linux/drivers/usb/host/ehci-timer.c

427 lines
13 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0+
/*
* Copyright (C) 2012 by Alan Stern
*/
/* This file is part of ehci-hcd.c */
/*-------------------------------------------------------------------------*/
/* Set a bit in the USBCMD register */
static void ehci_set_command_bit(struct ehci_hcd *ehci, u32 bit)
{
ehci->command |= bit;
ehci_writel(ehci, ehci->command, &ehci->regs->command);
/* unblock posted write */
ehci_readl(ehci, &ehci->regs->command);
}
/* Clear a bit in the USBCMD register */
static void ehci_clear_command_bit(struct ehci_hcd *ehci, u32 bit)
{
ehci->command &= ~bit;
ehci_writel(ehci, ehci->command, &ehci->regs->command);
/* unblock posted write */
ehci_readl(ehci, &ehci->regs->command);
}
/*-------------------------------------------------------------------------*/
/*
* EHCI timer support... Now using hrtimers.
*
* Lots of different events are triggered from ehci->hrtimer. Whenever
* the timer routine runs, it checks each possible event; events that are
* currently enabled and whose expiration time has passed get handled.
* The set of enabled events is stored as a collection of bitflags in
* ehci->enabled_hrtimer_events, and they are numbered in order of
* increasing delay values (ranging between 1 ms and 100 ms).
*
* Rather than implementing a sorted list or tree of all pending events,
* we keep track only of the lowest-numbered pending event, in
* ehci->next_hrtimer_event. Whenever ehci->hrtimer gets restarted, its
* expiration time is set to the timeout value for this event.
*
* As a result, events might not get handled right away; the actual delay
* could be anywhere up to twice the requested delay. This doesn't
* matter, because none of the events are especially time-critical. The
* ones that matter most all have a delay of 1 ms, so they will be
* handled after 2 ms at most, which is okay. In addition to this, we
* allow for an expiration range of 1 ms.
*/
/*
* Delay lengths for the hrtimer event types.
* Keep this list sorted by delay length, in the same order as
* the event types indexed by enum ehci_hrtimer_event in ehci.h.
*/
static unsigned event_delays_ns[] = {
1 * NSEC_PER_MSEC, /* EHCI_HRTIMER_POLL_ASS */
1 * NSEC_PER_MSEC, /* EHCI_HRTIMER_POLL_PSS */
1 * NSEC_PER_MSEC, /* EHCI_HRTIMER_POLL_DEAD */
1125 * NSEC_PER_USEC, /* EHCI_HRTIMER_UNLINK_INTR */
2 * NSEC_PER_MSEC, /* EHCI_HRTIMER_FREE_ITDS */
USB: EHCI: add a delay when unlinking an active QH Michael Reutman reports that an AMD/ATI EHCI host controller on one of his computers does not stop transferring data when an active bulk QH is unlinked from the async schedule. Apparently that host controller fails to implement the IAA mechanism correctly when an active QH is unlinked. This leads to data corruption, because the controller continues to update the QH in memory when the driver doesn't expect it. As a result, the next URB submitted for that QH can hang, because the link pointers for the TD queue have been messed up. This misbehavior is observed quite regularly. To be fair, the EHCI spec (section 4.8.2) says that active QHs should not be unlinked. It goes on to recommend a procedure that involves waiting for the QH to go inactive before unlinking it. In the real world this is impractical, not least because the QH may _never_ go inactive. (What were they thinking?) Sometimes we have no choice but to unlink an active