ugen - USB generic driver


Node Name@unit-address

#include <sys/usb/clients/ugen/usb_ugen.h>


ugen is a generic USBA (Solaris USB Architecture) compliant client
character driver that presents USB devices to applications through a
standard open(2), close(2), read(2), write(2), aioread(3C), aiowrite(3C)
Unix interface. Uninterpreted raw data are transferred to and from the
device via file descriptors created for each USB endpoint. Status is
obtained by reading file descriptors created for endpoint and full device

ugen supports control, bulk, isochronous and interrupt (in and out)
transfers. libusb uses ugen to access devices that do not contain drivers
(such as digital cameras and PDAs).


In general, no explicit binding of the ugen driver is necessary because
usb_mid(4D) is the default driver for devices without a class or vendor
unique driver. usb_mid(4D) creates the same logical device names as ugen,
but only if no child interfaces are explicitly bound to ugen. If it is
necessary to bind ugen explicitly to a device or interface, the following
section explains the necessary steps.

ugen can bind to a device with one or more interfaces in its entirety, or
to a single interface of that device. The binding type depends on
information that is passed to add_drv(8) or update_drv(8).

An add_drv(8) command binds ugen to a list of device types it is to
control. update_drv(8) adds an additional device type to the list of
device types being managed by the driver.

Names used to bind drivers can be found in /var/adm/messages. When a
device is on-lined after hot insertion, and no driver is found, there
will be an entry containing:

USB 2.0 device (usb<vid>,<pid>)...

where vid is the USB vendor identifier in hex and pid is the product
identifier in hex supplied by the device descriptor usb_dev_descr(9S).
Note, the USB device version may vary depending on the device.

When using ugen for the first time, you must add the driver utilizing
add_drv(8), using a command of the following form:

Assuming that the vid is 472 and pid is b0b0:

add_drv -n -m '* <device perms> <owner> <group>'
-i '"usb472,b0b0"' ugen

If the command fails with:

(ugen) already in use as a driver or alias.

...add the device using update_drv(8):

update_drv -a -m '* <device perms> <owner> <group>'
-i '"usb472,b0b0"' ugen

This binds ugen to the entire device.

If ugen only binds to one interface of the device, use the following
driver_alias instead of usb<vid>,<pid>:

usbif<vid>,<pid>.config<cfg value>.<interface number>

where cfg value is the value of bConfigurationValue in the configuration
descriptor (usb_cfg_descr(9S)). For example "usbif1234,4567.config1.0."

Note that you can use update_drv to also remove bindings. Please see
update_drv(8) for more information.

After a successful add_drv or update_drv, remove the device and reinsert.
Check with the prtconf(8) -D option to determine if ugen is successfully
bound to the device and the nodes created in /dev/usb/<vid>.<pid> (see

An example showing how to bind a child device representing interface 0 of
configuration 1 of a composite device follows:

update_drv -a -m '* 0666 root sys'
-i '"usbif472,b0b0.config1.0"' ugen

Note that if you uninstall the ugen driver, and later reinstall it, any
pre-existing ugen driver device-bindings will be reactivated. Likewise,
any pre-existing ugen driver device-bindings are preserved across
operating system updates or upgrades.


For each device or child device it manages, ugen creates one logical
device name for device-wide status and one logical device name for
endpoint 0. ugen also creates logical device names for all other
endpoints within the device node's binding scope (interface or device),
plus logical device names for their status.

If separate ugen instances control different interfaces of the same
device, the device-wide status and endpoint logical device names created
for each instance will share access to the same source or endpoint pipes.
For example, a device with two interfaces, each operated by their own
ugen instance, will show endpoint0 as if0cntrl0 to the first interface,
and will show it as if1cntrl0 to the second interface. Both of these
logical device names share endpoint0. Likewise for the same device, ugen
makes the device-wide status available as if0devstat to the first
interface and as if1devstat to the second interface. if0devstat and
if1devstat both return the same data.

Any ugen logical device name can be held open by only one user at a time,
regardless of whether the O_EXCL flag passed to open(2). When a single
pipe or data source is shared by multiple logical device names, such as
if[0,1]cntrl0 or if[0,1]devstat above, more than one logical device name
sharing the pipe or data source can be open at a time. However, only one
user may access the shared pipe or data source at a time, regardless of
the logical device name used for access.

