diff options
Diffstat (limited to 'drivers/usb/host/ehci-sched.c')
| -rw-r--r-- | drivers/usb/host/ehci-sched.c | 1883 |
1 files changed, 1062 insertions, 821 deletions
diff --git a/drivers/usb/host/ehci-sched.c b/drivers/usb/host/ehci-sched.c index 8a8e08a51ba..e113fd73aea 100644 --- a/drivers/usb/host/ehci-sched.c +++ b/drivers/usb/host/ehci-sched.c @@ -36,8 +36,6 @@ static int ehci_get_frame (struct usb_hcd *hcd); -/*-------------------------------------------------------------------------*/ - /* * periodic_next_shadow - return "next" pointer on shadow list * @periodic: host pointer to qh/itd/sitd @@ -60,6 +58,20 @@ periodic_next_shadow(struct ehci_hcd *ehci, union ehci_shadow *periodic, } } +static __hc32 * +shadow_next_periodic(struct ehci_hcd *ehci, union ehci_shadow *periodic, + __hc32 tag) +{ + switch (hc32_to_cpu(ehci, tag)) { + /* our ehci_shadow.qh is actually software part */ + case Q_TYPE_QH: + return &periodic->qh->hw->hw_next; + /* others are hw parts */ + default: + return periodic->hw_next; + } +} + /* caller must hold ehci->lock */ static void periodic_unlink (struct ehci_hcd *ehci, unsigned frame, void *ptr) { @@ -71,7 +83,8 @@ static void periodic_unlink (struct ehci_hcd *ehci, unsigned frame, void *ptr) while (here.ptr && here.ptr != ptr) { prev_p = periodic_next_shadow(ehci, prev_p, Q_NEXT_TYPE(ehci, *hw_p)); - hw_p = here.hw_next; + hw_p = shadow_next_periodic(ehci, &here, + Q_NEXT_TYPE(ehci, *hw_p)); here = *prev_p; } /* an interrupt entry (at list end) could have been shared */ @@ -83,81 +96,217 @@ static void periodic_unlink (struct ehci_hcd *ehci, unsigned frame, void *ptr) */ *prev_p = *periodic_next_shadow(ehci, &here, Q_NEXT_TYPE(ehci, *hw_p)); - *hw_p = *here.hw_next; + + if (!ehci->use_dummy_qh || + *shadow_next_periodic(ehci, &here, Q_NEXT_TYPE(ehci, *hw_p)) + != EHCI_LIST_END(ehci)) + *hw_p = *shadow_next_periodic(ehci, &here, + Q_NEXT_TYPE(ehci, *hw_p)); + else + *hw_p = cpu_to_hc32(ehci, ehci->dummy->qh_dma); } -/* how many of the uframe's 125 usecs are allocated? */ -static unsigned short -periodic_usecs (struct ehci_hcd *ehci, unsigned frame, unsigned uframe) +/*-------------------------------------------------------------------------*/ + +/* Bandwidth and TT management */ + +/* Find the TT data structure for this device; create it if necessary */ +static struct ehci_tt *find_tt(struct usb_device *udev) { - __hc32 *hw_p = &ehci->periodic [frame]; - union ehci_shadow *q = &ehci->pshadow [frame]; - unsigned usecs = 0; - - while (q->ptr) { - switch (hc32_to_cpu(ehci, Q_NEXT_TYPE(ehci, *hw_p))) { - case Q_TYPE_QH: - /* is it in the S-mask? */ - if (q->qh->hw_info2 & cpu_to_hc32(ehci, 1 << uframe)) - usecs += q->qh->usecs; - /* ... or C-mask? */ - if (q->qh->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; - break; - // case Q_TYPE_FSTN: - default: - /* for "save place" FSTNs, count the relevant INTR - * bandwidth from the previous frame - */ - if (q->fstn->hw_prev != EHCI_LIST_END(ehci)) { - ehci_dbg (ehci, "ignoring FSTN cost ...\n"); - } - hw_p = &q->fstn->hw_next; - q = &q->fstn->fstn_next; - break; - case Q_TYPE_ITD: - if (q->itd->hw_transaction[uframe]) - usecs += q->itd->stream->usecs; - hw_p = &q->itd->hw_next; - q = &q->itd->itd_next; - break; - case Q_TYPE_SITD: - /* is it in the S-mask? (count SPLIT, DATA) */ - if (q->sitd->hw_uframe & cpu_to_hc32(ehci, - 1 << uframe)) { - if (q->sitd->hw_fullspeed_ep & - cpu_to_hc32(ehci, 1<<31)) - usecs += q->sitd->stream->usecs; - else /* worst case for OUT start-split */ - usecs += HS_USECS_ISO (188); - } + struct usb_tt *utt = udev->tt; + struct ehci_tt *tt, **tt_index, **ptt; + unsigned port; + bool allocated_index = false; + + if (!utt) + return NULL; /* Not below a TT */ + + /* + * Find/create our data structure. + * For hubs with a single TT, we get it directly. + * For hubs with multiple TTs, there's an extra level of pointers. + */ + tt_index = NULL; + if (utt->multi) { + tt_index = utt->hcpriv; + if (!tt_index) { /* Create the index array */ + tt_index = kzalloc(utt->hub->maxchild * + sizeof(*tt_index), GFP_ATOMIC); + if (!tt_index) + return ERR_PTR(-ENOMEM); + utt->hcpriv = tt_index; + allocated_index = true; + } + port = udev->ttport - 1; + ptt = &tt_index[port]; + } else { + port = 0; + ptt = (struct ehci_tt **) &utt->hcpriv; + } + + tt = *ptt; + if (!tt) { /* Create the ehci_tt */ + struct ehci_hcd *ehci = + hcd_to_ehci(bus_to_hcd(udev->bus)); - /* ... C-mask? (count CSPLIT, DATA) */ - if (q->sitd->hw_uframe & - cpu_to_hc32(ehci, 1 << (8 + uframe))) { - /* worst case for IN complete-split */ - usecs += q->sitd->stream->c_usecs; + tt = kzalloc(sizeof(*tt), GFP_ATOMIC); + if (!tt) { + if (allocated_index) { + utt->hcpriv = NULL; + kfree(tt_index); } + return ERR_PTR(-ENOMEM); + } + list_add_tail(&tt->tt_list, &ehci->tt_list); + INIT_LIST_HEAD(&tt->ps_list); + tt->usb_tt = utt; + tt->tt_port = port; + *ptt = tt; + } - hw_p = &q->sitd->hw_next; - q = &q->sitd->sitd_next; - break; + return tt; +} + +/* Release