Commit 7b397f62 authored by Ben Walker's avatar Ben Walker Committed by Jim Harris
Browse files

nvmf: No longer tie subsystems to CPU cores 



The "Core" parameter in the configuration file has been removed. New
connections are handed out to available cores using round-robin.

Change-Id: I24527fa22a0b2738ebbf5fb030e3bb373ead5da2
Signed-off-by: default avatarBen Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/388295


Tested-by: default avatarSPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: default avatarDaniel Verkamp <daniel.verkamp@intel.com>
Reviewed-by: default avatarJim Harris <james.r.harris@intel.com>
parent 80016bd9

CHANGELOG.md

+5 −0
Original line number Diff line number Diff line
@@ -39,6 +39,11 @@ vfio-pci driver should trigger one "remove" uevent at the very beginning of the 
of pci driver, this will make sure the SPDK could detect the event and release the vfio-attached

device Fd and related resource when the device removed.



### NVMe-oF Target



Subsystems are no longer tied explicitly to CPU cores. Instead, connections are handed out to the available

cores round-robin. The "Core" option in the configuration file has been removed.



### Blobstore



A number of functions have been renamed:

app/nvmf_tgt/conf.c

+19 −110
Original line number Diff line number Diff line
@@ -52,7 +52,7 @@ 
#define ACCEPT_TIMEOUT_US		10000 /* 10ms */



struct spdk_nvmf_probe_ctx {

	struct nvmf_tgt_subsystem	*app_subsystem;

	struct spdk_nvmf_subsystem	*subsystem;

	bool				any;

	bool				found;

	struct spdk_nvme_transport_id	trid;
@@ -61,71 +61,19 @@ struct spdk_nvmf_probe_ctx { 
#define MAX_STRING_LEN 255



struct spdk_nvmf_tgt_conf g_spdk_nvmf_tgt_conf;

static int32_t g_last_core = -1;



static int

spdk_get_numa_node_value(const char *path)

{

	FILE *fd;

	int numa_node = -1;

	char buf[MAX_STRING_LEN];



	fd = fopen(path, "r");

	if (!fd) {

		return -1;

	}



	if (fgets(buf, sizeof(buf), fd) != NULL) {

		numa_node = strtoul(buf, NULL, 10);

	}

	fclose(fd);



	return numa_node;

}



static int

spdk_get_ifaddr_numa_node(const char *if_addr)

{

	int ret;

	struct ifaddrs *ifaddrs, *ifa;

	struct sockaddr_in addr, addr_in;

	char path[MAX_STRING_LEN];

	int numa_node = -1;



	addr_in.sin_addr.s_addr = inet_addr(if_addr);



	ret = getifaddrs(&ifaddrs);

	if (ret < 0) {

		return -1;

	}



	for (ifa = ifaddrs; ifa != NULL; ifa = ifa->ifa_next) {

		addr = *(struct sockaddr_in *)ifa->ifa_addr;

		if ((uint32_t)addr_in.sin_addr.s_addr != (uint32_t)addr.sin_addr.s_addr) {

			continue;

		}

		snprintf(path, MAX_STRING_LEN, "/sys/class/net/%s/device/numa_node", ifa->ifa_name);

		numa_node = spdk_get_numa_node_value(path);

		break;

	}

	freeifaddrs(ifaddrs);



	return numa_node;

}



static int

spdk_add_nvmf_discovery_subsystem(void)

{

	struct nvmf_tgt_subsystem *app_subsys;

	struct spdk_nvmf_subsystem *subsystem;



	app_subsys = nvmf_tgt_create_subsystem(SPDK_NVMF_DISCOVERY_NQN, SPDK_NVMF_SUBTYPE_DISCOVERY, 0,

					       spdk_env_get_current_core());

	if (app_subsys == NULL) {

	subsystem = nvmf_tgt_create_subsystem(SPDK_NVMF_DISCOVERY_NQN, SPDK_NVMF_SUBTYPE_DISCOVERY, 0);

	if (subsystem == NULL) {

		SPDK_ERRLOG("Failed creating discovery nvmf library subsystem\n");

		return -1;