When ugen is bound to an entire device, the following logical device
names are created (each on a single line). N represents the instance
number of the device type.

Endpoint 0 (default endpoint):


For example:


Configuration index 1, Endpoints > 0, alternate 0:


For example:


Configuration index 1, Endpoints > 0, alternate > 0:


For example:


Configuration index> 1, Endpoints > 0, alternate 0:

For example:


Note that the configuration value from the configuration
descriptor indexed by the configuration index is used in
the node name and not the configuration index itself.

Configuration index> 1, Endpoints > 0, alternate > 0:

For example:


Device status:


For example:


When ugen is bound to a single device interface, the following logical
device nodes are created:

Endpoint 0 (default endpoint):


For example:


Device status:

For example:

The format for all other logical device names is identical to the format
used when ugen is bound to the entire device.

Opening the endpoint of a different configuration or different alternate
interface will cause an implicit change of configuration or a switch to
an alternate interface. A configuration change is prohibited when any
non-zero endpoint device nodes are open. An alternate interface switch is
prohibited if any endpoint in the same interface is open.

A device may be hot-removed at any time. Following hot-removal, the
device status changes to USB_DEV_STAT_DISCONNECTED, the status of open
endpoints change to USB_LC_STAT_DISCONNECTED upon their access, and all
subsequent transfer requests fail. Endpoints are reactivated by first
reinserting the device and then closing and reopening all endpoints that
were open when the device was disconnected.

CPR (Checkpoint/Resume) may be initiated at any time and is treated
similarly to a hot-removal. Upon successful suspend and resume, all
subsequent transfer requests fail as an indication to the application to
reinitialize. Applications should close and reopen all endpoints to
reinstate them. All endpoint and device status on Resume (before close
and reopen) is USB_LC_STAT_SUSPENDED. A system suspend will fail while
ugen is performing a transfer.


Devices which support remote wakeup can be power managed when they have
no open logical device nodes. When an application opens the first logical
device node of a device, that application should assume that a
reinitialization of device state is required.


Applications can monitor device status changes by reading the device
status from the device status logical name. When opened without
O_NONBLOCK and O_NDELAY, all reads from that file descriptor (with the
exception of the initial read that follows the open) block until a device
status change occurs. Calls to read will always return immediately if
opened with O_NONBLOCK or O_NDELAY. Nonblocking calls to read which have
no data to return, return no error and zero bytes read.

Device statuses are:

Device is available.

Device has been disconnected.

Device has been resumed, however, endpoints
which were open on suspend have not yet been
closed and reopened.

Device has been reconnected, however,
endpoints which were open on disconnect have
not yet been closed and reopened.

The following code reads the device status device logical name:

int fd;
int status;

if ((fd = open("/dev/usb/472.b0b0/0/devstat",
O_RDONLY)) < 0) {
/* handle error */

if (read(fd, &status, sizeof(status)) != sizeof(status)) {
/* handle error */

switch (status) {
printf ("Terminating as device has been disconnected.\n");
exit (0);

* Close and reopen endpoints to reestablish device access,
* then reset device.


Use poll(2) to block on several logical names simultaneously, including
device status logical names. Poll indicates when reading a logical name
would return data. See poll(2) for details. Calls to read may be done
whether or not they follow calls to poll.


Each data endpoint has a corresponding status logical name. Use the
status logical name to retrieve the state of the data endpoint, including
detail on how its most recent transfer failed. Reads of the status file
descriptors always return immediately. See the ERRORS section for more
information on endpoint status values. All logical device name files
created for returning status must be opened with O_RDONLY.

The following code illustrates reading the status file descriptor of an
endpoint which just failed a data transfer in order to get more
information on the failure.

int data_xfered, status;
int ep1_data_fd, ep1_stat_fd;
uchar_t request[8];

ep1_data_fd = open ("/dev/usb/472.b0b0/0/if0out1", O_WRONLY);

if (ep1_data_fd < 0) {
/* Handle open error. */

ep1_stat_fd = open ("/dev/usb/472.b0b0/0/if0out1stat",
if (ep1_stat_fd < 0) {
/* Handle open error. */

data_xfered = write(ep1_data_fd, request, sizeof (request));

/* An error occurred during the data transfer. */
if (data_xfered != sizeof (request)) {

/* Read status file descriptor for details on failure. */
if (read(ep1_stat_fd, (int *)&status, sizeof (status)) !=
sizeof (status)) {

/* Take appropriate action. */
switch (status) {
printf ("Endpoint stalled.\n");
case ...



