]> git.kernelconcepts.de Git - karo-tx-uboot.git/blobdiff - drivers/usb/host/ehci-hcd.c
Merge 'u-boot-atmel/master' into 'u-boot-arm/master'
[karo-tx-uboot.git] / drivers / usb / host / ehci-hcd.c
index c7fba104899fe336a160b68f2a837c9cb35b11f0..7f98a6354ac4227515cb412f8ed5abde3ed63cdb 100644 (file)
  */
 #include <common.h>
 #include <asm/byteorder.h>
+#include <asm/unaligned.h>
 #include <usb.h>
 #include <asm/io.h>
 #include <malloc.h>
+#include <watchdog.h>
 
 #include "ehci.h"
 
-int rootdev;
-struct ehci_hccr *hccr;        /* R/O registers, not need for volatile */
-volatile struct ehci_hcor *hcor;
+#ifndef CONFIG_USB_MAX_CONTROLLER_COUNT
+#define CONFIG_USB_MAX_CONTROLLER_COUNT 1
+#endif
+
+static struct ehci_ctrl {
+       struct ehci_hccr *hccr; /* R/O registers, not need for volatile */
+       struct ehci_hcor *hcor;
+       int rootdev;
+       uint16_t portreset;
+       struct QH qh_list __attribute__((aligned(USB_DMA_MINALIGN)));
+} ehcic[CONFIG_USB_MAX_CONTROLLER_COUNT];
 
-static uint16_t portreset;
-static struct QH qh_list __attribute__((aligned(32)));
+#define ALIGN_END_ADDR(type, ptr, size)                        \
+       ((uint32_t)(ptr) + roundup((size) * sizeof(type), USB_DMA_MINALIGN))
 
 static struct descriptor {
        struct usb_hub_descriptor hub;
@@ -47,7 +57,7 @@ static struct descriptor {
                0x29,           /* bDescriptorType: hub descriptor */
                2,              /* bNrPorts -- runtime modified */
                0,              /* wHubCharacteristics */
-               0xff,           /* bPwrOn2PwrGood */
+               10,             /* bPwrOn2PwrGood */
                0,              /* bHubCntrCurrent */
                {},             /* Device removable */
                {}              /* at most 7 ports! XXX */
@@ -107,98 +117,13 @@ static struct descriptor {
 #define ehci_is_TDI()  (0)
 #endif
 
-#if defined(CONFIG_EHCI_DCACHE)
-/*
- * Routines to handle (flush/invalidate) the dcache for the QH and qTD
- * structures and data buffers. This is needed on platforms using this
- * EHCI support with dcache enabled.
- */
-static void flush_invalidate(u32 addr, int size, int flush)
-{
-       if (flush)
-               flush_dcache_range(addr, addr + size);
-       else
-               invalidate_dcache_range(addr, addr + size);
-}
-
-static void cache_qtd(struct qTD *qtd, int flush)
-{
-       u32 *ptr = (u32 *)qtd->qt_buffer[0];
-       int len = (qtd->qt_token & 0x7fff0000) >> 16;
-
-       flush_invalidate((u32)qtd, sizeof(struct qTD), flush);
-       if (ptr && len)
-               flush_invalidate((u32)ptr, len, flush);
-}
-
-
-static inline struct QH *qh_addr(struct QH *qh)
-{
-       return (struct QH *)((u32)qh & 0xffffffe0);
-}
-
-static void cache_qh(struct QH *qh, int flush)
-{
-       struct qTD *qtd;
-       struct qTD *next;
-       static struct qTD *first_qtd;
-
-       /*
-        * Walk the QH list and flush/invalidate all entries
-        */
-       while (1) {
-               flush_invalidate((u32)qh_addr(qh), sizeof(struct QH), flush);
-               if ((u32)qh & QH_LINK_TYPE_QH)
-                       break;
-               qh = qh_addr(qh);
-               qh = (struct QH *)qh->qh_link;
-       }
-       qh = qh_addr(qh);
-
-       /*
-        * Save first qTD pointer, needed for invalidating pass on this QH
-        */
-       if (flush)
-               first_qtd = qtd = (struct qTD *)(*(u32 *)&qh->qh_overlay &
-                                                0xffffffe0);
-       else
-               qtd = first_qtd;
-
-       /*
-        * Walk the qTD list and flush/invalidate all entries
-        */
-       while (1) {
-               if (qtd == NULL)
-                       break;
-               cache_qtd(qtd, flush);
-               next = (struct qTD *)((u32)qtd->qt_next & 0xffffffe0);
-               if (next == qtd)
-                       break;
-               qtd = next;
-       }
-}
-
-static inline void ehci_flush_dcache(struct QH *qh)
-{
-       cache_qh(qh, 1);
-}
-
-static inline void ehci_invalidate_dcache(struct QH *qh)
-{
-       cache_qh(qh, 0);
-}
-#else /* CONFIG_EHCI_DCACHE */
-/*
- *
- */
-static inline void ehci_flush_dcache(struct QH *qh)
+void __ehci_powerup_fixup(uint32_t *status_reg, uint32_t *reg)
 {
+       mdelay(50);
 }
 
-static inline void ehci_invalidate_dcache(struct QH *qh)
-{
-}
-#endif /* CONFIG_EHCI_DCACHE */
+void ehci_powerup_fixup(uint32_t *status_reg, uint32_t *reg)
+       __attribute__((weak, alias("__ehci_powerup_fixup")));
 
 static int handshake(uint32_t *ptr, uint32_t mask, uint32_t done, int usec)
 {
@@ -216,29 +141,25 @@ static int handshake(uint32_t *ptr, uint32_t mask, uint32_t done, int usec)
        return -1;
 }
 
-static void ehci_free(void *p, size_t sz)
-{
-
-}
-
-static int ehci_reset(void)
+static int ehci_reset(int index)
 {
        uint32_t cmd;
        uint32_t tmp;
        uint32_t *reg_ptr;
        int ret = 0;
 
