Skip to contentSkip to conversation

Introduce new max-queue-size configuration

PATCH In review Max Gurtovoy 3 participants 12 messages Jan 23, 2024 – Feb 26, 2024

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

v3 cover Introduce new max-queue-size configuration

Hi Christoph/Sagi/Keith,
This patch series is mainly for adding an interface for a user to configure the maximal queue size for fabrics via port configfs. Using this interface a user will be able to better control the system and HW resources.

Also, I've increased the maximal queue depth for RDMA controllers to be 256 after request from Guixin Liu. This new value will be valid only for controllers that don't support PI.

While developing this feature I've made some minor cleanups as well.

Changes from v2:
- collected Reviewed-by signatures (Sagi and Christoph)
- added local variable to simplify code in patch 6/8 (Christoph)

Changes from v1:
- collected Reviewed-by signatures (Sagi and Guixin Liu)
- removed the patches that unify fabric host and target max/min/default queue size definitions (Sagi)
- align MQES and SQ size according to the NVMe Spec (patch 2/8)

Max Gurtovoy (8):
nvme-rdma: move NVME_RDMA_IP_PORT from common file
nvmet: compare mqes and sqsize only for IO SQ
nvmet: set maxcmd to be per controller
nvmet: set ctrl pi_support cap before initializing cap reg
nvme-rdma: introduce NVME_RDMA_MAX_METADATA_QUEUE_SIZE definition
nvme-rdma: clamp queue size according to ctrl cap
nvmet: introduce new max queue size configuration entry
nvmet-rdma: set max_queue_size for RDMA transport

 drivers/nvme/host/rdma.c          | 14 ++++++++++----
 drivers/nvme/target/admin-cmd.c   |  2 +-
 drivers/nvme/target/configfs.c    | 28 ++++++++++++++++++++++++++++
 drivers/nvme/target/core.c        | 18 ++++++++++++++++--
 drivers/nvme/target/discovery.c   |  2 +-
 drivers/nvme/target/fabrics-cmd.c |  5 ++---
 drivers/nvme/target/nvmet.h       |  6 ++++--
 drivers/nvme/target/passthru.c    |  2 +-
 drivers/nvme/target/rdma.c        | 10 ++++++++++
 include/linux/nvme-rdma.h         |  6 +++++-
 include/linux/nvme.h              |  2 --
 11 files changed, 78 insertions(+), 17 deletions(-)
Signature · 1 line
2.18.1
#1 lore raw

PATCH v3 1/8 nvme-rdma: move NVME_RDMA_IP_PORT from common file

The correct place for this definition is the nvme rdma header file and not the common nvme header file.

---
 include/linux/nvme-rdma.h | 2 ++
 include/linux/nvme.h      | 2 --
 2 files changed, 2 insertions(+), 2 deletions(-)
include/linux/nvme-rdma.h +2 -0
index 4dd7e6fe92fb..146dd2223a5f 100644
@@ -6,6 +6,8 @@
 #ifndef _LINUX_NVME_RDMA_H
 #define _LINUX_NVME_RDMA_H
 
+#define NVME_RDMA_IP_PORT		4420
+
 #define NVME_RDMA_MAX_QUEUE_SIZE	128
 
 enum nvme_rdma_cm_fmt {
include/linux/nvme.h +0 -2
index 462c21e0e417..ee28acdb8d43 100644
@@ -23,8 +23,6 @@
 
 #define NVME_DISC_SUBSYS_NAME	"nqn.2014-08.org.nvmexpress.discovery"
 
-#define NVME_RDMA_IP_PORT	4420
-
 #define NVME_NSID_ALL		0xffffffff
 
 enum nvme_subsys_type {
Signature · 1 line
2.18.1
#2 lore raw

PATCH v3 2/8 nvmet: compare mqes and sqsize only for IO SQ

In reply to Max Gurtovoy

According to the NVMe Spec:
"
MQES: This field indicates the maximum individual queue size that the controller supports. For NVMe over PCIe implementations, this value applies to the I/O Submission Queues and I/O Completion Queues that the host creates. For NVMe over Fabrics implementations, this value applies to only the I/O Submission Queues that the host creates.
"

Align the target code to compare mqes and sqsize as mentioned in the NVMe Spec.

