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

1236 lines
34 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-1.0+
/*
* OHCI HCD (Host Controller Driver) for USB.
*
* (C) Copyright 1999 Roman Weissgaerber <weissg@vienna.at>
* (C) Copyright 2000-2002 David Brownell <dbrownell@users.sourceforge.net>
*
* This file is licenced under the GPL.
*/
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 13:55:46 +00:00
#include <linux/irq.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 08:04:11 +00:00
#include <linux/slab.h>
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 13:55:46 +00:00
static void urb_free_priv (struct ohci_hcd *hc, urb_priv_t *urb_priv)
{
int last = urb_priv->length - 1;
if (last >= 0) {
int i;
struct td *td;
for (i = 0; i <= last; i++) {
td = urb_priv->td [i];
if (td)
td_free (hc, td);
}
}
list_del (&urb_priv->pending);
kfree (urb_priv);
}
/*-------------------------------------------------------------------------*/
/*
* URB goes back to driver, and isn't reissued.
* It's completely gone from HC data structures.
* PRECONDITION: ohci lock held, irqs blocked.
*/
static void
finish_urb(struct ohci_hcd *ohci, struct urb *urb, int status)
__releases(ohci->lock)
__acquires(ohci->lock)
{
USB: OHCI: accept very late isochronous URBs Commit 24f531371de1 (USB: EHCI: accept very late isochronous URBs) changed the isochronous API provided by ehci-hcd. URBs submitted too late, so that the time slots for all their packets have already expired, are no longer rejected outright. Instead the submission is accepted, and the URB completes normally with a -EXDEV error for each packet. This is what client drivers expect. This patch implements the same policy in ohci-hcd. The change is more complicated than it was in ehci-hcd, because ohci-hcd doesn't scan for isochronous completions in the same way as ehci-hcd does. Rather, it depends on the hardware adding completed TDs to a "done queue". Some OHCI controller don't handle this properly when a TD's time slot has already expired, so we have to avoid adding such TDs to the schedule in the first place. As a result, if the URB was submitted too late then none of its TDs will get put on the schedule, so none of them will end up on the done queue, so the driver will never realize that the URB should be completed. To solve this problem, the patch adds one to urb_priv->td_cnt for such URBs, making it larger than urb_priv->length (td_cnt already gets set to the number of TD's that had to be skipped because their slots have expired). Each time an URB is given back, the finish_urb() routine looks to see if urb_priv->td_cnt for the next URB on the same endpoint is marked in this way. If so, it gives back the next URB right away. This should be applied to all kernels containing commit 815fa7b91761 (USB: OHCI: fix logic for scheduling isochronous URBs). Signed-off-by: Alan Stern <stern@rowland.harvard.edu> CC: <stable@vger.kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-09-24 19:46:45 +00:00
struct device *dev = ohci_to_hcd(ohci)->self.controller;
struct usb_host_endpoint *ep = urb->ep;
struct urb_priv *urb_priv;
// ASSERT (urb->hcpriv != 0);
USB: OHCI: accept very late isochronous URBs Commit 24f531371de1 (USB: EHCI: accept very late isochronous URBs) changed the isochronous API provided by ehci-hcd. URBs submitted too late, so that the time slots for all their packets have already expired, are no longer rejected outright. Instead the submission is accepted, and the URB completes normally with a -EXDEV error for each packet. This is what client drivers expect. This patch implements the same policy in ohci-hcd. The change is more complicated than it was in ehci-hcd, because ohci-hcd doesn't scan for isochronous completions in the same way as ehci-hcd does. Rather, it depends on the hardware adding completed TDs to a "done queue". Some OHCI controller don't handle this properly when a TD's time slot has already expired, so we have to avoid adding such TDs to the schedule in the first place. As a result, if the URB was submitted too late then none of its TDs will get put on the schedule, so none of them will end up on the done queue, so the driver will never realize that the URB should be completed. To solve this problem, the patch adds one to urb_priv->td_cnt for such URBs, making it larger than urb_priv->length (td_cnt already gets set to the number of TD's that had to be skipped because their slots have expired). Each time an URB is given back, the finish_urb() routine looks to see if urb_priv->td_cnt for the next URB on the same endpoint is marked in this way. If so, it gives back the next URB right away. This should be applied to all kernels containing commit 815fa7b91761 (USB: OHCI: fix logic for scheduling isochronous URBs). Signed-off-by: Alan Stern <stern@rowland.harvard.edu> CC: <stable@vger.kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-09-24 19:46:45 +00:00
restart:
urb_free_priv (ohci, urb->hcpriv);
urb->hcpriv = NULL;
if (likely(status == -EINPROGRESS))
status = 0;
switch (usb_pipetype (urb->pipe)) {
case PIPE_ISOCHRONOUS:
ohci_to_hcd(ohci)->self.bandwidth_isoc_reqs--;
if (ohci_to_hcd(ohci)->self.bandwidth_isoc_reqs == 0) {
if (quirk_amdiso(ohci))
USB host: Move AMD PLL quirk to pci-quirks.c This patch moves the AMD PLL quirk code in OHCI/EHCI driver to pci-quirks.c, and exports the functions to be used by xHCI driver later. AMD PLL quirk disable the optional PM feature inside specific SB700/SB800/Hudson-2/3 platforms under the following conditions: 1. If an isochronous device is connected to OHCI/EHCI/xHCI port and is active; 2. Optional PM feature that powers down the internal Bus PLL when the link is in low power state is enabled. Without AMD PLL quirk, USB isochronous stream may stutter or have breaks occasionally, which greatly impair the performance of audio/video streams. Currently AMD PLL quirk is implemented in OHCI and EHCI driver, and will be added to xHCI driver too. They are doing similar things actually, so move the quirk code to pci-quirks.c, which has several advantages: 1. Remove duplicate defines and functions in OHCI/EHCI (and xHCI) driver and make them cleaner; 2. AMD chipset information will be probed only once and then stored. Currently they're probed during every OHCI/EHCI initialization, move the detect code to pci-quirks.c saves the repeat detect cost; 3. Build up synchronization among OHCI/EHCI/xHCI driver. In current code, every host controller enable/disable PLL only according to its own status, and may enable PLL while there is still isoc transfer on other HCs. Move the quirk to pci-quirks.c prevents this issue. Signed-off-by: Andiry Xu <andiry.xu@amd.com> Cc: David Brownell <dbrownell@users.sourceforge.net> Cc: Alex He <alex.he@amd.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2011-03-01 06:57:05 +00:00
usb_amd_quirk_pll_enable();
if (quirk_amdprefetch(ohci))
sb800_prefetch(dev, 0);
}
break;
case PIPE_INTERRUPT:
ohci_to_hcd(ohci)->self.bandwidth_int_reqs--;
break;
}
/* urb->complete() can reenter this HCD */
usb_hcd_unlink_urb_from_ep(ohci_to_hcd(ohci), urb);
spin_unlock (&ohci->lock);
usb_hcd_giveback_urb(ohci_to_hcd(ohci), urb, status);
spin_lock (&ohci->lock);
/* stop periodic dma if it's not needed */
if (ohci_to_hcd(ohci)->self.bandwidth_isoc_reqs == 0
&& ohci_to_hcd(ohci)->self.bandwidth_int_reqs == 0) {
ohci->hc_control &= ~(OHCI_CTRL_PLE|OHCI_CTRL_IE);
ohci_writel (ohci, ohci->hc_control, &ohci->regs->control);
}
USB: OHCI: accept very late isochronous URBs Commit 24f531371de1 (USB: EHCI: accept very late isochronous URBs) changed the isochronous API provided by ehci-hcd. URBs submitted too late, so that the time slots for all their packets have already expired, are no longer rejected outright. Instead the submission is accepted, and the URB completes normally with a -EXDEV error for each packet. This is what client drivers expect. This patch implements the same policy in ohci-hcd. The change is more complicated than it was in ehci-hcd, because ohci-hcd doesn't scan for isochronous completions in the same way as ehci-hcd does. Rather, it depends on the hardware adding completed TDs to a "done queue". Some OHCI controller don't handle this properly when a TD's time slot has already expired, so we have to avoid adding such TDs to the schedule in the first place. As a result, if the URB was submitted too late then none of its TDs will get put on the schedule, so none of them will end up on the done queue, so the driver will never realize that the URB should be completed. To solve this problem, the patch adds one to urb_priv->td_cnt for such URBs, making it larger than urb_priv->length (td_cnt already gets set to the number of TD's that had to be skipped because their slots have expired). Each time an URB is given back, the finish_urb() routine looks to see if urb_priv->td_cnt for the next URB on the same endpoint is marked in this way. If so, it gives back the next URB right away. This should be applied to all kernels containing commit 815fa7b91761 (USB: OHCI: fix logic for scheduling isochronous URBs). Signed-off-by: Alan Stern <stern@rowland.harvard.edu> CC: <stable@vger.kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-09-24 19:46:45 +00:00
/*
* An isochronous URB that is sumitted too late won't have any TDs
* (marked by the fact that the td_cnt value is larger than the
* actual number of TDs). If the next URB on this endpoint is like
* that, give it back now.
*/
if (!list_empty(&ep->urb_list)) {
urb = list_first_entry(&ep->urb_list, struct urb, urb_list);
urb_priv = urb->hcpriv;
if (urb_priv->td_cnt > urb_priv->length) {
status = 0;
goto restart;
}
}
}
/*-------------------------------------------------------------------------*
* ED handling functions
*-------------------------------------------------------------------------*/
/* search for the right schedule branch to use for a periodic ed.
* does some load balancing; returns the branch, or negative errno.
*/
static int balance (struct ohci_hcd *ohci, int interval, int load)
{
int i, branch = -ENOSPC;
/* iso periods can be huge; iso tds specify frame numbers */
if (interval > NUM_INTS)
interval = NUM_INTS;
/* search for the least loaded schedule branch of that period
* that has enough bandwidth left unreserved.
