engines/libaio: add preliminary support for pre-mapped IO buffers

Experimental kernel features that allows us to register IO buffers
when the io_context is setup, eliminating the need to perform
get_user_pages() + put_page() for each IO. This dramatically
increases performance and lowers latency.

Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/backend.c b/backend.c
index 2f103b3abb15f7d25e45d82f536b925075518bdd..2f46329376451f18a49c94696e84687dda4390d8 100644 (file)
--- a/backend.c
+++ b/backend.c
@@ -1704,6 +1704,9 @@ static void *thread_main(void *data)
        if (init_io_u(td))
                goto err;
 
+       if (td->io_ops->post_init && td->io_ops->post_init(td))
+               goto err;
+
        if (o->verify_async && verify_async_init(td))
                goto err;
 

diff --git a/engines/libaio.c b/engines/libaio.c
index 74238e4e939e55cab3b349925725554456269fb4..e0a073bc29add192e1eb24618d8a8d536a40f81e 100644 (file)
--- a/engines/libaio.c
+++ b/engines/libaio.c
@@ -24,6 +24,9 @@
 #ifndef IOCTX_FLAG_IOPOLL
 #define IOCTX_FLAG_IOPOLL      (1 << 1)
 #endif
+#ifndef IOCTX_FLAG_FIXEDBUFS
+#define IOCTX_FLAG_FIXEDBUFS   (1 << 2)
+#endif
 
 static int fio_libaio_commit(struct thread_data *td);
 
@@ -56,6 +59,7 @@ struct libaio_options {
        unsigned int userspace_reap;
        unsigned int hipri;
        unsigned int useriocb;
+       unsigned int fixedbufs;
 };
 
 static struct fio_option options[] = {
@@ -86,6 +90,15 @@ static struct fio_option options[] = {
                .category = FIO_OPT_C_ENGINE,
                .group  = FIO_OPT_G_LIBAIO,
        },
+       {
+               .name   = "fixedbufs",
+               .lname  = "Fixed (pre-mapped) IO buffers",
+               .type   = FIO_OPT_STR_SET,
+               .off1   = offsetof(struct libaio_options, fixedbufs),
+               .help   = "Pre map IO buffers",
+               .category = FIO_OPT_C_ENGINE,
+               .group  = FIO_OPT_G_LIBAIO,
+       },
        {
                .name   = NULL,
        },
@@ -399,7 +412,7 @@ static void fio_libaio_cleanup(struct thread_data *td)
 }
 
 static int fio_libaio_old_queue_init(struct libaio_data *ld, unsigned int depth,
-                                    bool hipri, bool useriocb)
+                                    bool hipri, bool useriocb, bool fixedbufs)
 {
        if (hipri) {
                log_err("fio: polled aio not available on your platform\n");
@@ -409,12 +422,16 @@ static int fio_libaio_old_queue_init(struct libaio_data *ld, unsigned int depth,
                log_err("fio: user mapped iocbs not available on your platform\n");
                return 1;
        }
+       if (fixedbufs) {
+               log_err("fio: fixed buffers not available on your platform\n");
+               return 1;
+       }
 
        return io_queue_init(depth, &ld->aio_ctx);
 }
 
 static int fio_libaio_queue_init(struct libaio_data *ld, unsigned int depth,
-                                bool hipri, bool useriocb)
+                                bool hipri, bool useriocb, bool fixedbufs)
 {
 #ifdef __NR_sys_io_setup2
        int ret, flags = 0;
@@ -423,6 +440,8 @@ static int fio_libaio_queue_init(struct libaio_data *ld, unsigned int depth,
                flags |= IOCTX_FLAG_IOPOLL;
        if (useriocb)
                flags |= IOCTX_FLAG_USERIOCB;
+       if (fixedbufs)
+               flags |= IOCTX_FLAG_FIXEDBUFS;
 
