linux/drivers/usb/host/uhci-q.c

1542 lines
40 KiB
C
Raw Normal View History

/*
* Universal Host Controller Interface driver for USB.
*
* Maintainer: Alan Stern <stern@rowland.harvard.edu>
*
* (C) Copyright 1999 Linus Torvalds
* (C) Copyright 1999-2002 Johannes Erdfelt, johannes@erdfelt.com
* (C) Copyright 1999 Randy Dunlap
* (C) Copyright 1999 Georg Acher, acher@in.tum.de
* (C) Copyright 1999 Deti Fliegl, deti@fliegl.de
* (C) Copyright 1999 Thomas Sailer, sailer@ife.ee.ethz.ch
* (C) Copyright 1999 Roman Weissgaerber, weissg@vienna.at
* (C) Copyright 2000 Yggdrasil Computing, Inc. (port of new PCI interface
* support from usb-ohci.c by Adam Richter, adam@yggdrasil.com).
* (C) Copyright 1999 Gregory P. Smith (from usb-ohci.c)
* (C) Copyright 2004-2006 Alan Stern, stern@rowland.harvard.edu
*/
/*
* Technically, updating td->status here is a race, but it's not really a
* problem. The worst that can happen is that we set the IOC bit again
* generating a spurious interrupt. We could fix this by creating another
* QH and leaving the IOC bit always set, but then we would have to play
* games with the FSBR code to make sure we get the correct order in all
* the cases. I don't think it's worth the effort
*/
static void uhci_set_next_interrupt(struct uhci_hcd *uhci)
{
if (uhci->is_stopped)
mod_timer(&uhci_to_hcd(uhci)->rh_timer, jiffies);
uhci->term_td->status |= cpu_to_le32(TD_CTRL_IOC);
}
static inline void uhci_clear_next_interrupt(struct uhci_hcd *uhci)
{
uhci->term_td->status &= ~cpu_to_le32(TD_CTRL_IOC);
}
/*
* Full-Speed Bandwidth Reclamation (FSBR).
* We turn on FSBR whenever a queue that wants it is advancing,
* and leave it on for a short time thereafter.
*/
static void uhci_fsbr_on(struct uhci_hcd *uhci)
{
uhci->fsbr_is_on = 1;
uhci->skel_term_qh->link = cpu_to_le32(
uhci->skel_fs_control_qh->dma_handle) | UHCI_PTR_QH;
}
static void uhci_fsbr_off(struct uhci_hcd *uhci)
{
uhci->fsbr_is_on = 0;
uhci->skel_term_qh->link = UHCI_PTR_TERM;
}
static void uhci_add_fsbr(struct uhci_hcd *uhci, struct urb *urb)
{
struct urb_priv *urbp = urb->hcpriv;
if (!(urb->transfer_flags & URB_NO_FSBR))
urbp->fsbr = 1;
}
static void uhci_urbp_wants_fsbr(struct uhci_hcd *uhci, struct urb_priv *urbp)
{
if (urbp->fsbr) {
uhci->fsbr_is_wanted = 1;
if (!uhci->fsbr_is_on)
uhci_fsbr_on(uhci);
else if (uhci->fsbr_expiring) {
uhci->fsbr_expiring = 0;
del_timer(&uhci->fsbr_timer);
}
}
}
static void uhci_fsbr_timeout(unsigned long _uhci)
{
struct uhci_hcd *uhci = (struct uhci_hcd *) _uhci;
unsigned long flags;
spin_lock_irqsave(&uhci->lock, flags);
if (uhci->fsbr_expiring) {
uhci->fsbr_expiring = 0;
uhci_fsbr_off(uhci);
}
spin_unlock_irqrestore(&uhci->lock, flags);
}
static struct uhci_td *uhci_alloc_td(struct uhci_hcd *uhci)
{
dma_addr_t dma_handle;
struct uhci_td *td;
td = dma_pool_alloc(uhci->td_pool, GFP_ATOMIC, &dma_handle);
if (!td)
return NULL;
td->dma_handle = dma_handle;
td->frame = -1;
INIT_LIST_HEAD(&td->list);
INIT_LIST_HEAD(&td->fl_list);
return td;
}
static void uhci_free_td(struct uhci_hcd *uhci, struct uhci_td *td)
{
if (!list_empty(&td->list))
dev_warn(uhci_dev(uhci), "td %p still in list!\n", td);
if (!list_empty(&td->fl_list))
dev_warn(uhci_dev(uhci), "td %p still in fl_list!\n", td);
dma_pool_free(uhci->td_pool, td, td->dma_handle);
}
static inline void uhci_fill_td(struct uhci_td *td, u32 status,
u32 token, u32 buffer)
{
td->status = cpu_to_le32(status);
td->token = cpu_to_le32(token);
td->buffer = cpu_to_le32(buffer);
}
static void uhci_add_td_to_urbp(struct uhci_td *td, struct urb_priv *urbp)
{
list_add_tail(&td->list, &urbp->td_list);
}
static void uhci_remove_td_from_urbp(struct uhci_td *td)
{
list_del_init(&td->list);
}
/*
* We insert Isochronous URBs directly into the frame list at the beginning
*/
static inline void uhci_insert_td_in_frame_list(struct uhci_hcd *uhci,
struct uhci_td *td, unsigned framenum)
{
framenum &= (UHCI_NUMFRAMES - 1);
td->frame = framenum;
/* Is there a TD already mapped there? */
if (uhci->frame_cpu[framenum]) {
struct uhci_td *ftd, *ltd;
ftd = uhci->frame_cpu[framenum];
ltd = list_entry(ftd->fl_list.prev, struct uhci_td, fl_list);
list_add_tail(&td->fl_list, &ftd->fl_list);
td->link = ltd->link;
wmb();
ltd->link = cpu_to_le32(td->dma_handle);
} else {
td->link = uhci->frame[framenum];
wmb();
uhci->frame[framenum] = cpu_to_le32(td->dma_handle);
uhci->frame_cpu[framenum] = td;
}
}
static inline void uhci_remove_td_from_frame_list(struct uhci_hcd *uhci,
