/*
 * xHCI host controller driver
 *
 * Copyright (C) 2008 Intel Corp.
 *
 * Author: Sarah Sharp
 * Some code borrowed from the Linux EHCI driver.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 *
 * 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/pci.h>
#include <linux/irq.h>
#include <linux/log2.h>
#include <linux/module.h>
#include <linux/moduleparam.h>
#include <linux/slab.h>
#include <linux/dmi.h>
#include <linux/dma-mapping.h>

#include "xhci.h"
#include "xhci-trace.h"

#define DRIVER_AUTHOR "Sarah Sharp"
#define DRIVER_DESC "'eXtensible' Host Controller (xHC) Driver"

#define PORT_WAKE_BITS  (PORT_WKOC_E | PORT_WKDISC_E | PORT_WKCONN_E)

/* Some 0.95 hardware can't handle the chain bit on a Link TRB being cleared */
static int link_quirk;
module_param(link_quirk, int, S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(link_quirk, "Don't clear the chain bit on a link TRB");

static unsigned int quirks;
module_param(quirks, uint, S_IRUGO);
MODULE_PARM_DESC(quirks, "Bit flags for quirks to be enabled as default");

/* TODO: copied from ehci-hcd.c - can this be refactored? */
/*
 * xhci_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).
 */
int xhci_handshake(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;
}

/*
 * Disable interrupts and begin the xHCI halting process.
 */
void xhci_quiesce(struct xhci_hcd *xhci)
{
        u32 halted;
        u32 cmd;
        u32 mask;

        mask = ~(XHCI_IRQS);
        halted = readl(&xhci->op_regs->status) & STS_HALT;
        if (!halted)
                mask &= ~CMD_RUN;

        cmd = readl(&xhci->op_regs->command);
        cmd &= mask;
        writel(cmd, &xhci->op_regs->command);
}

/*
 * Force HC into halt state.
 *
 * Disable any IRQs and clear the run/stop bit.
 * HC will complete any current and actively pipelined transactions, and
 * should halt within 16 ms of the run/stop bit being cleared.
 * Read HC Halted bit in the status register to see when the HC is finished.
 */
int xhci_halt(struct xhci_hcd *xhci)
{
        int ret;
        xhci_dbg_trace(xhci, trace_xhci_dbg_init, "// Halt the HC");
        xhci_quiesce(xhci);

        ret = xhci_handshake(&xhci->op_regs->status,
                        STS_HALT, STS_HALT, XHCI_MAX_HALT_USEC);
        if (!ret) {
                xhci->xhc_state |= XHCI_STATE_HALTED;
                xhci->cmd_ring_state = CMD_RING_STATE_STOPPED;
        } else
                xhci_warn(xhci, "Host not halted after %u microseconds.\n",
                                XHCI_MAX_HALT_USEC);
        return ret;
}

/*
 * Set the run bit and wait for the host to be running.
 */
static int xhci_start(struct xhci_hcd *xhci)
{
        u32 temp;
        int ret;

        temp = readl(&xhci->op_regs->command);
        temp |= (CMD_RUN);
        xhci_dbg_trace(xhci, trace_xhci_dbg_init, "// Turn on HC, cmd = 0x%x.",
                        temp);
        writel(temp, &xhci->op_regs->command);

        /*
         * Wait for the HCHalted Status bit to be 0 to indicate the host is
         * running.
         */
        ret = xhci_handshake(&xhci->op_regs->status,
                        STS_HALT, 0, XHCI_MAX_HALT_USEC);
        if (ret == -ETIMEDOUT)
                xhci_err(xhci, "Host took too long to start, "
                                "waited %u microseconds.\n",
                                XHCI_MAX_HALT_USEC);
        if (!ret)
                xhci->xhc_state &= ~(XHCI_STATE_HALTED | XHCI_STATE_DYING);

        return ret;
}

/*
 * Reset a halted HC.
 *
 * This resets pipelines, timers, counters, state machines, etc.
 * Transactions will be terminated immediately, and operational registers
 * will be set to their defaults.
 */
int xhci_reset(struct xhci_hcd *xhci)
{
        u32 command;
        u32 state;
        int ret, i;

        state = readl(&xhci->op_regs->status);
        if ((state & STS_HALT) == 0) {
                xhci_warn(xhci, "Host controller not halted, aborting reset.\n");
                return 0;
        }

        xhci_dbg_trace(xhci, trace_xhci_dbg_init, "// Reset the HC");
        command = readl(&xhci->op_regs->command);
        command |= CMD_RESET;
        writel(command, &xhci->op_regs->command);

        ret = xhci_handshake(&xhci->op_regs->command,
                        CMD_RESET, 0, 10 * 1000 * 1000);
        if (ret)
                return ret;

        xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                         "Wait for controller to be ready for doorbell rings");
        /*
         * xHCI cannot write to any doorbells or operational registers other
         * than status until the "Controller Not Ready" flag is cleared.
         */
        ret = xhci_handshake(&xhci->op_regs->status,
                        STS_CNR, 0, 10 * 1000 * 1000);

        for (i = 0; i < 2; ++i) {
                xhci->bus_state[i].port_c_suspend = 0;
                xhci->bus_state[i].suspended_ports = 0;
                xhci->bus_state[i].resuming_ports = 0;
        }

        return ret;
}

#ifdef CONFIG_PCI
static int xhci_free_msi(struct xhci_hcd *xhci)
{
        int i;

        if (!xhci->msix_entries)
                return -EINVAL;

        for (i = 0; i < xhci->msix_count; i++)
                if (xhci->msix_entries[i].vector)
                        free_irq(xhci->msix_entries[i].vector,
                                        xhci_to_hcd(xhci));
        return 0;
}

/*
 * Set up MSI
 */
static int xhci_setup_msi(struct xhci_hcd *xhci)
{
        int ret;
        struct pci_dev  *pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);

        ret = pci_enable_msi(pdev);
        if (ret) {
                xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                                "failed to allocate MSI entry");
                return ret;
        }

        ret = request_irq(pdev->irq, xhci_msi_irq,
                                0, "xhci_hcd", xhci_to_hcd(xhci));
        if (ret) {
                xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                                "disable MSI interrupt");
                pci_disable_msi(pdev);
        }

        return ret;
}

/*
 * Free IRQs
 * free all IRQs request
 */
static void xhci_free_irq(struct xhci_hcd *xhci)
{
        struct pci_dev *pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
        int ret;

        /* return if using legacy interrupt */
        if (xhci_to_hcd(xhci)->irq > 0)
                return;

        ret = xhci_free_msi(xhci);
        if (!ret)
                return;
        if (pdev->irq > 0)
                free_irq(pdev->irq, xhci_to_hcd(xhci));

        return;
}

/*
 * Set up MSI-X
 */
static int xhci_setup_msix(struct xhci_hcd *xhci)
{
        int i, ret = 0;
        struct usb_hcd *hcd = xhci_to_hcd(xhci);
        struct pci_dev *pdev = to_pci_dev(hcd->self.controller);

        /*
         * calculate number of msi-x vectors supported.
         * - HCS_MAX_INTRS: the max number of interrupts the host can handle,
         *   with max number of interrupters based on the xhci HCSPARAMS1.
         * - num_online_cpus: maximum msi-x vectors per CPUs core.
         *   Add additional 1 vector to ensure always available interrupt.
         */
        xhci->msix_count = min(num_online_cpus() + 1,
                                HCS_MAX_INTRS(xhci->hcs_params1));

        xhci->msix_entries =
                kmalloc((sizeof(struct msix_entry))*xhci->msix_count,
                                GFP_KERNEL);
        if (!xhci->msix_entries) {
                xhci_err(xhci, "Failed to allocate MSI-X entries\n");
                return -ENOMEM;
        }

        for (i = 0; i < xhci->msix_count; i++) {
                xhci->msix_entries[i].entry = i;
                xhci->msix_entries[i].vector = 0;
        }

        ret = pci_enable_msix_exact(pdev, xhci->msix_entries, xhci->msix_count);
        if (ret) {
                xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                                "Failed to enable MSI-X");
                goto free_entries;
        }

        for (i = 0; i < xhci->msix_count; i++) {
                ret = request_irq(xhci->msix_entries[i].vector,
                                xhci_msi_irq,
                                0, "xhci_hcd", xhci_to_hcd(xhci));
                if (ret)
                        goto disable_msix;
        }

        hcd->msix_enabled = 1;
        return ret;

disable_msix:
        xhci_dbg_trace(xhci, trace_xhci_dbg_init, "disable MSI-X interrupt");
        xhci_free_irq(xhci);
        pci_disable_msix(pdev);
free_entries:
        kfree(xhci->msix_entries);
        xhci->msix_entries = NULL;
        return ret;
}

/* Free any IRQs and disable MSI-X */
static void xhci_cleanup_msix(struct xhci_hcd *xhci)
{
        struct usb_hcd *hcd = xhci_to_hcd(xhci);
        struct pci_dev *pdev = to_pci_dev(hcd->self.controller);

        if (xhci->quirks & XHCI_PLAT)
                return;

        xhci_free_irq(xhci);

        if (xhci->msix_entries) {
                pci_disable_msix(pdev);
                kfree(xhci->msix_entries);
                xhci->msix_entries = NULL;
        } else {
                pci_disable_msi(pdev);
        }

        hcd->msix_enabled = 0;
        return;
}

static void __maybe_unused xhci_msix_sync_irqs(struct xhci_hcd *xhci)
{
        int i;

        if (xhci->msix_entries) {
                for (i = 0; i < xhci->msix_count; i++)
                        synchronize_irq(xhci->msix_entries[i].vector);
        }
}

static int xhci_try_enable_msi(struct usb_hcd *hcd)
{
        struct xhci_hcd *xhci = hcd_to_xhci(hcd);
        struct pci_dev  *pdev;
        int ret;

        /* The xhci platform device has set up IRQs through usb_add_hcd. */
        if (xhci->quirks & XHCI_PLAT)
                return 0;

        pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
        /*
         * Some Fresco Logic host controllers advertise MSI, but fail to
         * generate interrupts.  Don't even try to enable MSI.
         */
        if (xhci->quirks & XHCI_BROKEN_MSI)
                goto legacy_irq;

        /* unregister the legacy interrupt */
        if (hcd->irq)
                free_irq(hcd->irq, hcd);
        hcd->irq = 0;

        ret = xhci_setup_msix(xhci);
        if (ret)
                /* fall back to msi*/
                ret = xhci_setup_msi(xhci);

        if (!ret)
                /* hcd->irq is 0, we have MSI */
                return 0;

        if (!pdev->irq) {
                xhci_err(xhci, "No msi-x/msi found and no IRQ in BIOS\n");
                return -EINVAL;
        }

 legacy_irq:
        if (!strlen(hcd->irq_descr))
                snprintf(hcd->irq_descr, sizeof(hcd->irq_descr), "%s:usb%d",
                         hcd->driver->description, hcd->self.busnum);

        /* fall back to legacy interrupt*/
        ret = request_irq(pdev->irq, &usb_hcd_irq, IRQF_SHARED,
                        hcd->irq_descr, hcd);
        if (ret) {
                xhci_err(xhci, "request interrupt %d failed\n",
                                pdev->irq);
                return ret;
        }
        hcd->irq = pdev->irq;
        return 0;
}

#else

static inline int xhci_try_enable_msi(struct usb_hcd *hcd)
{
        return 0;
}

static inline void xhci_cleanup_msix(struct xhci_hcd *xhci)
{
}

static inline void xhci_msix_sync_irqs(struct xhci_hcd *xhci)
{
}

#endif

static void compliance_mode_recovery(unsigned long arg)
{
        struct xhci_hcd *xhci;
        struct usb_hcd *hcd;
        u32 temp;
        int i;

        xhci = (struct xhci_hcd *)arg;

        for (i = 0; i < xhci->num_usb3_ports; i++) {
                temp = readl(xhci->usb3_ports[i]);
                if ((temp & PORT_PLS_MASK) == USB_SS_PORT_LS_COMP_MOD) {
                        /*
                         * Compliance Mode Detected. Letting USB Core
                         * handle the Warm Reset
                         */
                        xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                                        "Compliance mode detected->port %d",
                                        i + 1);
                        xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                                        "Attempting compliance mode recovery");
                        hcd = xhci->shared_hcd;

                        if (hcd->state == HC_STATE_SUSPENDED)
                                usb_hcd_resume_root_hub(hcd);

                        usb_hcd_poll_rh_status(hcd);
                }
        }

        if (xhci->port_status_u0 != ((1 << xhci->num_usb3_ports)-1))
                mod_timer(&xhci->comp_mode_recovery_timer,
                        jiffies + msecs_to_jiffies(COMP_MODE_RCVRY_MSECS));
}

/*
 * Quirk to work around issue generated by the SN65LVPE502CP USB3.0 re-driver
 * that causes ports behind that hardware to enter compliance mode sometimes.
 * The quirk creates a timer that polls every 2 seconds the link state of
 * each host controller's port and recovers it by issuing a Warm reset
 * if Compliance mode is detected, otherwise the port will become "dead" (no
 * device connections or disconnections will be detected anymore). Becasue no
 * status event is generated when entering compliance mode (per xhci spec),
 * this quirk is needed on systems that have the failing hardware installed.
 */
static void compliance_mode_recovery_timer_init(struct xhci_hcd *xhci)
{
        xhci->port_status_u0 = 0;
        setup_timer(&xhci->comp_mode_recovery_timer,
                    compliance_mode_recovery, (unsigned long)xhci);
        xhci->comp_mode_recovery_timer.expires = jiffies +
                        msecs_to_jiffies(COMP_MODE_RCVRY_MSECS);

        set_timer_slack(&xhci->comp_mode_recovery_timer,
                        msecs_to_jiffies(COMP_MODE_RCVRY_MSECS));
        add_timer(&xhci->comp_mode_recovery_timer);
        xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                        "Compliance mode recovery timer initialized");
}

