Skip to contentSkip to conversation

NVMeTCP Offload ULP

PATCH In review Prabhakar Kushwaha 6 participants 37 messages Jun 29, 2021 – Jul 16, 2021 NVMe

lore
Share & tools
lore
#0 lore raw

v4 cover NVMeTCP Offload ULP

With the goal of enabling a generic infrastructure that allows NVMe/TCP offload devices like NICs to seamlessly plug into the NVMe-oF stack, this patch series introduces the nvme-tcp-offload ULP host layer, which will be a new transport type called "tcp-offload" and will serve as an abstraction layer to work with device specific nvme-tcp offload drivers.

NVMeTCP offload is a full offload of the NVMeTCP protocol, this includes both the TCP level and the NVMeTCP level.

The nvme-tcp-offload transport can co-exist with the existing tcp and other transports. The tcp offload was designed so that stack changes are kept to a bare minimum: only registering new transports.
All other APIs, ops etc. are identical to the regular tcp transport. Representing the TCP offload as a new transport allows clear and manageable differentiation between the connections which should use the offload path and those that are not offloaded (even on the same device).

The nvme-tcp-offload layers and API compared to nvme-tcp and nvme-rdma:

* NVMe layer: *

       [ nvme/nvme-fabrics/blk-mq ]
             |
        (nvme API and blk-mq API)
             |
             |			 

* Transport layer: *

[ nvme-rdma ] [ nvme-tcp ] [ nvme-tcp-offload ]
| | |
(Verbs)
| | |
| (Socket)
| | |
| | (nvme-tcp-offload API)
| | |
| | |
* Transport Driver: *

             |        |             |
      [ RDMA driver ]       
                      |             |

[ Network driver ]
|
[ NVMeTCP Offload driver ]

Upstream plan:
==============
As discussed in RFCV7, "NVMeTCP Offload ULP and QEDN Device Driver" contains 3 parts:
https://lore.kernel.org/linux-nvme/[email protected]/

This series contains part1 and part3, intended for linux-nvme:
- Part 1: The nvme-tcp-offload patches
- Part 3: Marvell's Offload device driver(qedn) patches. It has "compilation dependency" on both Part 1 and Part 2.

Part 2 is already accepted in net-next.git:
https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=eda1bc65b0dc1b03006e427430ba23746ec44714

Usage:
======
The user will interact with the network-device in order to configure the ip/vlan - Logically similar to the RDMA model.
The NVMeTCP configuration is populated as part of the nvme connect command.

Example:
Assign IP to the net-device (from any existing Linux tool):

    ip addr add 100.100.0.101/24 dev p1p1

This IP will be used by both net-device and offload-device.

In order to connect from "sw" nvme-tcp through the net-device:

    nvme connect -t tcp -s 4420 -a 100.100.0.100 -n testnqn

In order to connect from "offload" nvme-tcp through the offload-device:

    nvme connect -t tcp_offload -s 4420 -a 100.100.0.100 -n testnqn

An alternative approach, and as a future enhancement that will not impact this series will be to modify nvme-cli with a new flag that will determine if "-t tcp" should be the regular nvme-tcp (which will be the default) or nvme-tcp-offload.
Exmaple:

    nvme connect -t tcp -s 4420 -a 100.100.0.100 -n testnqn -[new flag]

Queue Initialization Design:
============================
The nvme-tcp-offload ULP module shall register with the existing nvmf_transport_ops (.name = "tcp_offload"), nvme_ctrl_ops and blk_mq_ops. The nvme-tcp-offload driver shall register to nvme-tcp-offload ULP with the following ops:
- claim_dev() - in order to resolve the route to the target according to the paired net_dev.
- create_queue() - in order to create offloaded nvme-tcp queue.

The nvme-tcp-offload ULP module shall manage all the controller level functionalities, call claim_dev and based on the return values shall call the relevant module create_queue in order to create the admin queue and the IO queues.

IO-path Design:
===============
The nvme-tcp-offload shall work at the IO-level - the nvme-tcp-offload ULP module shall pass the request (the IO) to the nvme-tcp-offload driver and later, the nvme-tcp-offload driver returns the request completion (the IO completion).
No additional handling is needed in between; this design will reduce the CPU utilization as we will describe below.

The nvme-tcp-offload driver shall register to nvme-tcp-offload ULP with the following IO-path ops:
- send_req() - in order to pass the request to the handling of the offload driver that shall pass it to the offload device specific device.
- poll_queue()

Once the IO completes, the nvme-tcp-offload driver shall call command.done() that will invoke the nvme-tcp-offload ULP layer to complete the request.

Teardown and errors:
====================
In case of NVMeTCP queue error the nvme-tcp-offload driver shall call the nvme_tcp_ofld_report_queue_err.
The nvme-tcp-offload driver shall register to nvme-tcp-offload ULP with the following teardown ops:
- drain_queue()
- destroy_queue()

The Marvell qedn driver:
========================
The new driver will be added under "drivers/nvme/hw" and will be enabled by the Kconfig "Marvell NVM Express over Fabrics TCP offload". As part of the qedn init, the driver will register as a pci device driver and will work with the Marvell fastlinQ NIC.
As part of the probe, the driver will register to the nvme_tcp_offload (ULP) and to the qed module (qed_nvmetcp_ops) - similar to other "qed_*_ops" which are used by the qede, qedr, qedf and qedi device drivers.

Changes since RFC v1:
=====================
- nvme-tcp-offload: Fix nvme_tcp_ofld_ops return values.
- nvme-tcp-offload: Remove NVMF_TRTYPE_TCP_OFFLOAD.
- nvme-tcp-offload: Add nvme_tcp_ofld_poll() implementation.
- nvme-tcp-offload: Fix nvme_tcp_ofld_queue_rq() to check map_sg() and send_req() return values.

Changes since RFC v2:
=====================
- nvme-tcp-offload: Fixes in controller and queue level (patches 3-6).
- qedn: Add the Marvell's NVMeTCP HW offload device driver init and probe (patches 8-11).

Changes since RFC v3:
=====================
- nvme-tcp-offload: Add the full implementation of the nvme-tcp-offload layer including the new ops: setup_ctrl(), release_ctrl(), commit_rqs() and new flows (ASYNC and timeout).
- nvme-tcp-offload: Add device maximums: max_hw_sectors, max_segments.
- nvme-tcp-offload: layer design and optimization changes.
- qedn: Add full implementation for the conn level, IO path and error handling.

Changes since RFC v4:
=====================
(Many thanks to Hannes Reinecke for his feedback)
- nvme_tcp_offload: Add num_hw_vectors in order to limit the number of queues.
- nvme_tcp_offload: Add per device private_data.
- nvme_tcp_offload: Fix header digest, data digest and tos initialization.
- qedn: Remove the qedn_global list.
- qedn: Remove the workqueue flow from send_req.
- qedn: Add db_recovery support.

Changes since RFC v5:
=====================
(Many thanks to Sagi Grimberg for his feedback)
- nvme-fabrics: Expose nvmf_check_required_opts() globally (as a new patch).
- nvme_tcp_offload: Remove io-queues BLK_MQ_F_BLOCKING.
- nvme_tcp_offload: Fix the nvme_tcp_ofld_stop_queue (drain_queue) flow.
- nvme_tcp_offload: Fix the nvme_tcp_ofld_free_queue (destroy_queue) flow.
- nvme_tcp_offload: Change rwsem to mutex.
- nvme_tcp_offload: remove redundant fields.
- nvme_tcp_offload: Remove the "new" from setup_ctrl().
- nvme_tcp_offload: Remove the init_req() and commit_rqs() ops.
- nvme_tcp_offload: Minor fixes in nvme_tcp_ofld_create_ctrl() ansd nvme_tcp_ofld_free_queue().
- nvme_tcp_offload: Patch 8 (timeout and async) was squeashed into patch 7 (io level).
- qedn: Fix the free_queue flow and the destroy_queue flow.
- qedn: Remove version number.

Changes since RFC v6:
=====================
- No changes in nvme_tcp_offload
- qedn: Remove redundant logic in the io-queues core affinity initialization.
- qedn: Remove qedn_validate_cccid_in_range().

Changes since v1:
=====================
- nvme_tcp_offload: Add support for NVME_OPT_HOST_IFACE.
- nvme_tcp_offload: Kconfig fix (thanks to Petr Mladek).
- nvme_tcp_offload: return code fix (thanks to Dan Carpenter).

Changes since v2:
=====================
- nvme_tcp_offload: Fix overly long lines.
- nvme_tcp_offload: use correct terminology for vendor driver.
- qedn: Added qedn driver as part of series.

Changes since v3:
=====================
- nvme_tcp_offload: Rename nvme_tcp_ofld_map_data() to nvme_tcp_ofld_set_sg_host_data().

Arie Gershberg (2):
nvme-tcp-offload: Add controller level implementation
nvme-tcp-offload: Add controller level error recovery implementation

Dean Balandin (3):
nvme-tcp-offload: Add device scan implementation
nvme-tcp-offload: Add queue level implementation
nvme-tcp-offload: Add IO level implementation

Nikolay Assa (1):
qedn: Add qedn_claim_dev API support

Prabhakar Kushwaha (7):
nvme-fabrics: Move NVMF_ALLOWED_OPTS and NVMF_REQUIRED_OPTS
definitions
nvme-fabrics: Expose nvmf_check_required_opts() globally
qedn: Add connection-level slowpath functionality
qedn: Add support of configuring HW filter block
qedn: Add support of Task and SGL
qedn: Add support of NVME ICReq & ICResp
qedn: Add support of ASYNC

Shai Malin (7):
nvme-tcp-offload: Add nvme-tcp-offload - NVMeTCP HW offload ULP
qedn: Add qedn - Marvell's NVMeTCP HW offload device driver
qedn: Add qedn probe
qedn: Add IRQ and fast-path resources initializations
qedn: Add IO level qedn_send_req and fw_cq workqueue
qedn: Add IO level fastpath functionality
qedn: Add Connection and IO level recovery flows

 MAINTAINERS                      |   18 +
 drivers/nvme/Kconfig             |    1 +
 drivers/nvme/Makefile            |    1 +
 drivers/nvme/host/Kconfig        |   15 +
 drivers/nvme/host/Makefile       |    3 +
 drivers/nvme/host/fabrics.c      |   12 +-
 drivers/nvme/host/fabrics.h      |    9 +
 drivers/nvme/host/tcp-offload.c  | 1346 ++++++++++++++++++++++++++++++
 drivers/nvme/host/tcp-offload.h  |  207 +++++
 drivers/nvme/hw/Kconfig          |    9 +
 drivers/nvme/hw/Makefile         |    3 +
 drivers/nvme/hw/qedn/Makefile    |    4 +
 drivers/nvme/hw/qedn/qedn.h      |  402 +++++++++
 drivers/nvme/hw/qedn/qedn_conn.c | 1076 ++++++++++++++++++++++++
 drivers/nvme/hw/qedn/qedn_main.c | 1109 ++++++++++++++++++++++++
 drivers/nvme/hw/qedn/qedn_task.c |  873 +++++++++++++++++++
 16 files changed, 5079 insertions(+), 9 deletions(-)
 create mode 100644 drivers/nvme/host/tcp-offload.c
 create mode 100644 drivers/nvme/host/tcp-offload.h
 create mode 100644 drivers/nvme/hw/Kconfig
 create mode 100644 drivers/nvme/hw/Makefile
 create mode 100644 drivers/nvme/hw/qedn/Makefile
 create mode 100644 drivers/nvme/hw/qedn/qedn.h
 create mode 100644 drivers/nvme/hw/qedn/qedn_conn.c
 create mode 100644 drivers/nvme/hw/qedn/qedn_main.c
 create mode 100644 drivers/nvme/hw/qedn/qedn_task.c
Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#1 lore raw

PATCH v4 1/20 nvme-tcp-offload: Add nvme-tcp-offload - NVMeTCP HW offload ULP

From: Shai Malin <[email protected]>

This patch will present the structure for the NVMeTCP offload common layer driver. This module is added under "drivers/nvme/host/" and future offload drivers which will register to it will be placed under "drivers/nvme/hw".
This new driver will be enabled by the Kconfig "NVM Express over Fabrics TCP offload commmon layer".
In order to support the new transport type, for host mode, no change is needed.

Each new offload device specific driver will register to this ULP during its probe function, by filling out the nvme_tcp_ofld_dev->ops and nvme_tcp_ofld_dev->private_data and calling nvme_tcp_ofld_register_dev with the initialized struct.

The internal implementation:
- tcp-offload.h:
Includes all common structs and ops to be used and shared by offload drivers.

- tcp-offload.c:
Includes the init function which registers as a NVMf transport just like any other transport.

---
 MAINTAINERS                     |   8 ++
 drivers/nvme/host/Kconfig       |  15 +++
 drivers/nvme/host/Makefile      |   3 +
 drivers/nvme/host/tcp-offload.c | 124 ++++++++++++++++++++
 drivers/nvme/host/tcp-offload.h | 199 ++++++++++++++++++++++++++++++++
 5 files changed, 349 insertions(+)
 create mode 100644 drivers/nvme/host/tcp-offload.c
 create mode 100644 drivers/nvme/host/tcp-offload.h
MAINTAINERS +8 -0
index 81e1edeceae4..01fbebdc7722 100644
@@ -13093,6 +13093,14 @@ F:	drivers/nvme/host/
 F:	include/linux/nvme.h
 F:	include/uapi/linux/nvme_ioctl.h
 
+NVM EXPRESS TCP OFFLOAD TRANSPORT DRIVERS
+M:	Shai Malin <[email protected]>
+M:	Ariel Elior <[email protected]>
+L:	[email protected]
+S:	Supported
+F:	drivers/nvme/host/tcp-offload.c
+F:	drivers/nvme/host/tcp-offload.h
+
 NVM EXPRESS FC TRANSPORT DRIVERS
 M:	James Smart <[email protected]>
 L:	[email protected]
drivers/nvme/host/Kconfig +15 -0
index 102292289cdf..1993734d0104 100644
@@ -84,3 +84,18 @@ config NVME_TCP
 	  from https://github.com/linux-nvme/nvme-cli.
 
 	  If unsure, say N.
+
+config NVME_TCP_OFFLOAD
+	tristate "NVM Express over Fabrics TCP offload common layer"
+	depends on INET
+	depends on BLK_DEV_NVME
+	select NVME_FABRICS
+	help
+	  This provides support for the NVMe over Fabrics protocol using
+	  the TCP offload transport. This allows you to use remote block devices
+	  exported using the NVMe protocol set.
+
+	  To configure a NVMe over Fabrics controller use the nvme-cli tool
+	  from https://github.com/linux-nvme/nvme-cli.
+
+	  If unsure, say N.
drivers/nvme/host/Makefile +3 -0
index cbc509784b2e..3c3fdf83ce38 100644
@@ -8,6 +8,7 @@ obj-$(CONFIG_NVME_FABRICS)		+= nvme-fabrics.o
 obj-$(CONFIG_NVME_RDMA)			+= nvme-rdma.o
 obj-$(CONFIG_NVME_FC)			+= nvme-fc.o
 obj-$(CONFIG_NVME_TCP)			+= nvme-tcp.o
+obj-$(CONFIG_NVME_TCP_OFFLOAD)	+= nvme-tcp-offload.o
 
 nvme-core-y				:= core.o ioctl.o
 nvme-core-$(CONFIG_TRACING)		+= trace.o
@@ -26,3 +27,5 @@ nvme-rdma-y				+= rdma.o
 nvme-fc-y				+= fc.o
 
 nvme-tcp-y				+= tcp.o
+
+nvme-tcp-offload-y		+= tcp-offload.o
drivers/nvme/host/tcp-offload.c new file +124 -0

124 lines not on this page · raw message

drivers/nvme/host/tcp-offload.h new file +199 -0

199 lines not on this page · raw message

Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#2 lore raw

Re: [PATCH v4 01/20] nvme-tcp-offload: Add nvme-tcp-offload - NVMeTCP HW offload ULP

drivers/nvme/host/tcp-offload.c line 10

Quoted 4 lines
drivers/nvme/host/tcp-offload.c new file +2 -0
+/* Kernel includes */
 
 
+/* Driver includes */

I think these comments are a little pointless.

Quoted 1 line

+int nvme_tcp_ofld_register_dev(struct nvme_tcp_ofld_dev *dev)

Can you spell out offload everywhere?

drivers/nvme/host/tcp-offload.c line 79

Quoted 6 lines
drivers/nvme/host/tcp-offload.c new file +6 -0
+int nvme_tcp_ofld_report_queue_err(struct nvme_tcp_ofld_queue *queue)
+{
+	/* Placeholder - invoke error recovery flow */
+
+	return 0;
+}

Please don't add any placeholders like this. The whole file is still pretty small with all the patches applied, so no need to split it.

Quoted 12 lines
drivers/nvme/host/tcp-offload.c new file +12 -0
+static struct nvmf_transport_ops nvme_tcp_ofld_transport = {
+	.name		= "tcp_offload",
+	.module		= THIS_MODULE,
+	.required_opts	= NVMF_OPT_TRADDR,
+	.allowed_opts	= NVMF_OPT_TRSVCID | NVMF_OPT_NR_WRITE_QUEUES  |
+			  NVMF_OPT_HOST_TRADDR | NVMF_OPT_CTRL_LOSS_TMO |
+			  NVMF_OPT_RECONNECT_DELAY | NVMF_OPT_HDR_DIGEST |
+			  NVMF_OPT_DATA_DIGEST | NVMF_OPT_NR_POLL_QUEUES |
+			  NVMF_OPT_TOS | NVMF_OPT_HOST_IFACE,
+};
+
+static int __init nvme_tcp_ofld_init_module(void)

drivers/nvme/host/tcp-offload.c line 120

Quoted 10 lines
drivers/nvme/host/tcp-offload.c new file +10 -0
+{
+	nvmf_register_transport(&nvme_tcp_ofld_transport);
+
+	return 0;
+}
+
+static void __exit nvme_tcp_ofld_cleanup_module(void)
+{
+	nvmf_unregister_transport(&nvme_tcp_ofld_transport);
+}

Looking at the final result this doesn't do much. Assuming we want to support these kinds of whacky offloads (which I'd rather not do), the proper way would be to allow registering multiple transport_ops structures for a given name rather adding an indirection that duplicates a whole lot of code.

drivers/nvme/host/tcp-offload.h line 30

Quoted 3 lines
drivers/nvme/host/tcp-offload.h new file +3 -0
+	/* Offload device specific driver context */
+	int num_hw_vectors;
+};

As far as I can tell this is about queues, not "vectors" of some kind.

Quoted 7 lines
drivers/nvme/host/tcp-offload.h new file +7 -0
+struct nvme_tcp_ofld_req {
+	struct nvme_request req;
+	struct nvme_command nvme_cmd;
+	struct list_head queue_entry;
+	struct nvme_tcp_ofld_queue *queue;
+
+	/* Offload device specific driver context */

drivers/nvme/host/tcp-offload.h line 51

Quoted 10 lines
drivers/nvme/host/tcp-offload.h new file +10 -0
+	void *private_data;
+
+	/* async flag is used to distinguish between async and IO flow
+	 * in common send_req() of nvme_tcp_ofld_ops.
+	 */
+	bool async;
+
+	void (*done)(struct nvme_tcp_ofld_req *req,
+		     union nvme_result *result,
+		     __le16 status);

This always points to nvme_tcp_ofld_req_done, why the costly indirection?

drivers/nvme/host/tcp-offload.h line 76

Quoted 3 lines
drivers/nvme/host/tcp-offload.h new file +3 -0
+	/* Error callback function */
+	int (*report_err)(struct nvme_tcp_ofld_queue *queue);
+};

This seems to always point to nvme_tcp_ofld_report_queue_err, why the indirection?

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#3 lore raw

Re: [PATCH v4 01/20] nvme-tcp-offload: Add nvme-tcp-offload - NVMeTCP HW offload ULP

Quoted 7 lines · Christoph Hellwig

drivers/nvme/host/tcp-offload.c line 10

Quoted 5 lines
drivers/nvme/host/tcp-offload.c new file +2 -0
+/* Kernel includes */
 
 
+/* Driver includes */

I think these comments are a little pointless.

Will be removed.

Quoted 4 lines
Quoted 2 lines

+int nvme_tcp_ofld_register_dev(struct nvme_tcp_ofld_dev *dev)

Can you spell out offload everywhere?

Sure.

Quoted 10 lines

drivers/nvme/host/tcp-offload.c line 79

Quoted 7 lines
drivers/nvme/host/tcp-offload.c new file +6 -0
+int nvme_tcp_ofld_report_queue_err(struct nvme_tcp_ofld_queue *queue)
+{
+     /* Placeholder - invoke error recovery flow */
+
+     return 0;
+}

Please don't add any placeholders like this. The whole file is still pretty small with all the patches applied, so no need to split it.

Sure.

