From d690d89cc7487bece9897728d277eee1283e59f7 Mon Sep 17 00:00:00 2001 From: Dinh Nguyen Date: Wed, 16 Sep 2009 12:56:38 -0500 Subject: [PATCH] usb: imx usb buffers in iram work-around Some i.MX parts (MX31, MX37) need to have usb host queue descriptors and/or buffers in on-chip IRAM in cases of heavily loaded memory system (i.e. large LCDs). Signed-off-by: Dinh Nguyen --- drivers/usb/host/ehci-hcd.c | 5 + drivers/usb/host/ehci-mem-iram.c | 514 ++++++++++++ drivers/usb/host/ehci-q-iram.c | 1345 ++++++++++++++++++++++++++++++ drivers/usb/host/ehci.h | 15 + 4 files changed, 1879 insertions(+) create mode 100644 drivers/usb/host/ehci-mem-iram.c create mode 100644 drivers/usb/host/ehci-q-iram.c diff --git a/drivers/usb/host/ehci-hcd.c b/drivers/usb/host/ehci-hcd.c index f89f77f1b339..bf3418362df6 100644 --- a/drivers/usb/host/ehci-hcd.c +++ b/drivers/usb/host/ehci-hcd.c @@ -335,8 +335,13 @@ static void ehci_work(struct ehci_hcd *ehci); #include "ehci-hub.c" #include "ehci-lpm.c" +#ifdef CONFIG_USB_STATIC_IRAM +#include "ehci-mem-iram.c" +#include "ehci-q-iram.c" +#else #include "ehci-mem.c" #include "ehci-q.c" +#endif #include "ehci-sched.c" /*-------------------------------------------------------------------------*/ diff --git a/drivers/usb/host/ehci-mem-iram.c b/drivers/usb/host/ehci-mem-iram.c new file mode 100644 index 000000000000..fd7b42b22758 --- /dev/null +++ b/drivers/usb/host/ehci-mem-iram.c @@ -0,0 +1,514 @@ +/* + * Copyright (c) 2001 by David Brownell + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License as published by the + * Free Software Foundation; either version 2 of the License, or (at your + * option) any later version. + * + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software Foundation, + * Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + */ + +/* this file is part of ehci-hcd.c */ + +/*-------------------------------------------------------------------------*/ + +/* + * There's basically three types of memory: + * - data used only by the HCD ... kmalloc is fine + * - async and periodic schedules, shared by HC and HCD ... these + * need to use dma_pool or dma_alloc_coherent + * - driver buffers, read/written by HC ... single shot DMA mapped + * + * There's also "register" data (e.g. PCI or SOC), which is memory mapped. + * No memory seen by this driver is pageable. + */ + +/*-------------------------------------------------------------------------*/ + +/* Allocate the key transfer structures from the previously allocated pool */ +#include +#include + +bool use_iram_qtd; + +struct memDesc { + u32 start; + u32 end; + struct memDesc *next; +} ; + +static u32 g_usb_pool_start; +static s32 g_usb_pool_count; +static u32 g_total_pages; +static u32 g_alignment = 32; +struct memDesc *g_allocated_desc; +static spinlock_t g_usb_sema; +static u32 g_debug_qtd_allocated; +static u32 g_debug_qH_allocated; +static int g_alloc_map; +static unsigned long g_iram_base; +static __iomem void *g_iram_addr; + +/*! + * usb_pool_initialize + * + * @param memPool start address of the pool + * @param poolSize memory pool size + * @param alignment alignment for example page alignmnet will be 4K + * + * @return 0 for success -1 for errors + */ +static int usb_pool_initialize(u32 memPool, u32 poolSize, u32 alignment) +{ + if (g_usb_pool_count) { + printk(KERN_INFO "usb_pool_initialize : already initialzed.\n"); + return 0; + } + + g_alignment = alignment; + if (g_alignment == 0) { + printk(KERN_INFO + "usb_pool_initialize : g_alignment can not be zero.\n"); + g_alignment = 32; + } + + g_total_pages = poolSize / g_alignment; + g_usb_pool_start = (u32) memPool; + + g_allocated_desc = kmalloc(sizeof(struct memDesc), GFP_KERNEL); + if (!g_allocated_desc) { + printk(KERN_ALERT "usb_pool_initialize : kmalloc failed \n"); + return (-1); + } + + g_allocated_desc->start = 0; + g_allocated_desc->end = 0; + g_allocated_desc->next = NULL; + + spin_lock_init(&g_usb_sema); + g_usb_pool_count++; + return (0); +} + +static void usb_pool_deinit() +{ + if (--g_usb_pool_count < 0) + g_usb_pool_count = 0; +} + +/*! + * usb_malloc + * + * @param size memory pool size + * + * @return physical address, 0 for error + */ +static u32 usb_malloc(u32 size, gfp_t mem_flags) +{ + unsigned long flags; + struct memDesc *prevDesc = NULL; + struct memDesc *nextDesc = NULL; + struct memDesc *currentDesc = NULL; + u32 pages = (size + g_alignment - 1) / g_alignment; + + if ((size == 0) || (pages > g_total_pages)) + return 0; + + currentDesc = kmalloc(sizeof(struct memDesc), mem_flags); + if (!currentDesc) { + printk(KERN_ALERT "usb_malloc: kmalloc failed \n"); + return 0; + } + + spin_lock_irqsave(&g_usb_sema, flags); + + /* Create the first Allocated descriptor */ + if (!g_allocated_desc->next) { + g_allocated_desc->next = currentDesc; + currentDesc->start = 0; + currentDesc->end = pages; + currentDesc->next = NULL; + spin_unlock_irqrestore(&g_usb_sema, flags); + return (g_usb_pool_start + currentDesc->start * g_alignment); + } + + /* Find the free spot */ + prevDesc = g_allocated_desc; + while (prevDesc->next) { + nextDesc = prevDesc->next; + if (pages <= nextDesc->start - prevDesc->end) { + currentDesc->start = prevDesc->end; + currentDesc->end = currentDesc->start + pages; + currentDesc->next = nextDesc; + prevDesc->next = currentDesc; + break; + } + prevDesc = nextDesc; + } + + /* Do not find the free spot inside the chain, append to the end */ + if (!prevDesc->next) { + if (pages > (g_total_pages - prevDesc->end)) { + spin_unlock_irqrestore(&g_usb_sema, flags); + kfree(currentDesc); + return 0; + } else { + currentDesc->start = prevDesc->end; + currentDesc->end = currentDesc->start + pages; + currentDesc->next = NULL; + prevDesc->next = currentDesc; + } + } + + spin_unlock_irqrestore(&g_usb_sema, flags); + return (g_usb_pool_start + currentDesc->start * g_alignment); +} + +/*! + * usb_free + * + * @param physical physical address try to free + * + */ +static void usb_free(u32 physical) +{ + unsigned long flags; + struct memDesc *prevDesc = NULL; + struct memDesc *nextDesc = NULL; + u32 pages = (physical - g_usb_pool_start) / g_alignment; + + /* Protect the memory pool data structures. */ + spin_lock_irqsave(&g_usb_sema, flags); + + prevDesc = g_allocated_desc; + while (prevDesc->next) { + nextDesc = prevDesc->next; + if (nextDesc->start == pages) { + prevDesc->next = nextDesc->next; + kfree(nextDesc); + break; + } + prevDesc = prevDesc->next; + } + /* All done with memory pool data structures. */ + spin_unlock_irqrestore(&g_usb_sema, flags); +} + +static int address_to_buffer(struct ehci_hcd *ehci, int address) +{ + int i; + + for (i = 0; i < IRAM_NTD; i++) { + if (ehci->usb_address[i] == address) + return i; + } + return IRAM_NTD; +} + +static void use_buffer(struct ehci_hcd *ehci, int address) +{ + int i; + + for (i = 0; i < IRAM_NTD; i++) { + if (ehci->usb_address[i] == address) + return; + } + + if (ehci->usb_address[0] == 0) { + ehci->usb_address[0] = address; + printk(KERN_INFO "usb_address[0] %x\n", address); + return; + } else if (ehci->usb_address[1] == 0) { + ehci->usb_address[1] = address; + printk(KERN_INFO "usb_address[1] %x\n", address); + return; + } else + printk(KERN_ALERT "qh_make run out of iRAM, already be used"); +} + +static u32 alloc_iram_buf(void) +{ + int i; + + for (i = 0; i < IRAM_NTD; i++) { + if (!