the TT above udev, if it's not in use */ +static void drop_tt(struct usb_device *udev) +{ + struct usb_tt *utt = udev->tt; + struct ehci_tt *tt, **tt_index, **ptt; + int cnt, i; + + if (!utt || !utt->hcpriv) + return; /* Not below a TT, or never allocated */ + + cnt = 0; + if (utt->multi) { + tt_index = utt->hcpriv; + ptt = &tt_index[udev->ttport - 1]; + + /* How many entries are left in tt_index? */ + for (i = 0; i < utt->hub->maxchild; ++i) + cnt += !!tt_index[i]; + } else { + tt_index = NULL; + ptt = (struct ehci_tt **) &utt->hcpriv; + } + + tt = *ptt; + if (!tt || !list_empty(&tt->ps_list)) + return; /* never allocated, or still in use */ + + list_del(&tt->tt_list); + *ptt = NULL; + kfree(tt); + if (cnt == 1) { + utt->hcpriv = NULL; + kfree(tt_index); + } +} + +static void bandwidth_dbg(struct ehci_hcd *ehci, int sign, char *type, + struct ehci_per_sched *ps) +{ + dev_dbg(&ps->udev->dev, + "ep %02x: %s %s @ %u+%u (%u.%u+%u) [%u/%u us] mask %04x\n", + ps->ep->desc.bEndpointAddress, + (sign >= 0 ? "reserve" : "release"), type, + (ps->bw_phase << 3) + ps->phase_uf, ps->bw_uperiod, + ps->phase, ps->phase_uf, ps->period, + ps->usecs, ps->c_usecs, ps->cs_mask); +} + +static void reserve_release_intr_bandwidth(struct ehci_hcd *ehci, + struct ehci_qh *qh, int sign) +{ + unsigned start_uf; + unsigned i, j, m; + int usecs = qh->ps.usecs; + int c_usecs = qh->ps.c_usecs; + int tt_usecs = qh->ps.tt_usecs; + struct ehci_tt *tt; + + if (qh->ps.phase == NO_FRAME) /* Bandwidth wasn't reserved */ + return; + start_uf = qh->ps.bw_phase << 3; + + bandwidth_dbg(ehci, sign, "intr", &qh->ps); + + if (sign < 0) { /* Release bandwidth */ + usecs = -usecs; + c_usecs = -c_usecs; + tt_usecs = -tt_usecs; + } + + /* Entire transaction (high speed) or start-split (full/low speed) */ + for (i = start_uf + qh->ps.phase_uf; i < EHCI_BANDWIDTH_SIZE; + i += qh->ps.bw_uperiod) + ehci->bandwidth[i] += usecs; + + /* Complete-split (full/low speed) */ + if (qh->ps.c_usecs) { + /* NOTE: adjustments needed for FSTN */ + for (i = start_uf; i < EHCI_BANDWIDTH_SIZE; + i += qh->ps.bw_uperiod) { + for ((j = 2, m = 1 << (j+8)); j < 8; (++j, m <<= 1)) { + if (qh->ps.cs_mask & m) + ehci->bandwidth[i+j] += c_usecs; + } } } -#ifdef DEBUG - if (usecs > 100) - ehci_err (ehci, "uframe %d sched overrun: %d usecs\n", - frame * 8 + uframe, usecs); -#endif - return usecs; + + /* FS/LS bus bandwidth */ + if (tt_usecs) { + tt = find_tt(qh->ps.udev); + if (sign > 0) + list_add_tail(&qh->ps.ps_list, &tt->ps_list); + else + list_del(&qh->ps.ps_list); + + for (i = start_uf >> 3; i < EHCI_BANDWIDTH_FRAMES; + i += qh->ps.bw_period) + tt->bandwidth[i] += tt_usecs; + } } /*-------------------------------------------------------------------------*/ -static int same_tt (struct usb_device *dev1, struct usb_device *dev2) +static void compute_tt_budget(u8 budget_table[EHCI_BANDWIDTH_SIZE], + struct ehci_tt *tt) +{ + struct ehci_per_sched *ps; + unsigned uframe, uf, x; + u8 *budget_line; + + if (!tt) + return; + memset(budget_table, 0, EHCI_BANDWIDTH_SIZE); + + /* Add up the contributions from all the endpoints using this TT */ + list_for_each_entry(ps, &tt->ps_list, ps_list) { + for (uframe = ps->bw_phase << 3; uframe < EHCI_BANDWIDTH_SIZE; + uframe += ps->bw_uperiod) { + budget_line = &budget_table[uframe]; + x = ps->tt_usecs; + + /* propagate the time forward */ + for (uf = ps->phase_uf; uf < 8; ++uf) { + x += budget_line[uf]; + + /* Each microframe lasts 125 us */ + if (x <= 125) { + budget_line[uf] = x; + break; + } else { + budget_line[uf] = 125; + x -= 125; + } + } + } + } +} + +static int __maybe_unused same_tt(struct usb_device *dev1, + struct usb_device *dev2) { if (!dev1->tt || !dev2->tt) return 0; @@ -205,68 +354,6 @@ static inline void carryover_tt_bandwidth(unsigned short tt_usecs[8]) } } -/* How many of the tt's periodic downstream 1000 usecs are allocated? - * - * While this measures the bandwidth in terms of usecs/uframe, - * the low/fullspeed bus has no notion of uframes, so any particular - * low/fullspeed transfer can "carry over" from one uframe to the next, - * since the TT just performs downstream transfers in sequence. - * - * For example two separate 100 usec transfers can start in the same uframe, - * and the second one would "carry over" 75 usecs into the next uframe. - */ -static void -periodic_tt_usecs ( - struct ehci_hcd *ehci, - struct usb_device *dev, - unsigned frame, - unsigned short tt_usecs[8] -) -{ - __hc32 *hw_p = &ehci->periodic [frame]; - union ehci_shadow *q = &ehci->pshadow [frame]; - unsigned char uf; - - memset(tt_usecs, 0, 16); - - while (q->ptr) { - switch (hc32_to_cpu(ehci, Q_NEXT_TYPE(ehci, *hw_p))) { - case Q_TYPE_ITD: - hw_p = &q->itd->hw_next; - 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; - } - hw_p = &q->qh->hw_next; - q = &q->qh->qh_next; - continue; - case Q_TYPE_SITD: - if (same_tt(dev, q->sitd->urb->dev)) { - uf = tt_start_uframe(ehci, q->sitd->hw_uframe); - tt_usecs[uf] += q->sitd->stream->tt_usecs; - } - hw_p = &q->sitd->hw_next; - q = &q->sitd->sitd_next; - continue; - // case Q_TYPE_FSTN: - default: - ehci_dbg(ehci, "ignoring periodic frame %d FSTN\n", - frame); - hw_p = &q->fstn->hw_next; - q = &q->fstn->fstn_next; - } - } - - carryover_tt_bandwidth(tt_usecs); - - if (max_tt_usecs[7] < tt_usecs[7]) - ehci_err(ehci, "frame %d tt sched overrun: %d usecs\n", - frame, tt_usecs[7] - max_tt_usecs[7]); -} - /* * Return true if the device's tt's downstream bus is available for a * periodic transfer of the specified length (usecs), starting at the @@ -290,32 +377,32 @@ periodic_tt_usecs ( */ static int tt_available ( struct ehci_hcd *ehci, - unsigned period, - struct usb_device *dev, + struct ehci_per_sched *ps, + struct ehci_tt *tt, unsigned frame, - unsigned uframe, - u16 usecs + unsigned uframe ) { + unsigned period = ps->bw_period; + unsigned usecs = ps->tt_usecs; + if ((period == 0) || (uframe >= 7)) /* error */ return 0; - for (; frame < ehci->periodic_size; frame += period) { - unsigned short tt_usecs[8]; + for (frame &= period - 1; frame < EHCI_BANDWIDTH_FRAMES; + frame += period) { + unsigned i, uf; + unsigned short tt_usecs[8]; - periodic_tt_usecs (ehci, dev, frame, tt_usecs); + if (tt->bandwidth[frame] + usecs > 900) + return 0; - ehci_vdbg(ehci, "tt frame %d check %d usecs start uframe %d in" - " schedule %d/%d/%d/%d/%d/%d/%d/%d\n", - frame, usecs, uframe, - tt_usecs[0], tt_usecs[1], tt_usecs[2], tt_usecs[3], - tt_usecs[4], tt_usecs[5], tt_usecs[6], tt_usecs[7]); + uf = frame << 3; + for (i = 0; i < 8; (++i, ++uf)) + tt_usecs[i] = ehci->tt_budget[uf]; - if (max_tt_usecs[uframe] <= tt_usecs[uframe]) { - ehci_vdbg(ehci, "frame %d uframe %d fully scheduled\n", - frame, uframe); + if (max_tt_usecs[uframe] <= tt_usecs[uframe]) return 0; - } /* special case for isoc transfers larger than 125us: * the first and each subsequent fully used uframe @@ -323,16 +410,11 @@ static int tt_available ( * already scheduled transactions */ if (125 < usecs) { - int ufs = (usecs / 125) - 1; - int i; + int ufs = (usecs / 125); + for (i = uframe; i < (uframe + ufs) && i < 8; i++) - if (0 < tt_usecs[i]) { - ehci_vdbg(ehci, - "multi-uframe xfer can't fit " - "in frame %d uframe %d\n", - frame, i); + if (0 < tt_usecs[i]) return 0; - } } tt_usecs[uframe] += usecs; @@ -340,12 +422,8 @@ static int tt_available ( carryover_tt_bandwidth(tt_usecs); /* fail if the carryover pushed bw past the last uframe's limit */ - if (max_tt_usecs[7] < tt_usecs[7]) { - ehci_vdbg(ehci, - "tt unavailable usecs %d frame %d uframe %d\n", - usecs, frame, uframe); + if (max_tt_usecs[7] < tt_usecs[7]) return 0; - } } return 1; @@ -375,6 +453,7 @@ static int tt_no_collision ( for (; frame < ehci->periodic_size; frame += period) { union ehci_shadow here; __hc32 type; + struct ehci_qh_hw *hw; here = ehci->pshadow [frame]; type = Q_NEXT_TYPE(ehci, ehci->periodic [frame]); @@ -385,17 +464,18 @@ static int tt_no_collision ( here = here.itd->itd_next; continue; case Q_TYPE_QH: - if (same_tt (dev, here.qh->dev)) { + hw = here.qh->hw; + if (same_tt(dev, here.qh->ps.udev)) { u32 mask; mask = hc32_to_cpu(ehci, - here.qh->hw_info2); + 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); + type = Q_NEXT_TYPE(ehci, hw->hw_next); here = here.qh->qh_next; continue; case Q_TYPE_SITD: @@ -432,51 +512,26 @@ static int tt_no_collision ( /*-------------------------------------------------------------------------*/ -static int enable_periodic (struct ehci_hcd *ehci) +static void enable_periodic(struct ehci_hcd *ehci) { - u32 cmd; - int status; - - /* did clearing PSE did take effect yet? - * takes effect only at frame boundaries... - */ - status = handshake(ehci, &ehci->regs->status, STS_PSS, 0, 9 * 125); - if (status != 0) { - ehci_to_hcd(ehci)->state = HC_STATE_HALT; - return status; - } + if (ehci->periodic_count++) + return; - cmd = ehci_readl(ehci, &ehci->regs->command) | CMD_PSE; - ehci_writel(ehci, cmd, &ehci->regs->command); - /* posted write ... PSS happens later */ - ehci_to_hcd(ehci)->state = HC_STATE_RUNNING; + /* Stop waiting to turn off the periodic schedule */ + ehci->enabled_hrtimer_events &= ~BIT(EHCI_HRTIMER_DISABLE_PERIODIC); - /* make sure ehci_work scans these */ - ehci->next_uframe = ehci_readl(ehci, &ehci->regs->frame_index) - % (ehci->periodic_size << 3); - return 0; + /* Don't start the schedule until PSS is 0 */ + ehci_poll_PSS(ehci); + turn_on_io_watchdog(ehci); } -static int disable_periodic (struct ehci_hcd *ehci) +static void disable_periodic(struct ehci_hcd *ehci) { - u32 cmd; - int status; - - /* did setting PSE not take effect yet? - * takes effect only at frame boundaries... - */ - status = handshake(ehci, &ehci->regs->status, STS_PSS, STS_PSS, 9 * 125); - if (status != 0) { - ehci_to_hcd(ehci)->state = HC_STATE_HALT; - return status; - } - - cmd = ehci_readl(ehci, &ehci->regs->command) & ~CMD_PSE; - ehci_writel(ehci, cmd, &ehci->regs->command); - /* posted write ... */ + if (--ehci->periodic_count) + return; - ehci->next_uframe = -1; - return 