The control endpoint is typically used to set up the device and to query
device status or configuration.

Applications requiring I/O on a control endpoint should open the
corresponding logical device name and use regular UNIX I/O system calls.
For example: read(2), write(2), aioread(3C) and aiowrite(3C). poll(2) is
not supported on control endpoints.

A control endpoint must be opened with O_RDWR since it is bidirectional.
It cannot be opened with O_NONBLOCK or O_NDELAY.

For example:

fd = open("/dev/usb/472.b0b0/0/cntrl0", O_RDWR);

fdstat = open("/dev/usb/472.b0b0/0/cntrl0stat", O_RDONLY);

Control endpoints can be read and written. A read operation receives data
from the device and a write operation sends data to the device.

To perform a control-IN transfer, perform a write(2) of USB setup data
(see section 9.3 of the USB 1.1 or 2.0 specifications) followed by a
read(2) on the same control endpoint to fetch the desired data. For

void init_cntrl_req(
uchar_t *req, uchar_t bmRequestType, uchar_t bRequest,
ushort_t wValue, ushort_t wIndex, ushort_t wLength) {
req[0] = bmRequestType;
req[1] = bRequest;
req[2] = 0xFF & wValue;
req[3] = 0xFF & (wValue >> 8);
req[4] = 0xFF & wIndex;
req[5] = 0xFF & (wIndex >> 8);
req[6] = 0xFF & wLength;
req[7] = 0xFF & (wLength >> 8);


uchar_t dev_descr_req[8];
usb_dev_descr_t descr;

USB_DESCR_TYPE_SETUP_DEV, 0, sizeof (descr));

count = write(fd, dev_descr_req, sizeof (dev_descr_req));
if (count != sizeof (dev_descr_req)) {
/* do some error recovery */

count = read(fd, &descr, sizeof (descr));
if (count != sizeof (descr)) {
/* do some error recovery */

The application can issue any number of reads to read data received on a
control endpoint. ugen successfully completes all reads, returning the
number of bytes transferred. Zero is returned when there is no data to

If the read/write fails and returns -1, you can access the endpoint's
status device logical name for precise error information:

int status;

count = read(fdstat, &status, sizeof (status));
if (count == sizeof (status)) {
switch (status) {
/* close all endpoints */

Refer to the ERRORS section for all possible error values.

To perform a control-OUT transfer, send in a single transfer, the USB
setup data followed by any accompanying data bytes.

/* 1st 8 bytes of wbuf are setup. */
init_cntrl_req(wbuf, .......);

/* Data bytes begin at byte 8 of wbuf. */
bcopy(data, &wuf[8], sizeof (data));

/* Send it all in a single transfer. */
count = write(fd, wbuf, sizeof (wbuf));

A write(2) returns the number of bytes (both setup and data) actually
transferred, (whether or not the write is completely successful),
provided that some data is actually transferred. When no data is
transferred, write(2) returns -1. Applications can read the corresponding
endpoint status to retrieve detailed error information. Note that it is
an error to specify a size different than:

(number of data bytes + number of setup bytes).

Here is a more extensive example which gets all descriptors of a device
configuration. For sake of brevity, uninteresting parts are omitted.

#include <sys/usb/usba.h>
#include <sys/usb/clients/ugen/usb_ugen.h>

uchar_t *config_cloud;
uchar_t *curr_descr;

uchar_t *bytes;

int curr_descr_len;
int curr_descr_type;

usb_cfg_descr_t cfg_descr;
usb_if_descr_t if_descr;
usb_ep_descr_t ep_descr;

/* See 9.13 of USB 2.0 spec for ordering. */
static char *pipetypes[] = {
"Control", "Isochronous", "Bulk", "Interrupt"

* Setup to send a request to read just the config descriptor. The
* size of the whole cloud, containing all cfg, interface, endpoint,
* class and vendor-specific descriptors, will be returned as part of
* the config descriptor.
init_cntrl_req(&setup_data, USB_DEV_REQ_DEV_TO_HOST, USB_REQ_GET_DESCR,

* Write setup data. USB device will prepare to return the whole
* config cloud as a response to this. We will read this separately.
count = write(ctrl_fd, &setup_data, sizeof (setup_data));
if (count != sizeof (setup_data)) {
/* Error recovery. */
} else {
count = read(ctrl_fd, &cfg_descr, USB_CFG_DESCR_SIZE);
if (count != USB_CFG_DESCR_SIZE) {
/* Error recovery. */

