]> pilppa.com Git - linux-2.6-omap-h63xx.git/commitdiff
usb host: Oxford OXU210HP HCD driver.
authorRodolfo Giometti <giometti@linux.it>
Thu, 23 Oct 2008 08:08:07 +0000 (10:08 +0200)
committerGreg Kroah-Hartman <gregkh@suse.de>
Wed, 7 Jan 2009 17:59:50 +0000 (09:59 -0800)
This driver implements the support for Oxford OXU210HP USB high-speed host,
no peripheral nor OTG.

Signed-off-by: Rodolfo Giometti <giometti@linux.it>
Cc: Kan Liu <kan.k.liu@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
drivers/usb/host/Kconfig
drivers/usb/host/Makefile
drivers/usb/host/oxu210hp-hcd.c [new file with mode: 0644]
drivers/usb/host/oxu210hp.h [new file with mode: 0644]
include/linux/oxu210hp.h [new file with mode: 0644]

index f3a75a929e0aa26dcdbaee0c75c0cfa10403f8bf..2b476b6b3d4d3e1ed6ad2b24ff082178e0d3e147 100644 (file)
@@ -96,6 +96,19 @@ config USB_EHCI_HCD_PPC_OF
          Enables support for the USB controller present on the PowerPC
          OpenFirmware platform bus.
 
+config USB_OXU210HP_HCD
+       tristate "OXU210HP HCD support"
+       depends on USB
+       ---help---
+         The OXU210HP is an USB host/OTG/device controller. Enable this
+         option if your board has this chip. If unsure, say N.
+
+         This driver does not support isochronous transfers and doesn't
+         implement OTG nor USB device controllers.
+
+         To compile this driver as a module, choose M here: the
+         module will be called oxu210hp-hcd.
+
 config USB_ISP116X_HCD
        tristate "ISP116X HCD support"
        depends on USB
index 23be222240447bbc60c902617e77a6cc5b4476a4..e5f3f20787e4ffd665cbc2ba004bc5a3577abe8e 100644 (file)
@@ -13,6 +13,7 @@ obj-$(CONFIG_USB_WHCI_HCD)    += whci/
 obj-$(CONFIG_PCI)              += pci-quirks.o
 
 obj-$(CONFIG_USB_EHCI_HCD)     += ehci-hcd.o
+obj-$(CONFIG_USB_OXU210HP_HCD) += oxu210hp-hcd.o
 obj-$(CONFIG_USB_ISP116X_HCD)  += isp116x-hcd.o
 obj-$(CONFIG_USB_OHCI_HCD)     += ohci-hcd.o
 obj-$(CONFIG_USB_UHCI_HCD)     += uhci-hcd.o
diff --git a/drivers/usb/host/oxu210hp-hcd.c b/drivers/usb/host/oxu210hp-hcd.c
new file mode 100644 (file)
index 0000000..251123c
--- /dev/null
@@ -0,0 +1,3985 @@
+/*
+ * Copyright (c) 2008 Rodolfo Giometti <giometti@linux.it>
+ * Copyright (c) 2008 Eurotech S.p.A. <info@eurtech.it>
+ *
+ * This code is *strongly* based on EHCI-HCD code by David Brownell since
+ * the chip is a quasi-EHCI compatible.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the
+ * Free Software Foundation; either version 2 of the License, or (at your
+ * option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/dmapool.h>
+#include <linux/kernel.h>
+#include <linux/delay.h>
+#include <linux/ioport.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/errno.h>
+#include <linux/init.h>
+#include <linux/timer.h>
+#include <linux/list.h>
+#include <linux/interrupt.h>
+#include <linux/reboot.h>
+#include <linux/usb.h>
+#include <linux/moduleparam.h>
+#include <linux/dma-mapping.h>
+#include <linux/io.h>
+
+#include "../core/hcd.h"
+
+#include <asm/irq.h>
+#include <asm/system.h>
+#include <asm/unaligned.h>
+
+#include <linux/irq.h>
+#include <linux/platform_device.h>
+
+#include "oxu210hp.h"
+
+#define DRIVER_VERSION "0.0.50"
+
+/*
+ * Main defines
+ */
+
+#define oxu_dbg(oxu, fmt, args...) \
+               dev_dbg(oxu_to_hcd(oxu)->self.controller , fmt , ## args)
+#define oxu_err(oxu, fmt, args...) \
+               dev_err(oxu_to_hcd(oxu)->self.controller , fmt , ## args)
+#define oxu_info(oxu, fmt, args...) \
+               dev_info(oxu_to_hcd(oxu)->self.controller , fmt , ## args)
+
+static inline struct usb_hcd *oxu_to_hcd(struct oxu_hcd *oxu)
+{
+       return container_of((void *) oxu, struct usb_hcd, hcd_priv);
+}
+
+static inline struct oxu_hcd *hcd_to_oxu(struct usb_hcd *hcd)
+{
+       return (struct oxu_hcd *) (hcd->hcd_priv);
+}
+
+/*
+ * Debug stuff
+ */
+
+#undef OXU_URB_TRACE
+#undef OXU_VERBOSE_DEBUG
+
+#ifdef OXU_VERBOSE_DEBUG
+#define oxu_vdbg                       oxu_dbg
+#else
+#define oxu_vdbg(oxu, fmt, args...)    /* Nop */
+#endif
+
+#ifdef DEBUG
+
+static int __attribute__((__unused__))
+dbg_status_buf(char *buf, unsigned len, const char *label, u32 status)
+{
+       return scnprintf(buf, len, "%s%sstatus %04x%s%s%s%s%s%s%s%s%s%s",
+               label, label[0] ? " " : "", status,
+               (status & STS_ASS) ? " Async" : "",
+               (status & STS_PSS) ? " Periodic" : "",
+               (status & STS_RECL) ? " Recl" : "",
+               (status & STS_HALT) ? " Halt" : "",
+               (status & STS_IAA) ? " IAA" : "",
+               (status & STS_FATAL) ? " FATAL" : "",
+               (status & STS_FLR) ? " FLR" : "",
+               (status & STS_PCD) ? " PCD" : "",
+               (status & STS_ERR) ? " ERR" : "",
+               (status & STS_INT) ? " INT" : ""
+               );
+}
+
+static int __attribute__((__unused__))
+dbg_intr_buf(char *buf, unsigned len, const char *label, u32 enable)
+{
+       return scnprintf(buf, len, "%s%sintrenable %02x%s%s%s%s%s%s",
+               label, label[0] ? " " : "", enable,
+               (enable & STS_IAA) ? " IAA" : "",
+               (enable & STS_FATAL) ? " FATAL" : "",
+               (enable & STS_FLR) ? " FLR" : "",
+               (enable & STS_PCD) ? " PCD" : "",
+               (enable & STS_ERR) ? " ERR" : "",
+               (enable & STS_INT) ? " INT" : ""
+               );
+}
+
+static const char *const fls_strings[] =
+    { "1024", "512", "256", "??" };
+
+static int dbg_command_buf(char *buf, unsigned len,
+                               const char *label, u32 command)
+{
+       return scnprintf(buf, len,
+               "%s%scommand %06x %s=%d ithresh=%d%s%s%s%s period=%s%s %s",
+               label, label[0] ? " " : "", command,
+               (command & CMD_PARK) ? "park" : "(park)",
+               CMD_PARK_CNT(command),
+               (command >> 16) & 0x3f,
+               (command & CMD_LRESET) ? " LReset" : "",
+               (command & CMD_IAAD) ? " IAAD" : "",
+               (command & CMD_ASE) ? " Async" : "",
+               (command & CMD_PSE) ? " Periodic" : "",
+               fls_strings[(command >> 2) & 0x3],
+               (command & CMD_RESET) ? " Reset" : "",
+               (command & CMD_RUN) ? "RUN" : "HALT"
+               );
+}
+
+static int dbg_port_buf(char *buf, unsigned len, const char *label,
+                               int port, u32 status)
+{
+       char    *sig;
+
+       /* signaling state */
+       switch (status & (3 << 10)) {
+       case 0 << 10:
+               sig = "se0";
+               break;
+       case 1 << 10:
+               sig = "k";      /* low speed */
+               break;
+       case 2 << 10:
+               sig = "j";
+               break;
+       default:
+               sig = "?";
+               break;
+       }
+
+       return scnprintf(buf, len,
+               "%s%sport %d status %06x%s%s sig=%s%s%s%s%s%s%s%s%s%s",
+               label, label[0] ? " " : "", port, status,
+               (status & PORT_POWER) ? " POWER" : "",
+               (status & PORT_OWNER) ? " OWNER" : "",
+               sig,
+               (status & PORT_RESET) ? " RESET" : "",
+               (status & PORT_SUSPEND) ? " SUSPEND" : "",
+               (status & PORT_RESUME) ? " RESUME" : "",
+               (status & PORT_OCC) ? " OCC" : "",
+               (status & PORT_OC) ? " OC" : "",
+               (status & PORT_PEC) ? " PEC" : "",
+               (status & PORT_PE) ? " PE" : "",
+               (status & PORT_CSC) ? " CSC" : "",
+               (status & PORT_CONNECT) ? " CONNECT" : ""
+           );
+}
+
+#else
+
+static inline int __attribute__((__unused__))
+dbg_status_buf(char *buf, unsigned len, const char *label, u32 status)
+{ return 0; }
+
+static inline int __attribute__((__unused__))
+dbg_command_buf(char *buf, unsigned len, const char *label, u32 command)
+{ return 0; }
+
+static inline int __attribute__((__unused__))
+dbg_intr_buf(char *buf, unsigned len, const char *label, u32 enable)
+{ return 0; }
+
+static inline int __attribute__((__unused__))
+dbg_port_buf(char *buf, unsigned len, const char *label, int port, u32 status)
+{ return 0; }
+
+#endif /* DEBUG */
+
+/* functions have the "wrong" filename when they're output... */
+#define dbg_status(oxu, label, status) { \
+       char _buf[80]; \
+       dbg_status_buf(_buf, sizeof _buf, label, status); \
+       oxu_dbg(oxu, "%s\n", _buf); \
+}
+
+#define dbg_cmd(oxu, label, command) { \
+       char _buf[80]; \
+       dbg_command_buf(_buf, sizeof _buf, label, command); \
+       oxu_dbg(oxu, "%s\n", _buf); \
+}
+
+#define dbg_port(oxu, label, port, status) { \
+       char _buf[80]; \
+       dbg_port_buf(_buf, sizeof _buf, label, port, status); \
+       oxu_dbg(oxu, "%s\n", _buf); \
+}
+
+/*
+ * Module parameters
+ */
+
+/* Initial IRQ latency: faster than hw default */
+static int log2_irq_thresh;                    /* 0 to 6 */
+module_param(log2_irq_thresh, int, S_IRUGO);
+MODULE_PARM_DESC(log2_irq_thresh, "log2 IRQ latency, 1-64 microframes");
+
+/* Initial park setting: slower than hw default */
+static unsigned park;
+module_param(park, uint, S_IRUGO);
+MODULE_PARM_DESC(park, "park setting; 1-3 back-to-back async packets");
+
+/* For flakey hardware, ignore overcurrent indicators */
+static int ignore_oc;
+module_param(ignore_oc, bool, S_IRUGO);
+MODULE_PARM_DESC(ignore_oc, "ignore bogus hardware overcurrent indications");
+
+
+static void ehci_work(struct oxu_hcd *oxu);
+static int oxu_hub_control(struct usb_hcd *hcd,
+                               u16 typeReq, u16 wValue, u16 wIndex,
+                               char *buf, u16 wLength);
+
+/*
+ * Local functions
+ */
+
+/* Low level read/write registers functions */
+static inline u32 oxu_readl(void *base, u32 reg)
+{
+       return readl(base + reg);
+}
+
+static inline void oxu_writel(void *base, u32 reg, u32 val)
+{
+       writel(val, base + reg);
+}
+
+static inline void timer_action_done(struct oxu_hcd *oxu,
+                                       enum ehci_timer_action action)
+{
+       clear_bit(action, &oxu->actions);
+}
+
+static inline void timer_action(struct oxu_hcd *oxu,
+                                       enum ehci_timer_action action)
+{
+       if (!test_and_set_bit(action, &oxu->actions)) {
+               unsigned long t;
+
+               switch (action) {
+               case TIMER_IAA_WATCHDOG:
+                       t = EHCI_IAA_JIFFIES;
+                       break;
+               case TIMER_IO_WATCHDOG:
+                       t = EHCI_IO_JIFFIES;
+                       break;
+               case TIMER_ASYNC_OFF:
+                       t = EHCI_ASYNC_JIFFIES;
+                       break;
+               case TIMER_ASYNC_SHRINK:
+               default:
+                       t = EHCI_SHRINK_JIFFIES;
+                       break;
+               }
+               t += jiffies;
+               /* all timings except IAA watchdog can be overridden.
+                * async queue SHRINK often precedes IAA.  while it's ready
+                * to go OFF neither can matter, and afterwards the IO
+                * watchdog stops unless there's still periodic traffic.
+                */
+               if (action != TIMER_IAA_WATCHDOG
+                               && t > oxu->watchdog.expires
+                               && timer_pending(&oxu->watchdog))
+                       return;
+               mod_timer(&oxu->watchdog, t);
+       }
+}
+
+/*
+ * handshake - spin reading hc until handshake completes or fails
+ * @ptr: address of hc register to be read
+ * @mask: bits to look at in result of read
+ * @done: value of those bits when handshake succeeds
+ * @usec: timeout in microseconds
+ *
+ * Returns negative errno, or zero on success
+ *
+ * Success happens when the "mask" bits have the specified value (hardware
+ * handshake done).  There are two failure modes:  "usec" have passed (major
+ * hardware flakeout), or the register reads as all-ones (hardware removed).
+ *
+ * That last failure should_only happen in cases like physical cardbus eject
+ * before driver shutdown. But it also seems to be caused by bugs in cardbus
+ * bridge shutdown:  shutting down the bridge before the devices using it.
+ */
+static int handshake(struct oxu_hcd *oxu, void __iomem *ptr,
+                                       u32 mask, u32 done, int usec)
+{
+       u32 result;
+
+       do {
+               result = readl(ptr);
+               if (result == ~(u32)0)          /* card removed */
+                       return -ENODEV;
+               result &= mask;
+               if (result == done)
+                       return 0;
+               udelay(1);
+               usec--;
+       } while (usec > 0);
+       return -ETIMEDOUT;
+}
+
+/* Force HC to halt state from unknown (EHCI spec section 2.3) */
+static int ehci_halt(struct oxu_hcd *oxu)
+{
+       u32     temp = readl(&oxu->regs->status);
+
+       /* disable any irqs left enabled by previous code */
+       writel(0, &oxu->regs->intr_enable);
+
+       if ((temp & STS_HALT) != 0)
+               return 0;
+
+       temp = readl(&oxu->regs->command);
+       temp &= ~CMD_RUN;
+       writel(temp, &oxu->regs->command);
+       return handshake(oxu, &oxu->regs->status,
+                         STS_HALT, STS_HALT, 16 * 125);
+}
+
+/* Put TDI/ARC silicon into EHCI mode */
+static void tdi_reset(struct oxu_hcd *oxu)
+{
+       u32 __iomem *reg_ptr;
+       u32 tmp;
+
+       reg_ptr = (u32 __iomem *)(((u8 __iomem *)oxu->regs) + 0x68);
+       tmp = readl(reg_ptr);
+       tmp |= 0x3;
+       writel(tmp, reg_ptr);
+}
+
+/* Reset a non-running (STS_HALT == 1) controller */
+static int ehci_reset(struct oxu_hcd *oxu)
+{
+       int     retval;
+       u32     command = readl(&oxu->regs->command);
+
+       command |= CMD_RESET;
+       dbg_cmd(oxu, "reset", command);
+       writel(command, &oxu->regs->command);
+       oxu_to_hcd(oxu)->state = HC_STATE_HALT;
+       oxu->next_statechange = jiffies;
+       retval = handshake(oxu, &oxu->regs->command,
+                           CMD_RESET, 0, 250 * 1000);
+
+       if (retval)
+               return retval;
+
+       tdi_reset(oxu);
+
+       return retval;
+}
+
+/* Idle the controller (from running) */
+static void ehci_quiesce(struct oxu_hcd *oxu)
+{
+       u32     temp;
+
+#ifdef DEBUG
+       if (!HC_IS_RUNNING(oxu_to_hcd(oxu)->state))
+               BUG();
+#endif
+
+       /* wait for any schedule enables/disables to take effect */
+       temp = readl(&oxu->regs->command) << 10;
+       temp &= STS_ASS | STS_PSS;
+       if (handshake(oxu, &oxu->regs->status, STS_ASS | STS_PSS,
+                               temp, 16 * 125) != 0) {
+               oxu_to_hcd(oxu)->state = HC_STATE_HALT;
+               return;
+       }
+
+       /* then disable anything that's still active */
+       temp = readl(&oxu->regs->command);
+       temp &= ~(CMD_ASE | CMD_IAAD | CMD_PSE);
+       writel(temp, &oxu->regs->command);
+
+       /* hardware can take 16 microframes to turn off ... */
+       if (handshake(oxu, &oxu->regs->status, STS_ASS | STS_PSS,
+                               0, 16 * 125) != 0) {
+               oxu_to_hcd(oxu)->state = HC_STATE_HALT;
+               return;
+       }
+}
+
+static int check_reset_complete(struct oxu_hcd *oxu, int index,
+                               u32 __iomem *status_reg, int port_status)
+{
+       if (!(port_status & PORT_CONNECT)) {
+               oxu->reset_done[index] = 0;
+               return port_status;
+       }
+
+       /* if reset finished and it's still not enabled -- handoff */
+       if (!(port_status & PORT_PE)) {
+               oxu_dbg(oxu, "Failed to enable port %d on root hub TT\n",
+                               index+1);
+               return port_status;
+       } else
+               oxu_dbg(oxu, "port %d high speed\n", index + 1);
+
+       return port_status;
+}
+
+static void ehci_hub_descriptor(struct oxu_hcd *oxu,
+                               struct usb_hub_descriptor *desc)
+{
+       int ports = HCS_N_PORTS(oxu->hcs_params);
+       u16 temp;
+
+       desc->bDescriptorType = 0x29;
+       desc->bPwrOn2PwrGood = 10;      /* oxu 1.0, 2.3.9 says 20ms max */
+       desc->bHubContrCurrent = 0;
+
+       desc->bNbrPorts = ports;
+       temp = 1 + (ports / 8);
+       desc->bDescLength = 7 + 2 * temp;
+
+       /* two bitmaps:  ports removable, and usb 1.0 legacy PortPwrCtrlMask */
+       memset(&desc->bitmap[0], 0, temp);
+       memset(&desc->bitmap[temp], 0xff, temp);
+
+       temp = 0x0008;                  /* per-port overcurrent reporting */
+       if (HCS_PPC(oxu->hcs_params))
+               temp |= 0x0001;         /* per-port power control */
+       else
+               temp |= 0x0002;         /* no power switching */
+       desc->wHubCharacteristics = (__force __u16)cpu_to_le16(temp);
+}
+
+
+/* Allocate an OXU210HP on-chip memory data buffer
+ *
+ * An on-chip memory data buffer is required for each OXU210HP USB transfer.
+ * Each transfer descriptor has one or more on-chip memory data buffers.
+ *
+ * Data buffers are allocated from a fix sized pool of data blocks.
+ * To minimise fragmentation and give reasonable memory utlisation,
+ * data buffers are allocated with sizes the power of 2 multiples of
+ * the block size, starting on an address a multiple of the allocated size.
+ *
+ * FIXME: callers of this function require a buffer to be allocated for
+ * len=0. This is a waste of on-chip memory and should be fix. Then this
+ * function should be changed to not allocate a buffer for len=0.
