* message.c - synchronous message handling
*/
+#include <linux/pci.h> /* for scatterlist macros */
#include <linux/usb.h>
#include <linux/module.h>
#include <linux/slab.h>
return usb_start_wait_urb(urb,timeout,actual_length);
}
+/*-------------------------------------------------------------------*/
+
+static void sg_clean (struct usb_sg_request *io)
+{
+ if (io->urbs) {
+ while (io->entries--)
+ usb_free_urb (io->urbs [io->entries]);
+ kfree (io->urbs);
+ io->urbs = 0;
+ }
+ usb_buffer_unmap_sg (io->dev, io->pipe, io->sg, io->nents);
+ io->dev = 0;
+}
+
+static void sg_complete (struct urb *urb)
+{
+ struct usb_sg_request *io = (struct usb_sg_request *) urb->context;
+ unsigned long flags;
+
+ spin_lock_irqsave (&io->lock, flags);
+
+ /* In 2.5 we require hcds' endpoint queues not to progress after fault
+ * reports, until the competion callback (this!) returns. That lets
+ * device driver code (like this routine) unlink queued urbs first,
+ * if it needs to, since the HC won't work on them at all. So it's
+ * not possible for page N+1 to overwrite page N, and so on.
+ */
+ if (io->status && urb->actual_length) {
+ err ("driver for bus %s dev %s ep %d-%s corrupted data!",
+ io->dev->bus->bus_name, io->dev->devpath,
+ usb_pipeendpoint (urb->pipe),
+ usb_pipein (urb->pipe) ? "in" : "out");
+ // BUG ();
+ }
+
+ if (urb->status && urb->status != -ECONNRESET) {
+ int i, found, status;
+
+ io->status = urb->status;
+
+ /* the previous urbs, and this one, completed already.
+ * unlink the later ones so they won't rx/tx bad data,
+ *
+ * FIXME don't bother unlinking urbs that haven't yet been
+ * submitted; those non-error cases shouldn't be syslogged
+ */
+ for (i = 0, found = 0; i < io->entries; i++) {
+ if (found) {
+ status = usb_unlink_urb (io->urbs [i]);
+ if (status && status != -EINPROGRESS)
+ err ("sg_complete, unlink --> %d",
+ status);
+ } else if (urb == io->urbs [i])
+ found = 1;
+ }
+ }
+
+ /* on the last completion, signal usb_sg_wait() */
+ io->bytes += urb->actual_length;
+ io->count--;
+ if (!io->count)
+ complete (&io->complete);
+
+ spin_unlock_irqrestore (&io->lock, flags);
+}
+
+
+/**
+ * usb_sg_init - initializes scatterlist-based bulk/interrupt I/O request
+ * @io: request block being initialized. until usb_sg_wait() returns,
+ * treat this as a pointer to an opaque block of memory,
+ * @dev: the usb device that will send or receive the data
+ * @pipe: endpoint "pipe" used to transfer the data
+ * @period: polling rate for interrupt endpoints, in frames or
+ * (for high speed endpoints) microframes; ignored for bulk
+ * @sg: scatterlist entries
+ * @nents: how many entries in the scatterlist
+ * @length: how many bytes to send from the scatterlist, or zero to
+ * send every byte identified in the list.
+ * @mem_flags: SLAB_* flags affecting memory allocations in this call
+ *
+ * Returns zero for success, else a negative errno value. This initializes a
+ * scatter/gather request, allocating resources such as I/O mappings and urb
+ * memory (except maybe memory used by USB controller drivers).
+ *
+ * The request must be issued using usb_sg_wait(), which waits for the I/O to
+ * complete (or to be canceled) and then cleans up all resources allocated by
+ * usb_sg_init().
+ *
+ * The request may be canceled with usb_sg_cancel(), either before or after
+ * usb_sg_wait() is called.
+ *
+ * NOTE:
+ *
+ * At this writing, don't use the interrupt transfer mode, since the old old
+ * "automagic resubmit" mode hasn't yet been removed. It should be removed
+ * by the time 2.5 finalizes.