Quoted 21 lines
Quoted 21 lines
drivers/nvme/host/tcp-offload.c new file +20 -0
+static struct nvmf_transport_ops nvme_tcp_ofld_transport = {
+     .name           = "tcp_offload",
+     .module         = THIS_MODULE,
+     .required_opts  = NVMF_OPT_TRADDR,
+     .allowed_opts   = NVMF_OPT_TRSVCID | NVMF_OPT_NR_WRITE_QUEUES  |
+                       NVMF_OPT_HOST_TRADDR | NVMF_OPT_CTRL_LOSS_TMO |
+                       NVMF_OPT_RECONNECT_DELAY | NVMF_OPT_HDR_DIGEST |
+                       NVMF_OPT_DATA_DIGEST | NVMF_OPT_NR_POLL_QUEUES |
+                       NVMF_OPT_TOS | NVMF_OPT_HOST_IFACE,
+};
+
+static int __init nvme_tcp_ofld_init_module(void)
+{
+     nvmf_register_transport(&nvme_tcp_ofld_transport);
+
+     return 0;
+}
+
+static void __exit nvme_tcp_ofld_cleanup_module(void)
+{
Quoted 8 lines

drivers/nvme/host/tcp-offload.c line 120

Quoted 2 lines
drivers/nvme/host/tcp-offload.c new file +2 -0
+     nvmf_unregister_transport(&nvme_tcp_ofld_transport);
+}

Looking at the final result this doesn't do much. Assuming we want to support these kinds of whacky offloads (which I'd rather not do), the proper way would be to allow registering multiple transport_ops structures for a given name rather adding an indirection that duplicates a whole lot of code.

In that case, would you prefer that we invoke the tcp-offload from within the tcp flow?
Should it be with the same transport name (“tcp”) or with a different transport name (“tcp_offload”)?

Also, would you prefer that we register the offload device driver directly to blk_mq layer or through the tcp-offload layer?

Quoted 6 lines

drivers/nvme/host/tcp-offload.h line 30

Quoted 4 lines
drivers/nvme/host/tcp-offload.h new file +3 -0
+     /* Offload device specific driver context */
+     int num_hw_vectors;
+};

As far as I can tell this is about queues, not "vectors" of some kind.

It's the num irqs which the device support. We can rename it "num_hw_queues".

Quoted 12 lines
Quoted 12 lines
drivers/nvme/host/tcp-offload.h new file +11 -0
+struct nvme_tcp_ofld_req {
+     struct nvme_request req;
+     struct nvme_command nvme_cmd;
+     struct list_head queue_entry;
+     struct nvme_tcp_ofld_queue *queue;
+
+     /* Offload device specific driver context */
+     void *private_data;
+
+     /* async flag is used to distinguish between async and IO flow
+      * in common send_req() of nvme_tcp_ofld_ops.
Quoted 8 lines

drivers/nvme/host/tcp-offload.h line 51

Quoted 6 lines
drivers/nvme/host/tcp-offload.h new file +6 -0
+      */
+     bool async;
+
+     void (*done)(struct nvme_tcp_ofld_req *req,
+                  union nvme_result *result,
+                  __le16 status);

This always points to nvme_tcp_ofld_req_done, why the costly indirection?

Will be fixed.

Quoted 7 lines

drivers/nvme/host/tcp-offload.h line 76

Quoted 4 lines
drivers/nvme/host/tcp-offload.h new file +3 -0
+     /* Error callback function */
+     int (*report_err)(struct nvme_tcp_ofld_queue *queue);
+};

This seems to always point to nvme_tcp_ofld_report_queue_err, why the indirection?

Will be fixed.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#4 lore raw

Re: [PATCH v4 01/20] nvme-tcp-offload: Add nvme-tcp-offload - NVMeTCP HW offload ULP

Hi Christoph,

Quoted 3 lines · Shai Malin
Quoted 1 line · Christoph Hellwig

[snip]

Quoted 31 lines
Quoted 22 lines
Quoted 22 lines
drivers/nvme/host/tcp-offload.c new file +20 -0
+static struct nvmf_transport_ops nvme_tcp_ofld_transport = {
+     .name           = "tcp_offload",
+     .module         = THIS_MODULE,
+     .required_opts  = NVMF_OPT_TRADDR,
+     .allowed_opts   = NVMF_OPT_TRSVCID | NVMF_OPT_NR_WRITE_QUEUES  |
+                       NVMF_OPT_HOST_TRADDR | NVMF_OPT_CTRL_LOSS_TMO |
+                       NVMF_OPT_RECONNECT_DELAY | NVMF_OPT_HDR_DIGEST |
+                       NVMF_OPT_DATA_DIGEST | NVMF_OPT_NR_POLL_QUEUES |
+                       NVMF_OPT_TOS | NVMF_OPT_HOST_IFACE,
+};
+
+static int __init nvme_tcp_ofld_init_module(void)
+{
+     nvmf_register_transport(&nvme_tcp_ofld_transport);
+
+     return 0;
+}
+
+static void __exit nvme_tcp_ofld_cleanup_module(void)
+{
Quoted 8 lines

drivers/nvme/host/tcp-offload.c line 120

Quoted 2 lines
drivers/nvme/host/tcp-offload.c new file +2 -0
+     nvmf_unregister_transport(&nvme_tcp_ofld_transport);
+}

Looking at the final result this doesn't do much. Assuming we want to support these kinds of whacky offloads (which I'd rather not do), the proper way would be to allow registering multiple transport_ops structures for a given name rather adding an indirection that duplicates a whole lot of code.

Quoted 8 lines

In that case, would you prefer that we invoke the tcp-offload from within the tcp flow?
Should it be with the same transport name (“tcp”) or with a different transport name (“tcp_offload”)?

Also, would you prefer that we register the offload device driver directly to blk_mq layer or through the tcp-offload layer?

I hope you got some time to look into the above queries. We would appreciate your feedback in order to finalize the design.

--pk

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#5 lore raw

Re: [PATCH v4 01/20] nvme-tcp-offload: Add nvme-tcp-offload - NVMeTCP HW offload ULP

In reply to Shai Malin
Quoted 11 lines · Shai Malin
Quoted 10 lines

drivers/nvme/host/tcp-offload.c line 120

Quoted 4 lines
drivers/nvme/host/tcp-offload.c new file +4 -0
+static void __exit nvme_tcp_ofld_cleanup_module(void)
+{
+     nvmf_unregister_transport(&nvme_tcp_ofld_transport);
+}

Looking at the final result this doesn't do much. Assuming we want to support these kinds of whacky offloads (which I'd rather not do), the proper way would be to allow registering multiple transport_ops structures for a given name rather adding an indirection that duplicates a whole lot of code.

Quoted 7 lines

In that case, would you prefer that we invoke the tcp-offload from within the tcp flow?
Should it be with the same transport name (“tcp”) or with a different transport name (“tcp_offload”)?

Also, would you prefer that we register the offload device driver directly to blk_mq layer or through the tcp-offload layer?

As said, we should allow the different offload drivers register as te offload transport and just iterate through the multiple instances of the transport until we find a match instead of duplicating the registration infrastructure.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#6 lore raw

PATCH v4 2/20 nvme-fabrics: Move NVMF_ALLOWED_OPTS and NVMF_REQUIRED_OPTS definitions

In reply to Prabhakar Kushwaha

Move NVMF_ALLOWED_OPTS and NVMF_REQUIRED_OPTS definitions to header file, so it can be used by the different HW devices.

NVMeTCP offload devices might have different limitations of the allowed options, for example, a device that does not support all the queue types. With tcp and rdma, only the nvme-tcp and nvme-rdma layers handle those attributes and the HW devices do not create any limitations for the allowed options.

An alternative design could be to add separate fields in nvme_tcp_ofld_ops such as max_hw_sectors and max_segments that we already have in this series.

---
 drivers/nvme/host/fabrics.c | 7 -------
 drivers/nvme/host/fabrics.h | 7 +++++++
 2 files changed, 7 insertions(+), 7 deletions(-)
drivers/nvme/host/fabrics.c +0 -7
index 1e6a7cc056ca..21d67775d2af 100644
@@ -947,13 +947,6 @@ void nvmf_free_options(struct nvmf_ctrl_options *opts)
 }
 EXPORT_SYMBOL_GPL(nvmf_free_options);
 
-#define NVMF_REQUIRED_OPTS	(NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
-#define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
-				 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
-				 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
-				 NVMF_OPT_DISABLE_SQFLOW |\
-				 NVMF_OPT_FAIL_FAST_TMO)
-
 static struct nvme_ctrl *
 nvmf_create_ctrl(struct device *dev, const char *buf)
 {
drivers/nvme/host/fabrics.h +7 -0
index c31dad69a773..38ac7b757d78 100644
@@ -69,6 +69,13 @@ enum {
 	NVMF_OPT_HOST_IFACE	= 1 << 21,
 };
 
+#define NVMF_REQUIRED_OPTS	(NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
+#define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
+				 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
+				 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
+				 NVMF_OPT_DISABLE_SQFLOW |\
+				 NVMF_OPT_FAIL_FAST_TMO)
+
 /**
  * struct nvmf_ctrl_options - Used to hold the options specified
  *			      with the parsing opts enum.
Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#7 lore raw

PATCH v4 3/20 nvme-fabrics: Expose nvmf_check_required_opts() globally

In reply to Prabhakar Kushwaha

nvmf_check_required_opts() is used to check if user provided opts has the required_opts or not. if not, it will log which options are not provided.

It can be leveraged by nvme-tcp-offload to check if provided opts are supported by this specific offload driver or not.

So expose nvmf_check_required_opts() globally.

---
 drivers/nvme/host/fabrics.c | 5 +++--
 drivers/nvme/host/fabrics.h | 2 ++
 2 files changed, 5 insertions(+), 2 deletions(-)
drivers/nvme/host/fabrics.c +3 -2
index 21d67775d2af..7830d2540492 100644
@@ -864,8 +864,8 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 	return ret;
 }
 
-static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
-		unsigned int required_opts)
+int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
+			     unsigned int required_opts)
 {
 	if ((opts->mask & required_opts) != required_opts) {
 		int i;
@@ -883,6 +883,7 @@ static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
 
 	return 0;
 }
+EXPORT_SYMBOL_GPL(nvmf_check_required_opts);
 
 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
 		struct nvmf_ctrl_options *opts)
drivers/nvme/host/fabrics.h +2 -0
index 38ac7b757d78..15d9c15ef8a6 100644
@@ -197,5 +197,7 @@ int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size);
 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl);
 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
 		struct nvmf_ctrl_options *opts);
+int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
+			     unsigned int required_opts);
 
 #endif /* _NVME_FABRICS_H */
Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#9 lore raw

Re: [PATCH v4 03/20] nvme-fabrics: Expose nvmf_check_required_opts() globally

Quoted 3 lines · Christoph Hellwig

This and the previous patch are pretty much two sides of the same coin and belong together. But without the registration indirection they wouldn't even be needed.

Sure.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#10 lore raw

PATCH v4 4/20 nvme-tcp-offload: Add device scan implementation

In reply to Prabhakar Kushwaha

From: Dean Balandin <[email protected]>

As part of create_ctrl(), it scans the registered devices and calls the claim_dev op on each of them, to find the first devices that matches the connection params. Once the correct devices is found (claim_dev returns true), we raise the refcnt of that device and return that device as the device to be used for ctrl currently being created.

---
 drivers/nvme/host/tcp-offload.c | 91 +++++++++++++++++++++++++++++++++
 drivers/nvme/host/tcp-offload.h |  1 +
 2 files changed, 92 insertions(+)
drivers/nvme/host/tcp-offload.c +91 -0
index 10b87f5b875b..d0f4b83549b9 100644
@@ -13,6 +13,11 @@
 static LIST_HEAD(nvme_tcp_ofld_devices);
 static DEFINE_MUTEX(nvme_tcp_ofld_devices_mutex);
 
+static inline struct nvme_tcp_ofld_ctrl *to_tcp_ofld_ctrl(struct nvme_ctrl *nc)
+{
+	return container_of(nc, struct nvme_tcp_ofld_ctrl, nctrl);
+}
+
 /**
  * nvme_tcp_ofld_register_dev() - NVMeTCP Offload Library registration
  * function.
@@ -96,6 +101,91 @@ void nvme_tcp_ofld_req_done(struct nvme_tcp_ofld_req *req,
 	/* Placeholder - complete request with/without error */
 }
 
+static struct nvme_tcp_ofld_dev *
+nvme_tcp_ofld_lookup_dev(struct nvme_tcp_ofld_ctrl *ctrl)
+{
+	struct nvme_ctrl *nctrl = &ctrl->nctrl;
+	struct nvme_tcp_ofld_dev *dev;
+	char *iface;
+
+	if (nctrl->opts->mask & NVMF_OPT_HOST_IFACE) {
+		iface = nctrl->opts->host_iface;
+		ctrl->ndev = __dev_get_by_name(&init_net, iface);
+		if (!ctrl->ndev) {
+			pr_err("invalid interface passed: %s\n", iface);
+			return NULL;
+		}
+	}
+
+	mutex_lock(&nvme_tcp_ofld_devices_mutex);
+	list_for_each_entry(dev, &nvme_tcp_ofld_devices, entry) {
+	/* ctrl includes the destination ip, source ip (if provided) and
+	 * network interface (if provided).
+	 */
+		if (dev->ops->claim_dev(dev, ctrl))
+			goto out;
+	}
+
+	dev = NULL;
+out:
+	mutex_unlock(&nvme_tcp_ofld_devices_mutex);
+
+	return dev;
+}
+
+static struct nvme_ctrl *
+nvme_tcp_ofld_create_ctrl(struct device *ndev, struct nvmf_ctrl_options *opts)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl;
+	struct nvme_tcp_ofld_dev *dev;
+	struct nvme_ctrl *nctrl;
+	int rc = 0;
+
+	ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
+	if (!ctrl)
+		return ERR_PTR(-ENOMEM);
+
+	nctrl = &ctrl->nctrl;
+
+	/* Init nvme_tcp_ofld_ctrl & nvme_ctrl params based on received opts */
+
+	/* Find device that can reach the dest addr */
+	dev = nvme_tcp_ofld_lookup_dev(ctrl);
+	if (!dev) {
+		pr_info("no device found for addr %s:%s.\n",
+			opts->traddr, opts->trsvcid);
+		rc = -EINVAL;
+		goto out_free_ctrl;
+	}
+
+	/* Increase driver refcnt */
+	if (!try_module_get(dev->ops->module)) {
+		pr_err("try_module_get failed\n");
+		rc = -ENODEV;
+		goto out_free_ctrl;
+	}
+
+	ctrl->dev = dev;
+
+	if (ctrl->dev->ops->max_hw_sectors)
+		nctrl->max_hw_sectors = ctrl->dev->ops->max_hw_sectors;
+	if (ctrl->dev->ops->max_segments)
+		nctrl->max_segments = ctrl->dev->ops->max_segments;
+
+	/* Init queues */
+
+	/* Call nvme_init_ctrl */
+
+	/* Setup ctrl */
+
+	return nctrl;
+
+out_free_ctrl:
+	kfree(ctrl);
+
+	return ERR_PTR(rc);
+}
+
 static struct nvmf_transport_ops nvme_tcp_ofld_transport = {
 	.name		= "tcp_offload",
 	.module		= THIS_MODULE,
@@ -105,6 +195,7 @@ static struct nvmf_transport_ops nvme_tcp_ofld_transport = {
 			  NVMF_OPT_RECONNECT_DELAY | NVMF_OPT_HDR_DIGEST |
 			  NVMF_OPT_DATA_DIGEST | NVMF_OPT_NR_POLL_QUEUES |
 			  NVMF_OPT_TOS | NVMF_OPT_HOST_IFACE,
+	.create_ctrl	= nvme_tcp_ofld_create_ctrl,
 };
 
 static int __init nvme_tcp_ofld_init_module(void)
drivers/nvme/host/tcp-offload.h +1 -0
index bfa759177a07..d1c2c6171897 100644
@@ -8,6 +8,7 @@
 #include <linux/scatterlist.h>
 #include <linux/types.h>
 #include <linux/nvme-tcp.h>
+#include <linux/netdevice.h>
 
 /* Driver includes */
 #include "nvme.h"
Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#11 lore raw

Re: [PATCH v4 04/20] nvme-tcp-offload: Add device scan implementation

drivers/nvme/host/tcp-offload.c void nvme_tcp_ofld_req_done(struct nvme_tcp_ofld_req *req, · line 124

Quoted 5 lines
drivers/nvme/host/tcp-offload.c +5 -0
+	mutex_lock(&nvme_tcp_ofld_devices_mutex);
+	list_for_each_entry(dev, &nvme_tcp_ofld_devices, entry) {
+	/* ctrl includes the destination ip, source ip (if provided) and
+	 * network interface (if provided).
+	 */

This is not the normal kernel comment style, and also incorrectly indented.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#12 lore raw

Re: [PATCH v4 04/20] nvme-tcp-offload: Add device scan implementation

Quoted 8 lines · Christoph Hellwig

drivers/nvme/host/tcp-offload.c void nvme_tcp_ofld_req_done(struct nvme_tcp_ofld_req *req, · line 124

Quoted 5 lines
drivers/nvme/host/tcp-offload.c +5 -0
+     mutex_lock(&nvme_tcp_ofld_devices_mutex);
+     list_for_each_entry(dev, &nvme_tcp_ofld_devices, entry) {
+     /* ctrl includes the destination ip, source ip (if provided) and
+      * network interface (if provided).
+      */

This is not the normal kernel comment style, and also incorrectly indented.

Will be fixed.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#13 lore raw

PATCH v4 5/20 nvme-tcp-offload: Add controller level implementation

In reply to Prabhakar Kushwaha

From: Arie Gershberg <[email protected]>

In this patch we implement controller level functionality including:
- create_ctrl.
- delete_ctrl.
- free_ctrl.

The implementation is similar to other nvme fabrics modules, the main difference being that the nvme-tcp-offload ULP calls the offload specific claim_dev() op with the given TCP/IP parameters to determine which device will be used for this controller.
Once found, the offload specific device and controller will be paired and kept in a controller list managed by the ULP.

---
 drivers/nvme/host/tcp-offload.c | 487 +++++++++++++++++++++++++++++++-
 1 file changed, 482 insertions(+), 5 deletions(-)
drivers/nvme/host/tcp-offload.c +482 -5
index d0f4b83549b9..806247722a35 100644
@@ -12,6 +12,10 @@
 
 static LIST_HEAD(nvme_tcp_ofld_devices);
 static DEFINE_MUTEX(nvme_tcp_ofld_devices_mutex);
+static LIST_HEAD(nvme_tcp_ofld_ctrl_list);
+static DEFINE_MUTEX(nvme_tcp_ofld_ctrl_mutex);
+static struct blk_mq_ops nvme_tcp_ofld_admin_mq_ops;
+static struct blk_mq_ops nvme_tcp_ofld_mq_ops;
 
 static inline struct nvme_tcp_ofld_ctrl *to_tcp_ofld_ctrl(struct nvme_ctrl *nc)
 {
@@ -133,21 +137,441 @@ nvme_tcp_ofld_lookup_dev(struct nvme_tcp_ofld_ctrl *ctrl)
 	return dev;
 }
 
