM_MASK) >> MGRET_MEM_POS;
+ mgr.time_ovf = (retval >> MGRET_OVER_TIM_POS) & 1;
+ mgr.mem_ovf = (retval >> MGRET_OVER_MEM_POS) & 1;
+ return mgr;
+}
+
+void compute_kernel(int time_us)
+{
+ struct timespec ts;
+ uint64_t current_us, end_us;
clock_gettime(CLOCK_MONOTONIC, &ts);
- current_us = ts.tv_sec * 1000000 + ts.tv_nsec / 1000;
- } while (current_us < end_us);
+ end_us = ts.tv_sec * 1000000 + ts.tv_nsec / 1000 + time_us;
+ do {
+ clock_gettime(CLOCK_MONOTONIC, &ts);
+ current_us = ts.tv_sec * 1000000 + ts.tv_nsec / 1000;
+ } while (current_us < end_us);
}
-long memory_kernel(long lines){
- for(int i = 0; i < lines; i++){
- sum += random_values[(i*8) % RANDOM_ARRAY_SIZE];
- }
- return sum;
+void read_memory(long lines)
+{
+ uint64_t sum = 0;
+
+ for (uint64_t i = 0; i < lines; i++) {
+ sum += memory[(i * 64) % sizeof(memory)];
+ }
+ /* Do not optimize this function out */
+ volatile uint64_t x = sum;
+ (void)x;
}
-void random_memory_kernel(long lines) {
- uint32_t index;
+void write_memory(long lines)
+{
+ uint64_t *ptr;
+ for (uint64_t i = 0; i < lines; i++) {
+ ptr = (void*)&memory[(i * 64) % sizeof(memory)];
+ *ptr = i;
+ }
+}
+
+void read_memory_rnd(long lines)
+{
+ static unsigned rnd;
+ uint64_t sum = 0;
- for(long i=0; i<lines; i++) {
- index = (uint32_t)rand();
- index = index % RANDOM_ARRAY_SIZE;
- sum += random_values[index];
- }
+ for (long i = 0; i < lines; i++) {
+ rnd = (rnd + 523) * 253573;
+ sum += memory[rnd % sizeof(memory)];
+ }
+ /* Do not optimize this function out */
+ volatile uint64_t x = sum;
+ (void)x;
}
/* Memguard flags */
-#define MGF_PERIODIC (1 << 0) /* Chooses between periodic or one-shot budget replenishment */
-#define MGF_RET_STATS (1 << 1) /* Return statistics since the last call (cache misses, time, ...?) */
-#define MGF_MASK_INT (1 << 2) /* Mask (disable) low priority interrupts until next memguard call */
+#define MGF_PERIODIC \
+ (1 << 0) /* Chooses between periodic or one-shot budget replenishment */
+#define MGF_MASK_INT \
+ (1 \
+ << 1) /* Mask (disable) low priority interrupts until next memguard call */
-long prem_memguard_check(unsigned long timeout,
- unsigned long memory_budget,
- unsigned long flags)
+long memguard(unsigned long timeout, unsigned long memory_budget,
+ unsigned long flags)
{
- return syscall(SYS_prem_memguard_check, timeout, memory_budget, flags);
+ return syscall(SYS_memguard, timeout, memory_budget, flags);
+}
+
+void wvtest_pass(bool cond, const char* file, int line, const char* str)
+{
+ printf("! %s:%d %s %s\n", file, line, str, cond ? "ok" : "FAILURE");
+}
+
+#define WVPASS(cond) wvtest_pass(cond, __FILE__, __LINE__, #cond)
+
+static void print_test_info(uint64_t timeout_us, uint64_t mem_budget, uint64_t flags,
+ int64_t retval, const char *code)
+{
+ char call[80];
+ snprintf(call, sizeof(call), "memguard(%luus, %lumiss, %c%c):",
+ timeout_us, mem_budget,
