#define IOCTX_FLAG_SQTHREAD (1 << 3) /* Use SQ thread */
#define IOCTX_FLAG_SQWQ (1 << 4) /* Use SQ wq */
+#define IOEV_RES2_CACHEHIT (1 << 0)
+
#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;
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_GETEVENTS (1 << 1)
#define DEPTH 32
-#define RING_SIZE (DEPTH + 1)
#define BATCH_SUBMIT 8
#define BATCH_COMPLETE 8
#define BS 4096
+static unsigned sq_ring_mask = DEPTH - 1;
+static unsigned cq_ring_mask = (2 * DEPTH) - 1;
+
struct submitter {
pthread_t thread;
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;
unsigned long calls;
+ unsigned long cachehit, cachemiss;
volatile int finish;
char filename[128];
};
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;
- u32 tail, next_tail, prepped = 0;
+ struct aio_sq_ring *ring = s->sq_ring;
+ u32 index, tail, next_tail, prepped = 0;
next_tail = tail = ring->tail;
do {
next_tail++;
- if (next_tail == ring->nr_events)
- next_tail = 0;
-
barrier();
if (next_tail == ring->head)
break;
- iocb = &s->sq_ring->iocbs[tail];
- init_io(s, fd, iocb);
+ index = tail & sq_ring_mask;
+ init_io(s, fd, &s->iocbs[index]);
+ s->sq_ring->array[index] = index;
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;
barrier();
if (head == ring->tail)
break;
- ev = &ring->events[head];
+ ev = &ring->events[head & cq_ring_mask];
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);
return -1;
}
+ if (ev->res2 & IOEV_RES2_CACHEHIT)
+ s->cachehit++;
+ else
+ s->cachemiss++;
reaped++;
head++;
- if (head == ring->nr_events)
- head = 0;
} while (1);
s->inflight -= reaped;
int main(int argc, char *argv[])
{
struct submitter *s = &submitters[0];
- unsigned long done, calls, reap;
+ unsigned long done, calls, reap, cache_hit, cache_miss;
int flags = 0, err;
int j;
size_t size;
arm_sig_int();
- size = sizeof(struct aio_iocb_ring) + RING_SIZE * sizeof(struct iocb);
+ size = sizeof(struct iocb) * DEPTH;
+ if (posix_memalign(&p, 4096, size))
+ return 1;
+ memset(p, 0, size);
+ s->iocbs = p;
+
+ size = sizeof(struct aio_sq_ring) + DEPTH * sizeof(u32);
if (posix_memalign(&p, 4096, size))
return 1;
s->sq_ring = p;
- s->sq_ring->nr_events = RING_SIZE;
memset(p, 0, size);
+ s->sq_ring->nr_events = DEPTH;
+ s->sq_ring->iocbs = (u64) s->iocbs;
/* CQ ring must be twice as big */
- size = sizeof(struct aio_io_event_ring) +
- 2 * RING_SIZE * sizeof(struct io_event);
+ size = sizeof(struct aio_cq_ring) +
+ 2 * DEPTH * sizeof(struct io_event);
if (posix_memalign(&p, 4096, size))
return 1;
s->cq_ring = p;
- s->cq_ring->nr_events = 2 * RING_SIZE;
memset(p, 0, size);
+ s->cq_ring->nr_events = 2 * DEPTH;
- for (j = 0; j < RING_SIZE; j++) {
- struct iocb *iocb = &s->sq_ring->iocbs[j];
+ for (j = 0; j < DEPTH; 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(DEPTH, flags, s->sq_ring, s->cq_ring, &s->ioc);
if (err) {
printf("ctx_init failed: %s, %d\n", strerror(errno), err);
return 1;
pthread_create(&s->thread, NULL, submitter_fn, s);
- reap = calls = done = 0;
+ cache_hit = cache_miss = reap = calls = done = 0;
do {
unsigned long this_done = 0;
unsigned long this_reap = 0;
unsigned long this_call = 0;
+ unsigned long this_cache_hit = 0;
+ unsigned long this_cache_miss = 0;
unsigned long rpc = 0, ipc = 0;
+ double hit = 0.0;
sleep(1);
this_done += s->done;
this_call += s->calls;
this_reap += s->reaps;
+ this_cache_hit += s->cachehit;
+ this_cache_miss += s->cachemiss;
+ if (this_cache_hit && this_cache_miss) {
+ unsigned long hits, total;
+
+ hits = this_cache_hit - cache_hit;
+ total = hits + this_cache_miss - cache_miss;
+ hit = (double) hits / (double) total;
+ hit *= 100.0;
+ }
if (this_call - calls) {
rpc = (this_done - done) / (this_call - calls);
ipc = (this_reap - reap) / (this_call - calls);
}
- printf("IOPS=%lu, IOS/call=%lu/%lu, inflight=%u (head=%d tail=%d), %lu, %lu\n",
+ printf("IOPS=%lu, IOS/call=%lu/%lu, inflight=%u (head=%u tail=%u), Cachehit=%0.2f%%\n",
this_done - done, rpc, ipc, s->inflight,
- s->cq_ring->head, s->cq_ring->tail, s->reaps, s->done);
+ s->cq_ring->head, s->cq_ring->tail, hit);
done = this_done;
calls = this_call;
reap = this_reap;
+ cache_hit = s->cachehit;
+ cache_miss = s->cachemiss;
} while (!finish);
pthread_join(s->thread, &ret);