/*
 * This function identifies the systems that have installed the SN65LVPE502CP
 * USB3.0 re-driver and that need the Compliance Mode Quirk.
 * Systems:
 * Vendor: Hewlett-Packard -> System Models: Z420, Z620 and Z820
 */
static bool xhci_compliance_mode_recovery_timer_quirk_check(void)
{
        const char *dmi_product_name, *dmi_sys_vendor;

        dmi_product_name = dmi_get_system_info(DMI_PRODUCT_NAME);
        dmi_sys_vendor = dmi_get_system_info(DMI_SYS_VENDOR);
        if (!dmi_product_name || !dmi_sys_vendor)
                return false;

        if (!(strstr(dmi_sys_vendor, "Hewlett-Packard")))
                return false;

        if (strstr(dmi_product_name, "Z420") ||
                        strstr(dmi_product_name, "Z620") ||
                        strstr(dmi_product_name, "Z820") ||
                        strstr(dmi_product_name, "Z1 Workstation"))
                return true;

        return false;
}

static int xhci_all_ports_seen_u0(struct xhci_hcd *xhci)
{
        return (xhci->port_status_u0 == ((1 << xhci->num_usb3_ports)-1));
}


/*
 * Initialize memory for HCD and xHC (one-time init).
 *
 * Program the PAGESIZE register, initialize the device context array, create
 * device contexts (?), set up a command ring segment (or two?), create event
 * ring (one for now).
 */
int xhci_init(struct usb_hcd *hcd)
{
        struct xhci_hcd *xhci = hcd_to_xhci(hcd);
        int retval = 0;

        xhci_dbg_trace(xhci, trace_xhci_dbg_init, "xhci_init");
        spin_lock_init(&xhci->lock);
        if (xhci->hci_version == 0x95 && link_quirk) {
                xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                                "QUIRK: Not clearing Link TRB chain bits.");
                xhci->quirks |= XHCI_LINK_TRB_QUIRK;
        } else {
                xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                                "xHCI doesn't need link TRB QUIRK");
        }
        retval = xhci_mem_init(xhci, GFP_KERNEL);
        xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Finished xhci_init");

        /* Initializing Compliance Mode Recovery Data If Needed */
        if (xhci_compliance_mode_recovery_timer_quirk_check()) {
                xhci->quirks |= XHCI_COMP_MODE_QUIRK;
                compliance_mode_recovery_timer_init(xhci);
        }

        return retval;
}

/*-------------------------------------------------------------------------*/


static int xhci_run_finished(struct xhci_hcd *xhci)
{
        if (xhci_start(xhci)) {
                xhci_halt(xhci);
                return -ENODEV;
        }
        xhci->shared_hcd->state = HC_STATE_RUNNING;
        xhci->cmd_ring_state = CMD_RING_STATE_RUNNING;

        if (xhci->quirks & XHCI_NEC_HOST)
                xhci_ring_cmd_db(xhci);

        xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                        "Finished xhci_run for USB3 roothub");
        return 0;
}

/*
 * Start the HC after it was halted.
 *
 * This function is called by the USB core when the HC driver is added.
 * Its opposite is xhci_stop().
 *
 * xhci_init() must be called once before this function can be called.
 * Reset the HC, enable device slot contexts, program DCBAAP, and
 * set command ring pointer and event ring pointer.
 *
 * Setup MSI-X vectors and enable interrupts.
 */
int xhci_run(struct usb_hcd *hcd)
{
        u32 temp;
        u64 temp_64;
        int ret;
        struct xhci_hcd *xhci = hcd_to_xhci(hcd);

        /* Start the xHCI host controller running only after the USB 2.0 roothub
         * is setup.
         */

        hcd->uses_new_polling = 1;
        if (!usb_hcd_is_primary_hcd(hcd))
                return xhci_run_finished(xhci);

        xhci_dbg_trace(xhci, trace_xhci_dbg_init, "xhci_run");

        ret = xhci_try_enable_msi(hcd);
        if (ret)
                return ret;

        xhci_dbg(xhci, "Command ring memory map follows:\n");
        xhci_debug_ring(xhci, xhci->cmd_ring);
        xhci_dbg_ring_ptrs(xhci, xhci->cmd_ring);
        xhci_dbg_cmd_ptrs(xhci);

        xhci_dbg(xhci, "ERST memory map follows:\n");
        xhci_dbg_erst(xhci, &xhci->erst);
        xhci_dbg(xhci, "Event ring:\n");
        xhci_debug_ring(xhci, xhci->event_ring);
        xhci_dbg_ring_ptrs(xhci, xhci->event_ring);
        temp_64 = xhci_read_64(xhci, &xhci->ir_set->erst_dequeue);
        temp_64 &= ~ERST_PTR_MASK;
        xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                        "ERST deq = 64'h%0lx", (long unsigned int) temp_64);

        xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                        "// Set the interrupt modulation register");
        temp = readl(&xhci->ir_set->irq_control);
        temp &= ~ER_IRQ_INTERVAL_MASK;
        temp |= (u32) 160;
        writel(temp, &xhci->ir_set->irq_control);

        /* Set the HCD state before we enable the irqs */
        temp = readl(&xhci->op_regs->command);
        temp |= (CMD_EIE);
        xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                        "// Enable interrupts, cmd = 0x%x.", temp);
        writel(temp, &xhci->op_regs->command);

        temp = readl(&xhci->ir_set->irq_pending);
        xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                        "// Enabling event ring interrupter %p by writing 0x%x to irq_pending",
                        xhci->ir_set, (unsigned int) ER_IRQ_ENABLE(temp));
        writel(ER_IRQ_ENABLE(temp), &xhci->ir_set->irq_pending);
        xhci_print_ir_set(xhci, 0);

        if (xhci->quirks & XHCI_NEC_HOST) {
                struct xhci_command *command;
                command = xhci_alloc_command(xhci, false, false, GFP_KERNEL);
                if (!command)
                        return -ENOMEM;
                xhci_queue_vendor_command(xhci, command, 0, 0, 0,
                                TRB_TYPE(TRB_NEC_GET_FW));
        }
        xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                        "Finished xhci_run for USB2 roothub");
        return 0;
}
EXPORT_SYMBOL_GPL(xhci_run);

/*
 * Stop xHCI driver.
 *
 * This function is called by the USB core when the HC driver is removed.
 * Its opposite is xhci_run().
 *
 * Disable device contexts, disable IRQs, and quiesce the HC.
 * Reset the HC, finish any completed transactions, and cleanup memory.
 */
void xhci_stop(struct usb_hcd *hcd)
{
        u32 temp;
        struct xhci_hcd *xhci = hcd_to_xhci(hcd);

        if (xhci->xhc_state & XHCI_STATE_HALTED)
                return;

        mutex_lock(&xhci->mutex);
        spin_lock_irq(&xhci->lock);
        xhci->xhc_state |= XHCI_STATE_HALTED;
        xhci->cmd_ring_state = CMD_RING_STATE_STOPPED;

        /* Make sure the xHC is halted for a USB3 roothub
         * (xhci_stop() could be called as part of failed init).
         */
        xhci_halt(xhci);
        xhci_reset(xhci);
        spin_unlock_irq(&xhci->lock);

        xhci_cleanup_msix(xhci);

        /* Deleting Compliance Mode Recovery Timer */
        if ((xhci->quirks & XHCI_COMP_MODE_QUIRK) &&
                        (!(xhci_all_ports_seen_u0(xhci)))) {
                del_timer_sync(&xhci->comp_mode_recovery_timer);
                xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                                "%s: compliance mode recovery timer deleted",
                                __func__);
        }

        if (xhci->quirks & XHCI_AMD_PLL_FIX)
                usb_amd_dev_put();

        xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                        "// Disabling event ring interrupts");
        temp = readl(&xhci->op_regs->status);
        writel(temp & ~STS_EINT, &xhci->op_regs->status);
        temp = readl(&xhci->ir_set->irq_pending);
        writel(ER_IRQ_DISABLE(temp), &xhci->ir_set->irq_pending);
        xhci_print_ir_set(xhci, 0);

        xhci_dbg_trace(xhci, trace_xhci_dbg_init, "cleaning up memory");
        xhci_mem_cleanup(xhci);
        xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                        "xhci_stop completed - status = %x",
                        readl(&xhci->op_regs->status));
        mutex_unlock(&xhci->mutex);
}

/*
 * Shutdown HC (not bus-specific)
 *
 * This is called when the machine is rebooting or halting.  We assume that the
 * machine will be powered off, and the HC's internal state will be reset.
 * Don't bother to free memory.
 *
 * This will only ever be called with the main usb_hcd (the USB3 roothub).
 */
void xhci_shutdown(struct usb_hcd *hcd)
{
        struct xhci_hcd *xhci = hcd_to_xhci(hcd);

        if (xhci->quirks & XHCI_SPURIOUS_REBOOT)
                usb_disable_xhci_ports(to_pci_dev(hcd->self.controller));

        spin_lock_irq(&xhci->lock);
        xhci_halt(xhci);
        /* Workaround for spurious wakeups at shutdown with HSW */
        if (xhci->quirks & XHCI_SPURIOUS_WAKEUP)
                xhci_reset(xhci);
        spin_unlock_irq(&xhci->lock);

        xhci_cleanup_msix(xhci);

        xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                        "xhci_shutdown completed - status = %x",
                        readl(&xhci->op_regs->status));

        /* Yet another workaround for spurious wakeups at shutdown with HSW */
        if (xhci->quirks & XHCI_SPURIOUS_WAKEUP)
                pci_set_power_state(to_pci_dev(hcd->self.controller), PCI_D3hot);
}

#ifdef CONFIG_PM
static void xhci_save_registers(struct xhci_hcd *xhci)
{
        xhci->s3.command = readl(&xhci->op_regs->command);
        xhci->s3.dev_nt = readl(&xhci->op_regs->dev_notification);
        xhci->s3.dcbaa_ptr = xhci_read_64(xhci, &xhci->op_regs->dcbaa_ptr);
        xhci->s3.config_reg = readl(&xhci->op_regs->config_reg);
        xhci->s3.erst_size = readl(&xhci->ir_set->erst_size);
        xhci->s3.erst_base = xhci_read_64(xhci, &xhci->ir_set->erst_base);
        xhci->s3.erst_dequeue = xhci_read_64(xhci, &xhci->ir_set->erst_dequeue);
        xhci->s3.irq_pending = readl(&xhci->ir_set->irq_pending);
        xhci->s3.irq_control = readl(&xhci->ir_set->irq_control);
}

static void xhci_restore_registers(struct xhci_hcd *xhci)
{
        writel(xhci->s3.command, &xhci->op_regs->command);
        writel(xhci->s3.dev_nt, &xhci->op_regs->dev_notification);
        xhci_write_64(xhci, xhci->s3.dcbaa_ptr, &xhci->op_regs->dcbaa_ptr);
        writel(xhci->s3.config_reg, &xhci->op_regs->config_reg);
        writel(xhci->s3.erst_size, &xhci->ir_set->erst_size);
        xhci_write_64(xhci, xhci->s3.erst_base, &xhci->ir_set->erst_base);
        xhci_write_64(xhci, xhci->s3.erst_dequeue, &xhci->ir_set->erst_dequeue);
        writel(xhci->s3.irq_pending, &xhci->ir_set->irq_pending);
        writel(xhci->s3.irq_control, &xhci->ir_set->irq_control);
}

static void xhci_set_cmd_ring_deq(struct xhci_hcd *xhci)
{
        u64     val_64;

        /* step 2: initialize command ring buffer */
        val_64 = xhci_read_64(xhci, &xhci->op_regs->cmd_ring);
        val_64 = (val_64 & (u64) CMD_RING_RSVD_BITS) |
                (xhci_trb_virt_to_dma(xhci->cmd_ring->deq_seg,
                                      xhci->cmd_ring->dequeue) &
                 (u64) ~CMD_RING_RSVD_BITS) |
                xhci->cmd_ring->cycle_state;
        xhci_dbg_trace(xhci, trace_xhci_dbg_init,
                        "// Setting command ring address to 0x%llx",
                        (long unsigned long) val_64);
        xhci_write_64(xhci, val_64, &xhci->op_regs->cmd_ring);
}

/*
 * The whole command ring must be cleared to zero when we suspend the host.
 *
 * The host doesn't save the command ring pointer in the suspend well, so we
 * need to re-program it on resume.  Unfortunately, the pointer must be 64-byte
 * aligned, because of the reserved bits in the command ring dequeue pointer
 * register.  Therefore, we can't just set the dequeue pointer back in the
 * middle of the ring (TRBs are 16-byte aligned).
 */
static void xhci_clear_command_ring(struct xhci_hcd *xhci)
{
        struct xhci_ring *ring;
        struct xhci_segment *seg;

        ring = xhci->cmd_ring;
        seg = ring->deq_seg;
        do {
                memset(seg->trbs, 0,
                        sizeof(union xhci_trb) * (TRBS_PER_SEGMENT - 1));
                seg->trbs[TRBS_PER_SEGMENT - 1].link.control &=
                        cpu_to_le32(~TRB_CYCLE);
                seg = seg->next;
        } while (seg != ring->deq_seg);

        /* Reset the software enqueue and dequeue pointers */
        ring->deq_seg = ring->first_seg;
        ring->dequeue = ring->first_seg->trbs;
        ring->enq_seg = ring->deq_seg;
        ring->enqueue = ring->dequeue;

        ring->num_trbs_free = ring->num_segs * (TRBS_PER_SEGMENT - 1) - 1;
        /*
         * Ring is now zeroed, so the HW should look for change of ownership
         * when the cycle bit is set to 1.
         */
        ring->cycle_state = 1;

        /*
         * Reset the hardware dequeue pointer.
         * Yes, this will need to be re-written after resume, but we're paranoid
         * and want to make sure the hardware doesn't access bogus memory
         * because, say, the BIOS or an SMI started the host without changing
         * the command ring pointers.
         */
        xhci_set_cmd_ring_deq(xhci);
}

static void xhci_disable_port_wake_on_bits(struct xhci_hcd *xhci)
{
        int port_index;
        __le32 __iomem **port_array;
        unsigned long flags;
        u32 t1, t2;

        spin_lock_irqsave(&xhci->lock, flags);