/* USB data is little endian. */
bytes = (uchar_t *)(&cfg_descr.wTotalLength);
totalLength = bytes[0] + (bytes[1] << 8);

* The size of the whole cloud is in the bLength field. Set up
* to read this amount of data, to get the whole cloud.
config_cloud = malloc(totalLength);

init_cntrl_req(&setup_data, USB_DEV_REQ_DEV_TO_HOST, USB_REQ_GET_DESCR,
USB_DESCR_TYPE_SETUP_CFG, 0, totalLength);

count = write(ctrl_fd, &setup_data, sizeof (setup_data));
if (count != sizeof (setup_data)) {
/* Error recovery. */
} else {
count = read(ctrl_fd, config_cloud, totalLength);
if (count != totalLength) {
/* Error recovery. */

/* Got the data. Now loop, dumping out the descriptors found. */

curr_descr = config_cloud;
offset = 0;
while (offset < totalLength) {

/* All descr have length and type at offset 0 and 1 */
curr_descr_len = curr_descr[0];
curr_descr_type = curr_descr[1];

switch (curr_descr_type) {

* Copy data into separate structure, needed for
* proper alignment of all non char fields. Note:
* non-char fields of all descriptors begin on aligned
* boundaries. The issue is that some structures may
* be adjacent to others which have an odd-numbered
* byte size, and may thus start on an odd-numbered
* boundary. */
bcopy(curr_descr, &cfg_descr, curr_descr_len);

/* Remember to read any words in endian-neutral way. */

(void) printf("\nConfig %d found.\n",

bcopy(curr_descr, &if_descr, curr_descr_len);
(void) printf("\n\tInterface %d, Alt %d found.\n",

bcopy(curr_descr, &ep_descr, curr_descr_len);
(void) printf("\n\t\tEndpoint %d (%s-%s) found.\n",
(ep_descr.bEndpointAddress & USB_EP_NUM_MASK),
ep_descr.bmAttributes & USB_EP_ATTR_MASK]),
((ep_descr.bEndpointAddress &
USB_EP_DIR_IN) ? "IN" : "OUT"));

(void) printf(
"\n\t\t\tOther descriptor found. Type:%d\n",

offset += curr_descr_len;
curr_descr = &config_cloud[offset];

Applications requiring data from an interrupt-IN endpoint should open the
corresponding logical device name and use read(2), aioread(3C) and
poll(2) system calls.

An interrupt-IN endpoint must be opened with O_RDONLY. It can also be
opened using O_NONBLOCK or O_NDELAY if desired.

fd = open("/dev/usb/472.b0b0/0/if0in1", O_RDONLY);

fdstat = open("/dev/usb/472.b0b0/0/if0in1stat", O_RDONLY);

ugen starts polling interrupt--IN endpoints immediately upon opening them
and stops polling them upon closure. (Polling refers to interrogation of
the device by the driver and should not be confused with poll(2), which
is an interrogation of the driver by the application.)

A read(2) of an endpoint opened with the O_NONBLOCK or O_NDELAY flags set
will not block when there is insufficient data available to satisfy the
request. The read simply returns what it can without signifying any

Applications should continuously check for and consume interrupt data.
ugen enables buffering of up to one second of incoming data. In case of
buffer overflow, ugen stops polling the interrupt-IN endpoint until the
application consumes all the data. In this case, a read(2) of an empty
buffer returns -1, sets the endpoint status to USB_LC_STAT_INTR_BUF_FULL
(to indicate that the buffer had been full and polling had been stopped)
and causes ugen to start polling the endpoint again. To retrieve the
status, the application can open and read the corresponding endpoint's
status device logical name.

for (;;) {
count = read(fd, buf, sizeof(buf));
if (count == -1) {
int cnt, status;

cnt = read(fdstat, &status, sizeof (status));
if (cnt == -1) {
/* more error recovery here */
} else {
switch (status) {
/* process the data */

ugen will never drop data. However, the device may drop data if the
application cannot read it at the rate that it is produced.

Applications requiring unbuffered data from an interrupt-IN endpoint
should open the associated status endpoint with O_RDWR before opening the
associated interrupt-IN endpoint and write a control byte with
USB_EP_INTR_ONE_XFER set. All other bits are reserved and should be 0.

