#define IOCB_FLAG_HIPRI (1 << 2)
-#define IOCTX_FLAG_IOPOLL (1 << 1)
-#define IOCTX_FLAG_USERIOCB (1 << 0)
+#define IOCTX_FLAG_IOPOLL (1 << 0)
+#define IOCTX_FLAG_SCQRING (1 << 1) /* Use SQ/CQ rings */
#define IOCTX_FLAG_FIXEDBUFS (1 << 2)
-#define IOCTX_FLAG_SCQRING (1 << 3) /* Use SQ/CQ rings */
-#define IOCTX_FLAG_SQTHREAD (1 << 4) /* Use SQ thread */
-#define IOCTX_FLAG_SQWQ (1 << 5) /* Use SQ wq */
+#define IOCTX_FLAG_SQTHREAD (1 << 3) /* Use SQ thread */
+#define IOCTX_FLAG_SQWQ (1 << 4) /* Use SQ wq */
#define barrier() __asm__ __volatile__("": : :"memory")
#define min(a, b) ((a < b) ? (a) : (b))
+typedef uint64_t u64;
typedef uint32_t u32;
+typedef uint16_t u16;
-struct aio_iocb_ring {
+struct aio_sq_ring {
union {
struct {
- u32 head, tail;
+ u32 head;
+ u32 tail;
u32 nr_events;
- u32 sq_thread_cpu;
+ u16 sq_thread_cpu;
+ u64 iocbs;
};
- struct iocb pad_iocb;
+ u32 pad[16];
};
- struct iocb iocbs[0];
+ u32 array[0];
};
-struct aio_io_event_ring {
+struct aio_cq_ring {
union {
struct {
- u32 head, tail;
+ u32 head;
+ u32 tail;
u32 nr_events;
};
- struct io_event pad_event;
+ struct io_event pad;
};
struct io_event events[0];
};
#define IORING_FLAG_SUBMIT (1 << 0)
#define IORING_FLAG_GETEVENTS (1 << 1)
-#define container_of(ptr, type, member) ({ \
- const __typeof__( ((type *)0)->member ) *__mptr = (ptr); \
- (type *)( (char *)__mptr - offsetof(type,member) );})
-
#define DEPTH 32
#define RING_SIZE (DEPTH + 1)
unsigned long max_blocks;
io_context_t ioc;
struct drand48_data rand;
- struct aio_iocb_ring *sq_ring;
- struct aio_io_event_ring *cq_ring;
+ struct aio_sq_ring *sq_ring;
+ struct iocb *iocbs;
+ struct aio_cq_ring *cq_ring;
int inflight;
unsigned long reaps;
unsigned long done;
static int polled = 1; /* use IO polling */
static int fixedbufs = 1; /* use fixed user buffers */
static int buffered = 0; /* use buffered IO, not O_DIRECT */
-static int sq_thread = 0;
-static int sq_thread_cpu = 0;
+static int sq_thread = 0; /* use kernel submission thread */
+static int sq_thread_cpu = 0; /* pin above thread to this CPU */
static int io_setup2(unsigned int nr_events, unsigned int flags,
- struct iocb *iocbs, struct aio_iocb_ring *sq_ring,
- struct aio_io_event_ring *cq_ring, io_context_t *ctx_idp)
+ struct aio_sq_ring *sq_ring, struct aio_cq_ring *cq_ring,
+ io_context_t *ctx_idp)
{
- return syscall(335, nr_events, flags, iocbs, sq_ring, cq_ring, ctx_idp);
+ return syscall(335, nr_events, flags, sq_ring, cq_ring, ctx_idp);
}
static int io_ring_enter(io_context_t ctx, unsigned int to_submit,
static int prep_more_ios(struct submitter *s, int fd, int max_ios)
{
- struct aio_iocb_ring *ring = s->sq_ring;
- struct iocb *iocb;
+ struct aio_sq_ring *ring = s->sq_ring;
u32 tail, next_tail, prepped = 0;
next_tail = tail = ring->tail;
if (next_tail == ring->head)
break;
- iocb = &s->sq_ring->iocbs[tail];
- init_io(s, fd, iocb);
+ init_io(s, fd, &s->iocbs[tail]);
+ s->sq_ring->array[tail] = tail;
prepped++;
tail = next_tail;
} while (prepped < max_ios);
static int reap_events(struct submitter *s)
{
- struct aio_io_event_ring *ring = s->cq_ring;
+ struct aio_cq_ring *ring = s->cq_ring;
struct io_event *ev;
u32 head, reaped = 0;
break;
ev = &ring->events[head];
if (ev->res != BS) {
- int index = (int) (uintptr_t) ev->obj;
- struct iocb *iocb = &s->sq_ring->iocbs[index];
+ struct iocb *iocb = ev->obj;
printf("io: unexpected ret=%ld\n", ev->res);
printf("offset=%lu, size=%lu\n", (unsigned long) iocb->u.c.offset, (unsigned long) iocb->u.c.nbytes);
break;
if (this_reap)
goto submit;
- printf("EAGAIN reap=%d\n", reap_events(s));
- printf("%d/%d, %d/%d\n", s->sq_ring->head, s->sq_ring->tail, s->cq_ring->head, s->cq_ring->tail);
to_submit = 0;
goto submit;
}
arm_sig_int();
- size = sizeof(struct aio_iocb_ring) + RING_SIZE * sizeof(struct iocb);
+ size = sizeof(struct iocb) * RING_SIZE;
+ if (posix_memalign(&p, 4096, size))
+ return 1;
+ memset(p, 0, size);
+ s->iocbs = p;
+
+ size = sizeof(struct aio_sq_ring) + RING_SIZE * sizeof(u32);
if (posix_memalign(&p, 4096, size))
return 1;
s->sq_ring = p;
memset(p, 0, size);
+ s->sq_ring->nr_events = RING_SIZE;
+ s->sq_ring->iocbs = (u64) s->iocbs;
- size = sizeof(struct aio_io_event_ring) + RING_SIZE * sizeof(struct io_event);
+ /* CQ ring must be twice as big */
+ size = sizeof(struct aio_cq_ring) +
+ 2 * RING_SIZE * sizeof(struct io_event);
if (posix_memalign(&p, 4096, size))
return 1;
s->cq_ring = p;
memset(p, 0, size);
+ s->cq_ring->nr_events = 2 * RING_SIZE;
for (j = 0; j < RING_SIZE; j++) {
- struct iocb *iocb = &s->sq_ring->iocbs[j];
+ struct iocb *iocb = &s->iocbs[j];
if (posix_memalign(&iocb->u.c.buf, BS, BS)) {
printf("failed alloc\n");
s->sq_ring->sq_thread_cpu = sq_thread_cpu;
}
- err = io_setup2(RING_SIZE, flags, s->sq_ring->iocbs, s->sq_ring, s->cq_ring, &s->ioc);
+ err = io_setup2(RING_SIZE, flags, s->sq_ring, s->cq_ring, &s->ioc);
if (err) {
printf("ctx_init failed: %s, %d\n", strerror(errno), err);
return 1;