        /* disble usb3 ports Wake bits*/
        port_index = xhci->num_usb3_ports;
        port_array = xhci->usb3_ports;
        while (port_index--) {
                t1 = readl(port_array[port_index]);
                t1 = xhci_port_state_to_neutral(t1);
                t2 = t1 & ~PORT_WAKE_BITS;
                if (t1 != t2)
                        writel(t2, port_array[port_index]);
        }

        /* disble usb2 ports Wake bits*/
        port_index = xhci->num_usb2_ports;
        port_array = xhci->usb2_ports;
        while (port_index--) {
                t1 = readl(port_array[port_index]);
                t1 = xhci_port_state_to_neutral(t1);
                t2 = t1 & ~PORT_WAKE_BITS;
                if (t1 != t2)
                        writel(t2, port_array[port_index]);
        }

        spin_unlock_irqrestore(&xhci->lock, flags);
}

/*
 * Stop HC (not bus-specific)
 *
 * This is called when the machine transition into S3/S4 mode.
 *
 */
int xhci_suspend(struct xhci_hcd *xhci, bool do_wakeup)
{
        int                     rc = 0;
        unsigned int            delay = XHCI_MAX_HALT_USEC;
        struct usb_hcd          *hcd = xhci_to_hcd(xhci);
        u32                     command;

        if (!hcd->state)
                return 0;

        if (hcd->state != HC_STATE_SUSPENDED ||
                        xhci->shared_hcd->state != HC_STATE_SUSPENDED)
                return -EINVAL;

        /* Clear root port wake on bits if wakeup not allowed. */
        if (!do_wakeup)
                xhci_disable_port_wake_on_bits(xhci);

        /* Don't poll the roothubs on bus suspend. */
        xhci_dbg(xhci, "%s: stopping port polling.\n", __func__);
        clear_bit(HCD_FLAG_POLL_RH, &hcd->flags);
        del_timer_sync(&hcd->rh_timer);
        clear_bit(HCD_FLAG_POLL_RH, &xhci->shared_hcd->flags);
        del_timer_sync(&xhci->shared_hcd->rh_timer);

        spin_lock_irq(&xhci->lock);
        clear_bit(HCD_FLAG_HW_ACCESSIBLE, &hcd->flags);
        clear_bit(HCD_FLAG_HW_ACCESSIBLE, &xhci->shared_hcd->flags);
        /* step 1: stop endpoint */
        /* skipped assuming that port suspend has done */

        /* step 2: clear Run/Stop bit */
        command = readl(&xhci->op_regs->command);
        command &= ~CMD_RUN;
        writel(command, &xhci->op_regs->command);

        /* Some chips from Fresco Logic need an extraordinary delay */
        delay *= (xhci->quirks & XHCI_SLOW_SUSPEND) ? 10 : 1;

        if (xhci_handshake(&xhci->op_regs->status,
                      STS_HALT, STS_HALT, delay)) {
                xhci_warn(xhci, "WARN: xHC CMD_RUN timeout\n");
                spin_unlock_irq(&xhci->lock);
                return -ETIMEDOUT;
        }
        xhci_clear_command_ring(xhci);

        /* step 3: save registers */
        xhci_save_registers(xhci);

        /* step 4: set CSS flag */
        command = readl(&xhci->op_regs->command);
        command |= CMD_CSS;
        writel(command, &xhci->op_regs->command);
        if (xhci_handshake(&xhci->op_regs->status,
                                STS_SAVE, 0, 10 * 1000)) {
                xhci_warn(xhci, "WARN: xHC save state timeout\n");
                spin_unlock_irq(&xhci->lock);
                return -ETIMEDOUT;
        }
        spin_unlock_irq(&xhci->lock);

        /*
         * Deleting Compliance Mode Recovery Timer because the xHCI Host
         * is about to be suspended.
         */
        if ((xhci->quirks & XHCI_COMP_MODE_QUIRK) &&
                        (!(xhci_all_ports_seen_u0(xhci)))) {
                del_timer_sync(&xhci->comp_mode_recovery_timer);
                xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                                "%s: compliance mode recovery timer deleted",
                                __func__);
        }

        /* step 5: remove core well power */
        /* synchronize irq when using MSI-X */
        xhci_msix_sync_irqs(xhci);

        return rc;
}
EXPORT_SYMBOL_GPL(xhci_suspend);

/*
 * start xHC (not bus-specific)
 *
 * This is called when the machine transition from S3/S4 mode.
 *
 */
int xhci_resume(struct xhci_hcd *xhci, bool hibernated)
{
        u32                     command, temp = 0, status;
        struct usb_hcd          *hcd = xhci_to_hcd(xhci);
        struct usb_hcd          *secondary_hcd;
        int                     retval = 0;
        bool                    comp_timer_running = false;

        if (!hcd->state)
                return 0;

        /* Wait a bit if either of the roothubs need to settle from the
         * transition into bus suspend.
         */
        if (time_before(jiffies, xhci->bus_state[0].next_statechange) ||
                        time_before(jiffies,
                                xhci->bus_state[1].next_statechange))
                msleep(100);

        set_bit(HCD_FLAG_HW_ACCESSIBLE, &hcd->flags);
        set_bit(HCD_FLAG_HW_ACCESSIBLE, &xhci->shared_hcd->flags);

        spin_lock_irq(&xhci->lock);
        if (xhci->quirks & XHCI_RESET_ON_RESUME)
                hibernated = true;

        if (!hibernated) {
                /* step 1: restore register */
                xhci_restore_registers(xhci);
                /* step 2: initialize command ring buffer */
                xhci_set_cmd_ring_deq(xhci);

                /* step 3: restore state and start state*/
                /* step 3: set CRS flag */
                command = readl(&xhci->op_regs->command);
                command |= CMD_CRS;
                writel(command, &xhci->op_regs->command);
                if (xhci_handshake(&xhci->op_regs->status,
                              STS_RESTORE, 0, 10 * 1000)) {
                        xhci_warn(xhci, "WARN: xHC restore state timeout\n");
                        spin_unlock_irq(&xhci->lock);
                        return -ETIMEDOUT;
                }
                temp = readl(&xhci->op_regs->status);
        }

        /* If restore operation fails, re-initialize the HC during resume */
        if ((temp & STS_SRE) || hibernated) {

                if ((xhci->quirks & XHCI_COMP_MODE_QUIRK) &&
                                !(xhci_all_ports_seen_u0(xhci))) {
                        del_timer_sync(&xhci->comp_mode_recovery_timer);
                        xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                                "Compliance Mode Recovery Timer deleted!");
                }

                /* Let the USB core know _both_ roothubs lost power. */
                usb_root_hub_lost_power(xhci->main_hcd->self.root_hub);
                usb_root_hub_lost_power(xhci->shared_hcd->self.root_hub);

                xhci_dbg(xhci, "Stop HCD\n");
                xhci_halt(xhci);
                xhci_reset(xhci);
                spin_unlock_irq(&xhci->lock);
                xhci_cleanup_msix(xhci);

                xhci_dbg(xhci, "// Disabling event ring interrupts\n");
                temp = readl(&xhci->op_regs->status);
                writel(temp & ~STS_EINT, &xhci->op_regs->status);
                temp = readl(&xhci->ir_set->irq_pending);
                writel(ER_IRQ_DISABLE(temp), &xhci->ir_set->irq_pending);
                xhci_print_ir_set(xhci, 0);

                xhci_dbg(xhci, "cleaning up memory\n");
                xhci_mem_cleanup(xhci);
                xhci_dbg(xhci, "xhci_stop completed - status = %x\n",
                            readl(&xhci->op_regs->status));

                /* USB core calls the PCI reinit and start functions twice:
                 * first with the primary HCD, and then with the secondary HCD.
                 * If we don't do the same, the host will never be started.
                 */
                if (!usb_hcd_is_primary_hcd(hcd))
                        secondary_hcd = hcd;
                else
                        secondary_hcd = xhci->shared_hcd;

                xhci_dbg(xhci, "Initialize the xhci_hcd\n");
                retval = xhci_init(hcd->primary_hcd);
                if (retval)
                        return retval;
                comp_timer_running = true;

                xhci_dbg(xhci, "Start the primary HCD\n");
                retval = xhci_run(hcd->primary_hcd);
                if (!retval) {
                        xhci_dbg(xhci, "Start the secondary HCD\n");
                        retval = xhci_run(secondary_hcd);
                }
                hcd->state = HC_STATE_SUSPENDED;
                xhci->shared_hcd->state = HC_STATE_SUSPENDED;
                goto done;
        }

        /* step 4: set Run/Stop bit */
        command = readl(&xhci->op_regs->command);
        command |= CMD_RUN;
        writel(command, &xhci->op_regs->command);
        xhci_handshake(&xhci->op_regs->status, STS_HALT,
                  0, 250 * 1000);

        /* step 5: walk topology and initialize portsc,
         * portpmsc and portli
         */
        /* this is done in bus_resume */

        /* step 6: restart each of the previously
         * Running endpoints by ringing their doorbells
         */

        spin_unlock_irq(&xhci->lock);

 done:
        if (retval == 0) {
                /* Resume root hubs only when have pending events. */
                status = readl(&xhci->op_regs->status);
                if (status & STS_EINT) {
                        usb_hcd_resume_root_hub(hcd);
                        usb_hcd_resume_root_hub(xhci->shared_hcd);
                }
        }

        /*
         * If system is subject to the Quirk, Compliance Mode Timer needs to
         * be re-initialized Always after a system resume. Ports are subject
         * to suffer the Compliance Mode issue again. It doesn't matter if
         * ports have entered previously to U0 before system's suspension.
         */
        if ((xhci->quirks & XHCI_COMP_MODE_QUIRK) && !comp_timer_running)
                compliance_mode_recovery_timer_init(xhci);

        /* Re-enable port polling. */
        xhci_dbg(xhci, "%s: starting port polling.\n", __func__);
        set_bit(HCD_FLAG_POLL_RH, &hcd->flags);
        usb_hcd_poll_rh_status(hcd);
        set_bit(HCD_FLAG_POLL_RH, &xhci->shared_hcd->flags);
        usb_hcd_poll_rh_status(xhci->shared_hcd);

        return retval;
}
EXPORT_SYMBOL_GPL(xhci_resume);
#endif  /* CONFIG_PM */

/*-------------------------------------------------------------------------*/

/**
 * xhci_get_endpoint_index - Used for passing endpoint bitmasks between the core and
 * HCDs.  Find the index for an endpoint given its descriptor.  Use the return
 * value to right shift 1 for the bitmask.
 *
 * Index  = (epnum * 2) + direction - 1,
 * where direction = 0 for OUT, 1 for IN.
 * For control endpoints, the IN index is used (OUT index is unused), so
 * index = (epnum * 2) + direction - 1 = (epnum * 2) + 1 - 1 = (epnum * 2)
 */
unsigned int xhci_get_endpoint_index(struct usb_endpoint_descriptor *desc)
{
        unsigned int index;
        if (usb_endpoint_xfer_control(desc))
                index = (unsigned int) (usb_endpoint_num(desc)*2);
        else
                index = (unsigned int) (usb_endpoint_num(desc)*2) +
                        (usb_endpoint_dir_in(desc) ? 1 : 0) - 1;
        return index;
}

/* The reverse operation to xhci_get_endpoint_index. Calculate the USB endpoint
 * address from the XHCI endpoint index.
 */
unsigned int xhci_get_endpoint_address(unsigned int ep_index)
{
        unsigned int number = DIV_ROUND_UP(ep_index, 2);
        unsigned int direction = ep_index % 2 ? USB_DIR_OUT : USB_DIR_IN;
        return direction | number;
}

/* Find the flag for this endpoint (for use in the control context).  Use the
 * endpoint index to create a bitmask.  The slot context is bit 0, endpoint 0 is
 * bit 1, etc.
 */
unsigned int xhci_get_endpoint_flag(struct usb_endpoint_descriptor *desc)
{
        return 1 << (xhci_get_endpoint_index(desc) + 1);
}

/* Find the flag for this endpoint (for use in the control context).  Use the
 * endpoint index to create a bitmask.  The slot context is bit 0, endpoint 0 is
 * bit 1, etc.
 */
unsigned int xhci_get_endpoint_flag_from_index(unsigned int ep_index)
{
        return 1 << (ep_index + 1);
}

/* Compute the last valid endpoint context index.  Basically, this is the
 * endpoint index plus one.  For slot contexts with more than valid endpoint,
 * we find the most significant bit set in the added contexts flags.
 * e.g. ep 1 IN (with epnum 0x81) => added_ctxs = 0b1000
 * fls(0b1000) = 4, but the endpoint context index is 3, so subtract one.
 */
unsigned int xhci_last_valid_endpoint(u32 added_ctxs)
{
        return fls(added_ctxs) - 1;
}

/* Returns 1 if the arguments are OK;
 * returns 0 this is a root hub; returns -EINVAL for NULL pointers.
 */
static int xhci_check_args(struct usb_hcd *hcd, struct usb_device *udev,
                struct usb_host_endpoint *ep, int check_ep, bool check_virt_dev,
                const char *func) {
        struct xhci_hcd *xhci;
        struct xhci_virt_device *virt_dev;

        if (!hcd || (check_ep && !ep) || !udev) {
                pr_debug("xHCI %s called with invalid args\n", func);
                return -EINVAL;
        }
        if (!udev->parent) {
                pr_debug("xHCI %s called for root hub\n", func);
                return 0;
        }

        xhci = hcd_to_xhci(hcd);
        if (check_virt_dev) {
                if (!udev->slot_id || !xhci->devs[udev->slot_id]) {
                        xhci_dbg(xhci, "xHCI %s called with unaddressed device\n",
                                        func);
                        return -EINVAL;
                }

                virt_dev = xhci->devs[udev->slot_id];
                if (virt_dev->udev != udev) {
                        xhci_dbg(xhci, "xHCI %s called with udev and "
                                          "virt_dev does not match\n", func);
                        return -EINVAL;
                }
        }

        if (xhci->xhc_state & XHCI_STATE_HALTED)
                return -ENODEV;

        return 1;
}

static int xhci_configure_endpoint(struct xhci_hcd *xhci,
                struct usb_device *udev, struct xhci_command *command,
                bool ctx_change, bool must_succeed);

