1USERSPACE MAD ACCESS
2
3Device files
4
5  Each port of each InfiniBand device has a "umad" device and an
6  "issm" device attached.  For example, a two-port HCA will have two
7  umad devices and two issm devices, while a switch will have one
8  device of each type (for switch port 0).
9
10Creating MAD agents
11
12  A MAD agent can be created by filling in a struct ib_user_mad_reg_req
13  and then calling the IB_USER_MAD_REGISTER_AGENT ioctl on a file
14  descriptor for the appropriate device file.  If the registration
15  request succeeds, a 32-bit id will be returned in the structure.
16  For example:
17
18	struct ib_user_mad_reg_req req = { /* ... */ };
19	ret = ioctl(fd, IB_USER_MAD_REGISTER_AGENT, (char *) &req);
20        if (!ret)
21		my_agent = req.id;
22	else
23		perror("agent register");
24
25  Agents can be unregistered with the IB_USER_MAD_UNREGISTER_AGENT
26  ioctl.  Also, all agents registered through a file descriptor will
27  be unregistered when the descriptor is closed.
28
29  2014 -- a new registration ioctl is now provided which allows additional
30       fields to be provided during registration.
31       Users of this registration call are implicitly setting the use of
32       pkey_index (see below).
33
34Receiving MADs
35
36  MADs are received using read().  The receive side now supports
37  RMPP. The buffer passed to read() must be at least one
38  struct ib_user_mad + 256 bytes. For example:
39
40  If the buffer passed is not large enough to hold the received
41  MAD (RMPP), the errno is set to ENOSPC and the length of the
42  buffer needed is set in mad.length.
43
44  Example for normal MAD (non RMPP) reads:
45	struct ib_user_mad *mad;
46	mad = malloc(sizeof *mad + 256);
47	ret = read(fd, mad, sizeof *mad + 256);
48	if (ret != sizeof mad + 256) {
49		perror("read");
50		free(mad);
51	}
52
53  Example for RMPP reads:
54	struct ib_user_mad *mad;
55	mad = malloc(sizeof *mad + 256);
56	ret = read(fd, mad, sizeof *mad + 256);
57	if (ret == -ENOSPC)) {
58		length = mad.length;
59		free(mad);
60		mad = malloc(sizeof *mad + length);
61		ret = read(fd, mad, sizeof *mad + length);
62	}
63	if (ret < 0) {
64		perror("read");
65		free(mad);
66	}
67
68  In addition to the actual MAD contents, the other struct ib_user_mad
69  fields will be filled in with information on the received MAD.  For
70  example, the remote LID will be in mad.lid.
71
72  If a send times out, a receive will be generated with mad.status set
73  to ETIMEDOUT.  Otherwise when a MAD has been successfully received,
74  mad.status will be 0.
75
76  poll()/select() may be used to wait until a MAD can be read.
77
78Sending MADs
79
80  MADs are sent using write().  The agent ID for sending should be
81  filled into the id field of the MAD, the destination LID should be
82  filled into the lid field, and so on.  The send side does support
83  RMPP so arbitrary length MAD can be sent. For example:
84
85	struct ib_user_mad *mad;
86
87	mad = malloc(sizeof *mad + mad_length);
88
89	/* fill in mad->data */
90
91	mad->hdr.id  = my_agent;	/* req.id from agent registration */
92	mad->hdr.lid = my_dest;		/* in network byte order... */
93	/* etc. */
94
95	ret = write(fd, &mad, sizeof *mad + mad_length);
96	if (ret != sizeof *mad + mad_length)
97		perror("write");
98
99Transaction IDs
100
101  Users of the umad devices can use the lower 32 bits of the
102  transaction ID field (that is, the least significant half of the
103  field in network byte order) in MADs being sent to match
104  request/response pairs.  The upper 32 bits are reserved for use by
105  the kernel and will be overwritten before a MAD is sent.
106
107P_Key Index Handling
108
109  The old ib_umad interface did not allow setting the P_Key index for
110  MADs that are sent and did not provide a way for obtaining the P_Key
111  index of received MADs.  A new layout for struct ib_user_mad_hdr
112  with a pkey_index member has been defined; however, to preserve binary
113  compatibility with older applications, this new layout will not be used
114  unless one of IB_USER_MAD_ENABLE_PKEY or IB_USER_MAD_REGISTER_AGENT2 ioctl's
115  are called before a file descriptor is used for anything else.
116
117  In September 2008, the IB_USER_MAD_ABI_VERSION will be incremented
118  to 6, the new layout of struct ib_user_mad_hdr will be used by
119  default, and the IB_USER_MAD_ENABLE_PKEY ioctl will be removed.
120
121Setting IsSM Capability Bit
122
123  To set the IsSM capability bit for a port, simply open the
124  corresponding issm device file.  If the IsSM bit is already set,
125  then the open call will block until the bit is cleared (or return
126  immediately with errno set to EAGAIN if the O_NONBLOCK flag is
127  passed to open()).  The IsSM bit will be cleared when the issm file
128  is closed.  No read, write or other operations can be performed on
129  the issm file.
130
131/dev files
132
133  To create the appropriate character device files automatically with
134  udev, a rule like
135
136    KERNEL=="umad*", NAME="infiniband/%k"
137    KERNEL=="issm*", NAME="infiniband/%k"
138
139  can be used.  This will create device nodes named
140
141    /dev/infiniband/umad0
142    /dev/infiniband/issm0
143
144  for the first port, and so on.  The InfiniBand device and port
145  associated with these devices can be determined from the files
146
147    /sys/class/infiniband_mad/umad0/ibdev
148    /sys/class/infiniband_mad/umad0/port
149
150  and
151
152    /sys/class/infiniband_mad/issm0/ibdev
153    /sys/class/infiniband_mad/issm0/port
154