struct uhci_td *td)
{
/* If it's not inserted, don't remove it */
if (td->frame == -1) {
WARN_ON(!list_empty(&td->fl_list));
return;
}
if (uhci->frame_cpu[td->frame] == td) {
if (list_empty(&td->fl_list)) {
uhci->frame[td->frame] = td->link;
uhci->frame_cpu[td->frame] = NULL;
} else {
struct uhci_td *ntd;
ntd = list_entry(td->fl_list.next, struct uhci_td, fl_list);
uhci->frame[td->frame] = cpu_to_le32(ntd->dma_handle);
uhci->frame_cpu[td->frame] = ntd;
}
} else {
struct uhci_td *ptd;
ptd = list_entry(td->fl_list.prev, struct uhci_td, fl_list);
ptd->link = td->link;
}
list_del_init(&td->fl_list);
td->frame = -1;
}
static inline void uhci_remove_tds_from_frame(struct uhci_hcd *uhci,
unsigned int framenum)
{
struct uhci_td *ftd, *ltd;
framenum &= (UHCI_NUMFRAMES - 1);
ftd = uhci->frame_cpu[framenum];
if (ftd) {
ltd = list_entry(ftd->fl_list.prev, struct uhci_td, fl_list);
uhci->frame[framenum] = ltd->link;
uhci->frame_cpu[framenum] = NULL;
while (!list_empty(&ftd->fl_list))
list_del_init(ftd->fl_list.prev);
}
}
/*
* Remove all the TDs for an Isochronous URB from the frame list
*/
static void uhci_unlink_isochronous_tds(struct uhci_hcd *uhci, struct urb *urb)
{
struct urb_priv *urbp = (struct urb_priv *) urb->hcpriv;
struct uhci_td *td;
list_for_each_entry(td, &urbp->td_list, list)
uhci_remove_td_from_frame_list(uhci, td);
}
static struct uhci_qh *uhci_alloc_qh(struct uhci_hcd *uhci,
struct usb_device *udev, struct usb_host_endpoint *hep)
{
dma_addr_t dma_handle;
struct uhci_qh *qh;
qh = dma_pool_alloc(uhci->qh_pool, GFP_ATOMIC, &dma_handle);
if (!qh)
return NULL;
memset(qh, 0, sizeof(*qh));
qh->dma_handle = dma_handle;
qh->element = UHCI_PTR_TERM;
qh->link = UHCI_PTR_TERM;
INIT_LIST_HEAD(&qh->queue);
INIT_LIST_HEAD(&qh->node);
if (udev) { /* Normal QH */
qh->dummy_td = uhci_alloc_td(uhci);
if (!qh->dummy_td) {
dma_pool_free(uhci->qh_pool, qh, dma_handle);
return NULL;
}
qh->state = QH_STATE_IDLE;
qh->hep = hep;
qh->udev = udev;
hep->hcpriv = qh;
qh->type = hep->desc.bmAttributes & USB_ENDPOINT_XFERTYPE_MASK;
} else { /* Skeleton QH */
qh->state = QH_STATE_ACTIVE;
qh->type = -1;
}
return qh;
}
static void uhci_free_qh(struct uhci_hcd *uhci, struct uhci_qh *qh)
{
WARN_ON(qh->state != QH_STATE_IDLE && qh->udev);
if (!list_empty(&qh->queue))
dev_warn(uhci_dev(uhci), "qh %p list not empty!\n", qh);
list_del(&qh->node);
if (qh->udev) {
qh->hep->hcpriv = NULL;
uhci_free_td(uhci, qh->dummy_td);
}
dma_pool_free(uhci->qh_pool, qh, qh->dma_handle);
}
/*
* When a queue is stopped and a dequeued URB is given back, adjust
* the previous TD link (if the URB isn't first on the queue) or
* save its toggle value (if it is first and is currently executing).
*
* Returns 0 if the URB should not yet be given back, 1 otherwise.
*/
static int uhci_cleanup_queue(struct uhci_hcd *uhci, struct uhci_qh *qh,
struct urb *urb)
{
struct urb_priv *urbp = urb->hcpriv;
struct uhci_td *td;
int ret = 1;
/* Isochronous pipes don't use toggles and their TD link pointers
* get adjusted during uhci_urb_dequeue(). But since their queues
* cannot truly be stopped, we have to watch out for dequeues
* occurring after the nominal unlink frame. */
if (qh->type == USB_ENDPOINT_XFER_ISOC) {
ret = (uhci->frame_number + uhci->is_stopped !=
qh->unlink_frame);
goto done;
}
/* If the URB isn't first on its queue, adjust the link pointer
* of the last TD in the previous URB. The toggle doesn't need
* to be saved since this URB can't be executing yet. */
if (qh->queue.next != &urbp->node) {
struct urb_priv *purbp;
struct uhci_td *ptd;
purbp = list_entry(urbp->node.prev, struct urb_priv, node);
WARN_ON(list_empty(&purbp->td_list));
ptd = list_entry(purbp->td_list.prev, struct uhci_td,
list);
td = list_entry(urbp->td_list.prev, struct uhci_td,
list);
ptd->link = td->link;
goto done;
}
/* If the QH element pointer is UHCI_PTR_TERM then then currently
* executing URB has already been unlinked, so this one isn't it. */
if (qh_element(qh) == UHCI_PTR_TERM)
goto done;
qh->element = UHCI_PTR_TERM;
/* Control pipes have to worry about toggles */
if (qh->type == USB_ENDPOINT_XFER_CONTROL)
goto done;
/* Save the next toggle value */
WARN_ON(list_empty(&urbp->td_list));
td = list_entry(urbp->td_list.next, struct uhci_td, list);
qh->needs_fixup = 1;
qh->initial_toggle = uhci_toggle(td_token(td));
done:
return ret;
}
/*
* Fix up the data toggles for URBs in a queue, when one of them
* terminates early (short transfer, error, or dequeued).