"One transfer" mode will persist until disabled explicitly after the
associated interrupt-IN endpoint has been closed by writing a control
byte with USB_EP_INTR_ONE_XFER cleared.

"One transfer" mode is implicitly disabled when the status/control
endpoint is closed.

Attempts to change the "one transfer" mode while the endpoint is open
will result in EINVAL.

An application can open multiple interrupt-IN endpoints and can call
poll(2) to monitor the availability of new data. (Note: poll works with
interrupt-IN data endpoints, not their status endpoints.)

struct pollfd pfd[2];

bzero(pfd, sizeof (pfd));
pfd[0].fd = fd1; /* fd1 is one interrupt-IN endpoint. */
pfd[0].events = POLLIN;
pfd[1].fd = fd2; /* fd2 is another interrupt-IN endpoint. */
pfd[1].events = POLLIN;

for (;;) {
poll(pfd, 2, -1);

if (pfd[0].revents & POLLIN) {
count = read(fd1, buf, sizeof (buf));
if (pfd[1].revents & POLLIN) {
count = read(fd2, buf, sizeof (buf));

You can monitor the device status endpoint via poll(2) concurrently with
the multiple interrupt-IN endpoints. Simply add another pollfd element
to the pfd array in the previous code example, and initialize the new
element's fd field with the file descriptor of the device status endpoint
(opened without O_NONBLOCK or O_NDELAY). Set the new element's event
field to POLLIN like the other elements. Note that only interrupt-IN
endpoints and the device status endpoint can be monitored using poll(2).

Applications requiring output on an interrupt-OUT endpoint can open the
corresponding logical device name and perform regular UNIX I/O system
calls such as write(2) and aiowrite(3C).

An interrupt-OUT endpoint must be opened with O_WRONLY.

fd = open("/dev/usb/472.b0b0/0/if0out3", O_WRONLY);

fdstat = open("/dev/usb/472.b0b0/0/if0out3stat", O_RDONLY);

Data can be written to an interrupt-OUT endpoint as follows:

count = write(fd, buf, sizeof (buf)):
if (count == -1) {
/* error recovery */


Applications requiring I/O on a bulk endpoint can open the corresponding
logical device name and perform regular UNIX I/O system calls. For
example: read(2), write(2), aioread(3C) and aiowrite(3C). poll(2) is not
supported on bulk endpoints.

A bulk endpoint must be opened with O_RDONLY or O_WRONLY and cannot be
opened with O_NONBLOCK or O_NDELAY:

fd = open("/dev/usb/472.b0b0/0/if0in2", O_RDONLY);

fdstat = open("/dev/usb/472.b0b0/0/if0in2stat", O_RDONLY);

Data can be read from a bulk-IN endpoint as follows:

count = read(fd, buf, sizeof (buf)):
if (count == -1) {
/* error recovery */

Data can be written to a bulk-OUT endpoint as follows:

count = write(fd, buf, sizeof (buf)):
if (count == -1) {
/* error recovery */


Applications requiring I/O on an isochronous endpoint can open the
corresponding logical device name and perform regular UNIX I/O system
calls such as read(2), write(2), poll(2), aioread(3C) and aiowrite(3C).
An isochronous endpoint must be opened with O_RDWR.

fd = open("/dev/usb/472.b0b0/0/if0.3in2", O_RDWR);

fdstat = open("/dev/usb/472.b0b0/0/if0.3in2stat", O_RDONLY);

Applications can use the status logical name to retrieve the state of the
isochronous data endpoint, including details on why the most recent
transfer failed.

Applications have the flexibility to specify the number of isochronous
packets and the size of individual packets they want to transfer.
Applications should use the following data structures to exchange
isochronous packet information with the ugen driver:

typedef struct ugen_isoc_pkt_descr {
* Set by the application, for all isochro.
* requests, to the num. of bytes to xfer
* in a packet.
ushort_t dsc_isoc_pkt_len;

* Set by ugen to actual num. of bytes sent/received
* in a packet.
ushort_t dsc_isoc_pkt_actual_len;

* Per pkt. status set by ugen driver both for the
* isochronous IN and OUT requests. Application can
* use USB_LC_STAT_* to parse the status.
int dsc_isoc_pkt_status;
} ugen_isoc_pkt_descr_t;

typedef struct ugen_isoc_req_head {
/* pkt count of the isoc request */
int req_isoc_pkts_count;

/* pkt descriptors */
ugen_isoc_pkt_descr_t req_isoc_pkt_descrs[1];
} ugen_isoc_req_head_t;

req_isoc_pkts_count is limited by the capability of the USB host
controller driver. The current upper bound for the uhci and ohci drivers
is 512. The upper bound for the ehci driver is 1024.