---
 drivers/nvme/target/fabrics-cmd.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/nvme/target/fabrics-cmd.c +2 -1
index d8da840a1c0e..4d014c5d0b6a 100644
@@ -157,7 +157,8 @@ static u16 nvmet_install_queue(struct nvmet_ctrl *ctrl, struct nvmet_req *req)
 		return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
 	}
 
-	if (sqsize > mqes) {
+	/* for fabrics, this value applies to only the I/O Submission Queues */
+	if (qid && sqsize > mqes) {
 		pr_warn("sqsize %u is larger than MQES supported %u cntlid %d\n",
 				sqsize, mqes, ctrl->cntlid);
 		req->error_loc = offsetof(struct nvmf_connect_command, sqsize);
Signature · 1 line
2.18.1
#3 lore raw

PATCH v3 3/8 nvmet: set maxcmd to be per controller

In reply to Max Gurtovoy

This is a preparation for having a dynamic configuration of max queue size for a controller. Make sure that the maxcmd field stays the same as the MQES (+1) value as we do today.

---
 drivers/nvme/target/admin-cmd.c | 2 +-
 drivers/nvme/target/discovery.c | 2 +-
 drivers/nvme/target/nvmet.h     | 2 +-
 drivers/nvme/target/passthru.c  | 2 +-
 4 files changed, 4 insertions(+), 4 deletions(-)
drivers/nvme/target/admin-cmd.c +1 -1
index 39cb570f833d..f5b7054a4a05 100644
@@ -428,7 +428,7 @@ static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
 	id->cqes = (0x4 << 4) | 0x4;
 
 	/* no enforcement soft-limit for maxcmd - pick arbitrary high value */
-	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
+	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD(ctrl));
 
 	id->nn = cpu_to_le32(NVMET_MAX_NAMESPACES);
 	id->mnan = cpu_to_le32(NVMET_MAX_NAMESPACES);
drivers/nvme/target/discovery.c +1 -1
index 668d257fa986..0d5014905069 100644
@@ -282,7 +282,7 @@ static void nvmet_execute_disc_identify(struct nvmet_req *req)
 	id->lpa = (1 << 2);
 
 	/* no enforcement soft-limit for maxcmd - pick arbitrary high value */
-	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
+	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD(ctrl));
 
 	id->sgls = cpu_to_le32(1 << 0);	/* we always support SGLs */
 	if (ctrl->ops->flags & NVMF_KEYED_SGLS)
drivers/nvme/target/nvmet.h +1 -1
index 6c8acebe1a1a..144aca2fa6ad 100644
@@ -545,7 +545,7 @@ void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
 
 #define NVMET_QUEUE_SIZE	1024
 #define NVMET_NR_QUEUES		128
-#define NVMET_MAX_CMD		NVMET_QUEUE_SIZE
+#define NVMET_MAX_CMD(ctrl)	(NVME_CAP_MQES(ctrl->cap) + 1)
 
 /*
  * Nice round number that makes a list of nsids fit into a page.
drivers/nvme/target/passthru.c +1 -1
index f2d963e1fe94..bb4a69d538fd 100644
@@ -132,7 +132,7 @@ static u16 nvmet_passthru_override_id_ctrl(struct nvmet_req *req)
 
 	id->sqes = min_t(__u8, ((0x6 << 4) | 0x6), id->sqes);
 	id->cqes = min_t(__u8, ((0x4 << 4) | 0x4), id->cqes);
-	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
+	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD(ctrl));
 
 	/* don't support fuse commands */
 	id->fuses = 0;
Signature · 1 line
2.18.1
#4 lore raw

PATCH v3 4/8 nvmet: set ctrl pi_support cap before initializing cap reg

In reply to Max Gurtovoy

This is a preparation for setting the maximal queue size of a controller that supports PI.

