This patch is against latest linus git tree.
>From 7dc04f6d1feaa10f8e9c8d8edd4714f7dae8e92f Mon Sep 17 00:00:00 2001
From: Alek Du <alek.du@xxxxxxxxx>
Date: Thu, 18 Jun 2009 16:52:48 +0800
Subject: [PATCH] EHCI: split ehci_qh structure into hw and sw parts
The ehci_qh structure merged hw and sw together which is not good:
1. More and more items are being added into ehci_qh, the ehci_qh software
part are unnecessary to be allocated in DMA qh_pool.
2. If HCD has local SRAM, the sw part will consume it too, and it won't
bring any benefit.
3. For non-x86 system, the entire ehci_qh is uncachable, actually we
only need the hw part to be uncacheable. Split them will give the sw
part cacheable feature.
Signed-off-by: Alek Du <alek.du@xxxxxxxxx>
---
drivers/usb/host/ehci-dbg.c | 61 ++++++++++++++------------
drivers/usb/host/ehci-hcd.c | 21 +++++----
drivers/usb/host/ehci-hub.c | 2 +-
drivers/usb/host/ehci-mem.c | 26 ++++++++---
drivers/usb/host/ehci-q.c | 59 +++++++++++++------------
drivers/usb/host/ehci-sched.c | 97 +++++++++++++++++++++++------------------
drivers/usb/host/ehci.h | 19 ++++++--
7 files changed, 162 insertions(+), 123 deletions(-)
diff --git a/drivers/usb/host/ehci-dbg.c b/drivers/usb/host/ehci-dbg.c
index 7f4ace7..5b588f2 100644
--- a/drivers/usb/host/ehci-dbg.c
+++ b/drivers/usb/host/ehci-dbg.c
@@ -135,9 +135,9 @@ static void __maybe_unused
dbg_qh (const char *label, struct ehci_hcd *ehci, struct ehci_qh *qh)
{
ehci_dbg (ehci, "%s qh %p n%08x info %x %x qtd %x\n", label,
- qh, qh->hw_next, qh->hw_info1, qh->hw_info2,
- qh->hw_current);
- dbg_qtd ("overlay", ehci, (struct ehci_qtd *) &qh->hw_qtd_next);
+ qh, qh->hw->hw_next, qh->hw->hw_info1, qh->hw->hw_info2,
+ qh->hw->hw_current);
+ dbg_qtd("overlay", ehci, (struct ehci_qtd *) &qh->hw->hw_qtd_next);
}
static void __maybe_unused
@@ -401,30 +401,30 @@ static void qh_lines (
char mark;
__le32 list_end = EHCI_LIST_END(ehci);
- if (qh->hw_qtd_next == list_end) /* NEC does this */
+ if (qh->hw->hw_qtd_next == list_end) /* NEC does this */
mark = '@';
else
- mark = token_mark(ehci, qh->hw_token);
+ mark = token_mark(ehci, qh->hw->hw_token);
if (mark == '/') { /* qh_alt_next controls qh advance? */
- if ((qh->hw_alt_next & QTD_MASK(ehci))
- == ehci->async->hw_alt_next)
+ if ((qh->hw->hw_alt_next & QTD_MASK(ehci))
+ == ehci->async->hw->hw_alt_next)
mark = '#'; /* blocked */
- else if (qh->hw_alt_next == list_end)
+ else if (qh->hw->hw_alt_next == list_end)
mark = '.'; /* use hw_qtd_next */
/* else alt_next points to some other qtd */
}
- scratch = hc32_to_cpup(ehci, &qh->hw_info1);
- hw_curr = (mark == '*') ? hc32_to_cpup(ehci, &qh->hw_current) : 0;
+ scratch = hc32_to_cpup(ehci, &qh->hw->hw_info1);
+ hw_curr = (mark == '*') ? hc32_to_cpup(ehci, &qh->hw->hw_current) : 0;
temp = scnprintf (next, size,
"qh/%p dev%d %cs ep%d %08x %08x (%08x%c %s nak%d)",
qh, scratch & 0x007f,
speed_char (scratch),
(scratch >> 8) & 0x000f,
- scratch, hc32_to_cpup(ehci, &qh->hw_info2),
- hc32_to_cpup(ehci, &qh->hw_token), mark,
- (cpu_to_hc32(ehci, QTD_TOGGLE) & qh->hw_token)
+ scratch, hc32_to_cpup(ehci, &qh->hw->hw_info2),
+ hc32_to_cpup(ehci, &qh->hw->hw_token), mark,
+ (cpu_to_hc32(ehci, QTD_TOGGLE) & qh->hw->hw_token)