-       cmd = ehci_readl(&hcor->or_usbcmd);
-       cmd |= CMD_RESET;
-       ehci_writel(&hcor->or_usbcmd, cmd);
-       ret = handshake((uint32_t *)&hcor->or_usbcmd, CMD_RESET, 0, 250 * 1000);
+       cmd = ehci_readl(&ehcic[index].hcor->or_usbcmd);
+       cmd = (cmd & ~CMD_RUN) | CMD_RESET;
+       ehci_writel(&ehcic[index].hcor->or_usbcmd, cmd);
+       ret = handshake((uint32_t *)&ehcic[index].hcor->or_usbcmd,
+                       CMD_RESET, 0, 250 * 1000);
        if (ret < 0) {
                printf("EHCI fail to reset\n");
                goto out;
        }
 
        if (ehci_is_TDI()) {
-               reg_ptr = (uint32_t *)((u8 *)hcor + USBMODE);
+               reg_ptr = (uint32_t *)((u8 *)ehcic[index].hcor + USBMODE);
                tmp = ehci_readl(reg_ptr);
                tmp |= USBMODE_CM_HC;
 #if defined(CONFIG_EHCI_MMIO_BIG_ENDIAN)
@@ -246,50 +167,33 @@ static int ehci_reset(void)
 #endif
                ehci_writel(reg_ptr, tmp);
        }
+
+#ifdef CONFIG_USB_EHCI_TXFIFO_THRESH
+       cmd = ehci_readl(&ehcic[index].hcor->or_txfilltuning);
+       cmd &= ~TXFIFO_THRESH_MASK;
+       cmd |= TXFIFO_THRESH(CONFIG_USB_EHCI_TXFIFO_THRESH);
+       ehci_writel(&ehcic[index].hcor->or_txfilltuning, cmd);
+#endif
 out:
        return ret;
 }
 
-static void *ehci_alloc(size_t sz, size_t align)
-{
-       static struct QH qh __attribute__((aligned(32)));
-       static struct qTD td[3] __attribute__((aligned (32)));
-       static int ntds;
-       void *p;
-
-       switch (sz) {
-       case sizeof(struct QH):
-               p = &qh;
-               ntds = 0;
-               break;
-       case sizeof(struct qTD):
-               if (ntds == 3) {
-                       debug("out of TDs\n");
-                       return NULL;
-               }
-               p = &td[ntds];
-               ntds++;
-               break;
-       default:
-               debug("unknown allocation size\n");
-               return NULL;
-       }
-
-       memset(p, 0, sz);
-       return p;
-}
-
 static int ehci_td_buffer(struct qTD *td, void *buf, size_t sz)
 {
-       uint32_t addr, delta, next;
+       uint32_t delta, next;
+       uint32_t addr = (uint32_t)buf;
        int idx;
 
-       addr = (uint32_t) buf;
+       if (addr != ALIGN(addr, ARCH_DMA_MINALIGN))
+               debug("EHCI-HCD: Misaligned buffer address (%p)\n", buf);
+
+       flush_dcache_range(addr, ALIGN(addr + sz, ARCH_DMA_MINALIGN));
+
        idx = 0;
-       while (idx < 5) {
+       while (idx < QT_BUFFER_CNT) {
                td->qt_buffer[idx] = cpu_to_hc32(addr);
                td->qt_buffer_hi[idx] = 0;
-               next = (addr + 4096) & ~4095;
+               next = (addr + EHCI_PAGE_SIZE) & ~(EHCI_PAGE_SIZE - 1);
                delta = next - addr;
                if (delta >= sz)
                        break;
@@ -298,27 +202,43 @@ static int ehci_td_buffer(struct qTD *td, void *buf, size_t sz)
                idx++;
        }
 
-       if (idx == 5) {
-               debug("out of buffer pointers (%u bytes left)\n", sz);
+       if (idx == QT_BUFFER_CNT) {
+               printf("out of buffer pointers (%u bytes left)\n", sz);
                return -1;
        }
 
        return 0;
 }
 
+static inline u8 ehci_encode_speed(enum usb_device_speed speed)
+{
+       #define QH_HIGH_SPEED   2
+       #define QH_FULL_SPEED   0
+       #define QH_LOW_SPEED    1
+       if (speed == USB_SPEED_HIGH)
+               return QH_HIGH_SPEED;
+       if (speed == USB_SPEED_LOW)
+               return QH_LOW_SPEED;
+       return QH_FULL_SPEED;
+}
+
 static int
 ehci_submit_async(struct usb_device *dev, unsigned long pipe, void *buffer,
                   int length, struct devrequest *req)
 {
-       struct QH *qh;
-       struct qTD *td;
+       ALLOC_ALIGN_BUFFER(struct QH, qh, 1, USB_DMA_MINALIGN);
+       struct qTD *qtd;
+       int qtd_count = 0;
+       int qtd_counter = 0;
        volatile struct qTD *vtd;
        unsigned long ts;
        uint32_t *tdp;
-       uint32_t endpt, token, usbsts;
+       uint32_t endpt, maxpacket, token, usbsts;
        uint32_t c, toggle;
        uint32_t cmd;
+       int timeout;
        int ret = 0;
+       struct ehci_ctrl *ctrl = dev->controller;
 
        debug("dev=%p, pipe=%lx, buffer=%p, length=%d, req=%p\n", dev, pipe,
              buffer, length, req);
@@ -329,191 +249,336 @@ ehci_submit_async(struct usb_device *dev, unsigned long pipe, void *buffer,
                      le16_to_cpu(req->value), le16_to_cpu(req->value),
                      le16_to_cpu(req->index));
 