/*
 * Full speed devices may have a max packet size greater than 8 bytes, but the
 * USB core doesn't know that until it reads the first 8 bytes of the
 * descriptor.  If the usb_device's max packet size changes after that point,
 * we need to issue an evaluate context command and wait on it.
 */
static int xhci_check_maxpacket(struct xhci_hcd *xhci, unsigned int slot_id,
                unsigned int ep_index, struct urb *urb)
{
        struct xhci_container_ctx *out_ctx;
        struct xhci_input_control_ctx *ctrl_ctx;
        struct xhci_ep_ctx *ep_ctx;
        struct xhci_command *command;
        int max_packet_size;
        int hw_max_packet_size;
        int ret = 0;

        out_ctx = xhci->devs[slot_id]->out_ctx;
        ep_ctx = xhci_get_ep_ctx(xhci, out_ctx, ep_index);
        hw_max_packet_size = MAX_PACKET_DECODED(le32_to_cpu(ep_ctx->ep_info2));
        max_packet_size = usb_endpoint_maxp(&urb->dev->ep0.desc);
        if (hw_max_packet_size != max_packet_size) {
                xhci_dbg_trace(xhci,  trace_xhci_dbg_context_change,
                                "Max Packet Size for ep 0 changed.");
                xhci_dbg_trace(xhci,  trace_xhci_dbg_context_change,
                                "Max packet size in usb_device = %d",
                                max_packet_size);
                xhci_dbg_trace(xhci,  trace_xhci_dbg_context_change,
                                "Max packet size in xHCI HW = %d",
                                hw_max_packet_size);
                xhci_dbg_trace(xhci,  trace_xhci_dbg_context_change,
                                "Issuing evaluate context command.");

                /* Set up the input context flags for the command */
                /* FIXME: This won't work if a non-default control endpoint
                 * changes max packet sizes.
                 */

                command = xhci_alloc_command(xhci, false, true, GFP_KERNEL);
                if (!command)
                        return -ENOMEM;

                command->in_ctx = xhci->devs[slot_id]->in_ctx;
                ctrl_ctx = xhci_get_input_control_ctx(command->in_ctx);
                if (!ctrl_ctx) {
                        xhci_warn(xhci, "%s: Could not get input context, bad type.\n",
                                        __func__);
                        ret = -ENOMEM;
                        goto command_cleanup;
                }
                /* Set up the modified control endpoint 0 */
                xhci_endpoint_copy(xhci, xhci->devs[slot_id]->in_ctx,
                                xhci->devs[slot_id]->out_ctx, ep_index);

                ep_ctx = xhci_get_ep_ctx(xhci, command->in_ctx, ep_index);
                ep_ctx->ep_info2 &= cpu_to_le32(~MAX_PACKET_MASK);
                ep_ctx->ep_info2 |= cpu_to_le32(MAX_PACKET(max_packet_size));

                ctrl_ctx->add_flags = cpu_to_le32(EP0_FLAG);
                ctrl_ctx->drop_flags = 0;

                xhci_dbg(xhci, "Slot %d input context\n", slot_id);
                xhci_dbg_ctx(xhci, command->in_ctx, ep_index);
                xhci_dbg(xhci, "Slot %d output context\n", slot_id);
                xhci_dbg_ctx(xhci, out_ctx, ep_index);

                ret = xhci_configure_endpoint(xhci, urb->dev, command,
                                true, false);

                /* Clean up the input context for later use by bandwidth
                 * functions.
                 */
                ctrl_ctx->add_flags = cpu_to_le32(SLOT_FLAG);
command_cleanup:
                kfree(command->completion);
                kfree(command);
        }
        return ret;
}

/*
 * non-error returns are a promise to giveback() the urb later
 * we drop ownership so next owner (or urb unlink) can get it
 */
int xhci_urb_enqueue(struct usb_hcd *hcd, struct urb *urb, gfp_t mem_flags)
{
        struct xhci_hcd *xhci = hcd_to_xhci(hcd);
        struct xhci_td *buffer;
        unsigned long flags;
        int ret = 0;
        unsigned int slot_id, ep_index;
        struct urb_priv *urb_priv;
        int size, i;

        if (!urb || xhci_check_args(hcd, urb->dev, urb->ep,
                                        true, true, __func__) <= 0)
                return -EINVAL;

        slot_id = urb->dev->slot_id;
        ep_index = xhci_get_endpoint_index(&urb->ep->desc);

        if (!HCD_HW_ACCESSIBLE(hcd)) {
                if (!in_interrupt())
                        xhci_dbg(xhci, "urb submitted during PCI suspend\n");
                ret = -ESHUTDOWN;
                goto exit;
        }

        if (usb_endpoint_xfer_isoc(&urb->ep->desc))
                size = urb->number_of_packets;
        else if (usb_endpoint_is_bulk_out(&urb->ep->desc) &&
            urb->transfer_buffer_length > 0 &&
            urb->transfer_flags & URB_ZERO_PACKET &&
            !(urb->transfer_buffer_length % usb_endpoint_maxp(&urb->ep->desc)))
                size = 2;
        else
                size = 1;

        urb_priv = kzalloc(sizeof(struct urb_priv) +
                                  size * sizeof(struct xhci_td *), mem_flags);
        if (!urb_priv)
                return -ENOMEM;

        buffer = kzalloc(size * sizeof(struct xhci_td), mem_flags);
        if (!buffer) {
                kfree(urb_priv);
                return -ENOMEM;
        }

        for (i = 0; i < size; i++) {
                urb_priv->td[i] = buffer;
                buffer++;
        }

        urb_priv->length = size;
        urb_priv->td_cnt = 0;
        urb->hcpriv = urb_priv;

        if (usb_endpoint_xfer_control(&urb->ep->desc)) {
                /* Check to see if the max packet size for the default control
                 * endpoint changed during FS device enumeration
                 */
                if (urb->dev->speed == USB_SPEED_FULL) {
                        ret = xhci_check_maxpacket(xhci, slot_id,
                                        ep_index, urb);
                        if (ret < 0) {
                                xhci_urb_free_priv(urb_priv);
                                urb->hcpriv = NULL;
                                return ret;
                        }
                }

                /* We have a spinlock and interrupts disabled, so we must pass
                 * atomic context to this function, which may allocate memory.
                 */
                spin_lock_irqsave(&xhci->lock, flags);
                if (xhci->xhc_state & XHCI_STATE_DYING)
                        goto dying;
                ret = xhci_queue_ctrl_tx(xhci, GFP_ATOMIC, urb,
                                slot_id, ep_index);
                if (ret)
                        goto free_priv;
                spin_unlock_irqrestore(&xhci->lock, flags);
        } else if (usb_endpoint_xfer_bulk(&urb->ep->desc)) {
                spin_lock_irqsave(&xhci->lock, flags);
                if (xhci->xhc_state & XHCI_STATE_DYING)
                        goto dying;
                if (xhci->devs[slot_id]->eps[ep_index].ep_state &
                                EP_GETTING_STREAMS) {
                        xhci_warn(xhci, "WARN: Can't enqueue URB while bulk ep "
                                        "is transitioning to using streams.\n");
                        ret = -EINVAL;
                } else if (xhci->devs[slot_id]->eps[ep_index].ep_state &
                                EP_GETTING_NO_STREAMS) {
                        xhci_warn(xhci, "WARN: Can't enqueue URB while bulk ep "
                                        "is transitioning to "
                                        "not having streams.\n");
                        ret = -EINVAL;
                } else {
                        ret = xhci_queue_bulk_tx(xhci, GFP_ATOMIC, urb,
                                        slot_id, ep_index);
                }
                if (ret)
                        goto free_priv;
                spin_unlock_irqrestore(&xhci->lock, flags);
        } else if (usb_endpoint_xfer_int(&urb->ep->desc)) {
                spin_lock_irqsave(&xhci->lock, flags);
                if (xhci->xhc_state & XHCI_STATE_DYING)
                        goto dying;
                ret = xhci_queue_intr_tx(xhci, GFP_ATOMIC, urb,
                                slot_id, ep_index);
                if (ret)
                        goto free_priv;
                spin_unlock_irqrestore(&xhci->lock, flags);
        } else {
                spin_lock_irqsave(&xhci->lock, flags);
                if (xhci->xhc_state & XHCI_STATE_DYING)
                        goto dying;
                ret = xhci_queue_isoc_tx_prepare(xhci, GFP_ATOMIC, urb,
                                slot_id, ep_index);
                if (ret)
                        goto free_priv;
                spin_unlock_irqrestore(&xhci->lock, flags);
        }
exit:
        return ret;
dying:
        xhci_dbg(xhci, "Ep 0x%x: URB %p submitted for "
                        "non-responsive xHCI host.\n",
                        urb->ep->desc.bEndpointAddress, urb);
        ret = -ESHUTDOWN;
free_priv:
        xhci_urb_free_priv(urb_priv);
        urb->hcpriv = NULL;
        spin_unlock_irqrestore(&xhci->lock, flags);
        return ret;
}

/* Get the right ring for the given URB.
 * If the endpoint supports streams, boundary check the URB's stream ID.
 * If the endpoint doesn't support streams, return the singular endpoint ring.
 */
static struct xhci_ring *xhci_urb_to_transfer_ring(struct xhci_hcd *xhci,
                struct urb *urb)
{
        unsigned int slot_id;
        unsigned int ep_index;
        unsigned int stream_id;
        struct xhci_virt_ep *ep;

        slot_id = urb->dev->slot_id;
        ep_index = xhci_get_endpoint_index(&urb->ep->desc);
        stream_id = urb->stream_id;
        ep = &xhci->devs[slot_id]->eps[ep_index];
        /* Common case: no streams */
        if (!(ep->ep_state & EP_HAS_STREAMS))
                return ep->ring;

        if (stream_id == 0) {
                xhci_warn(xhci,
                                "WARN: Slot ID %u, ep index %u has streams, "
                                "but URB has no stream ID.\n",
                                slot_id, ep_index);
                return NULL;
        }

        if (stream_id < ep->stream_info->num_streams)
                return ep->stream_info->stream_rings[stream_id];

        xhci_warn(xhci,
                        "WARN: Slot ID %u, ep index %u has "
                        "stream IDs 1 to %u allocated, "
                        "but stream ID %u is requested.\n",
                        slot_id, ep_index,
                        ep->stream_info->num_streams - 1,
                        stream_id);
        return NULL;
}

/*
 * Remove the URB's TD from the endpoint ring.  This may cause the HC to stop
 * USB transfers, potentially stopping in the middle of a TRB buffer.  The HC
 * should pick up where it left off in the TD, unless a Set Transfer Ring
 * Dequeue Pointer is issued.
 *
 * The TRBs that make up the buffers for the canceled URB will be "removed" from
 * the ring.  Since the ring is a contiguous structure, they can't be physically
 * removed.  Instead, there are two options:
 *
 *  1) If the HC is in the middle of processing the URB to be canceled, we
 *     simply move the ring's dequeue pointer past those TRBs using the Set
 *     Transfer Ring Dequeue Pointer command.  This will be the common case,
 *     when drivers timeout on the last submitted URB and attempt to cancel.
 *
 *  2) If the HC is in the middle of a different TD, we turn the TRBs into a
 *     series of 1-TRB transfer no-op TDs.  (No-ops shouldn't be chained.)  The
 *     HC will need to invalidate the any TRBs it has cached after the stop
 *     endpoint command, as noted in the xHCI 0.95 errata.
 *
 *  3) The TD may have completed by the time the Stop Endpoint Command
 *     completes, so software needs to handle that case too.
 *
 * This function should protect against the TD enqueueing code ringing the
 * doorbell while this code is waiting for a Stop Endpoint command to complete.
 * It also needs to account for multiple cancellations on happening at the same
 * time for the same endpoint.
 *
 * Note that this function can be called in any context, or so says
 * usb_hcd_unlink_urb()
 */
int xhci_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status)
{
        unsigned long flags;
        int ret, i;
        u32 temp;
        struct xhci_hcd *xhci;
        struct urb_priv *urb_priv;
        struct xhci_td *td;
        unsigned int ep_index;
        struct xhci_ring *ep_ring;
        struct xhci_virt_ep *ep;
        struct xhci_command *command;

        xhci = hcd_to_xhci(hcd);
        spin_lock_irqsave(&xhci->lock, flags);
        /* Make sure the URB hasn't completed or been unlinked already */
        ret = usb_hcd_check_unlink_urb(hcd, urb, status);
        if (ret || !urb->hcpriv)
                goto done;
        temp = readl(&xhci->op_regs->status);
        if (temp == 0xffffffff || (xhci->xhc_state & XHCI_STATE_HALTED)) {
                xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
                                "HW died, freeing TD.");
                urb_priv = urb->hcpriv;
                for (i = urb_priv->td_cnt; i < urb_priv->length; i++) {
                        td = urb_priv->td[i];
                        if (!list_empty(&td->td_list))
                                list_del_init(&td->td_list);
                        if (!list_empty(&td->cancelled_td_list))
                                list_del_init(&td->cancelled_td_list);
                }

                usb_hcd_unlink_urb_from_ep(hcd, urb);
                spin_unlock_irqrestore(&xhci->lock, flags);
                usb_hcd_giveback_urb(hcd, urb, -ESHUTDOWN);
                xhci_urb_free_priv(urb_priv);
                return ret;
        }
        if ((xhci->xhc_state & XHCI_STATE_DYING) ||
                        (xhci->xhc_state & XHCI_STATE_HALTED)) {
                xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
                                "Ep 0x%x: URB %p to be canceled on "
                                "non-responsive xHCI host.",
                                urb->ep->desc.bEndpointAddress, urb);
                /* Let the stop endpoint command watchdog timer (which set this
                 * state) finish cleaning up the endpoint TD lists.  We must
                 * have caught it in the middle of dropping a lock and giving
                 * back an URB.
                 */
                goto done;
        }

        ep_index = xhci_get_endpoint_index(&urb->ep->desc);
        ep = &xhci->devs[urb->dev->slot_id]->eps[ep_index];
        ep_ring = xhci_urb_to_transfer_ring(xhci, urb);
        if (!ep_ring) {
                ret = -EINVAL;
                goto done;
        }

        urb_priv = urb->hcpriv;
        i = urb_priv->td_cnt;
        if (i < urb_priv->length)
                xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
                                "Cancel URB %p, dev %s, ep 0x%x, "
                                "starting at offset 0x%llx",
                                urb, urb->dev->devpath,
                                urb->ep->desc.bEndpointAddress,
                                (unsigned long long) xhci_trb_virt_to_dma(
                                        urb_priv->td[i]->start_seg,
                                        urb_priv->td[i]->first_trb));

        for (; i < urb_priv->length; i++) {
                td = urb_priv->td[i];
                list_add_tail(&td->cancelled_td_list, &ep->cancelled_td_list);
        }