+ */
+static int oxu_buf_alloc(struct oxu_hcd *oxu, struct ehci_qtd *qtd, int len)
+{
+       int n_blocks;   /* minium blocks needed to hold len */
+       int a_blocks;   /* blocks allocated */
+       int i, j;
+
+       /* Don't allocte bigger than supported */
+       if (len > BUFFER_SIZE * BUFFER_NUM) {
+               oxu_err(oxu, "buffer too big (%d)\n", len);
+               return -ENOMEM;
+       }
+
+       spin_lock(&oxu->mem_lock);
+
+       /* Number of blocks needed to hold len */
+       n_blocks = (len + BUFFER_SIZE - 1) / BUFFER_SIZE;
+
+       /* Round the number of blocks up to the power of 2 */
+       for (a_blocks = 1; a_blocks < n_blocks; a_blocks <<= 1)
+               ;
+
+       /* Find a suitable available data buffer */
+       for (i = 0; i < BUFFER_NUM;
+                       i += max(a_blocks, (int)oxu->db_used[i])) {
+
+               /* Check all the required blocks are available */
+               for (j = 0; j < a_blocks; j++)
+                       if (oxu->db_used[i + j])
+                               break;
+
+               if (j != a_blocks)
+                       continue;
+
+               /* Allocate blocks found! */
+               qtd->buffer = (void *) &oxu->mem->db_pool[i];
+               qtd->buffer_dma = virt_to_phys(qtd->buffer);
+
+               qtd->qtd_buffer_len = BUFFER_SIZE * a_blocks;
+               oxu->db_used[i] = a_blocks;
+
+               spin_unlock(&oxu->mem_lock);
+
+               return 0;
+       }
+
+       /* Failed */
+
+       spin_unlock(&oxu->mem_lock);
+
+       return -ENOMEM;
+}
+
+static void oxu_buf_free(struct oxu_hcd *oxu, struct ehci_qtd *qtd)
+{
+       int index;
+
+       spin_lock(&oxu->mem_lock);
+
+       index = (qtd->buffer - (void *) &oxu->mem->db_pool[0])
+                                                        / BUFFER_SIZE;
+       oxu->db_used[index] = 0;
+       qtd->qtd_buffer_len = 0;
+       qtd->buffer_dma = 0;
+       qtd->buffer = NULL;
+
+       spin_unlock(&oxu->mem_lock);
+
+       return;
+}
+
+static inline void ehci_qtd_init(struct ehci_qtd *qtd, dma_addr_t dma)
+{
+       memset(qtd, 0, sizeof *qtd);
+       qtd->qtd_dma = dma;
+       qtd->hw_token = cpu_to_le32(QTD_STS_HALT);
+       qtd->hw_next = EHCI_LIST_END;
+       qtd->hw_alt_next = EHCI_LIST_END;
+       INIT_LIST_HEAD(&qtd->qtd_list);
+}
+
+static inline void oxu_qtd_free(struct oxu_hcd *oxu, struct ehci_qtd *qtd)
+{
+       int index;
+
+       if (qtd->buffer)
+               oxu_buf_free(oxu, qtd);
+
+       spin_lock(&oxu->mem_lock);
+
+       index = qtd - &oxu->mem->qtd_pool[0];
+       oxu->qtd_used[index] = 0;
+
+       spin_unlock(&oxu->mem_lock);
+
+       return;
+}
+
+static struct ehci_qtd *ehci_qtd_alloc(struct oxu_hcd *oxu)
+{
+       int i;
+       struct ehci_qtd *qtd = NULL;
+
+       spin_lock(&oxu->mem_lock);
+
+       for (i = 0; i < QTD_NUM; i++)
+               if (!oxu->qtd_used[i])
+                       break;
+
+       if (i < QTD_NUM) {
+               qtd = (struct ehci_qtd *) &oxu->mem->qtd_pool[i];
+               memset(qtd, 0, sizeof *qtd);
+
+               qtd->hw_token = cpu_to_le32(QTD_STS_HALT);
+               qtd->hw_next = EHCI_LIST_END;
+               qtd->hw_alt_next = EHCI_LIST_END;
+               INIT_LIST_HEAD(&qtd->qtd_list);
+
+               qtd->qtd_dma = virt_to_phys(qtd);
+
+               oxu->qtd_used[i] = 1;
+       }
+
+       spin_unlock(&oxu->mem_lock);
+
+       return qtd;
+}
+
+static void oxu_qh_free(struct oxu_hcd *oxu, struct ehci_qh *qh)
+{
+       int index;
+
+       spin_lock(&oxu->mem_lock);
+
+       index = qh - &oxu->mem->qh_pool[0];
+       oxu->qh_used[index] = 0;
+
+       spin_unlock(&oxu->mem_lock);
+
+       return;
+}
+
+static void qh_destroy(struct kref *kref)
+{
+       struct ehci_qh *qh = container_of(kref, struct ehci_qh, kref);
+       struct oxu_hcd *oxu = qh->oxu;
+
+       /* clean qtds first, and know this is not linked */
+       if (!list_empty(&qh->qtd_list) || qh->qh_next.ptr) {
+               oxu_dbg(oxu, "unused qh not empty!\n");
+               BUG();
+       }
+       if (qh->dummy)
+               oxu_qtd_free(oxu, qh->dummy);
+       oxu_qh_free(oxu, qh);
+}
+
+static struct ehci_qh *oxu_qh_alloc(struct oxu_hcd *oxu)
+{
+       int i;
+       struct ehci_qh *qh = NULL;
+
+       spin_lock(&oxu->mem_lock);
+
+       for (i = 0; i < QHEAD_NUM; i++)
+               if (!oxu->qh_used[i])
+                       break;
+
+       if (i < QHEAD_NUM) {
+               qh = (struct ehci_qh *) &oxu->mem->qh_pool[i];
+               memset(qh, 0, sizeof *qh);
+
+               kref_init(&qh->kref);
+               qh->oxu = oxu;
+               qh->qh_dma = virt_to_phys(qh);
+               INIT_LIST_HEAD(&qh->qtd_list);
+
+               /* dummy td enables safe urb queuing */
+               qh->dummy = ehci_qtd_alloc(oxu);
+               if (qh->dummy == NULL) {
+                       oxu_dbg(oxu, "no dummy td\n");
+                       oxu->qh_used[i] = 0;
+
+                       return NULL;
+               }
+
+               oxu->qh_used[i] = 1;
+       }
+
+       spin_unlock(&oxu->mem_lock);
+
+       return qh;
+}
+
+/* to share a qh (cpu threads, or hc) */
+static inline struct ehci_qh *qh_get(struct ehci_qh *qh)
+{
+       kref_get(&qh->kref);
+       return qh;
+}
+
+static inline void qh_put(struct ehci_qh *qh)
+{
+       kref_put(&qh->kref, qh_destroy);
+}
+
+static void oxu_murb_free(struct oxu_hcd *oxu, struct oxu_murb *murb)
+{
+       int index;
+
+       spin_lock(&oxu->mem_lock);
+
+       index = murb - &oxu->murb_pool[0];
+       oxu->murb_used[index] = 0;
+
+       spin_unlock(&oxu->mem_lock);
+
+       return;
+}
+
+static struct oxu_murb *oxu_murb_alloc(struct oxu_hcd *oxu)
+
+{
+       int i;
+       struct oxu_murb *murb = NULL;
+
+       spin_lock(&oxu->mem_lock);
+
+       for (i = 0; i < MURB_NUM; i++)
+               if (!oxu->murb_used[i])
+                       break;
+
+       if (i < MURB_NUM) {
+               murb = &(oxu->murb_pool)[i];
+
+               oxu->murb_used[i] = 1;
+       }
+
+       spin_unlock(&oxu->mem_lock);
+
+       return murb;
+}
+
+/* The queue heads and transfer descriptors are managed from pools tied
+ * to each of the "per device" structures.
+ * This is the initialisation and cleanup code.
+ */
+static void ehci_mem_cleanup(struct oxu_hcd *oxu)
+{
+       kfree(oxu->murb_pool);
+       oxu->murb_pool = NULL;
+
+       if (oxu->async)
+               qh_put(oxu->async);
+       oxu->async = NULL;
+
+       del_timer(&oxu->urb_timer);
+
+       oxu->periodic = NULL;
+
+       /* shadow periodic table */
+       kfree(oxu->pshadow);
+       oxu->pshadow = NULL;
+}
+
+/* Remember to add cleanup code (above) if you add anything here.
+ */
+static int ehci_mem_init(struct oxu_hcd *oxu, gfp_t flags)
+{
+       int i;
+
+       for (i = 0; i < oxu->periodic_size; i++)
+               oxu->mem->frame_list[i] = EHCI_LIST_END;
+       for (i = 0; i < QHEAD_NUM; i++)
+               oxu->qh_used[i] = 0;
+       for (i = 0; i < QTD_NUM; i++)
+               oxu->qtd_used[i] = 0;
+
+       oxu->murb_pool = kcalloc(MURB_NUM, sizeof(struct oxu_murb), flags);
+       if (!oxu->murb_pool)
+               goto fail;
+
+       for (i = 0; i < MURB_NUM; i++)
+               oxu->murb_used[i] = 0;
+
+       oxu->async = oxu_qh_alloc(oxu);
+       if (!oxu->async)
+               goto fail;
+
+       oxu->periodic = (__le32 *) &oxu->mem->frame_list;
+       oxu->periodic_dma = virt_to_phys(oxu->periodic);
+
+       for (i = 0; i < oxu->periodic_size; i++)
+               oxu->periodic[i] = EHCI_LIST_END;
+
+       /* software shadow of hardware table */
+       oxu->pshadow = kcalloc(oxu->periodic_size, sizeof(void *), flags);
+       if (oxu->pshadow != NULL)
+               return 0;
+
+fail:
+       oxu_dbg(oxu, "couldn't init memory\n");
+       ehci_mem_cleanup(oxu);
+       return -ENOMEM;
+}
+
+/* Fill a qtd, returning how much of the buffer we were able to queue up.
+ */
+static int qtd_fill(struct ehci_qtd *qtd, dma_addr_t buf, size_t len,
+                               int token, int maxpacket)
+{
+       int i, count;
+       u64 addr = buf;
+
+       /* one buffer entry per 4K ... first might be short or unaligned */
+       qtd->hw_buf[0] = cpu_to_le32((u32)addr);
+       qtd->hw_buf_hi[0] = cpu_to_le32((u32)(addr >> 32));
+       count = 0x1000 - (buf & 0x0fff);        /* rest of that page */
+       if (likely(len < count))                /* ... iff needed */
+               count = len;
+       else {
+               buf +=  0x1000;
+               buf &= ~0x0fff;
+
+               /* per-qtd limit: from 16K to 20K (best alignment) */
+               for (i = 1; count < len && i < 5; i++) {
+                       addr = buf;
+                       qtd->hw_buf[i] = cpu_to_le32((u32)addr);
+                       qtd->hw_buf_hi[i] = cpu_to_le32((u32)(addr >> 32));
+                       buf += 0x1000;
+                       if ((count + 0x1000) < len)
+                               count += 0x1000;
+                       else
+                               count = len;
+               }
+
+               /* short packets may only terminate transfers */
+               if (count != len)
+                       count -= (count % maxpacket);
+       }
+       qtd->hw_token = cpu_to_le32((count << 16) | token);
+       qtd->length = count;
+
+       return count;
+}
+
+static inline void qh_update(struct oxu_hcd *oxu,
+                               struct ehci_qh *qh, struct ehci_qtd *qtd)
+{
+       /* writes to an active overlay are unsafe */
+       BUG_ON(qh->qh_state != QH_STATE_IDLE);
+
+       qh->hw_qtd_next = QTD_NEXT(qtd->qtd_dma);
+       qh->hw_alt_next = EHCI_LIST_END;
+
+       /* Except for control endpoints, we make hardware maintain data
+        * toggle (like OHCI) ... here (re)initialize the toggle in the QH,
+        * and set the pseudo-toggle in udev. Only usb_clear_halt() will
+        * ever clear it.
+        */
+       if (!(qh->hw_info1 & cpu_to_le32(1 << 14))) {
+               unsigned        is_out, epnum;
+
+               is_out = !(qtd->hw_token & cpu_to_le32(1 << 8));
+               epnum = (le32_to_cpup(&qh->hw_info1) >> 8) & 0x0f;
+               if (unlikely(!usb_gettoggle(qh->dev, epnum, is_out))) {
+                       qh->hw_token &= ~__constant_cpu_to_le32(QTD_TOGGLE);
+                       usb_settoggle(qh->dev, epnum, is_out, 1);
+               }
+       }
+
+       /* HC must see latest qtd and qh data before we clear ACTIVE+HALT */
+       wmb();
+       qh->hw_token &= __constant_cpu_to_le32(QTD_TOGGLE | QTD_STS_PING);
+}
+
+/* If it weren't for a common silicon quirk (writing the dummy into the qh
+ * overlay, so qh->hw_token wrongly becomes inactive/halted), only fault
+ * recovery (including urb dequeue) would need software changes to a QH...
+ */
+static void qh_refresh(struct oxu_hcd *oxu, struct ehci_qh *qh)
+{
+       struct ehci_qtd *qtd;
+
+       if (list_empty(&qh->qtd_list))
+               qtd = qh->dummy;
+       else {
+               qtd = list_entry(qh->qtd_list.next,
+                               struct ehci_qtd, qtd_list);
+               /* first qtd may already be partially processed */
+               if (cpu_to_le32(qtd->qtd_dma) == qh->hw_current)
+                       qtd = NULL;
+       }
+
+       if (qtd)
+               qh_update(oxu, qh, qtd);
+}
+
+static void qtd_copy_status(struct oxu_hcd *oxu, struct urb *urb,
+                               size_t length, u32 token)
+{
+       /* count IN/OUT bytes, not SETUP (even short packets) */
+       if (likely(QTD_PID(token) != 2))
+               urb->actual_length += length - QTD_LENGTH(token);
+
+       /* don't modify error codes */
+       if (unlikely(urb->status != -EINPROGRESS))
+               return;
+
+       /* force cleanup after short read; not always an error */
+       if (unlikely(IS_SHORT_READ(token)))
+               urb->status = -EREMOTEIO;
+
+       /* serious "can't proceed" faults reported by the hardware */
+       if (token & QTD_STS_HALT) {
+               if (token & QTD_STS_BABBLE) {
+                       /* FIXME "must" disable babbling device's port too */
+                       urb->status = -EOVERFLOW;
+               } else if (token & QTD_STS_MMF) {
+                       /* fs/ls interrupt xfer missed the complete-split */
+                       urb->status = -EPROTO;
+               } else if (token & QTD_STS_DBE) {
+                       urb->status = (QTD_PID(token) == 1) /* IN ? */
+                               ? -ENOSR  /* hc couldn't read data */
+                               : -ECOMM; /* hc couldn't write data */
+               } else if (token & QTD_STS_XACT) {
+                       /* timeout, bad crc, wrong PID, etc; retried */
+                       if (QTD_CERR(token))
+                               urb->status = -EPIPE;
+                       else {
+                               oxu_dbg(oxu, "devpath %s ep%d%s 3strikes\n",
+                                       urb->dev->devpath,
+                                       usb_pipeendpoint(urb->pipe),
+                                       usb_pipein(urb->pipe) ? "in" : "out");
+                               urb->status = -EPROTO;
+                       }
+               /* CERR nonzero + no errors + halt --> stall */
+               } else if (QTD_CERR(token))
+                       urb->status = -EPIPE;
+               else    /* unknown */
+                       urb->status = -EPROTO;
+
+               oxu_vdbg(oxu, "dev%d ep%d%s qtd token %08x --> status %d\n",
+                       usb_pipedevice(urb->pipe),
+                       usb_pipeendpoint(urb->pipe),
+                       usb_pipein(urb->pipe) ? "in" : "out",
+                       token, urb->status);
+       }
+}
+
+static void ehci_urb_done(struct oxu_hcd *oxu, struct urb *urb)
+__releases(oxu->lock)
+__acquires(oxu->lock)
+{
+       if (likely(urb->hcpriv != NULL)) {
+               struct ehci_qh  *qh = (struct ehci_qh *) urb->hcpriv;
+
+               /* S-mask in a QH means it's an interrupt urb */
+               if ((qh->hw_info2 & __constant_cpu_to_le32(QH_SMASK)) != 0) {
+
+                       /* ... update hc-wide periodic stats (for usbfs) */
+                       oxu_to_hcd(oxu)->self.bandwidth_int_reqs--;
+               }
+               qh_put(qh);
+       }
+
+       urb->hcpriv = NULL;
+       switch (urb->status) {
+       case -EINPROGRESS:              /* success */
+               urb->status = 0;
+       default:                        /* fault */
+               break;
+       case -EREMOTEIO:                /* fault or normal */
+               if (!(urb->transfer_flags & URB_SHORT_NOT_OK))
+                       urb->status = 0;
+               break;
+       case -ECONNRESET:               /* canceled */
+       case -ENOENT:
+               break;
+       }
+
+#ifdef OXU_URB_TRACE
+       oxu_dbg(oxu, "%s %s urb %p ep%d%s status %d len %d/%d\n",
+               __func__, urb->dev->devpath, urb,
+               usb_pipeendpoint(urb->pipe),
+               usb_pipein(urb->pipe) ? "in" : "out",
+               urb->status,
+               urb->actual_length, urb->transfer_buffer_length);
+#endif
+
+       /* complete() can reenter this HCD */
+       spin_unlock(&oxu->lock);
+       usb_hcd_giveback_urb(oxu_to_hcd(oxu), urb, urb->status);
+       spin_lock(&oxu->lock);
+}
+
+static void start_unlink_async(struct oxu_hcd *oxu, struct ehci_qh *qh);
+static void unlink_async(struct oxu_hcd *oxu, struct ehci_qh *qh);
+
+static void intr_deschedule(struct oxu_hcd *oxu, struct ehci_qh *qh);
+static int qh_schedule(struct oxu_hcd *oxu, struct ehci_qh *qh);
+
+#define HALT_BIT __constant_cpu_to_le32(QTD_STS_HALT)
+
+/* Process and free completed qtds for a qh, returning URBs to drivers.
+ * Chases up to qh->hw_current.  Returns number of completions called,
+ * indicating how much "real" work we did.
+ */
+static unsigned qh_completions(struct oxu_hcd *oxu, struct ehci_qh *qh)
+{
+       struct ehci_qtd *last = NULL, *end = qh->dummy;
+       struct list_head *entry, *tmp;
+       int stopped;
+       unsigned count = 0;
+       int do_status = 0;
+       u8 state;
+       struct oxu_murb *murb = NULL;
+
+       if (unlikely(list_empty(&qh->qtd_list)))
+               return count;
+
+       /* completions (or tasks on other cpus) must never clobber HALT
+        * till we've gone through and cleaned everything up, even when
+        * they add urbs to this qh's queue or mark them for unlinking.
+        *
+        * NOTE:  unlinking expects to be done in queue order.
+        */
+       state = qh->qh_state;
+       qh->qh_state = QH_STATE_COMPLETING;
+       stopped = (state == QH_STATE_IDLE);
+
+       /* remove de-activated QTDs from front of queue.
+        * after faults (including short reads), cleanup this urb
+        * then let the queue advance.
+        * if queue is stopped, handles unlinks.
+        */
+       list_for_each_safe(entry, tmp, &qh->qtd_list) {
+               struct ehci_qtd *qtd;
+               struct urb *urb;
+               u32 token = 0;
+
+               qtd = list_entry(entry, struct ehci_qtd, qtd_list);
+               urb = qtd->urb;
+
+               /* Clean up any state from previous QTD ...*/
+               if (last) {
+                       if (likely(last->urb != urb)) {
+                               if (last->urb->complete == NULL) {
+                                       murb = (struct oxu_murb *) last->urb;
+                                       last->urb = murb->main;
+                                       if (murb->last) {
+                                               ehci_urb_done(oxu, last->urb);
+                                               count++;
+                                       }
+                                       oxu_murb_free(oxu, murb);
+                               } else {
+                                       ehci_urb_done(oxu, last->urb);
+                                       count++;
+                               }
+                       }
+                       oxu_qtd_free(oxu, last);
+                       last = NULL;
+               }
+
+               /* ignore urbs submitted during completions we reported */
+               if (qtd == end)
+                       break;
+
+               /* hardware copies qtd out of qh overlay */
+               rmb();
+               token = le32_to_cpu(qtd->hw_token);
+
+               /* always clean up qtds the hc de-activated */
+               if ((token & QTD_STS_ACTIVE) == 0) {
+
+                       if ((token & QTD_STS_HALT) != 0) {
+                               stopped = 1;
+
+                       /* magic dummy for some short reads; qh won't advance.
+                        * that silicon quirk can kick in with this dummy too.
+                        */
+                       } else if (IS_SHORT_READ(token) &&
+                                       !(qtd->hw_alt_next & EHCI_LIST_END)) {
+                               stopped = 1;
+                               goto halt;
+                       }
+
+               /* stop scanning when we reach qtds the hc is using */
+               } else if (likely(!stopped &&
+                               HC_IS_RUNNING(oxu_to_hcd(oxu)->state))) {
+                       break;
+
+               } else {
+                       stopped = 1;
+
+                       if (unlikely(!HC_IS_RUNNING(oxu_to_hcd(oxu)->state)))
+                               urb->status = -ESHUTDOWN;
+
+                       /* ignore active urbs unless some previous qtd
+                        * for the urb faulted (including short read) or
+                        * its urb was canceled.  we may patch qh or qtds.
+                        */
+                       if (likely(urb->status == -EINPROGRESS))
+                               continue;
+
+                       /* issue status after short control reads */
+                       if (unlikely(do_status != 0)
+                                       && QTD_PID(token) == 0 /* OUT */) {
+                               do_status = 0;
+                               continue;
+                       }
+
+                       /* token in overlay may be most current */
+                       if (state == QH_STATE_IDLE
+                                       && cpu_to_le32(qtd->qtd_dma)
+                                               == qh->hw_current)
+                               token = le32_to_cpu(qh->hw_token);
+
+                       /* force halt for unlinked or blocked qh, so we'll
+                        * patch the qh later and so that completions can't
+                        * activate it while we "know" it's stopped.
+                        */
+                       if ((HALT_BIT & qh->hw_token) == 0) {
+halt:
+                               qh->hw_token |= HALT_BIT;
+                               wmb();
+                       }
+               }
+
+               /* Remove it from the queue */
+               qtd_copy_status(oxu, urb->complete ?