-       qh = ehci_alloc(sizeof(struct QH), 32);
-       if (qh == NULL) {
-               debug("unable to allocate QH\n");
+#define PKT_ALIGN      512
+       /*
+        * The USB transfer is split into qTD transfers. Eeach qTD transfer is
+        * described by a transfer descriptor (the qTD). The qTDs form a linked
+        * list with a queue head (QH).
+        *
+        * Each qTD transfer starts with a new USB packet, i.e. a packet cannot
+        * have its beginning in a qTD transfer and its end in the following
+        * one, so the qTD transfer lengths have to be chosen accordingly.
+        *
+        * Each qTD transfer uses up to QT_BUFFER_CNT data buffers, mapped to
+        * single pages. The first data buffer can start at any offset within a
+        * page (not considering the cache-line alignment issues), while the
+        * following buffers must be page-aligned. There is no alignment
+        * constraint on the size of a qTD transfer.
+        */
+       if (req != NULL)
+               /* 1 qTD will be needed for SETUP, and 1 for ACK. */
+               qtd_count += 1 + 1;
+       if (length > 0 || req == NULL) {
+               /*
+                * Determine the qTD transfer size that will be used for the
+                * data payload (not considering the first qTD transfer, which
+                * may be longer or shorter, and the final one, which may be
+                * shorter).
+                *
+                * In order to keep each packet within a qTD transfer, the qTD
+                * transfer size is aligned to PKT_ALIGN, which is a multiple of
+                * wMaxPacketSize (except in some cases for interrupt transfers,
+                * see comment in submit_int_msg()).
+                *
+                * By default, i.e. if the input buffer is aligned to PKT_ALIGN,
+                * QT_BUFFER_CNT full pages will be used.
+                */
+               int xfr_sz = QT_BUFFER_CNT;
+               /*
+                * However, if the input buffer is not aligned to PKT_ALIGN, the
+                * qTD transfer size will be one page shorter, and the first qTD
+                * data buffer of each transfer will be page-unaligned.
+                */
+               if ((uint32_t)buffer & (PKT_ALIGN - 1))
+                       xfr_sz--;
+               /* Convert the qTD transfer size to bytes. */
+               xfr_sz *= EHCI_PAGE_SIZE;
+               /*
+                * Approximate by excess the number of qTDs that will be
+                * required for the data payload. The exact formula is way more
+                * complicated and saves at most 2 qTDs, i.e. a total of 128
+                * bytes.
+                */
+               qtd_count += 2 + length / xfr_sz;
+       }
+/*
+ * Threshold value based on the worst-case total size of the allocated qTDs for
+ * a mass-storage transfer of 65535 blocks of 512 bytes.
+ */
+#if CONFIG_SYS_MALLOC_LEN <= 64 + 128 * 1024
+#warning CONFIG_SYS_MALLOC_LEN may be too small for EHCI
+#endif
+       qtd = memalign(USB_DMA_MINALIGN, qtd_count * sizeof(struct qTD));
+       if (qtd == NULL) {
+               printf("unable to allocate TDs\n");
                return -1;
        }
-       qh->qh_link = cpu_to_hc32((uint32_t)&qh_list | QH_LINK_TYPE_QH);
-       c = (usb_pipespeed(pipe) != USB_SPEED_HIGH &&
-            usb_pipeendpoint(pipe) == 0) ? 1 : 0;
-       endpt = (8 << 28) |
-           (c << 27) |
-           (usb_maxpacket(dev, pipe) << 16) |
-           (0 << 15) |
-           (1 << 14) |
-           (usb_pipespeed(pipe) << 12) |
-           (usb_pipeendpoint(pipe) << 8) |
-           (0 << 7) | (usb_pipedevice(pipe) << 0);
+
+       memset(qh, 0, sizeof(struct QH));
+       memset(qtd, 0, qtd_count * sizeof(*qtd));
+
+       toggle = usb_gettoggle(dev, usb_pipeendpoint(pipe), usb_pipeout(pipe));
+
+       /*
+        * Setup QH (3.6 in ehci-r10.pdf)
+        *
+        *   qh_link ................. 03-00 H
+        *   qh_endpt1 ............... 07-04 H
+        *   qh_endpt2 ............... 0B-08 H
+        * - qh_curtd
+        *   qh_overlay.qt_next ...... 13-10 H
+        * - qh_overlay.qt_altnext
+        */
+       qh->qh_link = cpu_to_hc32((uint32_t)&ctrl->qh_list | QH_LINK_TYPE_QH);
+       c = (dev->speed != USB_SPEED_HIGH) && !usb_pipeendpoint(pipe);
+       maxpacket = usb_maxpacket(dev, pipe);
+       endpt = QH_ENDPT1_RL(8) | QH_ENDPT1_C(c) |
+               QH_ENDPT1_MAXPKTLEN(maxpacket) | QH_ENDPT1_H(0) |
+               QH_ENDPT1_DTC(QH_ENDPT1_DTC_DT_FROM_QTD) |
+               QH_ENDPT1_EPS(ehci_encode_speed(dev->speed)) |
+               QH_ENDPT1_ENDPT(usb_pipeendpoint(pipe)) | QH_ENDPT1_I(0) |
+               QH_ENDPT1_DEVADDR(usb_pipedevice(pipe));
        qh->qh_endpt1 = cpu_to_hc32(endpt);
-       endpt = (1 << 30) |
-           (dev->portnr << 23) |
-           (dev->parent->devnum << 16) | (0 << 8) | (0 << 0);
+       endpt = QH_ENDPT2_MULT(1) | QH_ENDPT2_PORTNUM(dev->portnr) |
+               QH_ENDPT2_HUBADDR(dev->parent->devnum) |
+               QH_ENDPT2_UFCMASK(0) | QH_ENDPT2_UFSMASK(0);
        qh->qh_endpt2 = cpu_to_hc32(endpt);
        qh->qh_overlay.qt_next = cpu_to_hc32(QT_NEXT_TERMINATE);
 
-       td = NULL;
        tdp = &qh->qh_overlay.qt_next;
 