? "data1" : "data0",
- (hc32_to_cpup(ehci, &qh->hw_alt_next) >> 1) & 0x0f);
+ (hc32_to_cpup(ehci, &qh->hw->hw_alt_next) >> 1) & 0x0f);
size -= temp;
next += temp;
@@ -435,10 +435,10 @@ static void qh_lines (
mark = ' ';
if (hw_curr == td->qtd_dma)
mark = '*';
- else if (qh->hw_qtd_next == cpu_to_hc32(ehci, td->qtd_dma))
+ else if (qh->hw->hw_qtd_next == cpu_to_hc32(ehci, td->qtd_dma))
mark = '+';
else if (QTD_LENGTH (scratch)) {
- if (td->hw_alt_next == ehci->async->hw_alt_next)
+ if (td->hw_alt_next == ehci->async->hw->hw_alt_next)
mark = '#';
else if (td->hw_alt_next != list_end)
mark = '/';
@@ -495,7 +495,8 @@ static ssize_t fill_async_buffer(struct debug_buffer *buf)
* one QH per line, and TDs we know about
*/
spin_lock_irqsave (&ehci->lock, flags);
- for (qh = ehci->async->qh_next.qh; size > 0 && qh; qh = qh->qh_next.qh)
+ for (qh = to_qh(&ehci->async->qh_next); size > 0 && qh;
+ qh = to_qh(&qh->qh_next))
qh_lines (ehci, qh, &next, &size);
if (ehci->reclaim && size > 0) {
temp = scnprintf (next, size, "\nreclaim =\n");
@@ -521,6 +522,7 @@ static ssize_t fill_periodic_buffer(struct debug_buffer *buf)
char *next;
unsigned i;
__hc32 tag;
+ struct ehci_qh *qh;
if (!(seen = kmalloc (DBG_SCHED_LIMIT * sizeof *seen, GFP_ATOMIC)))
return 0;
@@ -552,20 +554,21 @@ static ssize_t fill_periodic_buffer(struct debug_buffer *buf)
do {
switch (hc32_to_cpu(ehci, tag)) {
case Q_TYPE_QH:
+ qh = to_qh(&p);
temp = scnprintf (next, size, " qh%d-%04x/%p",
- p.qh->period,
+ qh->period,
hc32_to_cpup(ehci,
- &p.qh->hw_info2)
+ &p.qh_hw->hw_info2)
/* uframe masks */
& (QH_CMASK | QH_SMASK),
- p.qh);
+ qh);
size -= temp;
next += temp;
/* don't repeat what follows this qh */
for (temp = 0; temp < seen_count; temp++) {
if (seen [temp].ptr != p.ptr)
continue;
- if (p.qh->qh_next.ptr) {
+ if (qh->qh_next.ptr) {
temp = scnprintf (next, size,
" ...");
size -= temp;
@@ -576,14 +579,14 @@ static ssize_t fill_periodic_buffer(struct debug_buffer *buf)
/* show more info the first time around */
if (temp == seen_count) {
u32 scratch = hc32_to_cpup(ehci,
- &p.qh->hw_info1);
+ &p.qh_hw->hw_info1);
struct ehci_qtd *qtd;
char *type = "";
/* count tds, get ep direction */
temp = 0;
list_for_each_entry (qtd,
- &p.qh->qtd_list,
+ &qh->qtd_list,
qtd_list) {
temp++;
switch (0x03 & (hc32_to_cpu(
@@ -600,17 +603,19 @@ static ssize_t fill_periodic_buffer(struct debug_buffer *buf)
speed_char (scratch),
scratch & 0x007f,
(scratch >> 8) & 0x000f, type,
- p.qh->usecs, p.qh->c_usecs,
+ qh->usecs, qh->c_usecs,
temp,
0x7ff & (scratch >> 16));
if (seen_count < DBG_SCHED_LIMIT)
- seen [seen_count++].qh = p.qh;
+ seen[seen_count++].qh_hw =
+ p.qh_hw;
} else
temp = 0;
- if (p.qh) {
- tag = Q_NEXT_TYPE(ehci, p.qh->hw_next);
- p = p.qh->qh_next;
+ if (qh) {
+ tag = Q_NEXT_TYPE(ehci,
+ p.qh_hw->hw_next);
+ p = qh->qh_next;
}
break;
case Q_TYPE_FSTN:
diff --git a/drivers/usb/host/ehci-hcd.c b/drivers/usb/host/ehci-hcd.c
index 2b72473..27361ae 100644
--- a/drivers/usb/host/ehci-hcd.c
+++ b/drivers/usb/host/ehci-hcd.c
@@ -543,13 +543,14 @@ static int ehci_init(struct usb_hcd *hcd)
* its dummy is used in hw_alt_next of many tds, to prevent the qh
* from automatically advancing to the next td after short reads.