+                                       urb : ((struct oxu_murb *) urb)->main,
+                               qtd->length, token);
+               if ((usb_pipein(qtd->urb->pipe)) &&
+                               (NULL != qtd->transfer_buffer))
+                       memcpy(qtd->transfer_buffer, qtd->buffer, qtd->length);
+               do_status = (urb->status == -EREMOTEIO)
+                               && usb_pipecontrol(urb->pipe);
+
+               if (stopped && qtd->qtd_list.prev != &qh->qtd_list) {
+                       last = list_entry(qtd->qtd_list.prev,
+                                       struct ehci_qtd, qtd_list);
+                       last->hw_next = qtd->hw_next;
+               }
+               list_del(&qtd->qtd_list);
+               last = qtd;
+       }
+
+       /* last urb's completion might still need calling */
+       if (likely(last != NULL)) {
+               if (last->urb->complete == NULL) {
+                       murb = (struct oxu_murb *) last->urb;
+                       last->urb = murb->main;
+                       if (murb->last) {
+                               ehci_urb_done(oxu, last->urb);
+                               count++;
+                       }
+                       oxu_murb_free(oxu, murb);
+               } else {
+                       ehci_urb_done(oxu, last->urb);
+                       count++;
+               }
+               oxu_qtd_free(oxu, last);
+       }
+
+       /* restore original state; caller must unlink or relink */
+       qh->qh_state = state;
+
+       /* be sure the hardware's done with the qh before refreshing
+        * it after fault cleanup, or recovering from silicon wrongly
+        * overlaying the dummy qtd (which reduces DMA chatter).
+        */
+       if (stopped != 0 || qh->hw_qtd_next == EHCI_LIST_END) {
+               switch (state) {
+               case QH_STATE_IDLE:
+                       qh_refresh(oxu, qh);
+                       break;
+               case QH_STATE_LINKED:
+                       /* should be rare for periodic transfers,
+                        * except maybe high bandwidth ...
+                        */
+                       if ((__constant_cpu_to_le32(QH_SMASK)
+                                       & qh->hw_info2) != 0) {
+                               intr_deschedule(oxu, qh);
+                               (void) qh_schedule(oxu, qh);
+                       } else
+                               unlink_async(oxu, qh);
+                       break;
+               /* otherwise, unlink already started */
+               }
+       }
+
+       return count;
+}
+
+/* High bandwidth multiplier, as encoded in highspeed endpoint descriptors */
+#define hb_mult(wMaxPacketSize)                (1 + (((wMaxPacketSize) >> 11) & 0x03))
+/* ... and packet size, for any kind of endpoint descriptor */
+#define max_packet(wMaxPacketSize)     ((wMaxPacketSize) & 0x07ff)
+
+/* Reverse of qh_urb_transaction: free a list of TDs.
+ * used for cleanup after errors, before HC sees an URB's TDs.
+ */
+static void qtd_list_free(struct oxu_hcd *oxu,
+                               struct urb *urb, struct list_head *qtd_list)
+{
+       struct list_head *entry, *temp;
+
+       list_for_each_safe(entry, temp, qtd_list) {
+               struct ehci_qtd *qtd;
+
+               qtd = list_entry(entry, struct ehci_qtd, qtd_list);
+               list_del(&qtd->qtd_list);
+               oxu_qtd_free(oxu, qtd);
+       }
+}
+
+/* Create a list of filled qtds for this URB; won't link into qh.
+ */
+static struct list_head *qh_urb_transaction(struct oxu_hcd *oxu,
+                                               struct urb *urb,
+                                               struct list_head *head,
+                                               gfp_t flags)
+{
+       struct ehci_qtd *qtd, *qtd_prev;
+       dma_addr_t buf;
+       int len, maxpacket;
+       int is_input;
+       u32 token;
+       void *transfer_buf = NULL;
+       int ret;
+
+       /*
+        * URBs map to sequences of QTDs: one logical transaction
+        */
+       qtd = ehci_qtd_alloc(oxu);
+       if (unlikely(!qtd))
+               return NULL;
+       list_add_tail(&qtd->qtd_list, head);
+       qtd->urb = urb;
+
+       token = QTD_STS_ACTIVE;
+       token |= (EHCI_TUNE_CERR << 10);
+       /* for split transactions, SplitXState initialized to zero */
+
+       len = urb->transfer_buffer_length;
+       is_input = usb_pipein(urb->pipe);
+       if (!urb->transfer_buffer && urb->transfer_buffer_length && is_input)
+               urb->transfer_buffer = phys_to_virt(urb->transfer_dma);
+
+       if (usb_pipecontrol(urb->pipe)) {
+               /* SETUP pid */
+               ret = oxu_buf_alloc(oxu, qtd, sizeof(struct usb_ctrlrequest));
+               if (ret)
+                       goto cleanup;
+
+               qtd_fill(qtd, qtd->buffer_dma, sizeof(struct usb_ctrlrequest),
+                               token | (2 /* "setup" */ << 8), 8);
+               memcpy(qtd->buffer, qtd->urb->setup_packet,
+                               sizeof(struct usb_ctrlrequest));
+
+               /* ... and always at least one more pid */
+               token ^= QTD_TOGGLE;
+               qtd_prev = qtd;
+               qtd = ehci_qtd_alloc(oxu);
+               if (unlikely(!qtd))
+                       goto cleanup;
+               qtd->urb = urb;
+               qtd_prev->hw_next = QTD_NEXT(qtd->qtd_dma);
+               list_add_tail(&qtd->qtd_list, head);
+
+               /* for zero length DATA stages, STATUS is always IN */
+               if (len == 0)
+                       token |= (1 /* "in" */ << 8);
+       }
+
+       /*
+        * Data transfer stage: buffer setup
+        */
+
+       ret = oxu_buf_alloc(oxu, qtd, len);
+       if (ret)
+               goto cleanup;
+
+       buf = qtd->buffer_dma;
+       transfer_buf = urb->transfer_buffer;
+
+       if (!is_input)
+               memcpy(qtd->buffer, qtd->urb->transfer_buffer, len);
+
+       if (is_input)
+               token |= (1 /* "in" */ << 8);
+       /* else it's already initted to "out" pid (0 << 8) */
+
+       maxpacket = max_packet(usb_maxpacket(urb->dev, urb->pipe, !is_input));
+
+       /*
+        * buffer gets wrapped in one or more qtds;
+        * last one may be "short" (including zero len)
+        * and may serve as a control status ack
+        */
+       for (;;) {
+               int this_qtd_len;
+
+               this_qtd_len = qtd_fill(qtd, buf, len, token, maxpacket);
+               qtd->transfer_buffer = transfer_buf;
+               len -= this_qtd_len;
+               buf += this_qtd_len;
+               transfer_buf += this_qtd_len;
+               if (is_input)
+                       qtd->hw_alt_next = oxu->async->hw_alt_next;
+
+               /* qh makes control packets use qtd toggle; maybe switch it */
+               if ((maxpacket & (this_qtd_len + (maxpacket - 1))) == 0)
+                       token ^= QTD_TOGGLE;
+
+               if (likely(len <= 0))
+                       break;
+
+               qtd_prev = qtd;
+               qtd = ehci_qtd_alloc(oxu);
+               if (unlikely(!qtd))
+                       goto cleanup;
+               if (likely(len > 0)) {
+                       ret = oxu_buf_alloc(oxu, qtd, len);
+                       if (ret)
+                               goto cleanup;
+               }
+               qtd->urb = urb;
+               qtd_prev->hw_next = QTD_NEXT(qtd->qtd_dma);
+               list_add_tail(&qtd->qtd_list, head);
+       }
+
+       /* unless the bulk/interrupt caller wants a chance to clean
+        * up after short reads, hc should advance qh past this urb
+        */
+       if (likely((urb->transfer_flags & URB_SHORT_NOT_OK) == 0
+                               || usb_pipecontrol(urb->pipe)))
+               qtd->hw_alt_next = EHCI_LIST_END;
+
+       /*
+        * control requests may need a terminating data "status" ack;
+        * bulk ones may need a terminating short packet (zero length).
+        */
+       if (likely(urb->transfer_buffer_length != 0)) {
+               int     one_more = 0;
+
+               if (usb_pipecontrol(urb->pipe)) {
+                       one_more = 1;
+                       token ^= 0x0100;        /* "in" <--> "out"  */
+                       token |= QTD_TOGGLE;    /* force DATA1 */
+               } else if (usb_pipebulk(urb->pipe)
+                               && (urb->transfer_flags & URB_ZERO_PACKET)
+                               && !(urb->transfer_buffer_length % maxpacket)) {
+                       one_more = 1;
+               }
+               if (one_more) {
+                       qtd_prev = qtd;
+                       qtd = ehci_qtd_alloc(oxu);
+                       if (unlikely(!qtd))
+                               goto cleanup;
+                       qtd->urb = urb;
+                       qtd_prev->hw_next = QTD_NEXT(qtd->qtd_dma);
+                       list_add_tail(&qtd->qtd_list, head);
+
+                       /* never any data in such packets */
+                       qtd_fill(qtd, 0, 0, token, 0);
+               }
+       }
+
+       /* by default, enable interrupt on urb completion */
+               qtd->hw_token |= __constant_cpu_to_le32(QTD_IOC);
+       return head;
+
+cleanup:
+       qtd_list_free(oxu, urb, head);
+       return NULL;
+}
+
+/* Each QH holds a qtd list; a QH is used for everything except iso.
+ *
+ * For interrupt urbs, the scheduler must set the microframe scheduling
+ * mask(s) each time the QH gets scheduled.  For highspeed, that's
+ * just one microframe in the s-mask.  For split interrupt transactions
+ * there are additional complications: c-mask, maybe FSTNs.
+ */
+static struct ehci_qh *qh_make(struct oxu_hcd *oxu,
+                               struct urb *urb, gfp_t flags)
+{
+       struct ehci_qh *qh = oxu_qh_alloc(oxu);
+       u32 info1 = 0, info2 = 0;
+       int is_input, type;
+       int maxp = 0;
+
+       if (!qh)
+               return qh;
+
+       /*
+        * init endpoint/device data for this QH
+        */
+       info1 |= usb_pipeendpoint(urb->pipe) << 8;
+       info1 |= usb_pipedevice(urb->pipe) << 0;
+
+       is_input = usb_pipein(urb->pipe);
+       type = usb_pipetype(urb->pipe);
+       maxp = usb_maxpacket(urb->dev, urb->pipe, !is_input);
+
+       /* Compute interrupt scheduling parameters just once, and save.
+        * - allowing for high bandwidth, how many nsec/uframe are used?
+        * - split transactions need a second CSPLIT uframe; same question
+        * - splits also need a schedule gap (for full/low speed I/O)
+        * - qh has a polling interval
+        *
+        * For control/bulk requests, the HC or TT handles these.
+        */
+       if (type == PIPE_INTERRUPT) {
+               qh->usecs = NS_TO_US(usb_calc_bus_time(USB_SPEED_HIGH,
+                                                               is_input, 0,
+                               hb_mult(maxp) * max_packet(maxp)));
+               qh->start = NO_FRAME;
+
+               if (urb->dev->speed == USB_SPEED_HIGH) {
+                       qh->c_usecs = 0;
+                       qh->gap_uf = 0;
+
+                       qh->period = urb->interval >> 3;
+                       if (qh->period == 0 && urb->interval != 1) {
+                               /* NOTE interval 2 or 4 uframes could work.
+                                * But interval 1 scheduling is simpler, and
+                                * includes high bandwidth.
+                                */
+                               dbg("intr period %d uframes, NYET!",
+                                               urb->interval);
+                               goto done;
+                       }
+               } else {
+                       struct usb_tt   *tt = urb->dev->tt;
+                       int             think_time;
+
+                       /* gap is f(FS/LS transfer times) */
+                       qh->gap_uf = 1 + usb_calc_bus_time(urb->dev->speed,
+                                       is_input, 0, maxp) / (125 * 1000);
+
+                       /* FIXME this just approximates SPLIT/CSPLIT times */
+                       if (is_input) {         /* SPLIT, gap, CSPLIT+DATA */
+                               qh->c_usecs = qh->usecs + HS_USECS(0);
+                               qh->usecs = HS_USECS(1);
+                       } else {                /* SPLIT+DATA, gap, CSPLIT */
+                               qh->usecs += HS_USECS(1);
+                               qh->c_usecs = HS_USECS(0);
+                       }
+
+                       think_time = tt ? tt->think_time : 0;
+                       qh->tt_usecs = NS_TO_US(think_time +
+                                       usb_calc_bus_time(urb->dev->speed,
+                                       is_input, 0, max_packet(maxp)));
+                       qh->period = urb->interval;
+               }
+       }
+
+       /* support for tt scheduling, and access to toggles */
+       qh->dev = urb->dev;
+
+       /* using TT? */
+       switch (urb->dev->speed) {
+       case USB_SPEED_LOW:
+               info1 |= (1 << 12);     /* EPS "low" */
+               /* FALL THROUGH */
+
+       case USB_SPEED_FULL:
+               /* EPS 0 means "full" */
+               if (type != PIPE_INTERRUPT)
+                       info1 |= (EHCI_TUNE_RL_TT << 28);
+               if (type == PIPE_CONTROL) {
+                       info1 |= (1 << 27);     /* for TT */
+                       info1 |= 1 << 14;       /* toggle from qtd */
+               }
+               info1 |= maxp << 16;
+
+               info2 |= (EHCI_TUNE_MULT_TT << 30);
+               info2 |= urb->dev->ttport << 23;
+
+               /* NOTE:  if (PIPE_INTERRUPT) { scheduler sets c-mask } */
+
+               break;
+
+       case USB_SPEED_HIGH:            /* no TT involved */
+               info1 |= (2 << 12);     /* EPS "high" */
+               if (type == PIPE_CONTROL) {
+                       info1 |= (EHCI_TUNE_RL_HS << 28);
+                       info1 |= 64 << 16;      /* usb2 fixed maxpacket */
+                       info1 |= 1 << 14;       /* toggle from qtd */
+                       info2 |= (EHCI_TUNE_MULT_HS << 30);
+               } else if (type == PIPE_BULK) {
+                       info1 |= (EHCI_TUNE_RL_HS << 28);
+                       info1 |= 512 << 16;     /* usb2 fixed maxpacket */
+                       info2 |= (EHCI_TUNE_MULT_HS << 30);
+               } else {                /* PIPE_INTERRUPT */
+                       info1 |= max_packet(maxp) << 16;
+                       info2 |= hb_mult(maxp) << 30;
+               }
+               break;
+       default:
+               dbg("bogus dev %p speed %d", urb->dev, urb->dev->speed);
+done:
+               qh_put(qh);
+               return NULL;
+       }
+
+       /* NOTE:  if (PIPE_INTERRUPT) { scheduler sets s-mask } */
+
+       /* init as live, toggle clear, advance to dummy */
+       qh->qh_state = QH_STATE_IDLE;
+       qh->hw_info1 = cpu_to_le32(info1);
+       qh->hw_info2 = cpu_to_le32(info2);
+       usb_settoggle(urb->dev, usb_pipeendpoint(urb->pipe), !is_input, 1);
+       qh_refresh(oxu, qh);
+       return qh;
+}
+
+/* Move qh (and its qtds) onto async queue; maybe enable queue.
+ */
+static void qh_link_async(struct oxu_hcd *oxu, struct ehci_qh *qh)
+{
+       __le32 dma = QH_NEXT(qh->qh_dma);
+       struct ehci_qh *head;
+
+       /* (re)start the async schedule? */
+       head = oxu->async;
+       timer_action_done(oxu, TIMER_ASYNC_OFF);
+       if (!head->qh_next.qh) {
+               u32     cmd = readl(&oxu->regs->command);
+
+               if (!(cmd & CMD_ASE)) {
+                       /* in case a clear of CMD_ASE didn't take yet */
+                       (void)handshake(oxu, &oxu->regs->status,
+                                       STS_ASS, 0, 150);
+                       cmd |= CMD_ASE | CMD_RUN;
+                       writel(cmd, &oxu->regs->command);
+                       oxu_to_hcd(oxu)->state = HC_STATE_RUNNING;
+                       /* posted write need not be known to HC yet ... */
+               }
+       }
+
+       /* clear halt and/or toggle; and maybe recover from silicon quirk */
+       if (qh->qh_state == QH_STATE_IDLE)
+               qh_refresh(oxu, qh);
+
+       /* splice right after start */
+       qh->qh_next = head->qh_next;
+       qh->hw_next = head->hw_next;
+       wmb();
+
+       head->qh_next.qh = qh;
+       head->hw_next = dma;
+
+       qh->qh_state = QH_STATE_LINKED;
+       /* qtd completions reported later by interrupt */
+}
+
+#define        QH_ADDR_MASK    __constant_cpu_to_le32(0x7f)
+
+/*
+ * For control/bulk/interrupt, return QH with these TDs appended.
+ * Allocates and initializes the QH if necessary.
+ * Returns null if it can't allocate a QH it needs to.
+ * If the QH has TDs (urbs) already, that's great.
+ */
+static struct ehci_qh *qh_append_tds(struct oxu_hcd *oxu,
+                               struct urb *urb, struct list_head *qtd_list,
+                               int epnum, void **ptr)
+{
+       struct ehci_qh *qh = NULL;
+
+       qh = (struct ehci_qh *) *ptr;
+       if (unlikely(qh == NULL)) {
+               /* can't sleep here, we have oxu->lock... */
+               qh = qh_make(oxu, urb, GFP_ATOMIC);
+               *ptr = qh;
+       }
+       if (likely(qh != NULL)) {
+               struct ehci_qtd *qtd;
+
+               if (unlikely(list_empty(qtd_list)))
+                       qtd = NULL;
+               else
+                       qtd = list_entry(qtd_list->next, struct ehci_qtd,
+                                       qtd_list);
+
+               /* control qh may need patching ... */
+               if (unlikely(epnum == 0)) {
+
+                       /* usb_reset_device() briefly reverts to address 0 */
+                       if (usb_pipedevice(urb->pipe) == 0)
+                               qh->hw_info1 &= ~QH_ADDR_MASK;
+               }
+
+               /* just one way to queue requests: swap with the dummy qtd.
+                * only hc or qh_refresh() ever modify the overlay.
+                */
+               if (likely(qtd != NULL)) {
+                       struct ehci_qtd *dummy;
+                       dma_addr_t dma;
+                       __le32 token;
+
+                       /* to avoid racing the HC, use the dummy td instead of
+                        * the first td of our list (becomes new dummy).  both
+                        * tds stay deactivated until we're done, when the
+                        * HC is allowed to fetch the old dummy (4.10.2).
+                        */
+                       token = qtd->hw_token;
+                       qtd->hw_token = HALT_BIT;
+                       wmb();
+                       dummy = qh->dummy;
+
+                       dma = dummy->qtd_dma;
+                       *dummy = *qtd;
+                       dummy->qtd_dma = dma;
+
+                       list_del(&qtd->qtd_list);
+                       list_add(&dummy->qtd_list, qtd_list);
+                       list_splice(qtd_list, qh->qtd_list.prev);
+
+                       ehci_qtd_init(qtd, qtd->qtd_dma);
+                       qh->dummy = qtd;
+
+                       /* hc must see the new dummy at list end */
+                       dma = qtd->qtd_dma;
+                       qtd = list_entry(qh->qtd_list.prev,
+                                       struct ehci_qtd, qtd_list);
+                       qtd->hw_next = QTD_NEXT(dma);
+
+                       /* let the hc process these next qtds */
+                       dummy->hw_token = (token & ~(0x80));
+                       wmb();
+                       dummy->hw_token = token;
+
+                       urb->hcpriv = qh_get(qh);
+               }
+       }
+       return qh;
+}
+
+static int submit_async(struct oxu_hcd *oxu, struct urb *urb,
+                       struct list_head *qtd_list, gfp_t mem_flags)
+{
+       struct ehci_qtd *qtd;
+       int epnum;
+       unsigned long flags;
+       struct ehci_qh *qh = NULL;
+       int rc = 0;
+
+       qtd = list_entry(qtd_list->next, struct ehci_qtd, qtd_list);
+       epnum = urb->ep->desc.bEndpointAddress;
+
+#ifdef OXU_URB_TRACE
+       oxu_dbg(oxu, "%s %s urb %p ep%d%s len %d, qtd %p [qh %p]\n",
+               __func__, urb->dev->devpath, urb,
+               epnum & 0x0f, (epnum & USB_DIR_IN) ? "in" : "out",
+               urb->transfer_buffer_length,
+               qtd, urb->ep->hcpriv);
+#endif
+
+       spin_lock_irqsave(&oxu->lock, flags);
+       if (unlikely(!test_bit(HCD_FLAG_HW_ACCESSIBLE,
+                              &oxu_to_hcd(oxu)->flags))) {
+               rc = -ESHUTDOWN;
+               goto done;
+       }
+
+       qh = qh_append_tds(oxu, urb, qtd_list, epnum, &urb->ep->hcpriv);
+       if (unlikely(qh == NULL)) {
+               rc = -ENOMEM;
+               goto done;
+       }
+
+       /* Control/bulk operations through TTs don't need scheduling,
+        * the HC and TT handle it when the TT has a buffer ready.