-       toggle =
-           usb_gettoggle(dev, usb_pipeendpoint(pipe), usb_pipeout(pipe));
-
        if (req != NULL) {
-               td = ehci_alloc(sizeof(struct qTD), 32);
-               if (td == NULL) {
-                       debug("unable to allocate SETUP td\n");
+               /*
+                * Setup request qTD (3.5 in ehci-r10.pdf)
+                *
+                *   qt_next ................ 03-00 H
+                *   qt_altnext ............. 07-04 H
+                *   qt_token ............... 0B-08 H
+                *
+                *   [ buffer, buffer_hi ] loaded with "req".
+                */
+               qtd[qtd_counter].qt_next = cpu_to_hc32(QT_NEXT_TERMINATE);
+               qtd[qtd_counter].qt_altnext = cpu_to_hc32(QT_NEXT_TERMINATE);
+               token = QT_TOKEN_DT(0) | QT_TOKEN_TOTALBYTES(sizeof(*req)) |
+                       QT_TOKEN_IOC(0) | QT_TOKEN_CPAGE(0) | QT_TOKEN_CERR(3) |
+                       QT_TOKEN_PID(QT_TOKEN_PID_SETUP) |
+                       QT_TOKEN_STATUS(QT_TOKEN_STATUS_ACTIVE);
+               qtd[qtd_counter].qt_token = cpu_to_hc32(token);
+               if (ehci_td_buffer(&qtd[qtd_counter], req, sizeof(*req))) {
+                       printf("unable to construct SETUP TD\n");
                        goto fail;
                }
-               td->qt_next = cpu_to_hc32(QT_NEXT_TERMINATE);
-               td->qt_altnext = cpu_to_hc32(QT_NEXT_TERMINATE);
-               token = (0 << 31) |
-                   (sizeof(*req) << 16) |
-                   (0 << 15) | (0 << 12) | (3 << 10) | (2 << 8) | (0x80 << 0);
-               td->qt_token = cpu_to_hc32(token);
-               if (ehci_td_buffer(td, req, sizeof(*req)) != 0) {
-                       debug("unable construct SETUP td\n");
-                       ehci_free(td, sizeof(*td));
-                       goto fail;
-               }
-               *tdp = cpu_to_hc32((uint32_t) td);
-               tdp = &td->qt_next;
+               /* Update previous qTD! */
+               *tdp = cpu_to_hc32((uint32_t)&qtd[qtd_counter]);
+               tdp = &qtd[qtd_counter++].qt_next;
                toggle = 1;
        }
 
        if (length > 0 || req == NULL) {
-               td = ehci_alloc(sizeof(struct qTD), 32);
-               if (td == NULL) {
-                       debug("unable to allocate DATA td\n");
-                       goto fail;
-               }
-               td->qt_next = cpu_to_hc32(QT_NEXT_TERMINATE);
-               td->qt_altnext = cpu_to_hc32(QT_NEXT_TERMINATE);
-               token = (toggle << 31) |
-                   (length << 16) |
-                   ((req == NULL ? 1 : 0) << 15) |
-                   (0 << 12) |
-                   (3 << 10) |
-                   ((usb_pipein(pipe) ? 1 : 0) << 8) | (0x80 << 0);
-               td->qt_token = cpu_to_hc32(token);
-               if (ehci_td_buffer(td, buffer, length) != 0) {
-                       debug("unable construct DATA td\n");
-                       ehci_free(td, sizeof(*td));
-                       goto fail;
-               }
-               *tdp = cpu_to_hc32((uint32_t) td);
-               tdp = &td->qt_next;
+               uint8_t *buf_ptr = buffer;
+               int left_length = length;
+
+               do {
+                       /*
+                        * Determine the size of this qTD transfer. By default,
+                        * QT_BUFFER_CNT full pages can be used.
+                        */
+                       int xfr_bytes = QT_BUFFER_CNT * EHCI_PAGE_SIZE;
+                       /*
+                        * However, if the input buffer is not page-aligned, the
+                        * portion of the first page before the buffer start
+                        * offset within that page is unusable.
+                        */
+                       xfr_bytes -= (uint32_t)buf_ptr & (EHCI_PAGE_SIZE - 1);
+                       /*
+                        * In order to keep each packet within a qTD transfer,
+                        * align the qTD transfer size to PKT_ALIGN.
+                        */
+                       xfr_bytes &= ~(PKT_ALIGN - 1);
+                       /*
+                        * This transfer may be shorter than the available qTD
+                        * transfer size that has just been computed.
+                        */
+                       xfr_bytes = min(xfr_bytes, left_length);
+
+                       /*
+                        * Setup request qTD (3.5 in ehci-r10.pdf)
+                        *
+                        *   qt_next ................ 03-00 H
+                        *   qt_altnext ............. 07-04 H
+                        *   qt_token ............... 0B-08 H
+                        *
+                        *   [ buffer, buffer_hi ] loaded with "buffer".
+                        */
+                       qtd[qtd_counter].qt_next =
+                                       cpu_to_hc32(QT_NEXT_TERMINATE);
+                       qtd[qtd_counter].qt_altnext =
+                                       cpu_to_hc32(QT_NEXT_TERMINATE);
+                       token = QT_TOKEN_DT(toggle) |
+                               QT_TOKEN_TOTALBYTES(xfr_bytes) |
+                               QT_TOKEN_IOC(req == NULL) | QT_TOKEN_CPAGE(0) |
+                               QT_TOKEN_CERR(3) |
+                               QT_TOKEN_PID(usb_pipein(pipe) ?
+                                       QT_TOKEN_PID_IN : QT_TOKEN_PID_OUT) |
+                               QT_TOKEN_STATUS(QT_TOKEN_STATUS_ACTIVE);
+                       qtd[qtd_counter].qt_token = cpu_to_hc32(token);
+                       if (ehci_td_buffer(&qtd[qtd_counter], buf_ptr,
+                                               xfr_bytes)) {
+                               printf("unable to construct DATA TD\n");
+                               goto fail;
+                       }
+                       /* Update previous qTD! */
+                       *tdp = cpu_to_hc32((uint32_t)&qtd[qtd_counter]);
+                       tdp = &qtd[qtd_counter++].qt_next;
+                       /*
+                        * Data toggle has to be adjusted since the qTD transfer
+                        * size is not always an even multiple of
+                        * wMaxPacketSize.
+                        */
+                       if ((xfr_bytes / maxpacket) & 1)
+                               toggle ^= 1;
+                       buf_ptr += xfr_bytes;
+                       left_length -= xfr_bytes;
+               } while (left_length > 0);
        }
 