---
 drivers/nvme/target/core.c        | 1 +
 drivers/nvme/target/fabrics-cmd.c | 2 --
 2 files changed, 1 insertion(+), 2 deletions(-)
drivers/nvme/target/core.c +1 -0
index d26aa30f8702..ade5a7bd7f6d 100644
@@ -1411,6 +1411,7 @@ u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
 
 	kref_init(&ctrl->ref);
 	ctrl->subsys = subsys;
+	ctrl->pi_support = ctrl->port->pi_enable && ctrl->subsys->pi_support;
 	nvmet_init_cap(ctrl);
 	WRITE_ONCE(ctrl->aen_enabled, NVMET_AEN_CFG_OPTIONAL);
 
drivers/nvme/target/fabrics-cmd.c +0 -2
index 4d014c5d0b6a..08e9c6b6f551 100644
@@ -252,8 +252,6 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)
 	if (status)
 		goto out;
 
-	ctrl->pi_support = ctrl->port->pi_enable && ctrl->subsys->pi_support;
-
 	uuid_copy(&ctrl->hostid, &d->hostid);
 
 	ret = nvmet_setup_auth(ctrl);
Signature · 1 line
2.18.1
#5 lore raw

PATCH v3 5/8 nvme-rdma: introduce NVME_RDMA_MAX_METADATA_QUEUE_SIZE definition

In reply to Max Gurtovoy

This definition will be used by controllers that are configured with metadata support. For now, both regular and metadata controllers have the same maximal queue size but later commit will increase the maximal queue size for regular RDMA controllers to 256.
We'll keep the maximal queue size for metadata controllers to be 128 since there are more resources that are needed for metadata operations and 128 is the optimal size found for metadata controllers base on testing.

---
 drivers/nvme/target/rdma.c | 2 ++
 include/linux/nvme-rdma.h  | 3 ++-
 2 files changed, 4 insertions(+), 1 deletion(-)
drivers/nvme/target/rdma.c +2 -0
index 667f9c04f35d..aaf9c891b2c7 100644
@@ -2015,6 +2015,8 @@ static u8 nvmet_rdma_get_mdts(const struct nvmet_ctrl *ctrl)
 
 static u16 nvmet_rdma_get_max_queue_size(const struct nvmet_ctrl *ctrl)
 {
+	if (ctrl->pi_support)
+		return NVME_RDMA_MAX_METADATA_QUEUE_SIZE;
 	return NVME_RDMA_MAX_QUEUE_SIZE;
 }
 
include/linux/nvme-rdma.h +2 -1
index 146dd2223a5f..d0b9941911a1 100644
@@ -8,7 +8,8 @@
 
 #define NVME_RDMA_IP_PORT		4420
 
-#define NVME_RDMA_MAX_QUEUE_SIZE	128
+#define NVME_RDMA_MAX_QUEUE_SIZE 128
+#define NVME_RDMA_MAX_METADATA_QUEUE_SIZE 128
 
 enum nvme_rdma_cm_fmt {
 	NVME_RDMA_CM_FMT_1_0 = 0x0,
Signature · 1 line
2.18.1
#6 lore raw

PATCH v3 6/8 nvme-rdma: clamp queue size according to ctrl cap

In reply to Max Gurtovoy

If a controller is configured with metadata support, clamp the maximal queue size to be 128 since there are more resources that are needed for metadata operations. Otherwise, clamp it to 256.