	}



	spdk_nvmf_subsystem_set_allow_any_host(app_subsys->subsystem, true);

	spdk_nvmf_subsystem_set_allow_any_host(subsystem, true);



	return 0;

}
@@ -196,27 +144,6 @@ spdk_nvmf_parse_nvmf_tgt(void) 
	return 0;

}



static int

spdk_nvmf_allocate_lcore(uint64_t mask, uint32_t lcore)

{

	uint32_t end;



	if (lcore == 0) {

		end = 0;

	} else {

		end = lcore - 1;

	}



	do {

		if (((mask >> lcore) & 1U) == 1U) {

			break;

		}

		lcore = (lcore + 1) % 64;

	} while (lcore != end);



	return lcore;

}



static int

spdk_nvmf_parse_subsystem(struct spdk_conf_section *sp)

{
@@ -253,6 +180,15 @@ spdk_nvmf_parse_subsystem(struct spdk_conf_section *sp) 
		}

	}



	/* Core is no longer a valid parameter, but print out a nice

	 * message if it exists to inform users.

	 */

	if (lcore >= 0) {

		SPDK_NOTICELOG("Core present in the [Subsystem] section of the config file.\n"

			       "Core was removed as an option. Subsystems can now run on all available cores.\n");

		SPDK_NOTICELOG("Please remove Core from your configuration file. Ignoring it and continuing.\n");

	}



	/* Parse Listen sections */

	num_listen_addrs = 0;

	for (i = 0; i < MAX_LISTEN_ADDRESSES; i++) {
@@ -322,7 +258,7 @@ spdk_nvmf_parse_subsystem(struct spdk_conf_section *sp) 
		num_ns++;

	}



	ret = spdk_nvmf_construct_subsystem(nqn, lcore,

	ret = spdk_nvmf_construct_subsystem(nqn,

					    num_listen_addrs, listen_addrs,

					    num_hosts, hosts, allow_any_host,

					    sn,
@@ -375,16 +311,14 @@ spdk_nvmf_parse_conf(void) 
}



int

spdk_nvmf_construct_subsystem(const char *name, int32_t lcore,

spdk_nvmf_construct_subsystem(const char *name,

			      int num_listen_addresses, struct rpc_listen_address *addresses,

			      int num_hosts, char *hosts[], bool allow_any_host,

			      const char *sn, size_t num_ns, struct spdk_nvmf_ns_params *ns_list)

{

	struct spdk_nvmf_subsystem *subsystem;

	struct nvmf_tgt_subsystem *app_subsys;

	int i, rc;

	size_t j;

	uint64_t mask;

	struct spdk_bdev *bdev;



	if (name == NULL) {
@@ -402,40 +336,16 @@ spdk_nvmf_construct_subsystem(const char *name, int32_t lcore, 
		return -1;

	}



	if (lcore < 0) {

		lcore = ++g_last_core;

	}



	/* Determine which core to assign to the subsystem */

	mask = spdk_app_get_core_mask();

	lcore = spdk_nvmf_allocate_lcore(mask, lcore);

	g_last_core = lcore;



	app_subsys = nvmf_tgt_create_subsystem(name, SPDK_NVMF_SUBTYPE_NVME, num_ns, lcore);

	if (app_subsys == NULL) {

	subsystem = nvmf_tgt_create_subsystem(name, SPDK_NVMF_SUBTYPE_NVME, num_ns);

	if (subsystem == NULL) {

		SPDK_ERRLOG("Subsystem creation failed\n");

		return -1;

	}

	subsystem = app_subsys->subsystem;