*/
static void uhci_fixup_toggles(struct uhci_qh *qh, int skip_first)
{
struct urb_priv *urbp = NULL;
struct uhci_td *td;
unsigned int toggle = qh->initial_toggle;
unsigned int pipe;
/* Fixups for a short transfer start with the second URB in the
* queue (the short URB is the first). */
if (skip_first)
urbp = list_entry(qh->queue.next, struct urb_priv, node);
/* When starting with the first URB, if the QH element pointer is
* still valid then we know the URB's toggles are okay. */
else if (qh_element(qh) != UHCI_PTR_TERM)
toggle = 2;
/* Fix up the toggle for the URBs in the queue. Normally this
* loop won't run more than once: When an error or short transfer
* occurs, the queue usually gets emptied. */
urbp = list_prepare_entry(urbp, &qh->queue, node);
list_for_each_entry_continue(urbp, &qh->queue, node) {
/* If the first TD has the right toggle value, we don't
* need to change any toggles in this URB */
td = list_entry(urbp->td_list.next, struct uhci_td, list);
if (toggle > 1 || uhci_toggle(td_token(td)) == toggle) {
td = list_entry(urbp->td_list.prev, struct uhci_td,
list);
toggle = uhci_toggle(td_token(td)) ^ 1;
/* Otherwise all the toggles in the URB have to be switched */
} else {
list_for_each_entry(td, &urbp->td_list, list) {
td->token ^= __constant_cpu_to_le32(
TD_TOKEN_TOGGLE);
toggle ^= 1;
}
}
}
wmb();
pipe = list_entry(qh->queue.next, struct urb_priv, node)->urb->pipe;
usb_settoggle(qh->udev, usb_pipeendpoint(pipe),
usb_pipeout(pipe), toggle);
qh->needs_fixup = 0;
}
/*
* Put a QH on the schedule in both hardware and software
*/
static void uhci_activate_qh(struct uhci_hcd *uhci, struct uhci_qh *qh)
{
struct uhci_qh *pqh;
WARN_ON(list_empty(&qh->queue));
/* Set the element pointer if it isn't set already.
* This isn't needed for Isochronous queues, but it doesn't hurt. */
if (qh_element(qh) == UHCI_PTR_TERM) {
struct urb_priv *urbp = list_entry(qh->queue.next,
struct urb_priv, node);
struct uhci_td *td = list_entry(urbp->td_list.next,
struct uhci_td, list);
qh->element = cpu_to_le32(td->dma_handle);
}
/* Treat the queue as if it has just advanced */
qh->wait_expired = 0;
qh->advance_jiffies = jiffies;
if (qh->state == QH_STATE_ACTIVE)
return;
qh->state = QH_STATE_ACTIVE;
/* Move the QH from its old list to the end of the appropriate
* skeleton's list */
if (qh == uhci->next_qh)
uhci->next_qh = list_entry(qh->node.next, struct uhci_qh,
node);
list_move_tail(&qh->node, &qh->skel->node);
/* Link it into the schedule */
pqh = list_entry(qh->node.prev, struct uhci_qh, node);
qh->link = pqh->link;
wmb();
pqh->link = UHCI_PTR_QH | cpu_to_le32(qh->dma_handle);
}
/*
* Take a QH off the hardware schedule
*/
static void uhci_unlink_qh(struct uhci_hcd *uhci, struct uhci_qh *qh)
{
struct uhci_qh *pqh;
if (qh->state == QH_STATE_UNLINKING)
return;
WARN_ON(qh->state != QH_STATE_ACTIVE || !qh->udev);
qh->state = QH_STATE_UNLINKING;
/* Unlink the QH from the schedule and record when we did it */
pqh = list_entry(qh->node.prev, struct uhci_qh, node);
pqh->link = qh->link;
mb();
uhci_get_current_frame_number(uhci);
qh->unlink_frame = uhci->frame_number;
/* Force an interrupt so we know when the QH is fully unlinked */
if (list_empty(&uhci->skel_unlink_qh->node))
uhci_set_next_interrupt(uhci);
/* Move the QH from its old list to the end of the unlinking list */
if (qh == uhci->next_qh)
uhci->next_qh = list_entry(qh->node.next, struct uhci_qh,
node);
list_move_tail(&qh->node, &uhci->skel_unlink_qh->node);
}
/*
* When we and the controller are through with a QH, it becomes IDLE.
* This happens when a QH has been off the schedule (on the unlinking
* list) for more than one frame, or when an error occurs while adding
* the first URB onto a new QH.
*/
static void uhci_make_qh_idle(struct uhci_hcd *uhci, struct uhci_qh *qh)
{
WARN_ON(qh->state == QH_STATE_ACTIVE);
if (qh == uhci->next_qh)
uhci->next_qh = list_entry(qh->node.next, struct uhci_qh,
node);
list_move(&qh->node, &uhci->idle_qh_list);
qh->state = QH_STATE_IDLE;
/* Now that the QH is idle, its post_td isn't being used */
if (qh->post_td) {
uhci_free_td(uhci, qh->post_td);
qh->post_td = NULL;
}
/* If anyone is waiting for a QH to become idle, wake them up */
if (uhci->num_waiting)
wake_up_all(&uhci->waitqh);
}
static inline struct urb_priv *uhci_alloc_urb_priv(struct uhci_hcd *uhci,
struct urb *urb)
{
struct urb_priv *urbp;
urbp = kmem_cache_alloc(uhci_up_cachep, GFP_ATOMIC);
if (!urbp)
return NULL;
memset((void *)urbp, 0, sizeof(*urbp));
urbp->urb = urb;
urb->hcpriv = urbp;
INIT_LIST_HEAD(&urbp->node);
INIT_LIST_HEAD(&urbp->td_list);
return urbp;
}
static void uhci_free_urb_priv(struct uhci_hcd *uhci,
struct urb_priv *urbp)
{
struct uhci_td *td, *tmp;
if (!list_empty(&urbp->node))
dev_warn(uhci_dev(uhci), "urb %p still on QH's list!\n",
urbp->urb);
list_for_each_entry_safe(td, tmp, &urbp->td_list, list) {
uhci_remove_td_from_urbp(td);
uhci_free_td(uhci, td);
}
urbp->urb->hcpriv = NULL;
kmem_cache_free(uhci_up_cachep, urbp);
}
/*
* Map status to standard result codes
*
* <status> is (td_status(td) & 0xF60000), a.k.a.
* uhci_status_bits(td_status(td)).
* Note: <status> does not include the TD_CTRL_NAK bit.
* <dir_out> is True for output TDs and False for input TDs.
