xref: /illumos-gate/usr/src/man/man4d/mlxcx.4d (revision c464e8d1c46f2eb4848495bcfccd020b8481787d)
1.\"
2.\" This file and its contents are supplied under the terms of the
3.\" Common Development and Distribution License ("CDDL"), version 1.0.
4.\" You may only use this file in accordance with the terms of version
5.\" 1.0 of the CDDL.
6.\"
7.\" A full copy of the text of the CDDL should have accompanied this
8.\" source.  A copy of the CDDL is also available via the Internet at
9.\" http://www.illumos.org/license/CDDL.
10.\"
11.\"
12.\" Copyright 2020 the University of Queensland
13.\"
14.Dd August 27, 2020
15.Dt MLXCX 4D
16.Os
17.Sh NAME
18.Nm mlxcx
19.Nd Mellanox ConnectX-4/5/6 Ethernet controller driver
20.Sh SYNOPSIS
21.Pa /dev/net/mlxcx*
22.Sh DESCRIPTION
23The
24.Sy mlxcx
25driver is a GLDv3 NIC driver for the ConnectX-4, ConnectX-4 Lx, ConnectX-5 and
26ConnectX-6 families of ethernet controllers from Mellanox.
27It supports the Data Link Provider Interface,
28.Xr dlpi 4P .
29.Pp
30This driver supports:
31.Bl -dash -offset indent
32.It
33Jumbo frames up to 9000 bytes.
34.It
35Checksum offload for TCP, UDP, IPv4 and IPv6.
36.It
37Group support with VLAN and MAC steering to avoid software classification
38when using VNICs.
39.It
40Promiscuous access via
41.Xr snoop 8 and
42.Xr dlpi 4P
43.It
44LED control
45.It
46Transceiver information
47.It
48Internal temperature sensors
49.El
50.Pp
51At this time, the driver does not support Large Send Offload (LSO), energy
52efficient Ethernet (EEE), or the use of flow control through hardware pause
53frames.
54.Sh CONFIGURATION
55The
56.Sy mlxcx.conf
57file contains user configurable parameters, including the ability to set the
58number of rings and groups advertised to MAC, the sizes of rings and groups,
59and the maximum number of MAC address filters available.
60.Sh PROPERTIES
61The driver supports the following device properties which may be tuned through
62its driver.conf file,
63.Pa /kernel/drv/mlxcx.conf .
64These properties cannot be changed after the driver has been attached.
65.Pp
66These properties are not considered stable at this time, and may change.
67.Bl -hang -width Ds
68.It Sy eq_size_shift
69.Bd -filled -compact
70Minimum:
71.Sy 2 |
72Maximum:
73.Sy device dependent (up to 255)
74.Ed
75.Bd -filled
76The
77.Sy eq_size_shift
78property determines the number of entries on Event Queues for the device.
79The number of entries is calculated as
80.Dv (1 << eq_size_shift) ,
81so a value of 9 would mean 512 entries are created on each Event Queue.
82The default value is
83.Sy 9 .
84.Ed
85.It Sy cq_size_shift
86.Bd -filled -compact
87Minimum:
88.Sy 2 |
89Maximum:
90.Sy device dependent (up to 255)
91.Ed
92.Bd -filled
93The
94.Sy cq_size_shift
95property determines the number of entries on Completion Queues for the device.
96The number of entries is calculated as
97.Li (1 << cq_size_shift) ,
98so a value of 9 would mean 512 entries are created on each Event Queue.
99The default value is device dependent,
100.Sy 10
101for devices with maximum supported speed of 10Gb/s or less and
102.Sy 12
103for devices with higher supported speeds.
104This should be kept very close to the value set for
105.Sy rq_size_shift
106and
107.Sy sq_size_shift .
108.Ed
109.It Sy rq_size_shift
110.Bd -filled -compact
111Minimum:
112.Sy 2 |
113Maximum:
114.Sy device dependent (up to 255)
115.Ed
116.Bd -filled
117The
118.Sy rq_size_shift
119property determines the number of descriptors on Receive Queues for the device.
120The number of descriptors is calculated as
121.Dv (1 << rq_size_shift) ,
122so a value of 9 would mean 512 descriptors are created on each Receive Queue.
123This sets the number of packets on RX rings advertised to MAC.
124The default value is device dependent,
125.Sy 10
126for devices with maximum supported speed of 10Gb/s or less and
127.Sy 12
128for devices with higher supported speeds.
129.Ed
130.It Sy sq_size_shift
131.Bd -filled -compact
132Minimum:
133.Sy 2 |
134Maximum:
135.Sy device dependent (up to 255)
136.Ed
137.Bd -filled
138The
139.Sy sq_size_shift
140property determines the number of descriptors on Send Queues for the device.
141The number of descriptors is calculated as
142.Dv (1 << sq_size_shift) ,
143so a value of 9 would mean 512 descriptors are created on each Send Queue.
144This sets the number of packets on RX rings advertised to MAC.
145The default value is device dependent,
146.Sy 11
147for devices with maximum supported speed of 10Gb/s or less and
148.Sy 13
149for devices with higher supported speeds.
150Note that large packets often occupy more than one descriptor slot on the SQ,
151so it is sometimes a good idea to increase this if using a large MTU.
152.Ed
153.It Sy tx_ngroups
154.Bd -filled -compact
155Minimum:
156.Sy 1 |
157Maximum:
158.Sy device dependent
159.Ed
160.Bd -filled
161The
162.Sy tx_ngroups
163property determines the number of TX groups advertised to MAC.
164The default value is
165.Sy 1 .