*/
- ehci->async->qh_next.qh = NULL;
- ehci->async->hw_next = QH_NEXT(ehci, ehci->async->qh_dma);
- ehci->async->hw_info1 = cpu_to_hc32(ehci, QH_HEAD);
- ehci->async->hw_token = cpu_to_hc32(ehci, QTD_STS_HALT);
- ehci->async->hw_qtd_next = EHCI_LIST_END(ehci);
+ ehci->async->qh_next.qh_hw = NULL;
+ ehci->async->hw->hw_next = QH_NEXT(ehci, ehci->async->qh_dma);
+ ehci->async->hw->hw_info1 = cpu_to_hc32(ehci, QH_HEAD);
+ ehci->async->hw->hw_token = cpu_to_hc32(ehci, QTD_STS_HALT);
+ ehci->async->hw->hw_qtd_next = EHCI_LIST_END(ehci);
ehci->async->qh_state = QH_STATE_LINKED;
- ehci->async->hw_alt_next = QTD_NEXT(ehci, ehci->async->dummy->qtd_dma);
+ ehci->async->hw->hw_alt_next = QTD_NEXT(ehci,
+ ehci->async->dummy->qtd_dma);
/* clear interrupt enables, set irq latency */
if (log2_irq_thresh < 0 || log2_irq_thresh > 6)
@@ -978,7 +979,7 @@ rescan:
/* endpoints can be iso streams. for now, we don't
* accelerate iso completions ... so spin a while.
*/
- if (qh->hw_info1 == 0) {
+ if (qh->hw->hw_info1 == 0) {
ehci_vdbg (ehci, "iso delay\n");
goto idle_timeout;
}
@@ -987,9 +988,9 @@ rescan:
qh->qh_state = QH_STATE_IDLE;
switch (qh->qh_state) {
case QH_STATE_LINKED:
- for (tmp = ehci->async->qh_next.qh;
+ for (tmp = to_qh(&ehci->async->qh_next);
tmp && tmp != qh;
- tmp = tmp->qh_next.qh)
+ tmp = to_qh(&tmp->qh_next))
continue;
/* periodic qh self-unlinks on empty */
if (!tmp)
@@ -1047,7 +1048,7 @@ ehci_endpoint_reset(struct usb_hcd *hcd, struct usb_host_endpoint *ep)
if (!list_empty(&qh->qtd_list)) {
WARN_ONCE(1, "clear_halt for a busy endpoint\n");
} else if (qh->qh_state == QH_STATE_IDLE) {
- qh->hw_token &= ~cpu_to_hc32(ehci, QTD_TOGGLE);
+ qh->hw->hw_token &= ~cpu_to_hc32(ehci, QTD_TOGGLE);
} else {
/* It's not safe to write into the overlay area
* while the QH is active. Unlink it first and
diff --git a/drivers/usb/host/ehci-hub.c b/drivers/usb/host/ehci-hub.c
index f46ad27..e93b9c4 100644
--- a/drivers/usb/host/ehci-hub.c
+++ b/drivers/usb/host/ehci-hub.c
@@ -267,7 +267,7 @@ static int ehci_bus_resume (struct usb_hcd *hcd)
/* maybe re-activate the schedule(s) */
temp = 0;
- if (ehci->async->qh_next.qh)
+ if (ehci->async->qh_next.qh_hw)
temp |= CMD_ASE;
if (ehci->periodic_sched)
temp |= CMD_PSE;
diff --git a/drivers/usb/host/ehci-mem.c b/drivers/usb/host/ehci-mem.c
index 10d5291..8fc0d3e 100644
--- a/drivers/usb/host/ehci-mem.c
+++ b/drivers/usb/host/ehci-mem.c
@@ -75,7 +75,8 @@ static void qh_destroy(struct ehci_qh *qh)
}
if (qh->dummy)
ehci_qtd_free (ehci, qh->dummy);
- dma_pool_free (ehci->qh_pool, qh, qh->qh_dma);
+ dma_pool_free(ehci->qh_pool, qh->hw, qh->qh_dma);
+ kfree(qh);
}
static struct ehci_qh *ehci_qh_alloc (struct ehci_hcd *ehci, gfp_t flags)
@@ -83,15 +84,19 @@ static struct ehci_qh *ehci_qh_alloc (struct ehci_hcd *ehci, gfp_t flags)
struct ehci_qh *qh;
dma_addr_t dma;
- qh = (struct ehci_qh *)
- dma_pool_alloc (ehci->qh_pool, flags, &dma);
+ qh = kzalloc(sizeof *qh, GFP_ATOMIC);
if (!qh)
- return qh;
+ goto done;
+ qh->hw = (struct ehci_qh_hw *)
+ dma_pool_alloc(ehci->qh_pool, flags, &dma);
+ if (!qh->hw)