QH. In an attempt to avoid the problems that can ensue, this patch changes how the driver decides when the unlink is complete. In addition to waiting through two IAA cycles, in cases where the QH was not known to be inactive beforehand we now wait until a 2-ms period has elapsed with the host controller making no change to the QH data structure (the hw_current and hw_token fields in particular). The intuition here is that after such a long period, the endpoint must be NAKing and hopefully the QH has been dropped from the host controller's internal cache. There's no way to know if this reasoning is really valid -- the spec is no help in this regard -- but at least this approach fixes Michael's problem. The test for whether the QH is already known to be inactive involves the reason for unlinking the QH originally. If it was unlinked because it had halted, or it stopped in response to a short read, or it overlaid a dummy TD (a silicon bug), then it certainly is inactive. If it was unlinked because the TD queue was empty and no TDs have been added to the queue in the meantime, then it must be inactive. Or if the hardware status indicates that the QH is currently halted (even if that wasn't the reason for unlinking it), then it is inactive. Otherwise, if none of those checks apply, we go through the 2-ms delay. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Reported-by: Michael Reutman <mreutman@epiqsolutions.com> Tested-by: Michael Reutman <mreutman@epiqsolutions.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2016-01-25 20:45:25 +00:00
2 * NSEC_PER_MSEC, /* EHCI_HRTIMER_ACTIVE_UNLINK */
5 * NSEC_PER_MSEC, /* EHCI_HRTIMER_START_UNLINK_INTR */
6 * NSEC_PER_MSEC, /* EHCI_HRTIMER_ASYNC_UNLINKS */
10 * NSEC_PER_MSEC, /* EHCI_HRTIMER_IAA_WATCHDOG */
10 * NSEC_PER_MSEC, /* EHCI_HRTIMER_DISABLE_PERIODIC */
15 * NSEC_PER_MSEC, /* EHCI_HRTIMER_DISABLE_ASYNC */
100 * NSEC_PER_MSEC, /* EHCI_HRTIMER_IO_WATCHDOG */
};
/* Enable a pending hrtimer event */
static void ehci_enable_event(struct ehci_hcd *ehci, unsigned event,
bool resched)
{
ktime_t *timeout = &ehci->hr_timeouts[event];
if (resched)
*timeout = ktime_add(ktime_get(), event_delays_ns[event]);
ehci->enabled_hrtimer_events |= (1 << event);
/* Track only the lowest-numbered pending event */
if (event < ehci->next_hrtimer_event) {
ehci->next_hrtimer_event = event;
hrtimer_start_range_ns(&ehci->hrtimer, *timeout,
NSEC_PER_MSEC, HRTIMER_MODE_ABS);
}
}
/* Poll the STS_ASS status bit; see when it agrees with CMD_ASE */
static void ehci_poll_ASS(struct ehci_hcd *ehci)
{
unsigned actual, want;
/* Don't enable anything if the controller isn't running (e.g., died) */
if (ehci->rh_state != EHCI_RH_RUNNING)
return;
want = (ehci->command & CMD_ASE) ? STS_ASS : 0;
actual = ehci_readl(ehci, &ehci->regs->status) & STS_ASS;
if (want != actual) {
/* Poll again later, but give up after about 2-4 ms */
if (ehci->ASS_poll_count++ < 2) {
ehci_enable_event(ehci, EHCI_HRTIMER_POLL_ASS, true);
return;
}
ehci_dbg(ehci, "Waited too long for the async schedule status (%x/%x), giving up\n",
want, actual);
}
ehci->ASS_poll_count = 0;
/* The status is up-to-date; restart or stop the schedule as needed */
if (want == 0) { /* Stopped */
if (ehci->async_count > 0)
ehci_set_command_bit(ehci, CMD_ASE);
} else { /* Running */
if (ehci->async_count == 0) {
/* Turn off the schedule after a while */
ehci_enable_event(ehci, EHCI_HRTIMER_DISABLE_ASYNC,
