bdev_xnvme: add option to conserve cpu and set default to false

To achieve the highest possible IO-rates and lowest latency, then CPU
cycles must be spent processing IO. This commit introduces three
different techniques dependendent on the 'io_mechanism' used.

For 'libaio', xNVMe is instructued via 'opts.poll_io' to not wait for
completions, thus xnvme_queue_poke() returns immmediatly, the user can
then call xnvme_queue_poke() as frequently as desired. This requires
xNVMe v0.5.0. Earlier versions will ignore 'opts.poll_io' for 'libaio'.

For 'io_uring', xNVMe is instructed via 'opts.poll_io' to enable
IORING_SETUP_IOPOLL.

For 'io_uring_cmd', xNVMe is instructued via 'opts.poll_sq' to enable
IORING_SETUP_SQPOLL, this sets up a kernel-side thread reaping
completions from user-space on the ring. Note that 'io_uring_cmd'
requires liburing >= 2.2.

This commit enables the above by default. The above can be disabled by
setting 'conserve_cpu' to true on the bdev-setup.

Signed-off-by: Simon A. F. Lund <simon.lund@samsung.com>
Change-Id: Id54f1e59733ce9ae3b174ad4562904d868d4ef4f
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/14678
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>

doc/jsonrpc.md

@@ -5737,6 +5737,7 @@ Name                    | Optional | Type        | Description
 name                    | Required | string      | name of xNVMe bdev to create
 filename                | Required | string      | path to device or file (ex: /dev/nvme0n1)
 io_mechanism            | Required | string      | IO mechanism to use (ex: libaio, io_uring, io_uring_cmd, etc.)
+conserve_cpu            | Optional | boolean     | Whether or not to conserve CPU when polling (default: false)
 