+ goto fail;
- memset (qh, 0, sizeof *qh);
+ memset(qh->hw, 0, sizeof *qh->hw);
qh->refcount = 1;
qh->ehci = ehci;
qh->qh_dma = dma;
+ qh->hw->sw = qh;
// INIT_LIST_HEAD (&qh->qh_list);
INIT_LIST_HEAD (&qh->qtd_list);
@@ -99,10 +104,15 @@ static struct ehci_qh *ehci_qh_alloc (struct ehci_hcd *ehci, gfp_t flags)
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;
+ goto fail1;
}
+done:
return qh;
+fail1:
+ dma_pool_free(ehci->qh_pool, qh->hw, qh->qh_dma);
+fail:
+ kfree(qh);
+ return NULL;
}
/* to share a qh (cpu threads, or hc) */
@@ -180,7 +190,7 @@ static int ehci_mem_init (struct ehci_hcd *ehci, gfp_t flags)
/* QHs for control/bulk/intr transfers */
ehci->qh_pool = dma_pool_create ("ehci_qh",
ehci_to_hcd(ehci)->self.controller,
- sizeof (struct ehci_qh),
+ sizeof(struct ehci_qh_hw),
32 /* byte alignment (for hw parts) */,
4096 /* can't cross 4K */);
if (!ehci->qh_pool) {
diff --git a/drivers/usb/host/ehci-q.c b/drivers/usb/host/ehci-q.c
index 3192f68..77d7d81 100644
--- a/drivers/usb/host/ehci-q.c
+++ b/drivers/usb/host/ehci-q.c
@@ -90,16 +90,16 @@ 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);
+ qh->hw->hw_qtd_next = QTD_NEXT(ehci, qtd->qtd_dma);
+ qh->hw->hw_alt_next = EHCI_LIST_END(ehci);
/* 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);
+ qh->hw->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
+ * overlay, so qh->hw->hw_token wrongly becomes inactive/halted), only fault
* recovery (including urb dequeue) would need software changes to a QH...
*/
static void
@@ -113,7 +113,7 @@ qh_refresh (struct ehci_hcd *ehci, struct ehci_qh *qh)
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)
+ if (cpu_to_hc32(ehci, qtd->qtd_dma) == qh->hw->hw_current)
qtd = NULL;
}
@@ -215,7 +215,7 @@ __acquires(ehci->lock)
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) {
+ if ((qh->hw->hw_info2 & cpu_to_hc32(ehci, QH_SMASK)) != 0) {
/* ... update hc-wide periodic stats (for usbfs) */
ehci_to_hcd(ehci)->self.bandwidth_int_reqs--;
@@ -257,7 +257,7 @@ 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,
+ * Chases up to qh->hw->hw_current. Returns number of completions called,
* indicating how much "real" work we did.
*/
static unsigned
@@ -348,7 +348,8 @@ qh_completions (struct ehci_hcd *ehci, struct ehci_qh *qh)
qtd->hw_token = cpu_to_hc32(ehci,
token);
wmb();
- qh->hw_token = cpu_to_hc32(ehci, token);
+ qh->hw->hw_token =
+ cpu_to_hc32(ehci, token);
goto retry_xacterr;
}
stopped = 1;
@@ -391,16 +392,16 @@ qh_completions (struct ehci_hcd *ehci, struct ehci_qh *qh)
/* 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->hw->hw_current)
+ token = hc32_to_cpu(ehci, qh->hw->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) {
+ if ((halt & qh->hw->hw_token) == 0) {
halt:
- qh->hw_token |= halt;
+ qh->hw->hw_token |= halt;
wmb ();
}
}
@@ -452,7 +453,7 @@ halt:
* 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)) {
+ if (stopped != 0 || qh->hw->hw_qtd_next == EHCI_LIST_END(ehci)) {
switch (state) {
case QH_STATE_IDLE:
qh_refresh(ehci, qh);
@@ -470,7 +471,7 @@ halt:
* except maybe high bandwidth ...