        /* Queue a stop endpoint command, but only if this is
         * the first cancellation to be handled.
         */
        if (!(ep->ep_state & EP_HALT_PENDING)) {
                command = xhci_alloc_command(xhci, false, false, GFP_ATOMIC);
                if (!command) {
                        ret = -ENOMEM;
                        goto done;
                }
                ep->ep_state |= EP_HALT_PENDING;
                ep->stop_cmds_pending++;
                ep->stop_cmd_timer.expires = jiffies +
                        XHCI_STOP_EP_CMD_TIMEOUT * HZ;
                add_timer(&ep->stop_cmd_timer);
                xhci_queue_stop_endpoint(xhci, command, urb->dev->slot_id,
                                         ep_index, 0);
                xhci_ring_cmd_db(xhci);
        }
done:
        spin_unlock_irqrestore(&xhci->lock, flags);
        return ret;
}

/* Drop an endpoint from a new bandwidth configuration for this device.
 * Only one call to this function is allowed per endpoint before
 * check_bandwidth() or reset_bandwidth() must be called.
 * A call to xhci_drop_endpoint() followed by a call to xhci_add_endpoint() will
 * add the endpoint to the schedule with possibly new parameters denoted by a
 * different endpoint descriptor in usb_host_endpoint.
 * A call to xhci_add_endpoint() followed by a call to xhci_drop_endpoint() is
 * not allowed.
 *
 * The USB core will not allow URBs to be queued to an endpoint that is being
 * disabled, so there's no need for mutual exclusion to protect
 * the xhci->devs[slot_id] structure.
 */
int xhci_drop_endpoint(struct usb_hcd *hcd, struct usb_device *udev,
                struct usb_host_endpoint *ep)
{
        struct xhci_hcd *xhci;
        struct xhci_container_ctx *in_ctx, *out_ctx;
        struct xhci_input_control_ctx *ctrl_ctx;
        unsigned int ep_index;
        struct xhci_ep_ctx *ep_ctx;
        u32 drop_flag;
        u32 new_add_flags, new_drop_flags;
        int ret;

        ret = xhci_check_args(hcd, udev, ep, 1, true, __func__);
        if (ret <= 0)
                return ret;
        xhci = hcd_to_xhci(hcd);
        if (xhci->xhc_state & XHCI_STATE_DYING)
                return -ENODEV;

        xhci_dbg(xhci, "%s called for udev %p\n", __func__, udev);
        drop_flag = xhci_get_endpoint_flag(&ep->desc);
        if (drop_flag == SLOT_FLAG || drop_flag == EP0_FLAG) {
                xhci_dbg(xhci, "xHCI %s - can't drop slot or ep 0 %#x\n",
                                __func__, drop_flag);
                return 0;
        }

        in_ctx = xhci->devs[udev->slot_id]->in_ctx;
        out_ctx = xhci->devs[udev->slot_id]->out_ctx;
        ctrl_ctx = xhci_get_input_control_ctx(in_ctx);
        if (!ctrl_ctx) {
                xhci_warn(xhci, "%s: Could not get input context, bad type.\n",
                                __func__);
                return 0;
        }

        ep_index = xhci_get_endpoint_index(&ep->desc);
        ep_ctx = xhci_get_ep_ctx(xhci, out_ctx, ep_index);
        /* If the HC already knows the endpoint is disabled,
         * or the HCD has noted it is disabled, ignore this request
         */
        if (((ep_ctx->ep_info & cpu_to_le32(EP_STATE_MASK)) ==
             cpu_to_le32(EP_STATE_DISABLED)) ||
            le32_to_cpu(ctrl_ctx->drop_flags) &
            xhci_get_endpoint_flag(&ep->desc)) {
                /* Do not warn when called after a usb_device_reset */
                if (xhci->devs[udev->slot_id]->eps[ep_index].ring != NULL)
                        xhci_warn(xhci, "xHCI %s called with disabled ep %p\n",
                                  __func__, ep);
                return 0;
        }

        ctrl_ctx->drop_flags |= cpu_to_le32(drop_flag);
        new_drop_flags = le32_to_cpu(ctrl_ctx->drop_flags);

        ctrl_ctx->add_flags &= cpu_to_le32(~drop_flag);
        new_add_flags = le32_to_cpu(ctrl_ctx->add_flags);

        xhci_endpoint_zero(xhci, xhci->devs[udev->slot_id], ep);

        xhci_dbg(xhci, "drop ep 0x%x, slot id %d, new drop flags = %#x, new add flags = %#x\n",
                        (unsigned int) ep->desc.bEndpointAddress,
                        udev->slot_id,
                        (unsigned int) new_drop_flags,
                        (unsigned int) new_add_flags);
        return 0;
}

/* Add an endpoint to a new possible bandwidth configuration for this device.
 * Only one call to this function is allowed per endpoint before
 * check_bandwidth() or reset_bandwidth() must be called.
 * A call to xhci_drop_endpoint() followed by a call to xhci_add_endpoint() will
 * add the endpoint to the schedule with possibly new parameters denoted by a
 * different endpoint descriptor in usb_host_endpoint.
 * A call to xhci_add_endpoint() followed by a call to xhci_drop_endpoint() is
 * not allowed.
 *
 * The USB core will not allow URBs to be queued to an endpoint until the
 * configuration or alt setting is installed in the device, so there's no need
 * for mutual exclusion to protect the xhci->devs[slot_id] structure.
 */
int xhci_add_endpoint(struct usb_hcd *hcd, struct usb_device *udev,
                struct usb_host_endpoint *ep)
{
        struct xhci_hcd *xhci;
        struct xhci_container_ctx *in_ctx;
        unsigned int ep_index;
        struct xhci_input_control_ctx *ctrl_ctx;
        u32 added_ctxs;
        u32 new_add_flags, new_drop_flags;
        struct xhci_virt_device *virt_dev;
        int ret = 0;

        ret = xhci_check_args(hcd, udev, ep, 1, true, __func__);
        if (ret <= 0) {
                /* So we won't queue a reset ep command for a root hub */
                ep->hcpriv = NULL;
                return ret;
        }
        xhci = hcd_to_xhci(hcd);
        if (xhci->xhc_state & XHCI_STATE_DYING)
                return -ENODEV;

        added_ctxs = xhci_get_endpoint_flag(&ep->desc);
        if (added_ctxs == SLOT_FLAG || added_ctxs == EP0_FLAG) {
                /* FIXME when we have to issue an evaluate endpoint command to
                 * deal with ep0 max packet size changing once we get the
                 * descriptors
                 */
                xhci_dbg(xhci, "xHCI %s - can't add slot or ep 0 %#x\n",
                                __func__, added_ctxs);
                return 0;
        }

        virt_dev = xhci->devs[udev->slot_id];
        in_ctx = virt_dev->in_ctx;
        ctrl_ctx = xhci_get_input_control_ctx(in_ctx);
        if (!ctrl_ctx) {
                xhci_warn(xhci, "%s: Could not get input context, bad type.\n",
                                __func__);
                return 0;
        }

        ep_index = xhci_get_endpoint_index(&ep->desc);
        /* If this endpoint is already in use, and the upper layers are trying
         * to add it again without dropping it, reject the addition.
         */
        if (virt_dev->eps[ep_index].ring &&
                        !(le32_to_cpu(ctrl_ctx->drop_flags) & added_ctxs)) {
                xhci_warn(xhci, "Trying to add endpoint 0x%x "
                                "without dropping it.\n",
                                (unsigned int) ep->desc.bEndpointAddress);
                return -EINVAL;
        }

        /* If the HCD has already noted the endpoint is enabled,
         * ignore this request.
         */
        if (le32_to_cpu(ctrl_ctx->add_flags) & added_ctxs) {
                xhci_warn(xhci, "xHCI %s called with enabled ep %p\n",
                                __func__, ep);
                return 0;
        }

        /*
         * Configuration and alternate setting changes must be done in
         * process context, not interrupt context (or so documenation
         * for usb_set_interface() and usb_set_configuration() claim).
         */
        if (xhci_endpoint_init(xhci, virt_dev, udev, ep, GFP_NOIO) < 0) {
                dev_dbg(&udev->dev, "%s - could not initialize ep %#x\n",
                                __func__, ep->desc.bEndpointAddress);
                return -ENOMEM;
        }

        ctrl_ctx->add_flags |= cpu_to_le32(added_ctxs);
        new_add_flags = le32_to_cpu(ctrl_ctx->add_flags);

        /* If xhci_endpoint_disable() was called for this endpoint, but the
         * xHC hasn't been notified yet through the check_bandwidth() call,
         * this re-adds a new state for the endpoint from the new endpoint
         * descriptors.  We must drop and re-add this endpoint, so we leave the
         * drop flags alone.
         */
        new_drop_flags = le32_to_cpu(ctrl_ctx->drop_flags);

        /* Store the usb_device pointer for later use */
        ep->hcpriv = udev;

        xhci_dbg(xhci, "add ep 0x%x, slot id %d, new drop flags = %#x, new add flags = %#x\n",
                        (unsigned int) ep->desc.bEndpointAddress,
                        udev->slot_id,
                        (unsigned int) new_drop_flags,
                        (unsigned int) new_add_flags);
        return 0;
}

static void xhci_zero_in_ctx(struct xhci_hcd *xhci, struct xhci_virt_device *virt_dev)
{
        struct xhci_input_control_ctx *ctrl_ctx;
        struct xhci_ep_ctx *ep_ctx;
        struct xhci_slot_ctx *slot_ctx;
        int i;

        ctrl_ctx = xhci_get_input_control_ctx(virt_dev->in_ctx);
        if (!ctrl_ctx) {
                xhci_warn(xhci, "%s: Could not get input context, bad type.\n",
                                __func__);
                return;
        }

        /* When a device's add flag and drop flag are zero, any subsequent
         * configure endpoint command will leave that endpoint's state
         * untouched.  Make sure we don't leave any old state in the input
         * endpoint contexts.
         */
        ctrl_ctx->drop_flags = 0;
        ctrl_ctx->add_flags = 0;
        slot_ctx = xhci_get_slot_ctx(xhci, virt_dev->in_ctx);
        slot_ctx->dev_info &= cpu_to_le32(~LAST_CTX_MASK);
        /* Endpoint 0 is always valid */
        slot_ctx->dev_info |= cpu_to_le32(LAST_CTX(1));
        for (i = 1; i < 31; ++i) {
                ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, i);
                ep_ctx->ep_info = 0;
                ep_ctx->ep_info2 = 0;
                ep_ctx->deq = 0;
                ep_ctx->tx_info = 0;
        }
}

static int xhci_configure_endpoint_result(struct xhci_hcd *xhci,
                struct usb_device *udev, u32 *cmd_status)
{
        int ret;

        switch (*cmd_status) {
        case COMP_CMD_ABORT:
        case COMP_CMD_STOP:
                xhci_warn(xhci, "Timeout while waiting for configure endpoint command\n");
                ret = -ETIME;
                break;
        case COMP_ENOMEM:
                dev_warn(&udev->dev,
                         "Not enough host controller resources for new device state.\n");
                ret = -ENOMEM;
                /* FIXME: can we allocate more resources for the HC? */
                break;
        case COMP_BW_ERR:
        case COMP_2ND_BW_ERR:
                dev_warn(&udev->dev,
                         "Not enough bandwidth for new device state.\n");
                ret = -ENOSPC;
                /* FIXME: can we go back to the old state? */
                break;
        case COMP_TRB_ERR:
                /* the HCD set up something wrong */
                dev_warn(&udev->dev, "ERROR: Endpoint drop flag = 0, "
                                "add flag = 1, "
                                "and endpoint is not disabled.\n");
                ret = -EINVAL;
                break;
        case COMP_DEV_ERR:
                dev_warn(&udev->dev,
                         "ERROR: Incompatible device for endpoint configure command.\n");
                ret = -ENODEV;
                break;
        case COMP_SUCCESS:
                xhci_dbg_trace(xhci, trace_xhci_dbg_context_change,
                                "Successful Endpoint Configure command");
                ret = 0;
                break;
        default:
                xhci_err(xhci, "ERROR: unexpected command completion code 0x%x.\n",
                                *cmd_status);
                ret = -EINVAL;
                break;
        }
        return ret;
}

static int xhci_evaluate_context_result(struct xhci_hcd *xhci,
                struct usb_device *udev, u32 *cmd_status)
{
        int ret;
        struct xhci_virt_device *virt_dev = xhci->devs[udev->slot_id];

        switch (*cmd_status) {
        case COMP_CMD_ABORT:
        case COMP_CMD_STOP:
                xhci_warn(xhci, "Timeout while waiting for evaluate context command\n");
                ret = -ETIME;
                break;
        case COMP_EINVAL:
                dev_warn(&udev->dev,
                         "WARN: xHCI driver setup invalid evaluate context command.\n");
                ret = -EINVAL;
                break;
        case COMP_EBADSLT:
                dev_warn(&udev->dev,
                        "WARN: slot not enabled for evaluate context command.\n");
                ret = -EINVAL;
                break;
        case COMP_CTX_STATE:
                dev_warn(&udev->dev,
                        "WARN: invalid context state for evaluate context command.\n");
                xhci_dbg_ctx(xhci, virt_dev->out_ctx, 1);
                ret = -EINVAL;
                break;
        case COMP_DEV_ERR:
                dev_warn(&udev->dev,
                        "ERROR: Incompatible device for evaluate context command.\n");
                ret = -ENODEV;
                break;
        case COMP_MEL_ERR:
                /* Max Exit Latency too large error */
                dev_warn(&udev->dev, "WARN: Max Exit Latency too large\n");
                ret = -EINVAL;
                break;
        case COMP_SUCCESS:
                xhci_dbg_trace(xhci, trace_xhci_dbg_context_change,
                                "Successful evaluate context command");
                ret = 0;
                break;
        default:
                xhci_err(xhci, "ERROR: unexpected command completion code 0x%x.\n",
                        *cmd_status);
                ret = -EINVAL;
                break;
        }
        return ret;
}

static u32 xhci_count_num_new_endpoints(struct xhci_hcd *xhci,
                struct xhci_input_control_ctx *ctrl_ctx)
{
        u32 valid_add_flags;
        u32 valid_drop_flags;