(g_alloc_map & (1 << i))) { + g_alloc_map |= (1 << i); + return g_iram_base + i * (IRAM_TD_SIZE * 2); + } + } + panic("Out of IRAM buffers\n"); +} + +void free_iram_buf(u32 buf) +{ + int i = (buf - g_iram_base) / (IRAM_TD_SIZE * 2); + + g_alloc_map &= ~(1 << i); +} + +static inline void ehci_qtd_init(struct ehci_hcd *ehci, struct ehci_qtd *qtd, + dma_addr_t dma) +{ + memset(qtd, 0, sizeof *qtd); + qtd->qtd_dma = dma; + qtd->hw_token = cpu_to_le32(QTD_STS_HALT); + qtd->hw_next = EHCI_LIST_END(ehci); + qtd->hw_alt_next = EHCI_LIST_END(ehci); + INIT_LIST_HEAD(&qtd->qtd_list); +} + +static struct ehci_qtd *ehci_qtd_alloc(struct ehci_hcd *ehci, gfp_t flags) +{ + struct ehci_qtd *qtd; + dma_addr_t dma; + + if (use_iram_qtd) { + dma = usb_malloc(sizeof(struct ehci_qtd), flags); + if (dma != 0) + qtd = (struct ehci_qtd *)(g_iram_addr + (dma - g_iram_base)); + else + qtd = dma_pool_alloc(ehci->qtd_pool, flags, &dma); + } + else + qtd = dma_pool_alloc(ehci->qtd_pool, flags, &dma); + + if (qtd != NULL) { + ehci_qtd_init(ehci, qtd, dma); + ++g_debug_qtd_allocated; + } else { + panic + ("out of i-ram for qtd allocation g_debug_qtd_allocated %d \ + size%d \n", g_debug_qtd_allocated, + sizeof(struct ehci_qtd)); + } + return qtd; +} + +static inline void ehci_qtd_free(struct ehci_hcd *ehci, struct ehci_qtd *qtd) +{ + if ((qtd->qtd_dma & (g_iram_base & 0xFFF00000)) == + (g_iram_base & 0xFFF00000)) + usb_free(qtd->qtd_dma); + else + dma_pool_free(ehci->qtd_pool, qtd, qtd->qtd_dma); + --g_debug_qtd_allocated; +} + +static void qh_destroy(struct ehci_qh *qh) +{ + struct ehci_hcd *ehci = qh->ehci; + + /* clean qtds first, and know this is not linked */ + if (!list_empty(&qh->qtd_list) || qh->qh_next.ptr) { + ehci_dbg(ehci, "unused qh not empty!\n"); + BUG(); + } + if (qh->dummy) + ehci_qtd_free(ehci, qh->dummy); + int i; + for (i = 0; i < IRAM_NTD; i++) { + if (ehci->usb_address[i] == (qh->hw_info1 & 0x7F)) + ehci->usb_address[i] = 0; + } + + if ((qh->qh_dma & (g_iram_base & 0xFFF00000)) == + (g_iram_base & 0xFFF00000)) + usb_free(qh->qh_dma); + else + dma_pool_free(ehci->qh_pool, qh, qh->qh_dma); + --g_debug_qH_allocated; +} + +static struct ehci_qh *ehci_qh_alloc(struct ehci_hcd *ehci, gfp_t flags) +{ + struct ehci_qh *qh; + dma_addr_t dma; + + dma = usb_malloc(sizeof(struct ehci_qh), flags); + if (dma != 0) + qh = (struct ehci_qh *)(g_iram_addr + (dma - g_iram_base)); + else + qh = (struct ehci_qh *) + dma_pool_alloc(ehci->qh_pool, flags, &dma); + ++g_debug_qH_allocated; + if (qh == NULL) { + panic("run out of i-ram for qH allocation\n"); + return qh; + } + + memset(qh, 0, sizeof *qh); + qh->refcount = 1; + qh->ehci = ehci; + qh->qh_dma = dma; + INIT_LIST_HEAD(&qh->qtd_list); + + /* dummy td enables safe urb queuing */ + qh->dummy = ehci_qtd_alloc(ehci, flags); + if (qh->dummy == NULL) { + ehci_dbg(ehci, "no dummy td\n"); + dma_pool_free(ehci->qh_pool, qh, qh->qh_dma); + qh = NULL; + } + return qh; +} + +/* to share a qh (cpu threads, or hc) */ +static inline struct ehci_qh *qh_get(struct ehci_qh *qh) +{ + WARN_ON(!qh->refcount); + qh->refcount++; + return qh; +} + +static inline void qh_put(struct ehci_qh *qh) +{ + if (!--qh->refcount) + qh_destroy(qh); +} + +/*-------------------------------------------------------------------------*/ + +/* The queue heads and transfer descriptors are managed from pools tied + * to each of the "per device" structures. + * This is the initialisation and cleanup code. + */ + +static void ehci_mem_cleanup(struct ehci_hcd *ehci) +{ + if (ehci->async) + qh_put(ehci->async); + ehci->async = NULL; + + /* DMA consistent memory and pools */ + if (ehci->qtd_pool) + dma_pool_destroy(ehci->qtd_pool); + ehci->qtd_pool = NULL; + + if (ehci->qh_pool) { + dma_pool_destroy(ehci->qh_pool); + ehci->qh_pool = NULL; + } + + if (ehci->itd_pool) + dma_pool_destroy(ehci->itd_pool); + ehci->itd_pool = NULL; + + if (ehci->sitd_pool) + dma_pool_destroy(ehci->sitd_pool); + ehci->sitd_pool = NULL; + + if (ehci->periodic) + dma_free_coherent(ehci_to_hcd(ehci)->self.controller, + ehci->periodic_size * sizeof(u32), + ehci->periodic, ehci->periodic_dma); + ehci->periodic = NULL; + + if (ehci->iram_buffer[0]) + free_iram_buf(ehci->iram_buffer[0]); + if (ehci->iram_buffer[1]) + free_iram_buf(ehci->iram_buffer[1]); + + iounmap(g_iram_addr); + iram_free(g_iram_base, USB_IRAM_SIZE); + + /* shadow periodic table */ + kfree(ehci->pshadow); + ehci->pshadow = NULL; + usb_pool_deinit(); +} + +/* remember to add cleanup code (above) if you add anything here */ +static int ehci_mem_init(struct ehci_hcd *ehci, gfp_t flags) +{ + int i; + g_usb_pool_count = 0; + g_debug_qtd_allocated = 0; + g_debug_qH_allocated = 0; + g_alloc_map = 0; + + if (cpu_is_mx37()) + use_iram_qtd = 0; + else + use_iram_qtd = 1; + + g_iram_addr = iram_alloc(USB_IRAM_SIZE, &g_iram_base); + + usb_pool_initialize(g_iram_base + IRAM_TD_SIZE * IRAM_NTD * 2, + USB_IRAM_SIZE - IRAM_TD_SIZE * IRAM_NTD * 2, 32); + + if (!ehci->iram_buffer[0]) { + ehci->iram_buffer[0] = alloc_iram_buf(); + ehci->iram_buffer_v[0] = g_iram_addr + (ehci->iram_buffer[0] - g_iram_base); + ehci->iram_buffer[1] = alloc_iram_buf(); + ehci->iram_buffer_v[1] = g_iram_addr + (ehci->iram_buffer[1] - g_iram_base); + } + + /* QTDs for control/bulk/intr transfers */ + ehci->qtd_pool = dma_pool_create("ehci_qtd", + ehci_to_hcd(ehci)->self.controller, + sizeof(struct ehci_qtd), + 32/* byte alignment (for hw parts) */ + , 4096 /* can't cross 4K */); + if (!ehci->qtd_pool) + goto fail; + + /* QHs for control/bulk/intr transfers */ + ehci->qh_pool = dma_pool_create("ehci_qh", + ehci_to_hcd(ehci)->self.controller, + sizeof(struct ehci_qh), + 32 /* byte alignment (for hw parts) */ , + 4096 /* can't cross 4K */); + if (!ehci->qh_pool) + goto fail; + + ehci->async = ehci_qh_alloc(ehci, flags); + if (!ehci->async) + goto fail; + + /* ITD for high speed ISO transfers */ + ehci->itd_pool = dma_pool_create("ehci_itd", + ehci_to_hcd(ehci)->self.controller, + sizeof(struct ehci_itd), + 32/* byte alignment (for hw parts) */ + , 4096 /* can't cross 4K */); + if (!ehci->itd_pool) + goto fail; + + /* SITD for full/low speed split ISO transfers */ + ehci->sitd_pool = dma_pool_create("ehci_sitd", + ehci_to_hcd(ehci)->self.controller, + sizeof(struct ehci_sitd), + 32/* byte alignment (for hw parts) */ + , 4096 /* can't cross 4K */); + if (!ehci->sitd_pool) + goto fail; + + ehci->periodic = (__le32 *) + dma_alloc_coherent(ehci_to_hcd(ehci)->self.controller, + ehci->periodic_size * sizeof(__le32), + &ehci->periodic_dma, 0); + + if (ehci->periodic == NULL) + goto fail; + + for (i = 0; i < ehci->periodic_size; i++) + ehci->periodic[i] = EHCI_LIST_END(ehci); + + /* software shadow of hardware table */ + ehci->pshadow = kcalloc(ehci->periodic_size, sizeof(void *), flags); + if (ehci->pshadow != NULL) + return 0; + +fail: + ehci_dbg(ehci, "couldn't init memory\n"); + ehci_mem_cleanup(ehci); + return -ENOMEM; +} diff --git a/drivers/usb/host/ehci-q-iram.c b/drivers/usb/host/ehci-q-iram.c new file mode 100644 index 000000000000..318888563380 --- /dev/null +++ b/drivers/usb/host/ehci-q-iram.c @@ -0,0 +1,1345 @@ +/* + * Copyright (C) 2001-2004 by David Brownell + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License as published by the + * Free Software Foundation; either version 2 of the License, or (at your + * option) any later version. + * + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software Foundation, + * Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + */ +#undef EHCI_NO_ERR_COUNT +static size_t g_iram_size = IRAM_TD_SIZE; + +/* this file is part of ehci-hcd.c */ + +/*-------------------------------------------------------------------------*/ + +/* + * EHCI hardware queue manipulation ... the core. QH/QTD manipulation. + * + * Control, bulk, and interrupt traffic all use "qh" lists. They list "qtd" + * entries describing USB transactions, max 16-20kB/entry (with 4kB-aligned + * buffers needed for the larger number). We use one QH per endpoint, queue + * multiple urbs (all three types) per endpoint. URBs may need several qtds. + * + * ISO traffic uses "ISO TD" (itd, and sitd) records, and (along with + * interrupts) needs careful scheduling. Performance improvements can be + * an ongoing challenge. That's in "ehci-sched.c". + * + * USB 1.1 devices are handled (a) by "companion" OHCI or UHCI root hubs, + * or otherwise through transaction translators (TTs) in USB 2.0 hubs using + * (b) special fields in qh entries or (c) split iso entries. TTs will + * buffer low/full speed data so the host collects it at high speed. + */ + +/*-------------------------------------------------------------------------*/ +/* fill a qtd, returning how much of the buffer we were able to queue up */ +static int qtd_fill(struct ehci_hcd *ehci, struct ehci_qtd *qtd, dma_addr_t buf, + size_t len, int token, int maxpacket) +{ + int i, count; + u64 addr = buf; + struct urb *urb = qtd->urb; + + if (usb_pipebulk(urb->pipe) && + (address_to_buffer(ehci, usb_pipedevice(urb->pipe)) != 2)) { + urb->use_iram = 1; + qtd->buffer_offset = (size_t) (buf - urb->transfer_dma); + token |= QTD_IOC; + if (usb_pipeout(urb->pipe)) { + addr = ehci->iram_buffer[address_to_buffer(ehci, + usb_pipedevice(urb->pipe))]; + } else if (usb_pipein(urb->pipe)) { + addr = ehci->iram_buffer[address_to_buffer(ehci, + usb_pipedevice(urb->pipe))] + + g_iram_size; + } + } else { + urb->use_iram = 0; + addr = buf; + } + len = min(g_iram_size, len); + + /* one buffer entry per 4K ... first might be short or unaligned */ + qtd->hw_buf[0] = cpu_to_hc32(ehci, (u32) addr); + qtd->hw_buf_hi[0] = cpu_to_hc32(ehci, (u32) (addr >> 32)); + count = 0x1000 - (buf & 0x0fff); /* rest of that page */ + if (likely(len < count)) /* ... iff needed */ + count = len; + else { + buf += 0x1000; + buf &= ~0x0fff; + + /* per-qtd limit: from 16K to 20K (best alignment) */ + for (i = 1; count < len && i < 5; i++) { + addr = buf; + qtd->hw_buf[i] = cpu_to_hc32(ehci, (u32) addr); + qtd->hw_buf_hi[i] = + cpu_to_hc32(ehci, (u32) (addr >> 32)); + buf += 0x1000; + if ((count + 0x1000) < len) + count += 0x1000; + else + count = len; + } + + /* short packets may only terminate transfers */ + if (count != len) + count -= (count % maxpacket); + } + qtd->hw_token = cpu_to_hc32(ehci, (count << 16) | token); + qtd->length = count; + + return count; +} + +/*-------------------------------------------------------------------------*/ + +static inline void +qh_update(struct ehci_hcd *ehci, struct ehci_qh *qh, struct ehci_qtd *qtd) +{ + /* writes to an active overlay are unsafe */ + BUG_ON(qh->qh_state != QH_STATE_IDLE); + + qh->hw_qtd_next = QTD_NEXT(ehci, qtd->qtd_dma); + qh->hw_alt_next = EHCI_LIST_END(ehci); + + /* Except for control endpoints, we make hardware maintain data + * toggle (like OHCI) ... here (re)initialize the toggle in the QH, + * and set the pseudo-toggle in udev. Only usb_clear_halt() will + * ever clear it. + */ + if (!(qh->hw_info1 & cpu_to_hc32(ehci, 1 << 14))) { + unsigned is_out, epnum; + + is_out = !(qtd->hw_token & cpu_to_hc32(ehci, 1 << 8)); + epnum = (hc32_to_cpup(ehci, &qh->hw_info1) >> 8) & 0x0f; + if (unlikely(!usb_gettoggle(qh->dev, epnum, is_out))) { + qh->hw_token &= ~cpu_to_hc32(ehci, QTD_TOGGLE); + usb_settoggle(qh->dev, epnum, is_out, 1); + } + } + + /* HC must see latest qtd and qh data before we clear ACTIVE+HALT */ + wmb(); + qh->hw_token &= cpu_to_hc32(ehci, QTD_TOGGLE | QTD_STS_PING); +} + +/* if it weren't for a common silicon quirk (writing the dummy into the qh + * overlay, so qh->hw_token wrongly becomes inactive/halted), only fault + * recovery (including urb dequeue) would need software changes to a QH... + */ +static void qh_refresh(struct ehci_hcd *ehci, struct ehci_qh *qh) +{ + struct ehci_qtd *qtd; + + if (list_empty(&qh->qtd_list)) + qtd = qh->dummy; + else { + qtd = list_entry(qh->qtd_list.next, struct ehci_qtd, qtd_list); + /* first qtd may already be partially processed */ + if (cpu_to_hc32(ehci, qtd->qtd_dma) == qh->hw_current) + qtd = NULL; + } + + if (qtd) + qh_update(ehci, qh, qtd); +} + +/*-------------------------------------------------------------------------*/ + +static int qtd_copy_status(struct ehci_hcd *ehci, + struct urb *urb, size_t length, u32 token) +{ + int status = -EINPROGRESS; + + /* count IN/OUT bytes, not SETUP (even short packets) */ + if (likely(QTD_PID(token) != 2)) + urb->actual_length += length - QTD_LENGTH(token); + + /* don't modify error codes */ + if (unlikely(urb->unlinked)) + return status; + + /* force cleanup after short read; not always an error */ + if (unlikely(IS_SHORT_READ(token))) + status = -EREMOTEIO; + + /* serious "can't proceed" faults reported by the hardware */ + if (token & QTD_STS_HALT) { + if (token & QTD_STS_BABBLE) { + /* FIXME "must" disable babbling device's port too */ + status = -EOVERFLOW; + } else if (token & QTD_STS_MMF) { + /* fs/ls interrupt xfer missed the complete-split */ + status = -EPROTO; + } else if (token & QTD_STS_DBE) { + status = (QTD_PID(token) == 1) /* IN ? */ + ? -ENOSR /* hc couldn't read data */ + : -ECOMM; /* hc couldn't write data */ + } else if (token & QTD_STS_XACT) { + /* timeout, bad crc, wrong PID, etc; retried */ + if (QTD_CERR(token)) + status = -EPIPE; + else { + ehci_dbg(ehci, "devpath %s ep%d%s 3strikes\n", + urb->dev->devpath, + usb_pipeendpoint(urb->pipe), + usb_pipein(urb->pipe) ? "in" : "out"); + status = -EPROTO; + } + /* CERR nonzero + no errors + halt --> stall */ + } else if (QTD_CERR(token)) + status = -EPIPE; + else /* unknown */ + status = -EPROTO; + + ehci_vdbg(ehci, + "dev%d ep%d%s qtd token %08x --> status %d\n", + usb_pipedevice(urb->pipe), + usb_pipeendpoint(urb->pipe), + usb_pipein(urb->pipe) ? "in" : "out", token, status); + + /* if async CSPLIT failed, try cleaning out the TT buffer */ + if (status != -EPIPE && urb->dev->tt && !usb_pipeint(urb->pipe) + && ((token & QTD_STS_MMF) != 0 || QTD_CERR(token) == 0) + && (!ehci_is_TDI(ehci) + || urb->dev->tt->hub != + ehci_to_hcd(ehci)->self.root_hub)) { +#ifdef DEBUG + struct usb_device *tt = urb->dev->tt->hub; + dev_dbg(&tt->dev, + "clear tt buffer port %d, a%d ep%d t%08x\n", + urb->dev->ttport, urb->dev->devnum, + usb_pipeendpoint(urb->pipe), token); +#endif /* DEBUG */ + /* REVISIT ARC-derived cores don't clear the root + * hub TT buffer in this way... + */ + usb_hub_tt_clear_buffer(urb->dev, urb->pipe); + } + } + + return status; +} + +static void +ehci_urb_done(struct ehci_hcd *ehci, struct urb *urb, int status) +__releases(ehci->lock) __acquires(ehci->lock) +{ + if (likely(urb->hcpriv != NULL)) { + struct ehci_qh *qh = (struct ehci_qh *)urb->hcpriv; + + /* S-mask in a QH means it's an interrupt urb */ + if ((qh->hw_info2 & cpu_to_hc32(ehci, QH_SMASK)) != 0) { + + /* ... update hc-wide periodic stats (for usbfs) */ + ehci_to_hcd(ehci)->self.bandwidth_int_reqs--; + } + qh_put(qh); + } + + if (unlikely(urb->unlinked)) { + COUNT(ehci->stats.unlink); + } else { + /* report non-error and short read status as zero */ + if (status == -EINPROGRESS || status == -EREMOTEIO) + status = 0; + COUNT(ehci->stats.complete); + } + +#ifdef EHCI_URB_TRACE + ehci_dbg(ehci, + "%s %s urb %p ep%d%s status %d len %d/%d\n", + __func__, urb->dev->devpath, urb, + usb_pipeendpoint(urb->pipe), + usb_pipein(urb->pipe) ? "in" : "out", + status, urb->actual_length, urb->transfer_buffer_length); +#endif + + /* complete() can reenter this HCD */ + usb_hcd_unlink_urb_from_ep(ehci_to_hcd(ehci), urb); + spin_unlock(&ehci->lock); + usb_hcd_giveback_urb(ehci_to_hcd(ehci), urb, status); + spin_lock(&ehci->lock); +} + +static void start_unlink_async(struct ehci_hcd *ehci, struct ehci_qh *qh); +static void unlink_async(struct ehci_hcd *ehci, struct ehci_qh *qh); + +static void intr_deschedule(struct ehci_hcd *ehci, struct ehci_qh *qh); +static int qh_schedule(struct ehci_hcd *ehci, struct ehci_qh *qh); + +/* + * Process and free completed qtds for a qh, returning URBs to drivers. + * Chases up to qh->hw_current. Returns number of completions called, + * indicating how much "real" work we did. + */ +static unsigned qh_completions(struct ehci_hcd *ehci, struct ehci_qh *qh) +{ + struct ehci_qtd *last = NULL, *end = qh->dummy; + struct list_head *entry, *tmp; + int last_status = -EINPROGRESS; + int stopped; + unsigned count = 0; + u8 state; + __le32 halt = HALT_BIT(ehci); + __hc32 temp_hw_qtd_next = 0; + + if (unlikely(list_empty(&qh->qtd_list))) + return count; + + /* completions (or tasks on other cpus) must never clobber HALT + * till we've gone through and cleaned everything up, even when + * they add urbs to this qh's queue or mark them for unlinking. + * + * NOTE: unlinking expects to be done in queue order. + */ + state = qh->qh_state; + qh->qh_state = QH_STATE_COMPLETING; + stopped = (state == QH_STATE_IDLE); + + /* remove de-activated QTDs from front of queue. + * after faults (including short reads), cleanup this urb + * then let the queue advance. + * if queue is stopped, handles unlinks. + */ + list_for_each_safe(entry, tmp, &qh->qtd_list) { + struct ehci_qtd *qtd; + struct urb *urb; + struct ehci_qtd *qtd2; + struct urb *urb2; + + u32 token = 0; + + qtd = list_entry(entry, struct ehci_qtd, qtd_list); + urb = qtd->urb; + + /* clean up any state from previous QTD ... */ + if (last) { + if (likely(last->urb != urb)) { + ehci_urb_done(ehci, last->urb, last_status); + count++; + last_status = -EINPROGRESS; + } + ehci_qtd_free(ehci, last); + last = NULL; + } + + /* ignore urbs submitted during completions we reported */ + if (qtd == end) + break; + + /* hardware copies qtd out of qh overlay */ + rmb(); + token = hc32_to_cpu(ehci, qtd->hw_token); + + /* always clean up qtds the hc de-activated */ + if ((token & QTD_STS_ACTIVE) == 0) { + + /* on STALL, error, and short reads this urb must + * complete and all its qtds must be recycled. + */ + if ((token & QTD_STS_HALT) != 0) { + stopped = 1; + + /* magic dummy for some short reads; qh won't advance. + * that silicon quirk can kick in with this dummy too. + * + * other short reads won't stop the queue, including + * control transfers (status stage handles that) or + * most other single-qtd reads ... the queue stops if + * URB_SHORT_NOT_OK was set so the driver submitting + * the urbs could clean it up. + */ + } else if (IS_SHORT_READ(token) + && !