true);
}
}
}
/* Turn off the async schedule after a brief delay */
static void ehci_disable_ASE(struct ehci_hcd *ehci)
{
ehci_clear_command_bit(ehci, CMD_ASE);
}
/* Poll the STS_PSS status bit; see when it agrees with CMD_PSE */
static void ehci_poll_PSS(struct ehci_hcd *ehci)
{
unsigned actual, want;
/* Don't do anything if the controller isn't running (e.g., died) */
if (ehci->rh_state != EHCI_RH_RUNNING)
return;
want = (ehci->command & CMD_PSE) ? STS_PSS : 0;
actual = ehci_readl(ehci, &ehci->regs->status) & STS_PSS;
if (want != actual) {
/* Poll again later, but give up after about 2-4 ms */
if (ehci->PSS_poll_count++ < 2) {
ehci_enable_event(ehci, EHCI_HRTIMER_POLL_PSS, true);
return;
}
ehci_dbg(ehci, "Waited too long for the periodic schedule status (%x/%x), giving up\n",
want, actual);
}
ehci->PSS_poll_count = 0;
/* The status is up-to-date; restart or stop the schedule as needed */
if (want == 0) { /* Stopped */
if (ehci->periodic_count > 0)
ehci_set_command_bit(ehci, CMD_PSE);
} else { /* Running */
if (ehci->periodic_count == 0) {
/* Turn off the schedule after a while */
ehci_enable_event(ehci, EHCI_HRTIMER_DISABLE_PERIODIC,
true);
}
}
}
/* Turn off the periodic schedule after a brief delay */
static void ehci_disable_PSE(struct ehci_hcd *ehci)
{
ehci_clear_command_bit(ehci, CMD_PSE);
}
/* Poll the STS_HALT status bit; see when a dead controller stops */
static void ehci_handle_controller_death(struct ehci_hcd *ehci)
{
if (!(ehci_readl(ehci, &ehci->regs->status) & STS_HALT)) {
/* Give up after a few milliseconds */
if (ehci->died_poll_count++ < 5) {
/* Try again later */
ehci_enable_event(ehci, EHCI_HRTIMER_POLL_DEAD, true);
return;
}
ehci_warn(ehci, "Waited too long for the controller to stop, giving up\n");
}
/* Clean up the mess */
ehci->rh_state = EHCI_RH_HALTED;
ehci_writel(ehci, 0, &ehci->regs->configured_flag);
ehci_writel(ehci, 0, &ehci->regs->intr_enable);
ehci_work(ehci);
end_unlink_async(ehci);
/* Not in process context, so don't try to reset the controller */
}
/* start to unlink interrupt QHs */
static void ehci_handle_start_intr_unlinks(struct ehci_hcd *ehci)
{
bool stopped = (ehci->rh_state < EHCI_RH_RUNNING);
/*
* Process all the QHs on the intr_unlink list that were added
* before the current unlink cycle began. The list is in
* temporal order, so stop when we reach the first entry in the
* current cycle. But if the root hub isn't running then
* process all the QHs on the list.
*/
while (!list_empty(&ehci->intr_unlink_wait)) {
struct ehci_qh *qh;
qh = list_first_entry(&ehci->intr_unlink_wait,
struct ehci_qh, unlink_node);
if (!stopped && (qh->unlink_cycle ==
ehci->intr_unlink_wait_cycle))
break;
list_del_init(&qh->unlink_node);
qh->unlink_reason |= QH_UNLINK_QUEUE_EMPTY;
start_unlink_intr(ehci, qh);
}
/* Handle remaining entries later */
if (!list_empty(&ehci->intr_unlink_wait)) {
ehci_enable_event(ehci, EHCI_HRTIMER_START_UNLINK_INTR, true);
++ehci->intr_unlink_wait_cycle;
}
}
/* Handle unlinked interrupt QHs once they are gone from the hardware */
static void ehci_handle_intr_unlinks(struct ehci_hcd *ehci)
{
bool stopped = (ehci->rh_state < EHCI_RH_RUNNING);
/*
* Process all the QHs on the intr_unlink list that were added
* before the current unlink cycle began. The list is in
* temporal order, so stop when we reach the first entry in the
* current cycle. But if the root hub isn't running then
* process all the QHs on the list.