	/* Parse Listen sections */

	for (i = 0; i < num_listen_addresses; i++) {

		int nic_numa_node = spdk_get_ifaddr_numa_node(addresses[i].traddr);

		unsigned subsys_numa_node = spdk_env_get_socket_id(app_subsys->lcore);

		struct spdk_nvme_transport_id trid = {};



		if (nic_numa_node >= 0) {

			if (subsys_numa_node != (unsigned)nic_numa_node) {

				SPDK_WARNLOG("Subsystem %s is configured to run on a CPU core %d belonging "

					     "to a different NUMA node than the associated NIC. "

					     "This may result in reduced performance.\n",

					     name, lcore);

				SPDK_WARNLOG("The NIC is on socket %d\n", nic_numa_node);

				SPDK_WARNLOG("The Subsystem is on socket %u\n",

					     subsys_numa_node);

			}

		}



		if (spdk_nvme_transport_id_parse_trtype(&trid.trtype, addresses[i].transport)) {

			SPDK_ERRLOG("Missing listen address transport type\n");

			goto error;
@@ -502,7 +412,6 @@ spdk_nvmf_construct_subsystem(const char *name, int32_t lcore, 
	return 0;



error:

	spdk_nvmf_delete_subsystem(app_subsys->subsystem);

	app_subsys->subsystem = NULL;

	spdk_nvmf_delete_subsystem(subsystem);

	return -1;

}

app/nvmf_tgt/nvmf_rpc.c

+16 −11
Original line number Diff line number Diff line
@@ -44,17 +44,13 @@ 
#include "nvmf_tgt.h"



static void

dump_nvmf_subsystem(struct spdk_json_write_ctx *w, struct nvmf_tgt_subsystem *tgt_subsystem)

dump_nvmf_subsystem(struct spdk_json_write_ctx *w, struct spdk_nvmf_subsystem *subsystem)

{

	struct spdk_nvmf_host		*host;

	struct spdk_nvmf_subsystem	*subsystem = tgt_subsystem->subsystem;

	struct spdk_nvmf_listener 	*listener;



	spdk_json_write_object_begin(w);



	spdk_json_write_name(w, "core");

	spdk_json_write_int32(w, tgt_subsystem->lcore);



	spdk_json_write_name(w, "nqn");

	spdk_json_write_string(w, spdk_nvmf_subsystem_get_nqn(subsystem));

	spdk_json_write_name(w, "subtype");
@@ -143,7 +139,7 @@ spdk_rpc_get_nvmf_subsystems(struct spdk_jsonrpc_request *request, 
			     const struct spdk_json_val *params)

{

	struct spdk_json_write_ctx *w;

	struct nvmf_tgt_subsystem	*tgt_subsystem;

	struct spdk_nvmf_subsystem *subsystem;



	if (params != NULL) {

		spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS,
@@ -157,10 +153,10 @@ spdk_rpc_get_nvmf_subsystems(struct spdk_jsonrpc_request *request, 
	}



	spdk_json_write_array_begin(w);

	tgt_subsystem = nvmf_tgt_subsystem_first();

	while (tgt_subsystem) {

		dump_nvmf_subsystem(w, tgt_subsystem);

		tgt_subsystem = nvmf_tgt_subsystem_next(tgt_subsystem);

	subsystem = spdk_nvmf_subsystem_get_first(g_tgt.tgt);

	while (subsystem) {

		dump_nvmf_subsystem(w, subsystem);

		subsystem = spdk_nvmf_subsystem_get_next(subsystem);

	}

	spdk_json_write_array_end(w);

	spdk_jsonrpc_end_result(request, w);
@@ -379,7 +375,16 @@ spdk_rpc_construct_nvmf_subsystem(struct spdk_jsonrpc_request *request, 
		}

	}



	ret = spdk_nvmf_construct_subsystem(req.nqn, req.core,

	/* Core is no longer a valid parameter, but print out a nice

	 * message if it exists to inform users.