---
 drivers/nvme/host/rdma.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)
drivers/nvme/host/rdma.c +10 -4
index 2e77c0f25f71..ef1cb9a0b90d 100644
@@ -1005,6 +1005,7 @@ static int nvme_rdma_setup_ctrl(struct nvme_rdma_ctrl *ctrl, bool new)
 {
 	int ret;
 	bool changed;
+	u16 max_queue_size;
 
 	ret = nvme_rdma_configure_admin_queue(ctrl, new);
 	if (ret)
@@ -1029,11 +1030,16 @@ static int nvme_rdma_setup_ctrl(struct nvme_rdma_ctrl *ctrl, bool new)
 			ctrl->ctrl.opts->queue_size, ctrl->ctrl.sqsize + 1);
 	}
 
-	if (ctrl->ctrl.sqsize + 1 > NVME_RDMA_MAX_QUEUE_SIZE) {
+	if (ctrl->ctrl.max_integrity_segments)
+		max_queue_size = NVME_RDMA_MAX_METADATA_QUEUE_SIZE;
+	else
+		max_queue_size = NVME_RDMA_MAX_QUEUE_SIZE;
+
+	if (ctrl->ctrl.sqsize + 1 > max_queue_size) {
 		dev_warn(ctrl->ctrl.device,
-			"ctrl sqsize %u > max queue size %u, clamping down\n",
-			ctrl->ctrl.sqsize + 1, NVME_RDMA_MAX_QUEUE_SIZE);
-		ctrl->ctrl.sqsize = NVME_RDMA_MAX_QUEUE_SIZE - 1;
+			 "ctrl sqsize %u > max queue size %u, clamping down\n",
+			 ctrl->ctrl.sqsize + 1, max_queue_size);
+		ctrl->ctrl.sqsize = max_queue_size - 1;
 	}
 
 	if (ctrl->ctrl.sqsize + 1 > ctrl->ctrl.maxcmd) {
Signature · 1 line
2.18.1
#8 lore raw

PATCH v3 7/8 nvmet: introduce new max queue size configuration entry

In reply to Max Gurtovoy

Using this port configuration, one will be able to set the maximal queue size to be used for any controller that will be associated to the configured port.

The default value stayed 1024 but each transport will be able to set the its own values before enabling the port.

Introduce lower limit of 16 for minimal queue depth (same as we use in the host fabrics drivers).

---
 drivers/nvme/target/configfs.c | 28 ++++++++++++++++++++++++++++
 drivers/nvme/target/core.c     | 17 +++++++++++++++--
 drivers/nvme/target/nvmet.h    |  4 +++-
 3 files changed, 46 insertions(+), 3 deletions(-)
drivers/nvme/target/configfs.c +28 -0
index bd514d4c4a5b..f8df2ef715ba 100644
@@ -272,6 +272,32 @@ static ssize_t nvmet_param_inline_data_size_store(struct config_item *item,
 
 CONFIGFS_ATTR(nvmet_, param_inline_data_size);
 
+static ssize_t nvmet_param_max_queue_size_show(struct config_item *item,
+		char *page)
+{
+	struct nvmet_port *port = to_nvmet_port(item);
+
+	return snprintf(page, PAGE_SIZE, "%d\n", port->max_queue_size);
+}
+
+static ssize_t nvmet_param_max_queue_size_store(struct config_item *item,
+		const char *page, size_t count)
+{
+	struct nvmet_port *port = to_nvmet_port(item);
+	int ret;
+
+	if (nvmet_is_port_enabled(port, __func__))
+		return -EACCES;
+	ret = kstrtoint(page, 0, &port->max_queue_size);
+	if (ret) {
+		pr_err("Invalid value '%s' for max_queue_size\n", page);
+		return -EINVAL;
+	}
+	return count;
+}
+
+CONFIGFS_ATTR(nvmet_, param_max_queue_size);
+
 #ifdef CONFIG_BLK_DEV_INTEGRITY
 static ssize_t nvmet_param_pi_enable_show(struct config_item *item,
 		char *page)
@@ -1856,6 +1882,7 @@ static struct configfs_attribute *nvmet_port_attrs[] = {
 	&nvmet_attr_addr_trtype,
 	&nvmet_attr_addr_tsas,
 	&nvmet_attr_param_inline_data_size,
+	&nvmet_attr_param_max_queue_size,
 #ifdef CONFIG_BLK_DEV_INTEGRITY
 	&nvmet_attr_param_pi_enable,
 #endif
@@ -1914,6 +1941,7 @@ static struct config_group *nvmet_ports_make(struct config_group *group,
 	INIT_LIST_HEAD(&port->subsystems);
 	INIT_LIST_HEAD(&port->referrals);
 	port->inline_data_size = -1;	/* < 0 == let the transport choose */
+	port->max_queue_size = -1;	/* < 0 == let the transport choose */
 