*/
static int uhci_map_status(int status, int dir_out)
{
if (!status)
return 0;
if (status & TD_CTRL_BITSTUFF) /* Bitstuff error */
return -EPROTO;
if (status & TD_CTRL_CRCTIMEO) { /* CRC/Timeout */
if (dir_out)
return -EPROTO;
else
return -EILSEQ;
}
if (status & TD_CTRL_BABBLE) /* Babble */
return -EOVERFLOW;
if (status & TD_CTRL_DBUFERR) /* Buffer error */
return -ENOSR;
if (status & TD_CTRL_STALLED) /* Stalled */
return -EPIPE;
return 0;
}
/*
* Control transfers
*/
static int uhci_submit_control(struct uhci_hcd *uhci, struct urb *urb,
struct uhci_qh *qh)
{
struct uhci_td *td;
unsigned long destination, status;
int maxsze = le16_to_cpu(qh->hep->desc.wMaxPacketSize);
int len = urb->transfer_buffer_length;
dma_addr_t data = urb->transfer_dma;
__le32 *plink;
struct urb_priv *urbp = urb->hcpriv;
/* The "pipe" thing contains the destination in bits 8--18 */
destination = (urb->pipe & PIPE_DEVEP_MASK) | USB_PID_SETUP;
/* 3 errors, dummy TD remains inactive */
status = uhci_maxerr(3);
if (urb->dev->speed == USB_SPEED_LOW)
status |= TD_CTRL_LS;
/*
* Build the TD for the control request setup packet
*/
td = qh->dummy_td;
uhci_add_td_to_urbp(td, urbp);
uhci_fill_td(td, status, destination | uhci_explen(8),
urb->setup_dma);
plink = &td->link;
status |= TD_CTRL_ACTIVE;
/*
* If direction is "send", change the packet ID from SETUP (0x2D)
* to OUT (0xE1). Else change it from SETUP to IN (0x69) and
* set Short Packet Detect (SPD) for all data packets.
*/
if (usb_pipeout(urb->pipe))
destination ^= (USB_PID_SETUP ^ USB_PID_OUT);
else {
destination ^= (USB_PID_SETUP ^ USB_PID_IN);
status |= TD_CTRL_SPD;
}
/*
* Build the DATA TDs
*/
while (len > 0) {
int pktsze = min(len, maxsze);
td = uhci_alloc_td(uhci);
if (!td)
goto nomem;
*plink = cpu_to_le32(td->dma_handle);
/* Alternate Data0/1 (start with Data1) */
destination ^= TD_TOKEN_TOGGLE;
uhci_add_td_to_urbp(td, urbp);
uhci_fill_td(td, status, destination | uhci_explen(pktsze),
data);
plink = &td->link;
data += pktsze;
len -= pktsze;
}
/*
* Build the final TD for control status
*/
td = uhci_alloc_td(uhci);
if (!td)
goto nomem;
*plink = cpu_to_le32(td->dma_handle);
/*
* It's IN if the pipe is an output pipe or we're not expecting
* data back.
*/
destination &= ~TD_TOKEN_PID_MASK;
if (usb_pipeout(urb->pipe) || !urb->transfer_buffer_length)
destination |= USB_PID_IN;
else
destination |= USB_PID_OUT;
destination |= TD_TOKEN_TOGGLE; /* End in Data1 */
status &= ~TD_CTRL_SPD;
uhci_add_td_to_urbp(td, urbp);
uhci_fill_td(td, status | TD_CTRL_IOC,
destination | uhci_explen(0), 0);
plink = &td->link;
/*
* Build the new dummy TD and activate the old one
*/
td = uhci_alloc_td(uhci);
if (!td)
goto nomem;
*plink = cpu_to_le32(td->dma_handle);
uhci_fill_td(td, 0, USB_PID_OUT | uhci_explen(0), 0);
wmb();
qh->dummy_td->status |= __constant_cpu_to_le32(TD_CTRL_ACTIVE);
qh->dummy_td = td;
/* Low-speed transfers get a different queue, and won't hog the bus.
* Also, some devices enumerate better without FSBR; the easiest way
* to do that is to put URBs on the low-speed queue while the device
* isn't in the CONFIGURED state. */
if (urb->dev->speed == USB_SPEED_LOW ||
urb->dev->state != USB_STATE_CONFIGURED)
qh->skel = uhci->skel_ls_control_qh;
else {
qh->skel = uhci->skel_fs_control_qh;
uhci_add_fsbr(uhci, urb);
}
urb->actual_length = -8; /* Account for the SETUP packet */
return 0;
nomem:
/* Remove the dummy TD from the td_list so it doesn't get freed */
uhci_remove_td_from_urbp(qh->dummy_td);
return -ENOMEM;
}
/*
* Common submit for bulk and interrupt
*/
static int uhci_submit_common(struct uhci_hcd *uhci, struct urb *urb,
struct uhci_qh *qh)
{
struct uhci_td *td;
unsigned long destination, status;
int maxsze = le16_to_cpu(qh->hep->desc.wMaxPacketSize);
int len = urb->transfer_buffer_length;
dma_addr_t data = urb->transfer_dma;
__le32 *plink;
struct urb_priv *urbp = urb->hcpriv;
unsigned int toggle;
if (len < 0)
return -EINVAL;
/* The "pipe" thing contains the destination in bits 8--18 */
destination = (urb->pipe & PIPE_DEVEP_MASK) | usb_packetid(urb->pipe);
toggle = usb_gettoggle(urb->dev, usb_pipeendpoint(urb->pipe),
usb_pipeout(urb->pipe));
/* 3 errors, dummy TD remains inactive */
status = uhci_maxerr(3);
if (urb->dev->speed == USB_SPEED_LOW)
status |= TD_CTRL_LS;
if (usb_pipein(urb->pipe))
status |= TD_CTRL_SPD;
/*
* Build the DATA TDs
*/
plink = NULL;
td = qh->dummy_td;
do { /* Allow zero length packets */
int pktsze = maxsze;
if (len <= pktsze) { /* The last packet */
pktsze = len;
if (!(urb->transfer_flags & URB_SHORT_NOT_OK))
status &= ~TD_CTRL_SPD;
}
if (plink) {
td = uhci_alloc_td(uhci);
if (!td)
goto nomem;
*plink = cpu_to_le32(td->dma_handle);
}
uhci_add_td_to_urbp(td, urbp);
uhci_fill_td(td, status,
destination | uhci_explen(pktsze) |
(toggle << TD_TOKEN_TOGGLE_SHIFT),
data);
plink = &td->link;
status |= TD_CTRL_ACTIVE;
data += pktsze;
len -= maxsze;
toggle ^= 1;
} while (len > 0);
/*
* URB_ZERO_PACKET means adding a 0-length packet, if direction
* is OUT and the transfer_length was an exact multiple of maxsze,
* hence (len = transfer_length - N * maxsze) == 0
* however, if transfer_length == 0, the zero packet was already
* prepared above.