+        */
+       if (likely(qh->qh_state == QH_STATE_IDLE))
+               qh_link_async(oxu, qh_get(qh));
+done:
+       spin_unlock_irqrestore(&oxu->lock, flags);
+       if (unlikely(qh == NULL))
+               qtd_list_free(oxu, urb, qtd_list);
+       return rc;
+}
+
+/* The async qh for the qtds being reclaimed are now unlinked from the HC */
+
+static void end_unlink_async(struct oxu_hcd *oxu)
+{
+       struct ehci_qh *qh = oxu->reclaim;
+       struct ehci_qh *next;
+
+       timer_action_done(oxu, TIMER_IAA_WATCHDOG);
+
+       qh->qh_state = QH_STATE_IDLE;
+       qh->qh_next.qh = NULL;
+       qh_put(qh);                     /* refcount from reclaim */
+
+       /* other unlink(s) may be pending (in QH_STATE_UNLINK_WAIT) */
+       next = qh->reclaim;
+       oxu->reclaim = next;
+       oxu->reclaim_ready = 0;
+       qh->reclaim = NULL;
+
+       qh_completions(oxu, qh);
+
+       if (!list_empty(&qh->qtd_list)
+                       && HC_IS_RUNNING(oxu_to_hcd(oxu)->state))
+               qh_link_async(oxu, qh);
+       else {
+               qh_put(qh);             /* refcount from async list */
+
+               /* it's not free to turn the async schedule on/off; leave it
+                * active but idle for a while once it empties.
+                */
+               if (HC_IS_RUNNING(oxu_to_hcd(oxu)->state)
+                               && oxu->async->qh_next.qh == NULL)
+                       timer_action(oxu, TIMER_ASYNC_OFF);
+       }
+
+       if (next) {
+               oxu->reclaim = NULL;
+               start_unlink_async(oxu, next);
+       }
+}
+
+/* makes sure the async qh will become idle */
+/* caller must own oxu->lock */
+
+static void start_unlink_async(struct oxu_hcd *oxu, struct ehci_qh *qh)
+{
+       int cmd = readl(&oxu->regs->command);
+       struct ehci_qh *prev;
+
+#ifdef DEBUG
+       assert_spin_locked(&oxu->lock);
+       if (oxu->reclaim || (qh->qh_state != QH_STATE_LINKED
+                               && qh->qh_state != QH_STATE_UNLINK_WAIT))
+               BUG();
+#endif
+
+       /* stop async schedule right now? */
+       if (unlikely(qh == oxu->async)) {
+               /* can't get here without STS_ASS set */
+               if (oxu_to_hcd(oxu)->state != HC_STATE_HALT
+                               && !oxu->reclaim) {
+                       /* ... and CMD_IAAD clear */
+                       writel(cmd & ~CMD_ASE, &oxu->regs->command);
+                       wmb();
+                       /* handshake later, if we need to */
+                       timer_action_done(oxu, TIMER_ASYNC_OFF);
+               }
+               return;
+       }
+
+       qh->qh_state = QH_STATE_UNLINK;
+       oxu->reclaim = qh = qh_get(qh);
+
+       prev = oxu->async;
+       while (prev->qh_next.qh != qh)
+               prev = prev->qh_next.qh;
+
+       prev->hw_next = qh->hw_next;
+       prev->qh_next = qh->qh_next;
+       wmb();
+
+       if (unlikely(oxu_to_hcd(oxu)->state == HC_STATE_HALT)) {
+               /* if (unlikely(qh->reclaim != 0))
+                *      this will recurse, probably not much
+                */
+               end_unlink_async(oxu);
+               return;
+       }
+
+       oxu->reclaim_ready = 0;
+       cmd |= CMD_IAAD;
+       writel(cmd, &oxu->regs->command);
+       (void) readl(&oxu->regs->command);
+       timer_action(oxu, TIMER_IAA_WATCHDOG);
+}
+
+static void scan_async(struct oxu_hcd *oxu)
+{
+       struct ehci_qh *qh;
+       enum ehci_timer_action action = TIMER_IO_WATCHDOG;
+
+       if (!++(oxu->stamp))
+               oxu->stamp++;
+       timer_action_done(oxu, TIMER_ASYNC_SHRINK);
+rescan:
+       qh = oxu->async->qh_next.qh;
+       if (likely(qh != NULL)) {
+               do {
+                       /* clean any finished work for this qh */
+                       if (!list_empty(&qh->qtd_list)
+                                       && qh->stamp != oxu->stamp) {
+                               int temp;
+
+                               /* unlinks could happen here; completion
+                                * reporting drops the lock.  rescan using
+                                * the latest schedule, but don't rescan
+                                * qhs we already finished (no looping).
+                                */
+                               qh = qh_get(qh);
+                               qh->stamp = oxu->stamp;
+                               temp = qh_completions(oxu, qh);
+                               qh_put(qh);
+                               if (temp != 0)
+                                       goto rescan;
+                       }
+
+                       /* unlink idle entries, reducing HC PCI usage as well
+                        * as HCD schedule-scanning costs.  delay for any qh
+                        * we just scanned, there's a not-unusual case that it
+                        * doesn't stay idle for long.
+                        * (plus, avoids some kind of re-activation race.)
+                        */
+                       if (list_empty(&qh->qtd_list)) {
+                               if (qh->stamp == oxu->stamp)
+                                       action = TIMER_ASYNC_SHRINK;
+                               else if (!oxu->reclaim
+                                           && qh->qh_state == QH_STATE_LINKED)
+                                       start_unlink_async(oxu, qh);
+                       }
+
+                       qh = qh->qh_next.qh;
+               } while (qh);
+       }
+       if (action == TIMER_ASYNC_SHRINK)
+               timer_action(oxu, TIMER_ASYNC_SHRINK);
+}
+
+/*
+ * periodic_next_shadow - return "next" pointer on shadow list
+ * @periodic: host pointer to qh/itd/sitd
+ * @tag: hardware tag for type of this record
+ */
+static union ehci_shadow *periodic_next_shadow(union ehci_shadow *periodic,
+                                               __le32 tag)
+{
+       switch (tag) {
+       default:
+       case Q_TYPE_QH:
+               return &periodic->qh->qh_next;
+       }
+}
+
+/* caller must hold oxu->lock */
+static void periodic_unlink(struct oxu_hcd *oxu, unsigned frame, void *ptr)
+{
+       union ehci_shadow *prev_p = &oxu->pshadow[frame];
+       __le32 *hw_p = &oxu->periodic[frame];
+       union ehci_shadow here = *prev_p;
+
+       /* find predecessor of "ptr"; hw and shadow lists are in sync */
+       while (here.ptr && here.ptr != ptr) {
+               prev_p = periodic_next_shadow(prev_p, Q_NEXT_TYPE(*hw_p));
+               hw_p = here.hw_next;
+               here = *prev_p;
+       }
+       /* an interrupt entry (at list end) could have been shared */
+       if (!here.ptr)
+               return;
+
+       /* update shadow and hardware lists ... the old "next" pointers
+        * from ptr may still be in use, the caller updates them.
+        */
+       *prev_p = *periodic_next_shadow(&here, Q_NEXT_TYPE(*hw_p));
+       *hw_p = *here.hw_next;
+}
+
+/* how many of the uframe's 125 usecs are allocated? */
+static unsigned short periodic_usecs(struct oxu_hcd *oxu,
+                                       unsigned frame, unsigned uframe)
+{
+       __le32 *hw_p = &oxu->periodic[frame];
+       union ehci_shadow *q = &oxu->pshadow[frame];
+       unsigned usecs = 0;
+
+       while (q->ptr) {
+               switch (Q_NEXT_TYPE(*hw_p)) {
+               case Q_TYPE_QH:
+               default:
+                       /* is it in the S-mask? */
+                       if (q->qh->hw_info2 & cpu_to_le32(1 << uframe))
+                               usecs += q->qh->usecs;
+                       /* ... or C-mask? */
+                       if (q->qh->hw_info2 & cpu_to_le32(1 << (8 + uframe)))
+                               usecs += q->qh->c_usecs;
+                       hw_p = &q->qh->hw_next;
+                       q = &q->qh->qh_next;
+                       break;
+               }
+       }
+#ifdef DEBUG
+       if (usecs > 100)
+               oxu_err(oxu, "uframe %d sched overrun: %d usecs\n",
+                                               frame * 8 + uframe, usecs);
+#endif
+       return usecs;
+}
+
+static int enable_periodic(struct oxu_hcd *oxu)
+{
+       u32 cmd;
+       int status;
+
+       /* did clearing PSE did take effect yet?
+        * takes effect only at frame boundaries...
+        */
+       status = handshake(oxu, &oxu->regs->status, STS_PSS, 0, 9 * 125);
+       if (status != 0) {
+               oxu_to_hcd(oxu)->state = HC_STATE_HALT;
+               return status;
+       }
+
+       cmd = readl(&oxu->regs->command) | CMD_PSE;
+       writel(cmd, &oxu->regs->command);
+       /* posted write ... PSS happens later */
+       oxu_to_hcd(oxu)->state = HC_STATE_RUNNING;
+
+       /* make sure ehci_work scans these */
+       oxu->next_uframe = readl(&oxu->regs->frame_index)
+               % (oxu->periodic_size << 3);
+       return 0;
+}
+
+static int disable_periodic(struct oxu_hcd *oxu)
+{
+       u32 cmd;
+       int status;
+
+       /* did setting PSE not take effect yet?
+        * takes effect only at frame boundaries...
+        */
+       status = handshake(oxu, &oxu->regs->status, STS_PSS, STS_PSS, 9 * 125);
+       if (status != 0) {
+               oxu_to_hcd(oxu)->state = HC_STATE_HALT;
+               return status;
+       }
+
+       cmd = readl(&oxu->regs->command) & ~CMD_PSE;
+       writel(cmd, &oxu->regs->command);
+       /* posted write ... */
+
+       oxu->next_uframe = -1;
+       return 0;
+}
+
+/* periodic schedule slots have iso tds (normal or split) first, then a
+ * sparse tree for active interrupt transfers.
+ *
+ * this just links in a qh; caller guarantees uframe masks are set right.
+ * no FSTN support (yet; oxu 0.96+)
+ */
+static int qh_link_periodic(struct oxu_hcd *oxu, struct ehci_qh *qh)
+{
+       unsigned i;
+       unsigned period = qh->period;
+
+       dev_dbg(&qh->dev->dev,
+               "link qh%d-%04x/%p start %d [%d/%d us]\n",
+               period, le32_to_cpup(&qh->hw_info2) & (QH_CMASK | QH_SMASK),
+               qh, qh->start, qh->usecs, qh->c_usecs);
+
+       /* high bandwidth, or otherwise every microframe */
+       if (period == 0)
+               period = 1;
+
+       for (i = qh->start; i < oxu->periodic_size; i += period) {
+               union ehci_shadow       *prev = &oxu->pshadow[i];
+               __le32                  *hw_p = &oxu->periodic[i];
+               union ehci_shadow       here = *prev;
+               __le32                  type = 0;
+
+               /* skip the iso nodes at list head */
+               while (here.ptr) {
+                       type = Q_NEXT_TYPE(*hw_p);
+                       if (type == Q_TYPE_QH)
+                               break;
+                       prev = periodic_next_shadow(prev, type);
+                       hw_p = &here.qh->hw_next;
+                       here = *prev;
+               }
+
+               /* sorting each branch by period (slow-->fast)
+                * enables sharing interior tree nodes
+                */
+               while (here.ptr && qh != here.qh) {
+                       if (qh->period > here.qh->period)
+                               break;
+                       prev = &here.qh->qh_next;
+                       hw_p = &here.qh->hw_next;
+                       here = *prev;
+               }
+               /* link in this qh, unless some earlier pass did that */
+               if (qh != here.qh) {
+                       qh->qh_next = here;
+                       if (here.qh)
+                               qh->hw_next = *hw_p;
+                       wmb();
+                       prev->qh = qh;
+                       *hw_p = QH_NEXT(qh->qh_dma);
+               }
+       }
+       qh->qh_state = QH_STATE_LINKED;
+       qh_get(qh);
+
+       /* update per-qh bandwidth for usbfs */
+       oxu_to_hcd(oxu)->self.bandwidth_allocated += qh->period
+               ? ((qh->usecs + qh->c_usecs) / qh->period)
+               : (qh->usecs * 8);
+
+       /* maybe enable periodic schedule processing */
+       if (!oxu->periodic_sched++)
+               return enable_periodic(oxu);
+
+       return 0;
+}
+
+static void qh_unlink_periodic(struct oxu_hcd *oxu, struct ehci_qh *qh)
+{
+       unsigned i;
+       unsigned period;
+
+       /* FIXME:
+        *   IF this isn't high speed
+        *   and this qh is active in the current uframe
+        *   (and overlay token SplitXstate is false?)
+        * THEN
+        *   qh->hw_info1 |= __constant_cpu_to_le32(1 << 7 "ignore");
+        */
+
+       /* high bandwidth, or otherwise part of every microframe */
+       period = qh->period;
+       if (period == 0)
+               period = 1;
+
+       for (i = qh->start; i < oxu->periodic_size; i += period)
+               periodic_unlink(oxu, i, qh);
+
+       /* update per-qh bandwidth for usbfs */
+       oxu_to_hcd(oxu)->self.bandwidth_allocated -= qh->period
+               ? ((qh->usecs + qh->c_usecs) / qh->period)
+               : (qh->usecs * 8);
+
+       dev_dbg(&qh->dev->dev,
+               "unlink qh%d-%04x/%p start %d [%d/%d us]\n",
+               qh->period,
+               le32_to_cpup(&qh->hw_info2) & (QH_CMASK | QH_SMASK),
+               qh, qh->start, qh->usecs, qh->c_usecs);
+
+       /* qh->qh_next still "live" to HC */
+       qh->qh_state = QH_STATE_UNLINK;
+       qh->qh_next.ptr = NULL;
+       qh_put(qh);
+
+       /* maybe turn off periodic schedule */
+       oxu->periodic_sched--;
+       if (!oxu->periodic_sched)
+               (void) disable_periodic(oxu);
+}
+
+static void intr_deschedule(struct oxu_hcd *oxu, struct ehci_qh *qh)
+{
+       unsigned wait;
+
+       qh_unlink_periodic(oxu, qh);
+
+       /* simple/paranoid:  always delay, expecting the HC needs to read
+        * qh->hw_next or finish a writeback after SPLIT/CSPLIT ... and
+        * expect khubd to clean up after any CSPLITs we won't issue.
+        * active high speed queues may need bigger delays...
+        */
+       if (list_empty(&qh->qtd_list)
+               || (__constant_cpu_to_le32(QH_CMASK) & qh->hw_info2) != 0)
+               wait = 2;
+       else
+               wait = 55;      /* worst case: 3 * 1024 */
+
+       udelay(wait);
+       qh->qh_state = QH_STATE_IDLE;
+       qh->hw_next = EHCI_LIST_END;
+       wmb();
+}
+
+static int check_period(struct oxu_hcd *oxu,
+                       unsigned frame, unsigned uframe,
+                       unsigned period, unsigned usecs)
+{
+       int claimed;
+
+       /* complete split running into next frame?
+        * given FSTN support, we could sometimes check...
+        */
+       if (uframe >= 8)
+               return 0;
+
+       /*
+        * 80% periodic == 100 usec/uframe available
+        * convert "usecs we need" to "max already claimed"
+        */
+       usecs = 100 - usecs;
+
+       /* we "know" 2 and 4 uframe intervals were rejected; so
+        * for period 0, check _every_ microframe in the schedule.
+        */
+       if (unlikely(period == 0)) {
+               do {
+                       for (uframe = 0; uframe < 7; uframe++) {
+                               claimed = periodic_usecs(oxu, frame, uframe);
+                               if (claimed > usecs)
+                                       return 0;
+                       }
+               } while ((frame += 1) < oxu->periodic_size);
+
+       /* just check the specified uframe, at that period */
+       } else {
+               do {
+                       claimed = periodic_usecs(oxu, frame, uframe);
+                       if (claimed > usecs)
+                               return 0;
+               } while ((frame += period) < oxu->periodic_size);
+       }
+
+       return 1;
+}
+
+static int check_intr_schedule(struct oxu_hcd  *oxu,
+                               unsigned frame, unsigned uframe,
+                               const struct ehci_qh *qh, __le32 *c_maskp)
+{
+       int retval = -ENOSPC;
+
+       if (qh->c_usecs && uframe >= 6)         /* FSTN territory? */
+               goto done;
+
+       if (!check_period(oxu, frame, uframe, qh->period, qh->usecs))
+               goto done;
+       if (!qh->c_usecs) {
+               retval = 0;
+               *c_maskp = 0;
+               goto done;
+       }
+
+done:
+       return retval;
+}
+
+/* "first fit" scheduling policy used the first time through,
+ * or when the previous schedule slot can't be re-used.
+ */
+static int qh_schedule(struct oxu_hcd *oxu, struct ehci_qh *qh)
+{
+       int             status;
+       unsigned        uframe;
+       __le32          c_mask;
+       unsigned        frame;          /* 0..(qh->period - 1), or NO_FRAME */
+
+       qh_refresh(oxu, qh);
+       qh->hw_next = EHCI_LIST_END;
+       frame = qh->start;
+
+       /* reuse the previous schedule slots, if we can */
+       if (frame < qh->period) {
+               uframe = ffs(le32_to_cpup(&qh->hw_info2) & QH_SMASK);
+               status = check_intr_schedule(oxu, frame, --uframe,
+                               qh, &c_mask);
+       } else {
+               uframe = 0;
+               c_mask = 0;
+               status = -ENOSPC;
+       }
+
+       /* else scan the schedule to find a group of slots such that all
+        * uframes have enough periodic bandwidth available.
+        */
+       if (status) {
+               /* "normal" case, uframing flexible except with splits */
+               if (qh->period) {
+                       frame = qh->period - 1;
+                       do {
+                               for (uframe = 0; uframe < 8; uframe++) {
+                                       status = check_intr_schedule(oxu,
+                                                       frame, uframe, qh,
+                                                       &c_mask);
+                                       if (status == 0)
+                                               break;
+                               }
+                       } while (status && frame--);
+
+               /* qh->period == 0 means every uframe */
+               } else {
+                       frame = 0;
+                       status = check_intr_schedule(oxu, 0, 0, qh, &c_mask);
+               }
+               if (status)
+                       goto done;
+               qh->start = frame;
+
+               /* reset S-frame and (maybe) C-frame masks */
+               qh->hw_info2 &= __constant_cpu_to_le32(~(QH_CMASK | QH_SMASK));
+               qh->hw_info2 |= qh->period
+                       ? cpu_to_le32(1 << uframe)
+                       : __constant_cpu_to_le32(QH_SMASK);
+               qh->hw_info2 |= c_mask;
+       } else
+               oxu_dbg(oxu, "reused qh %p schedule\n", qh);
+
+       /* stuff into the periodic schedule */
+       status = qh_link_periodic(oxu, qh);
+done:
+       return status;
+}
+
+static int intr_submit(struct oxu_hcd *oxu, struct urb *urb,
+                       struct list_head *qtd_list, gfp_t mem_flags)
+{
+       unsigned epnum;
+       unsigned long flags;
+       struct ehci_qh *qh;
+       int status = 0;
+       struct list_head        empty;
+
+       /* get endpoint and transfer/schedule data */
+       epnum = urb->ep->desc.bEndpointAddress;
+
+       spin_lock_irqsave(&oxu->lock, flags);
+
+       if (unlikely(!test_bit(HCD_FLAG_HW_ACCESSIBLE,
+                              &oxu_to_hcd(oxu)->flags))) {
+               status = -ESHUTDOWN;
+               goto done;
+       }
+
+       /* get qh and force any scheduling errors */
+       INIT_LIST_HEAD(&empty);
+       qh = qh_append_tds(oxu, urb, &empty, epnum, &urb->ep->hcpriv);
+       if (qh == NULL) {
+               status = -ENOMEM;
+               goto done;
+       }
+       if (qh->qh_state == QH_STATE_IDLE) {
+               status = qh_schedule(oxu, qh);
+               if (status != 0)
+                       goto done;
+       }
+
+       /* then queue the urb's tds to the qh */
+       qh = qh_append_tds(oxu, urb, qtd_list, epnum, &urb->ep->hcpriv);
+       BUG_ON(qh == NULL);
+
+       /* ... update usbfs periodic stats */
+       oxu_to_hcd(oxu)->self.bandwidth_int_reqs++;
+
+done:
+       spin_unlock_irqrestore(&oxu->lock, flags);
+       if (status)
+               qtd_list_free(oxu, urb, qtd_list);
+
+       return status;
+}
+
+static inline int itd_submit(struct oxu_hcd *oxu, struct urb *urb,
+                                               gfp_t mem_flags)
+{
+       oxu_dbg(oxu, "iso support is missing!\n");
+       return -ENOSYS;
+}
+
+static inline int sitd_submit(struct oxu_hcd *oxu, struct urb *urb,
+                                               gfp_t mem_flags)
+{
+       oxu_dbg(oxu, "split iso support is missing!\n");
+       return -ENOSYS;
+}
+
+static void scan_periodic(struct oxu_hcd *oxu)
+{
+       unsigned frame, clock, now_uframe, mod;
+       unsigned modified;
+
+       mod = oxu->periodic_size << 3;
+
+       /*
+        * When running, scan from last scan point up to "now"
+        * else clean up by scanning everything that's left.