*/
if ((cpu_to_hc32(ehci, QH_SMASK)
- & qh->hw_info2) != 0) {
+ & qh->hw->hw_info2) != 0) {
intr_deschedule (ehci, qh);
(void) qh_schedule (ehci, qh);
} else
@@ -591,7 +592,7 @@ qh_urb_transaction (
* (this will usually be overridden later.)
*/
if (is_input)
- qtd->hw_alt_next = ehci->async->hw_alt_next;
+ qtd->hw_alt_next = ehci->async->hw->hw_alt_next;
/* qh makes control packets use qtd toggle; maybe switch it */
if ((maxpacket & (this_qtd_len + (maxpacket - 1))) == 0)
@@ -832,8 +833,8 @@ done:
/* 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);
+ qh->hw->hw_info1 = cpu_to_hc32(ehci, info1);
+ qh->hw->hw_info2 = cpu_to_hc32(ehci, info2);
qh_refresh (ehci, qh);
return qh;
}
@@ -850,7 +851,7 @@ static void qh_link_async (struct ehci_hcd *ehci, struct ehci_qh *qh)
/* (re)start the async schedule? */
head = ehci->async;
timer_action_done (ehci, TIMER_ASYNC_OFF);
- if (!head->qh_next.qh) {
+ if (!head->qh_next.qh_hw) {
u32 cmd = ehci_readl(ehci, &ehci->regs->command);
if (!(cmd & CMD_ASE)) {
@@ -870,11 +871,11 @@ static void qh_link_async (struct ehci_hcd *ehci, struct ehci_qh *qh)
/* splice right after start */
qh->qh_next = head->qh_next;
- qh->hw_next = head->hw_next;
+ qh->hw->hw_next = head->hw->hw_next;
wmb ();
- head->qh_next.qh = qh;
- head->hw_next = dma;
+ head->qh_next.qh_hw = qh->hw;
+ head->hw->hw_next = dma;
qh->xacterrs = QH_XACTERR_MAX;
qh->qh_state = QH_STATE_LINKED;
@@ -920,7 +921,7 @@ static struct ehci_qh *qh_append_tds (
/* usb_reset_device() briefly reverts to address 0 */
if (usb_pipedevice (urb->pipe) == 0)
- qh->hw_info1 &= ~qh_addr_mask;
+ qh->hw->hw_info1 &= ~qh_addr_mask;
}
/* just one way to queue requests: swap with the dummy qtd.
@@ -1037,7 +1038,7 @@ static void end_unlink_async (struct ehci_hcd *ehci)
// qh->hw_next = cpu_to_hc32(qh->qh_dma);
qh->qh_state = QH_STATE_IDLE;
- qh->qh_next.qh = NULL;
+ qh->qh_next.qh_hw = NULL;
qh_put (qh); // refcount from reclaim
/* other unlink(s) may be pending (in QH_STATE_UNLINK_WAIT) */
@@ -1057,7 +1058,7 @@ static void end_unlink_async (struct ehci_hcd *ehci)
* active but idle for a while once it empties.
*/
if (HC_IS_RUNNING (ehci_to_hcd(ehci)->state)
- && ehci->async->qh_next.qh == NULL)
+ && ehci->async->qh_next.qh_hw == NULL)
timer_action (ehci, TIMER_ASYNC_OFF);
}
@@ -1103,10 +1104,10 @@ static void start_unlink_async (struct ehci_hcd *ehci, struct ehci_qh *qh)
ehci->reclaim = qh = qh_get (qh);
prev = ehci->async;
- while (prev->qh_next.qh != qh)
- prev = prev->qh_next.qh;
+ while (prev->qh_next.qh_hw != qh->hw)
+ prev = to_qh(&prev->qh_next);
- prev->hw_next = qh->hw_next;
+ prev->hw->hw_next = qh->hw->hw_next;
prev->qh_next = qh->qh_next;
wmb ();
@@ -1135,7 +1136,7 @@ static void scan_async (struct ehci_hcd *ehci)