*/
if ((urb->transfer_flags & URB_ZERO_PACKET) &&
usb_pipeout(urb->pipe) && len == 0 &&
urb->transfer_buffer_length > 0) {
td = uhci_alloc_td(uhci);
if (!td)
goto nomem;
*plink = cpu_to_le32(td->dma_handle);
uhci_add_td_to_urbp(td, urbp);
uhci_fill_td(td, status,
destination | uhci_explen(0) |
(toggle << TD_TOKEN_TOGGLE_SHIFT),
data);
plink = &td->link;
toggle ^= 1;
}
/* Set the interrupt-on-completion flag on the last packet.
* A more-or-less typical 4 KB URB (= size of one memory page)
* will require about 3 ms to transfer; that's a little on the
* fast side but not enough to justify delaying an interrupt
* more than 2 or 3 URBs, so we will ignore the URB_NO_INTERRUPT
* flag setting. */
td->status |= __constant_cpu_to_le32(TD_CTRL_IOC);
/*
* Build the new dummy TD and activate the old one
*/
td = uhci_alloc_td(uhci);
if (!td)
goto nomem;
*plink = cpu_to_le32(td->dma_handle);
uhci_fill_td(td, 0, USB_PID_OUT | uhci_explen(0), 0);
wmb();
qh->dummy_td->status |= __constant_cpu_to_le32(TD_CTRL_ACTIVE);
qh->dummy_td = td;
qh->period = urb->interval;
usb_settoggle(urb->dev, usb_pipeendpoint(urb->pipe),
usb_pipeout(urb->pipe), toggle);
return 0;
nomem:
/* Remove the dummy TD from the td_list so it doesn't get freed */
uhci_remove_td_from_urbp(qh->dummy_td);
return -ENOMEM;
}
static inline int uhci_submit_bulk(struct uhci_hcd *uhci, struct urb *urb,
struct uhci_qh *qh)
{
int ret;
/* Can't have low-speed bulk transfers */
if (urb->dev->speed == USB_SPEED_LOW)
return -EINVAL;
qh->skel = uhci->skel_bulk_qh;
ret = uhci_submit_common(uhci, urb, qh);
if (ret == 0)
uhci_add_fsbr(uhci, urb);
return ret;
}
static int uhci_submit_interrupt(struct uhci_hcd *uhci, struct urb *urb,
struct uhci_qh *qh)
{
int exponent;
/* USB 1.1 interrupt transfers only involve one packet per interval.
* Drivers can submit URBs of any length, but longer ones will need
* multiple intervals to complete.
*/
/* Figure out which power-of-two queue to use */
for (exponent = 7; exponent >= 0; --exponent) {
if ((1 << exponent) <= urb->interval)
break;
}
if (exponent < 0)
return -EINVAL;
urb->interval = 1 << exponent;
if (qh->period == 0)
qh->skel = uhci->skelqh[UHCI_SKEL_INDEX(exponent)];
else if (qh->period != urb->interval)
return -EINVAL; /* Can't change the period */
return uhci_submit_common(uhci, urb, qh);
}
/*
* Fix up the data structures following a short transfer
*/
static int uhci_fixup_short_transfer(struct uhci_hcd *uhci,
struct uhci_qh *qh, struct urb_priv *urbp)
{
struct uhci_td *td;
struct list_head *tmp;
int ret;
td = list_entry(urbp->td_list.prev, struct uhci_td, list);
if (qh->type == USB_ENDPOINT_XFER_CONTROL) {
/* When a control transfer is short, we have to restart
* the queue at the status stage transaction, which is
* the last TD. */
WARN_ON(list_empty(&urbp->td_list));
qh->element = cpu_to_le32(td->dma_handle);
tmp = td->list.prev;
ret = -EINPROGRESS;
} else {
/* When a bulk/interrupt transfer is short, we have to
* fix up the toggles of the following URBs on the queue
* before restarting the queue at the next URB. */
qh->initial_toggle = uhci_toggle(td_token(qh->post_td)) ^ 1;
uhci_fixup_toggles(qh, 1);
if (list_empty(&urbp->td_list))
td = qh->post_td;
qh->element = td->link;
tmp = urbp->td_list.prev;
ret = 0;
}
/* Remove all the TDs we skipped over, from tmp back to the start */
while (tmp != &urbp->td_list) {
td = list_entry(tmp, struct uhci_td, list);
tmp = tmp->prev;
uhci_remove_td_from_urbp(td);
uhci_free_td(uhci, td);
}
return ret;
}
/*
* Common result for control, bulk, and interrupt
*/
static int uhci_result_common(struct uhci_hcd *uhci, struct urb *urb)
{
struct urb_priv *urbp = urb->hcpriv;
struct uhci_qh *qh = urbp->qh;
struct uhci_td *td, *tmp;
unsigned status;
int ret = 0;
list_for_each_entry_safe(td, tmp, &urbp->td_list, list) {
unsigned int ctrlstat;
int len;
ctrlstat = td_status(td);
status = uhci_status_bits(ctrlstat);
if (status & TD_CTRL_ACTIVE)
return -EINPROGRESS;
len = uhci_actual_length(ctrlstat);
urb->actual_length += len;
if (status) {
ret = uhci_map_status(status,
uhci_packetout(td_token(td)));
if ((debug == 1 && ret != -EPIPE) || debug > 1) {
/* Some debugging code */
dev_dbg(&urb->dev->dev,
"%s: failed with status %x\n",
__FUNCTION__, status);
if (debug > 1 && errbuf) {
/* Print the chain for debugging */
uhci_show_qh(urbp->qh, errbuf,
ERRBUF_LEN, 0);
lprintk(errbuf);
}
}
} else if (len < uhci_expected_length(td_token(td))) {
/* We received a short packet */
if (urb->transfer_flags & URB_SHORT_NOT_OK)
ret = -EREMOTEIO;
/* Fixup needed only if this isn't the URB's last TD */
else if (&td->list != urbp->td_list.prev)
ret = 1;
}
uhci_remove_td_from_urbp(td);
if (qh->post_td)
uhci_free_td(uhci, qh->post_td);
qh->post_td = td;
if (ret != 0)
goto err;
}
return ret;
err:
if (ret < 0) {
/* In case a control transfer gets an error
* during the setup stage */
urb->actual_length = max(urb->actual_length, 0);
/* Note that the queue has stopped and save