+        * Touches as few pages as possible:  cache-friendly.
+        */
+       now_uframe = oxu->next_uframe;
+       if (HC_IS_RUNNING(oxu_to_hcd(oxu)->state))
+               clock = readl(&oxu->regs->frame_index);
+       else
+               clock = now_uframe + mod - 1;
+       clock %= mod;
+
+       for (;;) {
+               union ehci_shadow       q, *q_p;
+               __le32                  type, *hw_p;
+               unsigned                uframes;
+
+               /* don't scan past the live uframe */
+               frame = now_uframe >> 3;
+               if (frame == (clock >> 3))
+                       uframes = now_uframe & 0x07;
+               else {
+                       /* safe to scan the whole frame at once */
+                       now_uframe |= 0x07;
+                       uframes = 8;
+               }
+
+restart:
+               /* scan each element in frame's queue for completions */
+               q_p = &oxu->pshadow[frame];
+               hw_p = &oxu->periodic[frame];
+               q.ptr = q_p->ptr;
+               type = Q_NEXT_TYPE(*hw_p);
+               modified = 0;
+
+               while (q.ptr != NULL) {
+                       union ehci_shadow temp;
+                       int live;
+
+                       live = HC_IS_RUNNING(oxu_to_hcd(oxu)->state);
+                       switch (type) {
+                       case Q_TYPE_QH:
+                               /* handle any completions */
+                               temp.qh = qh_get(q.qh);
+                               type = Q_NEXT_TYPE(q.qh->hw_next);
+                               q = q.qh->qh_next;
+                               modified = qh_completions(oxu, temp.qh);
+                               if (unlikely(list_empty(&temp.qh->qtd_list)))
+                                       intr_deschedule(oxu, temp.qh);
+                               qh_put(temp.qh);
+                               break;
+                       default:
+                               dbg("corrupt type %d frame %d shadow %p",
+                                       type, frame, q.ptr);
+                               q.ptr = NULL;
+                       }
+
+                       /* assume completion callbacks modify the queue */
+                       if (unlikely(modified))
+                               goto restart;
+               }
+
+               /* Stop when we catch up to the HC */
+
+               /* FIXME:  this assumes we won't get lapped when
+                * latencies climb; that should be rare, but...
+                * detect it, and just go all the way around.
+                * FLR might help detect this case, so long as latencies
+                * don't exceed periodic_size msec (default 1.024 sec).
+                */
+
+               /* FIXME: likewise assumes HC doesn't halt mid-scan */
+
+               if (now_uframe == clock) {
+                       unsigned        now;
+
+                       if (!HC_IS_RUNNING(oxu_to_hcd(oxu)->state))
+                               break;
+                       oxu->next_uframe = now_uframe;
+                       now = readl(&oxu->regs->frame_index) % mod;
+                       if (now_uframe == now)
+                               break;
+
+                       /* rescan the rest of this frame, then ... */
+                       clock = now;
+               } else {
+                       now_uframe++;
+                       now_uframe %= mod;
+               }
+       }
+}
+
+/* On some systems, leaving remote wakeup enabled prevents system shutdown.
+ * The firmware seems to think that powering off is a wakeup event!
+ * This routine turns off remote wakeup and everything else, on all ports.
+ */
+static void ehci_turn_off_all_ports(struct oxu_hcd *oxu)
+{
+       int port = HCS_N_PORTS(oxu->hcs_params);
+
+       while (port--)
+               writel(PORT_RWC_BITS, &oxu->regs->port_status[port]);
+}
+
+static void ehci_port_power(struct oxu_hcd *oxu, int is_on)
+{
+       unsigned port;
+
+       if (!HCS_PPC(oxu->hcs_params))
+               return;
+
+       oxu_dbg(oxu, "...power%s ports...\n", is_on ? "up" : "down");
+       for (port = HCS_N_PORTS(oxu->hcs_params); port > 0; )
+               (void) oxu_hub_control(oxu_to_hcd(oxu),
+                               is_on ? SetPortFeature : ClearPortFeature,
+                               USB_PORT_FEAT_POWER,
+                               port--, NULL, 0);
+       msleep(20);
+}
+
+/* Called from some interrupts, timers, and so on.
+ * It calls driver completion functions, after dropping oxu->lock.
+ */
+static void ehci_work(struct oxu_hcd *oxu)
+{
+       timer_action_done(oxu, TIMER_IO_WATCHDOG);
+       if (oxu->reclaim_ready)
+               end_unlink_async(oxu);
+
+       /* another CPU may drop oxu->lock during a schedule scan while
+        * it reports urb completions.  this flag guards against bogus
+        * attempts at re-entrant schedule scanning.
+        */
+       if (oxu->scanning)
+               return;
+       oxu->scanning = 1;
+       scan_async(oxu);
+       if (oxu->next_uframe != -1)
+               scan_periodic(oxu);
+       oxu->scanning = 0;
+
+       /* the IO watchdog guards against hardware or driver bugs that
+        * misplace IRQs, and should let us run completely without IRQs.
+        * such lossage has been observed on both VT6202 and VT8235.
+        */
+       if (HC_IS_RUNNING(oxu_to_hcd(oxu)->state) &&
+                       (oxu->async->qh_next.ptr != NULL ||
+                        oxu->periodic_sched != 0))
+               timer_action(oxu, TIMER_IO_WATCHDOG);
+}
+
+static void unlink_async(struct oxu_hcd *oxu, struct ehci_qh *qh)
+{
+       /* if we need to use IAA and it's busy, defer */
+       if (qh->qh_state == QH_STATE_LINKED
+                       && oxu->reclaim
+                       && HC_IS_RUNNING(oxu_to_hcd(oxu)->state)) {
+               struct ehci_qh          *last;
+
+               for (last = oxu->reclaim;
+                               last->reclaim;
+                               last = last->reclaim)
+                       continue;
+               qh->qh_state = QH_STATE_UNLINK_WAIT;
+               last->reclaim = qh;
+
+       /* bypass IAA if the hc can't care */
+       } else if (!HC_IS_RUNNING(oxu_to_hcd(oxu)->state) && oxu->reclaim)
+               end_unlink_async(oxu);
+
+       /* something else might have unlinked the qh by now */
+       if (qh->qh_state == QH_STATE_LINKED)
+               start_unlink_async(oxu, qh);
+}
+
+/*
+ * USB host controller methods
+ */
+
+static irqreturn_t oxu210_hcd_irq(struct usb_hcd *hcd)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+       u32 status, pcd_status = 0;
+       int bh;
+
+       spin_lock(&oxu->lock);
+
+       status = readl(&oxu->regs->status);
+
+       /* e.g. cardbus physical eject */
+       if (status == ~(u32) 0) {
+               oxu_dbg(oxu, "device removed\n");
+               goto dead;
+       }
+
+       status &= INTR_MASK;
+       if (!status) {                  /* irq sharing? */
+               spin_unlock(&oxu->lock);
+               return IRQ_NONE;
+       }
+
+       /* clear (just) interrupts */
+       writel(status, &oxu->regs->status);
+       readl(&oxu->regs->command);     /* unblock posted write */
+       bh = 0;
+
+#ifdef OXU_VERBOSE_DEBUG
+       /* unrequested/ignored: Frame List Rollover */
+       dbg_status(oxu, "irq", status);
+#endif
+
+       /* INT, ERR, and IAA interrupt rates can be throttled */
+
+       /* normal [4.15.1.2] or error [4.15.1.1] completion */
+       if (likely((status & (STS_INT|STS_ERR)) != 0))
+               bh = 1;
+
+       /* complete the unlinking of some qh [4.15.2.3] */
+       if (status & STS_IAA) {
+               oxu->reclaim_ready = 1;
+               bh = 1;
+       }
+
+       /* remote wakeup [4.3.1] */
+       if (status & STS_PCD) {
+               unsigned i = HCS_N_PORTS(oxu->hcs_params);
+               pcd_status = status;
+
+               /* resume root hub? */
+               if (!(readl(&oxu->regs->command) & CMD_RUN))
+                       usb_hcd_resume_root_hub(hcd);
+
+               while (i--) {
+                       int pstatus = readl(&oxu->regs->port_status[i]);
+
+                       if (pstatus & PORT_OWNER)
+                               continue;
+                       if (!(pstatus & PORT_RESUME)
+                                       || oxu->reset_done[i] != 0)
+                               continue;
+
+                       /* start 20 msec resume signaling from this port,
+                        * and make khubd collect PORT_STAT_C_SUSPEND to
+                        * stop that signaling.
+                        */
+                       oxu->reset_done[i] = jiffies + msecs_to_jiffies(20);
+                       oxu_dbg(oxu, "port %d remote wakeup\n", i + 1);
+                       mod_timer(&hcd->rh_timer, oxu->reset_done[i]);
+               }
+       }
+
+       /* PCI errors [4.15.2.4] */
+       if (unlikely((status & STS_FATAL) != 0)) {
+               /* bogus "fatal" IRQs appear on some chips... why?  */
+               status = readl(&oxu->regs->status);
+               dbg_cmd(oxu, "fatal", readl(&oxu->regs->command));
+               dbg_status(oxu, "fatal", status);
+               if (status & STS_HALT) {
+                       oxu_err(oxu, "fatal error\n");
+dead:
+                       ehci_reset(oxu);
+                       writel(0, &oxu->regs->configured_flag);
+                       /* generic layer kills/unlinks all urbs, then
+                        * uses oxu_stop to clean up the rest
+                        */
+                       bh = 1;
+               }
+       }
+
+       if (bh)
+               ehci_work(oxu);
+       spin_unlock(&oxu->lock);
+       if (pcd_status & STS_PCD)
+               usb_hcd_poll_rh_status(hcd);
+       return IRQ_HANDLED;
+}
+
+static irqreturn_t oxu_irq(struct usb_hcd *hcd)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+       int ret = IRQ_HANDLED;
+
+       u32 status = oxu_readl(hcd->regs, OXU_CHIPIRQSTATUS);
+       u32 enable = oxu_readl(hcd->regs, OXU_CHIPIRQEN_SET);
+
+       /* Disable all interrupt */
+       oxu_writel(hcd->regs, OXU_CHIPIRQEN_CLR, enable);
+
+       if ((oxu->is_otg && (status & OXU_USBOTGI)) ||
+               (!oxu->is_otg && (status & OXU_USBSPHI)))
+               oxu210_hcd_irq(hcd);
+       else
+               ret = IRQ_NONE;
+
+       /* Enable all interrupt back */
+       oxu_writel(hcd->regs, OXU_CHIPIRQEN_SET, enable);
+
+       return ret;
+}
+
+static void oxu_watchdog(unsigned long param)
+{
+       struct oxu_hcd  *oxu = (struct oxu_hcd *) param;
+       unsigned long flags;
+
+       spin_lock_irqsave(&oxu->lock, flags);
+
+       /* lost IAA irqs wedge things badly; seen with a vt8235 */
+       if (oxu->reclaim) {
+               u32 status = readl(&oxu->regs->status);
+               if (status & STS_IAA) {
+                       oxu_vdbg(oxu, "lost IAA\n");
+                       writel(STS_IAA, &oxu->regs->status);
+                       oxu->reclaim_ready = 1;
+               }
+       }
+
+       /* stop async processing after it's idled a bit */
+       if (test_bit(TIMER_ASYNC_OFF, &oxu->actions))
+               start_unlink_async(oxu, oxu->async);
+
+       /* oxu could run by timer, without IRQs ... */
+       ehci_work(oxu);
+
+       spin_unlock_irqrestore(&oxu->lock, flags);
+}
+
+/* One-time init, only for memory state.
+ */
+static int oxu_hcd_init(struct usb_hcd *hcd)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+       u32 temp;
+       int retval;
+       u32 hcc_params;
+
+       spin_lock_init(&oxu->lock);
+
+       init_timer(&oxu->watchdog);
+       oxu->watchdog.function = oxu_watchdog;
+       oxu->watchdog.data = (unsigned long) oxu;
+
+       /*
+        * hw default: 1K periodic list heads, one per frame.
+        * periodic_size can shrink by USBCMD update if hcc_params allows.
+        */
+       oxu->periodic_size = DEFAULT_I_TDPS;
+       retval = ehci_mem_init(oxu, GFP_KERNEL);
+       if (retval < 0)
+               return retval;
+
+       /* controllers may cache some of the periodic schedule ... */
+       hcc_params = readl(&oxu->caps->hcc_params);
+       if (HCC_ISOC_CACHE(hcc_params))         /* full frame cache */
+               oxu->i_thresh = 8;
+       else                                    /* N microframes cached */
+               oxu->i_thresh = 2 + HCC_ISOC_THRES(hcc_params);
+
+       oxu->reclaim = NULL;
+       oxu->reclaim_ready = 0;
+       oxu->next_uframe = -1;
+
+       /*
+        * dedicate a qh for the async ring head, since we couldn't unlink
+        * a 'real' qh without stopping the async schedule [4.8].  use it
+        * as the 'reclamation list head' too.
+        * its dummy is used in hw_alt_next of many tds, to prevent the qh
+        * from automatically advancing to the next td after short reads.
+        */
+       oxu->async->qh_next.qh = NULL;
+       oxu->async->hw_next = QH_NEXT(oxu->async->qh_dma);
+       oxu->async->hw_info1 = cpu_to_le32(QH_HEAD);
+       oxu->async->hw_token = cpu_to_le32(QTD_STS_HALT);
+       oxu->async->hw_qtd_next = EHCI_LIST_END;
+       oxu->async->qh_state = QH_STATE_LINKED;
+       oxu->async->hw_alt_next = QTD_NEXT(oxu->async->dummy->qtd_dma);
+
+       /* clear interrupt enables, set irq latency */
+       if (log2_irq_thresh < 0 || log2_irq_thresh > 6)
+               log2_irq_thresh = 0;
+       temp = 1 << (16 + log2_irq_thresh);
+       if (HCC_CANPARK(hcc_params)) {
+               /* HW default park == 3, on hardware that supports it (like
+                * NVidia and ALI silicon), maximizes throughput on the async
+                * schedule by avoiding QH fetches between transfers.
+                *
+                * With fast usb storage devices and NForce2, "park" seems to
+                * make problems:  throughput reduction (!), data errors...
+                */
+               if (park) {
+                       park = min(park, (unsigned) 3);
+                       temp |= CMD_PARK;
+                       temp |= park << 8;
+               }
+               oxu_dbg(oxu, "park %d\n", park);
+       }
+       if (HCC_PGM_FRAMELISTLEN(hcc_params)) {
+               /* periodic schedule size can be smaller than default */
+               temp &= ~(3 << 2);
+               temp |= (EHCI_TUNE_FLS << 2);
+       }
+       oxu->command = temp;
+
+       return 0;
+}
+
+/* Called during probe() after chip reset completes.
+ */
+static int oxu_reset(struct usb_hcd *hcd)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+       int ret;
+
+       spin_lock_init(&oxu->mem_lock);
+       INIT_LIST_HEAD(&oxu->urb_list);
+       oxu->urb_len = 0;
+
+       /* FIMXE */
+       hcd->self.controller->dma_mask = 0UL;
+
+       if (oxu->is_otg) {
+               oxu->caps = hcd->regs + OXU_OTG_CAP_OFFSET;
+               oxu->regs = hcd->regs + OXU_OTG_CAP_OFFSET + \
+                       HC_LENGTH(readl(&oxu->caps->hc_capbase));
+
+               oxu->mem = hcd->regs + OXU_SPH_MEM;
+       } else {
+               oxu->caps = hcd->regs + OXU_SPH_CAP_OFFSET;
+               oxu->regs = hcd->regs + OXU_SPH_CAP_OFFSET + \
+                       HC_LENGTH(readl(&oxu->caps->hc_capbase));
+
+               oxu->mem = hcd->regs + OXU_OTG_MEM;
+       }
+
+       oxu->hcs_params = readl(&oxu->caps->hcs_params);
+       oxu->sbrn = 0x20;
+
+       ret = oxu_hcd_init(hcd);
+       if (ret)
+               return ret;
+
+       return 0;
+}
+
+static int oxu_run(struct usb_hcd *hcd)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+       int retval;
+       u32 temp, hcc_params;
+
+       hcd->uses_new_polling = 1;
+       hcd->poll_rh = 0;
+
+       /* EHCI spec section 4.1 */
+       retval = ehci_reset(oxu);
+       if (retval != 0) {
+               ehci_mem_cleanup(oxu);
+               return retval;
+       }
+       writel(oxu->periodic_dma, &oxu->regs->frame_list);
+       writel((u32) oxu->async->qh_dma, &oxu->regs->async_next);
+
+       /* hcc_params controls whether oxu->regs->segment must (!!!)
+        * be used; it constrains QH/ITD/SITD and QTD locations.
+        * pci_pool consistent memory always uses segment zero.
+        * streaming mappings for I/O buffers, like pci_map_single(),
+        * can return segments above 4GB, if the device allows.
+        *
+        * NOTE:  the dma mask is visible through dma_supported(), so
+        * drivers can pass this info along ... like NETIF_F_HIGHDMA,
+        * Scsi_Host.highmem_io, and so forth.  It's readonly to all
+        * host side drivers though.
+        */
+       hcc_params = readl(&oxu->caps->hcc_params);
+       if (HCC_64BIT_ADDR(hcc_params))
+               writel(0, &oxu->regs->segment);
+
+       oxu->command &= ~(CMD_LRESET | CMD_IAAD | CMD_PSE |
+                               CMD_ASE | CMD_RESET);
+       oxu->command |= CMD_RUN;
+       writel(oxu->command, &oxu->regs->command);
+       dbg_cmd(oxu, "init", oxu->command);
+
+       /*
+        * Start, enabling full USB 2.0 functionality ... usb 1.1 devices
+        * are explicitly handed to companion controller(s), so no TT is
+        * involved with the root hub.  (Except where one is integrated,
+        * and there's no companion controller unless maybe for USB OTG.)
+        */
+       hcd->state = HC_STATE_RUNNING;
+       writel(FLAG_CF, &oxu->regs->configured_flag);
+       readl(&oxu->regs->command);     /* unblock posted writes */
+
+       temp = HC_VERSION(readl(&oxu->caps->hc_capbase));
+       oxu_info(oxu, "USB %x.%x started, quasi-EHCI %x.%02x, driver %s%s\n",
+               ((oxu->sbrn & 0xf0)>>4), (oxu->sbrn & 0x0f),
+               temp >> 8, temp & 0xff, DRIVER_VERSION,
+               ignore_oc ? ", overcurrent ignored" : "");
+
+       writel(INTR_MASK, &oxu->regs->intr_enable); /* Turn On Interrupts */
+
+       return 0;
+}
+
+static void oxu_stop(struct usb_hcd *hcd)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+
+       /* Turn off port power on all root hub ports. */
+       ehci_port_power(oxu, 0);
+
+       /* no more interrupts ... */
+       del_timer_sync(&oxu->watchdog);
+
+       spin_lock_irq(&oxu->lock);
+       if (HC_IS_RUNNING(hcd->state))
+               ehci_quiesce(oxu);
+
+       ehci_reset(oxu);
+       writel(0, &oxu->regs->intr_enable);
+       spin_unlock_irq(&oxu->lock);
+
+       /* let companion controllers work when we aren't */
+       writel(0, &oxu->regs->configured_flag);
+
+       /* root hub is shut down separately (first, when possible) */
+       spin_lock_irq(&oxu->lock);
+       if (oxu->async)
+               ehci_work(oxu);
+       spin_unlock_irq(&oxu->lock);
+       ehci_mem_cleanup(oxu);
+
+       dbg_status(oxu, "oxu_stop completed", readl(&oxu->regs->status));
+}
+
+/* Kick in for silicon on any bus (not just pci, etc).
+ * This forcibly disables dma and IRQs, helping kexec and other cases
+ * where the next system software may expect clean state.
+ */
+static void oxu_shutdown(struct usb_hcd *hcd)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+
+       (void) ehci_halt(oxu);
+       ehci_turn_off_all_ports(oxu);
+
+       /* make BIOS/etc use companion controller during reboot */
+       writel(0, &oxu->regs->configured_flag);
+
+       /* unblock posted writes */
+       readl(&oxu->regs->configured_flag);
+}
+
+/* Non-error returns are a promise to giveback() the urb later
+ * we drop ownership so next owner (or urb unlink) can get it
+ *
+ * urb + dev is in hcd.self.controller.urb_list
+ * we're queueing TDs onto software and hardware lists
+ *
+ * hcd-specific init for hcpriv hasn't been done yet
+ *
+ * NOTE:  control, bulk, and interrupt share the same code to append TDs
+ * to a (possibly active) QH, and the same QH scanning code.