        /* Ignore the slot flag (bit 0), and the default control endpoint flag
         * (bit 1).  The default control endpoint is added during the Address
         * Device command and is never removed until the slot is disabled.
         */
        valid_add_flags = le32_to_cpu(ctrl_ctx->add_flags) >> 2;
        valid_drop_flags = le32_to_cpu(ctrl_ctx->drop_flags) >> 2;

        /* Use hweight32 to count the number of ones in the add flags, or
         * number of endpoints added.  Don't count endpoints that are changed
         * (both added and dropped).
         */
        return hweight32(valid_add_flags) -
                hweight32(valid_add_flags & valid_drop_flags);
}

static unsigned int xhci_count_num_dropped_endpoints(struct xhci_hcd *xhci,
                struct xhci_input_control_ctx *ctrl_ctx)
{
        u32 valid_add_flags;
        u32 valid_drop_flags;

        valid_add_flags = le32_to_cpu(ctrl_ctx->add_flags) >> 2;
        valid_drop_flags = le32_to_cpu(ctrl_ctx->drop_flags) >> 2;

        return hweight32(valid_drop_flags) -
                hweight32(valid_add_flags & valid_drop_flags);
}

/*
 * We need to reserve the new number of endpoints before the configure endpoint
 * command completes.  We can't subtract the dropped endpoints from the number
 * of active endpoints until the command completes because we can oversubscribe
 * the host in this case:
 *
 *  - the first configure endpoint command drops more endpoints than it adds
 *  - a second configure endpoint command that adds more endpoints is queued
 *  - the first configure endpoint command fails, so the config is unchanged
 *  - the second command may succeed, even though there isn't enough resources
 *
 * Must be called with xhci->lock held.
 */
static int xhci_reserve_host_resources(struct xhci_hcd *xhci,
                struct xhci_input_control_ctx *ctrl_ctx)
{
        u32 added_eps;

        added_eps = xhci_count_num_new_endpoints(xhci, ctrl_ctx);
        if (xhci->num_active_eps + added_eps > xhci->limit_active_eps) {
                xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                                "Not enough ep ctxs: "
                                "%u active, need to add %u, limit is %u.",

                                xhci->num_active_eps, added_eps,
                                xhci->limit_active_eps);
                return -ENOMEM;
        }
        xhci->num_active_eps += added_eps;
        xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                        "Adding %u ep ctxs, %u now active.", added_eps,
                        xhci->num_active_eps);
        return 0;
}

/*
 * The configure endpoint was failed by the xHC for some other reason, so we
 * need to revert the resources that failed configuration would have used.
 *
 * Must be called with xhci->lock held.
 */
static void xhci_free_host_resources(struct xhci_hcd *xhci,
                struct xhci_input_control_ctx *ctrl_ctx)
{
        u32 num_failed_eps;

        num_failed_eps = xhci_count_num_new_endpoints(xhci, ctrl_ctx);
        xhci->num_active_eps -= num_failed_eps;
        xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                        "Removing %u failed ep ctxs, %u now active.",
                        num_failed_eps,
                        xhci->num_active_eps);
}

/*
 * Now that the command has completed, clean up the active endpoint count by
 * subtracting out the endpoints that were dropped (but not changed).
 *
 * Must be called with xhci->lock held.
 */
static void xhci_finish_resource_reservation(struct xhci_hcd *xhci,
                struct xhci_input_control_ctx *ctrl_ctx)
{
        u32 num_dropped_eps;

        num_dropped_eps = xhci_count_num_dropped_endpoints(xhci, ctrl_ctx);
        xhci->num_active_eps -= num_dropped_eps;
        if (num_dropped_eps)
                xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                                "Removing %u dropped ep ctxs, %u now active.",
                                num_dropped_eps,
                                xhci->num_active_eps);
}

static unsigned int xhci_get_block_size(struct usb_device *udev)
{
        switch (udev->speed) {
        case USB_SPEED_LOW:
        case USB_SPEED_FULL:
                return FS_BLOCK;
        case USB_SPEED_HIGH:
                return HS_BLOCK;
        case USB_SPEED_SUPER:
                return SS_BLOCK;
        case USB_SPEED_UNKNOWN:
        case USB_SPEED_WIRELESS:
        default:
                /* Should never happen */
                return 1;
        }
}

static unsigned int
xhci_get_largest_overhead(struct xhci_interval_bw *interval_bw)
{
        if (interval_bw->overhead[LS_OVERHEAD_TYPE])
                return LS_OVERHEAD;
        if (interval_bw->overhead[FS_OVERHEAD_TYPE])
                return FS_OVERHEAD;
        return HS_OVERHEAD;
}

/* If we are changing a LS/FS device under a HS hub,
 * make sure (if we are activating a new TT) that the HS bus has enough
 * bandwidth for this new TT.
 */
static int xhci_check_tt_bw_table(struct xhci_hcd *xhci,
                struct xhci_virt_device *virt_dev,
                int old_active_eps)
{
        struct xhci_interval_bw_table *bw_table;
        struct xhci_tt_bw_info *tt_info;

        /* Find the bandwidth table for the root port this TT is attached to. */
        bw_table = &xhci->rh_bw[virt_dev->real_port - 1].bw_table;
        tt_info = virt_dev->tt_info;
        /* If this TT already had active endpoints, the bandwidth for this TT
         * has already been added.  Removing all periodic endpoints (and thus
         * making the TT enactive) will only decrease the bandwidth used.
         */
        if (old_active_eps)
                return 0;
        if (old_active_eps == 0 && tt_info->active_eps != 0) {
                if (bw_table->bw_used + TT_HS_OVERHEAD > HS_BW_LIMIT)
                        return -ENOMEM;
                return 0;
        }
        /* Not sure why we would have no new active endpoints...
         *
         * Maybe because of an Evaluate Context change for a hub update or a
         * control endpoint 0 max packet size change?
         * FIXME: skip the bandwidth calculation in that case.
         */
        return 0;
}

static int xhci_check_ss_bw(struct xhci_hcd *xhci,
                struct xhci_virt_device *virt_dev)
{
        unsigned int bw_reserved;

        bw_reserved = DIV_ROUND_UP(SS_BW_RESERVED*SS_BW_LIMIT_IN, 100);
        if (virt_dev->bw_table->ss_bw_in > (SS_BW_LIMIT_IN - bw_reserved))
                return -ENOMEM;

        bw_reserved = DIV_ROUND_UP(SS_BW_RESERVED*SS_BW_LIMIT_OUT, 100);
        if (virt_dev->bw_table->ss_bw_out > (SS_BW_LIMIT_OUT - bw_reserved))
                return -ENOMEM;

        return 0;
}

/*
 * This algorithm is a very conservative estimate of the worst-case scheduling
 * scenario for any one interval.  The hardware dynamically schedules the
 * packets, so we can't tell which microframe could be the limiting factor in
 * the bandwidth scheduling.  This only takes into account periodic endpoints.
 *
 * Obviously, we can't solve an NP complete problem to find the minimum worst
 * case scenario.  Instead, we come up with an estimate that is no less than
 * the worst case bandwidth used for any one microframe, but may be an
 * over-estimate.
 *
 * We walk the requirements for each endpoint by interval, starting with the
 * smallest interval, and place packets in the schedule where there is only one
 * possible way to schedule packets for that interval.  In order to simplify
 * this algorithm, we record the largest max packet size for each interval, and
 * assume all packets will be that size.
 *
 * For interval 0, we obviously must schedule all packets for each interval.
 * The bandwidth for interval 0 is just the amount of data to be transmitted
 * (the sum of all max ESIT payload sizes, plus any overhead per packet times
 * the number of packets).
 *
 * For interval 1, we have two possible microframes to schedule those packets
 * in.  For this algorithm, if we can schedule the same number of packets for
 * each possible scheduling opportunity (each microframe), we will do so.  The
 * remaining number of packets will be saved to be transmitted in the gaps in
 * the next interval's scheduling sequence.
 *
 * As we move those remaining packets to be scheduled with interval 2 packets,
 * we have to double the number of remaining packets to transmit.  This is
 * because the intervals are actually powers of 2, and we would be transmitting
 * the previous interval's packets twice in this interval.  We also have to be
 * sure that when we look at the largest max packet size for this interval, we
 * also look at the largest max packet size for the remaining packets and take
 * the greater of the two.
 *
 * The algorithm continues to evenly distribute packets in each scheduling
 * opportunity, and push the remaining packets out, until we get to the last
 * interval.  Then those packets and their associated overhead are just added
 * to the bandwidth used.
 */
static int xhci_check_bw_table(struct xhci_hcd *xhci,
                struct xhci_virt_device *virt_dev,
                int old_active_eps)
{
        unsigned int bw_reserved;
        unsigned int max_bandwidth;
        unsigned int bw_used;
        unsigned int block_size;
        struct xhci_interval_bw_table *bw_table;
        unsigned int packet_size = 0;
        unsigned int overhead = 0;
        unsigned int packets_transmitted = 0;
        unsigned int packets_remaining = 0;
        unsigned int i;

        if (virt_dev->udev->speed == USB_SPEED_SUPER)
                return xhci_check_ss_bw(xhci, virt_dev);

        if (virt_dev->udev->speed == USB_SPEED_HIGH) {
                max_bandwidth = HS_BW_LIMIT;
                /* Convert percent of bus BW reserved to blocks reserved */
                bw_reserved = DIV_ROUND_UP(HS_BW_RESERVED * max_bandwidth, 100);
        } else {
                max_bandwidth = FS_BW_LIMIT;
                bw_reserved = DIV_ROUND_UP(FS_BW_RESERVED * max_bandwidth, 100);
        }

        bw_table = virt_dev->bw_table;
        /* We need to translate the max packet size and max ESIT payloads into
         * the units the hardware uses.
         */
        block_size = xhci_get_block_size(virt_dev->udev);

        /* If we are manipulating a LS/FS device under a HS hub, double check
         * that the HS bus has enough bandwidth if we are activing a new TT.
         */
        if (virt_dev->tt_info) {
                xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                                "Recalculating BW for rootport %u",
                                virt_dev->real_port);
                if (xhci_check_tt_bw_table(xhci, virt_dev, old_active_eps)) {
                        xhci_warn(xhci, "Not enough bandwidth on HS bus for "
                                        "newly activated TT.\n");
                        return -ENOMEM;
                }
                xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                                "Recalculating BW for TT slot %u port %u",
                                virt_dev->tt_info->slot_id,
                                virt_dev->tt_info->ttport);
        } else {
                xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                                "Recalculating BW for rootport %u",
                                virt_dev->real_port);
        }

        /* Add in how much bandwidth will be used for interval zero, or the
         * rounded max ESIT payload + number of packets * largest overhead.
         */
        bw_used = DIV_ROUND_UP(bw_table->interval0_esit_payload, block_size) +
                bw_table->interval_bw[0].num_packets *
                xhci_get_largest_overhead(&bw_table->interval_bw[0]);

        for (i = 1; i < XHCI_MAX_INTERVAL; i++) {
                unsigned int bw_added;
                unsigned int largest_mps;
                unsigned int interval_overhead;

                /*
                 * How many packets could we transmit in this interval?
                 * If packets didn't fit in the previous interval, we will need
                 * to transmit that many packets twice within this interval.
                 */
                packets_remaining = 2 * packets_remaining +
                        bw_table->interval_bw[i].num_packets;

                /* Find the largest max packet size of this or the previous
                 * interval.
                 */
                if (list_empty(&bw_table->interval_bw[i].endpoints))
                        largest_mps = 0;
                else {
                        struct xhci_virt_ep *virt_ep;
                        struct list_head *ep_entry;

                        ep_entry = bw_table->interval_bw[i].endpoints.next;
                        virt_ep = list_entry(ep_entry,
                                        struct xhci_virt_ep, bw_endpoint_list);
                        /* Convert to blocks, rounding up */
                        largest_mps = DIV_ROUND_UP(
                                        virt_ep->bw_info.max_packet_size,
                                        block_size);
                }
                if (largest_mps > packet_size)
                        packet_size = largest_mps;

                /* Use the larger overhead of this or the previous interval. */
                interval_overhead = xhci_get_largest_overhead(
                                &bw_table->interval_bw[i]);
                if (interval_overhead > overhead)
                        overhead = interval_overhead;

                /* How many packets can we evenly distribute across
                 * (1 << (i + 1)) possible scheduling opportunities?
                 */
                packets_transmitted = packets_remaining >> (i + 1);

                /* Add in the bandwidth used for those scheduled packets */
                bw_added = packets_transmitted * (overhead + packet_size);

                /* How many packets do we have remaining to transmit? */
                packets_remaining = packets_remaining % (1 << (i + 1));