(qtd->hw_alt_next & EHCI_LIST_END(ehci))) { + if (urb->use_iram && usb_pipein(urb->pipe)) { + if (urb->transfer_buffer == NULL) { + memcpy(phys_to_virt + (urb->transfer_dma) + + qtd->buffer_offset, + ehci-> + iram_buffer_v + [address_to_buffer + (ehci, + usb_pipedevice(urb-> + pipe))] + + g_iram_size, + min(g_iram_size, + qtd->length)); + } else { + memcpy(urb->transfer_buffer + + qtd->buffer_offset, + ehci-> + iram_buffer_v + [address_to_buffer + (ehci, + usb_pipedevice(urb-> + pipe))] + + g_iram_size, + min(g_iram_size, + qtd->length)); + } + } + stopped = 1; + goto halt; + } else if (urb->use_iram && (!qtd->last_one) + && usb_pipeout(urb->pipe)) { + ehci-> + iram_in_use[address_to_buffer + (ehci, + usb_pipedevice(urb->pipe))] = + 1; + qtd2 = + list_entry(tmp, struct ehci_qtd, qtd_list); + if (urb->transfer_buffer == NULL) { + memcpy(ehci-> + iram_buffer_v[address_to_buffer + (ehci, + usb_pipedevice + (urb->pipe))], + phys_to_virt(urb->transfer_dma) + + qtd->buffer_offset + qtd->length, + min(g_iram_size, qtd2->length)); + } else { + memcpy(ehci-> + iram_buffer_v[address_to_buffer + (ehci, + usb_pipedevice + (urb->pipe))], + urb->transfer_buffer + + qtd->buffer_offset + qtd->length, + min(g_iram_size, qtd2->length)); + } + temp_hw_qtd_next = + QTD_NEXT(ehci, qtd->hw_next) & 0xFFFFFFFE; + } else if (urb->use_iram && (qtd->last_one) + && usb_pipeout(urb->pipe)) { + urb->use_iram = 0; + qtd2 = + list_entry(tmp, struct ehci_qtd, qtd_list); + if (tmp != &qh->qtd_list) { + urb2 = qtd2->urb; + if (urb2 && urb2->use_iram == 1) { + ehci-> + iram_in_use + [address_to_buffer + (ehci, + usb_pipedevice(urb-> + pipe))] = + 1; + if (urb2->transfer_buffer == + NULL) { + memcpy(ehci-> + iram_buffer_v + [address_to_buffer + (ehci, + usb_pipedevice + (urb->pipe))], + phys_to_virt + (urb2-> + transfer_dma), + min(g_iram_size, + qtd2-> + length)); + } else { + memcpy(ehci-> + iram_buffer_v + [address_to_buffer + (ehci, + usb_pipedevice + (urb->pipe))], + urb2-> + transfer_buffer, + min(g_iram_size, + qtd2-> + length)); + } + } else { + ehci-> + iram_in_use + [address_to_buffer + (ehci, + usb_pipedevice(urb-> + pipe))] = + 0; + } + } else { + ehci-> + iram_in_use[address_to_buffer + (ehci, + usb_pipedevice(urb-> + pipe))] + = 0; + } + temp_hw_qtd_next = + QTD_NEXT(ehci, qtd->hw_next) & 0xFFFFFFFE; + } else if (urb->use_iram && usb_pipein(urb->pipe)) { + if (urb->transfer_buffer == NULL) { + memcpy(phys_to_virt(urb->transfer_dma) + + qtd->buffer_offset, + ehci-> + iram_buffer_v[address_to_buffer + (ehci, + usb_pipedevice + (urb->pipe))] + + g_iram_size, min(g_iram_size, + qtd->length)); + } else { + memcpy(urb->transfer_buffer + + qtd->buffer_offset, + ehci-> + iram_buffer_v[address_to_buffer + (ehci, + usb_pipedevice + (urb->pipe))] + + g_iram_size, min(g_iram_size, + qtd->length)); + } + temp_hw_qtd_next = + QTD_NEXT(ehci, qtd->hw_next) & 0xFFFFFFFE; + } + /* stop scanning when we reach qtds the hc is using */ + } else if (likely(!stopped + && HC_IS_RUNNING(ehci_to_hcd(ehci)->state))) { + break; + + /* scan the whole queue for unlinks whenever it stops */ + } else { + stopped = 1; + + /* cancel everything if we halt, suspend, etc */ + if (!HC_IS_RUNNING(ehci_to_hcd(ehci)->state)) + last_status = -ESHUTDOWN; + + /* this qtd is active; skip it unless a previous qtd + * for its urb faulted, or its urb was canceled. + */ + else if (last_status == -EINPROGRESS && !urb->unlinked) + continue; + + /* qh unlinked; token in overlay may be most current */ + if (state == QH_STATE_IDLE + && cpu_to_hc32(ehci, qtd->qtd_dma) + == qh->hw_current) + token = hc32_to_cpu(ehci, qh->hw_token); + + /* qh unlinked; token in overlay may be most current */ + if (state == QH_STATE_IDLE + && cpu_to_hc32(ehci, qtd->qtd_dma) + == qh->hw_current) + token = hc32_to_cpu(ehci, qh->hw_token); + + /* force halt for unlinked or blocked qh, so we'll + * patch the qh later and so that completions can't + * activate it while we "know" it's stopped. + */ + if ((halt & qh->hw_token) == 0) { +halt: + qh->hw_token |= halt; + wmb(); + } + } + + /* unless we already know the urb's status, collect qtd status + * and update count of bytes transferred. in common short read + * cases with only one data qtd (including control transfers), + * queue processing won't halt. but with two or more qtds (for + * example, with a 32 KB transfer), when the first qtd gets a + * short read the second must be removed by hand. + */ + if (last_status == -EINPROGRESS) { + last_status = qtd_copy_status(ehci, urb, + qtd->length, token); + if (last_status == -EREMOTEIO + && (qtd->hw_alt_next + & EHCI_LIST_END(ehci))) + last_status = -EINPROGRESS; + } + + /* if we're removing something not at the queue head, + * patch the hardware queue pointer. + */ + + if (stopped && qtd->qtd_list.prev != &qh->qtd_list) { + last = list_entry(qtd->qtd_list.prev, + struct ehci_qtd, qtd_list); + last->hw_next = qtd->hw_next; + } + +/* remove qtd; it's recycled after possible urb completion */ + list_del(&qtd->qtd_list); + last = qtd; + } + + /* last urb's completion might still need calling */ + if (likely(last != NULL)) { + ehci_urb_done(ehci, last->urb, last_status); + count++; + ehci_qtd_free(ehci, last); + } + + /* restore original state; caller must unlink or relink */ + qh->qh_state = state; + + /* be sure the hardware's done with the qh before refreshing + * it after fault cleanup, or recovering from silicon wrongly + * overlaying the dummy qtd (which reduces DMA chatter). + */ + if ((stopped != 0) || (qh->hw_qtd_next == EHCI_LIST_END(ehci)) + && (temp_hw_qtd_next == 0)) { + switch (state) { + case QH_STATE_IDLE: + qh_refresh(ehci, qh); + break; + case QH_STATE_LINKED: + /* We won't refresh a QH that's linked (after the HC + * stopped the queue). That avoids a race: + * - HC reads first part of QH; + * - CPU updates that first part and the token; + * - HC reads rest of that QH, including token + * Result: HC gets an inconsistent image, and then + * DMAs to/from the wrong memory (corrupting it). + * + * That should be rare for interrupt transfers, + * except maybe high bandwidth ... + */ + if ((cpu_to_hc32(ehci, QH_SMASK) + & qh->hw_info2) != 0) { + intr_deschedule(ehci, qh); + (void)qh_schedule(ehci, qh); + } else + unlink_async(ehci, qh); + break; + /* otherwise, unlink already started */ + } + } + if (temp_hw_qtd_next) + qh->hw_qtd_next = temp_hw_qtd_next; + + return count; +} + +/*-------------------------------------------------------------------------*/ + +/* high bandwidth multiplier, as encoded in highspeed endpoint descriptors */ +#define hb_mult(wMaxPacketSize) (1 + (((wMaxPacketSize) >> 11) & 0x03)) +/* ... and packet size, for any kind of endpoint descriptor */ +#define max_packet(wMaxPacketSize) ((wMaxPacketSize) & 0x07ff) + +/* + * reverse of qh_urb_transaction: free a list of TDs. + * used for cleanup after errors, before HC sees an URB's TDs. + */ +static void qtd_list_free(struct ehci_hcd *ehci, + struct urb *urb, struct list_head *qtd_list) +{ + struct list_head *entry, *temp; + + list_for_each_safe(entry, temp, qtd_list) { + struct ehci_qtd *qtd; + + qtd = list_entry(entry, struct ehci_qtd, qtd_list); + list_del(&qtd->qtd_list); + ehci_qtd_free(ehci, qtd); + } +} + +/* + * create a list