+ */
+static int __oxu_urb_enqueue(struct usb_hcd *hcd, struct urb *urb,
+                               gfp_t mem_flags)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+       struct list_head qtd_list;
+
+       INIT_LIST_HEAD(&qtd_list);
+
+       switch (usb_pipetype(urb->pipe)) {
+       case PIPE_CONTROL:
+       case PIPE_BULK:
+       default:
+               if (!qh_urb_transaction(oxu, urb, &qtd_list, mem_flags))
+                       return -ENOMEM;
+               return submit_async(oxu, urb, &qtd_list, mem_flags);
+
+       case PIPE_INTERRUPT:
+               if (!qh_urb_transaction(oxu, urb, &qtd_list, mem_flags))
+                       return -ENOMEM;
+               return intr_submit(oxu, urb, &qtd_list, mem_flags);
+
+       case PIPE_ISOCHRONOUS:
+               if (urb->dev->speed == USB_SPEED_HIGH)
+                       return itd_submit(oxu, urb, mem_flags);
+               else
+                       return sitd_submit(oxu, urb, mem_flags);
+       }
+}
+
+/* This function is responsible for breaking URBs with big data size
+ * into smaller size and processing small urbs in sequence.
+ */
+static int oxu_urb_enqueue(struct usb_hcd *hcd, struct urb *urb,
+                               gfp_t mem_flags)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+       int num, rem;
+       int transfer_buffer_length;
+       void *transfer_buffer;
+       struct urb *murb;
+       int i, ret;
+
+       /* If not bulk pipe just enqueue the URB */
+       if (!usb_pipebulk(urb->pipe))
+               return __oxu_urb_enqueue(hcd, urb, mem_flags);
+
+       /* Otherwise we should verify the USB transfer buffer size! */
+       transfer_buffer = urb->transfer_buffer;
+       transfer_buffer_length = urb->transfer_buffer_length;
+
+       num = urb->transfer_buffer_length / 4096;
+       rem = urb->transfer_buffer_length % 4096;
+       if (rem != 0)
+               num++;
+
+       /* If URB is smaller than 4096 bytes just enqueue it! */
+       if (num == 1)
+               return __oxu_urb_enqueue(hcd, urb, mem_flags);
+
+       /* Ok, we have more job to do! :) */
+
+       for (i = 0; i < num - 1; i++) {
+               /* Get free micro URB poll till a free urb is recieved */
+
+               do {
+                       murb = (struct urb *) oxu_murb_alloc(oxu);
+                       if (!murb)
+                               schedule();
+               } while (!murb);
+
+               /* Coping the urb */
+               memcpy(murb, urb, sizeof(struct urb));
+
+               murb->transfer_buffer_length = 4096;
+               murb->transfer_buffer = transfer_buffer + i * 4096;
+
+               /* Null pointer for the encodes that this is a micro urb */
+               murb->complete = NULL;
+
+               ((struct oxu_murb *) murb)->main = urb;
+               ((struct oxu_murb *) murb)->last = 0;
+
+               /* This loop is to guarantee urb to be processed when there's
+                * not enough resources at a particular time by retrying.
+                */
+               do {
+                       ret  = __oxu_urb_enqueue(hcd, murb, mem_flags);
+                       if (ret)
+                               schedule();
+               } while (ret);
+       }
+
+       /* Last urb requires special handling  */
+
+       /* Get free micro URB poll till a free urb is recieved */
+       do {
+               murb = (struct urb *) oxu_murb_alloc(oxu);
+               if (!murb)
+                       schedule();
+       } while (!murb);
+
+       /* Coping the urb */
+       memcpy(murb, urb, sizeof(struct urb));
+
+       murb->transfer_buffer_length = rem > 0 ? rem : 4096;
+       murb->transfer_buffer = transfer_buffer + (num - 1) * 4096;
+
+       /* Null pointer for the encodes that this is a micro urb */
+       murb->complete = NULL;
+
+       ((struct oxu_murb *) murb)->main = urb;
+       ((struct oxu_murb *) murb)->last = 1;
+
+       do {
+               ret = __oxu_urb_enqueue(hcd, murb, mem_flags);
+               if (ret)
+                       schedule();
+       } while (ret);
+
+       return ret;
+}
+
+/* Remove from hardware lists.
+ * Completions normally happen asynchronously
+ */
+static int oxu_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+       struct ehci_qh *qh;
+       unsigned long flags;
+
+       spin_lock_irqsave(&oxu->lock, flags);
+       switch (usb_pipetype(urb->pipe)) {
+       case PIPE_CONTROL:
+       case PIPE_BULK:
+       default:
+               qh = (struct ehci_qh *) urb->hcpriv;
+               if (!qh)
+                       break;
+               unlink_async(oxu, qh);
+               break;
+
+       case PIPE_INTERRUPT:
+               qh = (struct ehci_qh *) urb->hcpriv;
+               if (!qh)
+                       break;
+               switch (qh->qh_state) {
+               case QH_STATE_LINKED:
+                       intr_deschedule(oxu, qh);
+                       /* FALL THROUGH */
+               case QH_STATE_IDLE:
+                       qh_completions(oxu, qh);
+                       break;
+               default:
+                       oxu_dbg(oxu, "bogus qh %p state %d\n",
+                                       qh, qh->qh_state);
+                       goto done;
+               }
+
+               /* reschedule QH iff another request is queued */
+               if (!list_empty(&qh->qtd_list)
+                               && HC_IS_RUNNING(hcd->state)) {
+                       int status;
+
+                       status = qh_schedule(oxu, qh);
+                       spin_unlock_irqrestore(&oxu->lock, flags);
+
+                       if (status != 0) {
+                               /* shouldn't happen often, but ...
+                                * FIXME kill those tds' urbs
+                                */
+                               err("can't reschedule qh %p, err %d",
+                                       qh, status);
+                       }
+                       return status;
+               }
+               break;
+       }
+done:
+       spin_unlock_irqrestore(&oxu->lock, flags);
+       return 0;
+}
+
+/* Bulk qh holds the data toggle */
+static void oxu_endpoint_disable(struct usb_hcd *hcd,
+                                       struct usb_host_endpoint *ep)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+       unsigned long           flags;
+       struct ehci_qh          *qh, *tmp;
+
+       /* ASSERT:  any requests/urbs are being unlinked */
+       /* ASSERT:  nobody can be submitting urbs for this any more */
+
+rescan:
+       spin_lock_irqsave(&oxu->lock, flags);
+       qh = ep->hcpriv;
+       if (!qh)
+               goto done;
+
+       /* endpoints can be iso streams.  for now, we don't
+        * accelerate iso completions ... so spin a while.
+        */
+       if (qh->hw_info1 == 0) {
+               oxu_vdbg(oxu, "iso delay\n");
+               goto idle_timeout;
+       }
+
+       if (!HC_IS_RUNNING(hcd->state))
+               qh->qh_state = QH_STATE_IDLE;
+       switch (qh->qh_state) {
+       case QH_STATE_LINKED:
+               for (tmp = oxu->async->qh_next.qh;
+                               tmp && tmp != qh;
+                               tmp = tmp->qh_next.qh)
+                       continue;
+               /* periodic qh self-unlinks on empty */
+               if (!tmp)
+                       goto nogood;
+               unlink_async(oxu, qh);
+               /* FALL THROUGH */
+       case QH_STATE_UNLINK:           /* wait for hw to finish? */
+idle_timeout:
+               spin_unlock_irqrestore(&oxu->lock, flags);
+               schedule_timeout_uninterruptible(1);
+               goto rescan;
+       case QH_STATE_IDLE:             /* fully unlinked */
+               if (list_empty(&qh->qtd_list)) {
+                       qh_put(qh);
+                       break;
+               }
+               /* else FALL THROUGH */
+       default:
+nogood:
+               /* caller was supposed to have unlinked any requests;
+                * that's not our job.  just leak this memory.
+                */
+               oxu_err(oxu, "qh %p (#%02x) state %d%s\n",
+                       qh, ep->desc.bEndpointAddress, qh->qh_state,
+                       list_empty(&qh->qtd_list) ? "" : "(has tds)");
+               break;
+       }
+       ep->hcpriv = NULL;
+done:
+       spin_unlock_irqrestore(&oxu->lock, flags);
+       return;
+}
+
+static int oxu_get_frame(struct usb_hcd *hcd)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+
+       return (readl(&oxu->regs->frame_index) >> 3) %
+               oxu->periodic_size;
+}
+
+/* Build "status change" packet (one or two bytes) from HC registers */
+static int oxu_hub_status_data(struct usb_hcd *hcd, char *buf)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+       u32 temp, mask, status = 0;
+       int ports, i, retval = 1;
+       unsigned long flags;
+
+       /* if !USB_SUSPEND, root hub timers won't get shut down ... */
+       if (!HC_IS_RUNNING(hcd->state))
+               return 0;
+
+       /* init status to no-changes */
+       buf[0] = 0;
+       ports = HCS_N_PORTS(oxu->hcs_params);
+       if (ports > 7) {
+               buf[1] = 0;
+               retval++;
+       }
+
+       /* Some boards (mostly VIA?) report bogus overcurrent indications,
+        * causing massive log spam unless we completely ignore them.  It
+        * may be relevant that VIA VT8235 controlers, where PORT_POWER is
+        * always set, seem to clear PORT_OCC and PORT_CSC when writing to
+        * PORT_POWER; that's surprising, but maybe within-spec.
+        */
+       if (!ignore_oc)
+               mask = PORT_CSC | PORT_PEC | PORT_OCC;
+       else
+               mask = PORT_CSC | PORT_PEC;
+
+       /* no hub change reports (bit 0) for now (power, ...) */
+
+       /* port N changes (bit N)? */
+       spin_lock_irqsave(&oxu->lock, flags);
+       for (i = 0; i < ports; i++) {
+               temp = readl(&oxu->regs->port_status[i]);
+
+               /*
+                * Return status information even for ports with OWNER set.
+                * Otherwise khubd wouldn't see the disconnect event when a
+                * high-speed device is switched over to the companion
+                * controller by the user.
+                */
+
+               if (!(temp & PORT_CONNECT))
+                       oxu->reset_done[i] = 0;
+               if ((temp & mask) != 0 || ((temp & PORT_RESUME) != 0 &&
+                               time_after_eq(jiffies, oxu->reset_done[i]))) {
+                       if (i < 7)
+                               buf[0] |= 1 << (i + 1);
+                       else
+                               buf[1] |= 1 << (i - 7);
+                       status = STS_PCD;
+               }
+       }
+       /* FIXME autosuspend idle root hubs */
+       spin_unlock_irqrestore(&oxu->lock, flags);
+       return status ? retval : 0;
+}
+
+/* Returns the speed of a device attached to a port on the root hub. */
+static inline unsigned int oxu_port_speed(struct oxu_hcd *oxu,
+                                               unsigned int portsc)
+{
+       switch ((portsc >> 26) & 3) {
+       case 0:
+               return 0;
+       case 1:
+               return 1 << USB_PORT_FEAT_LOWSPEED;
+       case 2:
+       default:
+               return 1 << USB_PORT_FEAT_HIGHSPEED;
+       }
+}
+
+#define        PORT_WAKE_BITS  (PORT_WKOC_E|PORT_WKDISC_E|PORT_WKCONN_E)
+static int oxu_hub_control(struct usb_hcd *hcd, u16 typeReq,
+                               u16 wValue, u16 wIndex, char *buf, u16 wLength)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+       int ports = HCS_N_PORTS(oxu->hcs_params);
+       u32 __iomem *status_reg = &oxu->regs->port_status[wIndex - 1];
+       u32 temp, status;
+       unsigned long   flags;
+       int retval = 0;
+       unsigned selector;
+
+       /*
+        * FIXME:  support SetPortFeatures USB_PORT_FEAT_INDICATOR.
+        * HCS_INDICATOR may say we can change LEDs to off/amber/green.
+        * (track current state ourselves) ... blink for diagnostics,
+        * power, "this is the one", etc.  EHCI spec supports this.
+        */
+
+       spin_lock_irqsave(&oxu->lock, flags);
+       switch (typeReq) {
+       case ClearHubFeature:
+               switch (wValue) {
+               case C_HUB_LOCAL_POWER:
+               case C_HUB_OVER_CURRENT:
+                       /* no hub-wide feature/status flags */
+                       break;
+               default:
+                       goto error;
+               }
+               break;
+       case ClearPortFeature:
+               if (!wIndex || wIndex > ports)
+                       goto error;
+               wIndex--;
+               temp = readl(status_reg);
+
+               /*
+                * Even if OWNER is set, so the port is owned by the
+                * companion controller, khubd needs to be able to clear
+                * the port-change status bits (especially
+                * USB_PORT_FEAT_C_CONNECTION).
+                */
+
+               switch (wValue) {
+               case USB_PORT_FEAT_ENABLE:
+                       writel(temp & ~PORT_PE, status_reg);
+                       break;
+               case USB_PORT_FEAT_C_ENABLE:
+                       writel((temp & ~PORT_RWC_BITS) | PORT_PEC, status_reg);
+                       break;
+               case USB_PORT_FEAT_SUSPEND:
+                       if (temp & PORT_RESET)
+                               goto error;
+                       if (temp & PORT_SUSPEND) {
+                               if ((temp & PORT_PE) == 0)
+                                       goto error;
+                               /* resume signaling for 20 msec */
+                               temp &= ~(PORT_RWC_BITS | PORT_WAKE_BITS);
+                               writel(temp | PORT_RESUME, status_reg);
+                               oxu->reset_done[wIndex] = jiffies
+                                               + msecs_to_jiffies(20);
+                       }
+                       break;
+               case USB_PORT_FEAT_C_SUSPEND:
+                       /* we auto-clear this feature */
+                       break;
+               case USB_PORT_FEAT_POWER:
+                       if (HCS_PPC(oxu->hcs_params))
+                               writel(temp & ~(PORT_RWC_BITS | PORT_POWER),
+                                         status_reg);
+                       break;
+               case USB_PORT_FEAT_C_CONNECTION:
+                       writel((temp & ~PORT_RWC_BITS) | PORT_CSC, status_reg);
+                       break;
+               case USB_PORT_FEAT_C_OVER_CURRENT:
+                       writel((temp & ~PORT_RWC_BITS) | PORT_OCC, status_reg);
+                       break;
+               case USB_PORT_FEAT_C_RESET:
+                       /* GetPortStatus clears reset */
+                       break;
+               default:
+                       goto error;
+               }
+               readl(&oxu->regs->command);     /* unblock posted write */
+               break;
+       case GetHubDescriptor:
+               ehci_hub_descriptor(oxu, (struct usb_hub_descriptor *)
+                       buf);
+               break;
+       case GetHubStatus:
+               /* no hub-wide feature/status flags */
+               memset(buf, 0, 4);
+               break;
+       case GetPortStatus:
+               if (!wIndex || wIndex > ports)
+                       goto error;
+               wIndex--;
+               status = 0;
+               temp = readl(status_reg);
+
+               /* wPortChange bits */
+               if (temp & PORT_CSC)
+                       status |= 1 << USB_PORT_FEAT_C_CONNECTION;
+               if (temp & PORT_PEC)
+                       status |= 1 << USB_PORT_FEAT_C_ENABLE;
+               if ((temp & PORT_OCC) && !ignore_oc)
+                       status |= 1 << USB_PORT_FEAT_C_OVER_CURRENT;
+
+               /* whoever resumes must GetPortStatus to complete it!! */
+               if (temp & PORT_RESUME) {
+
+                       /* Remote Wakeup received? */
+                       if (!oxu->reset_done[wIndex]) {
+                               /* resume signaling for 20 msec */
+                               oxu->reset_done[wIndex] = jiffies
+                                               + msecs_to_jiffies(20);
+                               /* check the port again */
+                               mod_timer(&oxu_to_hcd(oxu)->rh_timer,
+                                               oxu->reset_done[wIndex]);
+                       }
+
+                       /* resume completed? */
+                       else if (time_after_eq(jiffies,
+                                       oxu->reset_done[wIndex])) {
+                               status |= 1 << USB_PORT_FEAT_C_SUSPEND;
+                               oxu->reset_done[wIndex] = 0;
+
+                               /* stop resume signaling */
+                               temp = readl(status_reg);
+                               writel(temp & ~(PORT_RWC_BITS | PORT_RESUME),
+                                       status_reg);
+                               retval = handshake(oxu, status_reg,
+                                          PORT_RESUME, 0, 2000 /* 2msec */);
+                               if (retval != 0) {
+                                       oxu_err(oxu,
+                                               "port %d resume error %d\n",
+                                               wIndex + 1, retval);
+                                       goto error;
+                               }
+                               temp &= ~(PORT_SUSPEND|PORT_RESUME|(3<<10));
+                       }
+               }
+
+               /* whoever resets must GetPortStatus to complete it!! */
+               if ((temp & PORT_RESET)
+                               && time_after_eq(jiffies,
+                                       oxu->reset_done[wIndex])) {
+                       status |= 1 << USB_PORT_FEAT_C_RESET;
+                       oxu->reset_done[wIndex] = 0;
+
+                       /* force reset to complete */
+                       writel(temp & ~(PORT_RWC_BITS | PORT_RESET),
+                                       status_reg);
+                       /* REVISIT:  some hardware needs 550+ usec to clear
+                        * this bit; seems too long to spin routinely...
+                        */
+                       retval = handshake(oxu, status_reg,
+                                       PORT_RESET, 0, 750);
+                       if (retval != 0) {
+                               oxu_err(oxu, "port %d reset error %d\n",
+                                       wIndex + 1, retval);
+                               goto error;
+                       }
+
+                       /* see what we found out */
+                       temp = check_reset_complete(oxu, wIndex, status_reg,
+                                       readl(status_reg));
+               }
+
+               /* transfer dedicated ports to the companion hc */
+               if ((temp & PORT_CONNECT) &&
+                               test_bit(wIndex, &oxu->companion_ports)) {
+                       temp &= ~PORT_RWC_BITS;
+                       temp |= PORT_OWNER;
+                       writel(temp, status_reg);
+                       oxu_dbg(oxu, "port %d --> companion\n", wIndex + 1);
+                       temp = readl(status_reg);
+               }
+
+               /*
+                * Even if OWNER is set, there's no harm letting khubd
+                * see the wPortStatus values (they should all be 0 except
+                * for PORT_POWER anyway).
+                */
+
+               if (temp & PORT_CONNECT) {
+                       status |= 1 << USB_PORT_FEAT_CONNECTION;
+                       /* status may be from integrated TT */
+                       status |= oxu_port_speed(oxu, temp);
+               }
+               if (temp & PORT_PE)
+                       status |= 1 << USB_PORT_FEAT_ENABLE;
+               if (temp & (PORT_SUSPEND|PORT_RESUME))
+                       status |= 1 << USB_PORT_FEAT_SUSPEND;
+               if (temp & PORT_OC)
+                       status |= 1 << USB_PORT_FEAT_OVER_CURRENT;
+               if (temp & PORT_RESET)
+                       status |= 1 << USB_PORT_FEAT_RESET;
+               if (temp & PORT_POWER)
+                       status |= 1 << USB_PORT_FEAT_POWER;
+
+#ifndef        OXU_VERBOSE_DEBUG
+       if (status & ~0xffff)   /* only if wPortChange is interesting */
+#endif
+               dbg_port(oxu, "GetStatus", wIndex + 1, temp);
+               put_unaligned(cpu_to_le32(status), (__le32 *) buf);
+               break;
+       case SetHubFeature:
+               switch (wValue) {
+               case C_HUB_LOCAL_POWER:
+               case C_HUB_OVER_CURRENT:
+                       /* no hub-wide feature/status flags */
+                       break;
+               default:
+                       goto error;
+               }
+               break;
+       case SetPortFeature:
+               selector = wIndex >> 8;
+               wIndex &= 0xff;
+               if (!wIndex || wIndex > ports)
+                       goto error;
+               wIndex--;
+               temp = readl(status_reg);
+               if (temp & PORT_OWNER)
+                       break;
+
+               temp &= ~PORT_RWC_BITS;
+               switch (wValue) {
+               case USB_PORT_FEAT_SUSPEND:
+                       if ((temp & PORT_PE) == 0
+                                       || (temp & PORT_RESET) != 0)
+                               goto error;
+                       if (device_may_wakeup(&hcd->self.root_hub->dev))
+                               temp |= PORT_WAKE_BITS;
+                       writel(temp | PORT_SUSPEND, status_reg);
+                       break;
+               case USB_PORT_FEAT_POWER:
+                       if (HCS_PPC(oxu->hcs_params))
+                               writel(temp | PORT_POWER, status_reg);
+                       break;
+               case USB_PORT_FEAT_RESET:
+                       if (temp & PORT_RESUME)
+                               goto error;
+                       /* line status bits may report this as low speed,
+                        * which can be fine if this root hub has a
+                        * transaction translator built in.