0; + /* Don't turn off the schedule until PSS is 1 */ + ehci_poll_PSS(ehci); } /*-------------------------------------------------------------------------*/ @@ -487,21 +542,22 @@ static int disable_periodic (struct ehci_hcd *ehci) * this just links in a qh; caller guarantees uframe masks are set right. * no FSTN support (yet; ehci 0.96+) */ -static int qh_link_periodic (struct ehci_hcd *ehci, struct ehci_qh *qh) +static void qh_link_periodic(struct ehci_hcd *ehci, struct ehci_qh *qh) { unsigned i; - unsigned period = qh->period; + unsigned period = qh->ps.period; - dev_dbg (&qh->dev->dev, + dev_dbg(&qh->ps.udev->dev, "link qh%d-%04x/%p start %d [%d/%d us]\n", - period, hc32_to_cpup(ehci, &qh->hw_info2) & (QH_CMASK | QH_SMASK), - qh, qh->start, qh->usecs, qh->c_usecs); + period, hc32_to_cpup(ehci, &qh->hw->hw_info2) + & (QH_CMASK | QH_SMASK), + qh, qh->ps.phase, qh->ps.usecs, qh->ps.c_usecs); /* high bandwidth, or otherwise every microframe */ if (period == 0) period = 1; - for (i = qh->start; i < ehci->periodic_size; i += period) { + for (i = qh->ps.phase; i < ehci->periodic_size; i += period) { union ehci_shadow *prev = &ehci->pshadow[i]; __hc32 *hw_p = &ehci->periodic[i]; union ehci_shadow here = *prev; @@ -513,7 +569,7 @@ 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 = shadow_next_periodic(ehci, &here, type); here = *prev; } @@ -521,100 +577,188 @@ static int qh_link_periodic (struct ehci_hcd *ehci, struct ehci_qh *qh) * enables sharing interior tree nodes */ while (here.ptr && qh != here.qh) { - if (qh->period > here.qh->period) + if (qh->ps.period > here.qh->ps.period) break; prev = &here.qh->qh_next; - hw_p = &here.qh->hw_next; + hw_p = &here.qh->hw->hw_next; here = *prev; } /* link in this qh, unless some earlier pass did that */ if (qh != here.qh) { qh->qh_next = here; if (here.qh) - qh->hw_next = *hw_p; + qh->hw->hw_next = *hw_p; wmb (); prev->qh = qh; *hw_p = QH_NEXT (ehci, qh->qh_dma); } } qh->qh_state = QH_STATE_LINKED; - qh_get (qh); + qh->xacterrs = 0; + qh->exception = 0; - /* update per-qh bandwidth for usbfs */ - ehci_to_hcd(ehci)->self.bandwidth_allocated += qh->period - ? ((qh->usecs + qh->c_usecs) / qh->period) - : (qh->usecs * 8); + /* update per-qh bandwidth for debugfs */ + ehci_to_hcd(ehci)->self.bandwidth_allocated += qh->ps.bw_period + ? ((qh->ps.usecs + qh->ps.c_usecs) / qh->ps.bw_period) + : (qh->ps.usecs * 8); - /* maybe enable periodic schedule processing */ - if (!ehci->periodic_sched++) - return enable_periodic (ehci); + list_add(&qh->intr_node, &ehci->intr_qh_list); - return 0; + /* maybe enable periodic schedule processing */ + ++ehci->intr_count; + enable_periodic(ehci); } -static void qh_unlink_periodic (struct ehci_hcd *ehci, struct ehci_qh *qh) +static void qh_unlink_periodic(struct ehci_hcd *ehci, struct ehci_qh *qh) { unsigned i; unsigned period; - // FIXME: - // IF this isn't high speed - // and this qh is active in the current uframe - // (and overlay token SplitXstate is false?) - // THEN - // qh->hw_info1 |= __constant_cpu_to_hc32(1 << 7 /* "ignore" */); + /* + * If qh is for a low/full-speed device, simply unlinking it + * could interfere with an ongoing split transaction. To unlink + * it safely would require setting the QH_INACTIVATE bit and + * waiting at least one frame, as described in EHCI 4.12.2.5. + * + * We won't bother with any of this. Instead, we assume that the + * only reason for unlinking an interrupt QH while the current URB + * is still active is to dequeue all the URBs (flush the whole + * endpoint queue). + * + * If rebalancing the periodic schedule is ever implemented, this + * approach will no longer be valid. + */ /* high bandwidth, or otherwise part of every microframe */ - if ((period = qh->period) == 0) - period = 1; + period = qh->ps.period ? : 1; - for (i = qh->start; i < ehci->periodic_size; i += period) + for (i = qh->ps.phase; i < ehci->periodic_size; i += period) periodic_unlink (ehci, i, qh); - /* update per-qh bandwidth for usbfs */ - ehci_to_hcd(ehci)->self.bandwidth_allocated -= qh->period - ? ((qh->usecs + qh->c_usecs) / qh->period) - : (qh->usecs * 8); + /* update per-qh bandwidth for debugfs */ + ehci_to_hcd(ehci)->self.bandwidth_allocated -= qh->ps.bw_period + ? ((qh->ps.usecs + qh->ps.c_usecs) / qh->ps.bw_period) + : (qh->ps.usecs * 8); - dev_dbg (&qh->dev->dev, + dev_dbg(&qh->ps.udev->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), - qh, qh->start, qh->usecs, qh->c_usecs); + qh->ps.period, + hc32_to_cpup(ehci, &qh->hw->hw_info2) & (QH_CMASK | QH_SMASK), + qh, qh->ps.phase, qh->ps.usecs, qh->ps.c_usecs); /* qh->qh_next still "live" to HC */ qh->qh_state = QH_STATE_UNLINK; qh->qh_next.ptr = NULL; - qh_put (qh); - /* maybe turn off periodic schedule */ - ehci->periodic_sched--; - if (!ehci->periodic_sched) - (void) disable_periodic (ehci); + if (ehci->qh_scan_next == qh) + ehci->qh_scan_next = list_entry(qh->intr_node.next, + struct ehci_qh, intr_node); + list_del(&qh->intr_node); } -static void