 #### Result
 
@@ -5754,7 +5755,8 @@ Example request:
   "params": {
     "name": "bdev_ng0n1",
     "filename": "/dev/ng0n1",
-    "io_mechanism": "io_uring_cmd"
+    "io_mechanism": "io_uring_cmd",
+    "conserve_cpu": false,
   }
 }
 ~~~

module/bdev/xnvme/bdev_xnvme.c

@@ -39,6 +39,7 @@ struct bdev_xnvme {
 	char			*io_mechanism;
 	struct xnvme_dev	*dev;
 	uint32_t		nsid;
+	bool			conserve_cpu;
 
 	TAILQ_ENTRY(bdev_xnvme) link;
 };
@@ -68,6 +69,7 @@ bdev_xnvme_config_json(struct spdk_json_write_ctx *w)
 		spdk_json_write_named_string(w, "name", xnvme->bdev.name);
 		spdk_json_write_named_string(w, "filename", xnvme->filename);
 		spdk_json_write_named_string(w, "io_mechanism", xnvme->io_mechanism);
+		spdk_json_write_named_bool(w, "conserve_cpu", xnvme->conserve_cpu);
 		spdk_json_write_object_end(w);
 
 		spdk_json_write_object_end(w);
@@ -304,7 +306,8 @@ bdev_xnvme_queue_destroy_cb(void *io_device, void *ctx_buf)
 }
 
 struct spdk_bdev *
-create_xnvme_bdev(const char *name, const char *filename, const char *io_mechanism)
+create_xnvme_bdev(const char *name, const char *filename, const char *io_mechanism,
+		  bool conserve_cpu)
 {
 	struct bdev_xnvme *xnvme;
 	uint32_t block_size;
@@ -328,8 +331,14 @@ create_xnvme_bdev(const char *name, const char *filename, const char *io_mechani
 		goto error_return;
 	}
 
-	if (!strcmp(xnvme->io_mechanism, "io_uring_cmd")) {
+	if (!conserve_cpu) {
-		opts.poll_sq = 1;
+		if (!strcmp(xnvme->io_mechanism, "libaio")) {
+			opts.poll_io = 1;
+		} else if (!strcmp(xnvme->io_mechanism, "io_uring")) {
+			opts.poll_io = 1;
+		} else if (!strcmp(xnvme->io_mechanism, "io_uring_cmd")) {
+			opts.poll_sq = 1;
+		}
 	}
 
 	xnvme->filename = strdup(filename);

module/bdev/xnvme/bdev_xnvme.h

@@ -16,7 +16,7 @@
 typedef void (*spdk_delete_xnvme_complete)(void *cb_arg, int bdeverrno);
 
 struct spdk_bdev *create_xnvme_bdev(const char *name, const char *filename,
-				    const char *io_mechanism);
+				    const char *io_mechanism, bool conserve_cpu);
 
 void delete_xnvme_bdev(struct spdk_bdev *bdev, spdk_delete_xnvme_complete cb_fn, void *cb_arg);
 

module/bdev/xnvme/bdev_xnvme_rpc.c

@@ -14,6 +14,7 @@ struct rpc_create_xnvme {
 	char *name;
 	char *filename;
 	char *io_mechanism;
+	bool conserve_cpu;
 };
 
 /* Free the allocated memory resource after the RPC handling. */
@@ -30,6 +31,7 @@ static const struct spdk_json_object_decoder rpc_create_xnvme_decoders[] = {
 	{"name", offsetof(struct rpc_create_xnvme, name), spdk_json_decode_string},
 	{"filename", offsetof(struct rpc_create_xnvme, filename), spdk_json_decode_string},
 	{"io_mechanism", offsetof(struct rpc_create_xnvme, io_mechanism), spdk_json_decode_string},
+	{"conserve_cpu", offsetof(struct rpc_create_xnvme, conserve_cpu), spdk_json_decode_bool, true},
 };
 
 static void
@@ -57,7 +59,7 @@ rpc_bdev_xnvme_create(struct spdk_jsonrpc_request *request,
 		goto cleanup;
 	}
 
-	bdev = create_xnvme_bdev(req.name, req.filename, req.io_mechanism);
+	bdev = create_xnvme_bdev(req.name, req.filename, req.io_mechanism, req.conserve_cpu);
 	if (!bdev) {
 		SPDK_ERRLOG("Unable to create xNVMe bdev from file %s\n", req.filename);
 		spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INTERNAL_ERROR,

python/spdk/rpc/bdev.py

@@ -492,20 +492,25 @@ def bdev_uring_delete(client, name):
     return client.call('bdev_uring_delete', params)
 
 
-def bdev_xnvme_create(client, filename, name, io_mechanism):
+def bdev_xnvme_create(client, filename, name, io_mechanism, conserve_cpu=None):
     """Create a bdev with xNVMe backend.
 
     Args:
         filename: path to device or file (ex: /dev/nvme0n1)
         name: name of xNVMe bdev to create
         io_mechanism: I/O mechanism to use (ex: io_uring, io_uring_cmd, etc.)
+        conserve_cpu: Whether or not to conserve CPU when polling (default: False)
 
     Returns:
         Name of created bdev.
     """
-    params = {'name': name,
+    params = {
-              'filename': filename,
+        'name': name,
-              'io_mechanism': io_mechanism}
+        'filename': filename,
+        'io_mechanism': io_mechanism,
+    }
+    if conserve_cpu:
+        params['conserve_cpu'] = conserve_cpu
 
     return client.call('bdev_xnvme_create', params)
 

scripts/rpc.py

@@ -525,6 +525,7 @@ if __name__ == "__main__":
     p.add_argument('filename', help='Path to device or file (ex: /dev/nvme0n1)')
     p.add_argument('name', help='name of xNVMe bdev to create')
     p.add_argument('io_mechanism', help='IO mechanism to use (ex: libaio, io_uring, io_uring_cmd, etc.)')
+    p.add_argument('conserve_cpu', action='store_true', help='Whether or not to conserve CPU when polling')
     p.set_defaults(func=bdev_xnvme_create)
 
     def bdev_xnvme_delete(args):