* the next toggle value */
qh->element = UHCI_PTR_TERM;
qh->is_stopped = 1;
qh->needs_fixup = (qh->type != USB_ENDPOINT_XFER_CONTROL);
qh->initial_toggle = uhci_toggle(td_token(td)) ^
(ret == -EREMOTEIO);
} else /* Short packet received */
ret = uhci_fixup_short_transfer(uhci, qh, urbp);
return ret;
}
/*
* Isochronous transfers
*/
static int uhci_submit_isochronous(struct uhci_hcd *uhci, struct urb *urb,
struct uhci_qh *qh)
{
struct uhci_td *td = NULL; /* Since urb->number_of_packets > 0 */
int i, frame;
unsigned long destination, status;
struct urb_priv *urbp = (struct urb_priv *) urb->hcpriv;
/* Values must not be too big (could overflow below) */
if (urb->interval >= UHCI_NUMFRAMES ||
urb->number_of_packets >= UHCI_NUMFRAMES)
return -EFBIG;
/* Check the period and figure out the starting frame number */
if (qh->period == 0) {
if (urb->transfer_flags & URB_ISO_ASAP) {
uhci_get_current_frame_number(uhci);
urb->start_frame = uhci->frame_number + 10;
} else {
i = urb->start_frame - uhci->last_iso_frame;
if (i <= 0 || i >= UHCI_NUMFRAMES)
return -EINVAL;
}
} else if (qh->period != urb->interval) {
return -EINVAL; /* Can't change the period */
} else { /* Pick up where the last URB leaves off */
if (list_empty(&qh->queue)) {
frame = qh->iso_frame;
} else {
struct urb *lurb;
lurb = list_entry(qh->queue.prev,
struct urb_priv, node)->urb;
frame = lurb->start_frame +
lurb->number_of_packets *
lurb->interval;
}
if (urb->transfer_flags & URB_ISO_ASAP)
urb->start_frame = frame;
else if (urb->start_frame != frame)
return -EINVAL;
}
/* Make sure we won't have to go too far into the future */
if (uhci_frame_before_eq(uhci->last_iso_frame + UHCI_NUMFRAMES,
urb->start_frame + urb->number_of_packets *
urb->interval))
return -EFBIG;
status = TD_CTRL_ACTIVE | TD_CTRL_IOS;
destination = (urb->pipe & PIPE_DEVEP_MASK) | usb_packetid(urb->pipe);
for (i = 0; i < urb->number_of_packets; i++) {
td = uhci_alloc_td(uhci);
if (!td)
return -ENOMEM;
uhci_add_td_to_urbp(td, urbp);
uhci_fill_td(td, status, destination |
uhci_explen(urb->iso_frame_desc[i].length),
urb->transfer_dma +
urb->iso_frame_desc[i].offset);
}
/* Set the interrupt-on-completion flag on the last packet. */
td->status |= __constant_cpu_to_le32(TD_CTRL_IOC);
qh->skel = uhci->skel_iso_qh;
qh->period = urb->interval;
/* Add the TDs to the frame list */
frame = urb->start_frame;
list_for_each_entry(td, &urbp->td_list, list) {
uhci_insert_td_in_frame_list(uhci, td, frame);
frame += qh->period;
}
if (list_empty(&qh->queue)) {
qh->iso_packet_desc = &urb->iso_frame_desc[0];
qh->iso_frame = urb->start_frame;
qh->iso_status = 0;
}
return 0;
}
static int uhci_result_isochronous(struct uhci_hcd *uhci, struct urb *urb)
{
struct uhci_td *td, *tmp;
struct urb_priv *urbp = urb->hcpriv;
struct uhci_qh *qh = urbp->qh;
list_for_each_entry_safe(td, tmp, &urbp->td_list, list) {
unsigned int ctrlstat;
int status;
int actlength;
if (uhci_frame_before_eq(uhci->cur_iso_frame, qh->iso_frame))
return -EINPROGRESS;
uhci_remove_tds_from_frame(uhci, qh->iso_frame);
ctrlstat = td_status(td);
if (ctrlstat & TD_CTRL_ACTIVE) {
status = -EXDEV; /* TD was added too late? */
} else {
status = uhci_map_status(uhci_status_bits(ctrlstat),
usb_pipeout(urb->pipe));
actlength = uhci_actual_length(ctrlstat);
urb->actual_length += actlength;
qh->iso_packet_desc->actual_length = actlength;
qh->iso_packet_desc->status = status;
}
if (status) {
urb->error_count++;
qh->iso_status = status;
}
uhci_remove_td_from_urbp(td);
uhci_free_td(uhci, td);
qh->iso_frame += qh->period;
++qh->iso_packet_desc;
}
return qh->iso_status;
}
static int uhci_urb_enqueue(struct usb_hcd *hcd,
struct usb_host_endpoint *hep,
struct urb *urb, gfp_t mem_flags)
{
int ret;
struct uhci_hcd *uhci = hcd_to_uhci(hcd);
unsigned long flags;
struct urb_priv *urbp;
struct uhci_qh *qh;
int bustime;
spin_lock_irqsave(&uhci->lock, flags);
ret = urb->status;
if (ret != -EINPROGRESS) /* URB already unlinked! */
goto done;
ret = -ENOMEM;
urbp = uhci_alloc_urb_priv(uhci, urb);
if (!urbp)
goto done;
if (hep->hcpriv)
qh = (struct uhci_qh *) hep->hcpriv;
else {
qh = uhci_alloc_qh(uhci, urb->dev, hep);
if (!qh)
goto err_no_qh;
}
urbp->qh = qh;
switch (qh->type) {
case USB_ENDPOINT_XFER_CONTROL:
ret = uhci_submit_control(uhci, urb, qh);
break;
case USB_ENDPOINT_XFER_BULK:
ret = uhci_submit_bulk(uhci, urb, qh);
break;
case USB_ENDPOINT_XFER_INT:
if (list_empty(&qh->queue)) {
bustime = usb_check_bandwidth(urb->dev, urb);
if (bustime < 0)
ret = bustime;
else {
ret = uhci_submit_interrupt(uhci, urb, qh);
if (ret == 0)
usb_claim_bandwidth(urb->dev, urb, bustime, 0);
}
} else { /* inherit from parent */
struct urb_priv *eurbp;
eurbp = list_entry(qh->queue.prev, struct urb_priv,
node);
urb->bandwidth = eurbp->urb->bandwidth;
ret = uhci_submit_interrupt(uhci, urb, qh);
}
break;
case USB_ENDPOINT_XFER_ISOC:
urb->error_count = 0;
bustime = usb_check_bandwidth(urb->dev, urb);
