prof: add user/kernel statistics

kernel-src/arch/x86-64/arch_prof.c

@@ -99,10 +99,8 @@ static bool intel_irq(context_t *context) {
 	uint8_t data[1 + sizeof(uintptr_t) * 256];
 
 	uint8_t size = get_backtrace(context, (uintptr_t *)(data + 1));
-	if (size) {
-		*data = size;
-		profiling_insert(size, data);
-	}
+	*data = size;
+	profiling_insert(size, data);
 
 	arch_apic_init_perf(); // SDM LVT figure says a PMC irq masks itself, so unmask it
 	reset_intel_pmc();
@@ -118,8 +116,6 @@ static void init_amd_pmc(void) {
 	__assert(!"Unimplemented");
 }
 
-
-
 bool arch_profiling_irq(context_t *context) {
 	if (!strcmp(current_cpu()->vendor, "AuthenticAMD"))
 		return amd_irq(context);

kernel-src/sys/prof.c

@@ -8,6 +8,7 @@
 #include <kernel/vfs.h>
 #include <arch/smp.h>
 #include <logging.h>
+#include <kernel/usercopy.h>
 
 #ifdef ENABLE_PROFILING
 
@@ -19,9 +20,17 @@ static MUTEX_DEFINE(mutex);
 static pollheader_t pollheader;
 static semaphore_t wait_sem;
 static int remaining;
+static size_t userspace_hits;
+static size_t kernel_hits;
 
 // called from a possibly NMI context!
 void profiling_insert(uint8_t size, void *data) {
+	if (!size) {
+		__atomic_fetch_add(&userspace_hits, 1, __ATOMIC_SEQ_CST);
+		return;
+	}
+
+	__atomic_fetch_add(&kernel_hits, 1, __ATOMIC_SEQ_CST);
 	size_t write_size = sizeof(uint8_t) + sizeof(uintptr_t) * size;
 	if (current_cpu()->prof_ringbuffer && RINGBUFFER_FREESPACE(current_cpu()->prof_ringbuffer) >= write_size) {
 		size_t written = ringbuffer_write(current_cpu()->prof_ringbuffer, data, write_size);
@@ -149,19 +158,31 @@ static int device_read(int minor, iovec_iterator_t *iovec_iterator, size_t size,
 	return error;
 }
 
-static int device_ioctl(int, unsigned long request, void *, int *result, cred_t *cred) {
-	if (request != 12345678)
-		return ENOTTY;
-
+static int device_ioctl(int, unsigned long request, void *addr, int *result, cred_t *cred) {
+	*result = 0;
 	// only supported ioctl is a truncation one: remove all previous data
 
-	MUTEX_ACQUIRE(&mutex);
+	switch (request) {
+		case 12345678:
+		MUTEX_ACQUIRE(&mutex);
 
-	ringbuffer_truncate(&global_ringbuffer, PROFILING_GLOBAL_BUFFER_SIZE);
+		ringbuffer_truncate(&global_ringbuffer, PROFILING_GLOBAL_BUFFER_SIZE);
 
-	MUTEX_RELEASE(&mutex);
+		MUTEX_RELEASE(&mutex);
+
+		__atomic_store_n(&userspace_hits, 0, __ATOMIC_SEQ_CST);
+		__atomic_store_n(&kernel_hits, 0, __ATOMIC_SEQ_CST);
+		break;
+		case 87654321:
+		size_t outputs[2] = {
+			__atomic_load_n(&userspace_hits, __ATOMIC_SEQ_CST), 
+			__atomic_load_n(&kernel_hits, __ATOMIC_SEQ_CST)
+		};
+
+		
+		return USERCOPY_POSSIBLY_TO_USER(addr, outputs, sizeof(outputs));
+	}
 
-	*result = 0;
 	return 0;
 }
 

tools/prof/main.c

@@ -22,8 +22,8 @@ static int open_or_die(const char *name, int flag, mode_t mode) {
 
 int main(int argc, char *argv[]) {
 	progname = argv[0];
-	if (argc != 2) {
-		fprintf(stderr, "%s: usage: %s FILE\n", argv[0], argv[0]);
+	if (argc < 2 || argc > 3) {
+		fprintf(stderr, "%s: usage: %s FILE [-n]\n", argv[0], argv[0]);
 		return EXIT_FAILURE;
 	}
 
@@ -42,10 +42,12 @@ int main(int argc, char *argv[]) {
 		tcsetattr(outfd, TCSANOW, &termios);
 	}
 
-	printf("%s: draining /dev/prof\n", argv[0]);
-	// empty it
-	if (ioctl(infd, 12345678, NULL) < 0)
-		return EXIT_FAILURE;
+	if (argc == 2 || (strcmp(argv[1], "-n") && strcmp(argv[2], "-n"))) {
+		printf("%s: draining /dev/prof\n", argv[0]);
+		// empty it
+		if (ioctl(infd, 12345678, NULL) < 0)
+			return EXIT_FAILURE;
+	}
 
 	struct pollfd pollfd[2];
 
@@ -68,6 +70,16 @@ int main(int argc, char *argv[]) {
 		write(outfd, buf, 8 * count);
 	}
 
+	size_t hits[2];
+	ioctl(infd, 87654321, hits);
+
+	size_t total = hits[0] + hits[1];
+	float u_ratio = (float)hits[0] / (float)total;
+	float k_ratio = 1.f - u_ratio;
+
+	printf("%s: %lu hits\n", argv[0], total);
+	printf("%s: %lu userspace hits (%01.02f%%)\n", argv[0], hits[0], u_ratio * 100);
+	printf("%s: %lu kernel hits    (%01.02f%%)\n", argv[0], hits[1], k_ratio * 100);
 	printf("%s: done\n", argv[0]);
 
 	return EXIT_SUCCESS;