*/
for (i = 0; i < interval ; i++) {
if (branch < 0 || ohci->load [branch] > ohci->load [i]) {
int j;
/* usb 1.1 says 90% of one frame */
for (j = i; j < NUM_INTS; j += interval) {
if ((ohci->load [j] + load) > 900)
break;
}
if (j < NUM_INTS)
continue;
branch = i;
}
}
return branch;
}
/*-------------------------------------------------------------------------*/
/* both iso and interrupt requests have periods; this routine puts them
* into the schedule tree in the apppropriate place. most iso devices use
* 1msec periods, but that's not required.
*/
static void periodic_link (struct ohci_hcd *ohci, struct ed *ed)
{
unsigned i;
ohci_dbg(ohci, "link %sed %p branch %d [%dus.], interval %d\n",
(ed->hwINFO & cpu_to_hc32 (ohci, ED_ISO)) ? "iso " : "",
ed, ed->branch, ed->load, ed->interval);
for (i = ed->branch; i < NUM_INTS; i += ed->interval) {
struct ed **prev = &ohci->periodic [i];
__hc32 *prev_p = &ohci->hcca->int_table [i];
struct ed *here = *prev;
/* sorting each branch by period (slow before fast)
* lets us share the faster parts of the tree.
* (plus maybe: put interrupt eds before iso)
*/
while (here && ed != here) {
if (ed->interval > here->interval)
break;
prev = &here->ed_next;
prev_p = &here->hwNextED;
here = *prev;
}
if (ed != here) {
ed->ed_next = here;
if (here)
ed->hwNextED = *prev_p;
wmb ();
*prev = ed;
*prev_p = cpu_to_hc32(ohci, ed->dma);
wmb();
}
ohci->load [i] += ed->load;
}
ohci_to_hcd(ohci)->self.bandwidth_allocated += ed->load / ed->interval;
}
/* link an ed into one of the HC chains */
static int ed_schedule (struct ohci_hcd *ohci, struct ed *ed)
{
int branch;
ed->ed_prev = NULL;
ed->ed_next = NULL;
ed->hwNextED = 0;
wmb ();
/* we care about rm_list when setting CLE/BLE in case the HC was at
* work on some TD when CLE/BLE was turned off, and isn't quiesced
* yet. finish_unlinks() restarts as needed, some upcoming INTR_SF.
*
* control and bulk EDs are doubly linked (ed_next, ed_prev), but
* periodic ones are singly linked (ed_next). that's because the
* periodic schedule encodes a tree like figure 3-5 in the ohci
* spec: each qh can have several "previous" nodes, and the tree
* doesn't have unused/idle descriptors.
*/
switch (ed->type) {
case PIPE_CONTROL:
if (ohci->ed_controltail == NULL) {
WARN_ON (ohci->hc_control & OHCI_CTRL_CLE);
ohci_writel (ohci, ed->dma,
&ohci->regs->ed_controlhead);
} else {
ohci->ed_controltail->ed_next = ed;
ohci->ed_controltail->hwNextED = cpu_to_hc32 (ohci,
ed->dma);
}
ed->ed_prev = ohci->ed_controltail;
if (!ohci->ed_controltail && !ohci->ed_rm_list) {
wmb();
ohci->hc_control |= OHCI_CTRL_CLE;
ohci_writel (ohci, 0, &ohci->regs->ed_controlcurrent);
ohci_writel (ohci, ohci->hc_control,
&ohci->regs->control);
}
ohci->ed_controltail = ed;
break;
case PIPE_BULK:
if (ohci->ed_bulktail == NULL) {
WARN_ON (ohci->hc_control & OHCI_CTRL_BLE);
ohci_writel (ohci, ed->dma, &ohci->regs->ed_bulkhead);
} else {
ohci->ed_bulktail->ed_next = ed;
ohci->ed_bulktail->hwNextED = cpu_to_hc32 (ohci,
ed->dma);
}
ed->ed_prev = ohci->ed_bulktail;
if (!ohci->ed_bulktail && !ohci->ed_rm_list) {
wmb();
ohci->hc_control |= OHCI_CTRL_BLE;
ohci_writel (ohci, 0, &ohci->regs->ed_bulkcurrent);
ohci_writel (ohci, ohci->hc_control,
&ohci->regs->control);
}
ohci->ed_bulktail = ed;
break;
// case PIPE_INTERRUPT:
// case PIPE_ISOCHRONOUS:
default:
branch = balance (ohci, ed->interval, ed->load);
if (branch < 0) {
ohci_dbg (ohci,
"ERR %d, interval %d msecs, load %d\n",
branch, ed->interval, ed->load);
// FIXME if there are TDs queued, fail them!
return branch;
}
ed->branch = branch;
periodic_link (ohci, ed);
}
/* the HC may not see the schedule updates yet, but if it does
* then they'll be properly ordered.
*/
ed->state = ED_OPER;
return 0;
}
/*-------------------------------------------------------------------------*/
/* scan the periodic table to find and unlink this ED */
static void periodic_unlink (struct ohci_hcd *ohci, struct ed *ed)
{
int i;
for (i = ed->branch; i < NUM_INTS; i += ed->interval) {
struct ed *temp;
struct ed **prev = &ohci->periodic [i];
__hc32 *prev_p = &ohci->hcca->int_table [i];
while (*prev && (temp = *prev) != ed) {
prev_p = &temp->hwNextED;
prev = &temp->ed_next;
}
if (*prev) {
*prev_p = ed->hwNextED;
*prev = ed->ed_next;
}
ohci->load [i] -= ed->load;
}
ohci_to_hcd(ohci)->self.bandwidth_allocated -= ed->load / ed->interval;
ohci_dbg(ohci, "unlink %sed %p branch %d [%dus.], interval %d\n",
(ed->hwINFO & cpu_to_hc32 (ohci, ED_ISO)) ? "iso " : "",
ed, ed->branch, ed->load, ed->interval);
}
/* unlink an ed from one of the HC chains.
* just the link to the ed is unlinked.
* the link from the ed still points to another operational ed or 0
* so the HC can eventually finish the processing of the unlinked ed
* (assuming it already started that, which needn't be true).
*
* ED_UNLINK is a transient state: the HC may still see this ED, but soon
* it won't. ED_SKIP means the HC will finish its current transaction,
* but won't start anything new. The TD queue may still grow; device
* drivers don't know about this HCD-internal state.
*
* When the HC can't see the ED, something changes ED_UNLINK to one of:
*
* - ED_OPER: when there's any request queued, the ED gets rescheduled
* immediately. HC should be working on them.
*
* - ED_IDLE: when there's no TD queue or the HC isn't running.
*
* When finish_unlinks() runs later, after SOF interrupt, it will often
* complete one or more URB unlinks before making that state change.
*/
static void ed_deschedule (struct ohci_hcd *ohci, struct ed *ed)
{
ed->hwINFO |= cpu_to_hc32 (ohci, ED_SKIP);
wmb ();
ed->state = ED_UNLINK;
/* To deschedule something from the control or bulk list, just
* clear CLE/BLE and wait. There's no safe way to scrub out list
* head/current registers until later, and "later" isn't very
* tightly specified. Figure 6-5 and Section 6.4.2.2 show how
* the HC is reading the ED queues (while we modify them).
*
* For now, ed_schedule() is "later". It might be good paranoia
* to scrub those registers in finish_unlinks(), in case of bugs
* that make the HC try to use them.
*/
switch (ed->type) {
case PIPE_CONTROL:
/* remove ED from the HC's list: */
if (ed->ed_prev == NULL) {
if (!ed->hwNextED) {
ohci->hc_control &= ~OHCI_CTRL_CLE;
ohci_writel (ohci, ohci->hc_control,
&ohci->regs->control);
// a ohci_readl() later syncs CLE with the HC
} else
ohci_writel (ohci,
hc32_to_cpup (ohci, &ed->hwNextED),
&ohci->regs->ed_controlhead);
} else {
ed->ed_prev->ed_next = ed->ed_next;
ed->ed_prev->hwNextED = ed->hwNextED;
}
/* remove ED from the HCD's list: */
if (ohci->ed_controltail == ed) {
ohci->ed_controltail = ed->ed_prev;
if (ohci->ed_controltail)
ohci->ed_controltail->ed_next = NULL;
} else if (ed->ed_next) {
ed->ed_next->ed_prev = ed->ed_prev;
}
break;
case PIPE_BULK:
/* remove ED from the HC's list: */
if (ed->ed_prev == NULL) {
if (!ed->hwNextED) {
ohci->hc_control &= ~OHCI_CTRL_BLE;
ohci_writel (ohci, ohci->hc_control,
&ohci->regs->control);
// a ohci_readl() later syncs BLE with the HC
} else
ohci_writel (ohci,
hc32_to_cpup (ohci, &ed->hwNextED),
&ohci->regs->ed_bulkhead);
} else {
ed->ed_prev->ed_next = ed->ed_next;
ed->ed_prev->hwNextED = ed->hwNextED;
}
/* remove ED from the HCD's list: */
if (ohci->ed_bulktail == ed) {
ohci->ed_bulktail = ed->ed_prev;
if (ohci->ed_bulktail)
ohci->ed_bulktail->ed_next = NULL;
} else if (ed->ed_next) {
ed->ed_next->ed_prev = ed->ed_prev;
}
break;
// case PIPE_INTERRUPT:
// case PIPE_ISOCHRONOUS:
default:
periodic_unlink (ohci, ed);
break;
}
}
/*-------------------------------------------------------------------------*/
/* get and maybe (re)init an endpoint. init _should_ be done only as part
* of enumeration, usb_set_configuration() or usb_set_interface().