+static struct blk_mq_tag_set *
+nvme_tcp_ofld_alloc_tagset(struct nvme_ctrl *nctrl, bool admin)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(nctrl);
+	struct blk_mq_tag_set *set;
+	int rc;
+
+	if (admin) {
+		set = &ctrl->admin_tag_set;
+		memset(set, 0, sizeof(*set));
+		set->ops = &nvme_tcp_ofld_admin_mq_ops;
+		set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
+		set->reserved_tags = NVMF_RESERVED_TAGS;
+		set->numa_node = nctrl->numa_node;
+		set->flags = BLK_MQ_F_BLOCKING;
+		set->cmd_size = sizeof(struct nvme_tcp_ofld_req);
+		set->driver_data = ctrl;
+		set->nr_hw_queues = 1;
+		set->timeout = NVME_ADMIN_TIMEOUT;
+	} else {
+		set = &ctrl->tag_set;
+		memset(set, 0, sizeof(*set));
+		set->ops = &nvme_tcp_ofld_mq_ops;
+		set->queue_depth = nctrl->sqsize + 1;
+		set->reserved_tags = NVMF_RESERVED_TAGS;
+		set->numa_node = nctrl->numa_node;
+		set->flags = BLK_MQ_F_SHOULD_MERGE;
+		set->cmd_size = sizeof(struct nvme_tcp_ofld_req);
+		set->driver_data = ctrl;
+		set->nr_hw_queues = nctrl->queue_count - 1;
+		set->timeout = NVME_IO_TIMEOUT;
+		set->nr_maps = nctrl->opts->nr_poll_queues ? HCTX_MAX_TYPES : 2;
+	}
+
+	rc = blk_mq_alloc_tag_set(set);
+	if (rc)
+		return ERR_PTR(rc);
+
+	return set;
+}
+
+static int nvme_tcp_ofld_configure_admin_queue(struct nvme_ctrl *nctrl,
+					       bool new)
+{
+	int rc;
+
+	/* Placeholder - alloc_admin_queue */
+	if (new) {
+		nctrl->admin_tagset =
+				nvme_tcp_ofld_alloc_tagset(nctrl, true);
+		if (IS_ERR(nctrl->admin_tagset)) {
+			rc = PTR_ERR(nctrl->admin_tagset);
+			nctrl->admin_tagset = NULL;
+			goto out_destroy_queue;
+		}
+
+		nctrl->fabrics_q = blk_mq_init_queue(nctrl->admin_tagset);
+		if (IS_ERR(nctrl->fabrics_q)) {
+			rc = PTR_ERR(nctrl->fabrics_q);
+			nctrl->fabrics_q = NULL;
+			goto out_free_tagset;
+		}
+
+		nctrl->admin_q = blk_mq_init_queue(nctrl->admin_tagset);
+		if (IS_ERR(nctrl->admin_q)) {
+			rc = PTR_ERR(nctrl->admin_q);
+			nctrl->admin_q = NULL;
+			goto out_cleanup_fabrics_q;
+		}
+	}
+
+	/* Placeholder - nvme_tcp_ofld_start_queue */
+
+	rc = nvme_enable_ctrl(nctrl);
+	if (rc)
+		goto out_stop_queue;
+
+	blk_mq_unquiesce_queue(nctrl->admin_q);
+
+	rc = nvme_init_ctrl_finish(nctrl);
+	if (rc)
+		goto out_quiesce_queue;
+
+	return 0;
+
+out_quiesce_queue:
+	blk_mq_quiesce_queue(nctrl->admin_q);
+	blk_sync_queue(nctrl->admin_q);
+
+out_stop_queue:
+	/* Placeholder - stop offload queue */
+	nvme_cancel_admin_tagset(nctrl);
+
+out_cleanup_fabrics_q:
+	if (new)
+		blk_cleanup_queue(nctrl->fabrics_q);
+out_free_tagset:
+	if (new)
+		blk_mq_free_tag_set(nctrl->admin_tagset);
+out_destroy_queue:
+	/* Placeholder - free admin queue */
+
+	return rc;
+}
+
+static int
+nvme_tcp_ofld_configure_io_queues(struct nvme_ctrl *nctrl, bool new)
+{
+	int rc;
+
+	/* Placeholder - alloc_io_queues */
+
+	if (new) {
+		nctrl->tagset = nvme_tcp_ofld_alloc_tagset(nctrl, false);
+		if (IS_ERR(nctrl->tagset)) {
+			rc = PTR_ERR(nctrl->tagset);
+			nctrl->tagset = NULL;
+			goto out_free_io_queues;
+		}
+
+		nctrl->connect_q = blk_mq_init_queue(nctrl->tagset);
+		if (IS_ERR(nctrl->connect_q)) {
+			rc = PTR_ERR(nctrl->connect_q);
+			nctrl->connect_q = NULL;
+			goto out_free_tag_set;
+		}
+	}
+
+	/* Placeholder - start_io_queues */
+
+	if (!new) {
+		nvme_start_queues(nctrl);
+		if (!nvme_wait_freeze_timeout(nctrl, NVME_IO_TIMEOUT)) {
+			/*
+			 * If we timed out waiting for freeze we are likely to
+			 * be stuck.  Fail the controller initialization just
+			 * to be safe.
+			 */
+			rc = -ENODEV;
+			goto out_wait_freeze_timed_out;
+		}
+		blk_mq_update_nr_hw_queues(nctrl->tagset,
+					   nctrl->queue_count - 1);
+		nvme_unfreeze(nctrl);
+	}
+
+	return 0;
+
+out_wait_freeze_timed_out:
+	nvme_stop_queues(nctrl);
+	nvme_sync_io_queues(nctrl);
+
+	/* Placeholder - Stop IO queues */
+
+	if (new)
+		blk_cleanup_queue(nctrl->connect_q);
+out_free_tag_set:
+	if (new)
+		blk_mq_free_tag_set(nctrl->tagset);
+out_free_io_queues:
+	/* Placeholder - free_io_queues */
+
+	return rc;
+}
+
+static int nvme_tcp_ofld_setup_ctrl(struct nvme_ctrl *nctrl, bool new)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(nctrl);
+	struct nvmf_ctrl_options *opts = nctrl->opts;
+	int rc = 0;
+
+	rc = ctrl->dev->ops->setup_ctrl(ctrl);
+	if (rc)
+		return rc;
+
+	rc = nvme_tcp_ofld_configure_admin_queue(nctrl, new);
+	if (rc)
+		goto out_release_ctrl;
+
+	if (nctrl->icdoff) {
+		dev_err(nctrl->device, "icdoff is not supported!\n");
+		rc = -EOPNOTSUPP;
+		goto destroy_admin;
+	}
+
+	if (!nvme_ctrl_sgl_supported(nctrl)) {
+		dev_err(nctrl->device, "Mandatory sgls are not supported!\n");
+		rc = -EOPNOTSUPP;
+		goto destroy_admin;
+	}
+
+	if (opts->queue_size > nctrl->sqsize + 1)
+		dev_warn(nctrl->device,
+			 "queue_size %zu > ctrl sqsize %u, clamping down\n",
+			 opts->queue_size, nctrl->sqsize + 1);
+
+	if (nctrl->sqsize + 1 > nctrl->maxcmd) {
+		dev_warn(nctrl->device,
+			 "sqsize %u > ctrl maxcmd %u, clamping down\n",
+			 nctrl->sqsize + 1, nctrl->maxcmd);
+		nctrl->sqsize = nctrl->maxcmd - 1;
+	}
+
+	if (nctrl->queue_count > 1) {
+		rc = nvme_tcp_ofld_configure_io_queues(nctrl, new);
+		if (rc)
+			goto destroy_admin;
+	}
+
+	if (!nvme_change_ctrl_state(nctrl, NVME_CTRL_LIVE)) {
+		/*
+		 * state change failure is ok if we started ctrl delete,
+		 * unless we're during creation of a new controller to
+		 * avoid races with teardown flow.
+		 */
+		WARN_ON_ONCE(nctrl->state != NVME_CTRL_DELETING &&
+			     nctrl->state != NVME_CTRL_DELETING_NOIO);
+		WARN_ON_ONCE(new);
+		rc = -EINVAL;
+		goto destroy_io;
+	}
+
+	nvme_start_ctrl(nctrl);
+
+	return 0;
+
+destroy_io:
+	/* Placeholder - stop and destroy io queues*/
+destroy_admin:
+	/* Placeholder - stop and destroy admin queue*/
+out_release_ctrl:
+	ctrl->dev->ops->release_ctrl(ctrl);
+
+	return rc;
+}
+
+static int
+nvme_tcp_ofld_check_dev_opts(struct nvmf_ctrl_options *opts,
+			     struct nvme_tcp_ofld_ops *ofld_ops)
+{
+	unsigned int nvme_tcp_ofld_opt_mask = NVMF_ALLOWED_OPTS |
+			ofld_ops->allowed_opts | ofld_ops->required_opts;
+	struct nvmf_ctrl_options dev_opts_mask;
+
+	if (opts->mask & ~nvme_tcp_ofld_opt_mask) {
+		pr_warn("One or more nvmf options missing from ofld drvr %s.\n",
+			ofld_ops->name);
+
+		dev_opts_mask.mask = nvme_tcp_ofld_opt_mask;
+
+		return nvmf_check_required_opts(&dev_opts_mask, opts->mask);
+	}
+
+	return 0;
+}
+
+static void nvme_tcp_ofld_free_ctrl(struct nvme_ctrl *nctrl)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(nctrl);
+	struct nvme_tcp_ofld_dev *dev = ctrl->dev;
+
+	if (list_empty(&ctrl->list))
+		goto free_ctrl;
+
+	ctrl->dev->ops->release_ctrl(ctrl);
+
+	mutex_lock(&nvme_tcp_ofld_ctrl_mutex);
+	list_del(&ctrl->list);
+	mutex_unlock(&nvme_tcp_ofld_ctrl_mutex);
+
+	nvmf_free_options(nctrl->opts);
+free_ctrl:
+	module_put(dev->ops->module);
+	kfree(ctrl->queues);
+	kfree(ctrl);
+}
+
+static void
+nvme_tcp_ofld_teardown_admin_queue(struct nvme_ctrl *ctrl, bool remove)
+{
+	/* Placeholder - teardown_admin_queue */
+}
+
+static void
+nvme_tcp_ofld_teardown_io_queues(struct nvme_ctrl *nctrl, bool remove)
+{
+	/* Placeholder - teardown_io_queues */
+}
+
+static void
+nvme_tcp_ofld_teardown_ctrl(struct nvme_ctrl *nctrl, bool shutdown)
+{
+	/* Placeholder - err_work and connect_work */
+	nvme_tcp_ofld_teardown_io_queues(nctrl, shutdown);
+	blk_mq_quiesce_queue(nctrl->admin_q);
+	if (shutdown)
+		nvme_shutdown_ctrl(nctrl);
+	else
+		nvme_disable_ctrl(nctrl);
+	nvme_tcp_ofld_teardown_admin_queue(nctrl, shutdown);
+}
+
+static void nvme_tcp_ofld_delete_ctrl(struct nvme_ctrl *nctrl)
+{
+	nvme_tcp_ofld_teardown_ctrl(nctrl, true);
+}
+
+static int
+nvme_tcp_ofld_init_request(struct blk_mq_tag_set *set,
+			   struct request *rq,
+			   unsigned int hctx_idx,
+			   unsigned int numa_node)
+{
+	struct nvme_tcp_ofld_req *req = blk_mq_rq_to_pdu(rq);
+
+	/* Placeholder - init request */
+
+	req->done = nvme_tcp_ofld_req_done;
+
+	return 0;
+}
+
+static blk_status_t
+nvme_tcp_ofld_queue_rq(struct blk_mq_hw_ctx *hctx,
+		       const struct blk_mq_queue_data *bd)
+{
+	/* Call nvme_setup_cmd(...) */
+
+	/* Call ops->send_req(...) */
+
+	return BLK_STS_OK;
+}
+
+static struct blk_mq_ops nvme_tcp_ofld_mq_ops = {
+	.queue_rq	= nvme_tcp_ofld_queue_rq,
+	.init_request	= nvme_tcp_ofld_init_request,
+	/*
+	 * All additional ops will be also implemented and registered similar to
+	 * tcp.c
+	 */
+};
+
+static struct blk_mq_ops nvme_tcp_ofld_admin_mq_ops = {
+	.queue_rq	= nvme_tcp_ofld_queue_rq,
+	.init_request	= nvme_tcp_ofld_init_request,
+	/*
+	 * All additional ops will be also implemented and registered similar to
+	 * tcp.c
+	 */
+};
+
+static const struct nvme_ctrl_ops nvme_tcp_ofld_ctrl_ops = {
+	.name			= "tcp_offload",
+	.module			= THIS_MODULE,
+	.flags			= NVME_F_FABRICS,
+	.reg_read32		= nvmf_reg_read32,
+	.reg_read64		= nvmf_reg_read64,
+	.reg_write32		= nvmf_reg_write32,
+	.free_ctrl		= nvme_tcp_ofld_free_ctrl,
+	.delete_ctrl		= nvme_tcp_ofld_delete_ctrl,
+	.get_address		= nvmf_get_address,
+};
+
+static bool
+nvme_tcp_ofld_existing_controller(struct nvmf_ctrl_options *opts)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl;
+	bool found = false;
+
+	mutex_lock(&nvme_tcp_ofld_ctrl_mutex);
+	list_for_each_entry(ctrl, &nvme_tcp_ofld_ctrl_list, list) {
+		found = nvmf_ip_options_match(&ctrl->nctrl, opts);
+		if (found)
+			break;
+	}
+	mutex_unlock(&nvme_tcp_ofld_ctrl_mutex);
+
+	return found;
+}
+
 static struct nvme_ctrl *
 nvme_tcp_ofld_create_ctrl(struct device *ndev, struct nvmf_ctrl_options *opts)
 {
+	struct nvme_tcp_ofld_queue *queue;
 	struct nvme_tcp_ofld_ctrl *ctrl;
 	struct nvme_tcp_ofld_dev *dev;
 	struct nvme_ctrl *nctrl;
-	int rc = 0;
+	int i, rc = 0;
 
 	ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
 	if (!ctrl)
 		return ERR_PTR(-ENOMEM);
 
+	INIT_LIST_HEAD(&ctrl->list);
 	nctrl = &ctrl->nctrl;
+	nctrl->opts = opts;
+	nctrl->queue_count = opts->nr_io_queues + opts->nr_write_queues +
+			     opts->nr_poll_queues + 1;
+	nctrl->sqsize = opts->queue_size - 1;
+	nctrl->kato = opts->kato;
+	if (!(opts->mask & NVMF_OPT_TRSVCID)) {
+		opts->trsvcid =
+			kstrdup(__stringify(NVME_TCP_DISC_PORT), GFP_KERNEL);
+		if (!opts->trsvcid) {
+			rc = -ENOMEM;
+			goto out_free_ctrl;
+		}
+		opts->mask |= NVMF_OPT_TRSVCID;
+	}
+
+	rc = inet_pton_with_scope(&init_net, AF_UNSPEC, opts->traddr,
+				  opts->trsvcid,
+				  &ctrl->conn_params.remote_ip_addr);
+	if (rc) {
+		pr_err("malformed address passed: %s:%s\n",
+		       opts->traddr, opts->trsvcid);
+		goto out_free_ctrl;
+	}
 
-	/* Init nvme_tcp_ofld_ctrl & nvme_ctrl params based on received opts */
+	if (opts->mask & NVMF_OPT_HOST_TRADDR) {
+		rc = inet_pton_with_scope(&init_net, AF_UNSPEC,
+					  opts->host_traddr, NULL,
+					  &ctrl->conn_params.local_ip_addr);
+		if (rc) {
+			pr_err("malformed src address passed: %s\n",
+			       opts->host_traddr);
+			goto out_free_ctrl;
+		}
+	}
+
+	if (!opts->duplicate_connect &&
+	    nvme_tcp_ofld_existing_controller(opts)) {
+		rc = -EALREADY;
+		goto out_free_ctrl;
+	}
 
 	/* Find device that can reach the dest addr */
 	dev = nvme_tcp_ofld_lookup_dev(ctrl);
@@ -165,6 +589,10 @@ nvme_tcp_ofld_create_ctrl(struct device *ndev, struct nvmf_ctrl_options *opts)
 		goto out_free_ctrl;
 	}
 
+	rc = nvme_tcp_ofld_check_dev_opts(opts, dev->ops);
+	if (rc)
+		goto out_module_put;
+
 	ctrl->dev = dev;
 
 	if (ctrl->dev->ops->max_hw_sectors)
@@ -172,14 +600,55 @@ nvme_tcp_ofld_create_ctrl(struct device *ndev, struct nvmf_ctrl_options *opts)
 	if (ctrl->dev->ops->max_segments)
 		nctrl->max_segments = ctrl->dev->ops->max_segments;
 
-	/* Init queues */
+	ctrl->queues = kcalloc(nctrl->queue_count,
+			       sizeof(struct nvme_tcp_ofld_queue),
+			       GFP_KERNEL);
+	if (!ctrl->queues) {
+		rc = -ENOMEM;
+		goto out_module_put;
+	}
+
+	for (i = 0; i < nctrl->queue_count; ++i) {
+		queue = &ctrl->queues[i];
+		queue->ctrl = ctrl;
+		queue->dev = dev;
+		queue->report_err = nvme_tcp_ofld_report_queue_err;
+	}
+
+	rc = nvme_init_ctrl(nctrl, ndev, &nvme_tcp_ofld_ctrl_ops, 0);
+	if (rc)
+		goto out_free_queues;
+
+	if (!nvme_change_ctrl_state(nctrl, NVME_CTRL_CONNECTING)) {
+		WARN_ON_ONCE(1);
+		rc = -EINTR;
+		goto out_uninit_ctrl;
+	}
 
-	/* Call nvme_init_ctrl */
+	rc = nvme_tcp_ofld_setup_ctrl(nctrl, true);
+	if (rc)
+		goto out_uninit_ctrl;
 
-	/* Setup ctrl */
+	dev_info(nctrl->device, "new ctrl: NQN \"%s\", addr %pISp\n",
+		 opts->subsysnqn, &ctrl->conn_params.remote_ip_addr);
+
+	mutex_lock(&nvme_tcp_ofld_ctrl_mutex);
+	list_add_tail(&ctrl->list, &nvme_tcp_ofld_ctrl_list);
+	mutex_unlock(&nvme_tcp_ofld_ctrl_mutex);
 
 	return nctrl;
 
+out_uninit_ctrl:
+	nvme_uninit_ctrl(nctrl);
+	nvme_put_ctrl(nctrl);
+	if (rc > 0)
+		rc = -EIO;
+
+	return ERR_PTR(rc);
+out_free_queues:
+	kfree(ctrl->queues);
+out_module_put:
+	module_put(dev->ops->module);
 out_free_ctrl:
 	kfree(ctrl);
 
@@ -207,7 +676,15 @@ static int __init nvme_tcp_ofld_init_module(void)
 
 static void __exit nvme_tcp_ofld_cleanup_module(void)
 {
+	struct nvme_tcp_ofld_ctrl *ctrl;
+
 	nvmf_unregister_transport(&nvme_tcp_ofld_transport);
+
+	mutex_lock(&nvme_tcp_ofld_ctrl_mutex);
+	list_for_each_entry(ctrl, &nvme_tcp_ofld_ctrl_list, list)
+		nvme_delete_ctrl(&ctrl->nctrl);
+	mutex_unlock(&nvme_tcp_ofld_ctrl_mutex);
+	flush_workqueue(nvme_delete_wq);
 }
 
 module_init(nvme_tcp_ofld_init_module);
Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#14 lore raw

PATCH v4 6/20 nvme-tcp-offload: Add controller level error recovery implementation

In reply to Prabhakar Kushwaha

From: Arie Gershberg <[email protected]>

In this patch, we implement controller level error handling and recovery. Upon an error discovered by the ULP or reset controller initiated by the nvme-core (using reset_ctrl workqueue), the ULP will initiate a controller recovery which includes teardown and re-connect of all queues.

---
 drivers/nvme/host/tcp-offload.c | 127 +++++++++++++++++++++++++++++++-
 drivers/nvme/host/tcp-offload.h |   1 +
 2 files changed, 126 insertions(+), 2 deletions(-)
drivers/nvme/host/tcp-offload.c +125 -2
index 806247722a35..3af89a938d1b 100644
@@ -72,6 +72,23 @@ void nvme_tcp_ofld_unregister_dev(struct nvme_tcp_ofld_dev *dev)
 }
 EXPORT_SYMBOL_GPL(nvme_tcp_ofld_unregister_dev);
 
+/**
+ * nvme_tcp_ofld_error_recovery() - NVMeTCP Offload library error recovery.
+ * function.
+ * @nctrl:	NVMe controller instance to change to resetting.
+ *
+ * API function that change the controller state to resseting.
+ * Part of the overall controller reset sequence.
+ */
+void nvme_tcp_ofld_error_recovery(struct nvme_ctrl *nctrl)
+{
+	if (!nvme_change_ctrl_state(nctrl, NVME_CTRL_RESETTING))
+		return;
+
+	queue_work(nvme_reset_wq, &to_tcp_ofld_ctrl(nctrl)->err_work);
+}
+EXPORT_SYMBOL_GPL(nvme_tcp_ofld_error_recovery);
+
 /**
  * nvme_tcp_ofld_report_queue_err() - NVMeTCP Offload report error event
  * callback function. Pointed to by nvme_tcp_ofld_queue->report_err.
@@ -82,7 +99,8 @@ EXPORT_SYMBOL_GPL(nvme_tcp_ofld_unregister_dev);
  */
 int nvme_tcp_ofld_report_queue_err(struct nvme_tcp_ofld_queue *queue)
 {
-	/* Placeholder - invoke error recovery flow */
+	pr_err("nvme-tcp-offload queue error\n");
+	nvme_tcp_ofld_error_recovery(&queue->ctrl->nctrl);
 
 	return 0;
 }
@@ -302,6 +320,28 @@ nvme_tcp_ofld_configure_io_queues(struct nvme_ctrl *nctrl, bool new)
 	return rc;
 }
 
+static void nvme_tcp_ofld_reconnect_or_remove(struct nvme_ctrl *nctrl)
+{
+	/* If we are resetting/deleting then do nothing */
+	if (nctrl->state != NVME_CTRL_CONNECTING) {
+		WARN_ON_ONCE(nctrl->state == NVME_CTRL_NEW ||
+			     nctrl->state == NVME_CTRL_LIVE);
+
+		return;
+	}
+
+	if (nvmf_should_reconnect(nctrl)) {
+		dev_info(nctrl->device, "Reconnecting in %d seconds...\n",
+			 nctrl->opts->reconnect_delay);
+		queue_delayed_work(nvme_wq,
+				   &to_tcp_ofld_ctrl(nctrl)->connect_work,
+				   nctrl->opts->reconnect_delay * HZ);
+	} else {
+		dev_info(nctrl->device, "Removing controller...\n");
+		nvme_delete_ctrl(nctrl);
+	}
+}
+
 static int nvme_tcp_ofld_setup_ctrl(struct nvme_ctrl *nctrl, bool new)
 {
 	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(nctrl);
@@ -426,10 +466,63 @@ nvme_tcp_ofld_teardown_io_queues(struct nvme_ctrl *nctrl, bool remove)
 	/* Placeholder - teardown_io_queues */
 }
 
