xref: /freebsd/share/man/man4/nvme.4 (revision 732a02b4e77866604a120a275c082bb6221bd2ff)
1.\"
2.\" Copyright (c) 2012-2016 Intel Corporation
3.\" All rights reserved.
4.\"
5.\" Redistribution and use in source and binary forms, with or without
6.\" modification, are permitted provided that the following conditions
7.\" are met:
8.\" 1. Redistributions of source code must retain the above copyright
9.\"    notice, this list of conditions, and the following disclaimer,
10.\"    without modification.
11.\" 2. Redistributions in binary form must reproduce at minimum a disclaimer
12.\"    substantially similar to the "NO WARRANTY" disclaimer below
13.\"    ("Disclaimer") and any redistribution must be conditioned upon
14.\"    including a substantially similar Disclaimer requirement for further
15.\"    binary redistribution.
16.\"
17.\" NO WARRANTY
18.\" THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19.\" "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20.\" LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
21.\" A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
22.\" HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
26.\" STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
27.\" IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28.\" POSSIBILITY OF SUCH DAMAGES.
29.\"
30.\" nvme driver man page.
31.\"
32.\" Author: Jim Harris <jimharris@FreeBSD.org>
33.\"
34.\" $FreeBSD$
35.\"
36.Dd January 6, 2020
37.Dt NVME 4
38.Os
39.Sh NAME
40.Nm nvme
41.Nd NVM Express core driver
42.Sh SYNOPSIS
43To compile this driver into your kernel,
44place the following line in your kernel configuration file:
45.Bd -ragged -offset indent
46.Cd "device nvme"
47.Ed
48.Pp
49Or, to load the driver as a module at boot, place the following line in
50.Xr loader.conf 5 :
51.Bd -literal -offset indent
52nvme_load="YES"
53.Ed
54.Pp
55Most users will also want to enable
56.Xr nvd 4
57to expose NVM Express namespaces as disk devices which can be
58partitioned.
59Note that in NVM Express terms, a namespace is roughly equivalent to a
60SCSI LUN.
61.Sh DESCRIPTION
62The
63.Nm
64driver provides support for NVM Express (NVMe) controllers, such as:
65.Bl -bullet
66.It
67Hardware initialization
68.It
69Per-CPU IO queue pairs
70.It
71API for registering NVMe namespace consumers such as
72.Xr nvd 4
73or
74.Xr nda 4
75.It
76API for submitting NVM commands to namespaces
77.It
78Ioctls for controller and namespace configuration and management
79.El
80.Pp
81The
82.Nm
83driver creates controller device nodes in the format
84.Pa /dev/nvmeX
85and namespace device nodes in
86the format
87.Pa /dev/nvmeXnsY .
88Note that the NVM Express specification starts numbering namespaces at 1,
89not 0, and this driver follows that convention.
90.Sh CONFIGURATION
91By default,
92.Nm
93will create an I/O queue pair for each CPU, provided enough MSI-X vectors
94and NVMe queue pairs can be allocated.
95If not enough vectors or queue
96pairs are available, nvme(4) will use a smaller number of queue pairs and
97assign multiple CPUs per queue pair.
98.Pp
99To force a single I/O queue pair shared by all CPUs, set the following
100tunable value in
101.Xr loader.conf 5 :
102.Bd -literal -offset indent
103hw.nvme.per_cpu_io_queues=0
104.Ed
105.Pp
106To assign more than one CPU per I/O queue pair, thereby reducing the number
107of MSI-X vectors consumed by the device, set the following tunable value in
108.Xr loader.conf 5 :
109.Bd -literal -offset indent
110hw.nvme.min_cpus_per_ioq=X
111.Ed
112.Pp
113To force legacy interrupts for all
114.Nm
115driver instances, set the following tunable value in
116.Xr loader.conf 5 :
117.Bd -literal -offset indent
118hw.nvme.force_intx=1
119.Ed
120.Pp
121Note that use of INTx implies disabling of per-CPU I/O queue pairs.
122.Pp
123To control maximum amount of system RAM in bytes to use as Host Memory
124Buffer for capable devices, set the following tunable:
125.Bd -literal -offset indent
126hw.nvme.hmb_max
127.Ed
128.Pp
129The default value is 5% of physical memory size per device.
130.Pp
131The
132.Xr nvd 4
133driver is used to provide a disk driver to the system by default.
134The
135.Xr nda 4
136driver can also be used instead.