	 */

	if (req.core != -1) {

		SPDK_NOTICELOG("Core present in the construct NVMe-oF subsystem RPC.\n"

			       "Core was removed as an option. Subsystems can now run on all available cores.\n");

		SPDK_NOTICELOG("Ignoring it and continuing.\n");

	}



	ret = spdk_nvmf_construct_subsystem(req.nqn,

					    req.listen_addresses.num_listen_address,

					    req.listen_addresses.addresses,

					    req.hosts.num_hosts, req.hosts.hosts, req.allow_any_host,

app/nvmf_tgt/nvmf_tgt.c

+13 −60
Original line number Diff line number Diff line
@@ -54,20 +54,8 @@ static size_t g_active_poll_groups = 0; 


static struct spdk_poller *g_acceptor_poller = NULL;



static TAILQ_HEAD(, nvmf_tgt_subsystem) g_subsystems = TAILQ_HEAD_INITIALIZER(g_subsystems);



static void nvmf_tgt_advance_state(void *arg1, void *arg2);



static void

nvmf_tgt_delete_subsystem(struct nvmf_tgt_subsystem *app_subsys)

{

	TAILQ_REMOVE(&g_subsystems, app_subsys, tailq);



	spdk_nvmf_delete_subsystem(app_subsys->subsystem);



	free(app_subsys);

}



static void

spdk_nvmf_shutdown_cb(void)

{
@@ -79,67 +67,41 @@ spdk_nvmf_shutdown_cb(void) 
	nvmf_tgt_advance_state(NULL, NULL);

}



struct nvmf_tgt_subsystem *

nvmf_tgt_create_subsystem(const char *name, enum spdk_nvmf_subtype subtype, uint32_t num_ns,

			  uint32_t lcore)

struct spdk_nvmf_subsystem *

nvmf_tgt_create_subsystem(const char *name, enum spdk_nvmf_subtype subtype, uint32_t num_ns)

{

	struct spdk_nvmf_subsystem *subsystem;

	struct nvmf_tgt_subsystem *app_subsys;



	if (spdk_nvmf_tgt_find_subsystem(g_tgt.tgt, name)) {

		SPDK_ERRLOG("Subsystem already exist\n");

		return NULL;

	}



	app_subsys = calloc(1, sizeof(*app_subsys));

	if (app_subsys == NULL) {

		SPDK_ERRLOG("Subsystem allocation failed\n");

		return NULL;

	}



	subsystem = spdk_nvmf_create_subsystem(g_tgt.tgt, name, subtype, num_ns);

	if (subsystem == NULL) {

		SPDK_ERRLOG("Subsystem creation failed\n");

		free(app_subsys);

		return NULL;

	}



	app_subsys->subsystem = subsystem;

	app_subsys->lcore = lcore;



	SPDK_NOTICELOG("allocated subsystem %s on lcore %u on socket %u\n", name, lcore,

		       spdk_env_get_socket_id(lcore));

	SPDK_NOTICELOG("allocated subsystem %s\n", name);



	TAILQ_INSERT_TAIL(&g_subsystems, app_subsys, tailq);



	return app_subsys;

	return subsystem;

}



struct nvmf_tgt_subsystem *

nvmf_tgt_subsystem_first(void)

int

nvmf_tgt_shutdown_subsystem_by_nqn(const char *nqn)

{

	return TAILQ_FIRST(&g_subsystems);

}

	struct spdk_nvmf_subsystem *subsystem;



struct nvmf_tgt_subsystem *

nvmf_tgt_subsystem_next(struct nvmf_tgt_subsystem *subsystem)

{

	return TAILQ_NEXT(subsystem, tailq);

	subsystem = spdk_nvmf_tgt_find_subsystem(g_tgt.tgt, nqn);

	if (!subsystem) {

		return -EINVAL;

	}



int

nvmf_tgt_shutdown_subsystem_by_nqn(const char *nqn)

{

	struct nvmf_tgt_subsystem *tgt_subsystem, *subsys_tmp;

	spdk_nvmf_delete_subsystem(subsystem);



	TAILQ_FOREACH_SAFE(tgt_subsystem, &g_subsystems, tailq, subsys_tmp) {

		if (strcmp(spdk_nvmf_subsystem_get_nqn(tgt_subsystem->subsystem), nqn) == 0) {

			nvmf_tgt_delete_subsystem(tgt_subsystem);

	return 0;