                /* What largest max packet size should those packets have? */
                /* If we've transmitted all packets, don't carry over the
                 * largest packet size.
                 */
                if (packets_remaining == 0) {
                        packet_size = 0;
                        overhead = 0;
                } else if (packets_transmitted > 0) {
                        /* Otherwise if we do have remaining packets, and we've
                         * scheduled some packets in this interval, take the
                         * largest max packet size from endpoints with this
                         * interval.
                         */
                        packet_size = largest_mps;
                        overhead = interval_overhead;
                }
                /* Otherwise carry over packet_size and overhead from the last
                 * time we had a remainder.
                 */
                bw_used += bw_added;
                if (bw_used > max_bandwidth) {
                        xhci_warn(xhci, "Not enough bandwidth. "
                                        "Proposed: %u, Max: %u\n",
                                bw_used, max_bandwidth);
                        return -ENOMEM;
                }
        }
        /*
         * Ok, we know we have some packets left over after even-handedly
         * scheduling interval 15.  We don't know which microframes they will
         * fit into, so we over-schedule and say they will be scheduled every
         * microframe.
         */
        if (packets_remaining > 0)
                bw_used += overhead + packet_size;

        if (!virt_dev->tt_info && virt_dev->udev->speed == USB_SPEED_HIGH) {
                unsigned int port_index = virt_dev->real_port - 1;

                /* OK, we're manipulating a HS device attached to a
                 * root port bandwidth domain.  Include the number of active TTs
                 * in the bandwidth used.
                 */
                bw_used += TT_HS_OVERHEAD *
                        xhci->rh_bw[port_index].num_active_tts;
        }

        xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                "Final bandwidth: %u, Limit: %u, Reserved: %u, "
                "Available: %u " "percent",
                bw_used, max_bandwidth, bw_reserved,
                (max_bandwidth - bw_used - bw_reserved) * 100 /
                max_bandwidth);

        bw_used += bw_reserved;
        if (bw_used > max_bandwidth) {
                xhci_warn(xhci, "Not enough bandwidth. Proposed: %u, Max: %u\n",
                                bw_used, max_bandwidth);
                return -ENOMEM;
        }

        bw_table->bw_used = bw_used;
        return 0;
}

static bool xhci_is_async_ep(unsigned int ep_type)
{
        return (ep_type != ISOC_OUT_EP && ep_type != INT_OUT_EP &&
                                        ep_type != ISOC_IN_EP &&
                                        ep_type != INT_IN_EP);
}

static bool xhci_is_sync_in_ep(unsigned int ep_type)
{
        return (ep_type == ISOC_IN_EP || ep_type == INT_IN_EP);
}

static unsigned int xhci_get_ss_bw_consumed(struct xhci_bw_info *ep_bw)
{
        unsigned int mps = DIV_ROUND_UP(ep_bw->max_packet_size, SS_BLOCK);

        if (ep_bw->ep_interval == 0)
                return SS_OVERHEAD_BURST +
                        (ep_bw->mult * ep_bw->num_packets *
                                        (SS_OVERHEAD + mps));
        return DIV_ROUND_UP(ep_bw->mult * ep_bw->num_packets *
                                (SS_OVERHEAD + mps + SS_OVERHEAD_BURST),
                                1 << ep_bw->ep_interval);

}

void xhci_drop_ep_from_interval_table(struct xhci_hcd *xhci,
                struct xhci_bw_info *ep_bw,
                struct xhci_interval_bw_table *bw_table,
                struct usb_device *udev,
                struct xhci_virt_ep *virt_ep,
                struct xhci_tt_bw_info *tt_info)
{
        struct xhci_interval_bw *interval_bw;
        int normalized_interval;

        if (xhci_is_async_ep(ep_bw->type))
                return;

        if (udev->speed == USB_SPEED_SUPER) {
                if (xhci_is_sync_in_ep(ep_bw->type))
                        xhci->devs[udev->slot_id]->bw_table->ss_bw_in -=
                                xhci_get_ss_bw_consumed(ep_bw);
                else
                        xhci->devs[udev->slot_id]->bw_table->ss_bw_out -=
                                xhci_get_ss_bw_consumed(ep_bw);
                return;
        }

        /* SuperSpeed endpoints never get added to intervals in the table, so
         * this check is only valid for HS/FS/LS devices.
         */
        if (list_empty(&virt_ep->bw_endpoint_list))
                return;
        /* For LS/FS devices, we need to translate the interval expressed in
         * microframes to frames.
         */
        if (udev->speed == USB_SPEED_HIGH)
                normalized_interval = ep_bw->ep_interval;
        else
                normalized_interval = ep_bw->ep_interval - 3;

        if (normalized_interval == 0)
                bw_table->interval0_esit_payload -= ep_bw->max_esit_payload;
        interval_bw = &bw_table->interval_bw[normalized_interval];
        interval_bw->num_packets -= ep_bw->num_packets;
        switch (udev->speed) {
        case USB_SPEED_LOW:
                interval_bw->overhead[LS_OVERHEAD_TYPE] -= 1;
                break;
        case USB_SPEED_FULL:
                interval_bw->overhead[FS_OVERHEAD_TYPE] -= 1;
                break;
        case USB_SPEED_HIGH:
                interval_bw->overhead[HS_OVERHEAD_TYPE] -= 1;
                break;
        case USB_SPEED_SUPER:
        case USB_SPEED_UNKNOWN:
        case USB_SPEED_WIRELESS:
                /* Should never happen because only LS/FS/HS endpoints will get
                 * added to the endpoint list.
                 */
                return;
        }
        if (tt_info)
                tt_info->active_eps -= 1;
        list_del_init(&virt_ep->bw_endpoint_list);
}

static void xhci_add_ep_to_interval_table(struct xhci_hcd *xhci,
                struct xhci_bw_info *ep_bw,
                struct xhci_interval_bw_table *bw_table,
                struct usb_device *udev,
                struct xhci_virt_ep *virt_ep,
                struct xhci_tt_bw_info *tt_info)
{
        struct xhci_interval_bw *interval_bw;
        struct xhci_virt_ep *smaller_ep;
        int normalized_interval;

        if (xhci_is_async_ep(ep_bw->type))
                return;

        if (udev->speed == USB_SPEED_SUPER) {
                if (xhci_is_sync_in_ep(ep_bw->type))
                        xhci->devs[udev->slot_id]->bw_table->ss_bw_in +=
                                xhci_get_ss_bw_consumed(ep_bw);
                else
                        xhci->devs[udev->slot_id]->bw_table->ss_bw_out +=
                                xhci_get_ss_bw_consumed(ep_bw);
                return;
        }

        /* For LS/FS devices, we need to translate the interval expressed in
         * microframes to frames.
         */
        if (udev->speed == USB_SPEED_HIGH)
                normalized_interval = ep_bw->ep_interval;
        else
                normalized_interval = ep_bw->ep_interval - 3;

        if (normalized_interval == 0)
                bw_table->interval0_esit_payload += ep_bw->max_esit_payload;
        interval_bw = &bw_table->interval_bw[normalized_interval];
        interval_bw->num_packets += ep_bw->num_packets;
        switch (udev->speed) {
        case USB_SPEED_LOW:
                interval_bw->overhead[LS_OVERHEAD_TYPE] += 1;
                break;
        case USB_SPEED_FULL:
                interval_bw->overhead[FS_OVERHEAD_TYPE] += 1;
                break;
        case USB_SPEED_HIGH:
                interval_bw->overhead[HS_OVERHEAD_TYPE] += 1;
                break;
        case USB_SPEED_SUPER:
        case USB_SPEED_UNKNOWN:
        case USB_SPEED_WIRELESS:
                /* Should never happen because only LS/FS/HS endpoints will get
                 * added to the endpoint list.
                 */
                return;
        }

        if (tt_info)
                tt_info->active_eps += 1;
        /* Insert the endpoint into the list, largest max packet size first. */
        list_for_each_entry(smaller_ep, &interval_bw->endpoints,
                        bw_endpoint_list) {
                if (ep_bw->max_packet_size >=
                                smaller_ep->bw_info.max_packet_size) {
                        /* Add the new ep before the smaller endpoint */
                        list_add_tail(&virt_ep->bw_endpoint_list,
                                        &smaller_ep->bw_endpoint_list);
                        return;
                }
        }
        /* Add the new endpoint at the end of the list. */
        list_add_tail(&virt_ep->bw_endpoint_list,
                        &interval_bw->endpoints);
}

void xhci_update_tt_active_eps(struct xhci_hcd *xhci,
                struct xhci_virt_device *virt_dev,
                int old_active_eps)
{
        struct xhci_root_port_bw_info *rh_bw_info;
        if (!virt_dev->tt_info)
                return;

        rh_bw_info = &xhci->rh_bw[virt_dev->real_port - 1];
        if (old_active_eps == 0 &&
                                virt_dev->tt_info->active_eps != 0) {
                rh_bw_info->num_active_tts += 1;
                rh_bw_info->bw_table.bw_used += TT_HS_OVERHEAD;
        } else if (old_active_eps != 0 &&
                                virt_dev->tt_info->active_eps == 0) {
                rh_bw_info->num_active_tts -= 1;
                rh_bw_info->bw_table.bw_used -= TT_HS_OVERHEAD;
        }
}

static int xhci_reserve_bandwidth(struct xhci_hcd *xhci,
                struct xhci_virt_device *virt_dev,
                struct xhci_container_ctx *in_ctx)
{
        struct xhci_bw_info ep_bw_info[31];
        int i;
        struct xhci_input_control_ctx *ctrl_ctx;
        int old_active_eps = 0;

        if (virt_dev->tt_info)
                old_active_eps = virt_dev->tt_info->active_eps;

        ctrl_ctx = xhci_get_input_control_ctx(in_ctx);
        if (!ctrl_ctx) {
                xhci_warn(xhci, "%s: Could not get input context, bad type.\n",
                                __func__);
                return -ENOMEM;
        }

        for (i = 0; i < 31; i++) {
                if (!EP_IS_ADDED(ctrl_ctx, i) && !EP_IS_DROPPED(ctrl_ctx, i))
                        continue;

                /* Make a copy of the BW info in case we need to revert this */
                memcpy(&ep_bw_info[i], &virt_dev->eps[i].bw_info,
                                sizeof(ep_bw_info[i]));
                /* Drop the endpoint from the interval table if the endpoint is
                 * being dropped or changed.
                 */
                if (EP_IS_DROPPED(ctrl_ctx, i))
                        xhci_drop_ep_from_interval_table(xhci,
                                        &virt_dev->eps[i].bw_info,
                                        virt_dev->bw_table,
                                        virt_dev->udev,
                                        &virt_dev->eps[i],
                                        virt_dev->tt_info);
        }
        /* Overwrite the information stored in the endpoints' bw_info */
        xhci_update_bw_info(xhci, virt_dev->in_ctx, ctrl_ctx, virt_dev);
        for (i = 0; i < 31; i++) {
                /* Add any changed or added endpoints to the interval table */
                if (EP_IS_ADDED(ctrl_ctx, i))
                        xhci_add_ep_to_interval_table(xhci,
                                        &virt_dev->eps[i].bw_info,
                                        virt_dev->bw_table,
                                        virt_dev->udev,
                                        &virt_dev->eps[i],
                                        virt_dev->tt_info);
        }

        if (!xhci_check_bw_table(xhci, virt_dev, old_active_eps)) {
                /* Ok, this fits in the bandwidth we have.
                 * Update the number of active TTs.
                 */
                xhci_update_tt_active_eps(xhci, virt_dev, old_active_eps);
                return 0;
        }

        /* We don't have enough bandwidth for this, revert the stored info. */
        for (i = 0; i < 31; i++) {
                if (!EP_IS_ADDED(ctrl_ctx, i) && !EP_IS_DROPPED(ctrl_ctx, i))
                        continue;

                /* Drop the new copies of any added or changed endpoints from
                 * the interval table.
                 */
                if (EP_IS_ADDED(ctrl_ctx, i)) {
                        xhci_drop_ep_from_interval_table(xhci,
                                        &virt_dev->eps[i].bw_info,
                                        virt_dev->bw_table,
                                        virt_dev->udev,
                                        &virt_dev->eps[i],
                                        virt_dev->tt_info);
                }
                /* Revert the endpoint back to its old information */
                memcpy(&virt_dev->eps[i].bw_info, &ep_bw_info[i],
                                sizeof(ep_bw_info[i]));
                /* Add any changed or dropped endpoints back into the table */
                if (EP_IS_DROPPED(ctrl_ctx, i))
                        xhci_add_ep_to_interval_table(xhci,
                                        &virt_dev->eps[i].bw_info,
                                        virt_dev->bw_table,
                                        virt_dev->udev,
                                        &virt_dev->eps[i],
                                        virt_dev->tt_info);
        }
        return -ENOMEM;
}


/* Issue a configure endpoint command or evaluate context command
 * and wait for it to finish.
 */
static int xhci_configure_endpoint(struct xhci_hcd *xhci,
                struct usb_device *udev,
                struct xhci_command *command,
                bool ctx_change, bool must_succeed)
{
        int ret;
        unsigned long flags;
        struct xhci_input_control_ctx *ctrl_ctx;
        struct xhci_virt_device *virt_dev;

        if (!command)
                return -EINVAL;

        spin_lock_irqsave(&xhci->lock, flags);
        virt_dev = xhci->devs[udev->slot_id];

        ctrl_ctx = xhci_get_input_control_ctx(command->in_ctx);
        if (!ctrl_ctx) {
                spin_unlock_irqrestore(&xhci->lock, flags);
                xhci_warn(xhci, "%s: Could not get input context, bad type.\n",
                                __func__);
                return -ENOMEM;
        }

        if ((xhci->quirks & XHCI_EP_LIMIT_QUIRK) &&
                        xhci_reserve_host_resources(xhci, ctrl_ctx)) {
                spin_unlock_irqrestore(&xhci->lock, flags);
                xhci_warn(xhci, "Not enough host resources, "
                                "active endpoint contexts = %u\n",
                                xhci->num_active_eps);
                return -ENOMEM;
        }
        if ((xhci->quirks & XHCI_SW_BW_CHECKING) &&
            xhci_reserve_bandwidth(xhci, virt_dev, command->in_ctx)) {
                if ((xhci->quirks & XHCI_EP_LIMIT_QUIRK))
                        xhci_free_host_resources(xhci, ctrl_ctx);
                spin_unlock_irqrestore(&xhci->lock, flags);
                xhci_warn(xhci, "Not enough bandwidth\n");
                return -ENOMEM;
        }

        if (!ctx_change)
                ret = xhci_queue_configure_endpoint(xhci, command,
                                command->in_ctx->dma,
                                udev->slot_id, must_succeed);
        else
                ret = xhci_queue_evaluate_context(xhci, command,
                                command->in_ctx->dma,
                                udev->slot_id, must_succeed);
        if (ret < 0) {
                if ((xhci->quirks & XHCI_EP_LIMIT_QUIRK))
                        xhci_free_host_resources(xhci, ctrl_ctx);
                spin_unlock_irqrestore(&xhci->lock, flags);
                xhci_dbg_trace(xhci,  trace_xhci_dbg_context_change,
                                "FIXME allocate a new ring segment");
                return -ENOMEM;
        }
        xhci_ring_cmd_db(xhci);
        spin_unlock_irqrestore(&xhci->lock, flags);