For an isochronous-IN endpoint, applications must first use the
ugen_isoc_req_head_t structure followed by ugen_isoc_pkt_descr_t to write
packet request information to the ugen node. The ugen driver then checks
the validity of the request. If it is valid, the driver immediately
begins isochronous polling on the IN endpoint and applications can
proceed with read(2) of the data on the isochronous-IN endpoint. Upon
successful return of read(2), isochronous packet descriptors (whose
dsc_isoc_pkt_actual_len and dsc_isoc_pkt_status fields were filled by the
driver) are returned, followed by the request's device payload data.

Applications should continuously check for and consume isochronous data.
The ugen driver enables buffering of up to eight seconds of incoming data
for full-speed isochronous endpoint, one second of data for high-speed
isochronous endpoints who request one transaction per microframe and 1/3
of a second of incoming data for high-speed high-bandwidth isochronous
endpoints who request three transactions per microframe. In case of
buffer overflow, ugen discards the oldest data.

The isochronous-IN polling can only be stopped by a close(2) associated
file descriptor. If applications want to change packet information, they
must first close(2) the endpoint to stop the isochronous-IN polling, then
open(2) the endpoint and write(2) new packets request.

The following example shows how to read an isochronous-IN endpoint:

#include <sys/usb/clients/ugen/usb_ugen.h>

char *buf, *p;
ushort_t pktlen;
int pktcnt, i;
int len;
ugen_isoc_req_head_t *req;
ugen_isoc_pkt_descr_t *pktdesc;
char rdbuf[5000];

pktcnt = 4; /* 4 packets in this request */

len = sizeof(int) +
sizeof(ugen_isoc_pkt_descr_t) * pktcount;

buf = malloc(len);
if (!buf) {
/* Error recovery. */

req = (ugen_isoc_req_head_t *)buf;
req->req_isoc_pkts_count = pktcnt;

pktdesc = (ugen_isoc_pkt_descr_t *)

for (i = 0; i < pktcnt; i++) {
* pktlen should not exceed xfer
* capability of an endpoint
pktdesc[i].dsc_isoc_pkt_len = pktlen;

pktdesc[i].dsc_isoc_pkt_actual_len = 0;
pktdesc[i].dsc_isoc_pkt_status = 0;

* write request info to driver and len must
* be exactly the sum of
* sizeof(int) + sizeof(ugen_isoc_pkt_descr_t) * pktcnt.
* Otherwise, an error is returned.
if (write(fd, buf, len) < 0) {
/* Error recovery. */

* Read length should be sum of all pkt descriptors
* length + payload data length of all pkts
* (sizeof(ugen_isoc_pkt_descr_t) + pktlen) * pktcnt
if (read(fd, rdbuf, (sizeof(ugen_isoc_pkt_descr_t) +
pktlen) * pktcnt) < 0) {
/* Error recovery. */

pktdesc = (ugen_isoc_pkt_descr_t *) rdbuf;

/* points to payload beginning */
p = rdbuf + pktcnt * sizeof(ugen_isoc_pkt_descr_t);

for (i = 0; i < pktcnt; i++) {
printf("packet %d len = %d,"
" actual_len = %d, status = 0x%x\n",
i, pktdesc->dsc_isoc_pkt_len,

/* Processing data */

* next packet data payload, do NOT use
* dsc_isoc_pkt_actual_len
p += pktdesc->dsc_isoc_pkt_len;


For an isochronous-OUT endpoint, applications use the same packet
descriptor and request structures to write request information to the
ugen node. Following the packet request head information is the packet
payload data. Upon successful return of write(2), applications can
read(2) the same ugen file immediately to retrieve the individual packet
transfer status of the last request. If the application isn't concerned
about the status, it can omit it.