}

	}

	return -1;

}



static void

nvmf_tgt_poll_group_add(void *arg1, void *arg2)
@@ -181,7 +143,7 @@ acceptor_poll(void *arg) 
static void

nvmf_tgt_destroy_poll_group_done(void *ctx)

{

	g_tgt.state = NVMF_TGT_FINI_SHUTDOWN_SUBSYSTEMS;

	g_tgt.state = NVMF_TGT_FINI_FREE_RESOURCES;

	nvmf_tgt_advance_state(NULL, NULL);

}
@@ -292,15 +254,6 @@ nvmf_tgt_advance_state(void *arg1, void *arg2) 
					     NULL,

					     nvmf_tgt_destroy_poll_group_done);

			break;

		case NVMF_TGT_FINI_SHUTDOWN_SUBSYSTEMS: {

			struct nvmf_tgt_subsystem *app_subsys, *tmp;



			TAILQ_FOREACH_SAFE(app_subsys, &g_subsystems, tailq, tmp) {

				nvmf_tgt_delete_subsystem(app_subsys);

			}

			g_tgt.state = NVMF_TGT_FINI_FREE_RESOURCES;

			break;

		}

		case NVMF_TGT_FINI_FREE_RESOURCES:

			spdk_nvmf_tgt_destroy(g_tgt.tgt);

			g_tgt.state = NVMF_TGT_STOPPED;

app/nvmf_tgt/nvmf_tgt.h

+2 −19
Original line number Diff line number Diff line
@@ -51,14 +51,6 @@ struct spdk_nvmf_tgt_conf { 
	uint32_t acceptor_poll_rate;

};



struct nvmf_tgt_subsystem {

	struct spdk_nvmf_subsystem *subsystem;



	TAILQ_ENTRY(nvmf_tgt_subsystem) tailq;



	uint32_t lcore;

};



enum nvmf_tgt_state {

	NVMF_TGT_INIT_NONE = 0,

	NVMF_TGT_INIT_PARSE_CONFIG,
@@ -67,7 +59,6 @@ enum nvmf_tgt_state { 
	NVMF_TGT_RUNNING,

	NVMF_TGT_FINI_STOP_ACCEPTOR,

	NVMF_TGT_FINI_DESTROY_POLL_GROUPS,

	NVMF_TGT_FINI_SHUTDOWN_SUBSYSTEMS,

	NVMF_TGT_FINI_FREE_RESOURCES,

	NVMF_TGT_STOPPED,

	NVMF_TGT_ERROR,
@@ -85,17 +76,10 @@ extern struct spdk_nvmf_tgt_conf g_spdk_nvmf_tgt_conf; 


extern struct nvmf_tgt g_tgt;



struct nvmf_tgt_subsystem *

nvmf_tgt_subsystem_first(void);



struct nvmf_tgt_subsystem *

nvmf_tgt_subsystem_next(struct nvmf_tgt_subsystem *subsystem);



int spdk_nvmf_parse_conf(void);



struct nvmf_tgt_subsystem *nvmf_tgt_create_subsystem(const char *name,

		enum spdk_nvmf_subtype subtype, uint32_t num_ns,

		uint32_t lcore);

struct spdk_nvmf_subsystem *nvmf_tgt_create_subsystem(const char *name,

		enum spdk_nvmf_subtype subtype, uint32_t num_ns);



struct spdk_nvmf_ns_params {

	char *bdev_name;
@@ -104,7 +88,6 @@ struct spdk_nvmf_ns_params { 


int

spdk_nvmf_construct_subsystem(const char *name,

			      int32_t lcore,

			      int num_listen_addresses, struct rpc_listen_address *addresses,

			      int num_hosts, char *hosts[], bool allow_any_host,

			      const char *sn, size_t num_ns, struct spdk_nvmf_ns_params *ns_list);