ehci->stamp = ehci_readl(ehci, &ehci->regs->frame_index);
timer_action_done (ehci, TIMER_ASYNC_SHRINK);
rescan:
- qh = ehci->async->qh_next.qh;
+ qh = to_qh(&ehci->async->qh_next);
if (likely (qh != NULL)) {
do {
/* clean any finished work for this qh */
@@ -1173,7 +1174,7 @@ rescan:
action = TIMER_ASYNC_SHRINK;
}
- qh = qh->qh_next.qh;
+ qh = to_qh(&qh->qh_next);
} while (qh);
}
if (action == TIMER_ASYNC_SHRINK)
diff --git a/drivers/usb/host/ehci-sched.c b/drivers/usb/host/ehci-sched.c
index 9d1babc..5489618 100644
--- a/drivers/usb/host/ehci-sched.c
+++ b/drivers/usb/host/ehci-sched.c
@@ -49,7 +49,7 @@ periodic_next_shadow(struct ehci_hcd *ehci, union ehci_shadow *periodic,
{
switch (hc32_to_cpu(ehci, tag)) {
case Q_TYPE_QH:
- return &periodic->qh->qh_next;
+ return &to_qh(periodic)->qh_next;
case Q_TYPE_FSTN:
return &periodic->fstn->fstn_next;
case Q_TYPE_ITD:
@@ -93,19 +93,21 @@ periodic_usecs (struct ehci_hcd *ehci, unsigned frame, unsigned uframe)
__hc32 *hw_p = &ehci->periodic [frame];
union ehci_shadow *q = &ehci->pshadow [frame];
unsigned usecs = 0;
+ struct ehci_qh *qh;
while (q->ptr) {
switch (hc32_to_cpu(ehci, Q_NEXT_TYPE(ehci, *hw_p))) {
case Q_TYPE_QH:
+ qh = to_qh(q);
/* is it in the S-mask? */
- if (q->qh->hw_info2 & cpu_to_hc32(ehci, 1 << uframe))
- usecs += q->qh->usecs;
+ if (q->qh_hw->hw_info2 & cpu_to_hc32(ehci, 1 << uframe))
+ usecs += qh->usecs;
/* ... or C-mask? */
- if (q->qh->hw_info2 & cpu_to_hc32(ehci,
+ if (q->qh_hw->hw_info2 & cpu_to_hc32(ehci,
1 << (8 + uframe)))
- usecs += q->qh->c_usecs;
- hw_p = &q->qh->hw_next;
- q = &q->qh->qh_next;
+ usecs += qh->c_usecs;
+ hw_p = &q->qh_hw->hw_next;
+ q = &qh->qh_next;
break;
// case Q_TYPE_FSTN:
default:
@@ -226,6 +228,7 @@ periodic_tt_usecs (
__hc32 *hw_p = &ehci->periodic [frame];
union ehci_shadow *q = &ehci->pshadow [frame];
unsigned char uf;
+ struct ehci_qh *qh;
memset(tt_usecs, 0, 16);
@@ -236,12 +239,13 @@ periodic_tt_usecs (
q = &q->itd->itd_next;
continue;
case Q_TYPE_QH:
- if (same_tt(dev, q->qh->dev)) {
- uf = tt_start_uframe(ehci, q->qh->hw_info2);
- tt_usecs[uf] += q->qh->tt_usecs;
+ qh = to_qh(q);
+ if (same_tt(dev, qh->dev)) {
+ uf = tt_start_uframe(ehci, q->qh_hw->hw_info2);
+ tt_usecs[uf] += qh->tt_usecs;
}
- hw_p = &q->qh->hw_next;
- q = &q->qh->qh_next;
+ hw_p = &q->qh_hw->hw_next;
+ q = &qh->qh_next;
continue;
case Q_TYPE_SITD:
if (same_tt(dev, q->sitd->urb->dev)) {
@@ -375,6 +379,7 @@ static int tt_no_collision (
for (; frame < ehci->periodic_size; frame += period) {
union ehci_shadow here;
__hc32 type;
+ struct ehci_qh *qh;
here = ehci->pshadow [frame];
type = Q_NEXT_TYPE(ehci, ehci->periodic [frame]);
@@ -385,18 +390,19 @@ static int tt_no_collision (