*/
static struct ed *ed_get (
struct ohci_hcd *ohci,
struct usb_host_endpoint *ep,
struct usb_device *udev,
unsigned int pipe,
int interval
) {
struct ed *ed;
unsigned long flags;
spin_lock_irqsave (&ohci->lock, flags);
ed = ep->hcpriv;
if (!ed) {
struct td *td;
int is_out;
u32 info;
ed = ed_alloc (ohci, GFP_ATOMIC);
if (!ed) {
/* out of memory */
goto done;
}
/* dummy td; end of td list for ed */
td = td_alloc (ohci, GFP_ATOMIC);
if (!td) {
/* out of memory */
ed_free (ohci, ed);
ed = NULL;
goto done;
}
ed->dummy = td;
ed->hwTailP = cpu_to_hc32 (ohci, td->td_dma);
ed->hwHeadP = ed->hwTailP; /* ED_C, ED_H zeroed */
ed->state = ED_IDLE;
is_out = !(ep->desc.bEndpointAddress & USB_DIR_IN);
/* FIXME usbcore changes dev->devnum before SET_ADDRESS
* succeeds ... otherwise we wouldn't need "pipe".
*/
info = usb_pipedevice (pipe);
ed->type = usb_pipetype(pipe);
info |= (ep->desc.bEndpointAddress & ~USB_DIR_IN) << 7;
USB: use usb_endpoint_maxp() instead of le16_to_cpu() Now ${LINUX}/drivers/usb/* can use usb_endpoint_maxp(desc) to get maximum packet size instead of le16_to_cpu(desc->wMaxPacketSize). This patch fix it up Cc: Armin Fuerst <fuerst@in.tum.de> Cc: Pavel Machek <pavel@ucw.cz> Cc: Johannes Erdfelt <johannes@erdfelt.com> Cc: Vojtech Pavlik <vojtech@suse.cz> Cc: Oliver Neukum <oliver@neukum.name> Cc: David Kubicek <dave@awk.cz> Cc: Johan Hovold <jhovold@gmail.com> Cc: Brad Hards <bhards@bigpond.net.au> Acked-by: Felipe Balbi <balbi@ti.com> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Thomas Dahlmann <dahlmann.thomas@arcor.de> Cc: David Brownell <david-b@pacbell.net> Cc: David Lopo <dlopo@chipidea.mips.com> Cc: Alan Stern <stern@rowland.harvard.edu> Cc: Michal Nazarewicz <m.nazarewicz@samsung.com> Cc: Xie Xiaobo <X.Xie@freescale.com> Cc: Li Yang <leoli@freescale.com> Cc: Jiang Bo <tanya.jiang@freescale.com> Cc: Yuan-hsin Chen <yhchen@faraday-tech.com> Cc: Darius Augulis <augulis.darius@gmail.com> Cc: Xiaochen Shen <xiaochen.shen@intel.com> Cc: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com> Cc: OKI SEMICONDUCTOR, <toshiharu-linux@dsn.okisemi.com> Cc: Robert Jarzmik <robert.jarzmik@free.fr> Cc: Ben Dooks <ben@simtec.co.uk> Cc: Thomas Abraham <thomas.ab@samsung.com> Cc: Herbert Pötzl <herbert@13thfloor.at> Cc: Arnaud Patard <arnaud.patard@rtp-net.org> Cc: Roman Weissgaerber <weissg@vienna.at> Acked-by: Sarah Sharp <sarah.a.sharp@linux.intel.com> Cc: Tony Olech <tony.olech@elandigitalsystems.com> Cc: Florian Floe Echtler <echtler@fs.tum.de> Cc: Christian Lucht <lucht@codemercs.com> Cc: Juergen Stuber <starblue@sourceforge.net> Cc: Georges Toth <g.toth@e-biz.lu> Cc: Bill Ryder <bryder@sgi.com> Cc: Kuba Ober <kuba@mareimbrium.org> Cc: Inaky Perez-Gonzalez <inaky.perez-gonzalez@intel.com> Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2011-08-23 10:12:03 +00:00
info |= usb_endpoint_maxp(&ep->desc) << 16;
if (udev->speed == USB_SPEED_LOW)
info |= ED_LOWSPEED;
/* only control transfers store pids in tds */
if (ed->type != PIPE_CONTROL) {
info |= is_out ? ED_OUT : ED_IN;
if (ed->type != PIPE_BULK) {
/* periodic transfers... */
if (ed->type == PIPE_ISOCHRONOUS)
info |= ED_ISO;
else if (interval > 32) /* iso can be bigger */
interval = 32;
ed->interval = interval;
ed->load = usb_calc_bus_time (
udev->speed, !is_out,
ed->type == PIPE_ISOCHRONOUS,
USB: use usb_endpoint_maxp() instead of le16_to_cpu() Now ${LINUX}/drivers/usb/* can use usb_endpoint_maxp(desc) to get maximum packet size instead of le16_to_cpu(desc->wMaxPacketSize). This patch fix it up Cc: Armin Fuerst <fuerst@in.tum.de> Cc: Pavel Machek <pavel@ucw.cz> Cc: Johannes Erdfelt <johannes@erdfelt.com> Cc: Vojtech Pavlik <vojtech@suse.cz> Cc: Oliver Neukum <oliver@neukum.name> Cc: David Kubicek <dave@awk.cz> Cc: Johan Hovold <jhovold@gmail.com> Cc: Brad Hards <bhards@bigpond.net.au> Acked-by: Felipe Balbi <balbi@ti.com> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Thomas Dahlmann <dahlmann.thomas@arcor.de> Cc: David Brownell <david-b@pacbell.net> Cc: David Lopo <dlopo@chipidea.mips.com> Cc: Alan Stern <stern@rowland.harvard.edu> Cc: Michal Nazarewicz <m.nazarewicz@samsung.com> Cc: Xie Xiaobo <X.Xie@freescale.com> Cc: Li Yang <leoli@freescale.com> Cc: Jiang Bo <tanya.jiang@freescale.com> Cc: Yuan-hsin Chen <yhchen@faraday-tech.com> Cc: Darius Augulis <augulis.darius@gmail.com> Cc: Xiaochen Shen <xiaochen.shen@intel.com> Cc: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com> Cc: OKI SEMICONDUCTOR, <toshiharu-linux@dsn.okisemi.com> Cc: Robert Jarzmik <robert.jarzmik@free.fr> Cc: Ben Dooks <ben@simtec.co.uk> Cc: Thomas Abraham <thomas.ab@samsung.com> Cc: Herbert Pötzl <herbert@13thfloor.at> Cc: Arnaud Patard <arnaud.patard@rtp-net.org> Cc: Roman Weissgaerber <weissg@vienna.at> Acked-by: Sarah Sharp <sarah.a.sharp@linux.intel.com> Cc: Tony Olech <tony.olech@elandigitalsystems.com> Cc: Florian Floe Echtler <echtler@fs.tum.de> Cc: Christian Lucht <lucht@codemercs.com> Cc: Juergen Stuber <starblue@sourceforge.net> Cc: Georges Toth <g.toth@e-biz.lu> Cc: Bill Ryder <bryder@sgi.com> Cc: Kuba Ober <kuba@mareimbrium.org> Cc: Inaky Perez-Gonzalez <inaky.perez-gonzalez@intel.com> Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2011-08-23 10:12:03 +00:00
usb_endpoint_maxp(&ep->desc))
/ 1000;
}
}
ed->hwINFO = cpu_to_hc32(ohci, info);
ep->hcpriv = ed;
}
done:
spin_unlock_irqrestore (&ohci->lock, flags);
return ed;
}
/*-------------------------------------------------------------------------*/
/* request unlinking of an endpoint from an operational HC.
* put the ep on the rm_list
* real work is done at the next start frame (SF) hardware interrupt
* caller guarantees HCD is running, so hardware access is safe,
* and that ed->state is ED_OPER
*/
static void start_ed_unlink (struct ohci_hcd *ohci, struct ed *ed)
{
ed->hwINFO |= cpu_to_hc32 (ohci, ED_DEQUEUE);
ed_deschedule (ohci, ed);
/* rm_list is just singly linked, for simplicity */
ed->ed_next = ohci->ed_rm_list;
ed->ed_prev = NULL;
ohci->ed_rm_list = ed;
/* enable SOF interrupt */
ohci_writel (ohci, OHCI_INTR_SF, &ohci->regs->intrstatus);
ohci_writel (ohci, OHCI_INTR_SF, &ohci->regs->intrenable);
// flush those writes, and get latest HCCA contents
(void) ohci_readl (ohci, &ohci->regs->control);
/* SF interrupt might get delayed; record the frame counter value that
* indicates when the HC isn't looking at it, so concurrent unlinks
* behave. frame_no wraps every 2^16 msec, and changes right before
* SF is triggered.
*/
ed->tick = ohci_frame_no(ohci) + 1;
}
/*-------------------------------------------------------------------------*
* TD handling functions
*-------------------------------------------------------------------------*/
/* enqueue next TD for this URB (OHCI spec 5.2.8.2) */
static void
td_fill (struct ohci_hcd *ohci, u32 info,
dma_addr_t data, int len,
struct urb *urb, int index)
{
struct td *td, *td_pt;
struct urb_priv *urb_priv = urb->hcpriv;
int is_iso = info & TD_ISO;
int hash;
// ASSERT (index < urb_priv->length);
/* aim for only one interrupt per urb. mostly applies to control
* and iso; other urbs rarely need more than one TD per urb.
* this way, only final tds (or ones with an error) cause IRQs.
* at least immediately; use DI=6 in case any control request is
* tempted to die part way through. (and to force the hc to flush
* its donelist soonish, even on unlink paths.)