if (bustime < 0) {
ret = bustime;
break;
}
ret = uhci_submit_isochronous(uhci, urb, qh);
if (ret == 0)
usb_claim_bandwidth(urb->dev, urb, bustime, 1);
break;
}
if (ret != 0)
goto err_submit_failed;
/* Add this URB to the QH */
urbp->qh = qh;
list_add_tail(&urbp->node, &qh->queue);
/* If the new URB is the first and only one on this QH then either
* the QH is new and idle or else it's unlinked and waiting to
* become idle, so we can activate it right away. But only if the
* queue isn't stopped. */
if (qh->queue.next == &urbp->node && !qh->is_stopped) {
uhci_activate_qh(uhci, qh);
uhci_urbp_wants_fsbr(uhci, urbp);
}
goto done;
err_submit_failed:
if (qh->state == QH_STATE_IDLE)
uhci_make_qh_idle(uhci, qh); /* Reclaim unused QH */
err_no_qh:
uhci_free_urb_priv(uhci, urbp);
done:
spin_unlock_irqrestore(&uhci->lock, flags);
return ret;
}
static int uhci_urb_dequeue(struct usb_hcd *hcd, struct urb *urb)
{
struct uhci_hcd *uhci = hcd_to_uhci(hcd);
unsigned long flags;
struct urb_priv *urbp;
struct uhci_qh *qh;
spin_lock_irqsave(&uhci->lock, flags);
urbp = urb->hcpriv;
if (!urbp) /* URB was never linked! */
goto done;
qh = urbp->qh;
/* Remove Isochronous TDs from the frame list ASAP */
if (qh->type == USB_ENDPOINT_XFER_ISOC) {
uhci_unlink_isochronous_tds(uhci, urb);
mb();
/* If the URB has already started, update the QH unlink time */
uhci_get_current_frame_number(uhci);
if (uhci_frame_before_eq(urb->start_frame, uhci->frame_number))
qh->unlink_frame = uhci->frame_number;
}
uhci_unlink_qh(uhci, qh);
done:
spin_unlock_irqrestore(&uhci->lock, flags);
return 0;
}
/*
* Finish unlinking an URB and give it back
*/
static void uhci_giveback_urb(struct uhci_hcd *uhci, struct uhci_qh *qh,
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 15:55:46 +02:00
struct urb *urb)
__releases(uhci->lock)
__acquires(uhci->lock)
{
struct urb_priv *urbp = (struct urb_priv *) urb->hcpriv;
/* When giving back the first URB in an Isochronous queue,
* reinitialize the QH's iso-related members for the next URB. */
if (qh->type == USB_ENDPOINT_XFER_ISOC &&
urbp->node.prev == &qh->queue &&
urbp->node.next != &qh->queue) {
struct urb *nurb = list_entry(urbp->node.next,
struct urb_priv, node)->urb;
qh->iso_packet_desc = &nurb->iso_frame_desc[0];
qh->iso_frame = nurb->start_frame;
qh->iso_status = 0;
}
/* Take the URB off the QH's queue. If the queue is now empty,
* this is a perfect time for a toggle fixup. */
list_del_init(&urbp->node);
if (list_empty(&qh->queue) && qh->needs_fixup) {
usb_settoggle(urb->dev, usb_pipeendpoint(urb->pipe),
usb_pipeout(urb->pipe), qh->initial_toggle);
qh->needs_fixup = 0;
}
uhci_free_urb_priv(uhci, urbp);
switch (qh->type) {
case USB_ENDPOINT_XFER_ISOC:
/* Release bandwidth for Interrupt or Isoc. transfers */
if (urb->bandwidth)
usb_release_bandwidth(urb->dev, urb, 1);
break;
case USB_ENDPOINT_XFER_INT:
/* Release bandwidth for Interrupt or Isoc. transfers */
/* Make sure we don't release if we have a queued URB */
if (list_empty(&qh->queue) && urb->bandwidth)
usb_release_bandwidth(urb->dev, urb, 0);
else
/* bandwidth was passed on to queued URB, */
/* so don't let usb_unlink_urb() release it */
urb->bandwidth = 0;
break;
}
spin_unlock(&uhci->lock);
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 15:55:46 +02:00
usb_hcd_giveback_urb(uhci_to_hcd(uhci), urb);
spin_lock(&uhci->lock);
/* If the queue is now empty, we can unlink the QH and give up its
* reserved bandwidth. */
if (list_empty(&qh->queue)) {
uhci_unlink_qh(uhci, qh);
/* Bandwidth stuff not yet implemented */
qh->period = 0;
}
}
/*
* Scan the URBs in a QH's queue
*/
#define QH_FINISHED_UNLINKING(qh) \
(qh->state == QH_STATE_UNLINKING && \
uhci->frame_number + uhci->is_stopped != qh->unlink_frame)
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 15:55:46 +02:00
static void uhci_scan_qh(struct uhci_hcd *uhci, struct uhci_qh *qh)
{
struct urb_priv *urbp;
struct urb *urb;
int status;
while (!list_empty(&qh->queue)) {
urbp = list_entry(qh->queue.next, struct urb_priv, node);
urb = urbp->urb;
if (qh->type == USB_ENDPOINT_XFER_ISOC)
status = uhci_result_isochronous(uhci, urb);
else
status = uhci_result_common(uhci, urb);
if (status == -EINPROGRESS)
break;
spin_lock(&urb->lock);
if (urb->status == -EINPROGRESS) /* Not dequeued */
urb->status = status;
else
status = ECONNRESET; /* Not -ECONNRESET */
spin_unlock(&urb->lock);
/* Dequeued but completed URBs can't be given back unless
* the QH is stopped or has finished unlinking. */
if (status == ECONNRESET) {
if (QH_FINISHED_UNLINKING(qh))
qh->is_stopped = 1;
else if (!qh->is_stopped)
return;
}
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 15:55:46 +02:00
uhci_giveback_urb(uhci, qh, urb);
if (status < 0 && qh->type != USB_ENDPOINT_XFER_ISOC)
break;
}
/* If the QH is neither stopped nor finished unlinking (normal case),
* our work here is done. */
if (QH_FINISHED_UNLINKING(qh))
qh->is_stopped = 1;
else if (!qh->is_stopped)
return;