+ flags & MGF_PERIODIC ? 'P' : '-',
+ flags & MGF_MASK_INT ? 'I' : '-'
+ );
+ struct mg_ret r = mgret(retval);
+ printf("Testing \"CPU%d: %-40s %-25s ⇒ time:%8u%c mem:%8u%c\" in %s:\n",
+ sched_getcpu(), call, code,
+ r.time, r.time_ovf ? '!' : ' ',
+ r.mem, r.mem_ovf ? '!' : ' ',
+ __FILE__);
+ WVPASS((retval & MGRET_ERROR_MASK) == 0);
+}
+
+#define MGTEST(timeout_us, mem_budget, flags, code) \
+ ({ \
+ long retval; \
+ retval = memguard(timeout_us, mem_budget, flags); \
+ code; \
+ retval = memguard(0, 0, 0); \
+ print_test_info(timeout_us, mem_budget, flags, retval, #code); \
+ mgret(retval); \
+ })
+
+void *test_thread(void *ptr)
+{
+ cpu_set_t set;
+ int cpu = (intptr_t)ptr;
+
+ /* Ensure that our test thread does not migrate to another CPU
+ * during memguarding */
+ CPU_ZERO(&set);
+ CPU_SET(cpu, &set);
+ if (sched_setaffinity(0, sizeof(set), &set) < 0)
+ err(1, "sched_setaffinity");
+
+ pthread_barrier_wait(&barrier);
+
+ struct mg_ret r;
+
+ for (uint64_t flags = 0; flags < 4; flags++) {
+ compute_kernel(1); /* warm up */
+
+ r = MGTEST(500, 10000, flags, compute_kernel(17*1000*1000));
+ continue;
+
+ ///////////////////////////////////////////////////////
+ r = MGTEST(5000, 10000, flags, compute_kernel(1000));
+ WVPASS(!r.time_ovf);
+ WVPASS(r.time > 900);
+
+ r = MGTEST(500, 10000, flags, compute_kernel(1000));
+ WVPASS(r.time_ovf);
+ WVPASS(r.time > 900);
+
+ r = MGTEST(5000, 10000, flags, compute_kernel(2000));
+ WVPASS(!r.time_ovf);
+ WVPASS(r.time > 1900);
+
+ r = MGTEST(500, 10000, flags, compute_kernel(2000));
+ WVPASS(r.time_ovf);
+ WVPASS(r.time > 1900);
+
+ ///////////////////////////////////////////////////////
+ r = MGTEST(100000, 500000, flags, read_memory(100000));
+ WVPASS(!r.mem_ovf);
+ WVPASS(r.mem >= 90000);
+
+ r = MGTEST(100000, 50000, flags, read_memory(100000));
+ WVPASS(r.mem_ovf);
+ WVPASS(r.mem >= 90000);
+
+ r = MGTEST(100000, 500000, flags, read_memory_rnd(100000));
+ WVPASS(!r.mem_ovf);
+ WVPASS(r.mem >= 90000);
+
+ r = MGTEST(100000, 50000, flags, read_memory_rnd(100000));
+ WVPASS(r.mem_ovf);
+ WVPASS(r.mem >= 90000);
+
+ r = MGTEST(100000, 500000, flags, write_memory(100000));
+ WVPASS(!r.mem_ovf);
+ WVPASS(r.mem >= 90000);
+
+ r = MGTEST(100000, 50000, flags, write_memory(100000));
+ WVPASS(r.mem_ovf);
+ WVPASS(r.mem >= 90000);
+
+
+ ///////////////////////////////////////////////////////
+ r = MGTEST(100000, 5000000, flags, read_memory(1000000));
+ WVPASS(!r.mem_ovf);
+ WVPASS(r.mem >= 900000);
+
+ r = MGTEST(100000, 500000, flags, read_memory(1000000));
+ WVPASS(r.mem_ovf);
+ WVPASS(r.mem >= 900000);
+
+ r = MGTEST(100000, 5000000, flags, read_memory_rnd(1000000));
+ WVPASS(!r.mem_ovf);
+ WVPASS(r.mem >= 900000);
+
+ r = MGTEST(100000, 500000, flags, read_memory_rnd(1000000));
+ WVPASS(r.mem_ovf);
+ WVPASS(r.mem >= 900000);
+
+ r = MGTEST(100000, 5000000, flags, write_memory(1000000));
+ WVPASS(!r.mem_ovf);
+ WVPASS(r.mem >= 900000);
+
+ r = MGTEST(100000, 500000, flags, write_memory(1000000));
+ WVPASS(r.mem_ovf);
+ WVPASS(r.mem >= 900000);
+
+ printf("\n");
+ }
+
+ /* Throttling tests */
+ struct mg_ret r1, r2, r3;
+ r1 = MGTEST(10000, 9000, MGF_PERIODIC | MGF_MASK_INT, read_memory(100000));
+ WVPASS(r1.time > 100*1000);
+ r2 = MGTEST(10000, 3000, MGF_PERIODIC | MGF_MASK_INT, read_memory(100000));
+ WVPASS(r2.time > 2 * r1.time);
+ r3 = MGTEST(10000, 1000, MGF_PERIODIC | MGF_MASK_INT, read_memory(100000));
+ WVPASS(r3.time > 2 * r2.time);
+
+ return NULL;
}
int main(int argc, char *argv[])
{
- cpu_set_t set;
- int cpu = 0;
-
- if (argc > 1)
- cpu = atoi(argv[1]);
-
- /* Ensure that memory phase starts and ends on the same CPU */
- CPU_ZERO(&set);
- CPU_SET(cpu, &set);
- if (sched_setaffinity(getpid(), sizeof(set), &set) < 0)
- err(1, "sched_setaffinity");
- printf("Pinned to CPU %d\n", cpu);
-
- printf("X1\n");
- fflush(stdout);
- //prem_memguard_check(950000, 1933440000, 1);
- prem_memguard_check(100000, 1000, MGF_PERIODIC);
- compute_kernel(1000);
- //prem_memguard_check(0, 0, 1);
-
- printf("X2\n");
- fflush(stdout);
- //prem_memguard_check(1050000, 2136960000, 1);
- prem_memguard_check(1050000, 2136960000, 0);
- compute_kernel(1000);
- //prem_memguard_check(0, 0, 1);
-
- printf("X3\n");
- fflush(stdout);
- //prem_memguard_check(1000000, 500, 1);
- prem_memguard_check(100000, 500, MGF_PERIODIC);
- memory_kernel(100000000);
- //prem_memguard_check(0, 0, 1);
-
- printf("X4\n");
- fflush(stdout);
- //prem_memguard_check(1000000, 1500, 1);
- prem_memguard_check(1000000, 15000000, 0);
- memory_kernel(100000000);
- //prem_memguard_check(0, 0, 1);
-
- printf("X5\n");
- fflush(stdout);
- //prem_memguard_check(1000000, 10000, 1);
- prem_memguard_check(1000000, 100000, 0);
- random_memory_kernel(100000000);
- //prem_memguard_check(0, 0, 1);
-
- printf("X6\n");
- fflush(stdout);
- //prem_memguard_check(1000000, 10000000, 1);
- prem_memguard_check(1000000, 100000000, 0);
- random_memory_kernel(100000000);
-
- prem_memguard_check(0, 0, 0);
-
- return 0;
+ int cpu_mask = 1;
+ int cpu_count = 0;
+ pthread_t threads[MAX_CORES];
+
+ /* TODO: currently shared memory */
+ for (int i = 0; i < sizeof(memory); i += 64)
+ memory[i] = 1;
+
+ if (argc > 1)
+ cpu_mask = strtol(argv[1], NULL, 16);
+
+ for (int i = 0; i < MAX_CORES; i++) {
+ if (cpu_mask & (1 << i)) {
+ cpu_count++;
+ }
+ }
+
+ printf("CPU count:%d CPU mask:%#x\n", cpu_count, cpu_mask);
+ int s = pthread_barrier_init(&barrier, NULL, cpu_count);
+ if (s != 0)
+ error(1, s, "pthread_barrier_init");
+
+ for (intptr_t i = 0; i < MAX_CORES; i++) {
+ if (cpu_mask & (1 << i))
+ pthread_create(&threads[i], NULL, test_thread, (void *)i);
+ }
+
+ for (int i = 0; i < MAX_CORES; i++) {
+ if (cpu_mask & (1 << i)) {
+ pthread_join(threads[i], NULL);
+ }
+ }
+
+ return 0;
}
projects / hercules2020 / jailhouse-build.git / blobdiff