*
* NOTE: could delay interrupts even for the last TD, and get fewer
* interrupts ... increasing per-urb latency by sharing interrupts.
* Drivers that queue bulk urbs may request that behavior.
*/
if (index != (urb_priv->length - 1)
|| (urb->transfer_flags & URB_NO_INTERRUPT))
info |= TD_DI_SET (6);
/* use this td as the next dummy */
td_pt = urb_priv->td [index];
/* fill the old dummy TD */
td = urb_priv->td [index] = urb_priv->ed->dummy;
urb_priv->ed->dummy = td_pt;
td->ed = urb_priv->ed;
td->next_dl_td = NULL;
td->index = index;
td->urb = urb;
td->data_dma = data;
if (!len)
data = 0;
td->hwINFO = cpu_to_hc32 (ohci, info);
if (is_iso) {
td->hwCBP = cpu_to_hc32 (ohci, data & 0xFFFFF000);
*ohci_hwPSWp(ohci, td, 0) = cpu_to_hc16 (ohci,
(data & 0x0FFF) | 0xE000);
} else {
td->hwCBP = cpu_to_hc32 (ohci, data);
}
if (data)
td->hwBE = cpu_to_hc32 (ohci, data + len - 1);
else
td->hwBE = 0;
td->hwNextTD = cpu_to_hc32 (ohci, td_pt->td_dma);
/* append to queue */
list_add_tail (&td->td_list, &td->ed->td_list);
/* hash it for later reverse mapping */
hash = TD_HASH_FUNC (td->td_dma);
td->td_hash = ohci->td_hash [hash];
ohci->td_hash [hash] = td;
/* HC might read the TD (or cachelines) right away ... */
wmb ();
td->ed->hwTailP = td->hwNextTD;
}
/*-------------------------------------------------------------------------*/
/* Prepare all TDs of a transfer, and queue them onto the ED.
* Caller guarantees HC is active.
* Usually the ED is already on the schedule, so TDs might be
* processed as soon as they're queued.
*/
static void td_submit_urb (
struct ohci_hcd *ohci,
struct urb *urb
) {
struct urb_priv *urb_priv = urb->hcpriv;
struct device *dev = ohci_to_hcd(ohci)->self.controller;
dma_addr_t data;
int data_len = urb->transfer_buffer_length;
int cnt = 0;
u32 info = 0;
int is_out = usb_pipeout (urb->pipe);
int periodic = 0;
int i, this_sg_len, n;
struct scatterlist *sg;
/* OHCI handles the bulk/interrupt data toggles itself. We just
* use the device toggle bits for resetting, and rely on the fact
* that resetting toggle is meaningless if the endpoint is active.
*/
if (!usb_gettoggle (urb->dev, usb_pipeendpoint (urb->pipe), is_out)) {
usb_settoggle (urb->dev, usb_pipeendpoint (urb->pipe),
is_out, 1);
urb_priv->ed->hwHeadP &= ~cpu_to_hc32 (ohci, ED_C);
}
list_add (&urb_priv->pending, &ohci->pending);
i = urb->num_mapped_sgs;
if (data_len > 0 && i > 0) {
sg = urb->sg;
data = sg_dma_address(sg);
/*
* urb->transfer_buffer_length may be smaller than the
* size of the scatterlist (or vice versa)
*/
this_sg_len = min_t(int, sg_dma_len(sg), data_len);
} else {
sg = NULL;
if (data_len)
data = urb->transfer_dma;
else
data = 0;
this_sg_len = data_len;
}
/* NOTE: TD_CC is set so we can tell which TDs the HC processed by
* using TD_CC_GET, as well as by seeing them on the done list.
* (CC = NotAccessed ... 0x0F, or 0x0E in PSWs for ISO.)
*/
switch (urb_priv->ed->type) {
/* Bulk and interrupt are identical except for where in the schedule
* their EDs live.
*/
case PIPE_INTERRUPT:
/* ... and periodic urbs have extra accounting */
periodic = ohci_to_hcd(ohci)->self.bandwidth_int_reqs++ == 0
&& ohci_to_hcd(ohci)->self.bandwidth_isoc_reqs == 0;
/* FALLTHROUGH */
case PIPE_BULK:
info = is_out
? TD_T_TOGGLE | TD_CC | TD_DP_OUT
: TD_T_TOGGLE | TD_CC | TD_DP_IN;
/* TDs _could_ transfer up to 8K each */
for (;;) {
n = min(this_sg_len, 4096);
/* maybe avoid ED halt on final TD short read */
if (n >= data_len || (i == 1 && n >= this_sg_len)) {
if (!(urb->transfer_flags & URB_SHORT_NOT_OK))
info |= TD_R;
}
td_fill(ohci, info, data, n, urb, cnt);
this_sg_len -= n;
data_len -= n;
data += n;
cnt++;
if (this_sg_len <= 0) {
if (--i <= 0 || data_len <= 0)
break;
sg = sg_next(sg);
data = sg_dma_address(sg);
this_sg_len = min_t(int, sg_dma_len(sg),
data_len);
}
}
if ((urb->transfer_flags & URB_ZERO_PACKET)
&& cnt < urb_priv->length) {
td_fill (ohci, info, 0, 0, urb, cnt);
cnt++;
}
/* maybe kickstart bulk list */
if (urb_priv->ed->type == PIPE_BULK) {
wmb ();
ohci_writel (ohci, OHCI_BLF, &ohci->regs->cmdstatus);
}
break;
/* control manages DATA0/DATA1 toggle per-request; SETUP resets it,
* any DATA phase works normally, and the STATUS ack is special.
*/
case PIPE_CONTROL:
info = TD_CC | TD_DP_SETUP | TD_T_DATA0;
td_fill (ohci, info, urb->setup_dma, 8, urb, cnt++);
if (data_len > 0) {
info = TD_CC | TD_R | TD_T_DATA1;
info |= is_out ? TD_DP_OUT : TD_DP_IN;
/* NOTE: mishandles transfers >8K, some >4K */
td_fill (ohci, info, data, data_len, urb, cnt++);
}
info = (is_out || data_len == 0)
? TD_CC | TD_DP_IN | TD_T_DATA1
: TD_CC | TD_DP_OUT | TD_T_DATA1;
td_fill (ohci, info, data, 0, urb, cnt++);
/* maybe kickstart control list */
wmb ();
ohci_writel (ohci, OHCI_CLF, &ohci->regs->cmdstatus);
break;
/* ISO has no retransmit, so no toggle; and it uses special TDs.
* Each TD could handle multiple consecutive frames (interval 1);
* we could often reduce the number of TDs here.
*/
case PIPE_ISOCHRONOUS:
for (cnt = urb_priv->td_cnt; cnt < urb->number_of_packets;
cnt++) {
int frame = urb->start_frame;
// FIXME scheduling should handle frame counter
// roll-around ... exotic case (and OHCI has
// a 2^16 iso range, vs other HCs max of 2^10)
frame += cnt * urb->interval;
frame &= 0xffff;
td_fill (ohci, TD_CC | TD_ISO | frame,
data + urb->iso_frame_desc [cnt].offset,
urb->iso_frame_desc [cnt].length, urb, cnt);
}
if (ohci_to_hcd(ohci)->self.bandwidth_isoc_reqs == 0) {
if (quirk_amdiso(ohci))
USB host: Move AMD PLL quirk to pci-quirks.c This patch moves the AMD PLL quirk code in OHCI/EHCI driver to pci-quirks.c, and exports the functions to be used by xHCI driver later. AMD PLL quirk disable the optional PM feature inside specific SB700/SB800/Hudson-2/3 platforms under the following conditions: 1. If an isochronous device is connected to OHCI/EHCI/xHCI port and is active; 2. Optional PM feature that powers down the internal Bus PLL when the link is in low power state is enabled. Without AMD PLL quirk, USB isochronous stream may stutter or have breaks occasionally, which greatly impair the performance of audio/video streams. Currently AMD PLL quirk is implemented in OHCI and EHCI driver, and will be added to xHCI driver too. They are doing similar things actually, so move the quirk code to pci-quirks.c, which has several advantages: 1. Remove duplicate defines and functions in OHCI/EHCI (and xHCI) driver and make them cleaner; 2. AMD chipset information will be probed only once and then stored. Currently they're probed during every OHCI/EHCI initialization, move the detect code to pci-quirks.c saves the repeat detect cost; 3. Build up synchronization among OHCI/EHCI/xHCI driver. In current code, every host controller enable/disable PLL only according to its own status, and may enable PLL while there is still isoc transfer on other HCs. Move the quirk to pci-quirks.c prevents this issue. Signed-off-by: Andiry Xu <andiry.xu@amd.com> Cc: David Brownell <dbrownell@users.sourceforge.net> Cc: Alex He <alex.he@amd.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2011-03-01 06:57:05 +00:00
usb_amd_quirk_pll_disable();
if (quirk_amdprefetch(ohci))
sb800_prefetch(dev, 1);
}
periodic = ohci_to_hcd(ohci)->self.bandwidth_isoc_reqs++ == 0
&& ohci_to_hcd(ohci)->self.bandwidth_int_reqs == 0;
break;
}
/* start periodic dma if needed */
if (periodic) {
wmb ();
ohci->hc_control |= OHCI_CTRL_PLE|OHCI_CTRL_IE;
ohci_writel (ohci, ohci->hc_control, &ohci->regs->control);
}
// ASSERT (urb_priv->length == cnt);
}
/*-------------------------------------------------------------------------*
* Done List handling functions
*-------------------------------------------------------------------------*/
/* calculate transfer length/status and update the urb */
static int td_done(struct ohci_hcd *ohci, struct urb *urb, struct td *td)
{
u32 tdINFO = hc32_to_cpup (ohci, &td->hwINFO);
int cc = 0;
int status = -EINPROGRESS;
list_del (&td->td_list);
/* ISO ... drivers see per-TD length/status */
if (tdINFO & TD_ISO) {
u16 tdPSW = ohci_hwPSW(ohci, td, 0);
int dlen = 0;
/* NOTE: assumes FC in tdINFO == 0, and that
* only the first of 0..MAXPSW psws is used.