*/
ehci->intr_unlinking = true;
while (!list_empty(&ehci->intr_unlink)) {
struct ehci_qh *qh;
qh = list_first_entry(&ehci->intr_unlink, struct ehci_qh,
unlink_node);
if (!stopped && qh->unlink_cycle == ehci->intr_unlink_cycle)
break;
list_del_init(&qh->unlink_node);
end_unlink_intr(ehci, qh);
}
/* Handle remaining entries later */
if (!list_empty(&ehci->intr_unlink)) {
ehci_enable_event(ehci, EHCI_HRTIMER_UNLINK_INTR, true);
++ehci->intr_unlink_cycle;
}
ehci->intr_unlinking = false;
}
/* Start another free-iTDs/siTDs cycle */
static void start_free_itds(struct ehci_hcd *ehci)
{
if (!(ehci->enabled_hrtimer_events & BIT(EHCI_HRTIMER_FREE_ITDS))) {
ehci->last_itd_to_free = list_entry(
ehci->cached_itd_list.prev,
struct ehci_itd, itd_list);
ehci->last_sitd_to_free = list_entry(
ehci->cached_sitd_list.prev,
struct ehci_sitd, sitd_list);
ehci_enable_event(ehci, EHCI_HRTIMER_FREE_ITDS, true);
}
}
/* Wait for controller to stop using old iTDs and siTDs */
static void end_free_itds(struct ehci_hcd *ehci)
{
struct ehci_itd *itd, *n;
struct ehci_sitd *sitd, *sn;
if (ehci->rh_state < EHCI_RH_RUNNING) {
ehci->last_itd_to_free = NULL;
ehci->last_sitd_to_free = NULL;
}
list_for_each_entry_safe(itd, n, &ehci->cached_itd_list, itd_list) {
list_del(&itd->itd_list);
dma_pool_free(ehci->itd_pool, itd, itd->itd_dma);
if (itd == ehci->last_itd_to_free)
break;
}
list_for_each_entry_safe(sitd, sn, &ehci->cached_sitd_list, sitd_list) {
list_del(&sitd->sitd_list);
dma_pool_free(ehci->sitd_pool, sitd, sitd->sitd_dma);
if (sitd == ehci->last_sitd_to_free)
break;
}
if (!list_empty(&ehci->cached_itd_list) ||
!list_empty(&ehci->cached_sitd_list))
start_free_itds(ehci);
}
/* Handle lost (or very late) IAA interrupts */
static void ehci_iaa_watchdog(struct ehci_hcd *ehci)
{
u32 cmd, status;
/*
* Lost IAA irqs wedge things badly; seen first with a vt8235.
* So we need this watchdog, but must protect it against both
* (a) SMP races against real IAA firing and retriggering, and
* (b) clean HC shutdown, when IAA watchdog was pending.
*/
if (!ehci->iaa_in_progress || ehci->rh_state != EHCI_RH_RUNNING)
return;
/* If we get here, IAA is *REALLY* late. It's barely
* conceivable that the system is so busy that CMD_IAAD
* is still legitimately set, so let's be sure it's
* clear before we read STS_IAA. (The HC should clear
* CMD_IAAD when it sets STS_IAA.)
*/
cmd = ehci_readl(ehci, &ehci->regs->command);
/*
* If IAA is set here it either legitimately triggered
* after the watchdog timer expired (_way_ late, so we'll
* still count it as lost) ... or a silicon erratum:
* - VIA seems to set IAA without triggering the IRQ;
* - IAAD potentially cleared without setting IAA.
*/
status = ehci_readl(ehci, &ehci->regs->status);
if ((status & STS_IAA) || !(cmd & CMD_IAAD)) {
INCR(ehci->stats.lost_iaa);
ehci_writel(ehci, STS_IAA, &ehci->regs->status);
}
ehci_dbg(ehci, "IAA watchdog: status %x cmd %x\n", status, cmd);
end_iaa_cycle(ehci);
}
/* Enable the I/O watchdog, if appropriate */
static void turn_on_io_watchdog(struct ehci_hcd *ehci)
{
/* Not needed if the controller isn't running or it's already enabled */
if (ehci->rh_state != EHCI_RH_RUNNING ||
(ehci->enabled_hrtimer_events &
BIT(EHCI_HRTIMER_IO_WATCHDOG)))
return;
/*
* Isochronous transfers always need the watchdog.
* For other sorts we use it only if the flag is set.
*/
if (ehci->isoc_count > 0 || (ehci->need_io_watchdog &&
ehci->async_count + ehci->intr_count > 0))
ehci_enable_event(ehci, EHCI_HRTIMER_IO_WATCHDOG, true);
}
/*
* Handler functions for the hrtimer event types.
* Keep this array in the same order as the event types indexed by
* enum ehci_hrtimer_event in ehci.h.