intr_deschedule (struct ehci_hcd *ehci, struct ehci_qh *qh) +static void cancel_unlink_wait_intr(struct ehci_hcd *ehci, struct ehci_qh *qh) { - unsigned wait; + if (qh->qh_state != QH_STATE_LINKED || + list_empty(&qh->unlink_node)) + return; + + list_del_init(&qh->unlink_node); + + /* + * TODO: disable the event of EHCI_HRTIMER_START_UNLINK_INTR for + * avoiding unnecessary CPU wakeup + */ +} + +static void start_unlink_intr(struct ehci_hcd *ehci, struct ehci_qh *qh) +{ + /* If the QH isn't linked then there's nothing we can do. */ + if (qh->qh_state != QH_STATE_LINKED) + return; + + /* if the qh is waiting for unlink, cancel it now */ + cancel_unlink_wait_intr(ehci, 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 - * expect khubd to clean up after any CSPLITs we won't issue. - * active high speed queues may need bigger delays... + /* Make sure the unlinks are visible before starting the timer */ + wmb(); + + /* + * The EHCI spec doesn't say how long it takes the controller to + * stop accessing an unlinked interrupt QH. The timer delay is + * 9 uframes; presumably that will be long enough. */ - if (list_empty (&qh->qtd_list) - || (cpu_to_hc32(ehci, QH_CMASK) - & qh->hw_info2) != 0) - wait = 2; - else - wait = 55; /* worst case: 3 * 1024 */ + qh->unlink_cycle = ehci->intr_unlink_cycle; + + /* New entries go at the end of the intr_unlink list */ + list_add_tail(&qh->unlink_node, &ehci->intr_unlink); + + if (ehci->intr_unlinking) + ; /* Avoid recursive calls */ + else if (ehci->rh_state < EHCI_RH_RUNNING) + ehci_handle_intr_unlinks(ehci); + else if (ehci->intr_unlink.next == &qh->unlink_node) { + ehci_enable_event(ehci, EHCI_HRTIMER_UNLINK_INTR, true); + ++ehci->intr_unlink_cycle; + } +} + +/* + * It is common only one intr URB is scheduled on one qh, and + * given complete() is run in tasklet context, introduce a bit + * delay to avoid unlink qh too early. + */ +static void start_unlink_intr_wait(struct ehci_hcd *ehci, + struct ehci_qh *qh) +{ + qh->unlink_cycle = ehci->intr_unlink_wait_cycle; + + /* New entries go at the end of the intr_unlink_wait list */ + list_add_tail(&qh->unlink_node, &ehci->intr_unlink_wait); + + if (ehci->rh_state < EHCI_RH_RUNNING) + ehci_handle_start_intr_unlinks(ehci); + else if (ehci->intr_unlink_wait.next == &qh->unlink_node) { + ehci_enable_event(ehci, EHCI_HRTIMER_START_UNLINK_INTR, true); + ++ehci->intr_unlink_wait_cycle; + } +} + +static void end_unlink_intr(struct ehci_hcd *ehci, struct ehci_qh *qh) +{ + struct ehci_qh_hw *hw = qh->hw; + int rc; - udelay (wait); qh->qh_state = QH_STATE_IDLE; - qh->hw_next = EHCI_LIST_END(ehci); - wmb (); + hw->hw_next = EHCI_LIST_END(ehci); + + if (!list_empty(&qh->qtd_list)) + qh_completions(ehci, qh); + + /* reschedule QH iff another request is queued */ + if (!list_empty(&qh->qtd_list) && ehci->rh_state == EHCI_RH_RUNNING) { + rc = qh_schedule(ehci, qh); + if (rc == 0) { + qh_refresh(ehci, qh); + qh_link_periodic(ehci, qh); + } + + /* An error here likely indicates handshake failure + * or no space left in the schedule. Neither fault + * should happen often ... + * + * FIXME kill the now-dysfunctional queued urbs + */ + else { + ehci_err(ehci, "can't reschedule qh %p, err %d\n", + qh, rc); + } + } + + /* maybe turn off periodic schedule */ + --ehci->intr_count; + disable_periodic(ehci); } /*-------------------------------------------------------------------------*/ @@ -623,42 +767,22 @@ static int check_period ( struct ehci_hcd *ehci, unsigned frame, unsigned uframe, - unsigned period, + unsigned uperiod, unsigned usecs ) { - int claimed; - /* complete split running into next frame? * given FSTN support, we could sometimes check... */ if (uframe >= 8) return 0; - /* - * 80% periodic == 100 usec/uframe available - * convert "usecs we need" to "max already claimed" - */ - usecs = 100 - usecs; + /* convert "usecs we need" to "max already claimed" */ + usecs = ehci->uframe_periodic_max - usecs; - /* we "know" 2 and 4 uframe intervals were rejected; so - * for period 0, check _every_ microframe in the schedule. - */ - if (unlikely (period == 0)) { - do { - for (uframe = 0; uframe < 7; uframe++) { - claimed = periodic_usecs (ehci, frame, uframe); - if (claimed > usecs) - return 0; - } - } while ((frame += 1) < ehci->periodic_size); - - /* just check the specified uframe, at that period */ - } else { - do { - claimed = periodic_usecs (ehci, frame, uframe); - if (claimed > usecs) - return 0; - } while ((frame += period) < ehci->periodic_size); + for (uframe += frame << 3; uframe < EHCI_BANDWIDTH_SIZE; + uframe += uperiod) { + if (ehci->bandwidth[uframe] > usecs) + return 0; } // success! @@ -669,40 +793,40 @@ static int check_intr_schedule ( struct ehci_hcd *ehci, unsigned frame, unsigned uframe, - const struct ehci_qh *qh, - __hc32 *c_maskp + struct ehci_qh *qh, + unsigned *c_maskp, + struct ehci_tt *tt ) { int retval = -ENOSPC; u8 mask = 0; - if (qh->c_usecs && uframe >= 6) /* FSTN territory? */ + if (qh->ps.c_usecs && uframe >= 6) /* FSTN