of filled qtds for this URB; won't link into qh. + */ +static struct list_head *qh_urb_transaction(struct ehci_hcd *ehci, + struct urb *urb, + struct list_head *head, gfp_t flags) +{ + struct ehci_qtd *qtd, *qtd_prev; + dma_addr_t buf; + int len, maxpacket; + int is_input; + u32 token; + + /* + * URBs map to sequences of QTDs: one logical transaction + */ + qtd = ehci_qtd_alloc(ehci, flags); + if (unlikely(!qtd)) + return NULL; + list_add_tail(&qtd->qtd_list, head); + qtd->urb = urb; + + token = QTD_STS_ACTIVE; + token |= (EHCI_TUNE_CERR << 10); + /* for split transactions, SplitXState initialized to zero */ + + len = urb->transfer_buffer_length; + is_input = usb_pipein(urb->pipe); + if (usb_pipecontrol(urb->pipe)) { + /* SETUP pid */ + qtd_fill(ehci, qtd, urb->setup_dma, + sizeof(struct usb_ctrlrequest), + token | (2 /* "setup" */ << 8), 8); + + /* ... and always at least one more pid */ + token ^= QTD_TOGGLE; + qtd_prev = qtd; + qtd = ehci_qtd_alloc(ehci, flags); + if (unlikely(!qtd)) + goto cleanup; + qtd->urb = urb; + qtd_prev->hw_next = QTD_NEXT(ehci, qtd->qtd_dma); + list_add_tail(&qtd->qtd_list, head); + + /* for zero length DATA stages, STATUS is always IN */ + if (len == 0) + token |= (1 /* "in" */ << 8); + } + + /* + * data transfer stage: buffer setup + */ + buf = urb->transfer_dma; + + if (is_input) + token |= (1 /* "in" */ << 8); + /* else it's already initted to "out" pid (0 << 8) */ + + maxpacket = max_packet(usb_maxpacket(urb->dev, urb->pipe, !is_input)); + + /* + * buffer gets wrapped in one or more qtds; + * last one may be "short" (including zero len) + * and may serve as a control status ack + */ + for (;;) { + int this_qtd_len; + this_qtd_len = qtd_fill(ehci, qtd, buf, len, token, maxpacket); + if (urb->use_iram && (!qtd->buffer_offset) + && usb_pipeout(urb->pipe) + && (ehci-> + iram_in_use[address_to_buffer + (ehci, usb_pipedevice(urb->pipe))] == 0)) { + ehci-> + iram_in_use[address_to_buffer + (ehci, usb_pipedevice(urb->pipe))] = 1; + if (urb->transfer_buffer == NULL) { + memcpy(ehci-> + iram_buffer_v[address_to_buffer + (ehci, + usb_pipedevice(urb-> + pipe))], + phys_to_virt(urb->transfer_dma), + min((int)g_iram_size, len)); + } else { + memcpy(ehci-> + iram_buffer_v[address_to_buffer + (ehci, + usb_pipedevice(urb-> + pipe))], + urb->transfer_buffer, + min((int)g_iram_size, len)); + } + } + len -= this_qtd_len; + buf += this_qtd_len; + + /* + * short reads advance to a "magic" dummy instead of the next + * qtd ... that forces the queue to stop, for manual cleanup. + * (this will usually be overridden later.) + */ + if (is_input) + qtd->hw_alt_next = ehci->async->hw_alt_next; + + /* qh makes control packets use qtd toggle; maybe switch it */ + if ((maxpacket & (this_qtd_len + (maxpacket - 1))) == 0) + token ^= QTD_TOGGLE; + + if (likely(len <= 0)) { + qtd->last_one = 1; + break; + } + qtd_prev = qtd; + qtd = ehci_qtd_alloc(ehci, flags); + if (unlikely(!qtd)) + goto cleanup; + qtd->urb = urb; + if (urb->use_iram) + qtd_prev->hw_next = QTD_NEXT(ehci, qtd->qtd_dma) | 0x1; + else + qtd_prev->hw_next = QTD_NEXT(ehci, qtd->qtd_dma); + + list_add_tail(&qtd->qtd_list, head); + } + + /* + * unless the caller requires manual cleanup after short reads, + * have the alt_next mechanism keep the queue running after the + * last data qtd (the only one, for control and most other cases). + */ + if (likely((urb->transfer_flags & URB_SHORT_NOT_OK) == 0 + || usb_pipecontrol(urb->pipe))) + qtd->hw_alt_next = EHCI_LIST_END(ehci); + + /* + * control requests may need a terminating data "status" ack; + * bulk ones may need a terminating short packet (zero length). + */ + if (likely(urb->transfer_buffer_length != 0)) { + int one_more = 0; + + if (usb_pipecontrol(urb->pipe)) { + one_more = 1; + token ^= 0x0100; /* "in" <--> "out" */ + token |= QTD_TOGGLE; /* force DATA1 */ + } else if (usb_pipebulk(urb->pipe) + && (urb->transfer_flags & URB_ZERO_PACKET) + && !(urb->transfer_buffer_length % maxpacket)) + one_more = 1; + if (one_more) { + qtd_prev = qtd; + qtd = ehci_qtd_alloc(ehci, flags); + if (unlikely(!qtd)) + goto cleanup; + qtd->urb = urb; + qtd_prev->hw_next = QTD_NEXT(ehci, qtd->qtd_dma); + list_add_tail(&qtd->qtd_list, head); + + /* never any data in such packets */ + qtd_fill(ehci, qtd, 0, 0, token, 0); + } + } + + /* by default, enable interrupt on urb completion */ + if (likely(!(urb->transfer_flags & URB_NO_INTERRUPT))) + qtd->hw_token |= cpu_to_hc32(ehci, QTD_IOC); + return head; + +cleanup: + qtd_list_free(ehci, urb, head); + return NULL; +} + +/*-------------------------------------------------------------------------*/ + +/* Would be best to create all qh's from config descriptors, + * when each interface/altsetting is established. Unlink + * any previous qh and cancel its urbs first; endpoints are + * implicitly reset then (data toggle too). + * That'd mean updating how usbcore talks to HCDs. (2.7?) + */ + +/* + * Each QH holds a qtd list; a QH is used for everything except iso. + * + * For interrupt urbs, the scheduler must set the microframe scheduling + * mask(s) each time the QH gets scheduled. For highspeed, that's + * just one microframe in the s-mask. For split interrupt transactions + * there are additional complications: c-mask, maybe FSTNs. + */ +static struct ehci_qh *qh_make(struct ehci_hcd *ehci, + struct urb *urb, gfp_t flags) +{ + struct ehci_qh *qh = ehci_qh_alloc(ehci, flags); + u32 info1 = 0, info2 = 0; + int is_input, type; + int maxp = 0; + struct usb_tt *tt = urb->dev->tt; + + if (!qh) + return qh; + + /* + * init endpoint/device data for this QH + */ + info1 |= usb_pipeendpoint(urb->pipe) << 8; + info1 |= usb_pipedevice(urb->pipe) << 0; + + is_input = usb_pipein(urb->pipe); + type = usb_pipetype(urb->pipe); + maxp = usb_maxpacket(urb->dev, urb->pipe, !is_input); + + /* 1024 byte maxpacket is a hardware ceiling. High bandwidth + * acts like up to 3KB, but is built from smaller packets. + */ + if (max_packet(maxp) > 1024) { + ehci_dbg(ehci, "bogus qh maxpacket %d\n", max_packet(maxp)); + goto done; + } + + /* Compute interrupt scheduling parameters just once, and save. + * - allowing for high bandwidth, how many nsec/uframe are used? + * - split transactions need a second CSPLIT uframe; same question + * - splits also need a schedule gap (for full/low speed I/O) + * - qh has a polling interval + * + * For control/bulk requests, the HC or TT handles these. + */ + if (type == PIPE_INTERRUPT) { + qh->usecs = + NS_TO_US(usb_calc_bus_time + (USB_SPEED_HIGH, is_input, 0, + hb_mult(maxp) * max_packet(maxp))); + qh->start = NO_FRAME; + + if (urb->dev->speed == USB_SPEED_HIGH) { + qh->c_usecs = 0; + qh->gap_uf = 0; + + qh->period = urb->interval >> 3; + if (qh->period == 0 && urb->interval != 1) { + /* NOTE interval 2 or 4 uframes could work. + * But interval 1 scheduling is simpler, and + * includes high bandwidth. + */ + dbg("intr period %d uframes, NYET!", + urb->interval); + goto done; + } + } else { + int think_time; + + /* gap is f(FS/LS transfer times) */ + qh->gap_uf = 1 + usb_calc_bus_time(urb->dev->speed, + is_input, 0, + maxp) / (125 * 1000); + + /* FIXME this just approximates SPLIT/CSPLIT times */ + if (is_input) { + qh->c_usecs = qh->usecs + HS_USECS(0); + qh->usecs = HS_USECS(1); + } else { + qh->usecs += HS_USECS(1); + qh->c_usecs = HS_USECS(0); + } + + think_time = tt ? tt->think_time : 0; + qh->tt_usecs = NS_TO_US(think_time + + usb_calc_bus_time(urb->dev-> + speed, + is_input, 0, + max_packet + (maxp))); + qh->period = urb->interval; + } + } + + /* support for tt scheduling, and access to toggles */ + qh->dev = urb->dev; + + /* using TT? */ + switch (urb->dev->speed) { + case USB_SPEED_LOW: + info1 |= (1 << 12); /* EPS "low" */ + /* FALL THROUGH */ + + case USB_SPEED_FULL: + /* EPS 0 means "full" */ + if (type != PIPE_INTERRUPT) + info1 |= (EHCI_TUNE_RL_TT << 28); + if (type == PIPE_CONTROL) { + info1 |= (1 << 27); /* for TT */ + info1 |= 1 << 14; /* toggle from qtd */ + } + info1 |= maxp << 16; + + info2 |= (EHCI_TUNE_MULT_TT << 30); + + /* Some Freescale processors have an erratum in which the + * port number in the queue head was 0..N-1 instead of 1..N. + */ + if (ehci_has_fsl_portno_bug(ehci)) + info2 |= (urb->dev->ttport - 1) << 23; + else + info2 |= urb->dev->ttport << 23; + + /* set the address of the TT; for TDI's integrated + * root hub tt, leave it zeroed. + */ + if (tt && tt->hub != ehci_to_hcd(ehci)->self.root_hub) + info2 |= tt->hub->devnum << 16; + + /* NOTE: if (PIPE_INTERRUPT) { scheduler sets c-mask } */ + + break; + + case USB_SPEED_HIGH: /* no TT involved */ + info1 |= (2 << 12); /* EPS "high" */ + if (type == PIPE_CONTROL) { + info1 |= (EHCI_TUNE_RL_HS << 28); + info1 |= 64 << 16; /* usb2 fixed maxpacket */ + info1 |= 1 << 14; /* toggle from qtd */ + info2 |= (EHCI_TUNE_MULT_HS << 30); + } else if (type == PIPE_BULK) { + info1 |= (EHCI_TUNE_RL_HS << 28); + /* The USB spec says that high speed bulk endpoints + * always use 512 byte maxpacket. But some device + * vendors decided to ignore that, and MSFT is happy + * to help them do so. So now people expect to use + * such nonconformant devices with Linux too; sigh. + */ + info1 |= max_packet(maxp) << 16; + info2 |= (EHCI_TUNE_MULT_HS << 30); + use_buffer(ehci, usb_pipedevice(urb->pipe)); + } else { /* PIPE_INTERRUPT */ + info1 |= max_packet(maxp) << 16; + info2 |= hb_mult(maxp) << 30; + } + break; + default: + dbg("bogus dev %p speed %d", urb->dev, urb->dev->speed); +done: + qh_put(qh); + return NULL; + } + + /* NOTE: if (PIPE_INTERRUPT) { scheduler sets s-mask } */ + + /* init as live, toggle clear, advance to dummy */ + qh->qh_state = QH_STATE_IDLE; + qh->hw_info1 = cpu_to_hc32(ehci, info1); + qh->hw_info2 = cpu_to_hc32(ehci, info2); + usb_settoggle(urb->dev, usb_pipeendpoint(urb->pipe), !is_input, 1); + qh_refresh(ehci, qh); + return qh; +} + +/*-------------------------------------------------------------------------*/ + +/* move qh (and its qtds) onto async queue; maybe enable queue. */ + +static void qh_link_async(struct ehci_hcd *ehci, struct ehci_qh *qh) +{ + __hc32 dma = QH_NEXT(ehci, qh->qh_dma); + struct ehci_qh *head; + + /* (re)start the async schedule? */ + head = ehci->async; + timer_action_done(ehci, TIMER_ASYNC_OFF); + if (!head->qh_next.qh) { + u32 cmd = ehci_readl(ehci, &ehci->regs->command); + + if (!(cmd & CMD_ASE)) { + /* in case a clear of CMD_ASE didn't take yet */ + (void)handshake(ehci, &ehci->regs->status, + STS_ASS, 0, 150); + cmd |= CMD_ASE | CMD_RUN; + ehci_writel(ehci, cmd, &ehci->regs->command); + ehci_to_hcd(ehci)->state = HC_STATE_RUNNING; + /* posted write need not be known to HC yet ... */ + } + } + + /* clear halt and/or toggle; and maybe recover from silicon quirk */ + if (qh->qh_state == QH_STATE_IDLE) + qh_refresh(ehci, qh); + + /* splice right after start */ + qh->qh_next = head->qh_next; + qh->hw_next = head->hw_next; + wmb(); + + head->qh_next.qh = qh; + head->hw_next = dma; + + qh->qh_state = QH_STATE_LINKED; + /* qtd completions reported later by interrupt */ +} + +/*-------------------------------------------------------------------------*/ + +/* + * For control/bulk/interrupt, return QH with these TDs appended. + * Allocates and initializes the QH if necessary. + * Returns null if it can't allocate a QH it needs to. + * If the QH has TDs (urbs) already, that's great. + */ +static struct ehci_qh *qh_append_tds(struct ehci_hcd *ehci, + struct urb *urb, + struct list_head *qtd_list, + int epnum, void **ptr) +{ + struct ehci_qh *qh = NULL; + __hc32 qh_addr_mask = cpu_to_hc32(ehci, 0x7f); + + qh = (struct ehci_qh *)*ptr; + if (unlikely(qh == NULL)) { + /* can't sleep here, we have ehci->lock... */ + qh = qh_make(ehci, urb, GFP_ATOMIC); + *ptr = qh; + } + if (likely(qh != NULL)) { + struct ehci_qtd *qtd; + + if (unlikely(list_empty(qtd_list))) + qtd = NULL; + else + qtd = list_entry(qtd_list->next, struct ehci_qtd, + qtd_list); + + /* control qh may need patching ... */ + if (unlikely(epnum == 0)) { + + /* usb_reset_device() briefly reverts to address 0 */ + if (usb_pipedevice(urb->pipe) == 0) + qh->hw_info1 &= ~qh_addr_mask; + } + + /* just one way to queue requests: swap with the dummy qtd. + * only hc or qh_refresh() ever modify the overlay. + */ + if (likely(qtd != NULL)) { + struct ehci_qtd *dummy; + dma_addr_t dma; + __hc32 token; + + /* to avoid racing the HC, use the dummy td instead of + * the first td of our list (becomes new dummy). both + * tds stay deactivated until we're done, when the + * HC is allowed to fetch the old dummy (4.10.2). + */ + token = qtd->hw_token; + qtd->hw_token = HALT_BIT(ehci); + wmb(); + dummy = qh->dummy; + + dma = dummy->qtd_dma; + *dummy = *qtd; + dummy->qtd_dma = dma; + + list_del(&qtd->qtd_list); + list_add(&dummy->qtd_list, qtd_list); + __list_splice(qtd_list, qh->qtd_list.prev); + + ehci_qtd_init(ehci, qtd, qtd->qtd_dma); + qh->dummy = qtd; + + /* hc must see the new dummy at list end */ + dma = qtd->qtd_dma; + qtd = list_entry(qh->qtd_list.prev, + struct ehci_qtd, qtd_list); + if (urb->use_iram) + qtd->hw_next = QTD_NEXT(ehci, dma) | 0x1; + else + qtd->hw_next = QTD_NEXT(ehci, dma); + + /* let the hc process