        if (req != NULL) {
-               td = ehci_alloc(sizeof(struct qTD), 32);
-               if (td == NULL) {
-                       debug("unable to allocate ACK td\n");
-                       goto fail;
-               }
-               td->qt_next = cpu_to_hc32(QT_NEXT_TERMINATE);
-               td->qt_altnext = cpu_to_hc32(QT_NEXT_TERMINATE);
-               token = (toggle << 31) |
-                   (0 << 16) |
-                   (1 << 15) |
-                   (0 << 12) |
-                   (3 << 10) |
-                   ((usb_pipein(pipe) ? 0 : 1) << 8) | (0x80 << 0);
-               td->qt_token = cpu_to_hc32(token);
-               *tdp = cpu_to_hc32((uint32_t) td);
-               tdp = &td->qt_next;
+               /*
+                * Setup request qTD (3.5 in ehci-r10.pdf)
+                *
+                *   qt_next ................ 03-00 H
+                *   qt_altnext ............. 07-04 H
+                *   qt_token ............... 0B-08 H
+                */
+               qtd[qtd_counter].qt_next = cpu_to_hc32(QT_NEXT_TERMINATE);
+               qtd[qtd_counter].qt_altnext = cpu_to_hc32(QT_NEXT_TERMINATE);
+               token = QT_TOKEN_DT(1) | QT_TOKEN_TOTALBYTES(0) |
+                       QT_TOKEN_IOC(1) | QT_TOKEN_CPAGE(0) | QT_TOKEN_CERR(3) |
+                       QT_TOKEN_PID(usb_pipein(pipe) ?
+                               QT_TOKEN_PID_OUT : QT_TOKEN_PID_IN) |
+                       QT_TOKEN_STATUS(QT_TOKEN_STATUS_ACTIVE);
+               qtd[qtd_counter].qt_token = cpu_to_hc32(token);
+               /* Update previous qTD! */
+               *tdp = cpu_to_hc32((uint32_t)&qtd[qtd_counter]);
+               tdp = &qtd[qtd_counter++].qt_next;
        }
 
-       qh_list.qh_link = cpu_to_hc32((uint32_t) qh | QH_LINK_TYPE_QH);
+       ctrl->qh_list.qh_link = cpu_to_hc32((uint32_t)qh | QH_LINK_TYPE_QH);
 
        /* Flush dcache */
-       ehci_flush_dcache(&qh_list);
+       flush_dcache_range((uint32_t)&ctrl->qh_list,
+               ALIGN_END_ADDR(struct QH, &ctrl->qh_list, 1));
+       flush_dcache_range((uint32_t)qh, ALIGN_END_ADDR(struct QH, qh, 1));
+       flush_dcache_range((uint32_t)qtd,
+                          ALIGN_END_ADDR(struct qTD, qtd, qtd_count));
+
+       /* Set async. queue head pointer. */
+       ehci_writel(&ctrl->hcor->or_asynclistaddr, (uint32_t)&ctrl->qh_list);
 
-       usbsts = ehci_readl(&hcor->or_usbsts);
-       ehci_writel(&hcor->or_usbsts, (usbsts & 0x3f));
+       usbsts = ehci_readl(&ctrl->hcor->or_usbsts);
+       ehci_writel(&ctrl->hcor->or_usbsts, (usbsts & 0x3f));
 
        /* Enable async. schedule. */
-       cmd = ehci_readl(&hcor->or_usbcmd);
+       cmd = ehci_readl(&ctrl->hcor->or_usbcmd);
        cmd |= CMD_ASE;
-       ehci_writel(&hcor->or_usbcmd, cmd);
+       ehci_writel(&ctrl->hcor->or_usbcmd, cmd);
 
-       ret = handshake((uint32_t *)&hcor->or_usbsts, STD_ASS, STD_ASS,
+       ret = handshake((uint32_t *)&ctrl->hcor->or_usbsts, STS_ASS, STS_ASS,
                        100 * 1000);
        if (ret < 0) {
-               printf("EHCI fail timeout STD_ASS set\n");
+               printf("EHCI fail timeout STS_ASS set\n");
                goto fail;
        }
 
        /* Wait for TDs to be processed. */
        ts = get_timer(0);
-       vtd = td;
+       vtd = &qtd[qtd_counter - 1];
+       timeout = USB_TIMEOUT_MS(pipe);
        do {
                /* Invalidate dcache */
-               ehci_invalidate_dcache(&qh_list);
+               invalidate_dcache_range((uint32_t)&ctrl->qh_list,
+                       ALIGN_END_ADDR(struct QH, &ctrl->qh_list, 1));
+               invalidate_dcache_range((uint32_t)qh,
+                       ALIGN_END_ADDR(struct QH, qh, 1));
+               invalidate_dcache_range((uint32_t)qtd,
+                       ALIGN_END_ADDR(struct qTD, qtd, qtd_count));
+
                token = hc32_to_cpu(vtd->qt_token);
-               if (!(token & 0x80))
+               if (!(QT_TOKEN_GET_STATUS(token) & QT_TOKEN_STATUS_ACTIVE))
                        break;
-       } while (get_timer(ts) < CONFIG_SYS_HZ);
+               WATCHDOG_RESET();
+       } while (get_timer(ts) < timeout);
+
+       /*
+        * Invalidate the memory area occupied by buffer
+        * Don't try to fix the buffer alignment, if it isn't properly
+        * aligned it's upper layer's fault so let invalidate_dcache_range()
+        * vow about it. But we have to fix the length as it's actual
+        * transfer length and can be unaligned. This is potentially
+        * dangerous operation, it's responsibility of the calling
+        * code to make sure enough space is reserved.
+        */
+       invalidate_dcache_range((uint32_t)buffer,
+               ALIGN((uint32_t)buffer + length, ARCH_DMA_MINALIGN));
+
+       /* Check that the TD processing happened */
+       if (QT_TOKEN_GET_STATUS(token) & QT_TOKEN_STATUS_ACTIVE)
+               printf("EHCI timed out on TD - token=%#x\n", token);
 
        /* Disable async schedule. */
-       cmd = ehci_readl(&hcor->or_usbcmd);
+       cmd = ehci_readl(&ctrl->hcor->or_usbcmd);
        cmd &= ~CMD_ASE;
-       ehci_writel(&hcor->or_usbcmd, cmd);
+       ehci_writel(&ctrl->hcor->or_usbcmd, cmd);
 
-       ret = handshake((uint32_t *)&hcor->or_usbsts, STD_ASS, 0,
+       ret = handshake((uint32_t *)&ctrl->hcor->or_usbsts, STS_ASS, 0,
                        100 * 1000);
        if (ret < 0) {
-               printf("EHCI fail timeout STD_ASS reset\n");
+               printf("EHCI fail timeout STS_ASS reset\n");
                goto fail;
        }
 