+ */
+int usb_sg_init (
+ struct usb_sg_request *io,
+ struct usb_device *dev,
+ unsigned pipe,
+ unsigned period,
+ struct scatterlist *sg,
+ int nents,
+ size_t length,
+ int mem_flags
+)
+{
+ int i;
+ int urb_flags;
+
+ if (!io || !dev || !sg
+ || usb_pipecontrol (pipe)
+ || usb_pipeisoc (pipe)
+ || nents <= 0)
+ return -EINVAL;
+
+ spin_lock_init (&io->lock);
+ io->dev = dev;
+ io->pipe = pipe;
+ io->sg = sg;
+ io->nents = nents;
+
+ /* initialize all the urbs we'll use */
+ io->entries = usb_buffer_map_sg (dev, pipe, sg, nents);
+ if (io->entries <= 0)
+ return io->entries;
+
+ io->count = 0;
+ io->urbs = kmalloc (io->entries * sizeof *io->urbs, mem_flags);
+ if (!io->urbs)
+ goto nomem;
+
+ urb_flags = USB_ASYNC_UNLINK | URB_NO_DMA_MAP | URB_NO_INTERRUPT;
+ if (usb_pipein (pipe))
+ urb_flags |= URB_SHORT_NOT_OK;
+
+ for (i = 0; i < io->entries; i++, io->count = i) {
+ unsigned len;
+
+ io->urbs [i] = usb_alloc_urb (0, mem_flags);
+ if (!io->urbs [i]) {
+ io->entries = i;
+ goto nomem;
+ }
+
+ io->urbs [i]->dev = dev;
+ io->urbs [i]->pipe = pipe;
+ io->urbs [i]->interval = period;
+ io->urbs [i]->transfer_flags = urb_flags;
+
+ io->urbs [i]->complete = sg_complete;
+ io->urbs [i]->context = io;
+ io->urbs [i]->status = -EINPROGRESS;
+ io->urbs [i]->actual_length = 0;
+
+ io->urbs [i]->transfer_dma = sg_dma_address (sg + i);
+ len = sg_dma_len (sg + i);
+ if (length) {
+ len = min_t (unsigned, len, length);
+ length -= len;
+ if (length == 0)
+ io->entries = i + 1;
+ }
+ io->urbs [i]->transfer_buffer_length = len;
+ }
+ io->urbs [--i]->transfer_flags &= ~URB_NO_INTERRUPT;
+
+ /* transaction state */
+ io->status = 0;
+ io->bytes = 0;
+ init_completion (&io->complete);
+ return 0;
+
+nomem:
+ sg_clean (io);
+ return -ENOMEM;
+}
+
+
+/**
+ * usb_sg_wait - synchronously execute scatter/gather request
+ * @io: request block handle, as initialized with usb_sg_init().
+ * some fields become accessible when this call returns.
+ * Context: !in_interrupt ()
+ *
+ * This function blocks until the specified I/O operation completes. It
+ * leverages the grouping of the related I/O requests to get good transfer
+ * rates, by queueing the requests. At higher speeds, such queuing can
+ * significantly improve USB throughput.
+ *
+ * There are three kinds of completion for this function.
+ * (1) success, where io->status is zero. The number of io->bytes
+ * transferred is as requested.
+ * (2) error, where io->status is a negative errno value. The number
+ * of io->bytes transferred before the error is usually less
+ * than requested, and can be nonzero.
+ * (3) cancelation, a type of error with status -ECONNRESET that
+ * is initiated by usb_sg_cancel().
+ *
+ * When this function returns, all memory allocated through usb_sg_init() or
+ * this call will have been freed. The request block parameter may still be
+ * passed to usb_sg_cancel(), or it may be freed. It could also be
+ * reinitialized and then reused.
+ *
+ * Data Transfer Rates:
+ *
+ * Bulk transfers are valid for full or high speed endpoints.
+ * The best full speed data rate is 19 packets of 64 bytes each
+ * per frame, or 1216 bytes per millisecond.
+ * The best high speed data rate is 13 packets of 512 bytes each
+ * per microframe, or 52 KBytes per millisecond.