+static void nvme_tcp_ofld_reconnect_ctrl_work(struct work_struct *work)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl =
+				container_of(to_delayed_work(work),
+					     struct nvme_tcp_ofld_ctrl,
+					     connect_work);
+	struct nvme_ctrl *nctrl = &ctrl->nctrl;
+
+	++nctrl->nr_reconnects;
+
+	if (nvme_tcp_ofld_setup_ctrl(nctrl, false))
+		goto requeue;
+
+	dev_info(nctrl->device, "Successfully reconnected (%d attempt)\n",
+		 nctrl->nr_reconnects);
+
+	nctrl->nr_reconnects = 0;
+
+	return;
+
+requeue:
+	dev_info(nctrl->device, "Failed reconnect attempt %d\n",
+		 nctrl->nr_reconnects);
+	nvme_tcp_ofld_reconnect_or_remove(nctrl);
+}
+
+static void nvme_tcp_ofld_error_recovery_work(struct work_struct *work)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl =
+		container_of(work, struct nvme_tcp_ofld_ctrl, err_work);
+	struct nvme_ctrl *nctrl = &ctrl->nctrl;
+
+	nvme_stop_keep_alive(nctrl);
+	nvme_tcp_ofld_teardown_io_queues(nctrl, false);
+	/* unquiesce to fail fast pending requests */
+	nvme_start_queues(nctrl);
+	nvme_tcp_ofld_teardown_admin_queue(nctrl, false);
+	blk_mq_unquiesce_queue(nctrl->admin_q);
+
+	if (!nvme_change_ctrl_state(nctrl, NVME_CTRL_CONNECTING)) {
+		/* state change failure is ok if we started nctrl delete */
+		WARN_ON_ONCE(nctrl->state != NVME_CTRL_DELETING &&
+			     nctrl->state != NVME_CTRL_DELETING_NOIO);
+
+		return;
+	}
+
+	nvme_tcp_ofld_reconnect_or_remove(nctrl);
+}
+
 static void
 nvme_tcp_ofld_teardown_ctrl(struct nvme_ctrl *nctrl, bool shutdown)
 {
-	/* Placeholder - err_work and connect_work */
+	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(nctrl);
+
+	cancel_work_sync(&ctrl->err_work);
+	cancel_delayed_work_sync(&ctrl->connect_work);
 	nvme_tcp_ofld_teardown_io_queues(nctrl, shutdown);
 	blk_mq_quiesce_queue(nctrl->admin_q);
 	if (shutdown)
@@ -444,6 +537,32 @@ static void nvme_tcp_ofld_delete_ctrl(struct nvme_ctrl *nctrl)
 	nvme_tcp_ofld_teardown_ctrl(nctrl, true);
 }
 
+static void nvme_tcp_ofld_reset_ctrl_work(struct work_struct *work)
+{
+	struct nvme_ctrl *nctrl =
+		container_of(work, struct nvme_ctrl, reset_work);
+
+	nvme_stop_ctrl(nctrl);
+	nvme_tcp_ofld_teardown_ctrl(nctrl, false);
+
+	if (!nvme_change_ctrl_state(nctrl, NVME_CTRL_CONNECTING)) {
+		/* state change failure is ok if we started ctrl delete */
+		WARN_ON_ONCE(nctrl->state != NVME_CTRL_DELETING &&
+			     nctrl->state != NVME_CTRL_DELETING_NOIO);
+
+		return;
+	}
+
+	if (nvme_tcp_ofld_setup_ctrl(nctrl, false))
+		goto out_fail;
+
+	return;
+
+out_fail:
+	++nctrl->nr_reconnects;
+	nvme_tcp_ofld_reconnect_or_remove(nctrl);
+}
+
 static int
 nvme_tcp_ofld_init_request(struct blk_mq_tag_set *set,
 			   struct request *rq,
@@ -537,6 +656,10 @@ nvme_tcp_ofld_create_ctrl(struct device *ndev, struct nvmf_ctrl_options *opts)
 			     opts->nr_poll_queues + 1;
 	nctrl->sqsize = opts->queue_size - 1;
 	nctrl->kato = opts->kato;
+	INIT_DELAYED_WORK(&ctrl->connect_work,
+			  nvme_tcp_ofld_reconnect_ctrl_work);
+	INIT_WORK(&ctrl->err_work, nvme_tcp_ofld_error_recovery_work);
+	INIT_WORK(&nctrl->reset_work, nvme_tcp_ofld_reset_ctrl_work);
 	if (!(opts->mask & NVMF_OPT_TRSVCID)) {
 		opts->trsvcid =
 			kstrdup(__stringify(NVME_TCP_DISC_PORT), GFP_KERNEL);
drivers/nvme/host/tcp-offload.h +1 -0
index d1c2c6171897..51fec632b72b 100644
@@ -198,3 +198,4 @@ struct nvme_tcp_ofld_ops {
 /* Exported functions for offload device specific drivers */
 int nvme_tcp_ofld_register_dev(struct nvme_tcp_ofld_dev *dev);
 void nvme_tcp_ofld_unregister_dev(struct nvme_tcp_ofld_dev *dev);
+void nvme_tcp_ofld_error_recovery(struct nvme_ctrl *nctrl);
Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#15 lore raw

PATCH v4 7/20 nvme-tcp-offload: Add queue level implementation

In reply to Prabhakar Kushwaha

From: Dean Balandin <[email protected]>

In this patch we implement queue level functionality.
The implementation is similar to the nvme-tcp module, the main difference being that we call the offload device specific create_queue op which creates the TCP connection, and NVMeTPC connection including icreq+icresp negotiation.
Once create_queue returns successfully, we can move on to the fabrics connect.

---
 drivers/nvme/host/tcp-offload.c | 420 +++++++++++++++++++++++++++++---
 drivers/nvme/host/tcp-offload.h |   4 +
 2 files changed, 396 insertions(+), 28 deletions(-)
drivers/nvme/host/tcp-offload.c +392 -28
index 3af89a938d1b..26253b107db2 100644
@@ -22,6 +22,11 @@ static inline struct nvme_tcp_ofld_ctrl *to_tcp_ofld_ctrl(struct nvme_ctrl *nc)
 	return container_of(nc, struct nvme_tcp_ofld_ctrl, nctrl);
 }
 
+static inline int nvme_tcp_ofld_qid(struct nvme_tcp_ofld_queue *queue)
+{
+	return queue - queue->ctrl->queues;
+}
+
 /**
  * nvme_tcp_ofld_register_dev() - NVMeTCP Offload Library registration
  * function.
@@ -196,19 +201,127 @@ nvme_tcp_ofld_alloc_tagset(struct nvme_ctrl *nctrl, bool admin)
 	return set;
 }
 
+static void __nvme_tcp_ofld_stop_queue(struct nvme_tcp_ofld_queue *queue)
+{
+	queue->dev->ops->drain_queue(queue);
+}
+
+static void nvme_tcp_ofld_stop_queue(struct nvme_ctrl *nctrl, int qid)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(nctrl);
+	struct nvme_tcp_ofld_queue *queue = &ctrl->queues[qid];
+
+	mutex_lock(&queue->queue_lock);
+	if (test_and_clear_bit(NVME_TCP_OFLD_Q_LIVE, &queue->flags))
+		__nvme_tcp_ofld_stop_queue(queue);
+	mutex_unlock(&queue->queue_lock);
+}
+
+static void nvme_tcp_ofld_stop_io_queues(struct nvme_ctrl *ctrl)
+{
+	int i;
+
+	for (i = 1; i < ctrl->queue_count; i++)
+		nvme_tcp_ofld_stop_queue(ctrl, i);
+}
+
+static void __nvme_tcp_ofld_free_queue(struct nvme_tcp_ofld_queue *queue)
+{
+	queue->dev->ops->destroy_queue(queue);
+}
+
+static void nvme_tcp_ofld_free_queue(struct nvme_ctrl *nctrl, int qid)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(nctrl);
+	struct nvme_tcp_ofld_queue *queue = &ctrl->queues[qid];
+
+	if (test_and_clear_bit(NVME_TCP_OFLD_Q_ALLOCATED, &queue->flags)) {
+		__nvme_tcp_ofld_free_queue(queue);
+		mutex_destroy(&queue->queue_lock);
+	}
+}
+
+static void
+nvme_tcp_ofld_free_io_queues(struct nvme_ctrl *nctrl)
+{
+	int i;
+
+	for (i = 1; i < nctrl->queue_count; i++)
+		nvme_tcp_ofld_free_queue(nctrl, i);
+}
+
+static void nvme_tcp_ofld_destroy_io_queues(struct nvme_ctrl *nctrl,
+					    bool remove)
+{
+	nvme_tcp_ofld_stop_io_queues(nctrl);
+	if (remove) {
+		blk_cleanup_queue(nctrl->connect_q);
+		blk_mq_free_tag_set(nctrl->tagset);
+	}
+	nvme_tcp_ofld_free_io_queues(nctrl);
+}
+
+static void nvme_tcp_ofld_destroy_admin_queue(struct nvme_ctrl *nctrl,
+					      bool remove)
+{
+	nvme_tcp_ofld_stop_queue(nctrl, 0);
+	if (remove) {
+		blk_cleanup_queue(nctrl->admin_q);
+		blk_cleanup_queue(nctrl->fabrics_q);
+		blk_mq_free_tag_set(nctrl->admin_tagset);
+	}
+	nvme_tcp_ofld_free_queue(nctrl, 0);
+}
+
+static int nvme_tcp_ofld_start_queue(struct nvme_ctrl *nctrl, int qid)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(nctrl);
+	struct nvme_tcp_ofld_queue *queue = &ctrl->queues[qid];
+	int rc;
+
+	queue = &ctrl->queues[qid];
+	if (qid) {
+		queue->cmnd_capsule_len = nctrl->ioccsz * 16;
+		rc = nvmf_connect_io_queue(nctrl, qid, false);
+	} else {
+		queue->cmnd_capsule_len = sizeof(struct nvme_command) +
+						NVME_TCP_ADMIN_CCSZ;
+		rc = nvmf_connect_admin_queue(nctrl);
+	}
+
+	if (!rc) {
+		set_bit(NVME_TCP_OFLD_Q_LIVE, &queue->flags);
+	} else {
+		if (test_bit(NVME_TCP_OFLD_Q_ALLOCATED, &queue->flags))
+			__nvme_tcp_ofld_stop_queue(queue);
+		dev_err(nctrl->device,
+			"failed to connect queue: %d ret=%d\n", qid, rc);
+	}
+
+	return rc;
+}
+
 static int nvme_tcp_ofld_configure_admin_queue(struct nvme_ctrl *nctrl,
 					       bool new)
 {
+	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(nctrl);
+	struct nvme_tcp_ofld_queue *queue = &ctrl->queues[0];
 	int rc;
 
-	/* Placeholder - alloc_admin_queue */
+	mutex_init(&queue->queue_lock);
+
+	rc = ctrl->dev->ops->create_queue(queue, 0, NVME_AQ_DEPTH);
+	if (rc)
+		return rc;
+
+	set_bit(NVME_TCP_OFLD_Q_ALLOCATED, &queue->flags);
 	if (new) {
 		nctrl->admin_tagset =
 				nvme_tcp_ofld_alloc_tagset(nctrl, true);
 		if (IS_ERR(nctrl->admin_tagset)) {
 			rc = PTR_ERR(nctrl->admin_tagset);
 			nctrl->admin_tagset = NULL;
-			goto out_destroy_queue;
+			goto out_free_queue;
 		}
 
 		nctrl->fabrics_q = blk_mq_init_queue(nctrl->admin_tagset);
@@ -226,7 +339,9 @@ static int nvme_tcp_ofld_configure_admin_queue(struct nvme_ctrl *nctrl,
 		}
 	}
 
-	/* Placeholder - nvme_tcp_ofld_start_queue */
+	rc = nvme_tcp_ofld_start_queue(nctrl, 0);
+	if (rc)
+		goto out_cleanup_queue;
 
 	rc = nvme_enable_ctrl(nctrl);
 	if (rc)
@@ -243,19 +358,143 @@ static int nvme_tcp_ofld_configure_admin_queue(struct nvme_ctrl *nctrl,
 out_quiesce_queue:
 	blk_mq_quiesce_queue(nctrl->admin_q);
 	blk_sync_queue(nctrl->admin_q);
-
 out_stop_queue:
-	/* Placeholder - stop offload queue */
+	nvme_tcp_ofld_stop_queue(nctrl, 0);
 	nvme_cancel_admin_tagset(nctrl);
-
+out_cleanup_queue:
+	if (new)
+		blk_cleanup_queue(nctrl->admin_q);
 out_cleanup_fabrics_q:
 	if (new)
 		blk_cleanup_queue(nctrl->fabrics_q);
 out_free_tagset:
 	if (new)
 		blk_mq_free_tag_set(nctrl->admin_tagset);
-out_destroy_queue:
-	/* Placeholder - free admin queue */
+out_free_queue:
+	nvme_tcp_ofld_free_queue(nctrl, 0);
+
+	return rc;
+}
+
+static unsigned int nvme_tcp_ofld_nr_io_queues(struct nvme_ctrl *nctrl)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(nctrl);
+	struct nvme_tcp_ofld_dev *dev = ctrl->dev;
+	u32 hw_vectors = dev->num_hw_vectors;
+	u32 nr_write_queues, nr_poll_queues;
+	u32 nr_io_queues, nr_total_queues;
+
+	nr_io_queues = min3(nctrl->opts->nr_io_queues, num_online_cpus(),
+			    hw_vectors);
+	nr_write_queues = min3(nctrl->opts->nr_write_queues, num_online_cpus(),
+			       hw_vectors);
+	nr_poll_queues = min3(nctrl->opts->nr_poll_queues, num_online_cpus(),
+			      hw_vectors);
+
+	nr_total_queues = nr_io_queues + nr_write_queues + nr_poll_queues;
+
+	return nr_total_queues;
+}
+
+static void
+nvme_tcp_ofld_set_io_queues(struct nvme_ctrl *nctrl, unsigned int nr_io_queues)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(nctrl);
+	struct nvmf_ctrl_options *opts = nctrl->opts;
+
+	if (opts->nr_write_queues && opts->nr_io_queues < nr_io_queues) {
+		/*
+		 * separate read/write queues
+		 * hand out dedicated default queues only after we have
+		 * sufficient read queues.
+		 */
+		ctrl->io_queues[HCTX_TYPE_READ] = opts->nr_io_queues;
+		nr_io_queues -= ctrl->io_queues[HCTX_TYPE_READ];
+		ctrl->io_queues[HCTX_TYPE_DEFAULT] =
+			min(opts->nr_write_queues, nr_io_queues);
+		nr_io_queues -= ctrl->io_queues[HCTX_TYPE_DEFAULT];
+	} else {
+		/*
+		 * shared read/write queues
+		 * either no write queues were requested, or we don't have
+		 * sufficient queue count to have dedicated default queues.
+		 */
+		ctrl->io_queues[HCTX_TYPE_DEFAULT] =
+			min(opts->nr_io_queues, nr_io_queues);
+		nr_io_queues -= ctrl->io_queues[HCTX_TYPE_DEFAULT];
+	}
+
+	if (opts->nr_poll_queues && nr_io_queues) {
+		/* map dedicated poll queues only if we have queues left */
+		ctrl->io_queues[HCTX_TYPE_POLL] =
+			min(opts->nr_poll_queues, nr_io_queues);
+	}
+}
+
+static int nvme_tcp_ofld_create_io_queues(struct nvme_ctrl *nctrl)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(nctrl);
+	int i, rc;
+
+	for (i = 1; i < nctrl->queue_count; i++) {
+		mutex_init(&ctrl->queues[i].queue_lock);
+
+		rc = ctrl->dev->ops->create_queue(&ctrl->queues[i],
+						  i, nctrl->sqsize + 1);
+		if (rc)
+			goto out_free_queues;
+
+		set_bit(NVME_TCP_OFLD_Q_ALLOCATED, &ctrl->queues[i].flags);
+	}
+
+	return 0;
+
+out_free_queues:
+	for (i--; i >= 1; i--)
+		nvme_tcp_ofld_free_queue(nctrl, i);
+
+	return rc;
+}
+
+static int nvme_tcp_ofld_alloc_io_queues(struct nvme_ctrl *nctrl)
+{
+	unsigned int nr_io_queues;
+	int rc;
+
+	nr_io_queues = nvme_tcp_ofld_nr_io_queues(nctrl);
+	rc = nvme_set_queue_count(nctrl, &nr_io_queues);
+	if (rc)
+		return rc;
+
+	nctrl->queue_count = nr_io_queues + 1;
+	if (nctrl->queue_count < 2) {
+		dev_err(nctrl->device,
+			"unable to set any I/O queues\n");
+
+		return -ENOMEM;
+	}
+
+	dev_info(nctrl->device, "creating %d I/O queues.\n", nr_io_queues);
+	nvme_tcp_ofld_set_io_queues(nctrl, nr_io_queues);
+
+	return nvme_tcp_ofld_create_io_queues(nctrl);
+}
+
+static int nvme_tcp_ofld_start_io_queues(struct nvme_ctrl *nctrl)
+{
+	int i, rc = 0;
+
+	for (i = 1; i < nctrl->queue_count; i++) {
+		rc = nvme_tcp_ofld_start_queue(nctrl, i);
+		if (rc)
+			goto out_stop_queues;
+	}
+
+	return 0;
+
+out_stop_queues:
+	for (i--; i >= 1; i--)
+		nvme_tcp_ofld_stop_queue(nctrl, i);
 
 	return rc;
 }
@@ -263,9 +502,10 @@ static int nvme_tcp_ofld_configure_admin_queue(struct nvme_ctrl *nctrl,
 static int
 nvme_tcp_ofld_configure_io_queues(struct nvme_ctrl *nctrl, bool new)
 {
-	int rc;
+	int rc = nvme_tcp_ofld_alloc_io_queues(nctrl);
 
-	/* Placeholder - alloc_io_queues */
+	if (rc)
+		return rc;
 
 	if (new) {
 		nctrl->tagset = nvme_tcp_ofld_alloc_tagset(nctrl, false);
@@ -283,7 +523,9 @@ nvme_tcp_ofld_configure_io_queues(struct nvme_ctrl *nctrl, bool new)
 		}
 	}
 
-	/* Placeholder - start_io_queues */
+	rc = nvme_tcp_ofld_start_io_queues(nctrl);
+	if (rc)
+		goto out_cleanup_connect_q;
 
 	if (!new) {
 		nvme_start_queues(nctrl);
@@ -306,16 +548,16 @@ nvme_tcp_ofld_configure_io_queues(struct nvme_ctrl *nctrl, bool new)
 out_wait_freeze_timed_out:
 	nvme_stop_queues(nctrl);
 	nvme_sync_io_queues(nctrl);
-
-	/* Placeholder - Stop IO queues */
-
+	nvme_tcp_ofld_stop_io_queues(nctrl);
+out_cleanup_connect_q:
+	nvme_cancel_tagset(nctrl);
 	if (new)
 		blk_cleanup_queue(nctrl->connect_q);
 out_free_tag_set:
 	if (new)
 		blk_mq_free_tag_set(nctrl->tagset);
 out_free_io_queues:
-	/* Placeholder - free_io_queues */
+	nvme_tcp_ofld_free_io_queues(nctrl);
 
 	return rc;
 }
@@ -342,6 +584,17 @@ static void nvme_tcp_ofld_reconnect_or_remove(struct nvme_ctrl *nctrl)
 	}
 }
 
+static int
+nvme_tcp_ofld_init_admin_hctx(struct blk_mq_hw_ctx *hctx, void *data,
+			      unsigned int hctx_idx)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl = data;
+
+	hctx->driver_data = &ctrl->queues[0];
+
+	return 0;
+}
+
 static int nvme_tcp_ofld_setup_ctrl(struct nvme_ctrl *nctrl, bool new)
 {
 	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(nctrl);
@@ -404,9 +657,19 @@ static int nvme_tcp_ofld_setup_ctrl(struct nvme_ctrl *nctrl, bool new)
 	return 0;
 
 destroy_io:
-	/* Placeholder - stop and destroy io queues*/
+	if (nctrl->queue_count > 1) {
+		nvme_stop_queues(nctrl);
+		nvme_sync_io_queues(nctrl);
+		nvme_tcp_ofld_stop_io_queues(nctrl);
+		nvme_cancel_tagset(nctrl);
+		nvme_tcp_ofld_destroy_io_queues(nctrl, new);
+	}
 destroy_admin:
-	/* Placeholder - stop and destroy admin queue*/
+	blk_mq_quiesce_queue(nctrl->admin_q);
+	blk_sync_queue(nctrl->admin_q);
+	nvme_tcp_ofld_stop_queue(nctrl, 0);
+	nvme_cancel_admin_tagset(nctrl);
+	nvme_tcp_ofld_destroy_admin_queue(nctrl, new);
 out_release_ctrl:
 	ctrl->dev->ops->release_ctrl(ctrl);
 
@@ -455,15 +718,37 @@ static void nvme_tcp_ofld_free_ctrl(struct nvme_ctrl *nctrl)
 }
 
 static void
-nvme_tcp_ofld_teardown_admin_queue(struct nvme_ctrl *ctrl, bool remove)
+nvme_tcp_ofld_teardown_admin_queue(struct nvme_ctrl *nctrl, bool remove)
 {
-	/* Placeholder - teardown_admin_queue */
+	blk_mq_quiesce_queue(nctrl->admin_q);
+	blk_sync_queue(nctrl->admin_q);
+
+	nvme_tcp_ofld_stop_queue(nctrl, 0);
+	nvme_cancel_admin_tagset(nctrl);
+
+	if (remove)
+		blk_mq_unquiesce_queue(nctrl->admin_q);
+
+	nvme_tcp_ofld_destroy_admin_queue(nctrl, remove);
 }
 
 static void
 nvme_tcp_ofld_teardown_io_queues(struct nvme_ctrl *nctrl, bool remove)
 {
-	/* Placeholder - teardown_io_queues */
+	if (nctrl->queue_count <= 1)
+		return;
+
+	blk_mq_quiesce_queue(nctrl->admin_q);
+	nvme_start_freeze(nctrl);
+	nvme_stop_queues(nctrl);
+	nvme_sync_io_queues(nctrl);
+	nvme_tcp_ofld_stop_io_queues(nctrl);
+	nvme_cancel_tagset(nctrl);
+
+	if (remove)
+		nvme_start_queues(nctrl);
+
+	nvme_tcp_ofld_destroy_io_queues(nctrl, remove);
 }
 
 static void nvme_tcp_ofld_reconnect_ctrl_work(struct work_struct *work)
@@ -578,6 +863,12 @@ nvme_tcp_ofld_init_request(struct blk_mq_tag_set *set,
 	return 0;
 }
 
