The USB subsystem now has a substantial section in "The Linux Kernel API"
guide (in Documentation/DocBook), generated from the current source
code. This particular documentation file isn't particularly current or
complete; don't rely on it except for a quick overview.
1.1. Basic concept or 'What is an URB?'
The basic idea of the new driver is message passing, the message itself is
called USB Request Block, or URB for short.
- An URB consists of all relevant information to execute any USB transaction
and deliver the data and status back.
- Execution of an URB is inherently an asynchronous operation, i.e. the
usb_submit_urb(urb) call returns immediately after it has successfully
queued the requested action.
- Transfers for one URB can be canceled with usb_unlink_urb(urb) at any time.
- Each URB has a completion handler, which is called after the action
has been successfully completed or canceled. The URB also contains a
context-pointer for passing information to the completion handler.
- Each endpoint for a device logically supports a queue of requests.
You can fill that queue, so that the USB hardware can still transfer
data to an endpoint while your driver handles completion of another.
This maximizes use of USB bandwidth, and supports seamless streaming
of data to (or from) devices when using periodic transfer modes.
1.2. The URB structure
Some of the fields in an URB are:
// (IN) device and pipe specify the endpoint queue
struct usb_device *dev; // pointer to associated USB device
unsigned int pipe; // endpoint information
unsigned int transfer_flags; // ISO_ASAP, SHORT_NOT_OK, etc.
// (IN) all urbs need completion routines
void *context; // context for completion routine
void (*complete)(struct urb *); // pointer to completion routine
// (OUT) status after each completion
int status; // returned status
// (IN) buffer used for data transfers
void *transfer_buffer; // associated data buffer
int transfer_buffer_length; // data buffer length
int number_of_packets; // size of iso_frame_desc
// (OUT) sometimes only part of CTRL/BULK/INTR transfer_buffer is used
int actual_length; // actual data buffer length
// (IN) setup stage for CTRL (pass a struct usb_ctrlrequest)
unsigned char* setup_packet; // setup packet (control only)
// Only for PERIODIC transfers (ISO, INTERRUPT)
// (IN/OUT) start_frame is set unless ISO_ASAP isn't set
int start_frame; // start frame
int interval; // polling interval
// ISO only: packets are only "best effort"; each can have errors
int error_count; // number of errors
struct usb_iso_packet_descriptor iso_frame_desc;
Your driver must create the "pipe" value using values from the appropriate
endpoint descriptor in an interface that it's claimed.
1.3. How to get an URB?
URBs are allocated with the following call
struct urb *usb_alloc_urb(int isoframes, int mem_flags)
Return value is a pointer to the allocated URB, 0 if allocation failed.
The parameter isoframes specifies the number of isochronous transfer frames
you want to schedule. For CTRL/BULK/INT, use 0. The mem_flags parameter
holds standard memory allocation flags, letting you control (among other
things) whether the underlying code may block or not.
To free an URB, use
void usb_free_urb(struct urb *urb)
You may free an urb that you've submitted, but which hasn't yet been
returned to you in a completion callback. It will automatically be
deallocated when it is no longer in use.
1.4. What has to be filled in?
Depending on the type of transaction, there are some inline functions
defined in <linux/usb.h> to simplify the initialization, such as
fill_control_urb() and fill_bulk_urb(). In general, they need the usb
device pointer, the pipe (usual format from usb.h), the transfer buffer,
the desired transfer length, the completion handler, and its context.
Take a look at the some existing drivers to see how they're used.
For ISO there are two startup behaviors: Specified start_frame or ASAP.
For ASAP set URB_ISO_ASAP in transfer_flags.
If short packets should NOT be tolerated, set URB_SHORT_NOT_OK in
1.5. How to submit an URB?
int usb_submit_urb(struct urb *urb, int mem_flags)
The mem_flags parameter, such as SLAB_ATOMIC, controls memory allocation,
such as whether the lower levels may block when memory is tight.
It immediately returns, either with status 0 (request queued) or some
error code, usually caused by the following:
- Out of memory (-ENOMEM)
- Unplugged device (-ENODEV)
- Stalled endpoint (-EPIPE)
- Too many queued ISO transfers (-EAGAIN)
- Too many requested ISO frames (-EFBIG)
- Invalid INT interval (-EINVAL)
- More than one packet for INT (-EINVAL)
After submission, urb->status is -EINPROGRESS; however, you should never
look at that value except in your completion callback.
For isochronous endpoints, your completion handlers should (re)submit
URBs to the same endpoint with the ISO_ASAP flag, using multi-buffering,
to get seamless ISO streaming.