+                        */
+                       oxu_vdbg(oxu, "port %d reset\n", wIndex + 1);
+                       temp |= PORT_RESET;
+                       temp &= ~PORT_PE;
+
+                       /*
+                        * caller must wait, then call GetPortStatus
+                        * usb 2.0 spec says 50 ms resets on root
+                        */
+                       oxu->reset_done[wIndex] = jiffies
+                                       + msecs_to_jiffies(50);
+                       writel(temp, status_reg);
+                       break;
+
+               /* For downstream facing ports (these):  one hub port is put
+                * into test mode according to USB2 11.24.2.13, then the hub
+                * must be reset (which for root hub now means rmmod+modprobe,
+                * or else system reboot).  See EHCI 2.3.9 and 4.14 for info
+                * about the EHCI-specific stuff.
+                */
+               case USB_PORT_FEAT_TEST:
+                       if (!selector || selector > 5)
+                               goto error;
+                       ehci_quiesce(oxu);
+                       ehci_halt(oxu);
+                       temp |= selector << 16;
+                       writel(temp, status_reg);
+                       break;
+
+               default:
+                       goto error;
+               }
+               readl(&oxu->regs->command);     /* unblock posted writes */
+               break;
+
+       default:
+error:
+               /* "stall" on error */
+               retval = -EPIPE;
+       }
+       spin_unlock_irqrestore(&oxu->lock, flags);
+       return retval;
+}
+
+#ifdef CONFIG_PM
+
+static int oxu_bus_suspend(struct usb_hcd *hcd)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+       int port;
+       int mask;
+
+       oxu_dbg(oxu, "suspend root hub\n");
+
+       if (time_before(jiffies, oxu->next_statechange))
+               msleep(5);
+
+       port = HCS_N_PORTS(oxu->hcs_params);
+       spin_lock_irq(&oxu->lock);
+
+       /* stop schedules, clean any completed work */
+       if (HC_IS_RUNNING(hcd->state)) {
+               ehci_quiesce(oxu);
+               hcd->state = HC_STATE_QUIESCING;
+       }
+       oxu->command = readl(&oxu->regs->command);
+       if (oxu->reclaim)
+               oxu->reclaim_ready = 1;
+       ehci_work(oxu);
+
+       /* Unlike other USB host controller types, EHCI doesn't have
+        * any notion of "global" or bus-wide suspend.  The driver has
+        * to manually suspend all the active unsuspended ports, and
+        * then manually resume them in the bus_resume() routine.
+        */
+       oxu->bus_suspended = 0;
+       while (port--) {
+               u32 __iomem *reg = &oxu->regs->port_status[port];
+               u32 t1 = readl(reg) & ~PORT_RWC_BITS;
+               u32 t2 = t1;
+
+               /* keep track of which ports we suspend */
+               if ((t1 & PORT_PE) && !(t1 & PORT_OWNER) &&
+                               !(t1 & PORT_SUSPEND)) {
+                       t2 |= PORT_SUSPEND;
+                       set_bit(port, &oxu->bus_suspended);
+               }
+
+               /* enable remote wakeup on all ports */
+               if (device_may_wakeup(&hcd->self.root_hub->dev))
+                       t2 |= PORT_WKOC_E|PORT_WKDISC_E|PORT_WKCONN_E;
+               else
+                       t2 &= ~(PORT_WKOC_E|PORT_WKDISC_E|PORT_WKCONN_E);
+
+               if (t1 != t2) {
+                       oxu_vdbg(oxu, "port %d, %08x -> %08x\n",
+                               port + 1, t1, t2);
+                       writel(t2, reg);
+               }
+       }
+
+       /* turn off now-idle HC */
+       del_timer_sync(&oxu->watchdog);
+       ehci_halt(oxu);
+       hcd->state = HC_STATE_SUSPENDED;
+
+       /* allow remote wakeup */
+       mask = INTR_MASK;
+       if (!device_may_wakeup(&hcd->self.root_hub->dev))
+               mask &= ~STS_PCD;
+       writel(mask, &oxu->regs->intr_enable);
+       readl(&oxu->regs->intr_enable);
+
+       oxu->next_statechange = jiffies + msecs_to_jiffies(10);
+       spin_unlock_irq(&oxu->lock);
+       return 0;
+}
+
+/* Caller has locked the root hub, and should reset/reinit on error */
+static int oxu_bus_resume(struct usb_hcd *hcd)
+{
+       struct oxu_hcd *oxu = hcd_to_oxu(hcd);
+       u32 temp;
+       int i;
+
+       if (time_before(jiffies, oxu->next_statechange))
+               msleep(5);
+       spin_lock_irq(&oxu->lock);
+
+       /* Ideally and we've got a real resume here, and no port's power
+        * was lost.  (For PCI, that means Vaux was maintained.)  But we
+        * could instead be restoring a swsusp snapshot -- so that BIOS was
+        * the last user of the controller, not reset/pm hardware keeping
+        * state we gave to it.
+        */
+       temp = readl(&oxu->regs->intr_enable);
+       oxu_dbg(oxu, "resume root hub%s\n", temp ? "" : " after power loss");
+
+       /* at least some APM implementations will try to deliver
+        * IRQs right away, so delay them until we're ready.
+        */
+       writel(0, &oxu->regs->intr_enable);
+
+       /* re-init operational registers */
+       writel(0, &oxu->regs->segment);
+       writel(oxu->periodic_dma, &oxu->regs->frame_list);
+       writel((u32) oxu->async->qh_dma, &oxu->regs->async_next);
+
+       /* restore CMD_RUN, framelist size, and irq threshold */
+       writel(oxu->command, &oxu->regs->command);
+
+       /* Some controller/firmware combinations need a delay during which
+        * they set up the port statuses.  See Bugzilla #8190. */
+       mdelay(8);
+
+       /* manually resume the ports we suspended during bus_suspend() */
+       i = HCS_N_PORTS(oxu->hcs_params);
+       while (i--) {
+               temp = readl(&oxu->regs->port_status[i]);
+               temp &= ~(PORT_RWC_BITS
+                       | PORT_WKOC_E | PORT_WKDISC_E | PORT_WKCONN_E);
+               if (test_bit(i, &oxu->bus_suspended) && (temp & PORT_SUSPEND)) {
+                       oxu->reset_done[i] = jiffies + msecs_to_jiffies(20);
+                       temp |= PORT_RESUME;
+               }
+               writel(temp, &oxu->regs->port_status[i]);
+       }
+       i = HCS_N_PORTS(oxu->hcs_params);
+       mdelay(20);
+       while (i--) {
+               temp = readl(&oxu->regs->port_status[i]);
+               if (test_bit(i, &oxu->bus_suspended) && (temp & PORT_SUSPEND)) {
+                       temp &= ~(PORT_RWC_BITS | PORT_RESUME);
+                       writel(temp, &oxu->regs->port_status[i]);
+                       oxu_vdbg(oxu, "resumed port %d\n", i + 1);
+               }
+       }
+       (void) readl(&oxu->regs->command);
+
+       /* maybe re-activate the schedule(s) */
+       temp = 0;
+       if (oxu->async->qh_next.qh)
+               temp |= CMD_ASE;
+       if (oxu->periodic_sched)
+               temp |= CMD_PSE;
+       if (temp) {
+               oxu->command |= temp;
+               writel(oxu->command, &oxu->regs->command);
+       }
+
+       oxu->next_statechange = jiffies + msecs_to_jiffies(5);
+       hcd->state = HC_STATE_RUNNING;
+
+       /* Now we can safely re-enable irqs */
+       writel(INTR_MASK, &oxu->regs->intr_enable);
+
+       spin_unlock_irq(&oxu->lock);
+       return 0;
+}
+
+#else
+
+static int oxu_bus_suspend(struct usb_hcd *hcd)
+{
+       return 0;
+}
+
+static int oxu_bus_resume(struct usb_hcd *hcd)
+{
+       return 0;
+}
+
+#endif /* CONFIG_PM */
+
+static const struct hc_driver oxu_hc_driver = {
+       .description =          "oxu210hp_hcd",
+       .product_desc =         "oxu210hp HCD",
+       .hcd_priv_size =        sizeof(struct oxu_hcd),
+
+       /*
+        * Generic hardware linkage
+        */
+       .irq =                  oxu_irq,
+       .flags =                HCD_MEMORY | HCD_USB2,
+
+       /*
+        * Basic lifecycle operations
+        */
+       .reset =                oxu_reset,
+       .start =                oxu_run,
+       .stop =                 oxu_stop,
+       .shutdown =             oxu_shutdown,
+
+       /*
+        * Managing i/o requests and associated device resources
+        */
+       .urb_enqueue =          oxu_urb_enqueue,
+       .urb_dequeue =          oxu_urb_dequeue,
+       .endpoint_disable =     oxu_endpoint_disable,
+
+       /*
+        * Scheduling support
+        */
+       .get_frame_number =     oxu_get_frame,
+
+       /*
+        * Root hub support
+        */
+       .hub_status_data =      oxu_hub_status_data,
+       .hub_control =          oxu_hub_control,
+       .bus_suspend =          oxu_bus_suspend,
+       .bus_resume =           oxu_bus_resume,
+};
+
+/*
+ * Module stuff
+ */
+
+static void oxu_configuration(struct platform_device *pdev, void *base)
+{
+       u32 tmp;
+
+       /* Initialize top level registers.
+        * First write ever
+        */
+       oxu_writel(base, OXU_HOSTIFCONFIG, 0x0000037D);
+       oxu_writel(base, OXU_SOFTRESET, OXU_SRESET);
+       oxu_writel(base, OXU_HOSTIFCONFIG, 0x0000037D);
+
+       tmp = oxu_readl(base, OXU_PIOBURSTREADCTRL);
+       oxu_writel(base, OXU_PIOBURSTREADCTRL, tmp | 0x0040);
+
+       oxu_writel(base, OXU_ASO, OXU_SPHPOEN | OXU_OVRCCURPUPDEN |
+                                       OXU_COMPARATOR | OXU_ASO_OP);
+
+       tmp = oxu_readl(base, OXU_CLKCTRL_SET);
+       oxu_writel(base, OXU_CLKCTRL_SET, tmp | OXU_SYSCLKEN | OXU_USBOTGCLKEN);
+
+       /* Clear all top interrupt enable */
+       oxu_writel(base, OXU_CHIPIRQEN_CLR, 0xff);
+
+       /* Clear all top interrupt status */
+       oxu_writel(base, OXU_CHIPIRQSTATUS, 0xff);
+
+       /* Enable all needed top interrupt except OTG SPH core */
+       oxu_writel(base, OXU_CHIPIRQEN_SET, OXU_USBSPHLPWUI | OXU_USBOTGLPWUI);
+}
+
+static int oxu_verify_id(struct platform_device *pdev, void *base)
+{
+       u32 id;
+       char *bo[] = {
+               "reserved",
+               "128-pin LQFP",
+               "84-pin TFBGA",
+               "reserved",
+       };
+
+       /* Read controller signature register to find a match */
+       id = oxu_readl(base, OXU_DEVICEID);
+       dev_info(&pdev->dev, "device ID %x\n", id);
+       if ((id & OXU_REV_MASK) != (OXU_REV_2100 << OXU_REV_SHIFT))
+               return -1;
+
+       dev_info(&pdev->dev, "found device %x %s (%04x:%04x)\n",
+               id >> OXU_REV_SHIFT,
+               bo[(id & OXU_BO_MASK) >> OXU_BO_SHIFT],
+               (id & OXU_MAJ_REV_MASK) >> OXU_MAJ_REV_SHIFT,
+               (id & OXU_MIN_REV_MASK) >> OXU_MIN_REV_SHIFT);
+
+       return 0;
+}
+
+static const struct hc_driver oxu_hc_driver;
+static struct usb_hcd *oxu_create(struct platform_device *pdev,
+                               unsigned long memstart, unsigned long memlen,
+                               void *base, int irq, int otg)
+{
+       struct device *dev = &pdev->dev;
+
+       struct usb_hcd *hcd;
+       struct oxu_hcd *oxu;
+       int ret;
+
+       /* Set endian mode and host mode */
+       oxu_writel(base + (otg ? OXU_OTG_CORE_OFFSET : OXU_SPH_CORE_OFFSET),
+                               OXU_USBMODE,
+                               OXU_CM_HOST_ONLY | OXU_ES_LITTLE | OXU_VBPS);
+
+       hcd = usb_create_hcd(&oxu_hc_driver, dev,
+                               otg ? "oxu210hp_otg" : "oxu210hp_sph");
+       if (!hcd)
+               return ERR_PTR(-ENOMEM);
+
+       hcd->rsrc_start = memstart;
+       hcd->rsrc_len = memlen;
+       hcd->regs = base;
+       hcd->irq = irq;
+       hcd->state = HC_STATE_HALT;
+
+       oxu = hcd_to_oxu(hcd);
+       oxu->is_otg = otg;
+
+       ret = usb_add_hcd(hcd, irq, IRQF_SHARED);
+       if (ret < 0)
+               return ERR_PTR(ret);
+
+       return hcd;
+}
+
+static int oxu_init(struct platform_device *pdev,
+                               unsigned long memstart, unsigned long memlen,
+                               void *base, int irq)
+{
+       struct oxu_info *info = platform_get_drvdata(pdev);
+       struct usb_hcd *hcd;
+       int ret;
+
+       /* First time configuration at start up */
+       oxu_configuration(pdev, base);
+
+       ret = oxu_verify_id(pdev, base);
+       if (ret) {
+               dev_err(&pdev->dev, "no devices found!\n");
+               return -ENODEV;
+       }
+
+       /* Create the OTG controller */
+       hcd = oxu_create(pdev, memstart, memlen, base, irq, 1);
+       if (IS_ERR(hcd)) {
+               dev_err(&pdev->dev, "cannot create OTG controller!\n");
+               ret = PTR_ERR(hcd);
+               goto error_create_otg;
+       }
+       info->hcd[0] = hcd;
+
+       /* Create the SPH host controller */
+       hcd = oxu_create(pdev, memstart, memlen, base, irq, 0);
+       if (IS_ERR(hcd)) {
+               dev_err(&pdev->dev, "cannot create SPH controller!\n");
+               ret = PTR_ERR(hcd);
+               goto error_create_sph;
+       }
+       info->hcd[1] = hcd;
+
+       oxu_writel(base, OXU_CHIPIRQEN_SET,
+               oxu_readl(base, OXU_CHIPIRQEN_SET) | 3);
+
+       return 0;
+
+error_create_sph:
+       usb_remove_hcd(info->hcd[0]);
+       usb_put_hcd(info->hcd[0]);
+
+error_create_otg:
+       return ret;
+}
+
+static int oxu_drv_probe(struct platform_device *pdev)
+{
+       struct resource *res;
+       void *base;
+       unsigned long memstart, memlen;
+       int irq, ret;
+       struct oxu_info *info;
+
+       if (usb_disabled())
+               return -ENODEV;
+
+       /*
+        * Get the platform resources
+        */
+       res = platform_get_resource(pdev, IORESOURCE_IRQ, 0);
+       if (!res) {
+               dev_err(&pdev->dev,
+                       "no IRQ! Check %s setup!\n", pdev->dev.bus_id);
+               return -ENODEV;
+       }
+       irq = res->start;
+       dev_dbg(&pdev->dev, "IRQ resource %d\n", irq);
+
+       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       if (!res) {
+               dev_err(&pdev->dev, "no registers address! Check %s setup!\n",
+                       pdev->dev.bus_id);
+               return -ENODEV;
+       }
+       memstart = res->start;
+       memlen = res->end - res->start + 1;
+       dev_dbg(&pdev->dev, "MEM resource %lx-%lx\n", memstart, memlen);
+       if (!request_mem_region(memstart, memlen,
+                               oxu_hc_driver.description)) {
+               dev_dbg(&pdev->dev, "memory area already in use\n");
+               return -EBUSY;
+       }
+
+       ret = set_irq_type(irq, IRQF_TRIGGER_FALLING);
+       if (ret) {
+               dev_err(&pdev->dev, "error setting irq type\n");
+               ret = -EFAULT;
+               goto error_set_irq_type;
+       }
+
+       base = ioremap(memstart, memlen);
+       if (!base) {
+               dev_dbg(&pdev->dev, "error mapping memory\n");
+               ret = -EFAULT;
+               goto error_ioremap;
+       }
+
+       /* Allocate a driver data struct to hold useful info for both
+        * SPH & OTG devices
+        */
+       info = kzalloc(sizeof(struct oxu_info), GFP_KERNEL);
+       if (!info) {
+               dev_dbg(&pdev->dev, "error allocating memory\n");
+               ret = -EFAULT;
+               goto error_alloc;
+       }
+       platform_set_drvdata(pdev, info);
+
+       ret = oxu_init(pdev, memstart, memlen, base, irq);
+       if (ret < 0) {
+               dev_dbg(&pdev->dev, "cannot init USB devices\n");
+               goto error_init;
+       }
+
+       dev_info(&pdev->dev, "devices enabled and running\n");
+       platform_set_drvdata(pdev, info);
+
+       return 0;
+
+error_init:
+       kfree(info);
+       platform_set_drvdata(pdev, NULL);
+
+error_alloc:
+       iounmap(base);
+
+error_set_irq_type:
+error_ioremap:
+       release_mem_region(memstart, memlen);
+
+       dev_err(&pdev->dev, "init %s fail, %d\n", pdev->dev.bus_id, ret);
+       return ret;
+}
+
+static void oxu_remove(struct platform_device *pdev, struct usb_hcd *hcd)
+{
+       usb_remove_hcd(hcd);
+       usb_put_hcd(hcd);
+}
+
+static int oxu_drv_remove(struct platform_device *pdev)
+{
+       struct oxu_info *info = platform_get_drvdata(pdev);
+       unsigned long memstart = info->hcd[0]->rsrc_start,
+                       memlen = info->hcd[0]->rsrc_len;
+       void *base = info->hcd[0]->regs;
+
+       oxu_remove(pdev, info->hcd[0]);
+       oxu_remove(pdev, info->hcd[1]);
+
+       iounmap(base);
+       release_mem_region(memstart, memlen);
+
+       kfree(info);
+       platform_set_drvdata(pdev, NULL);
+
+       return 0;
+}
+
+static void oxu_drv_shutdown(struct platform_device *pdev)
+{
+       oxu_drv_remove(pdev);
+}
+
+#if 0
+/* FIXME: TODO */
+static int oxu_drv_suspend(struct device *dev)
+{
+       struct platform_device *pdev = to_platform_device(dev);
+       struct usb_hcd *hcd = dev_get_drvdata(dev);
+
+       return 0;
+}
+
+static int oxu_drv_resume(struct device *dev)
+{
+       struct platform_device *pdev = to_platform_device(dev);
+       struct usb_hcd *hcd = dev_get_drvdata(dev);
+
+       return 0;
+}
+#else
+#define oxu_drv_suspend        NULL
+#define oxu_drv_resume NULL
+#endif
+
+static struct platform_driver oxu_driver = {
+       .probe          = oxu_drv_probe,
+       .remove         = oxu_drv_remove,
+       .shutdown       = oxu_drv_shutdown,
+       .suspend        = oxu_drv_suspend,
+       .resume         = oxu_drv_resume,
+       .driver = {
+               .name = "oxu210hp-hcd",
+               .bus = &platform_bus_type
+       }
+};
+
+static int __init oxu_module_init(void)
+{
+       int retval = 0;
+
+       retval = platform_driver_register(&oxu_driver);
+       if (retval < 0)
+               return retval;
+
+       return retval;
+}
+
+static void __exit oxu_module_cleanup(void)
+{
+       platform_driver_unregister(&oxu_driver);
+}
+
+module_init(oxu_module_init);
+module_exit(oxu_module_cleanup);
+
+MODULE_DESCRIPTION("Oxford OXU210HP HCD driver - ver. " DRIVER_VERSION);
+MODULE_AUTHOR("Rodolfo Giometti <giometti@linux.it>");
+MODULE_LICENSE("GPL");
diff --git a/drivers/usb/host/oxu210hp.h b/drivers/usb/host/oxu210hp.h
new file mode 100644 (file)
index 0000000..8910e27
--- /dev/null
@@ -0,0 +1,447 @@
+/*
+ * Host interface registers
+ */
+
+#define OXU_DEVICEID                   0x00
+       #define OXU_REV_MASK            0xffff0000
+       #define OXU_REV_SHIFT           16
+       #define OXU_REV_2100            0x2100
+       #define OXU_BO_SHIFT            8
+       #define OXU_BO_MASK             (0x3 << OXU_BO_SHIFT)
+       #define OXU_MAJ_REV_SHIFT       4
+       #define OXU_MAJ_REV_MASK        (0xf << OXU_MAJ_REV_SHIFT)
+       #define OXU_MIN_REV_SHIFT       0
+       #define OXU_MIN_REV_MASK        (0xf << OXU_MIN_REV_SHIFT)
+#define OXU_HOSTIFCONFIG               0x04
+#define OXU_SOFTRESET                  0x08
+       #define OXU_SRESET              (1 << 0)
+
+#define OXU_PIOBURSTREADCTRL           0x0C
+
+#define OXU_CHIPIRQSTATUS              0x10
+#define OXU_CHIPIRQEN_SET              0x14
+#define OXU_CHIPIRQEN_CLR              0x18
+       #define OXU_USBSPHLPWUI         0x00000080
+       #define OXU_USBOTGLPWUI         0x00000040
+       #define OXU_USBSPHI             0x00000002
+       #define OXU_USBOTGI             0x00000001
+
+#define OXU_CLKCTRL_SET                        0x1C
+       #define OXU_SYSCLKEN            0x00000008
+       #define OXU_USBSPHCLKEN         0x00000002
+       #define OXU_USBOTGCLKEN         0x00000001
+
+#define OXU_ASO                                0x68
+       #define OXU_SPHPOEN             0x00000100
+       #define OXU_OVRCCURPUPDEN       0x00000800
+       #define OXU_ASO_OP              (1 << 10)
+       #define OXU_COMPARATOR          0x000004000
+
+#define OXU_USBMODE                    0x1A8
+       #define OXU_VBPS                0x00000020
+       #define OXU_ES_LITTLE           0x00000000
+       #define OXU_CM_HOST_ONLY        0x00000003
+
+/*
+ * Proper EHCI structs & defines
+ */
+
+/* Magic numbers that can affect system performance */
+#define EHCI_TUNE_CERR         3       /* 0-3 qtd retries; 0 == don't stop */
+#define EHCI_TUNE_RL_HS                4       /* nak throttle; see 4.9 */
+#define EHCI_TUNE_RL_TT                0
+#define EHCI_TUNE_MULT_HS      1       /* 1-3 transactions/uframe; 4.10.3 */
+#define EHCI_TUNE_MULT_TT      1
+#define EHCI_TUNE_FLS          2       /* (small) 256 frame schedule */
+
+struct oxu_hcd;
+
+/* EHCI register interface, corresponds to EHCI Revision 0.95 specification */
+
+/* Section 2.2 Host Controller Capability Registers */
+struct ehci_caps {
+       /* these fields are specified as 8 and 16 bit registers,
+        * but some hosts can't perform 8 or 16 bit PCI accesses.