-       qh_list.qh_link = cpu_to_hc32((uint32_t)&qh_list | QH_LINK_TYPE_QH);
-
        token = hc32_to_cpu(qh->qh_overlay.qt_token);
-       if (!(token & 0x80)) {
+       if (!(QT_TOKEN_GET_STATUS(token) & QT_TOKEN_STATUS_ACTIVE)) {
                debug("TOKEN=%#x\n", token);
-               switch (token & 0xfc) {
+               switch (QT_TOKEN_GET_STATUS(token) &
+                       ~(QT_TOKEN_STATUS_SPLITXSTATE | QT_TOKEN_STATUS_PERR)) {
                case 0:
-                       toggle = token >> 31;
+                       toggle = QT_TOKEN_GET_DT(token);
                        usb_settoggle(dev, usb_pipeendpoint(pipe),
                                       usb_pipeout(pipe), toggle);
                        dev->status = 0;
                        break;
-               case 0x40:
+               case QT_TOKEN_STATUS_HALTED:
                        dev->status = USB_ST_STALLED;
                        break;
-               case 0xa0:
-               case 0x20:
+               case QT_TOKEN_STATUS_ACTIVE | QT_TOKEN_STATUS_DATBUFERR:
+               case QT_TOKEN_STATUS_DATBUFERR:
                        dev->status = USB_ST_BUF_ERR;
                        break;
-               case 0x50:
-               case 0x10:
+               case QT_TOKEN_STATUS_HALTED | QT_TOKEN_STATUS_BABBLEDET:
+               case QT_TOKEN_STATUS_BABBLEDET:
                        dev->status = USB_ST_BABBLE_DET;
                        break;
                default:
                        dev->status = USB_ST_CRC_ERR;
-                       if ((token & 0x40) == 0x40)
+                       if (QT_TOKEN_GET_STATUS(token) & QT_TOKEN_STATUS_HALTED)
                                dev->status |= USB_ST_STALLED;
                        break;
                }
-               dev->act_len = length - ((token >> 16) & 0x7fff);
+               dev->act_len = length - QT_TOKEN_GET_TOTALBYTES(token);
        } else {
                dev->act_len = 0;
                debug("dev=%u, usbsts=%#x, p[1]=%#x, p[2]=%#x\n",
-                     dev->devnum, ehci_readl(&hcor->or_usbsts),
-                     ehci_readl(&hcor->or_portsc[0]),
-                     ehci_readl(&hcor->or_portsc[1]));
+                     dev->devnum, ehci_readl(&ctrl->hcor->or_usbsts),
+                     ehci_readl(&ctrl->hcor->or_portsc[0]),
+                     ehci_readl(&ctrl->hcor->or_portsc[1]));
        }
 
+       free(qtd);
        return (dev->status != USB_ST_NOT_PROC) ? 0 : -1;
 
 fail:
-       td = (void *)hc32_to_cpu(qh->qh_overlay.qt_next);
-       while (td != (void *)QT_NEXT_TERMINATE) {
-               qh->qh_overlay.qt_next = td->qt_next;
-               ehci_free(td, sizeof(*td));
-               td = (void *)hc32_to_cpu(qh->qh_overlay.qt_next);
-       }
-       ehci_free(qh, sizeof(*qh));
+       free(qtd);
        return -1;
 }
 
@@ -537,13 +602,14 @@ ehci_submit_root(struct usb_device *dev, unsigned long pipe, void *buffer,
        int len, srclen;
        uint32_t reg;
        uint32_t *status_reg;
+       struct ehci_ctrl *ctrl = dev->controller;
 
        if (le16_to_cpu(req->index) > CONFIG_SYS_USB_EHCI_MAX_ROOT_PORTS) {
                printf("The request port(%d) is not configured\n",
                        le16_to_cpu(req->index) - 1);
                return -1;
        }
-       status_reg = (uint32_t *)&hcor->or_portsc[
+       status_reg = (uint32_t *)&ctrl->hcor->or_portsc[
                                                le16_to_cpu(req->index) - 1];
        srclen = 0;
 
@@ -560,12 +626,14 @@ ehci_submit_root(struct usb_device *dev, unsigned long pipe, void *buffer,
                case USB_DT_DEVICE:
                        debug("USB_DT_DEVICE request\n");
                        srcptr = &descriptor.device;
-                       srclen = 0x12;
+                       srclen = descriptor.device.bLength;
                        break;
                case USB_DT_CONFIG:
                        debug("USB_DT_CONFIG config\n");
                        srcptr = &descriptor.config;
-                       srclen = 0x19;
+                       srclen = descriptor.config.bLength +
+                                       descriptor.interface.bLength +
+                                       descriptor.endpoint.bLength;
                        break;
                case USB_DT_STRING:
                        debug("USB_DT_STRING config\n");
@@ -600,7 +668,7 @@ ehci_submit_root(struct usb_device *dev, unsigned long pipe, void *buffer,
                case USB_DT_HUB:
                        debug("USB_DT_HUB config\n");
                        srcptr = &descriptor.hub;
-                       srclen = 0x8;
+                       srclen = descriptor.hub.bLength;
                        break;
                default:
                        debug("unknown value %x\n", le16_to_cpu(req->value));
@@ -609,7 +677,7 @@ ehci_submit_root(struct usb_device *dev, unsigned long pipe, void *buffer,
                break;
        case USB_REQ_SET_ADDRESS | (USB_RECIP_DEVICE << 8):
                debug("USB_REQ_SET_ADDRESS\n");
-               rootdev = le16_to_cpu(req->value);
+               ctrl->rootdev = le16_to_cpu(req->value);
                break;
        case DeviceOutRequest | USB_REQ_SET_CONFIGURATION:
                debug("USB_REQ_SET_CONFIGURATION\n");
@@ -638,13 +706,13 @@ ehci_submit_root(struct usb_device *dev, unsigned long pipe, void *buffer,
                        tmpbuf[1] |= USB_PORT_STAT_POWER >> 8;
 