137The
138.Xr nvd 4
139driver performs better with smaller transactions and few TRIM
140commands.
141It sends all commands directly to the drive immediately.
142The
143.Xr nda 4
144driver performs better with larger transactions and also collapses
145TRIM commands giving better performance.
146It can queue commands to the drive; combine
147.Dv BIO_DELETE
148commands into a single trip; and
149use the CAM I/O scheduler to bias one type of operation over another.
150To select the
151.Xr nda 4
152driver, set the following tunable value in
153.Xr loader.conf 5 :
154.Bd -literal -offset indent
155hw.nvme.use_nvd=0
156.Ed
157.Pp
158This value may also be set in the kernel config file with
159.Bd -literal -offset indent
160.Cd options NVME_USE_NVD=0
161.Ed
162.Pp
163When there is an error,
164.Nm
165prints only the most relevant information about the command by default.
166To enable dumping of all information about the command, set the following tunable
167value in
168.Xr loader.conf 5 :
169.Bd -literal -offset indent
170hw.nvme.verbose_cmd_dump=1
171.Ed
172.Pp
173.Sh SYSCTL VARIABLES
174The following controller-level sysctls are currently implemented:
175.Bl -tag -width indent
176.It Va dev.nvme.0.num_cpus_per_ioq
177(R) Number of CPUs associated with each I/O queue pair.
178.It Va dev.nvme.0.int_coal_time
179(R/W) Interrupt coalescing timer period in microseconds.
180Set to 0 to disable.
181.It Va dev.nvme.0.int_coal_threshold
182(R/W) Interrupt coalescing threshold in number of command completions.
183Set to 0 to disable.
184.El
185.Pp
186The following queue pair-level sysctls are currently implemented.
187Admin queue sysctls take the format of dev.nvme.0.adminq and I/O queue sysctls
188take the format of dev.nvme.0.ioq0.
189.Bl -tag -width indent
190.It Va dev.nvme.0.ioq0.num_entries
191(R) Number of entries in this queue pair's command and completion queue.
192.It Va dev.nvme.0.ioq0.num_tr
193(R) Number of nvme_tracker structures currently allocated for this queue pair.
194.It Va dev.nvme.0.ioq0.num_prp_list
195(R) Number of nvme_prp_list structures currently allocated for this queue pair.
196.It Va dev.nvme.0.ioq0.sq_head
197(R) Current location of the submission queue head pointer as observed by
198the driver.
199The head pointer is incremented by the controller as it takes commands off
200of the submission queue.
201.It Va dev.nvme.0.ioq0.sq_tail
202(R) Current location of the submission queue tail pointer as observed by
203the driver.
204The driver increments the tail pointer after writing a command
205into the submission queue to signal that a new command is ready to be
206processed.
207.It Va dev.nvme.0.ioq0.cq_head
208(R) Current location of the completion queue head pointer as observed by
209the driver.
210The driver increments the head pointer after finishing
211with a completion entry that was posted by the controller.
212.It Va dev.nvme.0.ioq0.num_cmds
213(R) Number of commands that have been submitted on this queue pair.
214.It Va dev.nvme.0.ioq0.dump_debug
215(W) Writing 1 to this sysctl will dump the full contents of the submission
216and completion queues to the console.
217.El
218.Pp
219In addition to the typical pci attachment, the
220.Nm
221driver supports attaching to a
222.Xr ahci 4
223device.
224Intel's Rapid Storage Technology (RST) hides the nvme device
225behind the AHCI device due to limitations in Windows.
226However, this effectively hides it from the
227.Fx
228kernel.
229To work around this limitation,
230.Fx
231detects that the AHCI device supports RST and when it is enabled.
232See
233.Xr ahci 4
234for more details.
235.Sh SEE ALSO
236.Xr nda 4 ,
237.Xr nvd 4 ,
238.Xr pci 4 ,
239.Xr nvmecontrol 8 ,
240.Xr disk 9
241.Sh HISTORY
242The
243.Nm
244driver first appeared in
245.Fx 9.2 .
246.Sh AUTHORS
247.An -nosplit
248The
249.Nm
250driver was developed by Intel and originally written by
251.An Jim Harris Aq Mt jimharris@FreeBSD.org ,
252with contributions from
253.An Joe Golio
254at EMC.
255.Pp
256This man page was written by
257.An Jim Harris Aq Mt jimharris@FreeBSD.org .
258