/* Otherwise give back each of the dequeued URBs */
restart:
list_for_each_entry(urbp, &qh->queue, node) {
urb = urbp->urb;
if (urb->status != -EINPROGRESS) {
/* Fix up the TD links and save the toggles for
* non-Isochronous queues. For Isochronous queues,
* test for too-recent dequeues. */
if (!uhci_cleanup_queue(uhci, qh, urb)) {
qh->is_stopped = 0;
return;
}
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 15:55:46 +02:00
uhci_giveback_urb(uhci, qh, urb);
goto restart;
}
}
qh->is_stopped = 0;
/* There are no more dequeued URBs. If there are still URBs on the
* queue, the QH can now be re-activated. */
if (!list_empty(&qh->queue)) {
if (qh->needs_fixup)
uhci_fixup_toggles(qh, 0);
/* If the first URB on the queue wants FSBR but its time
* limit has expired, set the next TD to interrupt on
* completion before reactivating the QH. */
urbp = list_entry(qh->queue.next, struct urb_priv, node);
if (urbp->fsbr && qh->wait_expired) {
struct uhci_td *td = list_entry(urbp->td_list.next,
struct uhci_td, list);
td->status |= __cpu_to_le32(TD_CTRL_IOC);
}
uhci_activate_qh(uhci, qh);
}
/* The queue is empty. The QH can become idle if it is fully
* unlinked. */
else if (QH_FINISHED_UNLINKING(qh))
uhci_make_qh_idle(uhci, qh);
}
/*
* Check for queues that have made some forward progress.
* Returns 0 if the queue is not Isochronous, is ACTIVE, and
* has not advanced since last examined; 1 otherwise.
*
* Early Intel controllers have a bug which causes qh->element sometimes
* not to advance when a TD completes successfully. The queue remains
* stuck on the inactive completed TD. We detect such cases and advance
* the element pointer by hand.
*/
static int uhci_advance_check(struct uhci_hcd *uhci, struct uhci_qh *qh)
{
struct urb_priv *urbp = NULL;
struct uhci_td *td;
int ret = 1;
unsigned status;
if (qh->type == USB_ENDPOINT_XFER_ISOC)
goto done;
/* Treat an UNLINKING queue as though it hasn't advanced.
* This is okay because reactivation will treat it as though
* it has advanced, and if it is going to become IDLE then
* this doesn't matter anyway. Furthermore it's possible
* for an UNLINKING queue not to have any URBs at all, or
* for its first URB not to have any TDs (if it was dequeued
* just as it completed). So it's not easy in any case to
* test whether such queues have advanced. */
if (qh->state != QH_STATE_ACTIVE) {
urbp = NULL;
status = 0;
} else {
urbp = list_entry(qh->queue.next, struct urb_priv, node);
td = list_entry(urbp->td_list.next, struct uhci_td, list);
status = td_status(td);
if (!(status & TD_CTRL_ACTIVE)) {
/* We're okay, the queue has advanced */
qh->wait_expired = 0;
qh->advance_jiffies = jiffies;
goto done;
}
ret = 0;
}
/* The queue hasn't advanced; check for timeout */
if (qh->wait_expired)
goto done;
if (time_after(jiffies, qh->advance_jiffies + QH_WAIT_TIMEOUT)) {
/* Detect the Intel bug and work around it */
if (qh->post_td && qh_element(qh) ==
cpu_to_le32(qh->post_td->dma_handle)) {
qh->element = qh->post_td->link;
qh->advance_jiffies = jiffies;
ret = 1;
goto done;
}
qh->wait_expired = 1;
/* If the current URB wants FSBR, unlink it temporarily
* so that we can safely set the next TD to interrupt on
* completion. That way we'll know as soon as the queue
* starts moving again. */
if (urbp && urbp->fsbr && !(status & TD_CTRL_IOC))
uhci_unlink_qh(uhci, qh);
} else {
/* Unmoving but not-yet-expired queues keep FSBR alive */
if (urbp)
uhci_urbp_wants_fsbr(uhci, urbp);
}
done:
return ret;
}
/*
* Process events in the schedule, but only in one thread at a time
*/
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 15:55:46 +02:00
static void uhci_scan_schedule(struct uhci_hcd *uhci)
{
int i;
struct uhci_qh *qh;
/* Don't allow re-entrant calls */
if (uhci->scan_in_progress) {
uhci->need_rescan = 1;
return;
}
uhci->scan_in_progress = 1;
rescan:
uhci->need_rescan = 0;
uhci->fsbr_is_wanted = 0;
uhci_clear_next_interrupt(uhci);
uhci_get_current_frame_number(uhci);
uhci->cur_iso_frame = uhci->frame_number;
/* Go through all the QH queues and process the URBs in each one */
for (i = 0; i < UHCI_NUM_SKELQH - 1; ++i) {
uhci->next_qh = list_entry(uhci->skelqh[i]->node.next,
struct uhci_qh, node);
while ((qh = uhci->next_qh) != uhci->skelqh[i]) {
uhci->next_qh = list_entry(qh->node.next,
struct uhci_qh, node);
if (uhci_advance_check(uhci, qh)) {
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 15:55:46 +02:00
uhci_scan_qh(uhci, qh);
if (qh->state == QH_STATE_ACTIVE) {
uhci_urbp_wants_fsbr(uhci,
list_entry(qh->queue.next, struct urb_priv, node));
}
}
}
}
uhci->last_iso_frame = uhci->cur_iso_frame;
if (uhci->need_rescan)
goto rescan;
uhci->scan_in_progress = 0;
if (uhci->fsbr_is_on && !uhci->fsbr_is_wanted &&
!uhci->fsbr_expiring) {
uhci->fsbr_expiring = 1;
mod_timer(&uhci->fsbr_timer, jiffies + FSBR_OFF_DELAY);
}
if (list_empty(&uhci->skel_unlink_qh->node))
uhci_clear_next_interrupt(uhci);
else
uhci_set_next_interrupt(uhci);
}