+inline size_t nvme_tcp_ofld_inline_data_size(struct nvme_tcp_ofld_queue *queue)
+{
+	return queue->cmnd_capsule_len - sizeof(struct nvme_command);
+}
+EXPORT_SYMBOL_GPL(nvme_tcp_ofld_inline_data_size);
+
 static blk_status_t
 nvme_tcp_ofld_queue_rq(struct blk_mq_hw_ctx *hctx,
 		       const struct blk_mq_queue_data *bd)
@@ -589,22 +880,95 @@ nvme_tcp_ofld_queue_rq(struct blk_mq_hw_ctx *hctx,
 	return BLK_STS_OK;
 }
 
+static void
+nvme_tcp_ofld_exit_request(struct blk_mq_tag_set *set,
+			   struct request *rq, unsigned int hctx_idx)
+{
+	/*
+	 * Nothing is allocated in nvme_tcp_ofld_init_request,
+	 * hence empty.
+	 */
+}
+
+static int
+nvme_tcp_ofld_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
+			unsigned int hctx_idx)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl = data;
+
+	hctx->driver_data = &ctrl->queues[hctx_idx + 1];
+
+	return 0;
+}
+
+static int nvme_tcp_ofld_map_queues(struct blk_mq_tag_set *set)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl = set->driver_data;
+	struct nvmf_ctrl_options *opts = ctrl->nctrl.opts;
+
+	if (opts->nr_write_queues && ctrl->io_queues[HCTX_TYPE_READ]) {
+		/* separate read/write queues */
+		set->map[HCTX_TYPE_DEFAULT].nr_queues =
+			ctrl->io_queues[HCTX_TYPE_DEFAULT];
+		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
+		set->map[HCTX_TYPE_READ].nr_queues =
+			ctrl->io_queues[HCTX_TYPE_READ];
+		set->map[HCTX_TYPE_READ].queue_offset =
+			ctrl->io_queues[HCTX_TYPE_DEFAULT];
+	} else {
+		/* shared read/write queues */
+		set->map[HCTX_TYPE_DEFAULT].nr_queues =
+			ctrl->io_queues[HCTX_TYPE_DEFAULT];
+		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
+		set->map[HCTX_TYPE_READ].nr_queues =
+			ctrl->io_queues[HCTX_TYPE_DEFAULT];
+		set->map[HCTX_TYPE_READ].queue_offset = 0;
+	}
+	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
+	blk_mq_map_queues(&set->map[HCTX_TYPE_READ]);
+
+	if (opts->nr_poll_queues && ctrl->io_queues[HCTX_TYPE_POLL]) {
+		/* map dedicated poll queues only if we have queues left */
+		set->map[HCTX_TYPE_POLL].nr_queues =
+				ctrl->io_queues[HCTX_TYPE_POLL];
+		set->map[HCTX_TYPE_POLL].queue_offset =
+			ctrl->io_queues[HCTX_TYPE_DEFAULT] +
+			ctrl->io_queues[HCTX_TYPE_READ];
+		blk_mq_map_queues(&set->map[HCTX_TYPE_POLL]);
+	}
+
+	dev_info(ctrl->nctrl.device,
+		 "mapped %d/%d/%d default/read/poll queues.\n",
+		 ctrl->io_queues[HCTX_TYPE_DEFAULT],
+		 ctrl->io_queues[HCTX_TYPE_READ],
+		 ctrl->io_queues[HCTX_TYPE_POLL]);
+
+	return 0;
+}
+
+static int nvme_tcp_ofld_poll(struct blk_mq_hw_ctx *hctx)
+{
+	/* Placeholder - Implement polling mechanism */
+
+	return 0;
+}
+
 static struct blk_mq_ops nvme_tcp_ofld_mq_ops = {
 	.queue_rq	= nvme_tcp_ofld_queue_rq,
+	.complete	= nvme_complete_rq,
 	.init_request	= nvme_tcp_ofld_init_request,
-	/*
-	 * All additional ops will be also implemented and registered similar to
-	 * tcp.c
-	 */
+	.exit_request	= nvme_tcp_ofld_exit_request,
+	.init_hctx	= nvme_tcp_ofld_init_hctx,
+	.map_queues	= nvme_tcp_ofld_map_queues,
+	.poll		= nvme_tcp_ofld_poll,
 };
 
 static struct blk_mq_ops nvme_tcp_ofld_admin_mq_ops = {
 	.queue_rq	= nvme_tcp_ofld_queue_rq,
+	.complete	= nvme_complete_rq,
 	.init_request	= nvme_tcp_ofld_init_request,
-	/*
-	 * All additional ops will be also implemented and registered similar to
-	 * tcp.c
-	 */
+	.exit_request	= nvme_tcp_ofld_exit_request,
+	.init_hctx	= nvme_tcp_ofld_init_admin_hctx,
 };
 
 static const struct nvme_ctrl_ops nvme_tcp_ofld_ctrl_ops = {
drivers/nvme/host/tcp-offload.h +4 -0
index 51fec632b72b..b3502c01394e 100644
@@ -65,6 +65,9 @@ struct nvme_tcp_ofld_queue {
 	unsigned long flags;
 	size_t cmnd_capsule_len;
 
+	/* mutex used during stop_queue */
+	struct mutex queue_lock;
+
 	u8 hdr_digest;
 	u8 data_digest;
 	u8 tos;
@@ -199,3 +202,4 @@ struct nvme_tcp_ofld_ops {
 int nvme_tcp_ofld_register_dev(struct nvme_tcp_ofld_dev *dev);
 void nvme_tcp_ofld_unregister_dev(struct nvme_tcp_ofld_dev *dev);
 void nvme_tcp_ofld_error_recovery(struct nvme_ctrl *nctrl);
+inline size_t nvme_tcp_ofld_inline_data_size(struct nvme_tcp_ofld_queue *queue);
Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#16 lore raw

PATCH v4 8/20 nvme-tcp-offload: Add IO level implementation

In reply to Prabhakar Kushwaha

From: Dean Balandin <[email protected]>

In this patch, we present the IO level functionality.
The nvme-tcp-offload shall work on the IO-level, meaning the nvme-tcp-offload ULP module shall pass the request to the nvme-tcp-offload driver and shall expect for the request completion.
No additional handling is needed in between, this design will reduce the CPU utilization as we will describe below.

The nvme-tcp-offload driver shall register to nvme-tcp-offload ULP with the following IO-path ops:
- send_req - in order to pass the request to the handling of the offload driver that shall pass it to the offload specific device
- poll_queue

The offload device driver will manage the context from which the request will be executed and the request aggregations.
Once the IO completed, the nvme-tcp-offload driver shall call command.done() that shall invoke the nvme-tcp-offload ULP layer for completing the request.

This patch also add support for the nvme-tcp-offload timeout and nvme-tcp-offload ASYNC flow.

---
 drivers/nvme/host/tcp-offload.c | 181 ++++++++++++++++++++++++++++++--
 drivers/nvme/host/tcp-offload.h |   2 +
 2 files changed, 176 insertions(+), 7 deletions(-)
drivers/nvme/host/tcp-offload.c +174 -7
index 26253b107db2..501006ec9c97 100644
@@ -125,7 +125,30 @@ void nvme_tcp_ofld_req_done(struct nvme_tcp_ofld_req *req,
 			    union nvme_result *result,
 			    __le16 status)
 {
-	/* Placeholder - complete request with/without error */
+	struct request *rq = blk_mq_rq_from_pdu(req);
+
+	if (!nvme_try_complete_req(rq, cpu_to_le16(status << 1), *result))
+		nvme_complete_rq(rq);
+}
+
+/**
+ * nvme_tcp_ofld_async_req_done() - NVMeTCP Offload request done callback
+ * function for async request. Pointed to by nvme_tcp_ofld_req->done.
+ * Handles both NVME_TCP_F_DATA_SUCCESS flag and NVMe CQ.
+ * @req:	NVMeTCP offload request to complete.
+ * @result:     The nvme_result.
+ * @status:     The completion status.
+ *
+ * API function that allows the offload device specific driver to report
+ * request completions to the common offload layer.
+ */
+void nvme_tcp_ofld_async_req_done(struct nvme_tcp_ofld_req *req,
+				  union nvme_result *result, __le16 status)
+{
+	struct nvme_tcp_ofld_queue *queue = req->queue;
+	struct nvme_tcp_ofld_ctrl *ctrl = queue->ctrl;
+
+	nvme_complete_async_event(&ctrl->nctrl, status, result);
 }
 
 static struct nvme_tcp_ofld_dev *
@@ -717,6 +740,57 @@ static void nvme_tcp_ofld_free_ctrl(struct nvme_ctrl *nctrl)
 	kfree(ctrl);
 }
 
+static void nvme_tcp_ofld_set_sg_null(struct nvme_command *c)
+{
+	struct nvme_sgl_desc *sg = &c->common.dptr.sgl;
+
+	sg->addr = 0;
+	sg->length = 0;
+	sg->type = (NVME_TRANSPORT_SGL_DATA_DESC << 4) |
+			NVME_SGL_FMT_TRANSPORT_A;
+}
+
+inline void nvme_tcp_ofld_set_sg_inline(struct nvme_tcp_ofld_queue *queue,
+					struct nvme_command *c, u32 data_len)
+{
+	struct nvme_sgl_desc *sg = &c->common.dptr.sgl;
+
+	sg->addr = cpu_to_le64(queue->ctrl->nctrl.icdoff);
+	sg->length = cpu_to_le32(data_len);
+	sg->type = (NVME_SGL_FMT_DATA_DESC << 4) | NVME_SGL_FMT_OFFSET;
+}
+
+static void nvme_tcp_ofld_set_sg_host_data(struct nvme_command *c,
+					   u32 data_len)
+{
+	struct nvme_sgl_desc *sg = &c->common.dptr.sgl;
+
+	sg->addr = 0;
+	sg->length = cpu_to_le32(data_len);
+	sg->type = (NVME_TRANSPORT_SGL_DATA_DESC << 4) |
+			NVME_SGL_FMT_TRANSPORT_A;
+}
+
+static void nvme_tcp_ofld_submit_async_event(struct nvme_ctrl *arg)
+{
+	struct nvme_tcp_ofld_ctrl *ctrl = to_tcp_ofld_ctrl(arg);
+	struct nvme_tcp_ofld_queue *queue = &ctrl->queues[0];
+	struct nvme_tcp_ofld_dev *dev = queue->dev;
+	struct nvme_tcp_ofld_ops *ops = dev->ops;
+
+	ctrl->async_req.nvme_cmd.common.opcode = nvme_admin_async_event;
+	ctrl->async_req.nvme_cmd.common.command_id = NVME_AQ_BLK_MQ_DEPTH;
+	ctrl->async_req.nvme_cmd.common.flags |= NVME_CMD_SGL_METABUF;
+
+	nvme_tcp_ofld_set_sg_null(&ctrl->async_req.nvme_cmd);
+
+	ctrl->async_req.async = true;
+	ctrl->async_req.queue = queue;
+	ctrl->async_req.done = nvme_tcp_ofld_async_req_done;
+
+	ops->send_req(&ctrl->async_req);
+}
+
 static void
 nvme_tcp_ofld_teardown_admin_queue(struct nvme_ctrl *nctrl, bool remove)
 {
@@ -855,9 +929,13 @@ nvme_tcp_ofld_init_request(struct blk_mq_tag_set *set,
 			   unsigned int numa_node)
 {
 	struct nvme_tcp_ofld_req *req = blk_mq_rq_to_pdu(rq);
+	struct nvme_tcp_ofld_ctrl *ctrl = set->driver_data;
+	int qid;
 
-	/* Placeholder - init request */
-
+	qid = (set == &ctrl->tag_set) ? hctx_idx + 1 : 0;
+	req->queue = &ctrl->queues[qid];
+	nvme_req(rq)->ctrl = &ctrl->nctrl;
+	nvme_req(rq)->cmd = &req->nvme_cmd;
 	req->done = nvme_tcp_ofld_req_done;
 
 	return 0;
@@ -873,9 +951,46 @@ static blk_status_t
 nvme_tcp_ofld_queue_rq(struct blk_mq_hw_ctx *hctx,
 		       const struct blk_mq_queue_data *bd)
 {
-	/* Call nvme_setup_cmd(...) */
+	struct nvme_tcp_ofld_req *req = blk_mq_rq_to_pdu(bd->rq);
+	struct nvme_tcp_ofld_queue *queue = hctx->driver_data;
+	struct nvme_tcp_ofld_ctrl *ctrl = queue->ctrl;
+	struct nvme_ns *ns = hctx->queue->queuedata;
+	struct nvme_tcp_ofld_dev *dev = queue->dev;
+	struct nvme_tcp_ofld_ops *ops = dev->ops;
+	struct nvme_command *nvme_cmd;
+	struct request *rq = bd->rq;
+	bool queue_ready;
+	u32 data_len;
+	int rc;
+
+	queue_ready = test_bit(NVME_TCP_OFLD_Q_LIVE, &queue->flags);
+
+	req->async = false;
+
+	if (!nvme_check_ready(&ctrl->nctrl, rq, queue_ready))
+		return nvme_fail_nonready_command(&ctrl->nctrl, rq);
+
+	rc = nvme_setup_cmd(ns, rq);
+	if (unlikely(rc))
+		return rc;
 
-	/* Call ops->send_req(...) */
+	blk_mq_start_request(rq);
+
+	nvme_cmd = &req->nvme_cmd;
+	nvme_cmd->common.flags |= NVME_CMD_SGL_METABUF;
+
+	data_len = blk_rq_nr_phys_segments(rq) ? blk_rq_payload_bytes(rq) : 0;
+	if (!data_len)
+		nvme_tcp_ofld_set_sg_null(&req->nvme_cmd);
+	else if ((rq_data_dir(rq) == WRITE) &&
+		 data_len <= nvme_tcp_ofld_inline_data_size(queue))
+		nvme_tcp_ofld_set_sg_inline(queue, nvme_cmd, data_len);
+	else
+		nvme_tcp_ofld_set_sg_host_data(nvme_cmd, data_len);
+
+	rc = ops->send_req(req);
+	if (unlikely(rc))
+		return rc;
 
 	return BLK_STS_OK;
 }
@@ -948,9 +1063,58 @@ static int nvme_tcp_ofld_map_queues(struct blk_mq_tag_set *set)
 
 static int nvme_tcp_ofld_poll(struct blk_mq_hw_ctx *hctx)
 {
-	/* Placeholder - Implement polling mechanism */
+	struct nvme_tcp_ofld_queue *queue = hctx->driver_data;
+	struct nvme_tcp_ofld_dev *dev = queue->dev;
+	struct nvme_tcp_ofld_ops *ops = dev->ops;
 
-	return 0;
+	return ops->poll_queue(queue);
+}
+
+static void nvme_tcp_ofld_complete_timed_out(struct request *rq)
+{
+	struct nvme_tcp_ofld_req *req = blk_mq_rq_to_pdu(rq);
+	struct nvme_ctrl *nctrl = &req->queue->ctrl->nctrl;
+
+	nvme_tcp_ofld_stop_queue(nctrl, nvme_tcp_ofld_qid(req->queue));
+	if (blk_mq_request_started(rq) && !blk_mq_request_completed(rq)) {
+		nvme_req(rq)->status = NVME_SC_HOST_ABORTED_CMD;
+		blk_mq_complete_request(rq);
+	}
+}
+
+static enum blk_eh_timer_return nvme_tcp_ofld_timeout(struct request *rq,
+						      bool reserved)
+{
+	struct nvme_tcp_ofld_req *req = blk_mq_rq_to_pdu(rq);
+	struct nvme_tcp_ofld_ctrl *ctrl = req->queue->ctrl;
+
+	dev_warn(ctrl->nctrl.device,
+		 "queue %d: timeout request %#x type %d\n",
+		 nvme_tcp_ofld_qid(req->queue), rq->tag,
+		 req->nvme_cmd.common.opcode);
+
+	if (ctrl->nctrl.state != NVME_CTRL_LIVE) {
+		/*
+		 * If we are resetting, connecting or deleting we should
+		 * complete immediately because we may block controller
+		 * teardown or setup sequence
+		 * - ctrl disable/shutdown fabrics requests
+		 * - connect requests
+		 * - initialization admin requests
+		 * - I/O requests that entered after unquiescing and
+		 *   the controller stopped responding
+		 *
+		 * All other requests should be cancelled by the error
+		 * recovery work, so it's fine that we fail it here.
+		 */
+		nvme_tcp_ofld_complete_timed_out(rq);
+
+		return BLK_EH_DONE;
+	}
+
+	nvme_tcp_ofld_error_recovery(&ctrl->nctrl);
+
+	return BLK_EH_RESET_TIMER;
 }
 
 static struct blk_mq_ops nvme_tcp_ofld_mq_ops = {
@@ -959,6 +1123,7 @@ static struct blk_mq_ops nvme_tcp_ofld_mq_ops = {
 	.init_request	= nvme_tcp_ofld_init_request,
 	.exit_request	= nvme_tcp_ofld_exit_request,
 	.init_hctx	= nvme_tcp_ofld_init_hctx,
+	.timeout	= nvme_tcp_ofld_timeout,
 	.map_queues	= nvme_tcp_ofld_map_queues,
 	.poll		= nvme_tcp_ofld_poll,
 };
@@ -969,6 +1134,7 @@ static struct blk_mq_ops nvme_tcp_ofld_admin_mq_ops = {
 	.init_request	= nvme_tcp_ofld_init_request,
 	.exit_request	= nvme_tcp_ofld_exit_request,
 	.init_hctx	= nvme_tcp_ofld_init_admin_hctx,
+	.timeout	= nvme_tcp_ofld_timeout,
 };
 
 static const struct nvme_ctrl_ops nvme_tcp_ofld_ctrl_ops = {
@@ -979,6 +1145,7 @@ static const struct nvme_ctrl_ops nvme_tcp_ofld_ctrl_ops = {
 	.reg_read64		= nvmf_reg_read64,
 	.reg_write32		= nvmf_reg_write32,
 	.free_ctrl		= nvme_tcp_ofld_free_ctrl,
+	.submit_async_event     = nvme_tcp_ofld_submit_async_event,
 	.delete_ctrl		= nvme_tcp_ofld_delete_ctrl,
 	.get_address		= nvmf_get_address,
 };
drivers/nvme/host/tcp-offload.h +2 -0
index b3502c01394e..a4c28ddaf3ab 100644
@@ -115,6 +115,8 @@ struct nvme_tcp_ofld_ctrl {
 	/* Connectivity params */
 	struct nvme_tcp_ofld_ctrl_con_params conn_params;
 
+	struct nvme_tcp_ofld_req async_req;
+
 	/* Offload device driver context */
 	void *private_data;
 };
Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#17 lore raw

PATCH v4 9/20 qedn: Add qedn - Marvell's NVMeTCP HW offload device driver

In reply to Prabhakar Kushwaha

From: Shai Malin <[email protected]>

This patch will present the skeleton of the qedn driver. The new driver will be added under "drivers/nvme/hw/qedn" and will be enabled by the Kconfig "Marvell NVM Express over Fabrics TCP offload".

The internal implementation:
- qedn.h:
Includes all common structs to be used by the qedn offload device driver.

- qedn_main.c
Includes the qedn_init and qedn_cleanup implementation.
As part of the qedn init, the driver will register as a pci device and will work with the Marvell fastlinQ NICs.
As part of the probe, the driver will register to the nvme_tcp_offload (ULP).