 	port->disc_addr.portid = cpu_to_le16(portid);
 	port->disc_addr.adrfam = NVMF_ADDR_FAMILY_MAX;
drivers/nvme/target/core.c +15 -2
index ade5a7bd7f6d..d34520359cc9 100644
@@ -358,6 +358,18 @@ int nvmet_enable_port(struct nvmet_port *port)
 	if (port->inline_data_size < 0)
 		port->inline_data_size = 0;
 
+	/*
+	 * If the transport didn't set the max_queue_size properly, then clamp
+	 * it to the target limits. Also set default values in case the
+	 * transport didn't set it at all.
+	 */
+	if (port->max_queue_size < 0)
+		port->max_queue_size = NVMET_MAX_QUEUE_SIZE;
+	else
+		port->max_queue_size = clamp_t(int, port->max_queue_size,
+					       NVMET_MIN_QUEUE_SIZE,
+					       NVMET_MAX_QUEUE_SIZE);
+
 	port->enabled = true;
 	port->tr_ops = ops;
 	return 0;
@@ -1223,9 +1235,10 @@ static void nvmet_init_cap(struct nvmet_ctrl *ctrl)
 	ctrl->cap |= (15ULL << 24);
 	/* maximum queue entries supported: */
 	if (ctrl->ops->get_max_queue_size)
-		ctrl->cap |= ctrl->ops->get_max_queue_size(ctrl) - 1;
+		ctrl->cap |= min_t(u16, ctrl->ops->get_max_queue_size(ctrl),
+				   ctrl->port->max_queue_size) - 1;
 	else
-		ctrl->cap |= NVMET_QUEUE_SIZE - 1;
+		ctrl->cap |= ctrl->port->max_queue_size - 1;
 
 	if (nvmet_is_passthru_subsys(ctrl->subsys))
 		nvmet_passthrough_override_cap(ctrl);
drivers/nvme/target/nvmet.h +3 -1
index 144aca2fa6ad..7c6e7e65b032 100644
@@ -163,6 +163,7 @@ struct nvmet_port {
 	void				*priv;
 	bool				enabled;
 	int				inline_data_size;
+	int				max_queue_size;
 	const struct nvmet_fabrics_ops	*tr_ops;
 	bool				pi_enable;
 };
@@ -543,7 +544,8 @@ void nvmet_subsys_disc_changed(struct nvmet_subsys *subsys,
 void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
 		u8 event_info, u8 log_page);
 
-#define NVMET_QUEUE_SIZE	1024
+#define NVMET_MIN_QUEUE_SIZE	16
+#define NVMET_MAX_QUEUE_SIZE	1024
 #define NVMET_NR_QUEUES		128
 #define NVMET_MAX_CMD(ctrl)	(NVME_CAP_MQES(ctrl->cap) + 1)
 
Signature · 1 line
2.18.1
#9 lore raw

PATCH v3 8/8 nvmet-rdma: set max_queue_size for RDMA transport

In reply to Max Gurtovoy

A new port configuration was added to set max_queue_size. Clamp user configuration to RDMA transport limits.

Increase the maximal queue size of RDMA controllers from 128 to 256 (the default size stays 128 same as before).