here = here.itd->itd_next;
continue;
case Q_TYPE_QH:
- if (same_tt (dev, here.qh->dev)) {
+ qh = to_qh(&here);
+ if (same_tt(dev, qh->dev)) {
u32 mask;
mask = hc32_to_cpu(ehci,
- here.qh->hw_info2);
+ here.qh_hw->hw_info2);
/* "knows" no gap is needed */
mask |= mask >> 8;
if (mask & uf_mask)
break;
}
- type = Q_NEXT_TYPE(ehci, here.qh->hw_next);
- here = here.qh->qh_next;
+ type = Q_NEXT_TYPE(ehci, here.qh_hw->hw_next);
+ here = qh->qh_next;
continue;
case Q_TYPE_SITD:
if (same_tt (dev, here.sitd->urb->dev)) {
@@ -498,7 +504,8 @@ static int qh_link_periodic (struct ehci_hcd *ehci, struct ehci_qh *qh)
dev_dbg (&qh->dev->dev,
"link qh%d-%04x/%p start %d [%d/%d us]\n",
- period, hc32_to_cpup(ehci, &qh->hw_info2) & (QH_CMASK | QH_SMASK),
+ period, hc32_to_cpup(ehci,
+ &qh->hw->hw_info2) & (QH_CMASK | QH_SMASK),
qh, qh->start, qh->usecs, qh->c_usecs);
/* high bandwidth, or otherwise every microframe */
@@ -517,27 +524,29 @@ static int qh_link_periodic (struct ehci_hcd *ehci, struct ehci_qh *qh)
if (type == cpu_to_hc32(ehci, Q_TYPE_QH))
break;
prev = periodic_next_shadow(ehci, prev, type);
- hw_p = &here.qh->hw_next;
+ hw_p = &here.qh_hw->hw_next;
here = *prev;
}
/* sorting each branch by period (slow-->fast)
* enables sharing interior tree nodes
*/
- while (here.ptr && qh != here.qh) {
- if (qh->period > here.qh->period)
+ while (here.ptr && qh->hw != here.qh_hw) {
+ struct ehci_qh *qh = to_qh(&here);
+
+ if (qh->period > qh->period)
break;
- prev = &here.qh->qh_next;
- hw_p = &here.qh->hw_next;
+ prev = &qh->qh_next;
+ hw_p = &here.qh_hw->hw_next;
here = *prev;
}
/* link in this qh, unless some earlier pass did that */
- if (qh != here.qh) {
+ if (qh->hw != here.qh_hw) {
qh->qh_next = here;
- if (here.qh)
- qh->hw_next = *hw_p;
+ if (here.qh_hw)
+ qh->hw->hw_next = *hw_p;
wmb ();
- prev->qh = qh;
+ prev->qh_hw = qh->hw;
*hw_p = QH_NEXT (ehci, qh->qh_dma);
}
}
@@ -580,7 +589,7 @@ static int qh_unlink_periodic(struct ehci_hcd *ehci, struct ehci_qh *qh)
dev_dbg (&qh->dev->dev,
"unlink qh%d-%04x/%p start %d [%d/%d us]\n",
qh->period,
- hc32_to_cpup(ehci, &qh->hw_info2) & (QH_CMASK | QH_SMASK),
+ hc32_to_cpup(ehci, &qh->hw->hw_info2) & (QH_CMASK | QH_SMASK),
qh, qh->start, qh->usecs, qh->c_usecs);
/* qh->qh_next still "live" to HC */
@@ -599,20 +608,20 @@ static void intr_deschedule (struct ehci_hcd *ehci, struct ehci_qh *qh)
qh_unlink_periodic (ehci, qh);
/* simple/paranoid: always delay, expecting the HC needs to read
- * qh->hw_next or finish a writeback after SPLIT/CSPLIT ... and
+ * qh->hw->hw_next or finish a writeback after SPLIT/CSPLIT ... and
* expect khubd to clean up after any CSPLITs we won't issue.
* active high speed queues may need bigger delays...