---
 MAINTAINERS                      |  10 ++
 drivers/nvme/Kconfig             |   1 +
 drivers/nvme/Makefile            |   1 +
 drivers/nvme/hw/Kconfig          |   8 ++
 drivers/nvme/hw/Makefile         |   3 +
 drivers/nvme/hw/qedn/Makefile    |   5 +
 drivers/nvme/hw/qedn/qedn.h      |  19 +++
 drivers/nvme/hw/qedn/qedn_main.c | 200 +++++++++++++++++++++++++++++++
 8 files changed, 247 insertions(+)
 create mode 100644 drivers/nvme/hw/Kconfig
 create mode 100644 drivers/nvme/hw/Makefile
 create mode 100644 drivers/nvme/hw/qedn/Makefile
 create mode 100644 drivers/nvme/hw/qedn/qedn.h
 create mode 100644 drivers/nvme/hw/qedn/qedn_main.c
MAINTAINERS +10 -0
index 01fbebdc7722..207a62b768c5 100644
@@ -15006,6 +15006,16 @@ S:	Supported
 F:	drivers/infiniband/hw/qedr/
 F:	include/uapi/rdma/qedr-abi.h
 
+QLOGIC QL4xxx NVME-TCP-OFFLOAD DRIVER
+M:	Shai Malin <[email protected]>
+M:	Ariel Elior <[email protected]>
+L:	[email protected]
+S:	Supported
+W:	http://git.infradead.org/nvme.git
+T:	git://git.infradead.org/nvme.git
+F:	drivers/nvme/hw/qedn/
+F:	include/linux/qed/
+
 QLOGIC QLA1280 SCSI DRIVER
 M:	Michael Reed <[email protected]>
 L:	[email protected]
drivers/nvme/Kconfig +1 -0
index 87ae409a32b9..827c2c9f0ad1 100644
@@ -3,5 +3,6 @@ menu "NVME Support"
 
 source "drivers/nvme/host/Kconfig"
 source "drivers/nvme/target/Kconfig"
+source "drivers/nvme/hw/Kconfig"
 
 endmenu
drivers/nvme/Makefile +1 -0
index fb42c44609a8..14c569040ef2 100644
@@ -2,3 +2,4 @@
 
 obj-y		+= host/
 obj-y		+= target/
+obj-y		+= hw/
\ No newline at end of file
drivers/nvme/hw/Kconfig new file +8 -0
new file mode 100644
index 000000000000..374f1f9dbd3d
@@ -0,0 +1,8 @@
+# SPDX-License-Identifier: GPL-2.0-only
+config NVME_QEDN
+	tristate "Marvell NVM Express over Fabrics TCP offload"
+	depends on NVME_TCP_OFFLOAD
+	help
+	  This enables the Marvell NVMe TCP offload support (qedn).
+
+	  If unsure, say N.
drivers/nvme/hw/Makefile new file +3 -0
new file mode 100644
index 000000000000..2f38e0520795
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: GPL-2.0
+
+obj-$(CONFIG_NVME_QEDN)		+= qedn/
drivers/nvme/hw/qedn/Makefile new file +5 -0
new file mode 100644
index 000000000000..1422cd878680
@@ -0,0 +1,5 @@
+# SPDX-License-Identifier: GPL-2.0
+
+obj-$(CONFIG_NVME_QEDN) := qedn.o
+
+qedn-y := qedn_main.o
drivers/nvme/hw/qedn/qedn.h new file +19 -0
new file mode 100644
index 000000000000..bcd0748a10fd
@@ -0,0 +1,19 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright 2021 Marvell. All rights reserved.
+ */
+
+#ifndef _QEDN_H_
+#define _QEDN_H_
+
+/* Driver includes */
+#include "../../host/tcp-offload.h"
+
+#define QEDN_MODULE_NAME "qedn"
+
+struct qedn_ctx {
+	struct pci_dev *pdev;
+	struct nvme_tcp_ofld_dev qedn_ofld_dev;
+};
+
+#endif /* _QEDN_H_ */
drivers/nvme/hw/qedn/qedn_main.c new file +200 -0

200 lines not on this page · raw message

Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#18 lore raw

Re: [PATCH v4 09/20] qedn: Add qedn - Marvell's NVMeTCP HW offload device driver

Quoted 5 lines · Prabhakar Kushwaha

From: Shai Malin <[email protected]>

This patch will present the skeleton of the qedn driver. The new driver will be added under "drivers/nvme/hw/qedn" and will be enabled by the Kconfig "Marvell NVM Express over Fabrics TCP offload".

I don't see why we need a separate hw/ directory. nvme-pci.c already
is very much a hardware driver.

drivers/nvme/hw/Kconfig line 4

Quoted 3 lines
drivers/nvme/hw/Kconfig new file +3 -0
+config NVME_QEDN
+	tristate "Marvell NVM Express over Fabrics TCP offload"
+	depends on NVME_TCP_OFFLOAD

I think it also depends on PCI.

This whole patch is a bit pointless. In general splitting a driver submission into multiple patches is not very helpful unless the later patches add pretty much optional and clearly separatable bits. Otherwise it just becomes really hard to review.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#19 lore raw

Re: [PATCH v4 09/20] qedn: Add qedn - Marvell's NVMeTCP HW offload device driver

Quoted 9 lines · Christoph Hellwig
Quoted 5 lines · Prabhakar Kushwaha

From: Shai Malin <[email protected]>

This patch will present the skeleton of the qedn driver. The new driver will be added under "drivers/nvme/hw/qedn" and will be enabled by the Kconfig "Marvell NVM Express over Fabrics TCP offload".

I don't see why we need a separate hw/ directory. nvme-pci.c already
is very much a hardware driver.

We will restructure and keep only the bare minimum qedn.c under nvme/host.

Quoted 6 lines

drivers/nvme/hw/Kconfig line 4

Quoted 4 lines
drivers/nvme/hw/Kconfig new file +3 -0
+config NVME_QEDN
+     tristate "Marvell NVM Express over Fabrics TCP offload"
+     depends on NVME_TCP_OFFLOAD

I think it also depends on PCI.

Thanks!

Quoted 5 lines

This whole patch is a bit pointless. In general splitting a driver submission into multiple patches is not very helpful unless the later patches add pretty much optional and clearly separatable bits. Otherwise it just becomes really hard to review.

Understood. We will improve it.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#20 lore raw

PATCH v4 10/20 qedn: Add qedn probe

In reply to Prabhakar Kushwaha

From: Shai Malin <[email protected]>

This patch introduces the functionality of loading and unloading physical function.
qedn_probe() loads the offload device PF(physical function), and initialize the HW and the FW with the PF parameters using the HW ops->qed_nvmetcp_ops, which are similar to other "qed_*_ops" which are used by the qede, qedr, qedf and qedi device drivers. qedn_remove() unloads the offload device PF, re-initialize the HW and the FW with the PF parameters.

The struct qedn_ctx is per PF container for PF-specific attributes and resources.

---
 drivers/nvme/hw/Kconfig          |   1 +
 drivers/nvme/hw/qedn/qedn.h      |  26 ++++++
 drivers/nvme/hw/qedn/qedn_main.c | 155 ++++++++++++++++++++++++++++++-
 3 files changed, 177 insertions(+), 5 deletions(-)
drivers/nvme/hw/Kconfig +1 -0
index 374f1f9dbd3d..91b1bd6f07d8 100644
@@ -2,6 +2,7 @@
 config NVME_QEDN
 	tristate "Marvell NVM Express over Fabrics TCP offload"
 	depends on NVME_TCP_OFFLOAD
+	select QED_NVMETCP
 	help
 	  This enables the Marvell NVMe TCP offload support (qedn).
 
drivers/nvme/hw/qedn/qedn.h +26 -0
index bcd0748a10fd..931efc3afbaa 100644
@@ -6,14 +6,40 @@
 #ifndef _QEDN_H_
 #define _QEDN_H_
 
+#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_nvmetcp_if.h>
+
 /* Driver includes */
 #include "../../host/tcp-offload.h"
 
 #define QEDN_MODULE_NAME "qedn"
 
+#define QEDN_MAX_TASKS_PER_PF (16 * 1024)
+#define QEDN_MAX_CONNS_PER_PF (4 * 1024)
+#define QEDN_FW_CQ_SIZE (4 * 1024)
+#define QEDN_PROTO_CQ_PROD_IDX	0
+#define QEDN_NVMETCP_NUM_FW_CONN_QUEUE_PAGES 2
+
+enum qedn_state {
+	QEDN_STATE_CORE_PROBED = 0,
+	QEDN_STATE_CORE_OPEN,
+	QEDN_STATE_MFW_STATE,
+	QEDN_STATE_REGISTERED_OFFLOAD_DEV,
+	QEDN_STATE_MODULE_REMOVE_ONGOING,
+};
+
 struct qedn_ctx {
 	struct pci_dev *pdev;
+	struct qed_dev *cdev;
+	struct qed_dev_nvmetcp_info dev_info;
 	struct nvme_tcp_ofld_dev qedn_ofld_dev;
+	struct qed_pf_params pf_params;
+
+	/* Accessed with atomic bit ops, used with enum qedn_state */
+	unsigned long state;
+
+	/* Fast path queues */
+	u8 num_fw_cqs;
 };
 
 #endif /* _QEDN_H_ */
drivers/nvme/hw/qedn/qedn_main.c +150 -5

195 lines not on this page · raw message

Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#21 lore raw

Re: [PATCH v4 10/20] qedn: Add qedn probe

Quoted 33 lines · Prabhakar Kushwaha

From: Shai Malin <[email protected]>

This patch introduces the functionality of loading and unloading physical function.
qedn_probe() loads the offload device PF(physical function), and initialize the HW and the FW with the PF parameters using the HW ops->qed_nvmetcp_ops, which are similar to other "qed_*_ops" which are used by the qede, qedr, qedf and qedi device drivers. qedn_remove() unloads the offload device PF, re-initialize the HW and the FW with the PF parameters.

The struct qedn_ctx is per PF container for PF-specific attributes and resources.

---
 drivers/nvme/hw/Kconfig          |   1 +
 drivers/nvme/hw/qedn/qedn.h      |  26 ++++++
 drivers/nvme/hw/qedn/qedn_main.c | 155 ++++++++++++++++++++++++++++++-
 3 files changed, 177 insertions(+), 5 deletions(-)
drivers/nvme/hw/Kconfig +0 -0
index 374f1f9dbd3d..91b1bd6f07d8 100644

drivers/nvme/hw/Kconfig line 5

Quoted 4 lines
drivers/nvme/hw/Kconfig +1 -0
@@ -2,6 +2,7 @@
 config NVME_QEDN
 	tristate "Marvell NVM Express over Fabrics TCP offload"
 	depends on NVME_TCP_OFFLOAD
+	select QED_NVMETCP

This makes kconfig unhappy:

WARNING: unmet direct dependencies detected for QED_NVMETCP
Depends on [n]: NETDEVICES [=y] && ETHERNET [=y] && NET_VENDOR_QLOGIC [=n] Selected by [y]:
- NVME_QEDN [=y] && NVME_TCP_OFFLOAD [=y]

While we're at unhappy: without pinpointing it to the pointlessly split patches, this series generates tons of sparse warnings:

drivers/nvme/hw/qedn/qedn_main.c:17:30: warning: symbol 'qed_ops' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_main.c:38:24: warning: symbol 'qedn_add_llh_filter' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_main.c:90:6: warning: symbol 'qedn_dec_llh_filter' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_main.c:493:6: warning: symbol 'qedn_fw_cq_fp_handler' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_main.c:813:58: warning: incorrect type in assignment (different base types)
drivers/nvme/hw/qedn/qedn_main.c:813:58: expected restricted __le32 [usertype] hi
drivers/nvme/hw/qedn/qedn_main.c:813:58: got unsigned int [usertype]
drivers/nvme/hw/qedn/qedn_main.c:814:58: warning: incorrect type in assignment (different base types)
drivers/nvme/hw/qedn/qedn_main.c:814:58: expected restricted __le32 [usertype] lo
drivers/nvme/hw/qedn/qedn_main.c:814:58: got unsigned int [usertype]
drivers/nvme/hw/qedn/qedn_task.c:358:6: warning: symbol 'qedn_return_task_to_pool' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_task.c:422:5: warning: symbol 'qedn_send_read_cmd' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_task.c:502:32: warning: incorrect type in assignment (different base types)
drivers/nvme/hw/qedn/qedn_task.c:502:32: expected unsigned short [usertype] host_cccid
drivers/nvme/hw/qedn/qedn_task.c:502:32: got restricted __le16 [usertype]
drivers/nvme/hw/qedn/qedn_task.c:471:5: warning: symbol 'qedn_send_write_cmd' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_task.c:553:24: warning: incorrect type in return expression (different base types)
drivers/nvme/hw/qedn/qedn_task.c:553:24: expected int
drivers/nvme/hw/qedn/qedn_task.c:553:24: got restricted blk_status_t [usertype]
drivers/nvme/hw/qedn/qedn_task.c:576:24: warning: incorrect type in return expression (different base types)
drivers/nvme/hw/qedn/qedn_task.c:576:24: expected int
drivers/nvme/hw/qedn/qedn_task.c:576:24: got restricted blk_status_t [usertype]
drivers/nvme/hw/qedn/qedn_task.c:586:22: warning: symbol 'qedn_cqe_get_active_task' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_task.c:629:5: warning: symbol 'qedn_process_nvme_cqe' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_task.c:646:5: warning: symbol 'qedn_complete_c2h' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_task.c: note: in included file (through drivers/nvme/hw/qedn/qedn.h):
drivers/nvme/hw/qedn/../../host/tcp-offload.h:207:45: error: marked inline, but without a definition drivers/nvme/hw/qedn/qedn_conn.c:107:22: warning: incorrect type in assignment (different base types)
drivers/nvme/hw/qedn/qedn_conn.c:107:22: expected unsigned short [usertype] src_port
drivers/nvme/hw/qedn/qedn_conn.c:107:22: got restricted __be16 [usertype] sin_port
drivers/nvme/hw/qedn/qedn_conn.c:98:5: warning: symbol 'qedn_fill_ep_addr4' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_conn.c:126:22: warning: incorrect type in assignment (different base types)
drivers/nvme/hw/qedn/qedn_conn.c:126:22: expected unsigned short [usertype] src_port
drivers/nvme/hw/qedn/qedn_conn.c:126:22: got restricted __be16 [usertype] sin6_port
drivers/nvme/hw/qedn/qedn_conn.c:116:5: warning: symbol 'qedn_fill_ep_addr6' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_conn.c:557:23: warning: cast from restricted __le16 drivers/nvme/hw/qedn/qedn_conn.c:560:27: warning: cast from restricted __le32 drivers/nvme/hw/qedn/qedn_conn.c:629:6: warning: symbol 'qedn_error_recovery' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_conn.c:727:6: warning: symbol 'qedn_prep_db_data' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_conn.c:933:6: warning: symbol 'qedn_cleanup_all_fw_tasks' was not declared. Should it be static? drivers/nvme/hw/qedn/qedn_conn.c:976:6: warning: symbol 'qedn_destroy_connection' was not declared. Should it be static?

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#22 lore raw

Re: [PATCH v4 10/20] qedn: Add qedn probe

Quoted 41 lines · Christoph Hellwig
Quoted 33 lines · Prabhakar Kushwaha

From: Shai Malin <[email protected]>

This patch introduces the functionality of loading and unloading physical function.
qedn_probe() loads the offload device PF(physical function), and initialize the HW and the FW with the PF parameters using the HW ops->qed_nvmetcp_ops, which are similar to other "qed_*_ops" which are used by the qede, qedr, qedf and qedi device drivers. qedn_remove() unloads the offload device PF, re-initialize the HW and the FW with the PF parameters.

The struct qedn_ctx is per PF container for PF-specific attributes and resources.

---
 drivers/nvme/hw/Kconfig          |   1 +
 drivers/nvme/hw/qedn/qedn.h      |  26 ++++++
 drivers/nvme/hw/qedn/qedn_main.c | 155 ++++++++++++++++++++++++++++++-
 3 files changed, 177 insertions(+), 5 deletions(-)
drivers/nvme/hw/Kconfig +0 -0
index 374f1f9dbd3d..91b1bd6f07d8 100644

drivers/nvme/hw/Kconfig line 5

Quoted 4 lines
drivers/nvme/hw/Kconfig +1 -0
@@ -2,6 +2,7 @@
 config NVME_QEDN
      tristate "Marvell NVM Express over Fabrics TCP offload"
      depends on NVME_TCP_OFFLOAD
+     select QED_NVMETCP

This makes kconfig unhappy:

Quoted 8 lines

WARNING: unmet direct dependencies detected for QED_NVMETCP
Depends on [n]: NETDEVICES [=y] && ETHERNET [=y] && NET_VENDOR_QLOGIC [=n] Selected by [y]:
- NVME_QEDN [=y] && NVME_TCP_OFFLOAD [=y]

While we're at unhappy: without pinpointing it to the pointlessly split patches, this series generates tons of sparse warnings:

Sorry about that. Will be fixed.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#23 lore raw

PATCH v4 11/20 qedn: Add qedn_claim_dev API support

In reply to Prabhakar Kushwaha

From: Nikolay Assa <[email protected]>

This patch introduces the qedn_claim_dev() network service which the offload device (qedn) is using through the paired net-device (qede). qedn_claim_dev() returns true if the IP addr(IPv4 or IPv6) of the target server is reachable via the net-device which is paired with the offloaded device.

---
 drivers/nvme/hw/qedn/qedn.h      |  4 +++
 drivers/nvme/hw/qedn/qedn_main.c | 55 ++++++++++++++++++++++++++++++--
 2 files changed, 56 insertions(+), 3 deletions(-)
drivers/nvme/hw/qedn/qedn.h +4 -0
index 931efc3afbaa..0ce1e19d1ba8 100644
@@ -8,6 +8,10 @@
 
 #include <linux/qed/qed_if.h>
 #include <linux/qed/qed_nvmetcp_if.h>
+#include <linux/qed/qed_nvmetcp_ip_services_if.h>
+#include <linux/qed/qed_chain.h>
+#include <linux/qed/storage_common.h>
+#include <linux/qed/nvmetcp_common.h>
 
 /* Driver includes */
 #include "../../host/tcp-offload.h"
drivers/nvme/hw/qedn/qedn_main.c +52 -3

65 lines not on this page · raw message

Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#24 lore raw

PATCH v4 12/20 qedn: Add IRQ and fast-path resources initializations

In reply to Prabhakar Kushwaha

From: Shai Malin <[email protected]>

This patch will present the adding of qedn_fp_queue - this is a per cpu core element which handles all of the connections on that cpu core. The qedn_fp_queue will handle a group of connections (NVMeoF QPs) which are handled on the same cpu core, and will only use the same FW-driver resources with no need to be related to the same NVMeoF controller.

The per qedn_fq_queue resources are the FW CQ and FW status block:
- The FW CQ will be used for the FW to notify the driver that the the exchange has ended and the FW will pass the incoming NVMeoF CQE (if exist) to the driver.
- FW status block - which is used for the FW to notify the driver with the producer update of the FW CQE chain.

The FW fast-path queues are based on qed_chain.h

---
 drivers/nvme/hw/qedn/qedn.h      |  25 +++
 drivers/nvme/hw/qedn/qedn_main.c | 289 ++++++++++++++++++++++++++++++-
 2 files changed, 311 insertions(+), 3 deletions(-)
drivers/nvme/hw/qedn/qedn.h +25 -0
index 0ce1e19d1ba8..edb0836bca87 100644
@@ -24,17 +24,39 @@
 #define QEDN_PROTO_CQ_PROD_IDX	0
 #define QEDN_NVMETCP_NUM_FW_CONN_QUEUE_PAGES 2
 
+#define QEDN_PAGE_SIZE	4096 /* FW page size - Configurable */
+#define QEDN_IRQ_NAME_LEN 24
+#define QEDN_IRQ_NO_FLAGS 0
+
+#define QEDN_TCP_RTO_DEFAULT 280
+
 enum qedn_state {
 	QEDN_STATE_CORE_PROBED = 0,
 	QEDN_STATE_CORE_OPEN,
 	QEDN_STATE_MFW_STATE,
+	QEDN_STATE_NVMETCP_OPEN,
+	QEDN_STATE_IRQ_SET,
+	QEDN_STATE_FP_WORK_THREAD_SET,
 	QEDN_STATE_REGISTERED_OFFLOAD_DEV,
 	QEDN_STATE_MODULE_REMOVE_ONGOING,
 };
 
+/* Per CPU core params */
+struct qedn_fp_queue {
+	struct qed_chain cq_chain;
+	u16 *cq_prod;
+	struct mutex cq_mutex; /* cq handler mutex */
+	struct qedn_ctx	*qedn;
+	struct qed_sb_info *sb_info;
+	unsigned int cpu;
+	u16 sb_id;
+	char irqname[QEDN_IRQ_NAME_LEN];
+};
+
 struct qedn_ctx {
 	struct pci_dev *pdev;
 	struct qed_dev *cdev;
+	struct qed_int_info int_info;
 	struct qed_dev_nvmetcp_info dev_info;
 	struct nvme_tcp_ofld_dev qedn_ofld_dev;
 	struct qed_pf_params pf_params;
@@ -44,6 +66,9 @@ struct qedn_ctx {
 
 	/* Fast path queues */
 	u8 num_fw_cqs;
+	struct qedn_fp_queue *fp_q_arr;
+	struct nvmetcp_glbl_queue_entry *fw_cq_array_virt;
+	dma_addr_t fw_cq_array_phy; /* Physical address of fw_cq_array_virt */
 };
 
 #endif /* _QEDN_H_ */
drivers/nvme/hw/qedn/qedn_main.c +286 -3

348 lines not on this page · raw message

Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#25 lore raw

PATCH v4 13/20 qedn: Add connection-level slowpath functionality

In reply to Prabhakar Kushwaha

This patch will present the connection (queue) level slowpath implementation relevant for create_queue flow.

The internal implementation:
- Add per controller slowpath workqeueue via pre_setup_ctrl

- qedn_main.c:
Includes qedn's implementation of the create_queue op.