*/
cc = (tdPSW >> 12) & 0xF;
if (tdINFO & TD_CC) /* hc didn't touch? */
return status;
if (usb_pipeout (urb->pipe))
dlen = urb->iso_frame_desc [td->index].length;
else {
/* short reads are always OK for ISO */
if (cc == TD_DATAUNDERRUN)
cc = TD_CC_NOERROR;
dlen = tdPSW & 0x3ff;
}
urb->actual_length += dlen;
urb->iso_frame_desc [td->index].actual_length = dlen;
urb->iso_frame_desc [td->index].status = cc_to_error [cc];
if (cc != TD_CC_NOERROR)
ohci_dbg(ohci,
"urb %p iso td %p (%d) len %d cc %d\n",
urb, td, 1 + td->index, dlen, cc);
/* BULK, INT, CONTROL ... drivers see aggregate length/status,
* except that "setup" bytes aren't counted and "short" transfers
* might not be reported as errors.
*/
} else {
int type = usb_pipetype (urb->pipe);
u32 tdBE = hc32_to_cpup (ohci, &td->hwBE);
cc = TD_CC_GET (tdINFO);
/* update packet status if needed (short is normally ok) */
if (cc == TD_DATAUNDERRUN
&& !(urb->transfer_flags & URB_SHORT_NOT_OK))
cc = TD_CC_NOERROR;
if (cc != TD_CC_NOERROR && cc < 0x0E)
status = cc_to_error[cc];
/* count all non-empty packets except control SETUP packet */
if ((type != PIPE_CONTROL || td->index != 0) && tdBE != 0) {
if (td->hwCBP == 0)
urb->actual_length += tdBE - td->data_dma + 1;
else
urb->actual_length +=
hc32_to_cpup (ohci, &td->hwCBP)
- td->data_dma;
}
if (cc != TD_CC_NOERROR && cc < 0x0E)
ohci_dbg(ohci,
"urb %p td %p (%d) cc %d, len=%d/%d\n",
urb, td, 1 + td->index, cc,
urb->actual_length,
urb->transfer_buffer_length);
}
return status;
}
/*-------------------------------------------------------------------------*/
static void ed_halted(struct ohci_hcd *ohci, struct td *td, int cc)
{
struct urb *urb = td->urb;
urb_priv_t *urb_priv = urb->hcpriv;
struct ed *ed = td->ed;
struct list_head *tmp = td->td_list.next;
__hc32 toggle = ed->hwHeadP & cpu_to_hc32 (ohci, ED_C);
/* clear ed halt; this is the td that caused it, but keep it inactive
* until its urb->complete() has a chance to clean up.
*/
ed->hwINFO |= cpu_to_hc32 (ohci, ED_SKIP);
wmb ();
ed->hwHeadP &= ~cpu_to_hc32 (ohci, ED_H);
/* Get rid of all later tds from this urb. We don't have
* to be careful: no errors and nothing was transferred.
* Also patch the ed so it looks as if those tds completed normally.
*/
while (tmp != &ed->td_list) {
struct td *next;
next = list_entry (tmp, struct td, td_list);
tmp = next->td_list.next;
if (next->urb != urb)
break;
/* NOTE: if multi-td control DATA segments get supported,
* this urb had one of them, this td wasn't the last td
* in that segment (TD_R clear), this ed halted because
* of a short read, _and_ URB_SHORT_NOT_OK is clear ...
* then we need to leave the control STATUS packet queued
* and clear ED_SKIP.
*/
list_del(&next->td_list);
urb_priv->td_cnt++;
ed->hwHeadP = next->hwNextTD | toggle;
}
/* help for troubleshooting: report anything that
* looks odd ... that doesn't include protocol stalls
* (or maybe some other things)
*/
switch (cc) {
case TD_DATAUNDERRUN:
if ((urb->transfer_flags & URB_SHORT_NOT_OK) == 0)
break;
/* fallthrough */
case TD_CC_STALL:
if (usb_pipecontrol (urb->pipe))
break;
/* fallthrough */
default:
ohci_dbg (ohci,
"urb %p path %s ep%d%s %08x cc %d --> status %d\n",
urb, urb->dev->devpath,
usb_pipeendpoint (urb->pipe),
usb_pipein (urb->pipe) ? "in" : "out",
hc32_to_cpu (ohci, td->hwINFO),
cc, cc_to_error [cc]);
}
}
USB: OHCI: redesign the TD done list This patch changes the way ohci-hcd handles the TD done list. In addition to relying on the TD pointers stored by the controller hardware, we need to handle TDs that the hardware has forgotten about. This means the list has to exist even while the dl_done_list() routine isn't running. That function essentially gets split in two: update_done_list() reads the TD pointers stored by the hardware and adds the TDs to the done list, and process_done_list() scans through the list to handle URB completions. When we detect a TD that the hardware forgot about, we will be able to add it to the done list manually and then process it normally. Since the list is really a queue, and because there can be a lot of TDs, keep the existing singly linked implementation. To insure that URBs are given back in order of submission, whenever a TD is added to the done list, all the preceding TDs for the same endpoint must be added as well (going back to the first one that isn't already on the done list). The done list manipulations must all be protected by the private lock. The scope of the lock is expanded in preparation for the watchdog routine to be added in a later patch. We have to be more careful about giving back unlinked URBs. Since TDs may be added to the done list by the watchdog routine and not in response to a controller interrupt, we have to check explicitly to make sure all the URB's TDs that were added to the done list have been processed before giving back the URB. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-07-18 20:25:59 +00:00
/* Add a TD to the done list */
static void add_to_done_list(struct ohci_hcd *ohci, struct td *td)
{
struct td *td2, *td_prev;
struct ed *ed;
if (td->next_dl_td)
return; /* Already on the list */
/* Add all the TDs going back until we reach one that's on the list */
ed = td->ed;
td2 = td_prev = td;
list_for_each_entry_continue_reverse(td2, &ed->td_list, td_list) {
if (td2->next_dl_td)
break;
td2->next_dl_td = td_prev;
td_prev = td2;
}
if (ohci->dl_end)
ohci->dl_end->next_dl_td = td_prev;
else
ohci->dl_start = td_prev;
/*
* Make td->next_dl_td point to td itself, to mark the fact
* that td is on the done list.
*/
ohci->dl_end = td->next_dl_td = td;
/* Did we just add the latest pending TD? */
td2 = ed->pending_td;
if (td2 && td2->next_dl_td)
ed->pending_td = NULL;
USB: OHCI: redesign the TD done list This patch changes the way ohci-hcd handles the TD done list. In addition to relying on the TD pointers stored by the controller hardware, we need to handle TDs that the hardware has forgotten about. This means the list has to exist even while the dl_done_list() routine isn't running. That function essentially gets split in two: update_done_list() reads the TD pointers stored by the hardware and adds the TDs to the done list, and process_done_list() scans through the list to handle URB completions. When we detect a TD that the hardware forgot about, we will be able to add it to the done list manually and then process it normally. Since the list is really a queue, and because there can be a lot of TDs, keep the existing singly linked implementation. To insure that URBs are given back in order of submission, whenever a TD is added to the done list, all the preceding TDs for the same endpoint must be added as well (going back to the first one that isn't already on the done list). The done list manipulations must all be protected by the private lock. The scope of the lock is expanded in preparation for the watchdog routine to be added in a later patch. We have to be more careful about giving back unlinked URBs. Since TDs may be added to the done list by the watchdog routine and not in response to a controller interrupt, we have to check explicitly to make sure all the URB's TDs that were added to the done list have been processed before giving back the URB. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-07-18 20:25:59 +00:00
}
/* Get the entries on the hardware done queue and put them on our list */
static void update_done_list(struct ohci_hcd *ohci)
{
u32 td_dma;
struct td *td = NULL;
td_dma = hc32_to_cpup (ohci, &ohci->hcca->done_head);
ohci->hcca->done_head = 0;
wmb();
/* get TD from hc's singly linked list, and
USB: OHCI: redesign the TD done list This patch changes the way ohci-hcd handles the TD done list. In addition to relying on the TD pointers stored by the controller hardware, we need to handle TDs that the hardware has forgotten about. This means the list has to exist even while the dl_done_list() routine isn't running. That function essentially gets split in two: update_done_list() reads the TD pointers stored by the hardware and adds the TDs to the done list, and process_done_list() scans through the list to handle URB completions. When we detect a TD that the hardware forgot about, we will be able to add it to the done list manually and then process it normally. Since the list is really a queue, and because there can be a lot of TDs, keep the existing singly linked implementation. To insure that URBs are given back in order of submission, whenever a TD is added to the done list, all the preceding TDs for the same endpoint must be added as well (going back to the first one that isn't already on the done list). The done list manipulations must all be protected by the private lock. The scope of the lock is expanded in preparation for the watchdog routine to be added in a later patch. We have to be more careful about giving back unlinked URBs. Since TDs may be added to the done list by the watchdog routine and not in response to a controller interrupt, we have to check explicitly to make sure all the URB's TDs that were added to the done list have been processed before giving back the URB. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-07-18 20:25:59 +00:00
* add to ours. ed->td_list changes later.
*/
while (td_dma) {
int cc;
td = dma_to_td (ohci, td_dma);
if (!td) {
ohci_err (ohci, "bad entry %8x\n", td_dma);
break;
}
td->hwINFO |= cpu_to_hc32 (ohci, TD_DONE);
cc = TD_CC_GET (hc32_to_cpup (ohci, &td->hwINFO));
/* Non-iso endpoints can halt on error; un-halt,
* and dequeue any other TDs from this urb.