                if (ehci_is_TDI()) {
-                       switch ((reg >> 26) & 3) {
-                       case 0:
+                       switch (PORTSC_PSPD(reg)) {
+                       case PORTSC_PSPD_FS:
                                break;
-                       case 1:
+                       case PORTSC_PSPD_LS:
                                tmpbuf[1] |= USB_PORT_STAT_LOW_SPEED >> 8;
                                break;
-                       case 2:
+                       case PORTSC_PSPD_HS:
                        default:
                                tmpbuf[1] |= USB_PORT_STAT_HIGH_SPEED >> 8;
                                break;
@@ -659,7 +727,7 @@ ehci_submit_root(struct usb_device *dev, unsigned long pipe, void *buffer,
                        tmpbuf[2] |= USB_PORT_STAT_C_ENABLE;
                if (reg & EHCI_PS_OCC)
                        tmpbuf[2] |= USB_PORT_STAT_C_OVERCURRENT;
-               if (portreset & (1 << le16_to_cpu(req->index)))
+               if (ctrl->portreset & (1 << le16_to_cpu(req->index)))
                        tmpbuf[2] |= USB_PORT_STAT_C_RESET;
 
                srcptr = tmpbuf;
@@ -674,7 +742,7 @@ ehci_submit_root(struct usb_device *dev, unsigned long pipe, void *buffer,
                        ehci_writel(status_reg, reg);
                        break;
                case USB_PORT_FEAT_POWER:
-                       if (HCS_PPC(ehci_readl(&hccr->cr_hcsparams))) {
+                       if (HCS_PPC(ehci_readl(&ctrl->hccr->cr_hcsparams))) {
                                reg |= EHCI_PS_PP;
                                ehci_writel(status_reg, reg);
                        }
@@ -700,8 +768,8 @@ ehci_submit_root(struct usb_device *dev, unsigned long pipe, void *buffer,
                                 * usb 2.0 specification say 50 ms resets on
                                 * root
                                 */
-                               wait_ms(50);
-                               /* terminate the reset */
+                               ehci_powerup_fixup(status_reg, &reg);
+
                                ehci_writel(status_reg, reg & ~EHCI_PS_PR);
                                /*
                                 * A host controller must terminate the reset
@@ -711,7 +779,7 @@ ehci_submit_root(struct usb_device *dev, unsigned long pipe, void *buffer,
                                ret = handshake(status_reg, EHCI_PS_PR, 0,
                                                2 * 1000);
                                if (!ret)
-                                       portreset |=
+                                       ctrl->portreset |=
                                                1 << le16_to_cpu(req->index);
                                else
                                        printf("port(%d) reset error\n",
@@ -723,7 +791,7 @@ ehci_submit_root(struct usb_device *dev, unsigned long pipe, void *buffer,
                        goto unknown;
                }
                /* unblock posted writes */
-               (void) ehci_readl(&hcor->or_usbcmd);
+               (void) ehci_readl(&ctrl->hcor->or_usbcmd);
                break;
        case USB_REQ_CLEAR_FEATURE | ((USB_DIR_OUT | USB_RT_PORT) << 8):
                reg = ehci_readl(status_reg);
@@ -735,7 +803,7 @@ ehci_submit_root(struct usb_device *dev, unsigned long pipe, void *buffer,
                        reg = (reg & ~EHCI_PS_CLEAR) | EHCI_PS_PE;
                        break;
                case USB_PORT_FEAT_POWER:
-                       if (HCS_PPC(ehci_readl(&hccr->cr_hcsparams)))
+                       if (HCS_PPC(ehci_readl(&ctrl->hccr->cr_hcsparams)))
                                reg = reg & ~(EHCI_PS_CLEAR | EHCI_PS_PP);
                case USB_PORT_FEAT_C_CONNECTION:
                        reg = (reg & ~EHCI_PS_CLEAR) | EHCI_PS_CSC;
@@ -744,7 +812,7 @@ ehci_submit_root(struct usb_device *dev, unsigned long pipe, void *buffer,
                        reg = (reg & ~EHCI_PS_CLEAR) | EHCI_PS_OCC;
                        break;
                case USB_PORT_FEAT_C_RESET:
-                       portreset &= ~(1 << le16_to_cpu(req->index));
+                       ctrl->portreset &= ~(1 << le16_to_cpu(req->index));
                        break;
                default:
                        debug("unknown feature %x\n", le16_to_cpu(req->value));
@@ -752,14 +820,14 @@ ehci_submit_root(struct usb_device *dev, unsigned long pipe, void *buffer,
                }
                ehci_writel(status_reg, reg);
                /* unblock posted write */
-               (void) ehci_readl(&hcor->or_usbcmd);
+               (void) ehci_readl(&ctrl->hcor->or_usbcmd);
                break;
        default:
                debug("Unknown request\n");
                goto unknown;
        }
 
-       wait_ms(1);
+       mdelay(1);
        len = min3(srclen, le16_to_cpu(req->length), length);
        if (srcptr != NULL && len > 0)
                memcpy(buffer, srcptr, len);
@@ -780,72 +848,77 @@ unknown:
        return -1;
 }
 
-int usb_lowlevel_stop(void)
+int usb_lowlevel_stop(int index)
 {
-       return ehci_hcd_stop();
+       return ehci_hcd_stop(index);
 }
 
-int usb_lowlevel_init(void)
+int usb_lowlevel_init(int index, void **controller)
 {
        uint32_t reg;
        uint32_t cmd;
+       struct QH *qh_list;
 
-       if (ehci_hcd_init() != 0)
+       if (ehci_hcd_init(index, &ehcic[index].hccr, &ehcic[index].hcor))
                return -1;
 
        /* EHCI spec section 4.1 */
-       if (ehci_reset() != 0)
+       if (ehci_reset(index))
                return -1;
 
 #if defined(CONFIG_EHCI_HCD_INIT_AFTER_RESET)
-       if (ehci_hcd_init() != 0)
+       if (ehci_hcd_init(index, &ehcic[index].hccr, &ehcic[index].hcor))
                return -1;
 #endif
 
