Skip to contentSkip to conversation

traffic based keep alive support (TP 4024)

PATCH New Sagi Grimberg 2 participants 6 messages Nov 2, 2018 – Nov 8, 2018

Subsystems (2)
NVMe NVMe target
lore
Share & tools
lore

Sagi Grimberg

author maintainer
#0 lore raw

v3 cover traffic based keep alive support (TP 4024)

Traffic Based Keep Alive TP (4024) was ratified recently. The technical proposal introduces a mode where keep alive commands may be omitted based on the presence of admin or io traffic during the keep alive timeout period. The rational is that normal command execution is sufficient indication of host, controller and transport health so the keep alive command is redundant.

Moreover, There is no way to guarantee in the various transports that a keep alive command will not be starved by heavy traffic, causing the keep alive timeout to expire. There were a number bug reports on linux nvme community on this exact issue.

The implementation simply a controller-wide indicator that is cleared upon reset of the keep alive timer and set when a command (target) or completion (host) is processed. When the keep alive timer expires, we check if any command/completion were processed, we restart the keep alive timer even if a keep alive command was not executed.

Changes from v2:
- fixed debug messages for traffic based keep alive
- collected review tags

Changes from v1:
- omit comp_seen update for controllers that don't support keep alive

Sagi Grimberg (4):
nvme: introduce ctrl attributes enumeration
nvmet: support for traffic based keep-alive
nvme-core: cache controller attributes
nvme-core: support traffic based keep-alive based on controller
support

 drivers/nvme/host/core.c        | 14 ++++++++++++++
 drivers/nvme/host/nvme.h        |  2 ++
 drivers/nvme/target/admin-cmd.c |  3 ++-
 drivers/nvme/target/core.c      | 12 ++++++++++++
 drivers/nvme/target/nvmet.h     |  2 ++
 include/linux/nvme.h            |  5 +++++
 6 files changed, 37 insertions(+), 1 deletion(-)
Signature · 1 line
2.17.1
author maintainer
#1 lore raw

PATCH v3 1/4 nvme: introduce ctrl attributes enumeration

We are growing more controller attributes, so use
a proper enumeration for it. For now just add the
128-bit hostid which we support.

Reviewed-by Chaitanya Kulkarni <chaitanya.kulkarni at wdc.com>
Reviewed-by Hannes Reinecke <hare at suse.com>
Signed-off-by Sagi Grimberg <sagi at grimberg.me>
---
 drivers/nvme/target/admin-cmd.c | 2 +-
 include/linux/nvme.h            | 4 ++++
 2 files changed, 5 insertions(+), 1 deletion(-)
drivers/nvme/target/admin-cmd.c +1 -1
index 1179f6314323..30778ffc46f5 100644
@@ -304,7 +304,7 @@ static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
 
 	/* XXX: figure out what to do about RTD3R/RTD3 */
 	id->oaes = cpu_to_le32(NVMET_AEN_CFG_OPTIONAL);
-	id->ctratt = cpu_to_le32(1 << 0);
+	id->ctratt = cpu_to_le32(NVME_CTRL_ATTR_HID_128_BIT);
 
 	id->oacs = 0;
include/linux/nvme.h +4 -0
index 818dbe9331be..753c83a5c01f 100644
@@ -198,6 +198,10 @@ enum {
 	NVME_PS_FLAGS_NON_OP_STATE	= 1 << 1,
 };
 
+enum nvme_ctrl_attr {
+	NVME_CTRL_ATTR_HID_128_BIT	= (1 << 0),
+};
+
 struct nvme_id_ctrl {
 	__le16			vid;
 	__le16			ssvid;
Signature · 1 line
2.17.1
author maintainer
#2 lore raw

PATCH v3 2/4 nvmet: support for traffic based keep-alive

In reply to Sagi Grimberg

A controller that supports traffic based keep-alive can restart its keep alive timer even if keep-alive was not issued in the kato period but other admin or io commands was. For each command set ctrl->cmd_seen to true, and when keep-alive timer expires, if any commands were seen, resched ka_work instead of escalating to a fatal error.

Signed-off-by Sagi Grimberg <sagi at grimberg.me>
---
 drivers/nvme/target/admin-cmd.c |  3 ++-
 drivers/nvme/target/core.c      | 12 ++++++++++++
 drivers/nvme/target/nvmet.h     |  2 ++
 include/linux/nvme.h            |  1 +
 4 files changed, 17 insertions(+), 1 deletion(-)
drivers/nvme/target/admin-cmd.c +2 -1
index 30778ffc46f5..c9c6d25a3ec2 100644
@@ -304,7 +304,8 @@ static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
 
 	/* XXX: figure out what to do about RTD3R/RTD3 */
 	id->oaes = cpu_to_le32(NVMET_AEN_CFG_OPTIONAL);
-	id->ctratt = cpu_to_le32(NVME_CTRL_ATTR_HID_128_BIT);
+	id->ctratt = cpu_to_le32(NVME_CTRL_ATTR_HID_128_BIT |
+		NVME_CTRL_ATTR_TBKAS);
 
 	id->oacs = 0;
drivers/nvme/target/core.c +12 -0
index 0acdff9e6842..b9fbb60921a4 100644
@@ -298,6 +298,15 @@ static void nvmet_keep_alive_timer(struct work_struct *work)
 {
 	struct nvmet_ctrl *ctrl = container_of(to_delayed_work(work),
 			struct nvmet_ctrl, ka_work);
+	bool cmd_seen = ctrl->cmd_seen;
+
+	ctrl->cmd_seen = false;
+	if (cmd_seen) {
+		pr_debug("ctrl %d reschedule traffic based keep-alive timer\n",
+			ctrl->cntlid);
+		schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
+		return;
+	}
 
 	pr_err("ctrl %d keep-alive timer (%d seconds) expired!\n",
 		ctrl->cntlid, ctrl->kato);