In the following example, an application transfers data on an
isochronous-OUT endpoint:

#include <sys/usb/clients/ugen/usb_ugen.h>
char *buf, *p;
ushort_t i, pktlen;
int len, pktcnt;
ugen_isoc_req_head_t *req;
ugen_isoc_pkt_descr_t *pktdesc;
char rdbuf[4096];

pktcnt = 4;

* set packet length to a proper value, don't
* exceed endpoint's capability
pktlen = 1024;

len = sizeof(int) +
sizeof(ugen_isoc_pkt_descr_t) * pktcount;

len += pktlen * pktcnt;

buf = malloc(len);
if (!buf) {
/* Error recovery. */

req = (ugen_isoc_req_head_t *)buf;
req->req_isoc_pkts_count = pktcnt;

pktdesc =
(ugen_isoc_pkt_descr_t *)(req->req_isoc_pkt_descrs);

for (i = 0; i < pktcnt; i++) {
pktdesc[i].dsc_isoc_pkt_len = pktlen;
pktdesc[i].dsc_isoc_pkt_actual_len = 0;
pktdesc[i].dsc_isoc_pkt_status = 0;

/* moving to beginning of payload data */
p = buf + sizeof(int) + sizeof(*pktdesc) * pktcnt;
for (i = 0; i < pktcnt; i++) {

/* fill in the data buffer */

p += pktlen;

* write packet request information and data to ugen driver
* len should be the exact value of sizeof(int) +
* sizeof(ugen_isoc_pkt_descr_t) * pktcnt + payload length
if (write(fd, buf, len) < 0) {
/* Error recovery. */

/* read packet status */
if (read(fd, rdbuf, sizeof(*pktdesc) * pktcnt) < 0) {

/* Error recovery. */

} else {

/* Parse every packet's transfer status */



The following statuses are returned by endpoint status device logical


No error.


CRC error detected.


Bit stuffing error.


Data toggle did not match.


Endpoint returned stall.


Device not responding.


Unexpected Packet Identifier (PID).


Check bits on PID failed.


Data overrun.


Data underrun.


Buffer overrun.


Buffer underrun.


Command timed out.


Not accessed by the hardware.


Unspecified USBA or HCD error.


No bandwidth available.


Host Controller h/w error.


Device was suspended.


Device was disconnected.


Polling was stopped as the interrupt-IN data buffer was full. Buffer
is now empty and polling has been resumed.


Request was interrupted.


No resources available for request.


Failed to restart polling.


Failed to start isochronous polling.


Isochronous packet information not initialized.


All packets in this isochronous request have errors. The polling on
this isochronous-IN endpoint is suspended and can be resumed on next

The following system call errno values are returned:

An attempt was made to enable or disable "one transfer" mode
while the associated endpoint was open.

The endpoint has been opened and another open is attempted.

An endpoint open was attempted with incorrect flags.

Operation not supported.

Device associated with the file descriptor does not exist.

Device has been hot-removed or a suspend/resume happened
before this command.

An I/O error occurred. Send a read on the endpoint status
minor node to get the exact error information.

Interrupted system call.

No memory for the allocation of internal structures.



32-bit ELF kernel module. (x86).


64-bit ELF kernel module. (x86).


64-bit ELF kernel module. (SPARC).

/dev/usb/<vid>.<pid>/<N>/cntrl0 /dev/usb/<vid>.<pid>/<N>/cntrl0stat







where N is an integer representing the instance number of this type
of device. (All logical device names for a single device share the
same N.)


See attributes(7) for descriptions of the following attributes:

|Architecture | SPARC & x86, PCI-based systems |


close(2), poll(2), read(2), write(2), aioread(3C), aiowrite(3C),
usba(4D), usb_dev_descr(9S)


In addition to being logged, the following messages may appear on the
system console. All messages are formatted in the following manner:

Warning: <device path> (ugen<instance num>): Error Message...

Too many minor nodes.

Device has too many minor nodes. Not all are available.

Instance number too high (<number>).

Too many devices are using this driver.

Cannot access <device>. Please reconnect.

This device has been disconnected because a device other than the
original one has been inserted. The driver informs you of this fact
by displaying the name of the original device.

Device is not identical to the previous one on this port. Please
disconnect and reconnect.

Same condition as described above; however in this case, the driver
is unable to identify the original device with a name string.


ugen returns -1 for all commands and sets errno to ENODEV when device has
been hot-removed or resumed from a suspend. The application must close
and reopen all open minor nodes to reinstate successful communication.

May 13, 2017 UGEN(4D)