Merge branch 'iowait'

* iowait:
  Add suppport for toggling iowait

Author: axboe

src/include/liburing.h

@@ -320,6 +320,11 @@ int __io_uring_get_cqe(struct io_uring *ring,
 			struct io_uring_cqe **cqe_ptr, unsigned submit,
 			unsigned wait_nr, sigset_t *sigmask);
 
+/*
+ * Enable/disable setting of iowait by the kernel.
+ */
+int io_uring_set_iowait(struct io_uring *ring, bool enable_iowait);
+
 #define LIBURING_UDATA_TIMEOUT	((__u64) -1)
 
 /*

src/include/liburing/io_uring.h

@@ -527,6 +527,7 @@ struct io_cqring_offsets {
 #define IORING_ENTER_REGISTERED_RING	(1U << 4)
 #define IORING_ENTER_ABS_TIMER		(1U << 5)
 #define IORING_ENTER_EXT_ARG_REG	(1U << 6)
+#define IORING_ENTER_NO_IOWAIT		(1U << 7)
 
 /*
  * Passed in for io_uring_setup(2). Copied back with updated info on success
@@ -564,6 +565,7 @@ struct io_uring_params {
 #define IORING_FEAT_RECVSEND_BUNDLE	(1U << 14)
 #define IORING_FEAT_MIN_TIMEOUT		(1U << 15)
 #define IORING_FEAT_RW_ATTR		(1U << 16)
+#define IORING_FEAT_NO_IOWAIT		(1U << 17)
 
 /*
  * io_uring_register(2) opcodes and arguments

src/int_flags.h

@@ -2,10 +2,12 @@
 #ifndef LIBURING_INT_FLAGS
 #define LIBURING_INT_FLAGS
 
-#define INT_FLAGS_MASK		(IORING_ENTER_REGISTERED_RING)
+#define INT_FLAGS_MASK		(IORING_ENTER_REGISTERED_RING |	\
+				 IORING_ENTER_NO_IOWAIT)
 
 enum {
 	INT_FLAG_REG_RING	= IORING_ENTER_REGISTERED_RING,
+	INT_FLAG_NO_IOWAIT	= IORING_ENTER_NO_IOWAIT,
 	INT_FLAG_REG_REG_RING	= 1,
 	INT_FLAG_APP_MEM	= 2,
 	INT_FLAG_CQ_ENTER	= 4,

src/liburing-ffi.map

@@ -233,4 +233,5 @@ LIBURING_2.10 {
 		io_uring_prep_writev_fixed;
 		io_uring_prep_readv_fixed;
 		io_uring_prep_sendmsg_zc_fixed;
+		io_uring_set_iowait;
 } LIBURING_2.9;

src/liburing.map

@@ -115,4 +115,5 @@ LIBURING_2.9 {
 LIBURING_2.10 {
 	global:
 		io_uring_register_ifq;
+		io_uring_set_iowait;
 } LIBURING_2.9;

src/register.c

@@ -482,3 +482,14 @@ int io_uring_register_region(struct io_uring *ring,
 {
 	return do_register(ring, IORING_REGISTER_MEM_REGION, reg, 1);
 }
+
+int io_uring_set_iowait(struct io_uring *ring, bool enable_iowait)
+{
+	if (!(ring->features & IORING_FEAT_NO_IOWAIT))
+		return -EOPNOTSUPP;
+	if (enable_iowait)
+		ring->int_flags &= ~INT_FLAG_NO_IOWAIT;
+	else
+		ring->int_flags |= INT_FLAG_NO_IOWAIT;
+	return 0;
+}

test/Makefile

@@ -122,6 +122,7 @@ test_srcs := \
 	io_uring_passthrough.c \
 	io_uring_register.c \
 	io_uring_setup.c \
+	iowait.c \
 	kallsyms.c \
 	lfs-openat.c \
 	lfs-openat-write.c \