* No other TD could have caused the halt.
*/
if (cc != TD_CC_NOERROR
&& (td->ed->hwHeadP & cpu_to_hc32 (ohci, ED_H)))
ed_halted(ohci, td, cc);
td_dma = hc32_to_cpup (ohci, &td->hwNextTD);
USB: OHCI: redesign the TD done list This patch changes the way ohci-hcd handles the TD done list. In addition to relying on the TD pointers stored by the controller hardware, we need to handle TDs that the hardware has forgotten about. This means the list has to exist even while the dl_done_list() routine isn't running. That function essentially gets split in two: update_done_list() reads the TD pointers stored by the hardware and adds the TDs to the done list, and process_done_list() scans through the list to handle URB completions. When we detect a TD that the hardware forgot about, we will be able to add it to the done list manually and then process it normally. Since the list is really a queue, and because there can be a lot of TDs, keep the existing singly linked implementation. To insure that URBs are given back in order of submission, whenever a TD is added to the done list, all the preceding TDs for the same endpoint must be added as well (going back to the first one that isn't already on the done list). The done list manipulations must all be protected by the private lock. The scope of the lock is expanded in preparation for the watchdog routine to be added in a later patch. We have to be more careful about giving back unlinked URBs. Since TDs may be added to the done list by the watchdog routine and not in response to a controller interrupt, we have to check explicitly to make sure all the URB's TDs that were added to the done list have been processed before giving back the URB. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-07-18 20:25:59 +00:00
add_to_done_list(ohci, td);
}
}
/*-------------------------------------------------------------------------*/
/* there are some urbs/eds to unlink; called in_irq(), with HCD locked */
static void finish_unlinks(struct ohci_hcd *ohci)
{
unsigned tick = ohci_frame_no(ohci);
struct ed *ed, **last;
rescan_all:
for (last = &ohci->ed_rm_list, ed = *last; ed != NULL; ed = *last) {
struct list_head *entry, *tmp;
int completed, modified;
__hc32 *prev;
/* only take off EDs that the HC isn't using, accounting for
* frame counter wraps and EDs with partially retired TDs
*/
USB: OHCI: redesign the TD done list This patch changes the way ohci-hcd handles the TD done list. In addition to relying on the TD pointers stored by the controller hardware, we need to handle TDs that the hardware has forgotten about. This means the list has to exist even while the dl_done_list() routine isn't running. That function essentially gets split in two: update_done_list() reads the TD pointers stored by the hardware and adds the TDs to the done list, and process_done_list() scans through the list to handle URB completions. When we detect a TD that the hardware forgot about, we will be able to add it to the done list manually and then process it normally. Since the list is really a queue, and because there can be a lot of TDs, keep the existing singly linked implementation. To insure that URBs are given back in order of submission, whenever a TD is added to the done list, all the preceding TDs for the same endpoint must be added as well (going back to the first one that isn't already on the done list). The done list manipulations must all be protected by the private lock. The scope of the lock is expanded in preparation for the watchdog routine to be added in a later patch. We have to be more careful about giving back unlinked URBs. Since TDs may be added to the done list by the watchdog routine and not in response to a controller interrupt, we have to check explicitly to make sure all the URB's TDs that were added to the done list have been processed before giving back the URB. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-07-18 20:25:59 +00:00
if (likely(ohci->rh_state == OHCI_RH_RUNNING) &&
tick_before(tick, ed->tick)) {
skip_ed:
USB: OHCI: redesign the TD done list This patch changes the way ohci-hcd handles the TD done list. In addition to relying on the TD pointers stored by the controller hardware, we need to handle TDs that the hardware has forgotten about. This means the list has to exist even while the dl_done_list() routine isn't running. That function essentially gets split in two: update_done_list() reads the TD pointers stored by the hardware and adds the TDs to the done list, and process_done_list() scans through the list to handle URB completions. When we detect a TD that the hardware forgot about, we will be able to add it to the done list manually and then process it normally. Since the list is really a queue, and because there can be a lot of TDs, keep the existing singly linked implementation. To insure that URBs are given back in order of submission, whenever a TD is added to the done list, all the preceding TDs for the same endpoint must be added as well (going back to the first one that isn't already on the done list). The done list manipulations must all be protected by the private lock. The scope of the lock is expanded in preparation for the watchdog routine to be added in a later patch. We have to be more careful about giving back unlinked URBs. Since TDs may be added to the done list by the watchdog routine and not in response to a controller interrupt, we have to check explicitly to make sure all the URB's TDs that were added to the done list have been processed before giving back the URB. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-07-18 20:25:59 +00:00
last = &ed->ed_next;
continue;
}
if (!list_empty(&ed->td_list)) {
struct td *td;
u32 head;
USB: OHCI: redesign the TD done list This patch changes the way ohci-hcd handles the TD done list. In addition to relying on the TD pointers stored by the controller hardware, we need to handle TDs that the hardware has forgotten about. This means the list has to exist even while the dl_done_list() routine isn't running. That function essentially gets split in two: update_done_list() reads the TD pointers stored by the hardware and adds the TDs to the done list, and process_done_list() scans through the list to handle URB completions. When we detect a TD that the hardware forgot about, we will be able to add it to the done list manually and then process it normally. Since the list is really a queue, and because there can be a lot of TDs, keep the existing singly linked implementation. To insure that URBs are given back in order of submission, whenever a TD is added to the done list, all the preceding TDs for the same endpoint must be added as well (going back to the first one that isn't already on the done list). The done list manipulations must all be protected by the private lock. The scope of the lock is expanded in preparation for the watchdog routine to be added in a later patch. We have to be more careful about giving back unlinked URBs. Since TDs may be added to the done list by the watchdog routine and not in response to a controller interrupt, we have to check explicitly to make sure all the URB's TDs that were added to the done list have been processed before giving back the URB. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-07-18 20:25:59 +00:00
td = list_first_entry(&ed->td_list, struct td, td_list);
USB: OHCI: redesign the TD done list This patch changes the way ohci-hcd handles the TD done list. In addition to relying on the TD pointers stored by the controller hardware, we need to handle TDs that the hardware has forgotten about. This means the list has to exist even while the dl_done_list() routine isn't running. That function essentially gets split in two: update_done_list() reads the TD pointers stored by the hardware and adds the TDs to the done list, and process_done_list() scans through the list to handle URB completions. When we detect a TD that the hardware forgot about, we will be able to add it to the done list manually and then process it normally. Since the list is really a queue, and because there can be a lot of TDs, keep the existing singly linked implementation. To insure that URBs are given back in order of submission, whenever a TD is added to the done list, all the preceding TDs for the same endpoint must be added as well (going back to the first one that isn't already on the done list). The done list manipulations must all be protected by the private lock. The scope of the lock is expanded in preparation for the watchdog routine to be added in a later patch. We have to be more careful about giving back unlinked URBs. Since TDs may be added to the done list by the watchdog routine and not in response to a controller interrupt, we have to check explicitly to make sure all the URB's TDs that were added to the done list have been processed before giving back the URB. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-07-18 20:25:59 +00:00
/* INTR_WDH may need to clean up first */
head = hc32_to_cpu(ohci, ed->hwHeadP) & TD_MASK;
if (td->td_dma != head &&
ohci->rh_state == OHCI_RH_RUNNING)
goto skip_ed;
USB: OHCI: redesign the TD done list This patch changes the way ohci-hcd handles the TD done list. In addition to relying on the TD pointers stored by the controller hardware, we need to handle TDs that the hardware has forgotten about. This means the list has to exist even while the dl_done_list() routine isn't running. That function essentially gets split in two: update_done_list() reads the TD pointers stored by the hardware and adds the TDs to the done list, and process_done_list() scans through the list to handle URB completions. When we detect a TD that the hardware forgot about, we will be able to add it to the done list manually and then process it normally. Since the list is really a queue, and because there can be a lot of TDs, keep the existing singly linked implementation. To insure that URBs are given back in order of submission, whenever a TD is added to the done list, all the preceding TDs for the same endpoint must be added as well (going back to the first one that isn't already on the done list). The done list manipulations must all be protected by the private lock. The scope of the lock is expanded in preparation for the watchdog routine to be added in a later patch. We have to be more careful about giving back unlinked URBs. Since TDs may be added to the done list by the watchdog routine and not in response to a controller interrupt, we have to check explicitly to make sure all the URB's TDs that were added to the done list have been processed before giving back the URB. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-07-18 20:25:59 +00:00
/* Don't mess up anything already on the done list */
if (td->next_dl_td)
goto skip_ed;
}
/* ED's now officially unlinked, hc doesn't see */
ed->hwHeadP &= ~cpu_to_hc32(ohci, ED_H);
ed->hwNextED = 0;
wmb();
ed->hwINFO &= ~cpu_to_hc32(ohci, ED_SKIP | ED_DEQUEUE);
/* reentrancy: if we drop the schedule lock, someone might
* have modified this list. normally it's just prepending
* entries (which we'd ignore), but paranoia won't hurt.