territory? */ goto done; - if (!check_period (ehci, frame, uframe, qh->period, qh->usecs)) + if (!check_period(ehci, frame, uframe, qh->ps.bw_uperiod, qh->ps.usecs)) goto done; - if (!qh->c_usecs) { + if (!qh->ps.c_usecs) { retval = 0; *c_maskp = 0; goto done; } #ifdef CONFIG_USB_EHCI_TT_NEWSCHED - if (tt_available (ehci, qh->period, qh->dev, frame, uframe, - qh->tt_usecs)) { + if (tt_available(ehci, &qh->ps, tt, frame, uframe)) { unsigned i; /* TODO : this may need FSTN for SSPLIT in uframe 5. */ - for (i=uframe+1; i<8 && i<uframe+4; i++) - if (!check_period (ehci, frame, i, - qh->period, qh->c_usecs)) + for (i = uframe+2; i < 8 && i <= uframe+4; i++) + if (!check_period(ehci, frame, i, + qh->ps.bw_uperiod, qh->ps.c_usecs)) goto done; else mask |= 1 << i; retval = 0; - *c_maskp = cpu_to_hc32(ehci, mask << 8); + *c_maskp = mask; } #else /* Make sure this tt's buffer is also available for CSPLITs. @@ -713,15 +837,15 @@ static int check_intr_schedule ( * one smart pass... */ mask = 0x03 << (uframe + qh->gap_uf); - *c_maskp = cpu_to_hc32(ehci, mask << 8); + *c_maskp = mask; mask |= 1 << uframe; - if (tt_no_collision (ehci, qh->period, qh->dev, frame, mask)) { - if (!check_period (ehci, frame, uframe + qh->gap_uf + 1, - qh->period, qh->c_usecs)) + if (tt_no_collision(ehci, qh->ps.bw_period, qh->ps.udev, frame, mask)) { + if (!check_period(ehci, frame, uframe + qh->gap_uf + 1, + qh->ps.bw_uperiod, qh->ps.c_usecs)) goto done; - if (!check_period (ehci, frame, uframe + qh->gap_uf, - qh->period, qh->c_usecs)) + if (!check_period(ehci, frame, uframe + qh->gap_uf, + qh->ps.bw_uperiod, qh->ps.c_usecs)) goto done; retval = 0; } @@ -735,63 +859,68 @@ done: */ static int qh_schedule(struct ehci_hcd *ehci, struct ehci_qh *qh) { - int status; + int status = 0; unsigned uframe; - __hc32 c_mask; - unsigned frame; /* 0..(qh->period - 1), or NO_FRAME */ + unsigned c_mask; + struct ehci_qh_hw *hw = qh->hw; + struct ehci_tt *tt; - qh_refresh(ehci, qh); - qh->hw_next = EHCI_LIST_END(ehci); - frame = qh->start; + hw->hw_next = EHCI_LIST_END(ehci); /* reuse the previous schedule slots, if we can */ - if (frame < qh->period) { - uframe = ffs(hc32_to_cpup(ehci, &qh->hw_info2) & QH_SMASK); - status = check_intr_schedule (ehci, frame, --uframe, - qh, &c_mask); - } else { - uframe = 0; - c_mask = 0; - status = -ENOSPC; + if (qh->ps.phase != NO_FRAME) { + ehci_dbg(ehci, "reused qh %p schedule\n", qh); + return 0; + } + + uframe = 0; + c_mask = 0; + tt = find_tt(qh->ps.udev); + if (IS_ERR(tt)) { + status = PTR_ERR(tt); + goto done; } + compute_tt_budget(ehci->tt_budget, tt); /* else scan the schedule to find a group of slots such that all * uframes have enough periodic bandwidth available. */ - if (status) { - /* "normal" case, uframing flexible except with splits */ - if (qh->period) { - frame = qh->period - 1; - do { - for (uframe = 0; uframe < 8; uframe++) { - status = check_intr_schedule (ehci, - frame, uframe, qh, - &c_mask); - if (status == 0) - break; - } - } while (status && frame--); - - /* qh->period == 0 means every uframe */ - } else { - frame = 0; - status = check_intr_schedule (ehci, 0, 0, qh, &c_mask); + /* "normal" case, uframing flexible except with splits */ + if (qh->ps.bw_period) { + int i; + unsigned frame; + + for (i = qh->ps.bw_period; i > 0; --i) { + frame = ++ehci->random_frame & (qh->ps.bw_period - 1); + for (uframe = 0; uframe < 8; uframe++) { + status = check_intr_schedule(ehci, + frame, uframe, qh, &c_mask, tt); + if (status == 0) + goto got_it; + } } - if (status) - goto done; - 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 - ? cpu_to_hc32(ehci, 1 << uframe) - : cpu_to_hc32(ehci, QH_SMASK); - qh->hw_info2 |= c_mask; - } else - ehci_dbg (ehci, "reused qh %p schedule\n", qh); + /* qh->ps.bw_period == 0 means every uframe */ + } else { + status = check_intr_schedule(ehci, 0, 0, qh, &c_mask, tt); + } + if (status) + goto done; + + got_it: + qh->ps.phase = (qh->ps.period ? ehci->random_frame & + (qh->ps.period - 1) : 0); + qh->ps.bw_phase = qh->ps.phase & (qh->ps.bw_period - 1); + qh->ps.phase_uf = uframe; + qh->ps.cs_mask = qh->ps.period ? + (c_mask << 8) | (1 << uframe) : + QH_SMASK; + + /* reset S-frame and (maybe) C-frame masks */ + hw->hw_info2 &= cpu_to_hc32(ehci, ~(QH_CMASK | QH_SMASK)); + hw->hw_info2 |= cpu_to_hc32(ehci, qh->ps.cs_mask); + reserve_release_intr_bandwidth(ehci, qh, 1); - /* stuff into the periodic schedule */ - status = qh_link_periodic (ehci, qh); done: return status; } @@ -813,8 +942,7 @@ static int intr_submit ( spin_lock_irqsave (&ehci->lock, flags); - if (unlikely(!test_bit(HCD_FLAG_HW_ACCESSIBLE, - &ehci_to_hcd(ehci)->flags))) { + if (unlikely(!HCD_HW_ACCESSIBLE(ehci_to_hcd(ehci)))) { status = -ESHUTDOWN; goto done_not_linked; } @@ -838,6 +966,15 @@ static int intr_submit ( qh = qh_append_tds(ehci, urb, qtd_list, epnum, &urb->ep->hcpriv); BUG_ON (qh == NULL); + /* stuff into the periodic schedule */ + if (qh->qh_state == QH_STATE_IDLE) { + qh_refresh(ehci, qh); + qh_link_periodic(ehci, qh); + } else { + /* cancel unlink wait for the qh */ + cancel_unlink_wait_intr(ehci, qh); + } + /* ... update usbfs periodic stats */ ehci_to_hcd(ehci)->self.bandwidth_int_reqs++; @@ -852,6 +989,34 @@ done_not_linked: return status; } +static void scan_intr(struct ehci_hcd *ehci) +{ + struct ehci_qh *qh; + + list_for_each_entry_safe(qh, ehci->qh_scan_next, &ehci->intr_qh_list, + intr_node) { + + /* clean any finished work for this qh */ + if (!list_empty(&qh->qtd_list)) { + int temp; + + /* + * Unlinks could happen here; completion reporting + * drops the lock. That's why ehci->qh_scan_next + * always holds the next qh to scan; if the next qh + * gets unlinked then ehci->qh_scan_next is adjusted + * in qh_unlink_periodic(). + */ + temp = qh_completions(ehci, qh); + if (unlikely(temp)) + start_unlink_intr(ehci, qh); + else if (unlikely(list_empty(&qh->qtd_list) && + qh->qh_state == QH_STATE_LINKED)) + start_unlink_intr_wait(ehci, qh); + } + } +} + /*-------------------------------------------------------------------------*/ /* ehci_iso_stream ops work with both ITD and SITD */ @@ -865,8 +1030,8 @@ iso_stream_alloc (gfp_t mem_flags) if (likely (stream != NULL)) { INIT_LIST_HEAD(&stream->td_list); INIT_LIST_HEAD(&stream->free_list); - stream->next_uframe = -1; - stream->refcount = 1; + stream->next_uframe = NO_FRAME; + stream->ps.phase = NO_FRAME; } return stream; } @@ -875,25 +1040,24 @@ static void iso_stream_init ( struct ehci_hcd *ehci, struct ehci_iso_stream *stream, - struct usb_device *dev, - int pipe, - unsigned interval + struct urb *urb ) { static const u8 smask_out [] = { 0x01, 0x03, 0x07, 0x0f, 0x1f, 0x3f }; + struct usb_device *dev = urb->dev; u32 buf1; unsigned epnum, maxp; int is_input; - long bandwidth; + unsigned tmp; /* * this might be a "high bandwidth" highspeed endpoint, * as encoded in the ep descriptor's wMaxPacket field */ - epnum = usb_pipeendpoint (pipe); - is_input = usb_pipein (pipe) ? USB_DIR_IN : 0; - maxp = usb_maxpacket(dev, pipe, !is_input); + epnum = usb_pipeendpoint(urb->pipe); + is_input = usb_pipein(urb->pipe) ? USB_DIR_IN : 0; + maxp = usb_endpoint_maxp(&urb->ep->desc); if (is_input) { buf1 = (1 << 11); } else { @@ -917,9 +1081,19 @@ iso_stream_init ( /* usbfs wants to report the average usecs per frame tied up * when transfers on this endpoint are scheduled ... */ - stream->usecs = HS_USECS_ISO (maxp); - bandwidth = stream->usecs * 8; - bandwidth /= 1 << (interval - 1); + stream->ps.usecs = HS_USECS_ISO(maxp); + + /* period for bandwidth allocation */ + tmp = min_t(unsigned, EHCI_BANDWIDTH_SIZE, + 1 << (urb->ep->desc.bInterval - 1)); + + /* Allow urb->interval to override */ + stream->ps.bw_uperiod = min_t(unsigned, tmp, urb->interval); + + stream->uperiod = urb->interval; + stream->ps.period = urb->interval >> 3; + stream->bandwidth = stream->ps.usecs * 8 / + stream->ps.bw_uperiod; } else { u32 addr; @@ -933,101 +1107,49 @@ iso_stream_init ( addr |= dev->tt->hub->devnum << 16; addr |= epnum << 8; addr |= dev->devnum; - stream->usecs = HS_USECS_ISO (maxp); + stream->ps.usecs = HS_USECS_ISO(maxp); think_time = dev->tt ? dev->tt->think_time : 0; - stream->tt_usecs = NS_TO_US (think_time + usb_calc_bus_time ( + stream->ps.tt_usecs = NS_TO_US(think_time + usb_calc_bus_time( dev->speed, is_input, 1, maxp)); hs_transfers = max (1u, (maxp + 187) / 188); if (is_input) { u32 tmp; addr |= 1 << 31; - stream->c_usecs = stream->usecs; - stream->usecs = HS_USECS_ISO (1); - stream->raw_mask = 1; + stream->ps.c_usecs = stream->ps.usecs; + stream->ps.usecs = HS_USECS_ISO(1); + stream->ps.cs_mask = 1; /* c-mask as specified in USB 2.0 11.18.4 3.c */ tmp = (1 << (hs_transfers + 2)) - 1; - stream->raw_mask |= tmp << (8 + 2); + stream->ps.cs_mask |= tmp << (8 + 2); } else - stream->raw_mask = smask_out [hs_transfers - 1]; - bandwidth = stream->usecs + stream->c_usecs; - bandwidth /= 1 << (interval + 2); + stream->ps.cs_mask = smask_out[hs_transfers - 1]; + + /* period for bandwidth allocation */ + tmp = min_t(unsigned, EHCI_BANDWIDTH_FRAMES, + 1 << (urb->ep->desc.bInterval - 1)); + + /* Allow urb->interval to override */ + stream->ps.bw_period = min_t(unsigned, tmp, urb->interval); + stream->ps.bw_uperiod = stream->ps.bw_period << 3; - /* stream->splits gets created from raw_mask later */ + stream->ps.period = urb->interval; + stream->uperiod = urb->interval << 3; + stream->bandwidth = (stream->ps.usecs + stream->ps.c_usecs) / + stream->ps.bw_period; + + /* stream->splits gets created from cs_mask later */ stream->address = cpu_to_hc32(ehci, addr); } - stream->bandwidth = bandwidth; - stream->udev = dev; + stream->ps.udev = dev; + stream->ps.ep = urb->ep; stream->bEndpointAddress = is_input | epnum; - stream->interval = interval; stream->maxp = maxp; } -static void -iso_stream_put(struct ehci_hcd *ehci, struct ehci_iso_stream *stream) -{ - stream->refcount--; - - /* free whenever just a dev->ep reference remains. - * not like a QH -- no persistent state (toggle, halt) - */ - if (stream->refcount == 1) { - int is_in; - - // BUG_ON (!list_empty(&stream->td_list)); - - while (!list_empty (&stream->free_list)) { - struct list_head *entry; - - entry = stream->free_list.next; - list_del (entry); - - /* knows about ITD vs SITD */ - if (stream->highspeed) { - struct ehci_itd *itd; - - itd = list_entry (entry, struct ehci_itd, - itd_list); - dma_pool_free (ehci->itd_pool, itd, - itd->itd_dma); - } else { - struct ehci_sitd *sitd; - - sitd = list_entry (entry, struct ehci_sitd, - sitd_list); - dma_pool_free (ehci->sitd_pool, sitd, - sitd->sitd_dma); - } - } - - is_in = (stream->bEndpointAddress & USB_DIR_IN) ? 