@@ -700,6 +709,9 @@ bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
 		goto fail;
 	}
 
+	if (sq->ctrl)
+		sq->ctrl->cmd_seen = true;
+
 	return true;
 
 fail:
drivers/nvme/target/nvmet.h +2 -0
index 08f7b57a1203..c5255e743f63 100644
@@ -154,6 +154,8 @@ struct nvmet_ctrl {
 	struct nvmet_cq		**cqs;
 	struct nvmet_sq		**sqs;
 
+	bool			cmd_seen;
+
 	struct mutex		lock;
 	u64			cap;
 	u32			cc;
include/linux/nvme.h +1 -0
index 753c83a5c01f..429c4cf90899 100644
@@ -200,6 +200,7 @@ enum {
 
 enum nvme_ctrl_attr {
 	NVME_CTRL_ATTR_HID_128_BIT	= (1 << 0),
+	NVME_CTRL_ATTR_TBKAS		= (1 << 6),
 };
 
 struct nvme_id_ctrl {
Signature · 1 line
2.17.1
author maintainer
#3 lore raw

PATCH v3 3/4 nvme-core: cache controller attributes

In reply to Sagi Grimberg

We get the controller attributes in identify, cache them as we'll need them for traffic based keep alive support.

Reviewed-by Chaitanya Kulkarni <chaitanya.kulkarni at wdc.com>
Signed-off-by Sagi Grimberg <sagi at grimberg.me>
---
 drivers/nvme/host/core.c | 1 +
 drivers/nvme/host/nvme.h | 1 +
 2 files changed, 2 insertions(+)
drivers/nvme/host/core.c +1 -0
index 65c42448e904..ef609239c9c5 100644
@@ -2409,6 +2409,7 @@ int nvme_init_identify(struct nvme_ctrl *ctrl)
 	ctrl->sgls = le32_to_cpu(id->sgls);
 	ctrl->kas = le16_to_cpu(id->kas);
 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
+	ctrl->ctratt = le32_to_cpu(id->ctratt);
 
 	if (id->rtd3e) {
 		/* us -> s */
drivers/nvme/host/nvme.h +1 -0
index 9fefba039d1e..4dd7535caf1b 100644
@@ -193,6 +193,7 @@ struct nvme_ctrl {
 	u8 apsta;
 	u32 oaes;
 	u32 aen_result;
+	u32 ctratt;
 	unsigned int shutdown_timeout;
 	unsigned int kato;
 	bool subsystem;
Signature · 1 line
2.17.1
author maintainer
#4 lore raw

PATCH v3 4/4 nvme-core: support traffic based keep-alive based on controller support

In reply to Sagi Grimberg

If the controller supports traffic based keep alive, we restart the keep alive timer if any admin or io commands was completed during the kato period. This prevents a possible starvation of keep alive commands in the presence of heavy traffic as in such case, we already have a health indication from the host perspective.

Only set a comp_seen indicator in case the controller supports keep alive to minimize the overhead for pci controllers.

Signed-off-by Sagi Grimberg <sagi at grimberg.me>
---
 drivers/nvme/host/core.c | 13 +++++++++++++
 drivers/nvme/host/nvme.h |  1 +
 2 files changed, 14 insertions(+)
drivers/nvme/host/core.c +13 -0
index ef609239c9c5..a24e3321828d 100644
@@ -251,6 +251,9 @@ void nvme_complete_rq(struct request *req)
 
 	trace_nvme_complete_rq(req);
 
+	if (nvme_req(req)->ctrl->kas)
+		nvme_req(req)->ctrl->comp_seen = true;
+
 	if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 		if ((req->cmd_flags & REQ_NVME_MPATH) &&
 		    blk_path_error(status)) {

@@ -841,6 +844,7 @@ static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 		return;
 	}
 
+	ctrl->comp_seen = false;
 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 }
 

@@ -865,6 +869,15 @@ static void nvme_keep_alive_work(struct work_struct *work)
 {
 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 			struct nvme_ctrl, ka_work);
+	bool comp_seen = ctrl->comp_seen;
+
+	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
+		dev_dbg(ctrl->device,
+			"reschedule traffic based keep-alive timer\n");
+		ctrl->comp_seen = false;
+		schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
+		return;
+	}
 
 	if (nvme_keep_alive(ctrl)) {
 		/* allocation failure, reset the controller */
drivers/nvme/host/nvme.h +1 -0
index 4dd7535caf1b..6206d026d53b 100644
@@ -145,6 +145,7 @@ enum nvme_ctrl_state {
 };
 
 struct nvme_ctrl {
+	bool comp_seen;
 	enum nvme_ctrl_state state;
 	bool identified;
 	spinlock_t lock;
Signature · 1 line
2.17.1