*/
usb: ohci: Proper handling of ed_rm_list to handle race condition between usb_kill_urb() and finish_unlinks() There is a race condition between finish_unlinks->finish_urb() function and usb_kill_urb() in ohci controller case. The finish_urb calls spin_unlock(&ohci->lock) before usb_hcd_giveback_urb() function call, then if during this time, usb_kill_urb is called for another endpoint, then new ed will be added to ed_rm_list at beginning for unlink, and ed_rm_list will point to newly added. When finish_urb() is completed in finish_unlinks() and ed->td_list becomes empty as in below code (in finish_unlinks() function): if (list_empty(&ed->td_list)) { *last = ed->ed_next; ed->ed_next = NULL; } else if (ohci->rh_state == OHCI_RH_RUNNING) { *last = ed->ed_next; ed->ed_next = NULL; ed_schedule(ohci, ed); } The *last = ed->ed_next will make ed_rm_list to point to ed->ed_next and previously added ed by usb_kill_urb will be left unreferenced by ed_rm_list. This causes usb_kill_urb() hang forever waiting for finish_unlink to remove added ed from ed_rm_list. The main reason for hang in this race condtion is addition and removal of ed from ed_rm_list in the beginning during usb_kill_urb and later last* is modified in finish_unlinks(). As suggested by Alan Stern, the solution for proper handling of ohci->ed_rm_list is to remove ed from the ed_rm_list before finishing any URBs. Then at the end, we can add ed back to the list if necessary. This properly handle the updated ohci->ed_rm_list in usb_kill_urb(). Fixes: 977dcfdc6031 ("USB: OHCI: don't lose track of EDs when a controller dies") Acked-by: Alan Stern <stern@rowland.harvard.edu> CC: <stable@vger.kernel.org> Signed-off-by: Aman Deep <aman.deep@samsung.com> Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2018-02-08 03:55:01 +00:00
*last = ed->ed_next;
ed->ed_next = NULL;
modified = 0;
/* unlink urbs as requested, but rescan the list after
* we call a completion since it might have unlinked
* another (earlier) urb
*
* When we get here, the HC doesn't see this ed. But it
* must not be rescheduled until all completed URBs have
* been given back to the driver.
*/
rescan_this:
completed = 0;
prev = &ed->hwHeadP;
list_for_each_safe (entry, tmp, &ed->td_list) {
struct td *td;
struct urb *urb;
urb_priv_t *urb_priv;
__hc32 savebits;
u32 tdINFO;
td = list_entry (entry, struct td, td_list);
urb = td->urb;
urb_priv = td->urb->hcpriv;
if (!urb->unlinked) {
prev = &td->hwNextTD;
continue;
}
/* patch pointer hc uses */
savebits = *prev & ~cpu_to_hc32 (ohci, TD_MASK);
*prev = td->hwNextTD | savebits;
/* If this was unlinked, the TD may not have been
* retired ... so manually save the data toggle.
* The controller ignores the value we save for
* control and ISO endpoints.
*/
tdINFO = hc32_to_cpup(ohci, &td->hwINFO);
if ((tdINFO & TD_T) == TD_T_DATA0)
ed->hwHeadP &= ~cpu_to_hc32(ohci, ED_C);
else if ((tdINFO & TD_T) == TD_T_DATA1)
ed->hwHeadP |= cpu_to_hc32(ohci, ED_C);
/* HC may have partly processed this TD */
td_done (ohci, urb, td);
urb_priv->td_cnt++;
/* if URB is done, clean up */
USB: OHCI: accept very late isochronous URBs Commit 24f531371de1 (USB: EHCI: accept very late isochronous URBs) changed the isochronous API provided by ehci-hcd. URBs submitted too late, so that the time slots for all their packets have already expired, are no longer rejected outright. Instead the submission is accepted, and the URB completes normally with a -EXDEV error for each packet. This is what client drivers expect. This patch implements the same policy in ohci-hcd. The change is more complicated than it was in ehci-hcd, because ohci-hcd doesn't scan for isochronous completions in the same way as ehci-hcd does. Rather, it depends on the hardware adding completed TDs to a "done queue". Some OHCI controller don't handle this properly when a TD's time slot has already expired, so we have to avoid adding such TDs to the schedule in the first place. As a result, if the URB was submitted too late then none of its TDs will get put on the schedule, so none of them will end up on the done queue, so the driver will never realize that the URB should be completed. To solve this problem, the patch adds one to urb_priv->td_cnt for such URBs, making it larger than urb_priv->length (td_cnt already gets set to the number of TD's that had to be skipped because their slots have expired). Each time an URB is given back, the finish_urb() routine looks to see if urb_priv->td_cnt for the next URB on the same endpoint is marked in this way. If so, it gives back the next URB right away. This should be applied to all kernels containing commit 815fa7b91761 (USB: OHCI: fix logic for scheduling isochronous URBs). Signed-off-by: Alan Stern <stern@rowland.harvard.edu> CC: <stable@vger.kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-09-24 19:46:45 +00:00
if (urb_priv->td_cnt >= urb_priv->length) {
modified = completed = 1;
finish_urb(ohci, urb, 0);
}
}
if (completed && !list_empty (&ed->td_list))
goto rescan_this;
/*
usb: ohci: Proper handling of ed_rm_list to handle race condition between usb_kill_urb() and finish_unlinks() There is a race condition between finish_unlinks->finish_urb() function and usb_kill_urb() in ohci controller case. The finish_urb calls spin_unlock(&ohci->lock) before usb_hcd_giveback_urb() function call, then if during this time, usb_kill_urb is called for another endpoint, then new ed will be added to ed_rm_list at beginning for unlink, and ed_rm_list will point to newly added. When finish_urb() is completed in finish_unlinks() and ed->td_list becomes empty as in below code (in finish_unlinks() function): if (list_empty(&ed->td_list)) { *last = ed->ed_next; ed->ed_next = NULL; } else if (ohci->rh_state == OHCI_RH_RUNNING) { *last = ed->ed_next; ed->ed_next = NULL; ed_schedule(ohci, ed); } The *last = ed->ed_next will make ed_rm_list to point to ed->ed_next and previously added ed by usb_kill_urb will be left unreferenced by ed_rm_list. This causes usb_kill_urb() hang forever waiting for finish_unlink to remove added ed from ed_rm_list. The main reason for hang in this race condtion is addition and removal of ed from ed_rm_list in the beginning during usb_kill_urb and later last* is modified in finish_unlinks(). As suggested by Alan Stern, the solution for proper handling of ohci->ed_rm_list is to remove ed from the ed_rm_list before finishing any URBs. Then at the end, we can add ed back to the list if necessary. This properly handle the updated ohci->ed_rm_list in usb_kill_urb(). Fixes: 977dcfdc6031 ("USB: OHCI: don't lose track of EDs when a controller dies") Acked-by: Alan Stern <stern@rowland.harvard.edu> CC: <stable@vger.kernel.org> Signed-off-by: Aman Deep <aman.deep@samsung.com> Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2018-02-08 03:55:01 +00:00
* If no TDs are queued, ED is now idle.
* Otherwise, if the HC is running, reschedule.
usb: ohci: Proper handling of ed_rm_list to handle race condition between usb_kill_urb() and finish_unlinks() There is a race condition between finish_unlinks->finish_urb() function and usb_kill_urb() in ohci controller case. The finish_urb calls spin_unlock(&ohci->lock) before usb_hcd_giveback_urb() function call, then if during this time, usb_kill_urb is called for another endpoint, then new ed will be added to ed_rm_list at beginning for unlink, and ed_rm_list will point to newly added. When finish_urb() is completed in finish_unlinks() and ed->td_list becomes empty as in below code (in finish_unlinks() function): if (list_empty(&ed->td_list)) { *last = ed->ed_next; ed->ed_next = NULL; } else if (ohci->rh_state == OHCI_RH_RUNNING) { *last = ed->ed_next; ed->ed_next = NULL; ed_schedule(ohci, ed); } The *last = ed->ed_next will make ed_rm_list to point to ed->ed_next and previously added ed by usb_kill_urb will be left unreferenced by ed_rm_list. This causes usb_kill_urb() hang forever waiting for finish_unlink to remove added ed from ed_rm_list. The main reason for hang in this race condtion is addition and removal of ed from ed_rm_list in the beginning during usb_kill_urb and later last* is modified in finish_unlinks(). As suggested by Alan Stern, the solution for proper handling of ohci->ed_rm_list is to remove ed from the ed_rm_list before finishing any URBs. Then at the end, we can add ed back to the list if necessary. This properly handle the updated ohci->ed_rm_list in usb_kill_urb(). Fixes: 977dcfdc6031 ("USB: OHCI: don't lose track of EDs when a controller dies") Acked-by: Alan Stern <stern@rowland.harvard.edu> CC: <stable@vger.kernel.org> Signed-off-by: Aman Deep <aman.deep@samsung.com> Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2018-02-08 03:55:01 +00:00
* If the HC isn't running, add ED back to the
* start of the list for later processing.