- qedn_conn.c will include main slowpath connection level functions, including:
1. Per-queue resources allocation.
2. Creating a new connection.
3. Offloading the connection to the FW for TCP handshake.
4. Destroy of a connection.
5. Support of delete and free controller.
6. TCP port management via qed_fetch_tcp_port, qed_return_tcp_port

---
 drivers/nvme/hw/qedn/Makefile    |   5 +-
 drivers/nvme/hw/qedn/qedn.h      | 179 ++++++++++
 drivers/nvme/hw/qedn/qedn_conn.c | 564 +++++++++++++++++++++++++++++++
 drivers/nvme/hw/qedn/qedn_main.c | 210 +++++++++++-
 4 files changed, 948 insertions(+), 10 deletions(-)
 create mode 100644 drivers/nvme/hw/qedn/qedn_conn.c
drivers/nvme/hw/qedn/Makefile +2 -3
index 1422cd878680..ece84772d317 100644
@@ -1,5 +1,4 @@
 # SPDX-License-Identifier: GPL-2.0
 
-obj-$(CONFIG_NVME_QEDN) := qedn.o
-
-qedn-y := qedn_main.o
+obj-$(CONFIG_NVME_QEDN) += qedn.o
+qedn-y := qedn_main.o qedn_conn.o
drivers/nvme/hw/qedn/qedn.h +179 -0

202 lines not on this page · raw message

drivers/nvme/hw/qedn/qedn_conn.c new file +564 -0

564 lines not on this page · raw message

drivers/nvme/hw/qedn/qedn_main.c +203 -7

280 lines not on this page · raw message

Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#26 lore raw

PATCH v4 14/20 qedn: Add support of configuring HW filter block

In reply to Prabhakar Kushwaha

HW filter can be configured to filter TCP packets based on either source or target TCP port. QEDN leverage this feature to route NVMeTCP traffic.

This patch configures HW filter block based on source port for all receiving packets to deliver correct QEDN PF.

---
 drivers/nvme/hw/qedn/qedn.h      |  15 ++++
 drivers/nvme/hw/qedn/qedn_main.c | 113 ++++++++++++++++++++++++++++++-
 2 files changed, 127 insertions(+), 1 deletion(-)
drivers/nvme/hw/qedn/qedn.h +15 -0
index 38ab4ff88999..9e16297fa323 100644
@@ -29,6 +29,11 @@
 #define QEDN_IRQ_NAME_LEN 24
 #define QEDN_IRQ_NO_FLAGS 0
 
+/* HW defines */
+
+/* QEDN_MAX_LLH_PORTS will be extended in future */
+#define QEDN_MAX_LLH_PORTS 16
+
 /* Destroy connection defines */
 #define QEDN_NON_ABORTIVE_TERMINATION 0
 #define QEDN_ABORTIVE_TERMINATION 1
@@ -68,6 +73,7 @@
 enum qedn_state {
 	QEDN_STATE_CORE_PROBED = 0,
 	QEDN_STATE_CORE_OPEN,
+	QEDN_STATE_LLH_PORT_FILTER_SET,
 	QEDN_STATE_MFW_STATE,
 	QEDN_STATE_NVMETCP_OPEN,
 	QEDN_STATE_IRQ_SET,
@@ -99,6 +105,8 @@ struct qedn_ctx {
 	/* Accessed with atomic bit ops, used with enum qedn_state */
 	unsigned long state;
 
+	u8 num_llh_filters;
+	struct list_head llh_filter_list;
 	u8 local_mac_addr[ETH_ALEN];
 	u16 mtu;
 
@@ -165,6 +173,12 @@ enum qedn_conn_state {
 	CONN_STATE_DESTROY_COMPLETE
 };
 
+struct qedn_llh_filter {
+	struct list_head entry;
+	u16 port;
+	u16 ref_cnt;
+};
+
 struct qedn_ctrl {
 	struct list_head glb_entry;
 	struct list_head pf_entry;
@@ -249,5 +263,6 @@ int qedn_set_con_state(struct qedn_conn_ctx *conn_ctx,
 		       enum qedn_conn_state new_state);
 void qedn_terminate_connection(struct qedn_conn_ctx *conn_ctx);
 void qedn_cleanp_fw(struct qedn_conn_ctx *conn_ctx);
+__be16 qedn_get_in_port(struct sockaddr_storage *sa);
 
 #endif /* _QEDN_H_ */
drivers/nvme/hw/qedn/qedn_main.c +112 -1

151 lines not on this page · raw message

Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#27 lore raw

PATCH v4 15/20 qedn: Add IO level qedn_send_req and fw_cq workqueue

In reply to Prabhakar Kushwaha

From: Shai Malin <[email protected]>

This patch will present the IO level skeleton flows:

- qedn_send_req(): process new requests, similar to nvme_tcp_queue_rq().

- qedn_fw_cq_fp_wq(): process new FW completions, the flow starts from
the IRQ handler and for a single interrupt it will process all the pending NVMeoF Completions under polling mode.

---
 drivers/nvme/hw/qedn/Makefile    |  2 +-
 drivers/nvme/hw/qedn/qedn.h      | 14 ++++++
 drivers/nvme/hw/qedn/qedn_conn.c |  2 +
 drivers/nvme/hw/qedn/qedn_main.c | 84 +++++++++++++++++++++++++++++---
 drivers/nvme/hw/qedn/qedn_task.c | 79 ++++++++++++++++++++++++++++++
 5 files changed, 172 insertions(+), 9 deletions(-)
 create mode 100644 drivers/nvme/hw/qedn/qedn_task.c
drivers/nvme/hw/qedn/Makefile +1 -1
index ece84772d317..888d466fa5ed 100644
@@ -1,4 +1,4 @@
 # SPDX-License-Identifier: GPL-2.0
 
 obj-$(CONFIG_NVME_QEDN) += qedn.o
-qedn-y := qedn_main.o qedn_conn.o
+qedn-y := qedn_main.o qedn_conn.o qedn_task.o
\ No newline at end of file
drivers/nvme/hw/qedn/qedn.h +14 -0
index 9e16297fa323..13d63f420a23 100644
@@ -38,6 +38,8 @@
 #define QEDN_NON_ABORTIVE_TERMINATION 0
 #define QEDN_ABORTIVE_TERMINATION 1
 
+#define QEDN_FW_CQ_FP_WQ_WORKQUEUE "qedn_fw_cq_fp_wq"
+
 /*
  * TCP offload stack default configurations and defines.
  * Future enhancements will allow controlling the configurable
@@ -90,6 +92,7 @@ struct qedn_fp_queue {
 	struct qedn_ctx	*qedn;
 	struct qed_sb_info *sb_info;
 	unsigned int cpu;
+	struct work_struct fw_cq_fp_wq_entry;
 	u16 sb_id;
 	char irqname[QEDN_IRQ_NAME_LEN];
 };
@@ -118,6 +121,7 @@ struct qedn_ctx {
 	struct qedn_fp_queue *fp_q_arr;
 	struct nvmetcp_glbl_queue_entry *fw_cq_array_virt;
 	dma_addr_t fw_cq_array_phy; /* Physical address of fw_cq_array_virt */
+	struct workqueue_struct *fw_cq_fp_wq;
 };
 
 struct qedn_endpoint {
@@ -204,6 +208,12 @@ struct qedn_ctrl {
 
 /* Connection level struct */
 struct qedn_conn_ctx {
+	/* IO path */
+	struct qedn_fp_queue *fp_q;
+	/* mutex for queueing request */
+	struct mutex send_mutex;
+	int qid;
+
 	struct qedn_ctx *qedn;
 	struct nvme_tcp_ofld_queue *queue;
 	struct nvme_tcp_ofld_ctrl *ctrl;
@@ -264,5 +274,9 @@ int qedn_set_con_state(struct qedn_conn_ctx *conn_ctx,
 void qedn_terminate_connection(struct qedn_conn_ctx *conn_ctx);
 void qedn_cleanp_fw(struct qedn_conn_ctx *conn_ctx);
 __be16 qedn_get_in_port(struct sockaddr_storage *sa);
+int qedn_queue_request(struct qedn_conn_ctx *qedn_conn,
+		       struct nvme_tcp_ofld_req *req);
+void qedn_nvme_req_fp_wq_handler(struct work_struct *work);
+void qedn_io_work_cq(struct qedn_ctx *qedn, struct nvmetcp_fw_cqe *cqe);
 
 #endif /* _QEDN_H_ */
drivers/nvme/hw/qedn/qedn_conn.c +2 -0
index 04c45a6fa14b..962c88a4f345 100644
@@ -190,6 +190,7 @@ static void qedn_release_conn_ctx(struct qedn_conn_ctx *conn_ctx)
 		pr_err("Conn resources state isn't 0 as expected 0x%lx\n",
 		       conn_ctx->resrc_state);
 
+	mutex_destroy(&conn_ctx->send_mutex);
 	atomic_inc(&conn_ctx->destroy_conn_indicator);
 	qedn_set_con_state(conn_ctx, CONN_STATE_DESTROY_COMPLETE);
 	wake_up_interruptible(&conn_ctx->conn_waitq);
@@ -429,6 +430,7 @@ static int qedn_prep_and_offload_queue(struct qedn_conn_ctx *conn_ctx)
 	}
 
 	set_bit(QEDN_CONN_RESRC_FW_SQ, &conn_ctx->resrc_state);
+
 	rc = qed_ops->acquire_conn(qedn->cdev,
 				   &conn_ctx->conn_handle,
 				   &conn_ctx->fw_cid,
drivers/nvme/hw/qedn/qedn_main.c +76 -8

135 lines not on this page · raw message

drivers/nvme/hw/qedn/qedn_task.c new file +79 -0

79 lines not on this page · raw message

Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#28 lore raw

PATCH v4 16/20 qedn: Add support of Task and SGL

In reply to Prabhakar Kushwaha

This patch will add support of Task and SGL which is used for slowpath and fast path IO. here Task is IO granule used by firmware to perform tasks

The internal implementation:
- Create task/sgl resources used by all connection
- Provide APIs to allocate and free task.
- Add task support during connection establishment i.e. slowpath

---
 drivers/nvme/hw/qedn/qedn.h      |  61 ++++++
 drivers/nvme/hw/qedn/qedn_conn.c |  46 ++++-
 drivers/nvme/hw/qedn/qedn_main.c |  34 +++-
 drivers/nvme/hw/qedn/qedn_task.c | 326 +++++++++++++++++++++++++++++++
 4 files changed, 463 insertions(+), 4 deletions(-)
drivers/nvme/hw/qedn/qedn.h +61 -0
index 13d63f420a23..78efa8c02810 100644
@@ -40,6 +40,16 @@
 
 #define QEDN_FW_CQ_FP_WQ_WORKQUEUE "qedn_fw_cq_fp_wq"
 
+/* Protocol defines */
+#define QEDN_MAX_IO_SIZE QED_NVMETCP_MAX_IO_SIZE
+
+#define QEDN_FW_SLOW_IO_MIN_SGE_LIMIT (9700 / 6)
+
+#define QEDN_MAX_HW_SECTORS (QEDN_MAX_IO_SIZE / 512)
+#define QEDN_MAX_SEGMENTS 2048
+
+#define QEDN_INVALID_ITID 0xFFFF
+
 /*
  * TCP offload stack default configurations and defines.
  * Future enhancements will allow controlling the configurable
@@ -84,6 +94,15 @@ enum qedn_state {
 	QEDN_STATE_MODULE_REMOVE_ONGOING,
 };
 
+struct qedn_io_resources {
+	/* Lock for IO resources */
+	spinlock_t resources_lock;
+	struct list_head task_free_list;
+	u32 num_alloc_tasks;
+	u32 num_free_tasks;
+	u32 no_avail_resrc_cnt;
+};
+
 /* Per CPU core params */
 struct qedn_fp_queue {
 	struct qed_chain cq_chain;
@@ -93,6 +112,10 @@ struct qedn_fp_queue {
 	struct qed_sb_info *sb_info;
 	unsigned int cpu;
 	struct work_struct fw_cq_fp_wq_entry;
+
+	/* IO related resources for host */
+	struct qedn_io_resources host_resrc;
+
 	u16 sb_id;
 	char irqname[QEDN_IRQ_NAME_LEN];
 };
@@ -116,12 +139,35 @@ struct qedn_ctx {
 	/* Connections */
 	DECLARE_HASHTABLE(conn_ctx_hash, 16);
 
+	u32 num_tasks_per_pool;
+
 	/* Fast path queues */
 	u8 num_fw_cqs;
 	struct qedn_fp_queue *fp_q_arr;
 	struct nvmetcp_glbl_queue_entry *fw_cq_array_virt;
 	dma_addr_t fw_cq_array_phy; /* Physical address of fw_cq_array_virt */
 	struct workqueue_struct *fw_cq_fp_wq;
+
+	/* Fast Path Tasks */
+	struct qed_nvmetcp_tid	tasks;
+};
+
+struct qedn_task_ctx {
+	struct qedn_conn_ctx *qedn_conn;
+	struct qedn_ctx *qedn;
+	void *fw_task_ctx;
+	struct qedn_fp_queue *fp_q;
+	struct scatterlist *nvme_sg;
+	struct nvme_tcp_ofld_req *req; /* currently proccessed request */
+	struct list_head entry;
+	spinlock_t lock; /* To protect task resources */
+	bool valid;
+	unsigned long flags; /* Used by qedn_task_flags */
+	u32 task_size;
+	u16 itid;
+	u16 cccid;
+	int req_direction;
+	struct storage_sgl_task_params sgl_task_params;
 };
 
 struct qedn_endpoint {
@@ -219,6 +265,7 @@ struct qedn_conn_ctx {
 	struct nvme_tcp_ofld_ctrl *ctrl;
 	u32 conn_handle;
 	u32 fw_cid;
+	u8 default_cq;
 
 	atomic_t est_conn_indicator;
 	atomic_t destroy_conn_indicator;
@@ -236,6 +283,11 @@ struct qedn_conn_ctx {
 	dma_addr_t host_cccid_itid_phy_addr;
 	struct qedn_endpoint ep;
 	int abrt_flag;
+	/* Spinlock for accessing active_task_list */
+	spinlock_t task_list_lock;
+	struct list_head active_task_list;
+	atomic_t num_active_tasks;
+	atomic_t num_active_fw_tasks;
 
 	/* Connection resources - turned on to indicate what resource was
 	 * allocated, to that it can later be released.
@@ -255,6 +307,7 @@ struct qedn_conn_ctx {
 enum qedn_conn_resources_state {
 	QEDN_CONN_RESRC_FW_SQ,
 	QEDN_CONN_RESRC_ACQUIRE_CONN,
+	QEDN_CONN_RESRC_TASKS,
 	QEDN_CONN_RESRC_CCCID_ITID_MAP,
 	QEDN_CONN_RESRC_TCP_PORT,
 	QEDN_CONN_RESRC_DB_ADD,
@@ -278,5 +331,13 @@ int qedn_queue_request(struct qedn_conn_ctx *qedn_conn,
 		       struct nvme_tcp_ofld_req *req);
 void qedn_nvme_req_fp_wq_handler(struct work_struct *work);
 void qedn_io_work_cq(struct qedn_ctx *qedn, struct nvmetcp_fw_cqe *cqe);
+int qedn_alloc_tasks(struct qedn_conn_ctx *conn_ctx);
+inline int qedn_qid(struct nvme_tcp_ofld_queue *queue);
+void qedn_common_clear_fw_sgl(struct storage_sgl_task_params *sgl_task_params);
+void qedn_return_active_tasks(struct qedn_conn_ctx *conn_ctx);
+struct qedn_task_ctx *
+qedn_get_free_task_from_pool(struct qedn_conn_ctx *conn_ctx, u16 cccid);
+void qedn_destroy_free_tasks(struct qedn_fp_queue *fp_q,
+			     struct qedn_io_resources *io_resrc);
 
 #endif /* _QEDN_H_ */
drivers/nvme/hw/qedn/qedn_conn.c +45 -1

83 lines not on this page · raw message

drivers/nvme/hw/qedn/qedn_main.c +31 -3
index 975949ce6fb0..42c8ad6ac2d6 100644
@@ -29,6 +29,12 @@ __be16 qedn_get_in_port(struct sockaddr_storage *sa)
 		: ((struct sockaddr_in6 *)sa)->sin6_port;
 }
 
+static void qedn_init_io_resc(struct qedn_io_resources *io_resrc)
+{
+	spin_lock_init(&io_resrc->resources_lock);
+	INIT_LIST_HEAD(&io_resrc->task_free_list);
+}
+
 struct qedn_llh_filter *qedn_add_llh_filter(struct qedn_ctx *qedn, u16 tcp_port)
 {
 	struct qedn_llh_filter *llh_filter = NULL;
@@ -436,6 +442,8 @@ static struct nvme_tcp_ofld_ops qedn_ofld_ops = {
 		 *	NVMF_OPT_HDR_DIGEST | NVMF_OPT_DATA_DIGEST |
 		 *	NVMF_OPT_NR_POLL_QUEUES | NVMF_OPT_TOS
 		 */
+	.max_hw_sectors = QEDN_MAX_HW_SECTORS,
+	.max_segments = QEDN_MAX_SEGMENTS,
 	.claim_dev = qedn_claim_dev,
 	.setup_ctrl = qedn_setup_ctrl,
 	.release_ctrl = qedn_release_ctrl,
@@ -640,8 +648,24 @@ static inline int qedn_core_probe(struct qedn_ctx *qedn)
 	return rc;
 }
 
+static void qedn_call_destroy_free_tasks(struct qedn_fp_queue *fp_q,
+					 struct qedn_io_resources *io_resrc)
+{
+	if (list_empty(&io_resrc->task_free_list))
+		return;
+
+	if (io_resrc->num_alloc_tasks != io_resrc->num_free_tasks)
+		pr_err("Task Pool:Not all returned allocated=0x%x, free=0x%x\n",
+		       io_resrc->num_alloc_tasks, io_resrc->num_free_tasks);
+
+	qedn_destroy_free_tasks(fp_q, io_resrc);
+	if (io_resrc->num_free_tasks)
+		pr_err("Expected num_free_tasks to be 0\n");
+}
+
 static void qedn_free_function_queues(struct qedn_ctx *qedn)
 {
+	struct qedn_io_resources *host_resrc;
 	struct qed_sb_info *sb_info = NULL;
 	struct qedn_fp_queue *fp_q;
 	int i;
@@ -653,6 +677,9 @@ static void qedn_free_function_queues(struct qedn_ctx *qedn)
 	/* Free the fast path queues*/
 	for (i = 0; i < qedn->num_fw_cqs; i++) {
 		fp_q = &qedn->fp_q_arr[i];
+		host_resrc = &fp_q->host_resrc;
+
+		qedn_call_destroy_free_tasks(fp_q, host_resrc);
 
 		/* Free SB */
 		sb_info = fp_q->sb_info;
@@ -740,7 +767,8 @@ static int qedn_alloc_function_queues(struct qedn_ctx *qedn)
 		goto mem_alloc_failure;
 	}
 
-	/* placeholder - create task pools */
+	qedn->num_tasks_per_pool =
+		qedn->pf_params.nvmetcp_pf_params.num_tasks / qedn->num_fw_cqs;
 
 	for (i = 0; i < qedn->num_fw_cqs; i++) {
 		fp_q = &qedn->fp_q_arr[i];
@@ -782,7 +810,7 @@ static int qedn_alloc_function_queues(struct qedn_ctx *qedn)
 		fp_q->qedn = qedn;
 		INIT_WORK(&fp_q->fw_cq_fp_wq_entry, qedn_fw_cq_fq_wq_handler);
 
-		/* Placeholder - Init IO-path resources */
+		qedn_init_io_resc(&fp_q->host_resrc);
 	}
 
 	return 0;
@@ -964,7 +992,7 @@ static int __qedn_probe(struct pci_dev *pdev)
 
 	/* NVMeTCP start HW PF */
 	rc = qed_ops->start(qedn->cdev,
-			    NULL /* Placeholder for FW IO-path resources */,
+			    &qedn->tasks,
 			    qedn,
 			    qedn_event_cb);
 	if (rc) {
drivers/nvme/hw/qedn/qedn_task.c +326 -0

332 lines not on this page · raw message

Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#29 lore raw

PATCH v4 17/20 qedn: Add support of NVME ICReq & ICResp

In reply to Prabhakar Kushwaha

Once a TCP connection established, the host sends an Initialize Connection Request (ICReq) PDU to the controller. Further Initialize Connection Response (ICResp) PDU received from controller is processed by host to establish a connection and exchange connection configuration parameters.

This patch present support of generation of ICReq and processing of ICResp. It also update host configuration based on exchanged parameters.