---
 drivers/nvme/target/rdma.c | 8 ++++++++
 include/linux/nvme-rdma.h  | 3 ++-
 2 files changed, 10 insertions(+), 1 deletion(-)
drivers/nvme/target/rdma.c +8 -0
index aaf9c891b2c7..be57446f398b 100644
@@ -1956,6 +1956,14 @@ static int nvmet_rdma_add_port(struct nvmet_port *nport)
 		nport->inline_data_size = NVMET_RDMA_MAX_INLINE_DATA_SIZE;
 	}
 
+	if (nport->max_queue_size < 0) {
+		nport->max_queue_size = NVME_RDMA_DEFAULT_QUEUE_SIZE;
+	} else if (nport->max_queue_size > NVME_RDMA_MAX_QUEUE_SIZE) {
+		pr_warn("max_queue_size %u is too large, reducing to %u\n",
+			nport->max_queue_size, NVME_RDMA_MAX_QUEUE_SIZE);
+		nport->max_queue_size = NVME_RDMA_MAX_QUEUE_SIZE;
+	}
+
 	ret = inet_pton_with_scope(&init_net, af, nport->disc_addr.traddr,
 			nport->disc_addr.trsvcid, &port->addr);
 	if (ret) {
include/linux/nvme-rdma.h +2 -1
index d0b9941911a1..eb2f04d636c8 100644
@@ -8,8 +8,9 @@
 
 #define NVME_RDMA_IP_PORT		4420
 
-#define NVME_RDMA_MAX_QUEUE_SIZE 128
+#define NVME_RDMA_MAX_QUEUE_SIZE 256
 #define NVME_RDMA_MAX_METADATA_QUEUE_SIZE 128
+#define NVME_RDMA_DEFAULT_QUEUE_SIZE 128
 
 enum nvme_rdma_cm_fmt {
 	NVME_RDMA_CM_FMT_1_0 = 0x0,
Signature · 1 line
2.18.1
#10 lore raw
In reply to Max Gurtovoy

Hi Keith,
This series was fully reviewed by Sagi and Christoph for 6.8 but we probably somehow missed taking it.

can you please take it for the next 6.9 merge window and/or merge it to nvme-6.9 ?

Quoted 45 lines · Max Gurtovoy

Hi Christoph/Sagi/Keith,
This patch series is mainly for adding an interface for a user to configure the maximal queue size for fabrics via port configfs. Using this interface a user will be able to better control the system and HW resources.

Also, I've increased the maximal queue depth for RDMA controllers to be 256 after request from Guixin Liu. This new value will be valid only for controllers that don't support PI.

While developing this feature I've made some minor cleanups as well.

Changes from v2:
- collected Reviewed-by signatures (Sagi and Christoph)
- added local variable to simplify code in patch 6/8 (Christoph)

Changes from v1:
- collected Reviewed-by signatures (Sagi and Guixin Liu)
- removed the patches that unify fabric host and target max/min/default queue size definitions (Sagi)
- align MQES and SQ size according to the NVMe Spec (patch 2/8)

Max Gurtovoy (8):
nvme-rdma: move NVME_RDMA_IP_PORT from common file
nvmet: compare mqes and sqsize only for IO SQ
nvmet: set maxcmd to be per controller
nvmet: set ctrl pi_support cap before initializing cap reg
nvme-rdma: introduce NVME_RDMA_MAX_METADATA_QUEUE_SIZE definition
nvme-rdma: clamp queue size according to ctrl cap
nvmet: introduce new max queue size configuration entry
nvmet-rdma: set max_queue_size for RDMA transport

  drivers/nvme/host/rdma.c          | 14 ++++++++++----
  drivers/nvme/target/admin-cmd.c   |  2 +-
  drivers/nvme/target/configfs.c    | 28 ++++++++++++++++++++++++++++
  drivers/nvme/target/core.c        | 18 ++++++++++++++++--
  drivers/nvme/target/discovery.c   |  2 +-
  drivers/nvme/target/fabrics-cmd.c |  5 ++---
  drivers/nvme/target/nvmet.h       |  6 ++++--
  drivers/nvme/target/passthru.c    |  2 +-
  drivers/nvme/target/rdma.c        | 10 ++++++++++
  include/linux/nvme-rdma.h         |  6 +++++-
  include/linux/nvme.h              |  2 --
  11 files changed, 78 insertions(+), 17 deletions(-)