brintos

brintos / linux-shallow public Read only

0
0
Text · 5.3 KiB · d88abfc Raw
167 lines · plain
1====================2Userspace MAD access3====================4 5Device files6============7 8  Each port of each InfiniBand device has a "umad" device and an9  "issm" device attached.  For example, a two-port HCA will have two10  umad devices and two issm devices, while a switch will have one11  device of each type (for switch port 0).12 13Creating MAD agents14===================15 16  A MAD agent can be created by filling in a struct ib_user_mad_reg_req17  and then calling the IB_USER_MAD_REGISTER_AGENT ioctl on a file18  descriptor for the appropriate device file.  If the registration19  request succeeds, a 32-bit id will be returned in the structure.20  For example::21 22	struct ib_user_mad_reg_req req = { /* ... */ };23	ret = ioctl(fd, IB_USER_MAD_REGISTER_AGENT, (char *) &req);24        if (!ret)25		my_agent = req.id;26	else27		perror("agent register");28 29  Agents can be unregistered with the IB_USER_MAD_UNREGISTER_AGENT30  ioctl.  Also, all agents registered through a file descriptor will31  be unregistered when the descriptor is closed.32 33  201434       a new registration ioctl is now provided which allows additional35       fields to be provided during registration.36       Users of this registration call are implicitly setting the use of37       pkey_index (see below).38 39Receiving MADs40==============41 42  MADs are received using read().  The receive side now supports43  RMPP. The buffer passed to read() must be at least one44  struct ib_user_mad + 256 bytes. For example:45 46  If the buffer passed is not large enough to hold the received47  MAD (RMPP), the errno is set to ENOSPC and the length of the48  buffer needed is set in mad.length.49 50  Example for normal MAD (non RMPP) reads::51 52	struct ib_user_mad *mad;53	mad = malloc(sizeof *mad + 256);54	ret = read(fd, mad, sizeof *mad + 256);55	if (ret != sizeof mad + 256) {56		perror("read");57		free(mad);58	}59 60  Example for RMPP reads::61 62	struct ib_user_mad *mad;63	mad = malloc(sizeof *mad + 256);64	ret = read(fd, mad, sizeof *mad + 256);65	if (ret == -ENOSPC)) {66		length = mad.length;67		free(mad);68		mad = malloc(sizeof *mad + length);69		ret = read(fd, mad, sizeof *mad + length);70	}71	if (ret < 0) {72		perror("read");73		free(mad);74	}75 76  In addition to the actual MAD contents, the other struct ib_user_mad77  fields will be filled in with information on the received MAD.  For78  example, the remote LID will be in mad.lid.79 80  If a send times out, a receive will be generated with mad.status set81  to ETIMEDOUT.  Otherwise when a MAD has been successfully received,82  mad.status will be 0.83 84  poll()/select() may be used to wait until a MAD can be read.85 86Sending MADs87============88 89  MADs are sent using write().  The agent ID for sending should be90  filled into the id field of the MAD, the destination LID should be91  filled into the lid field, and so on.  The send side does support92  RMPP so arbitrary length MAD can be sent. For example::93 94	struct ib_user_mad *mad;95 96	mad = malloc(sizeof *mad + mad_length);97 98	/* fill in mad->data */99 100	mad->hdr.id  = my_agent;	/* req.id from agent registration */101	mad->hdr.lid = my_dest;		/* in network byte order... */102	/* etc. */103 104	ret = write(fd, &mad, sizeof *mad + mad_length);105	if (ret != sizeof *mad + mad_length)106		perror("write");107 108Transaction IDs109===============110 111  Users of the umad devices can use the lower 32 bits of the112  transaction ID field (that is, the least significant half of the113  field in network byte order) in MADs being sent to match114  request/response pairs.  The upper 32 bits are reserved for use by115  the kernel and will be overwritten before a MAD is sent.116 117P_Key Index Handling118====================119 120  The old ib_umad interface did not allow setting the P_Key index for121  MADs that are sent and did not provide a way for obtaining the P_Key122  index of received MADs.  A new layout for struct ib_user_mad_hdr123  with a pkey_index member has been defined; however, to preserve binary124  compatibility with older applications, this new layout will not be used125  unless one of IB_USER_MAD_ENABLE_PKEY or IB_USER_MAD_REGISTER_AGENT2 ioctl's126  are called before a file descriptor is used for anything else.127 128  In September 2008, the IB_USER_MAD_ABI_VERSION will be incremented129  to 6, the new layout of struct ib_user_mad_hdr will be used by130  default, and the IB_USER_MAD_ENABLE_PKEY ioctl will be removed.131 132Setting IsSM Capability Bit133===========================134 135  To set the IsSM capability bit for a port, simply open the136  corresponding issm device file.  If the IsSM bit is already set,137  then the open call will block until the bit is cleared (or return138  immediately with errno set to EAGAIN if the O_NONBLOCK flag is139  passed to open()).  The IsSM bit will be cleared when the issm file140  is closed.  No read, write or other operations can be performed on141  the issm file.142 143/dev files144==========145 146  To create the appropriate character device files automatically with147  udev, a rule like::148 149    KERNEL=="umad*", NAME="infiniband/%k"150    KERNEL=="issm*", NAME="infiniband/%k"151 152  can be used.  This will create device nodes named::153 154    /dev/infiniband/umad0155    /dev/infiniband/issm0156 157  for the first port, and so on.  The InfiniBand device and port158  associated with these devices can be determined from the files::159 160    /sys/class/infiniband_mad/umad0/ibdev161    /sys/class/infiniband_mad/umad0/port162 163  and::164 165    /sys/class/infiniband_mad/issm0/ibdev166    /sys/class/infiniband_mad/issm0/port167