*/
if (list_empty(&ed->td_list)) {
ed->state = ED_IDLE;
list_del(&ed->in_use_list);
} else if (ohci->rh_state == OHCI_RH_RUNNING) {
ed_schedule(ohci, ed);
} else {
usb: ohci: Proper handling of ed_rm_list to handle race condition between usb_kill_urb() and finish_unlinks() There is a race condition between finish_unlinks->finish_urb() function and usb_kill_urb() in ohci controller case. The finish_urb calls spin_unlock(&ohci->lock) before usb_hcd_giveback_urb() function call, then if during this time, usb_kill_urb is called for another endpoint, then new ed will be added to ed_rm_list at beginning for unlink, and ed_rm_list will point to newly added. When finish_urb() is completed in finish_unlinks() and ed->td_list becomes empty as in below code (in finish_unlinks() function): if (list_empty(&ed->td_list)) { *last = ed->ed_next; ed->ed_next = NULL; } else if (ohci->rh_state == OHCI_RH_RUNNING) { *last = ed->ed_next; ed->ed_next = NULL; ed_schedule(ohci, ed); } The *last = ed->ed_next will make ed_rm_list to point to ed->ed_next and previously added ed by usb_kill_urb will be left unreferenced by ed_rm_list. This causes usb_kill_urb() hang forever waiting for finish_unlink to remove added ed from ed_rm_list. The main reason for hang in this race condtion is addition and removal of ed from ed_rm_list in the beginning during usb_kill_urb and later last* is modified in finish_unlinks(). As suggested by Alan Stern, the solution for proper handling of ohci->ed_rm_list is to remove ed from the ed_rm_list before finishing any URBs. Then at the end, we can add ed back to the list if necessary. This properly handle the updated ohci->ed_rm_list in usb_kill_urb(). Fixes: 977dcfdc6031 ("USB: OHCI: don't lose track of EDs when a controller dies") Acked-by: Alan Stern <stern@rowland.harvard.edu> CC: <stable@vger.kernel.org> Signed-off-by: Aman Deep <aman.deep@samsung.com> Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2018-02-08 03:55:01 +00:00
ed->ed_next = ohci->ed_rm_list;
ohci->ed_rm_list = ed;
/* Don't loop on the same ED */
if (last == &ohci->ed_rm_list)
last = &ed->ed_next;
}
if (modified)
goto rescan_all;
}
/* maybe reenable control and bulk lists */
if (ohci->rh_state == OHCI_RH_RUNNING && !ohci->ed_rm_list) {
u32 command = 0, control = 0;
if (ohci->ed_controltail) {
command |= OHCI_CLF;
if (quirk_zfmicro(ohci))
mdelay(1);
if (!(ohci->hc_control & OHCI_CTRL_CLE)) {
control |= OHCI_CTRL_CLE;
ohci_writel (ohci, 0,
&ohci->regs->ed_controlcurrent);
}
}
if (ohci->ed_bulktail) {
command |= OHCI_BLF;
if (quirk_zfmicro(ohci))
mdelay(1);
if (!(ohci->hc_control & OHCI_CTRL_BLE)) {
control |= OHCI_CTRL_BLE;
ohci_writel (ohci, 0,
&ohci->regs->ed_bulkcurrent);
}
}
/* CLE/BLE to enable, CLF/BLF to (maybe) kickstart */
if (control) {
ohci->hc_control |= control;
if (quirk_zfmicro(ohci))
mdelay(1);
ohci_writel (ohci, ohci->hc_control,
&ohci->regs->control);
}
if (command) {
if (quirk_zfmicro(ohci))
mdelay(1);
ohci_writel (ohci, command, &ohci->regs->cmdstatus);
}
}
}
/*-------------------------------------------------------------------------*/
/* Take back a TD from the host controller */
static void takeback_td(struct ohci_hcd *ohci, struct td *td)
{
struct urb *urb = td->urb;
urb_priv_t *urb_priv = urb->hcpriv;
struct ed *ed = td->ed;
int status;
/* update URB's length and status from TD */
status = td_done(ohci, urb, td);
urb_priv->td_cnt++;
/* If all this urb's TDs are done, call complete() */
USB: OHCI: accept very late isochronous URBs Commit 24f531371de1 (USB: EHCI: accept very late isochronous URBs) changed the isochronous API provided by ehci-hcd. URBs submitted too late, so that the time slots for all their packets have already expired, are no longer rejected outright. Instead the submission is accepted, and the URB completes normally with a -EXDEV error for each packet. This is what client drivers expect. This patch implements the same policy in ohci-hcd. The change is more complicated than it was in ehci-hcd, because ohci-hcd doesn't scan for isochronous completions in the same way as ehci-hcd does. Rather, it depends on the hardware adding completed TDs to a "done queue". Some OHCI controller don't handle this properly when a TD's time slot has already expired, so we have to avoid adding such TDs to the schedule in the first place. As a result, if the URB was submitted too late then none of its TDs will get put on the schedule, so none of them will end up on the done queue, so the driver will never realize that the URB should be completed. To solve this problem, the patch adds one to urb_priv->td_cnt for such URBs, making it larger than urb_priv->length (td_cnt already gets set to the number of TD's that had to be skipped because their slots have expired). Each time an URB is given back, the finish_urb() routine looks to see if urb_priv->td_cnt for the next URB on the same endpoint is marked in this way. If so, it gives back the next URB right away. This should be applied to all kernels containing commit 815fa7b91761 (USB: OHCI: fix logic for scheduling isochronous URBs). Signed-off-by: Alan Stern <stern@rowland.harvard.edu> CC: <stable@vger.kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-09-24 19:46:45 +00:00
if (urb_priv->td_cnt >= urb_priv->length)
finish_urb(ohci, urb, status);
/* clean schedule: unlink EDs that are no longer busy */
if (list_empty(&ed->td_list)) {
if (ed->state == ED_OPER)
start_ed_unlink(ohci, ed);
/* ... reenabling halted EDs only after fault cleanup */
} else if ((ed->hwINFO & cpu_to_hc32(ohci, ED_SKIP | ED_DEQUEUE))
== cpu_to_hc32(ohci, ED_SKIP)) {
td = list_entry(ed->td_list.next, struct td, td_list);
if (!(td->hwINFO & cpu_to_hc32(ohci, TD_DONE))) {
ed->hwINFO &= ~cpu_to_hc32(ohci, ED_SKIP);
/* ... hc may need waking-up */
switch (ed->type) {
case PIPE_CONTROL:
ohci_writel(ohci, OHCI_CLF,
&ohci->regs->cmdstatus);
break;
case PIPE_BULK:
ohci_writel(ohci, OHCI_BLF,
&ohci->regs->cmdstatus);
break;
}
}
}
}
/*
* Process normal completions (error or success) and clean the schedules.
*
* This is the main path for handing urbs back to drivers. The only other
* normal path is finish_unlinks(), which unlinks URBs using ed_rm_list,
* instead of scanning the (re-reversed) donelist as this does.
*/
USB: OHCI: redesign the TD done list This patch changes the way ohci-hcd handles the TD done list. In addition to relying on the TD pointers stored by the controller hardware, we need to handle TDs that the hardware has forgotten about. This means the list has to exist even while the dl_done_list() routine isn't running. That function essentially gets split in two: update_done_list() reads the TD pointers stored by the hardware and adds the TDs to the done list, and process_done_list() scans through the list to handle URB completions. When we detect a TD that the hardware forgot about, we will be able to add it to the done list manually and then process it normally. Since the list is really a queue, and because there can be a lot of TDs, keep the existing singly linked implementation. To insure that URBs are given back in order of submission, whenever a TD is added to the done list, all the preceding TDs for the same endpoint must be added as well (going back to the first one that isn't already on the done list). The done list manipulations must all be protected by the private lock. The scope of the lock is expanded in preparation for the watchdog routine to be added in a later patch. We have to be more careful about giving back unlinked URBs. Since TDs may be added to the done list by the watchdog routine and not in response to a controller interrupt, we have to check explicitly to make sure all the URB's TDs that were added to the done list have been processed before giving back the URB. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-07-18 20:25:59 +00:00
static void process_done_list(struct ohci_hcd *ohci)
{
USB: OHCI: redesign the TD done list This patch changes the way ohci-hcd handles the TD done list. In addition to relying on the TD pointers stored by the controller hardware, we need to handle TDs that the hardware has forgotten about. This means the list has to exist even while the dl_done_list() routine isn't running. That function essentially gets split in two: update_done_list() reads the TD pointers stored by the hardware and adds the TDs to the done list, and process_done_list() scans through the list to handle URB completions. When we detect a TD that the hardware forgot about, we will be able to add it to the done list manually and then process it normally. Since the list is really a queue, and because there can be a lot of TDs, keep the existing singly linked implementation. To insure that URBs are given back in order of submission, whenever a TD is added to the done list, all the preceding TDs for the same endpoint must be added as well (going back to the first one that isn't already on the done list). The done list manipulations must all be protected by the private lock. The scope of the lock is expanded in preparation for the watchdog routine to be added in a later patch. We have to be more careful about giving back unlinked URBs. Since TDs may be added to the done list by the watchdog routine and not in response to a controller interrupt, we have to check explicitly to make sure all the URB's TDs that were added to the done list have been processed before giving back the URB. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-07-18 20:25:59 +00:00
struct td *td;
USB: OHCI: redesign the TD done list This patch changes the way ohci-hcd handles the TD done list. In addition to relying on the TD pointers stored by the controller hardware, we need to handle TDs that the hardware has forgotten about. This means the list has to exist even while the dl_done_list() routine isn't running. That function essentially gets split in two: update_done_list() reads the TD pointers stored by the hardware and adds the TDs to the done list, and process_done_list() scans through the list to handle URB completions. When we detect a TD that the hardware forgot about, we will be able to add it to the done list manually and then process it normally. Since the list is really a queue, and because there can be a lot of TDs, keep the existing singly linked implementation. To insure that URBs are given back in order of submission, whenever a TD is added to the done list, all the preceding TDs for the same endpoint must be added as well (going back to the first one that isn't already on the done list). The done list manipulations must all be protected by the private lock. The scope of the lock is expanded in preparation for the watchdog routine to be added in a later patch. We have to be more careful about giving back unlinked URBs. Since TDs may be added to the done list by the watchdog routine and not in response to a controller interrupt, we have to check explicitly to make sure all the URB's TDs that were added to the done list have been processed before giving back the URB. Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-07-18 20:25:59 +00:00
while (ohci->dl_start) {
td = ohci->dl_start;
if (td == ohci->dl_end)
ohci->dl_start = ohci->dl_end = NULL;
else
ohci->dl_start = td->next_dl_td;
takeback_td(ohci, td);
}
}
/*
* TD takeback and URB giveback must be single-threaded.
* This routine takes care of it all.
*/
static void ohci_work(struct ohci_hcd *ohci)
{
if (ohci->working) {
ohci->restart_work = 1;
return;
}
ohci->working = 1;
restart:
process_done_list(ohci);
if (ohci->ed_rm_list)
finish_unlinks(ohci);
if (ohci->restart_work) {
ohci->restart_work = 0;
goto restart;
}
ohci->working = 0;
}