166.Ed
167.It Sy tx_nrings_per_group
168.Bd -filled -compact
169Minimum:
170.Sy 1 |
171Maximum:
172.Sy device dependent
173.Ed
174.Bd -filled
175The
176.Sy tx_nrings_per_group
177property determines the number of rings in each TX group advertised to MAC.
178The default value is
179.Sy 64 .
180.Ed
181.It Sy rx_ngroups_large
182.Bd -filled -compact
183Minimum:
184.Sy 1 |
185Maximum:
186.Sy device dependent
187.Ed
188.Bd -filled
189The
190.Sy rx_ngroups_large
191property determines the number of "large" RX groups advertised to MAC.
192The size of "large" RX groups is set by the
193.Sy rx_nrings_per_large_group
194property.
195The default value is
196.Sy 2 .
197.Ed
198.It Sy rx_nrings_per_large_group
199.Bd -filled -compact
200Minimum:
201.Sy 1 |
202Maximum:
203.Sy device dependent
204.Ed
205.Bd -filled
206The
207.Sy rx_nrings_per_large_group
208property determines the number of rings in each "large" RX group advertised to
209MAC.
210The number of such groups is determined by the
211.Sy rx_ngroups_large
212property.
213The default value is
214.Sy 16 .
215.Ed
216.It Sy rx_ngroups_small
217.Bd -filled -compact
218Minimum:
219.Sy 1 |
220Maximum:
221.Sy device dependent
222.Ed
223.Bd -filled
224The
225.Sy rx_ngroups_small
226property determines the number of "small" RX groups advertised to MAC.
227The size of "small" RX groups is set by the
228.Sy rx_nrings_per_small_group
229property.
230It is recommended to use many small groups when using a large number of
231VNICs on top of the NIC (e.g. on a system with many zones).
232The default value is
233.Sy 256 .
234.Ed
235.It Sy rx_nrings_per_small_group
236.Bd -filled -compact
237Minimum:
238.Sy 1 |
239Maximum:
240.Sy device dependent
241.Ed
242.Bd -filled
243The
244.Sy rx_nrings_per_small_group
245property determines the number of rings in each "small" RX group advertised to
246MAC.
247The number of such groups is determined by the
248.Sy rx_ngroups_small
249property.
250The default value is
251.Sy 4 .
252.Ed
253.It Sy ftbl_root_size_shift
254.Bd -filled -compact
255Minimum:
256.Sy 4 |
257Maximum:
258.Sy device dependent
259.Ed
260.Bd -filled
261The
262.Sy ftbl_root_size_shift
263property determines the number of flow table entries on the root flow table,
264and therefore how many MAC addresses can be filtered into groups across the
265entire NIC.
266The number of flow entries is calculated as
267.Dv (1 << ftbl_root_size_shift) ,
268so a value of 9 would mean 512 entries are created in the root flow table.
269The default value is
270.Sy 12 .
271.Ed
272.It Sy cqemod_period_usec
273.Bd -filled -compact
274Minimum:
275.Sy 1 |
276Maximum:
277.Sy 65535
278.Ed
279.Bd -filled
280The
281.Sy cqemod_period_usec
282property determines the maximum delay after a completion event has occurred
283before an event queue entry (and thus an interrupt) is generated.
284The delay is measured in microseconds.
285The default value is
286.Sy 50 .
287.Ed
288.It Sy cqemod_count
289.Bd -filled -compact
290Minimum:
291.Sy 1 |
292Maximum:
293.Sy 65535
294.Ed
295.Bd -filled
296The
297.Sy cqemod_count
298property determines the maximum number of completion events that can have
299occurred before an event queue entry (and thus an interrupt) is generated.
300The default value is
301.Sy 80% of the CQ size .
302.Ed
303.It Sy intrmod_period_usec
304.Bd -filled -compact
305Minimum:
306.Sy 1 |
307Maximum:
308.Sy 65535
309.Ed
310.Bd -filled
311The
312.Sy intrmod_period_usec
313property determines the maximum delay after an event queue entry has been
314generated before an interrupt is raised.
315The delay is measured in microseconds.
316The default value is
317.Sy 10 .
318.Ed
319.It Sy tx_bind_threshold
320.Bd -filled -compact
321Minimum:
322.Sy 1 |
323Maximum:
324.Sy 65535
325.Ed
326.Bd -filled
327The
328.Sy tx_bind_threshold
329property determines the minimum number of bytes in a packet before the driver
330uses
331.Xr ddi_dma_addr_bind_handle 9F
332to bind the packet memory for DMA, rather than copying the memory as it does
333for small packets.
334DMA binds are expensive and involve taking locks in the PCI nexus driver, so it
335is seldom worth using them for small packets.
336The default value is
337.Sy 2048 .
338.Ed
339.It Sy rx_limit_per_completion
340.Bd -filled -compact
341Minimum:
342.Sy 16 |
343Maximum:
344.Sy 4096
345.Ed
346.Bd -filled
347The
348.Sy rx_limit_per_completion
349property determines the maximum number of packets that
350will be processed on a given completion ring during a single interrupt.
351This is done to try and guarantee some amount of liveness in the system.
352The default value is
353.Sy 256 .
354.Ed
355.El
356.Sh FILES
357.Bl -tag -width Pa
358.It Pa /kernel/drv/amd64/mlxcx
359Device driver (x86)
360.It Pa /kernel/drv/mlxcx.conf
361Driver configuration file containing user-configurable options
362.El
363.Sh SEE ALSO
364.Xr dlpi 4P ,
365.Xr driver.conf 5 ,
366.Xr dladm 8 ,
367.Xr snoop 8
368