*/
if (list_empty (&qh->qtd_list)
|| (cpu_to_hc32(ehci, QH_CMASK)
- & qh->hw_info2) != 0)
+ & qh->hw->hw_info2) != 0)
wait = 2;
else
wait = 55; /* worst case: 3 * 1024 */
udelay (wait);
qh->qh_state = QH_STATE_IDLE;
- qh->hw_next = EHCI_LIST_END(ehci);
+ qh->hw->hw_next = EHCI_LIST_END(ehci);
wmb ();
}
@@ -740,12 +749,12 @@ static int qh_schedule(struct ehci_hcd *ehci, struct ehci_qh *qh)
unsigned frame; /* 0..(qh->period - 1), or NO_FRAME */
qh_refresh(ehci, qh);
- qh->hw_next = EHCI_LIST_END(ehci);
+ qh->hw->hw_next = EHCI_LIST_END(ehci);
frame = qh->start;
/* reuse the previous schedule slots, if we can */
if (frame < qh->period) {
- uframe = ffs(hc32_to_cpup(ehci, &qh->hw_info2) & QH_SMASK);
+ uframe = ffs(hc32_to_cpup(ehci, &qh->hw->hw_info2) & QH_SMASK);
status = check_intr_schedule (ehci, frame, --uframe,
qh, &c_mask);
} else {
@@ -783,11 +792,11 @@ static int qh_schedule(struct ehci_hcd *ehci, struct ehci_qh *qh)
qh->start = frame;
/* reset S-frame and (maybe) C-frame masks */
- qh->hw_info2 &= cpu_to_hc32(ehci, ~(QH_CMASK | QH_SMASK));
- qh->hw_info2 |= qh->period
+ qh->hw->hw_info2 &= cpu_to_hc32(ehci, ~(QH_CMASK | QH_SMASK));
+ qh->hw->hw_info2 |= qh->period
? cpu_to_hc32(ehci, 1 << uframe)
: cpu_to_hc32(ehci, QH_SMASK);
- qh->hw_info2 |= c_mask;
+ qh->hw->hw_info2 |= c_mask;
} else
ehci_dbg (ehci, "reused qh %p schedule\n", qh);
@@ -2175,20 +2184,22 @@ restart:
while (q.ptr != NULL) {
unsigned uf;
- union ehci_shadow temp;
int live;
+ struct ehci_qh *qh;
+ struct ehci_qh *qh_tmp;
live = HC_IS_RUNNING (ehci_to_hcd(ehci)->state);
switch (hc32_to_cpu(ehci, type)) {
case Q_TYPE_QH:
+ qh = to_qh(&q);
/* handle any completions */
- temp.qh = qh_get (q.qh);
- type = Q_NEXT_TYPE(ehci, q.qh->hw_next);
- q = q.qh->qh_next;
- modified = qh_completions (ehci, temp.qh);
- if (unlikely (list_empty (&temp.qh->qtd_list)))
- intr_deschedule (ehci, temp.qh);
- qh_put (temp.qh);
+ qh_tmp = qh_get(qh);
+ type = Q_NEXT_TYPE(ehci, q.qh_hw->hw_next);
+ q = qh->qh_next;
+ modified = qh_completions(ehci, qh_tmp);
+ if (unlikely(list_empty(&qh_tmp->qtd_list)))
+ intr_deschedule(ehci, qh_tmp);
+ qh_put(qh_tmp);
break;
case Q_TYPE_FSTN:
/* for "save place" FSTNs, look at QH entries
diff --git a/drivers/usb/host/ehci.h b/drivers/usb/host/ehci.h
index 90ad339..a949d49 100644
--- a/drivers/usb/host/ehci.h
+++ b/drivers/usb/host/ehci.h
@@ -280,7 +280,7 @@ struct ehci_qtd {
* For entries in the async schedule, the type tag always says "qh".
*/
union ehci_shadow {
- struct ehci_qh *qh; /* Q_TYPE_QH */
+ struct ehci_qh_hw *qh_hw; /* Q_TYPE_QH */
struct ehci_itd *itd; /* Q_TYPE_ITD */
struct ehci_sitd *sitd; /* Q_TYPE_SITD */
struct ehci_fstn *fstn; /* Q_TYPE_FSTN */
@@ -298,8 +298,8 @@ union ehci_shadow {
* These appear in both the async and (for interrupt) periodic schedules.
*/
-struct ehci_qh {
- /* first part defined by EHCI spec */
+/* first part defined by EHCI spec */
+struct ehci_qh_hw {
__hc32 hw_next; /* see EHCI 3.6.1 */
__hc32 hw_info1; /* see EHCI 3.6.2 */
#define QH_HEAD 0x00008000
@@ -318,6 +318,12 @@ struct ehci_qh {
__hc32 hw_buf [5];
__hc32 hw_buf_hi [5];
+ /* we need a back ptr to software part for some case */
+ struct ehci_qh *sw;
+} __attribute__ ((aligned(32)));
+
+struct ehci_qh {
+ struct ehci_qh_hw *hw;
/* the rest is HCD-private */
dma_addr_t qh_dma; /* address of qh */
union ehci_shadow qh_next; /* ptr to qh; or periodic */
@@ -355,7 +361,12 @@ struct ehci_qh {
unsigned short start; /* where polling starts */
#define NO_FRAME ((unsigned short)~0) /* pick new start */
struct usb_device *dev; /* access to TT */
-} __attribute__ ((aligned (32)));
+};
+
+static inline struct ehci_qh *to_qh(union ehci_shadow *shadow)
+{
+ return shadow->qh_hw ? shadow->qh_hw->sw : NULL;
+}
/*-------------------------------------------------------------------------*/
--
1.6.0.4
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