*/
static void (*event_handlers[])(struct ehci_hcd *) = {
ehci_poll_ASS, /* EHCI_HRTIMER_POLL_ASS */
ehci_poll_PSS, /* EHCI_HRTIMER_POLL_PSS */
ehci_handle_controller_death, /* EHCI_HRTIMER_POLL_DEAD */
ehci_handle_intr_unlinks, /* EHCI_HRTIMER_UNLINK_INTR */
end_free_itds, /* EHCI_HRTIMER_FREE_ITDS */
USB: EHCI: add a delay when unlinking an active QH Michael Reutman reports that an AMD/ATI EHCI host controller on one of his computers does not stop transferring data when an active bulk QH is unlinked from the async schedule. Apparently that host controller fails to implement the IAA mechanism correctly when an active QH is unlinked. This leads to data corruption, because the controller continues to update the QH in memory when the driver doesn't expect it. As a result, the next URB submitted for that QH can hang, because the link pointers for the TD queue have been messed up. This misbehavior is observed quite regularly. To be fair, the EHCI spec (section 4.8.2) says that active QHs should not be unlinked. It goes on to recommend a procedure that involves waiting for the QH to go inactive before unlinking it. In the real world this is impractical, not least because the QH may _never_ go inactive. (What were they thinking?) Sometimes we have no choice but to unlink an active QH. In an attempt to avoid the problems that can ensue, this patch changes how the driver decides when the unlink is complete. In addition to waiting through two IAA cycles, in cases where the QH was not known to be inactive beforehand we now wait until a 2-ms period has elapsed with the host controller making no change to the QH data structure (the hw_current and hw_token fields in particular). The intuition here is that after such a long period, the endpoint must be NAKing and hopefully the QH has been dropped from the host controller's internal cache. There's no way to know if this reasoning is really valid -- the spec is no help in this regard -- but at least this approach fixes Michael's problem. The test for whether the QH is already known to be inactive involves the reason for unlinking the QH originally. If it was unlinked because it had halted, or it stopped in response to a short read, or it overlaid a dummy TD (a silicon bug), then it certainly is inactive. If it was unlinked because the TD queue was empty and no TDs have been added to the queue in the meantime, then it must be inactive. Or if the hardware status indicates that the QH is currently halted (even if that wasn't the reason for unlinking it), then it is inactive. Otherwise, if none of those checks apply, we go through the 2-ms delay. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Reported-by: Michael Reutman <mreutman@epiqsolutions.com> Tested-by: Michael Reutman <mreutman@epiqsolutions.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2016-01-25 20:45:25 +00:00
end_unlink_async, /* EHCI_HRTIMER_ACTIVE_UNLINK */
ehci_handle_start_intr_unlinks, /* EHCI_HRTIMER_START_UNLINK_INTR */
unlink_empty_async, /* EHCI_HRTIMER_ASYNC_UNLINKS */
ehci_iaa_watchdog, /* EHCI_HRTIMER_IAA_WATCHDOG */
ehci_disable_PSE, /* EHCI_HRTIMER_DISABLE_PERIODIC */
ehci_disable_ASE, /* EHCI_HRTIMER_DISABLE_ASYNC */
ehci_work, /* EHCI_HRTIMER_IO_WATCHDOG */
};
static enum hrtimer_restart ehci_hrtimer_func(struct hrtimer *t)
{
struct ehci_hcd *ehci = container_of(t, struct ehci_hcd, hrtimer);
ktime_t now;
unsigned long events;
unsigned long flags;
unsigned e;
spin_lock_irqsave(&ehci->lock, flags);
events = ehci->enabled_hrtimer_events;
ehci->enabled_hrtimer_events = 0;
ehci->next_hrtimer_event = EHCI_HRTIMER_NO_EVENT;
/*
* Check each pending event. If its time has expired, handle
* the event; otherwise re-enable it.
*/
now = ktime_get();
for_each_set_bit(e, &events, EHCI_HRTIMER_NUM_EVENTS) {
if (ktime_compare(now, ehci->hr_timeouts[e]) >= 0)
event_handlers[e](ehci);
else
ehci_enable_event(ehci, e, false);
}
spin_unlock_irqrestore(&ehci->lock, flags);
return HRTIMER_NORESTART;
}