---
 drivers/nvme/hw/qedn/qedn.h      |  38 ++++
 drivers/nvme/hw/qedn/qedn_conn.c | 334 ++++++++++++++++++++++++++++++-
 drivers/nvme/hw/qedn/qedn_main.c |  14 ++
 drivers/nvme/hw/qedn/qedn_task.c |   8 +-
 4 files changed, 390 insertions(+), 4 deletions(-)
drivers/nvme/hw/qedn/qedn.h +38 -0
index 78efa8c02810..829d474b3ab1 100644
@@ -16,6 +16,7 @@
 
 /* Driver includes */
 #include "../../host/tcp-offload.h"
+#include <linux/nvme-tcp.h>
 
 #define QEDN_MODULE_NAME "qedn"
 
@@ -42,6 +43,8 @@
 
 /* Protocol defines */
 #define QEDN_MAX_IO_SIZE QED_NVMETCP_MAX_IO_SIZE
+#define QEDN_MAX_PDU_SIZE 0x80000 /* 512KB */
+#define QEDN_MAX_OUTSTANDING_R2T_PDUS 0 /* 0 Based == 1 max R2T */
 
 #define QEDN_FW_SLOW_IO_MIN_SGE_LIMIT (9700 / 6)
 
@@ -50,6 +53,13 @@
 
 #define QEDN_INVALID_ITID 0xFFFF
 
+#define QEDN_ICREQ_FW_PAYLOAD (sizeof(struct nvme_tcp_icreq_pdu) -\
+			       QED_NVMETCP_NON_IO_HDR_SIZE)
+#define QEDN_ICREQ_FW_PAYLOAD_START 8
+
+/* The FW will handle the ICReq as CCCID 0 (FW internal design) */
+#define QEDN_ICREQ_CCCID 0
+
 /*
  * TCP offload stack default configurations and defines.
  * Future enhancements will allow controlling the configurable
@@ -120,6 +130,16 @@ struct qedn_fp_queue {
 	char irqname[QEDN_IRQ_NAME_LEN];
 };
 
+struct qedn_negotiation_params {
+	u32 maxh2cdata; /* Negotiation */
+	u32 maxr2t; /* Validation */
+	u16 pfv; /* Validation */
+	bool hdr_digest; /* Negotiation */
+	bool data_digest; /* Negotiation */
+	u8 cpda; /* Negotiation */
+	u8 hpda; /* Validation */
+};
+
 struct qedn_ctx {
 	struct pci_dev *pdev;
 	struct qed_dev *cdev;
@@ -176,6 +196,9 @@ struct qedn_endpoint {
 	struct nvmetcp_db_data db_data;
 	void __iomem *p_doorbell;
 
+	/* Spinlock for accessing FW queue */
+	spinlock_t doorbell_lock;
+
 	/* TCP Params */
 	__be32 dst_addr[4]; /* In network order */
 	__be32 src_addr[4]; /* In network order */
@@ -252,6 +275,12 @@ struct qedn_ctrl {
 	atomic_t host_num_active_conns;
 };
 
+struct qedn_icreq_padding {
+	u32 *buffer;
+	dma_addr_t pa;
+	struct nvmetcp_sge sge;
+};
+
 /* Connection level struct */
 struct qedn_conn_ctx {
 	/* IO path */
@@ -300,6 +329,11 @@ struct qedn_conn_ctx {
 
 	size_t sq_depth;
 
+	struct qedn_negotiation_params required_params;
+	struct qedn_negotiation_params pdu_params;
+	struct nvme_tcp_icresp_pdu icresp;
+	struct qedn_icreq_padding *icreq_pad;
+
 	/* "dummy" socket */
 	struct socket *sock;
 };
@@ -308,6 +342,7 @@ enum qedn_conn_resources_state {
 	QEDN_CONN_RESRC_FW_SQ,
 	QEDN_CONN_RESRC_ACQUIRE_CONN,
 	QEDN_CONN_RESRC_TASKS,
+	QEDN_CONN_RESRC_ICREQ_PAD,
 	QEDN_CONN_RESRC_CCCID_ITID_MAP,
 	QEDN_CONN_RESRC_TCP_PORT,
 	QEDN_CONN_RESRC_DB_ADD,
@@ -339,5 +374,8 @@ struct qedn_task_ctx *
 qedn_get_free_task_from_pool(struct qedn_conn_ctx *conn_ctx, u16 cccid);
 void qedn_destroy_free_tasks(struct qedn_fp_queue *fp_q,
 			     struct qedn_io_resources *io_resrc);
+void qedn_prep_icresp(struct qedn_conn_ctx *conn_ctx,
+		      struct nvmetcp_fw_cqe *cqe);
+void qedn_ring_doorbell(struct qedn_conn_ctx *conn_ctx);
 
 #endif /* _QEDN_H_ */
drivers/nvme/hw/qedn/qedn_conn.c +333 -1

388 lines not on this page · raw message

drivers/nvme/hw/qedn/qedn_main.c +14 -0
index 42c8ad6ac2d6..3cf913d527c0 100644
@@ -286,6 +286,19 @@ static int qedn_release_ctrl(struct nvme_tcp_ofld_ctrl *ctrl)
 	return 0;
 }
 
+static void qedn_set_pdu_params(struct qedn_conn_ctx *conn_ctx)
+{
+	/* Enable digest once supported */
+	conn_ctx->required_params.hdr_digest = 0;
+	conn_ctx->required_params.data_digest = 0;
+
+	conn_ctx->required_params.maxr2t = QEDN_MAX_OUTSTANDING_R2T_PDUS;
+	conn_ctx->required_params.pfv = NVME_TCP_PFV_1_0;
+	conn_ctx->required_params.cpda = 0;
+	conn_ctx->required_params.hpda = 0;
+	conn_ctx->required_params.maxh2cdata = QEDN_MAX_PDU_SIZE;
+}
+
 static int qedn_create_queue(struct nvme_tcp_ofld_queue *queue, int qid,
 			     size_t queue_size)
 {
@@ -317,6 +330,7 @@ static int qedn_create_queue(struct nvme_tcp_ofld_queue *queue, int qid,
 	conn_ctx->ctrl = ctrl;
 	conn_ctx->sq_depth = queue_size;
 	mutex_init(&conn_ctx->send_mutex);
+	qedn_set_pdu_params(conn_ctx);
 
 	init_waitqueue_head(&conn_ctx->conn_waitq);
 	atomic_set(&conn_ctx->est_conn_indicator, 0);
drivers/nvme/hw/qedn/qedn_task.c +5 -3
index 7b228b5c5169..e52460f9650e 100644
@@ -392,9 +392,11 @@ void qedn_io_work_cq(struct qedn_ctx *qedn, struct nvmetcp_fw_cqe *cqe)
 			break;
 
 		case NVMETCP_TASK_TYPE_INIT_CONN_REQUEST:
-
-			/* Placeholder - ICReq flow */
-
+			/* Clear ICReq-padding SGE from SGL */
+			qedn_common_clear_fw_sgl(&qedn_task->sgl_task_params);
+			/* Task is not required for icresp processing */
+			qedn_return_task_to_pool(conn_ctx, qedn_task);
+			qedn_prep_icresp(conn_ctx, cqe);
 			break;
 		default:
 			pr_info("Could not identify task type\n");
Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#30 lore raw

PATCH v4 18/20 qedn: Add IO level fastpath functionality

In reply to Prabhakar Kushwaha

From: Shai Malin <[email protected]>

This patch will present the IO level functionality of qedn nvme-tcp-offload host mode. The qedn_task_ctx structure is containing various params and state of the current IO, and is mapped 1x1 to the fw_task_ctx which is a HW and FW IO context.
A qedn_task is mapped directly to its parent connection.
For every new IO a qedn_task structure will be assigned and they will be linked for the entire IO's life span.

The patch will include 2 flows:
1. Send new command to the FW:
The flow is: nvme_tcp_ofld_queue_rq() which invokes qedn_send_req() which invokes qedn_queue_request() which will:
- Assign fw_task_ctx.
- Prepare the Read/Write SG buffer.
- Initialize the HW and FW context.
- Pass the IO to the FW.

2. Process the IO completion:
The flow is: qedn_irq_handler() which invokes qedn_fw_cq_fp_handler()
which invokes qedn_io_work_cq() which will:
- process the FW completion.
- Return the fw_task_ctx to the task pool.
- complete the nvme req.

---
 drivers/nvme/hw/qedn/qedn.h      |   5 +
 drivers/nvme/hw/qedn/qedn_conn.c |   1 +
 drivers/nvme/hw/qedn/qedn_main.c |   8 +
 drivers/nvme/hw/qedn/qedn_task.c | 317 ++++++++++++++++++++++++++++++-
 4 files changed, 327 insertions(+), 4 deletions(-)
drivers/nvme/hw/qedn/qedn.h +5 -0
index 829d474b3ab1..b36994be65cb 100644
@@ -172,6 +172,10 @@ struct qedn_ctx {
 	struct qed_nvmetcp_tid	tasks;
 };
 
+enum qedn_task_flags {
+	QEDN_TASK_USED_BY_FW,
+};
+
 struct qedn_task_ctx {
 	struct qedn_conn_ctx *qedn_conn;
 	struct qedn_ctx *qedn;
@@ -376,6 +380,7 @@ void qedn_destroy_free_tasks(struct qedn_fp_queue *fp_q,
 			     struct qedn_io_resources *io_resrc);
 void qedn_prep_icresp(struct qedn_conn_ctx *conn_ctx,
 		      struct nvmetcp_fw_cqe *cqe);
+void qedn_swap_bytes(u32 *p, int size);
 void qedn_ring_doorbell(struct qedn_conn_ctx *conn_ctx);
 
 #endif /* _QEDN_H_ */
drivers/nvme/hw/qedn/qedn_conn.c +1 -0
index b4c0a1a3e890..ea072eff34a6 100644
@@ -528,6 +528,7 @@ static int qedn_send_icreq(struct qedn_conn_ctx *conn_ctx)
 				     sgl_task_params,  NULL);
 
 	qedn_set_con_state(conn_ctx, CONN_STATE_WAIT_FOR_IC_COMP);
+	set_bit(QEDN_TASK_USED_BY_FW, &qedn_task->flags);
 	atomic_inc(&conn_ctx->num_active_fw_tasks);
 
 	/* spin_lock - doorbell is accessed  both Rx flow and response flow */
drivers/nvme/hw/qedn/qedn_main.c +8 -0
index 3cf913d527c0..fb47e315ab03 100644
@@ -1047,6 +1047,14 @@ static int qedn_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	return __qedn_probe(pdev);
 }
 
+void qedn_swap_bytes(u32 *p, int size)
+{
+	int i;
+
+	for (i = 0; i < size; ++i, ++p)
+		*p = __swab32(*p);
+}
+
 static struct pci_driver qedn_pci_driver = {
 	.name     = QEDN_MODULE_NAME,
 	.id_table = qedn_pci_tbl,
drivers/nvme/hw/qedn/qedn_task.c +313 -4

352 lines not on this page · raw message

Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#31 lore raw

PATCH v4 19/20 qedn: Add Connection and IO level recovery flows

In reply to Prabhakar Kushwaha

From: Shai Malin <[email protected]>

This patch will present the connection level functionalities:
- conn clear-sq: will release the FW restrictions in order to flush all the pending IOs.
- drain: in case clear-sq is stuck, will release all the device FW restrictions in order to flush all the pending IOs.
- task cleanup - will flush the IO level resources.

---
 drivers/nvme/hw/qedn/qedn.h      |   8 ++
 drivers/nvme/hw/qedn/qedn_conn.c | 135 ++++++++++++++++++++++++++++++-
 drivers/nvme/hw/qedn/qedn_main.c |   6 ++
 drivers/nvme/hw/qedn/qedn_task.c |  30 ++++++-
 4 files changed, 176 insertions(+), 3 deletions(-)
drivers/nvme/hw/qedn/qedn.h +8 -0
index b36994be65cb..5539791fc49b 100644
@@ -41,6 +41,8 @@
 
 #define QEDN_FW_CQ_FP_WQ_WORKQUEUE "qedn_fw_cq_fp_wq"
 
+#define QEDN_DRAIN_MAX_ATTEMPTS 3
+
 /* Protocol defines */
 #define QEDN_MAX_IO_SIZE QED_NVMETCP_MAX_IO_SIZE
 #define QEDN_MAX_PDU_SIZE 0x80000 /* 512KB */
@@ -91,6 +93,8 @@
 /* Timeouts and delay constants */
 #define QEDN_WAIT_CON_ESTABLSH_TMO 10000 /* 10 seconds */
 #define QEDN_RLS_CONS_TMO 5000 /* 5 sec */
+#define QEDN_TASK_CLEANUP_TMO 3000 /* 3 sec */
+#define QEDN_DRAIN_TMO 1000 /* 1 sec */
 
 enum qedn_state {
 	QEDN_STATE_CORE_PROBED = 0,
@@ -173,7 +177,9 @@ struct qedn_ctx {
 };
 
 enum qedn_task_flags {
+	QEDN_TASK_IS_ICREQ,
 	QEDN_TASK_USED_BY_FW,
+	QEDN_TASK_WAIT_FOR_CLEANUP,
 };
 
 struct qedn_task_ctx {
@@ -321,6 +327,8 @@ struct qedn_conn_ctx {
 	struct list_head active_task_list;
 	atomic_t num_active_tasks;
 	atomic_t num_active_fw_tasks;
+	atomic_t task_cleanups_cnt;
+	wait_queue_head_t cleanup_waitq;
 
 	/* Connection resources - turned on to indicate what resource was
 	 * allocated, to that it can later be released.
drivers/nvme/hw/qedn/qedn_conn.c +134 -1

161 lines not on this page · raw message

drivers/nvme/hw/qedn/qedn_main.c +6 -0
index fb47e315ab03..d18cc3624764 100644
@@ -262,12 +262,17 @@ static int qedn_setup_ctrl(struct nvme_tcp_ofld_ctrl *ctrl)
 
 static int qedn_release_ctrl(struct nvme_tcp_ofld_ctrl *ctrl)
 {
+	struct nvme_ctrl *nctrl = &ctrl->nctrl;
 	struct qedn_ctrl *qctrl;
 
 	qctrl = (struct qedn_ctrl *)ctrl->private_data;
 	if (!qctrl)
 		return -ENODEV;
 
+	if (nctrl->state == NVME_CTRL_CONNECTING ||
+	    nctrl->state == NVME_CTRL_RESETTING)
+		return 0;
+
 	if (test_and_clear_bit(LLH_FILTER, &qctrl->agg_state) &&
 	    qctrl->llh_filter) {
 		qedn_dec_llh_filter(qctrl->qedn, qctrl->llh_filter);
@@ -333,6 +338,7 @@ static int qedn_create_queue(struct nvme_tcp_ofld_queue *queue, int qid,
 	qedn_set_pdu_params(conn_ctx);
 
 	init_waitqueue_head(&conn_ctx->conn_waitq);
+	init_waitqueue_head(&conn_ctx->cleanup_waitq);
 	atomic_set(&conn_ctx->est_conn_indicator, 0);
 	atomic_set(&conn_ctx->destroy_conn_indicator, 0);
 
drivers/nvme/hw/qedn/qedn_task.c +28 -2
index dd0b5f31c052..6664cc587f06 100644
@@ -324,6 +324,17 @@ void qedn_return_active_tasks(struct qedn_conn_ctx *conn_ctx)
 	/* Return tasks that aren't "Used by FW" to the pool */
 	list_for_each_entry_safe(qedn_task, task_tmp,
 				 &conn_ctx->active_task_list, entry) {
+		/* If we got this far, cleanup was already done
+		 * in which case we want to return the task to the pool and
+		 * release it. So we make sure the cleanup indication is down
+		 */
+		clear_bit(QEDN_TASK_WAIT_FOR_CLEANUP, &qedn_task->flags);
+
+		/* Special handling in case of ICREQ task */
+		if (unlikely(conn_ctx->state ==	CONN_STATE_WAIT_FOR_IC_COMP &&
+			     test_bit(QEDN_TASK_IS_ICREQ, &(qedn_task)->flags)))
+			qedn_common_clear_fw_sgl(&qedn_task->sgl_task_params);
+
 		qedn_clear_task(conn_ctx, qedn_task);
 		num_returned_tasks++;
 	}
@@ -669,7 +680,9 @@ void qedn_io_work_cq(struct qedn_ctx *qedn, struct nvmetcp_fw_cqe *cqe)
 		return;
 
 	if (likely(cqe->cqe_type == NVMETCP_FW_CQE_TYPE_NORMAL)) {
-		/* Placeholder - verify the connection was established */
+		if (unlikely(test_bit(QEDN_TASK_WAIT_FOR_CLEANUP,
+				      &qedn_task->flags)))
+			return;
 
 		switch (cqe->task_type) {
 		case NVMETCP_TASK_TYPE_HOST_WRITE:
@@ -711,6 +724,19 @@ void qedn_io_work_cq(struct qedn_ctx *qedn, struct nvmetcp_fw_cqe *cqe)
 			pr_info("Could not identify task type\n");
 		}
 	} else {
-		/* Placeholder - Recovery flows */
+		if (cqe->cqe_type == NVMETCP_FW_CQE_TYPE_CLEANUP) {
+			clear_bit(QEDN_TASK_WAIT_FOR_CLEANUP,
+				  &qedn_task->flags);
+			qedn_return_task_to_pool(conn_ctx, qedn_task);
+			atomic_dec(&conn_ctx->task_cleanups_cnt);
+			wake_up_interruptible(&conn_ctx->cleanup_waitq);
+
+			return;
+		}
+
+		 /* The else is NVMETCP_FW_CQE_TYPE_DUMMY - in which don't
+		  * return the task. The task will return during
+		  * NVMETCP_FW_CQE_TYPE_CLEANUP.
+		  */
 	}
 }
Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#32 lore raw

PATCH v4 20/20 qedn: Add support of ASYNC

In reply to Prabhakar Kushwaha

This patch implement ASYNC request and response event notification handling at qedn driver level.

NVME Ofld layer's ASYNC request is treated similar to read with fake CCCID. This CCCID used to route ASYNC notification back to the NVME ofld layer.

---
 drivers/nvme/hw/qedn/qedn.h      |   8 ++
 drivers/nvme/hw/qedn/qedn_main.c |   1 +
 drivers/nvme/hw/qedn/qedn_task.c | 147 +++++++++++++++++++++++++++++--
 3 files changed, 148 insertions(+), 8 deletions(-)
drivers/nvme/hw/qedn/qedn.h +8 -0
index 5539791fc49b..b275d6b9bc59 100644
@@ -96,6 +96,9 @@
 #define QEDN_TASK_CLEANUP_TMO 3000 /* 3 sec */
 #define QEDN_DRAIN_TMO 1000 /* 1 sec */
 
+#define QEDN_MAX_OUTSTAND_ASYNC 32
+#define QEDN_INVALID_CCCID (-1)
+
 enum qedn_state {
 	QEDN_STATE_CORE_PROBED = 0,
 	QEDN_STATE_CORE_OPEN,
@@ -178,6 +181,7 @@ struct qedn_ctx {
 
 enum qedn_task_flags {
 	QEDN_TASK_IS_ICREQ,
+	QEDN_TASK_ASYNC,
 	QEDN_TASK_USED_BY_FW,
 	QEDN_TASK_WAIT_FOR_CLEANUP,
 };
@@ -346,6 +350,10 @@ struct qedn_conn_ctx {
 	struct nvme_tcp_icresp_pdu icresp;
 	struct qedn_icreq_padding *icreq_pad;
 
+	DECLARE_BITMAP(async_cccid_idx_map, QEDN_MAX_OUTSTAND_ASYNC);
+	/* Spinlock for fetching pseudo CCCID for async request */
+	spinlock_t async_cccid_bitmap_lock;
+
 	/* "dummy" socket */
 	struct socket *sock;
 };
drivers/nvme/hw/qedn/qedn_main.c +1 -0
index d18cc3624764..2aaf1af8f9a7 100644
@@ -343,6 +343,7 @@ static int qedn_create_queue(struct nvme_tcp_ofld_queue *queue, int qid,
 	atomic_set(&conn_ctx->destroy_conn_indicator, 0);
 
 	spin_lock_init(&conn_ctx->conn_state_lock);
+	spin_lock_init(&conn_ctx->async_cccid_bitmap_lock);
 
 	conn_ctx->qid = qid;
 
drivers/nvme/hw/qedn/qedn_task.c +139 -8

209 lines not on this page · raw message

Signature · 7 lines
2.24.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#33 lore raw
In reply to Prabhakar Kushwaha

I looked over it a bit (and will send some individual comments), and I have to say I really dislike how this layer and how the hardware works.

The whole point of NVMe is that we have a nicely standardized PCIe register level interface. One that you can trivially hide a TCP offload under with just a little control plane logic. But instead we come up with this gigant mess.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#34 lore raw
Quoted 8 lines · Christoph Hellwig

I looked over it a bit (and will send some individual comments), and I have to say I really dislike how this layer and how the hardware works.

The whole point of NVMe is that we have a nicely standardized PCIe register level interface. One that you can trivially hide a TCP offload under with just a little control plane logic. But instead we come up with this gigant mess.

I can't really see how this control plane logic should work; how would the entire NVMe-oF discovery be abstracted away to hide behind an NVMe-PCI device?

Cheers,

Hannes

Signature · 9 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
maintainer
#35 lore raw
Quoted 4 lines · Hannes Reinecke
Quoted 3 lines · Christoph Hellwig

I looked over it a bit (and will send some individual comments), and I have to say I really dislike how this layer and how the hardware works.

Quoted 8 lines
Quoted 5 lines

The whole point of NVMe is that we have a nicely standardized PCIe register level interface. One that you can trivially hide a TCP offload under with just a little control plane logic. But instead we come up with this gigant mess.

I can't really see how this control plane logic should work; how would the entire NVMe-oF discovery be abstracted away to hide behind an NVMe-PCI device?

Devices are already doing this. The discovery setup is device specific, though.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#36 lore raw
Quoted 10 lines · Keith Busch
Quoted 9 lines · Hannes Reinecke
Quoted 8 lines · Christoph Hellwig

I looked over it a bit (and will send some individual comments), and I have to say I really dislike how this layer and how the hardware works.

The whole point of NVMe is that we have a nicely standardized PCIe register level interface. One that you can trivially hide a TCP offload under with just a little control plane logic. But instead we come up with this gigant mess.

Quoted 7 lines
Quoted 3 lines

I can't really see how this control plane logic should work; how would the entire NVMe-oF discovery be abstracted away to hide behind an NVMe-PCI device?

Devices are already doing this. The discovery setup is device specific, though.

Oh, grand.
I hoped we could steer away from this after the horrible experience we had with iSCSI offload engines. I'd rather have a standardized way for this.

Cheers,

Hannes

Signature · 9 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme