perf_event_open: generalize to multiple events

path_properties.py

@@ -787,7 +787,14 @@ path_properties_tuples = (
                             'gem5_unimplemented_syscall': True
                         },
                         'pagemap_dump.c': {'requires_argument': True},
-                        'perf_event_open.c': {'extra_objs_lkmc_common': True},
+                        'perf_event_open.c': {
+                            # QEMU the syscall just fails on QEMU, presumably because QEMU
+                            # does not have a microarchitecture model, and so it must just set
+                            # CPU bits that inform the kernel that the feature is not available.
+                            # https://cirosantilli.com/linux-kernel-module-cheat#gem5-vs-qemu
+                            'allowed_emulators': {'gem5'},
+                            'extra_objs_lkmc_common': True,
+                        },
                         'poweroff.c': {'requires_sudo': True},
                         'proc_events.c': {'requires_sudo': True},
                         'proc_events.c': {'requires_sudo': True},

userland/linux/perf_event_open.c

@@ -1,7 +1,5 @@
 /* https://cirosantilli.com/linux-kernel-module-cheat#perf-event-open
- *
- * Malloc n bytes as given from the command line.
- */
+ * Adapted from `man perf_event_open` in manpages 5.05-1. */
 
 #include <asm/unistd.h>
 #include <linux/perf_event.h>
@@ -11,57 +9,95 @@
 #include <sys/ioctl.h>
 #include <unistd.h>
 
+#define LKMC_M5OPS_ENABLE 1
 #include <lkmc.h>
 
-static long
-perf_event_open(struct perf_event_attr *hw_event, pid_t pid,
-                int cpu, int group_fd, unsigned long flags)
-{
+static long perf_event_open(struct perf_event_attr *hw_event,
+    uint32_t type, uint64_t config, pid_t pid,
+    int cpu, int group_fd, unsigned long flags
+) {
     int ret;
 
+    memset(hw_event, 0, sizeof(struct perf_event_attr));
+    hw_event->type = type;
+    hw_event->size = sizeof(struct perf_event_attr);
+    hw_event->config = config;
+    hw_event->disabled = 1;
+    hw_event->exclude_kernel = 1;
+    /* Don't count hypervisor events. */
+    hw_event->exclude_hv = 1;
     ret = syscall(__NR_perf_event_open, hw_event, pid, cpu,
                     group_fd, flags);
+    if (ret == -1) {
+        fprintf(stderr, "Error opening leader %llx\n", hw_event->config);
+        exit(EXIT_FAILURE);
+    }
     return ret;
 }
 
-int
-main(int argc, char **argv)
-{
-    struct perf_event_attr pe;
-    long long count;
-    int fd;
+typedef struct {
+    char *name;
+    uint32_t type;
+    uint64_t config;
+} Desc;
 
+#define DESC(type,config) {#config, type, config}
+
+int
+main(int argc, char **argv) {
+    size_t i;
+    int gem5;
+    long long count;
     uint64_t n;
+    Desc descs[] = {
+        DESC(PERF_TYPE_HARDWARE, PERF_COUNT_HW_BRANCH_INSTRUCTIONS),
+        DESC(PERF_TYPE_HARDWARE, PERF_COUNT_HW_BRANCH_MISSES),
+        DESC(PERF_TYPE_HARDWARE, PERF_COUNT_HW_CPU_CYCLES),
+        DESC(PERF_TYPE_HARDWARE, PERF_COUNT_HW_INSTRUCTIONS),
+        DESC(PERF_TYPE_HW_CACHE, PERF_COUNT_HW_CACHE_L1D | PERF_COUNT_HW_CACHE_OP_READ << 8 | PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
+    };
+    struct perf_event_attr pes[LKMC_ARRAY_SIZE(descs)];
+    int fds[LKMC_ARRAY_SIZE(descs)];
+
     if (argc > 1) {
         n = strtoll(argv[1], NULL, 0);
     } else {
-        n = 100;
+        n = 10000;
+    }
+    if (argc > 2) {
+        gem5 = argv[2][0] == '1';
+    } else {
+        gem5 = 0;
     }
 
-    memset(&pe, 0, sizeof(struct perf_event_attr));
-    pe.type = PERF_TYPE_HARDWARE;
-    pe.size = sizeof(struct perf_event_attr);
-    pe.config = PERF_COUNT_HW_INSTRUCTIONS;
-    pe.disabled = 1;
-    pe.exclude_kernel = 1;
-    // Don't count hypervisor events.
-    pe.exclude_hv = 1;
+    for (i = 0; i < LKMC_ARRAY_SIZE(descs); i++)
+        fds[i] = perf_event_open(&pes[i],
+            descs[i].type, descs[i].config, 0, -1, -1, 0);
 
-    fd = perf_event_open(&pe, 0, -1, -1, 0);
-    if (fd == -1) {
-        fprintf(stderr, "Error opening leader %llx\n", pe.config);
-        exit(EXIT_FAILURE);
+    /* Start the counts. */
+    for (i = 0; i < LKMC_ARRAY_SIZE(descs); i++)
+        ioctl(fds[i], PERF_EVENT_IOC_RESET, 0);
+    for (i = 0; i < LKMC_ARRAY_SIZE(descs); i++)
+        ioctl(fds[i], PERF_EVENT_IOC_ENABLE, 0);
+    if (gem5) {
+        /* Cross check with gem5 stats to see if things make sense.. */
+        LKMC_M5OPS_RESETSTATS;
     }
 
-    ioctl(fd, PERF_EVENT_IOC_RESET, 0);
-    ioctl(fd, PERF_EVENT_IOC_ENABLE, 0);
-
+    /* Mesure this function. */
     lkmc_busy_loop(n, 1);
 
-    ioctl(fd, PERF_EVENT_IOC_DISABLE, 0);
-    read(fd, &count, sizeof(long long));
+    /* Stop the count. */
+    if (gem5) { LKMC_M5OPS_DUMPSTATS; }
+    for (i = 0; i < LKMC_ARRAY_SIZE(descs); i++)
+        ioctl(fds[i], PERF_EVENT_IOC_DISABLE, 0);
 
-    printf("Used %lld instructions\n", count);
+    /* Print results. */
+    for (i = 0; i < LKMC_ARRAY_SIZE(descs); i++) {
+        read(fds[i], &count, sizeof(long long));
+        printf("%s %lld\n", descs[i].name, count);
+    }
 
-    close(fd);
+    for (i = 0; i < LKMC_ARRAY_SIZE(descs); i++)
+        close(fds[i]);
 }