these next qtds */ + wmb(); + dummy->hw_token = token; + + urb->hcpriv = qh_get(qh); + } + } + return qh; +} + +/*-------------------------------------------------------------------------*/ + +static int +submit_async(struct ehci_hcd *ehci, + struct urb *urb, struct list_head *qtd_list, gfp_t mem_flags) +{ + struct ehci_qtd *qtd; + int epnum; + unsigned long flags; + struct ehci_qh *qh = NULL; + int rc; + + qtd = list_entry(qtd_list->next, struct ehci_qtd, qtd_list); + epnum = urb->ep->desc.bEndpointAddress; + +#ifdef EHCI_URB_TRACE + ehci_dbg(ehci, + "%s %s urb %p ep%d%s len %d, qtd %p [qh %p]\n", + __func__, urb->dev->devpath, urb, + epnum & 0x0f, (epnum & USB_DIR_IN) ? "in" : "out", + urb->transfer_buffer_length, qtd, urb->ep->hcpriv); +#endif + + spin_lock_irqsave(&ehci->lock, flags); + if (unlikely(!test_bit(HCD_FLAG_HW_ACCESSIBLE, + &ehci_to_hcd(ehci)->flags))) { + rc = -ESHUTDOWN; + goto done; + } + rc = usb_hcd_link_urb_to_ep(ehci_to_hcd(ehci), urb); + if (unlikely(rc)) + goto done; + + qh = qh_append_tds(ehci, urb, qtd_list, epnum, &urb->ep->hcpriv); + if (unlikely(qh == NULL)) { + usb_hcd_unlink_urb_from_ep(ehci_to_hcd(ehci), urb); + rc = -ENOMEM; + goto done; + } + + /* Control/bulk operations through TTs don't need scheduling, + * the HC and TT handle it when the TT has a buffer ready. + */ + if (likely(qh->qh_state == QH_STATE_IDLE)) + qh_link_async(ehci, qh_get(qh)); +done: + spin_unlock_irqrestore(&ehci->lock, flags); + if (unlikely(qh == NULL)) + qtd_list_free(ehci, urb, qtd_list); + return rc; +} + +/*-------------------------------------------------------------------------*/ + +/* the async qh for the qtds being reclaimed are now unlinked from the HC */ + +static void end_unlink_async(struct ehci_hcd *ehci) +{ + struct ehci_qh *qh = ehci->reclaim; + struct ehci_qh *next; + + iaa_watchdog_done(ehci); + + qh->qh_state = QH_STATE_IDLE; + qh->qh_next.qh = NULL; + qh_put(qh); /* refcount from reclaim */ + + /* other unlink(s) may be pending (in QH_STATE_UNLINK_WAIT) */ + next = qh->reclaim; + ehci->reclaim = next; + qh->reclaim = NULL; + + qh_completions(ehci, qh); + + if (!list_empty(&qh->qtd_list) + && HC_IS_RUNNING(ehci_to_hcd(ehci)->state)) + qh_link_async(ehci, qh); + else { + qh_put(qh); /* refcount from async list */ + + /* it's not free to turn the async schedule on/off; leave it + * active but idle for a while once it empties. + */ + if (HC_IS_RUNNING(ehci_to_hcd(ehci)->state) + && ehci->async->qh_next.qh == NULL) + timer_action(ehci, TIMER_ASYNC_OFF); + } + + if (next) { + ehci->reclaim = NULL; + start_unlink_async(ehci, next); + } +} + +/* makes sure the async qh will become idle */ +/* caller must own ehci->lock */ + +static void start_unlink_async(struct ehci_hcd *ehci, struct ehci_qh *qh) +{ + int cmd = ehci_readl(ehci, &ehci->regs->command); + struct ehci_qh *prev; + +#ifdef DEBUG + assert_spin_locked(&ehci->lock); + if (ehci->reclaim + || (qh->qh_state != QH_STATE_LINKED + && qh->qh_state != QH_STATE_UNLINK_WAIT) + ) + BUG(); +#endif + + /* stop async schedule right now? */ + if (unlikely(qh == ehci->async)) { + /* can't get here without STS_ASS set */ + if (ehci_to_hcd(ehci)->state != HC_STATE_HALT && + !ehci->reclaim) { + /* ... and CMD_IAAD clear */ + ehci_writel(ehci, cmd & ~CMD_ASE, &ehci->regs->command); + wmb(); + /* handshake later, if we need to */ + timer_action_done(ehci, TIMER_ASYNC_OFF); + } + return; + } + + qh->qh_state = QH_STATE_UNLINK; + ehci->reclaim = qh = qh_get(qh); + + prev = ehci->async; + while (prev->qh_next.qh != qh) + prev = prev->qh_next.qh; + + prev->hw_next = qh->hw_next; + prev->qh_next = qh->qh_next; + wmb(); + + if (unlikely(ehci_to_hcd(ehci)->state == HC_STATE_HALT)) { + /* if (unlikely (qh->reclaim != 0)) + * this will recurse, probably not much + */ + end_unlink_async(ehci); + return; + } + + cmd |= CMD_IAAD; + ehci_writel(ehci, cmd, &ehci->regs->command); + (void)ehci_readl(ehci, &ehci->regs->command); + iaa_watchdog_start(ehci); +} + +/*-------------------------------------------------------------------------*/ + +static void scan_async(struct ehci_hcd *ehci) +{ + struct ehci_qh *qh; + enum ehci_timer_action action = TIMER_IO_WATCHDOG; + + if (!++(ehci->stamp)) + ehci->stamp++; + timer_action_done(ehci, TIMER_ASYNC_SHRINK); +rescan: + qh = ehci->async->qh_next.qh; + if (likely(qh != NULL)) { + do { + /* clean any finished work for this qh */ + if (!list_empty(&qh->qtd_list) + && qh->stamp != ehci->stamp) { + int temp; + + /* unlinks could happen here; completion + * reporting drops the lock. rescan using + * the latest schedule, but don't rescan + * qhs we already finished (no looping). + */ + qh = qh_get(qh); + qh->stamp = ehci->stamp; + temp = qh_completions(ehci, qh); + qh_put(qh); + if (temp != 0) + goto rescan; + } + + /* unlink idle entries, reducing HC PCI usage as well + * as HCD schedule-scanning costs. delay for any qh + * we just scanned, there's a not-unusual case that it + * doesn't stay idle for long. + * (plus, avoids some kind of re-activation race.) + */ + if (list_empty(&qh->qtd_list)) { + if (qh->stamp == ehci->stamp) + action = TIMER_ASYNC_SHRINK; + else if (!ehci->reclaim + && qh->qh_state == QH_STATE_LINKED) + start_unlink_async(ehci, qh); + } + + qh = qh->qh_next.qh; + } while (qh); + } + if (action == TIMER_ASYNC_SHRINK) + timer_action(ehci, TIMER_ASYNC_SHRINK); +} diff --git a/drivers/usb/host/ehci.h b/drivers/usb/host/ehci.h index 9b9701fc1acc..3c3503acae58 100644 --- a/drivers/usb/host/ehci.h +++ b/drivers/usb/host/ehci.h @@ -157,6 +157,13 @@ struct ehci_hcd { /* one per controller */ * other external transceivers should be software-transparent */ struct otg_transceiver *transceiver; +#ifdef CONFIG_USB_STATIC_IRAM + u32 iram_buffer[2]; + u32 iram_buffer_v[2]; + int iram_in_use[2]; + int usb_address[2]; +#endif + /* irq statistics */ #ifdef EHCI_STATS struct ehci_stats stats; @@ -260,6 +267,10 @@ struct ehci_qtd { struct list_head qtd_list; /* sw qtd list */ struct urb *urb; /* qtd's urb */ size_t length; /* length of buffer */ +#ifdef CONFIG_USB_STATIC_IRAM + size_t buffer_offset; + int last_one; +#endif } __attribute__ ((aligned (32))); /* mask NakCnt+T in qh->hw_alt_next */ @@ -764,6 +775,10 @@ static inline unsigned ehci_read_frame_index(struct ehci_hcd *ehci) #define STUB_DEBUG_FILES #endif /* DEBUG */ +#ifdef CONFIG_USB_STATIC_IRAM +#define IRAM_TD_SIZE 1024 /* size of 1 qTD's buffer */ +#define IRAM_NTD 2 /* number of TDs in IRAM */ +#endif /*-------------------------------------------------------------------------*/ #endif /* __LINUX_EHCI_HCD_H */ -- 2.39.5