        /* Wait for the configure endpoint command to complete */
        wait_for_completion(command->completion);

        if (!ctx_change)
                ret = xhci_configure_endpoint_result(xhci, udev,
                                                     &command->status);
        else
                ret = xhci_evaluate_context_result(xhci, udev,
                                                   &command->status);

        if ((xhci->quirks & XHCI_EP_LIMIT_QUIRK)) {
                spin_lock_irqsave(&xhci->lock, flags);
                /* If the command failed, remove the reserved resources.
                 * Otherwise, clean up the estimate to include dropped eps.
                 */
                if (ret)
                        xhci_free_host_resources(xhci, ctrl_ctx);
                else
                        xhci_finish_resource_reservation(xhci, ctrl_ctx);
                spin_unlock_irqrestore(&xhci->lock, flags);
        }
        return ret;
}

static void xhci_check_bw_drop_ep_streams(struct xhci_hcd *xhci,
        struct xhci_virt_device *vdev, int i)
{
        struct xhci_virt_ep *ep = &vdev->eps[i];

        if (ep->ep_state & EP_HAS_STREAMS) {
                xhci_warn(xhci, "WARN: endpoint 0x%02x has streams on set_interface, freeing streams.\n",
                                xhci_get_endpoint_address(i));
                xhci_free_stream_info(xhci, ep->stream_info);
                ep->stream_info = NULL;
                ep->ep_state &= ~EP_HAS_STREAMS;
        }
}

/* Called after one or more calls to xhci_add_endpoint() or
 * xhci_drop_endpoint().  If this call fails, the USB core is expected
 * to call xhci_reset_bandwidth().
 *
 * Since we are in the middle of changing either configuration or
 * installing a new alt setting, the USB core won't allow URBs to be
 * enqueued for any endpoint on the old config or interface.  Nothing
 * else should be touching the xhci->devs[slot_id] structure, so we
 * don't need to take the xhci->lock for manipulating that.
 */
int xhci_check_bandwidth(struct usb_hcd *hcd, struct usb_device *udev)
{
        int i;
        int ret = 0;
        struct xhci_hcd *xhci;
        struct xhci_virt_device *virt_dev;
        struct xhci_input_control_ctx *ctrl_ctx;
        struct xhci_slot_ctx *slot_ctx;
        struct xhci_command *command;

        ret = xhci_check_args(hcd, udev, NULL, 0, true, __func__);
        if (ret <= 0)
                return ret;
        xhci = hcd_to_xhci(hcd);
        if (xhci->xhc_state & XHCI_STATE_DYING)
                return -ENODEV;

        xhci_dbg(xhci, "%s called for udev %p\n", __func__, udev);
        virt_dev = xhci->devs[udev->slot_id];

        command = xhci_alloc_command(xhci, false, true, GFP_KERNEL);
        if (!command)
                return -ENOMEM;

        command->in_ctx = virt_dev->in_ctx;

        /* See section 4.6.6 - A0 = 1; A1 = D0 = D1 = 0 */
        ctrl_ctx = xhci_get_input_control_ctx(command->in_ctx);
        if (!ctrl_ctx) {
                xhci_warn(xhci, "%s: Could not get input context, bad type.\n",
                                __func__);
                ret = -ENOMEM;
                goto command_cleanup;
        }
        ctrl_ctx->add_flags |= cpu_to_le32(SLOT_FLAG);
        ctrl_ctx->add_flags &= cpu_to_le32(~EP0_FLAG);
        ctrl_ctx->drop_flags &= cpu_to_le32(~(SLOT_FLAG | EP0_FLAG));

        /* Don't issue the command if there's no endpoints to update. */
        if (ctrl_ctx->add_flags == cpu_to_le32(SLOT_FLAG) &&
            ctrl_ctx->drop_flags == 0) {
                ret = 0;
                goto command_cleanup;
        }
        /* Fix up Context Entries field. Minimum value is EP0 == BIT(1). */
        slot_ctx = xhci_get_slot_ctx(xhci, virt_dev->in_ctx);
        for (i = 31; i >= 1; i--) {
                __le32 le32 = cpu_to_le32(BIT(i));

                if ((virt_dev->eps[i-1].ring && !(ctrl_ctx->drop_flags & le32))
                    || (ctrl_ctx->add_flags & le32) || i == 1) {
                        slot_ctx->dev_info &= cpu_to_le32(~LAST_CTX_MASK);
                        slot_ctx->dev_info |= cpu_to_le32(LAST_CTX(i));
                        break;
                }
        }
        xhci_dbg(xhci, "New Input Control Context:\n");
        xhci_dbg_ctx(xhci, virt_dev->in_ctx,
                     LAST_CTX_TO_EP_NUM(le32_to_cpu(slot_ctx->dev_info)));

        ret = xhci_configure_endpoint(xhci, udev, command,
                        false, false);
        if (ret)
                /* Callee should call reset_bandwidth() */
                goto command_cleanup;

        xhci_dbg(xhci, "Output context after successful config ep cmd:\n");
        xhci_dbg_ctx(xhci, virt_dev->out_ctx,
                     LAST_CTX_TO_EP_NUM(le32_to_cpu(slot_ctx->dev_info)));

        /* Free any rings that were dropped, but not changed. */
        for (i = 1; i < 31; ++i) {
                if ((le32_to_cpu(ctrl_ctx->drop_flags) & (1 << (i + 1))) &&
                    !(le32_to_cpu(ctrl_ctx->add_flags) & (1 << (i + 1)))) {
                        xhci_free_or_cache_endpoint_ring(xhci, virt_dev, i);
                        xhci_check_bw_drop_ep_streams(xhci, virt_dev, i);
                }
        }
        xhci_zero_in_ctx(xhci, virt_dev);
        /*
         * Install any rings for completely new endpoints or changed endpoints,
         * and free or cache any old rings from changed endpoints.
         */
        for (i = 1; i < 31; ++i) {
                if (!virt_dev->eps[i].new_ring)
                        continue;
                /* Only cache or free the old ring if it exists.
                 * It may not if this is the first add of an endpoint.
                 */
                if (virt_dev->eps[i].ring) {
                        xhci_free_or_cache_endpoint_ring(xhci, virt_dev, i);
                }
                xhci_check_bw_drop_ep_streams(xhci, virt_dev, i);
                virt_dev->eps[i].ring = virt_dev->eps[i].new_ring;
                virt_dev->eps[i].new_ring = NULL;
        }
command_cleanup:
        kfree(command->completion);
        kfree(command);

        return ret;
}

void xhci_reset_bandwidth(struct usb_hcd *hcd, struct usb_device *udev)
{
        struct xhci_hcd *xhci;
        struct xhci_virt_device *virt_dev;
        int i, ret;

        ret = xhci_check_args(hcd, udev, NULL, 0, true, __func__);
        if (ret <= 0)
                return;
        xhci = hcd_to_xhci(hcd);

        xhci_dbg(xhci, "%s called for udev %p\n", __func__, udev);
        virt_dev = xhci->devs[udev->slot_id];
        /* Free any rings allocated for added endpoints */
        for (i = 0; i < 31; ++i) {
                if (virt_dev->eps[i].new_ring) {
                        xhci_ring_free(xhci, virt_dev->eps[i].new_ring);
                        virt_dev->eps[i].new_ring = NULL;
                }
        }
        xhci_zero_in_ctx(xhci, virt_dev);
}

static void xhci_setup_input_ctx_for_config_ep(struct xhci_hcd *xhci,
                struct xhci_container_ctx *in_ctx,
                struct xhci_container_ctx *out_ctx,
                struct xhci_input_control_ctx *ctrl_ctx,
                u32 add_flags, u32 drop_flags)
{
        ctrl_ctx->add_flags = cpu_to_le32(add_flags);
        ctrl_ctx->drop_flags = cpu_to_le32(drop_flags);
        xhci_slot_copy(xhci, in_ctx, out_ctx);
        ctrl_ctx->add_flags |= cpu_to_le32(SLOT_FLAG);

        xhci_dbg(xhci, "Input Context:\n");
        xhci_dbg_ctx(xhci, in_ctx, xhci_last_valid_endpoint(add_flags));
}

static void xhci_setup_input_ctx_for_quirk(struct xhci_hcd *xhci,
                unsigned int slot_id, unsigned int ep_index,
                struct xhci_dequeue_state *deq_state)
{
        struct xhci_input_control_ctx *ctrl_ctx;
        struct xhci_container_ctx *in_ctx;
        struct xhci_ep_ctx *ep_ctx;
        u32 added_ctxs;
        dma_addr_t addr;

        in_ctx = xhci->devs[slot_id]->in_ctx;
        ctrl_ctx = xhci_get_input_control_ctx(in_ctx);
        if (!ctrl_ctx) {
                xhci_warn(xhci, "%s: Could not get input context, bad type.\n",
                                __func__);
                return;
        }

        xhci_endpoint_copy(xhci, xhci->devs[slot_id]->in_ctx,
                        xhci->devs[slot_id]->out_ctx, ep_index);
        ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, ep_index);
        addr = xhci_trb_virt_to_dma(deq_state->new_deq_seg,
                        deq_state->new_deq_ptr);
        if (addr == 0) {
                xhci_warn(xhci, "WARN Cannot submit config ep after "
                                "reset ep command\n");
                xhci_warn(xhci, "WARN deq seg = %p, deq ptr = %p\n",
                                deq_state->new_deq_seg,
                                deq_state->new_deq_ptr);
                return;
        }
        ep_ctx->deq = cpu_to_le64(addr | deq_state->new_cycle_state);

        added_ctxs = xhci_get_endpoint_flag_from_index(ep_index);
        xhci_setup_input_ctx_for_config_ep(xhci, xhci->devs[slot_id]->in_ctx,
                        xhci->devs[slot_id]->out_ctx, ctrl_ctx,
                        added_ctxs, added_ctxs);
}

void xhci_cleanup_stalled_ring(struct xhci_hcd *xhci,
                        unsigned int ep_index, struct xhci_td *td)
{
        struct xhci_dequeue_state deq_state;
        struct xhci_virt_ep *ep;
        struct usb_device *udev = td->urb->dev;

        xhci_dbg_trace(xhci, trace_xhci_dbg_reset_ep,
                        "Cleaning up stalled endpoint ring");
        ep = &xhci->devs[udev->slot_id]->eps[ep_index];
        /* We need to move the HW's dequeue pointer past this TD,
         * or it will attempt to resend it on the next doorbell ring.
         */
        xhci_find_new_dequeue_state(xhci, udev->slot_id,
                        ep_index, ep->stopped_stream, td, &deq_state);

        if (!deq_state.new_deq_ptr || !deq_state.new_deq_seg)
                return;

        /* HW with the reset endpoint quirk will use the saved dequeue state to
         * issue a configure endpoint command later.
         */
        if (!(xhci->quirks & XHCI_RESET_EP_QUIRK)) {
                xhci_dbg_trace(xhci, trace_xhci_dbg_reset_ep,
                                "Queueing new dequeue state");
                xhci_queue_new_dequeue_state(xhci, udev->slot_id,
                                ep_index, ep->stopped_stream, &deq_state);
        } else {
                /* Better hope no one uses the input context between now and the
                 * reset endpoint completion!
                 * XXX: No idea how this hardware will react when stream rings
                 * are enabled.
                 */
                xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
                                "Setting up input context for "
                                "configure endpoint command");
                xhci_setup_input_ctx_for_quirk(xhci, udev->slot_id,
                                ep_index, &deq_state);
        }
}

/* Called when clearing halted device. The core should have sent the control
 * message to clear the device halt condition. The host side of the halt should
 * already be cleared with a reset endpoint command issued when the STALL tx
 * event was received.
 *
 * Context: in_interrupt
 */

void xhci_endpoint_reset(struct usb_hcd *hcd,
                struct usb_host_endpoint *ep)
{
        struct xhci_hcd *xhci;

        xhci = hcd_to_xhci(hcd);

        /*
         * We might need to implement the config ep cmd in xhci 4.8.1 note:
         * The Reset Endpoint Command may only be issued to endpoints in the
         * Halted state. If software wishes reset the Data Toggle or Sequence
         * Number of an endpoint that isn't in the Halted state, then software
         * may issue a Configure Endpoint Command with the Drop and Add bits set
         * for the target endpoint. that is in the Stopped state.
         */

        /* For now just print debug to follow the situation */
        xhci_dbg(xhci, "Endpoint 0x%x ep reset callback called\n",
                 ep->desc.bEndpointAddress);
}

static int xhci_check_streams_endpoint(struct xhci_hcd *xhci,
                struct usb_device *udev, struct usb_host_endpoint *ep,
                unsigned int slot_id)
{
        int ret;
        unsigned int ep_index;
        unsigned int ep_state;

        if (!ep)
                return -EINVAL;
        ret = xhci_check_args(xhci_to_hcd(xhci), udev, ep, 1, true, __func__);
        if (ret <= 0)
                return -EINVAL;
        if (usb_ss_max_streams(&ep->ss_ep_comp) == 0) {
                xhci_warn(xhci, "WARN: SuperSpeed Endpoint Companion"
                                " descriptor for ep 0x%x does not support streams\n",
                                ep->desc.bEndpointAddress);
                return -EINVAL;
        }

        ep_index = xhci_get_endpoint_index(&ep->desc);