test/iowait.c

@@ -0,0 +1,136 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <string.h>
+
+#include "liburing.h"
+#include "helpers.h"
+
+static int get_iowait(int cpu)
+{
+	char cpu_buf[32], this_cpu[32];
+	int user, nice, system, idle, iowait, ret;
+	FILE *file;
+
+	file = fopen("/proc/stat", "r");
+	if (!file) {
+		perror("fopen stat");
+		return 0;
+	}
+
+	sprintf(this_cpu, "cpu%d", cpu);
+	ret = 0;
+	do {
+		int r;
+		r = fscanf(file, "%s %d %d %d %d %d", cpu_buf, &user, &nice,
+				&system, &idle, &iowait);
+		if (r != 6)
+			continue;
+		if (strncmp(cpu_buf, this_cpu, strlen(this_cpu)))
+			continue;
+		ret = iowait;
+		break;
+	} while (1);
+
+	fclose(file);
+	return ret;
+}
+
+static int test(struct io_uring *ring, int with_iowait, int cpu)
+{
+	struct io_uring_sqe *sqe;
+	struct io_uring_cqe *cqe;
+	int iowait_pre, iowait_post;
+	struct __kernel_timespec ts;
+	int ret, fds[2], diff;
+	char buf[32];
+
+	if (!(ring->features & IORING_FEAT_NO_IOWAIT))
+		return T_EXIT_SKIP;
+
+	if (pipe(fds) < 0) {
+		perror("pipe");
+		return T_EXIT_FAIL;
+	}
+
+	ret = io_uring_set_iowait(ring, with_iowait);
+	if (ret) {
+		fprintf(stderr, "set_iowait=%d\n", ret);
+		return T_EXIT_FAIL;
+	}
+
+	sqe = io_uring_get_sqe(ring);
+	io_uring_prep_read(sqe, fds[0], buf, sizeof(buf), 0);
+	io_uring_submit(ring);
+
+	ts.tv_sec = 1;
+	ts.tv_nsec = 0;
+	iowait_pre = get_iowait(cpu);
+	ret = io_uring_wait_cqe_timeout(ring, &cqe, &ts);
+	if (ret != -ETIME) {
+		fprintf(stderr, "Unexpected wait ret: %d\n", ret);
+		return T_EXIT_FAIL;
+	}
+	iowait_post = get_iowait(cpu);
+	diff = iowait_post - iowait_pre;
+
+	close(fds[0]);
+	close(fds[1]);
+
+	ret = io_uring_wait_cqe(ring, &cqe);
+	if (ret) {
+		fprintf(stderr, "Unexpected wait ret: %d\n", ret);
+		return T_EXIT_FAIL;
+	}
+	io_uring_cqe_seen(ring, cqe);
+
+	if (with_iowait) {
+		if (diff < 90) {
+			fprintf(stderr, "iowait diff too small: %d\n", diff);
+			return T_EXIT_FAIL;
+		}
+	} else {
+		if (diff > 10) {
+			fprintf(stderr, "iowait diff too large: %d\n", diff);
+			return T_EXIT_FAIL;
+		}
+	}
+	return T_EXIT_PASS;
+}
+
+int main(int argc, char *argv[])
+{
+	struct io_uring ring;
+	cpu_set_t cpuset;
+	int ret;
+
+	CPU_ZERO(&cpuset);
+	CPU_SET(0, &cpuset);
+	if (sched_setaffinity(0, sizeof(cpuset), &cpuset))
+		return T_EXIT_SKIP;
+
+	ret = t_create_ring(64, &ring, 0);
+	if (ret == T_SETUP_SKIP)
+		return T_EXIT_SKIP;
+	else if (ret != T_SETUP_OK)
+		return T_EXIT_FAIL;
+
+	ret = test(&ring, 0, 0);
+	if (ret == T_EXIT_SKIP) {
+		return T_EXIT_SKIP;
+	} else if (ret == T_EXIT_FAIL) {
+		fprintf(stderr, "test 0 failed\n");
+		return T_EXIT_FAIL;
+	}
+
+	ret = test(&ring, 1, 0);
+	if (ret == T_EXIT_SKIP) {
+		return T_EXIT_SKIP;
+	} else if (ret == T_EXIT_FAIL) {
+		fprintf(stderr, "test 1 failed\n");
+		return T_EXIT_FAIL;
+	}
+
+	io_uring_queue_exit(&ring);
+	return 0;
+}