+ *
+ * The reason to use interrupt transfers through this API would most likely
+ * be to reserve high speed bandwidth, where up to 24 KBytes per millisecond
+ * could be transferred. That capability is less useful for low or full
+ * speed interrupt endpoints, which allow at most one packet per millisecond,
+ * of at most 8 or 64 bytes (respectively).
+ */
+void usb_sg_wait (struct usb_sg_request *io)
+{
+ int i;
+ unsigned long flags;
+
+ /* queue the urbs. */
+ spin_lock_irqsave (&io->lock, flags);
+ for (i = 0; i < io->entries && !io->status; i++) {
+ int retval;
+
+ retval = usb_submit_urb (io->urbs [i], SLAB_ATOMIC);
+
+ /* after we submit, let completions or cancelations fire;
+ * we handshake using io->status.
+ */
+ spin_unlock_irqrestore (&io->lock, flags);
+ switch (retval) {
+ /* maybe we retrying will recover */
+ case -ENXIO: // hc didn't queue this one
+ case -EAGAIN:
+ case -ENOMEM:
+ retval = 0;
+ i--;
+ // FIXME: should it usb_sg_cancel() on INTERRUPT?
+ // how about imposing a backoff?
+ set_current_state (TASK_UNINTERRUPTIBLE);
+ schedule ();
+ break;
+
+ /* no error? continue immediately.
+ *
+ * NOTE: to work better with UHCI (4K I/O buffer may
+ * need 3K of TDs) it may be good to limit how many
+ * URBs are queued at once; N milliseconds?
+ */
+ case 0:
+ cpu_relax ();
+ break;
+
+ /* fail any uncompleted urbs */
+ default:
+ io->urbs [i]->status = retval;
+ dbg ("usb_sg_msg, submit --> %d", retval);
+ usb_sg_cancel (io);
+ }
+ spin_lock_irqsave (&io->lock, flags);
+ if (retval && io->status == -ECONNRESET)
+ io->status = retval;
+ }
+ spin_unlock_irqrestore (&io->lock, flags);
+
+ /* OK, yes, this could be packaged as non-blocking.
+ * So could the submit loop above ... but it's easier to
+ * solve neither problem than to solve both!
+ */
+ wait_for_completion (&io->complete);
+
+ sg_clean (io);
+}
+
+/**
+ * usb_sg_cancel - stop scatter/gather i/o issued by usb_sg_wait()
+ * @io: request block, initialized with usb_sg_init()
+ *
+ * This stops a request after it has been started by usb_sg_wait().
+ * It can also prevents one initialized by usb_sg_init() from starting,
+ * so that call just frees resources allocated to the request.
+ */
+void usb_sg_cancel (struct usb_sg_request *io)
+{
+ unsigned long flags;
+
+ spin_lock_irqsave (&io->lock, flags);
+
+ /* shut everything down, if it didn't already */
+ if (!io->status) {
+ int i;
+
+ io->status = -ECONNRESET;
+ for (i = 0; i < io->entries; i++) {
+ int retval;
+
+ if (!io->urbs [i]->dev)
+ continue;
+ retval = usb_unlink_urb (io->urbs [i]);
+ if (retval && retval != -EINPROGRESS)
+ warn ("usb_sg_cancel, unlink --> %d", retval);
+ // FIXME don't warn on "not yet submitted" error
+ }
+ }
+ spin_unlock_irqrestore (&io->lock, flags);
+}
+
+/*-------------------------------------------------------------------*/
+
/**
* usb_get_descriptor - issues a generic GET_DESCRIPTOR request
* @dev: the device whose descriptor is being retrieved
// synchronous request completion model
EXPORT_SYMBOL(usb_control_msg);
EXPORT_SYMBOL(usb_bulk_msg);
+
+EXPORT_SYMBOL(usb_sg_init);
+EXPORT_SYMBOL(usb_sg_cancel);
+EXPORT_SYMBOL(usb_sg_wait);
+
// synchronous control message convenience routines
EXPORT_SYMBOL(usb_get_descriptor);
EXPORT_SYMBOL(usb_get_device_descriptor);
projects / linux / commitdiff