-       /* Set head of reclaim list */
-       memset(&qh_list, 0, sizeof(qh_list));
-       qh_list.qh_link = cpu_to_hc32((uint32_t)&qh_list | QH_LINK_TYPE_QH);
-       qh_list.qh_endpt1 = cpu_to_hc32((1 << 15) | (USB_SPEED_HIGH << 12));
-       qh_list.qh_curtd = cpu_to_hc32(QT_NEXT_TERMINATE);
-       qh_list.qh_overlay.qt_next = cpu_to_hc32(QT_NEXT_TERMINATE);
-       qh_list.qh_overlay.qt_altnext = cpu_to_hc32(QT_NEXT_TERMINATE);
-       qh_list.qh_overlay.qt_token = cpu_to_hc32(0x40);
-
-       /* Set async. queue head pointer. */
-       ehci_writel(&hcor->or_asynclistaddr, (uint32_t)&qh_list);
+       qh_list = &ehcic[index].qh_list;
 
-       reg = ehci_readl(&hccr->cr_hcsparams);
+       /* Set head of reclaim list */
+       memset(qh_list, 0, sizeof(*qh_list));
+       qh_list->qh_link = cpu_to_hc32((uint32_t)qh_list | QH_LINK_TYPE_QH);
+       qh_list->qh_endpt1 = cpu_to_hc32(QH_ENDPT1_H(1) |
+                                               QH_ENDPT1_EPS(USB_SPEED_HIGH));
+       qh_list->qh_curtd = cpu_to_hc32(QT_NEXT_TERMINATE);
+       qh_list->qh_overlay.qt_next = cpu_to_hc32(QT_NEXT_TERMINATE);
+       qh_list->qh_overlay.qt_altnext = cpu_to_hc32(QT_NEXT_TERMINATE);
+       qh_list->qh_overlay.qt_token =
+                       cpu_to_hc32(QT_TOKEN_STATUS(QT_TOKEN_STATUS_HALTED));
+
+       reg = ehci_readl(&ehcic[index].hccr->cr_hcsparams);
        descriptor.hub.bNbrPorts = HCS_N_PORTS(reg);
-       printf("Register %x NbrPorts %d\n", reg, descriptor.hub.bNbrPorts);
+       debug("Register %x NbrPorts %d\n", reg, descriptor.hub.bNbrPorts);
        /* Port Indicators */
        if (HCS_INDICATOR(reg))
-               descriptor.hub.wHubCharacteristics |= 0x80;
+               put_unaligned(get_unaligned(&descriptor.hub.wHubCharacteristics)
+                               | 0x80, &descriptor.hub.wHubCharacteristics);
        /* Port Power Control */
        if (HCS_PPC(reg))
-               descriptor.hub.wHubCharacteristics |= 0x01;
+               put_unaligned(get_unaligned(&descriptor.hub.wHubCharacteristics)
+                               | 0x01, &descriptor.hub.wHubCharacteristics);
 
        /* Start the host controller. */
-       cmd = ehci_readl(&hcor->or_usbcmd);
+       cmd = ehci_readl(&ehcic[index].hcor->or_usbcmd);
        /*
         * Philips, Intel, and maybe others need CMD_RUN before the
         * root hub will detect new devices (why?); NEC doesn't
         */
        cmd &= ~(CMD_LRESET|CMD_IAAD|CMD_PSE|CMD_ASE|CMD_RESET);
        cmd |= CMD_RUN;
-       ehci_writel(&hcor->or_usbcmd, cmd);
+       ehci_writel(&ehcic[index].hcor->or_usbcmd, cmd);
 
        /* take control over the ports */
-       cmd = ehci_readl(&hcor->or_configflag);
+       cmd = ehci_readl(&ehcic[index].hcor->or_configflag);
        cmd |= FLAG_CF;
-       ehci_writel(&hcor->or_configflag, cmd);
+       ehci_writel(&ehcic[index].hcor->or_configflag, cmd);
        /* unblock posted write */
-       cmd = ehci_readl(&hcor->or_usbcmd);
-       wait_ms(5);
-       reg = HC_VERSION(ehci_readl(&hccr->cr_capbase));
+       cmd = ehci_readl(&ehcic[index].hcor->or_usbcmd);
+       mdelay(5);
+       reg = HC_VERSION(ehci_readl(&ehcic[index].hccr->cr_capbase));
        printf("USB EHCI %x.%02x\n", reg >> 8, reg & 0xff);
 
-       rootdev = 0;
+       ehcic[index].rootdev = 0;
 
+       *controller = &ehcic[index];
        return 0;
 }
 
@@ -865,14 +938,15 @@ int
 submit_control_msg(struct usb_device *dev, unsigned long pipe, void *buffer,
                   int length, struct devrequest *setup)
 {
+       struct ehci_ctrl *ctrl = dev->controller;
 
        if (usb_pipetype(pipe) != PIPE_CONTROL) {
                debug("non-control pipe (type=%lu)", usb_pipetype(pipe));
                return -1;
        }
 
-       if (usb_pipedevice(pipe) == rootdev) {
-               if (rootdev == 0)
+       if (usb_pipedevice(pipe) == ctrl->rootdev) {
+               if (!ctrl->rootdev)
                        dev->speed = USB_SPEED_HIGH;
                return ehci_submit_root(dev, pipe, buffer, length, setup);
        }
@@ -883,8 +957,24 @@ int
 submit_int_msg(struct usb_device *dev, unsigned long pipe, void *buffer,
               int length, int interval)
 {
-
        debug("dev=%p, pipe=%lu, buffer=%p, length=%d, interval=%d",
              dev, pipe, buffer, length, interval);
-       return -1;
+
+       /*
+        * Interrupt transfers requiring several transactions are not supported
+        * because bInterval is ignored.
+        *
+        * Also, ehci_submit_async() relies on wMaxPacketSize being a power of 2
+        * <= PKT_ALIGN if several qTDs are required, while the USB
+        * specification does not constrain this for interrupt transfers. That
+        * means that ehci_submit_async() would support interrupt transfers
+        * requiring several transactions only as long as the transfer size does
+        * not require more than a single qTD.
+        */
+       if (length > usb_maxpacket(dev, pipe)) {
+               printf("%s: Interrupt transfers requiring several transactions "
+                       "are not supported.\n", __func__);
+               return -1;
+       }
+       return ehci_submit_async(dev, pipe, buffer, length, NULL);
 }