1.6. How to cancel an already running URB?
There are two ways to cancel an URB you've submitted but which hasn't
been returned to your driver yet. For an asynchronous cancel, call
int usb_unlink_urb(struct urb *urb)
It removes the urb from the internal list and frees all allocated
HW descriptors. The status is changed to reflect unlinking. Note
that the URB will not normally have finished when usb_unlink_urb()
returns; you must still wait for the completion handler to be called.
To cancel an URB synchronously, call
void usb_kill_urb(struct urb *urb)
It does everything usb_unlink_urb does, and in addition it waits
until after the URB has been returned and the completion handler
has finished. It also marks the URB as temporarily unusable, so
that if the completion handler or anyone else tries to resubmit it
they will get a -EPERM error. Thus you can be sure that when
usb_kill_urb() returns, the URB is totally idle.
There is a lifetime issue to consider. An URB may complete at any
time, and the completion handler may free the URB. If this happens
while usb_unlink_urb or usb_kill_urb is running, it will cause a
memory-access violation. The driver is responsible for avoiding this,
which often means some sort of lock will be needed to prevent the URB
from being deallocated while it is still in use.
On the other hand, since usb_unlink_urb may end up calling the
completion handler, the handler must not take any lock that is held
when usb_unlink_urb is invoked. The general solution to this problem
is to increment the URB's reference count while holding the lock, then
drop the lock and call usb_unlink_urb or usb_kill_urb, and then
decrement the URB's reference count. You increment the reference
count by calling
struct urb *usb_get_urb(struct urb *urb)
(ignore the return value; it is the same as the argument) and
decrement the reference count by calling usb_free_urb. Of course,
none of this is necessary if there's no danger of the URB being freed
by the completion handler.
1.7. What about the completion handler?
The handler is of the following type:
typedef void (*usb_complete_t)(struct urb *, struct pt_regs *)
I.e., it gets the URB that caused the completion call, plus the
register values at the time of the corresponding interrupt (if any).
In the completion handler, you should have a look at urb->status to
detect any USB errors. Since the context parameter is included in the URB,
you can pass information to the completion handler.
Note that even when an error (or unlink) is reported, data may have been
transferred. That's because USB transfers are packetized; it might take
sixteen packets to transfer your 1KByte buffer, and ten of them might
have transferred successfully before the completion was called.
NOTE: ***** WARNING *****
NEVER SLEEP IN A COMPLETION HANDLER. These are normally called
during hardware interrupt processing. If you can, defer substantial
work to a tasklet (bottom half) to keep system latencies low. You'll
probably need to use spinlocks to protect data structures you manipulate
in completion handlers.
1.8. How to do isochronous (ISO) transfers?
For ISO transfers you have to fill a usb_iso_packet_descriptor structure,
allocated at the end of the URB by usb_alloc_urb(n,mem_flags), for each
packet you want to schedule. You also have to set urb->interval to say
how often to make transfers; it's often one per frame (which is once
every microframe for highspeed devices). The actual interval used will
be a power of two that's no bigger than what you specify.
The usb_submit_urb() call modifies urb->interval to the implemented interval
value that is less than or equal to the requested interval value. If
ISO_ASAP scheduling is used, urb->start_frame is also updated.
For each entry you have to specify the data offset for this frame (base is
transfer_buffer), and the length you want to write/expect to read.
After completion, actual_length contains the actual transferred length and
status contains the resulting status for the ISO transfer for this frame.
It is allowed to specify a varying length from frame to frame (e.g. for
audio synchronisation/adaptive transfer rates). You can also use the length
0 to omit one or more frames (striping).
For scheduling you can choose your own start frame or ISO_ASAP. As explained
earlier, if you always keep at least one URB queued and your completion
keeps (re)submitting a later URB, you'll get smooth ISO streaming (if usb
bandwidth utilization allows).
If you specify your own start frame, make sure it's several frames in advance
of the current frame. You might want this model if you're synchronizing
ISO data with some other event stream.
1.9. How to start interrupt (INT) transfers?
Interrupt transfers, like isochronous transfers, are periodic, and happen
in intervals that are powers of two (1, 2, 4 etc) units. Units are frames
for full and low speed devices, and microframes for high speed ones.
The usb_submit_urb() call modifies urb->interval to the implemented interval
value that is less than or equal to the requested interval value.
In Linux 2.6, unlike earlier versions, interrupt URBs are not automagically
restarted when they complete. They end when the completion handler is
called, just like other URBs. If you want an interrupt URB to be restarted,
your completion handler must resubmit it.