+        */
+       u32             hc_capbase;
+#define HC_LENGTH(p)           (((p)>>00)&0x00ff)      /* bits 7:0 */
+#define HC_VERSION(p)          (((p)>>16)&0xffff)      /* bits 31:16 */
+       u32             hcs_params;     /* HCSPARAMS - offset 0x4 */
+#define HCS_DEBUG_PORT(p)      (((p)>>20)&0xf) /* bits 23:20, debug port? */
+#define HCS_INDICATOR(p)       ((p)&(1 << 16)) /* true: has port indicators */
+#define HCS_N_CC(p)            (((p)>>12)&0xf) /* bits 15:12, #companion HCs */
+#define HCS_N_PCC(p)           (((p)>>8)&0xf)  /* bits 11:8, ports per CC */
+#define HCS_PORTROUTED(p)      ((p)&(1 << 7))  /* true: port routing */
+#define HCS_PPC(p)             ((p)&(1 << 4))  /* true: port power control */
+#define HCS_N_PORTS(p)         (((p)>>0)&0xf)  /* bits 3:0, ports on HC */
+
+       u32             hcc_params;      /* HCCPARAMS - offset 0x8 */
+#define HCC_EXT_CAPS(p)                (((p)>>8)&0xff) /* for pci extended caps */
+#define HCC_ISOC_CACHE(p)       ((p)&(1 << 7))  /* true: can cache isoc frame */
+#define HCC_ISOC_THRES(p)       (((p)>>4)&0x7)  /* bits 6:4, uframes cached */
+#define HCC_CANPARK(p)         ((p)&(1 << 2))  /* true: can park on async qh */
+#define HCC_PGM_FRAMELISTLEN(p) ((p)&(1 << 1))  /* true: periodic_size changes*/
+#define HCC_64BIT_ADDR(p)       ((p)&(1))       /* true: can use 64-bit addr */
+       u8              portroute[8];    /* nibbles for routing - offset 0xC */
+} __attribute__ ((packed));
+
+
+/* Section 2.3 Host Controller Operational Registers */
+struct ehci_regs {
+       /* USBCMD: offset 0x00 */
+       u32             command;
+/* 23:16 is r/w intr rate, in microframes; default "8" == 1/msec */
+#define CMD_PARK       (1<<11)         /* enable "park" on async qh */
+#define CMD_PARK_CNT(c)        (((c)>>8)&3)    /* how many transfers to park for */
+#define CMD_LRESET     (1<<7)          /* partial reset (no ports, etc) */
+#define CMD_IAAD       (1<<6)          /* "doorbell" interrupt async advance */
+#define CMD_ASE                (1<<5)          /* async schedule enable */
+#define CMD_PSE                (1<<4)          /* periodic schedule enable */
+/* 3:2 is periodic frame list size */
+#define CMD_RESET      (1<<1)          /* reset HC not bus */
+#define CMD_RUN                (1<<0)          /* start/stop HC */
+
+       /* USBSTS: offset 0x04 */
+       u32             status;
+#define STS_ASS                (1<<15)         /* Async Schedule Status */
+#define STS_PSS                (1<<14)         /* Periodic Schedule Status */
+#define STS_RECL       (1<<13)         /* Reclamation */
+#define STS_HALT       (1<<12)         /* Not running (any reason) */
+/* some bits reserved */
+       /* these STS_* flags are also intr_enable bits (USBINTR) */
+#define STS_IAA                (1<<5)          /* Interrupted on async advance */
+#define STS_FATAL      (1<<4)          /* such as some PCI access errors */
+#define STS_FLR                (1<<3)          /* frame list rolled over */
+#define STS_PCD                (1<<2)          /* port change detect */
+#define STS_ERR                (1<<1)          /* "error" completion (overflow, ...) */
+#define STS_INT                (1<<0)          /* "normal" completion (short, ...) */
+
+#define INTR_MASK (STS_IAA | STS_FATAL | STS_PCD | STS_ERR | STS_INT)
+
+       /* USBINTR: offset 0x08 */
+       u32             intr_enable;
+
+       /* FRINDEX: offset 0x0C */
+       u32             frame_index;    /* current microframe number */
+       /* CTRLDSSEGMENT: offset 0x10 */
+       u32             segment;        /* address bits 63:32 if needed */
+       /* PERIODICLISTBASE: offset 0x14 */
+       u32             frame_list;     /* points to periodic list */
+       /* ASYNCLISTADDR: offset 0x18 */
+       u32             async_next;     /* address of next async queue head */
+
+       u32             reserved[9];
+
+       /* CONFIGFLAG: offset 0x40 */
+       u32             configured_flag;
+#define FLAG_CF                (1<<0)          /* true: we'll support "high speed" */
+
+       /* PORTSC: offset 0x44 */
+       u32             port_status[0]; /* up to N_PORTS */
+/* 31:23 reserved */
+#define PORT_WKOC_E    (1<<22)         /* wake on overcurrent (enable) */
+#define PORT_WKDISC_E  (1<<21)         /* wake on disconnect (enable) */
+#define PORT_WKCONN_E  (1<<20)         /* wake on connect (enable) */
+/* 19:16 for port testing */
+#define PORT_LED_OFF   (0<<14)
+#define PORT_LED_AMBER (1<<14)
+#define PORT_LED_GREEN (2<<14)
+#define PORT_LED_MASK  (3<<14)
+#define PORT_OWNER     (1<<13)         /* true: companion hc owns this port */
+#define PORT_POWER     (1<<12)         /* true: has power (see PPC) */
+#define PORT_USB11(x) (((x)&(3<<10)) == (1<<10))       /* USB 1.1 device */
+/* 11:10 for detecting lowspeed devices (reset vs release ownership) */
+/* 9 reserved */
+#define PORT_RESET     (1<<8)          /* reset port */
+#define PORT_SUSPEND   (1<<7)          /* suspend port */
+#define PORT_RESUME    (1<<6)          /* resume it */
+#define PORT_OCC       (1<<5)          /* over current change */
+#define PORT_OC                (1<<4)          /* over current active */
+#define PORT_PEC       (1<<3)          /* port enable change */
+#define PORT_PE                (1<<2)          /* port enable */
+#define PORT_CSC       (1<<1)          /* connect status change */
+#define PORT_CONNECT   (1<<0)          /* device connected */
+#define PORT_RWC_BITS   (PORT_CSC | PORT_PEC | PORT_OCC)
+} __attribute__ ((packed));
+
+/* Appendix C, Debug port ... intended for use with special "debug devices"
+ * that can help if there's no serial console.  (nonstandard enumeration.)
+ */
+struct ehci_dbg_port {
+       u32     control;
+#define DBGP_OWNER     (1<<30)
+#define DBGP_ENABLED   (1<<28)
+#define DBGP_DONE      (1<<16)
+#define DBGP_INUSE     (1<<10)
+#define DBGP_ERRCODE(x)        (((x)>>7)&0x07)
+#      define DBGP_ERR_BAD     1
+#      define DBGP_ERR_SIGNAL  2
+#define DBGP_ERROR     (1<<6)
+#define DBGP_GO                (1<<5)
+#define DBGP_OUT       (1<<4)
+#define DBGP_LEN(x)    (((x)>>0)&0x0f)
+       u32     pids;
+#define DBGP_PID_GET(x)                (((x)>>16)&0xff)
+#define DBGP_PID_SET(data, tok)        (((data)<<8)|(tok))
+       u32     data03;
+       u32     data47;
+       u32     address;
+#define DBGP_EPADDR(dev, ep)   (((dev)<<8)|(ep))
+} __attribute__ ((packed));
+
+
+#define        QTD_NEXT(dma)   cpu_to_le32((u32)dma)
+
+/*
+ * EHCI Specification 0.95 Section 3.5
+ * QTD: describe data transfer components (buffer, direction, ...)
+ * See Fig 3-6 "Queue Element Transfer Descriptor Block Diagram".
+ *
+ * These are associated only with "QH" (Queue Head) structures,
+ * used with control, bulk, and interrupt transfers.
+ */
+struct ehci_qtd {
+       /* first part defined by EHCI spec */
+       __le32                  hw_next;                /* see EHCI 3.5.1 */
+       __le32                  hw_alt_next;            /* see EHCI 3.5.2 */
+       __le32                  hw_token;               /* see EHCI 3.5.3 */
+#define        QTD_TOGGLE      (1 << 31)       /* data toggle */
+#define        QTD_LENGTH(tok) (((tok)>>16) & 0x7fff)
+#define        QTD_IOC         (1 << 15)       /* interrupt on complete */
+#define        QTD_CERR(tok)   (((tok)>>10) & 0x3)
+#define        QTD_PID(tok)    (((tok)>>8) & 0x3)
+#define        QTD_STS_ACTIVE  (1 << 7)        /* HC may execute this */
+#define        QTD_STS_HALT    (1 << 6)        /* halted on error */
+#define        QTD_STS_DBE     (1 << 5)        /* data buffer error (in HC) */
+#define        QTD_STS_BABBLE  (1 << 4)        /* device was babbling (qtd halted) */
+#define        QTD_STS_XACT    (1 << 3)        /* device gave illegal response */
+#define        QTD_STS_MMF     (1 << 2)        /* incomplete split transaction */
+#define        QTD_STS_STS     (1 << 1)        /* split transaction state */
+#define        QTD_STS_PING    (1 << 0)        /* issue PING? */
+       __le32                  hw_buf[5];              /* see EHCI 3.5.4 */
+       __le32                  hw_buf_hi[5];           /* Appendix B */
+
+       /* the rest is HCD-private */
+       dma_addr_t              qtd_dma;                /* qtd address */
+       struct list_head        qtd_list;               /* sw qtd list */
+       struct urb              *urb;                   /* qtd's urb */
+       size_t                  length;                 /* length of buffer */
+
+       u32                     qtd_buffer_len;
+       void                    *buffer;
+       dma_addr_t              buffer_dma;
+       void                    *transfer_buffer;
+       void                    *transfer_dma;
+} __attribute__ ((aligned(32)));
+
+/* mask NakCnt+T in qh->hw_alt_next */
+#define QTD_MASK __constant_cpu_to_le32 (~0x1f)
+
+#define IS_SHORT_READ(token) (QTD_LENGTH(token) != 0 && QTD_PID(token) == 1)
+
+/* Type tag from {qh, itd, sitd, fstn}->hw_next */
+#define Q_NEXT_TYPE(dma) ((dma) & __constant_cpu_to_le32 (3 << 1))
+
+/* values for that type tag */
+#define Q_TYPE_QH      __constant_cpu_to_le32 (1 << 1)
+
+/* next async queue entry, or pointer to interrupt/periodic QH */
+#define        QH_NEXT(dma)    (cpu_to_le32(((u32)dma)&~0x01f)|Q_TYPE_QH)
+
+/* for periodic/async schedules and qtd lists, mark end of list */
+#define        EHCI_LIST_END   __constant_cpu_to_le32(1) /* "null pointer" to hw */
+
+/*
+ * Entries in periodic shadow table are pointers to one of four kinds
+ * of data structure.  That's dictated by the hardware; a type tag is
+ * encoded in the low bits of the hardware's periodic schedule.  Use
+ * Q_NEXT_TYPE to get the tag.
+ *
+ * For entries in the async schedule, the type tag always says "qh".
+ */
+union ehci_shadow {
+       struct ehci_qh          *qh;            /* Q_TYPE_QH */
+       __le32                  *hw_next;       /* (all types) */
+       void                    *ptr;
+};
+
+/*
+ * EHCI Specification 0.95 Section 3.6
+ * QH: describes control/bulk/interrupt endpoints
+ * See Fig 3-7 "Queue Head Structure Layout".
+ *
+ * These appear in both the async and (for interrupt) periodic schedules.
+ */
+
+struct ehci_qh {
+       /* first part defined by EHCI spec */
+       __le32                  hw_next;         /* see EHCI 3.6.1 */
+       __le32                  hw_info1;       /* see EHCI 3.6.2 */
+#define        QH_HEAD         0x00008000
+       __le32                  hw_info2;       /* see EHCI 3.6.2 */
+#define        QH_SMASK        0x000000ff
+#define        QH_CMASK        0x0000ff00
+#define        QH_HUBADDR      0x007f0000
+#define        QH_HUBPORT      0x3f800000
+#define        QH_MULT         0xc0000000
+       __le32                  hw_current;      /* qtd list - see EHCI 3.6.4 */
+
+       /* qtd overlay (hardware parts of a struct ehci_qtd) */
+       __le32                  hw_qtd_next;
+       __le32                  hw_alt_next;
+       __le32                  hw_token;
+       __le32                  hw_buf[5];
+       __le32                  hw_buf_hi[5];
+
+       /* the rest is HCD-private */
+       dma_addr_t              qh_dma;         /* address of qh */
+       union ehci_shadow       qh_next;        /* ptr to qh; or periodic */
+       struct list_head        qtd_list;       /* sw qtd list */
+       struct ehci_qtd         *dummy;
+       struct ehci_qh          *reclaim;       /* next to reclaim */
+
+       struct oxu_hcd          *oxu;
+       struct kref             kref;
+       unsigned                stamp;
+
+       u8                      qh_state;
+#define        QH_STATE_LINKED         1               /* HC sees this */
+#define        QH_STATE_UNLINK         2               /* HC may still see this */
+#define        QH_STATE_IDLE           3               /* HC doesn't see this */
+#define        QH_STATE_UNLINK_WAIT    4               /* LINKED and on reclaim q */
+#define        QH_STATE_COMPLETING     5               /* don't touch token.HALT */
+
+       /* periodic schedule info */
+       u8                      usecs;          /* intr bandwidth */
+       u8                      gap_uf;         /* uframes split/csplit gap */
+       u8                      c_usecs;        /* ... split completion bw */
+       u16                     tt_usecs;       /* tt downstream bandwidth */
+       unsigned short          period;         /* polling interval */
+       unsigned short          start;          /* where polling starts */
+#define NO_FRAME ((unsigned short)~0)                  /* pick new start */
+       struct usb_device       *dev;           /* access to TT */
+} __attribute__ ((aligned(32)));
+
+/*
+ * Proper OXU210HP structs
+ */
+
+#define OXU_OTG_CORE_OFFSET    0x00400
+#define OXU_OTG_CAP_OFFSET     (OXU_OTG_CORE_OFFSET + 0x100)
+#define OXU_SPH_CORE_OFFSET    0x00800
+#define OXU_SPH_CAP_OFFSET     (OXU_SPH_CORE_OFFSET + 0x100)
+
+#define OXU_OTG_MEM            0xE000
+#define OXU_SPH_MEM            0x16000
+
+/* Only how many elements & element structure are specifies here. */
+/* 2 host controllers are enabled - total size <= 28 kbytes */
+#define        DEFAULT_I_TDPS          1024
+#define QHEAD_NUM              16
+#define QTD_NUM                        32
+#define SITD_NUM               8
+#define MURB_NUM               8
+
+#define BUFFER_NUM             8
+#define BUFFER_SIZE            512
+
+struct oxu_info {
+       struct usb_hcd *hcd[2];
+};
+
+struct oxu_buf {
+       u8                      buffer[BUFFER_SIZE];
+} __attribute__ ((aligned(BUFFER_SIZE)));
+
+struct oxu_onchip_mem {
+       struct oxu_buf          db_pool[BUFFER_NUM];
+
+       u32                     frame_list[DEFAULT_I_TDPS];
+       struct ehci_qh          qh_pool[QHEAD_NUM];
+       struct ehci_qtd         qtd_pool[QTD_NUM];
+} __attribute__ ((aligned(4 << 10)));
+
+#define        EHCI_MAX_ROOT_PORTS     15              /* see HCS_N_PORTS */
+
+struct oxu_murb {
+       struct urb              urb;
+       struct urb              *main;
+       u8                      last;
+};
+
+struct oxu_hcd {                               /* one per controller */
+       unsigned int            is_otg:1;
+
+       u8                      qh_used[QHEAD_NUM];
+       u8                      qtd_used[QTD_NUM];
+       u8                      db_used[BUFFER_NUM];
+       u8                      murb_used[MURB_NUM];
+
+       struct oxu_onchip_mem   __iomem *mem;
+       spinlock_t              mem_lock;
+
+       struct timer_list       urb_timer;
+
+       struct ehci_caps __iomem *caps;
+       struct ehci_regs __iomem *regs;
+
+       __u32                   hcs_params;     /* cached register copy */
+       spinlock_t              lock;
+
+       /* async schedule support */
+       struct ehci_qh          *async;
+       struct ehci_qh          *reclaim;
+       unsigned                reclaim_ready:1;
+       unsigned                scanning:1;
+
+       /* periodic schedule support */
+       unsigned                periodic_size;
+       __le32                  *periodic;      /* hw periodic table */
+       dma_addr_t              periodic_dma;
+       unsigned                i_thresh;       /* uframes HC might cache */
+
+       union ehci_shadow       *pshadow;       /* mirror hw periodic table */
+       int                     next_uframe;    /* scan periodic, start here */
+       unsigned                periodic_sched; /* periodic activity count */
+
+       /* per root hub port */
+       unsigned long           reset_done[EHCI_MAX_ROOT_PORTS];
+       /* bit vectors (one bit per port) */
+       unsigned long           bus_suspended;  /* which ports were
+                                                * already suspended at the
+                                                * start of a bus suspend
+                                                */
+       unsigned long           companion_ports;/* which ports are dedicated
+                                                * to the companion controller
+                                                */
+
+       struct timer_list       watchdog;
+       unsigned long           actions;
+       unsigned                stamp;
+       unsigned long           next_statechange;
+       u32                     command;
+
+       /* SILICON QUIRKS */
+       struct list_head        urb_list;       /* this is the head to urb
+                                                * queue that didn't get enough
+                                                * resources
+                                                */
+       struct oxu_murb         *murb_pool;     /* murb per split big urb */
+       unsigned urb_len;
+
+       u8                      sbrn;           /* packed release number */
+};
+
+#define EHCI_IAA_JIFFIES       (HZ/100)        /* arbitrary; ~10 msec */
+#define EHCI_IO_JIFFIES                (HZ/10)         /* io watchdog > irq_thresh */
+#define EHCI_ASYNC_JIFFIES      (HZ/20)                /* async idle timeout */
+#define EHCI_SHRINK_JIFFIES     (HZ/200)       /* async qh unlink delay */
+
+enum ehci_timer_action {
+       TIMER_IO_WATCHDOG,
+       TIMER_IAA_WATCHDOG,
+       TIMER_ASYNC_SHRINK,
+       TIMER_ASYNC_OFF,
+};
+
+#include <linux/oxu210hp.h>
diff --git a/include/linux/oxu210hp.h b/include/linux/oxu210hp.h
new file mode 100644 (file)
index 0000000..0bf96ea
--- /dev/null
@@ -0,0 +1,7 @@
+/* platform data for the OXU210HP HCD */
+
+struct oxu210hp_platform_data {
+       unsigned int bus16:1;
+       unsigned int use_hcd_otg:1;
+       unsigned int use_hcd_sph:1;
+};