        ret = syscall(__NR_sys_io_setup2, depth, flags, ld->user_iocbs,
                        &ld->aio_ctx);
@@ -430,14 +449,42 @@ static int fio_libaio_queue_init(struct libaio_data *ld, unsigned int depth,
                return 0;
        /* fall through to old syscall */
 #endif
-       return fio_libaio_old_queue_init(ld, depth, hipri, useriocb);
+       return fio_libaio_old_queue_init(ld, depth, hipri, useriocb, fixedbufs);
+}
+
+static int fio_libaio_post_init(struct thread_data *td)
+{
+       struct libaio_data *ld = td->io_ops_data;
+       struct libaio_options *o = td->eo;
+       struct io_u *io_u;
+       struct iocb *iocb;
+       int err = 0;
+
+       if (o->fixedbufs) {
+               int i;
+
+               for (i = 0; i < td->o.iodepth; i++) {
+                       io_u = ld->io_u_index[i];
+                       iocb = &ld->user_iocbs[i];
+                       iocb->u.c.buf = io_u->buf;
+                       iocb->u.c.nbytes = io_u->buflen;
+               }
+       }
+
+       err = fio_libaio_queue_init(ld, td->o.iodepth, o->hipri, o->useriocb,
+                                       o->fixedbufs);
+       if (err) {
+               td_verror(td, -err, "io_queue_init");
+               return 1;
+       }
+
+       return 0;
 }
 
 static int fio_libaio_init(struct thread_data *td)
 {
        struct libaio_options *o = td->eo;
        struct libaio_data *ld;
-       int err = 0;
 
        ld = calloc(1, sizeof(*ld));
 
@@ -450,23 +497,6 @@ static int fio_libaio_init(struct thread_data *td)
                memset(ld->user_iocbs, 0, size);
        }
 
-       /*
-        * First try passing in 0 for queue depth, since we don't
-        * care about the user ring. If that fails, the kernel is too old
-        * and we need the right depth.
-        */
-       err = fio_libaio_queue_init(ld, td->o.iodepth, o->hipri, o->useriocb);
-       if (err) {
-               td_verror(td, -err, "io_queue_init");
-               log_err("fio: check /proc/sys/fs/aio-max-nr\n");
-               if (ld->user_iocbs) {
-                       size_t size = td->o.iodepth * sizeof(struct iocb);
-                       fio_memfree(ld->user_iocbs, size, false);
-               }
-               free(ld);
-               return 1;
-       }
-
        ld->entries = td->o.iodepth;
        ld->is_pow2 = is_power_of_2(ld->entries);
        ld->aio_events = calloc(ld->entries, sizeof(struct io_event));
@@ -494,6 +524,7 @@ static struct ioengine_ops ioengine = {
        .name                   = "libaio",
        .version                = FIO_IOOPS_VERSION,
        .init                   = fio_libaio_init,
+       .post_init              = fio_libaio_post_init,
        .io_u_init              = fio_libaio_io_u_init,
        .prep                   = fio_libaio_prep,
        .queue                  = fio_libaio_queue,

diff --git a/ioengines.h b/ioengines.h
index feb21db893456ae4ba5948d47f620f13e3ba3c46..b9cd33d52f5cf4e1252a6f946833c819b5edcdc3 100644 (file)
--- a/ioengines.h
+++ b/ioengines.h
@@ -7,7 +7,7 @@
 #include "flist.h"
 #include "io_u.h"
 
-#define FIO_IOOPS_VERSION      24
+#define FIO_IOOPS_VERSION      25
 
 /*
  * io_ops->queue() return values
@@ -25,6 +25,7 @@ struct ioengine_ops {
        int flags;
        int (*setup)(struct thread_data *);
        int (*init)(struct thread_data *);
+       int (*post_init)(struct thread_data *);
        int (*prep)(struct thread_data *, struct io_u *);
        enum fio_q_status (*queue)(struct thread_data *, struct io_u *);
        int (*commit)(struct thread_data *);