0x10 : 0; - stream->bEndpointAddress &= 0x0f; - stream->ep->hcpriv = NULL; - - if (stream->rescheduled) { - ehci_info (ehci, "ep%d%s-iso rescheduled " - "%lu times in %lu seconds\n", - stream->bEndpointAddress, is_in ? "in" : "out", - stream->rescheduled, - ((jiffies - stream->start)/HZ) - ); - } - - kfree(stream); - } -} - -static inline struct ehci_iso_stream * -iso_stream_get (struct ehci_iso_stream *stream) -{ - if (likely (stream != NULL)) - stream->refcount++; - return stream; -} - static struct ehci_iso_stream * iso_stream_find (struct ehci_hcd *ehci, struct urb *urb) { @@ -1048,24 +1170,18 @@ iso_stream_find (struct ehci_hcd *ehci, struct urb *urb) if (unlikely (stream == NULL)) { stream = iso_stream_alloc(GFP_ATOMIC); if (likely (stream != NULL)) { - /* dev->ep owns the initial refcount */ ep->hcpriv = stream; - stream->ep = ep; - iso_stream_init(ehci, stream, urb->dev, urb->pipe, - urb->interval); + iso_stream_init(ehci, stream, urb); } - /* if dev->ep [epnum] is a QH, info1.maxpacket is nonzero */ - } else if (unlikely (stream->hw_info1 != 0)) { + /* if dev->ep [epnum] is a QH, hw is set */ + } else if (unlikely (stream->hw != NULL)) { ehci_dbg (ehci, "dev %s ep%d%s, not iso??\n", urb->dev->devpath, epnum, usb_pipein(urb->pipe) ? "in" : "out"); stream = NULL; } - /* caller guarantees an eventual matching iso_stream_put */ - stream = iso_stream_get (stream); - spin_unlock_irqrestore (&ehci->lock, flags); return stream; } @@ -1100,7 +1216,7 @@ itd_sched_init( dma_addr_t dma = urb->transfer_dma; /* how many uframes are needed for these transfers */ - iso_sched->span = urb->number_of_packets * stream->interval; + iso_sched->span = urb->number_of_packets * stream->uperiod; /* figure out per-uframe itd fields that we'll need later * when we fit new itds into the schedule. @@ -1173,33 +1289,33 @@ itd_urb_transaction ( spin_lock_irqsave (&ehci->lock, flags); for (i = 0; i < num_itds; i++) { - /* free_list.next might be cache-hot ... but maybe - * the HC caches it too. avoid that issue for now. + /* + * Use iTDs from the free list, but not iTDs that may + * still be in use by the hardware. */ - - /* prefer previously-allocated itds */ - if (likely (!list_empty(&stream->free_list))) { - itd = list_entry (stream->free_list.prev, + if (likely(!list_empty(&stream->free_list))) { + itd = list_first_entry(&stream->free_list, struct ehci_itd, itd_list); + if (itd->frame == ehci->now_frame) + goto alloc_itd; list_del (&itd->itd_list); itd_dma = itd->itd_dma; - } else - itd = NULL; - - if (!itd) { + } else { + alloc_itd: spin_unlock_irqrestore (&ehci->lock, flags); itd = dma_pool_alloc (ehci->itd_pool, mem_flags, &itd_dma); spin_lock_irqsave (&ehci->lock, flags); + if (!itd) { + iso_sched_free(stream, sched); + spin_unlock_irqrestore(&ehci->lock, flags); + return -ENOMEM; + } } - if (unlikely (NULL == itd)) { - iso_sched_free (stream, sched); - spin_unlock_irqrestore (&ehci->lock, flags); - return -ENOMEM; - } memset (itd, 0, sizeof *itd); itd->itd_dma = itd_dma; + itd->frame = NO_FRAME; list_add (&itd->itd_list, &sched->td_list); } spin_unlock_irqrestore (&ehci->lock, flags); @@ -1212,102 +1328,155 @@ itd_urb_transaction ( /*-------------------------------------------------------------------------*/ +static void reserve_release_iso_bandwidth(struct ehci_hcd *ehci, + struct ehci_iso_stream *stream, int sign) +{ + unsigned uframe; + unsigned i, j; + unsigned s_mask, c_mask, m; + int usecs = stream->ps.usecs; + int c_usecs = stream->ps.c_usecs; + int tt_usecs = stream->ps.tt_usecs; + struct ehci_tt *tt; + + if (stream->ps.phase == NO_FRAME) /* Bandwidth wasn't reserved */ + return; + uframe = stream->ps.bw_phase << 3; + + bandwidth_dbg(ehci, sign, "iso", &stream->ps); + + if (sign < 0) { /* Release bandwidth */ + usecs = -usecs; + c_usecs = -c_usecs; + tt_usecs = -tt_usecs; + } + + if (!stream->splits) { /* High speed */ + for (i = uframe + stream->ps.phase_uf; i < EHCI_BANDWIDTH_SIZE; + i += stream->ps.bw_uperiod) + ehci->bandwidth[i] += usecs; + + } else { /* Full speed */ + s_mask = stream->ps.cs_mask; + c_mask = s_mask >> 8; + + /* NOTE: adjustment needed for frame overflow */ + for (i = uframe; i < EHCI_BANDWIDTH_SIZE; + i += stream->ps.bw_uperiod) { + for ((j = stream->ps.phase_uf, m = 1 << j); j < 8; + (++j, m <<= 1)) { + if (s_mask & m) + ehci->bandwidth[i+j] += usecs; + else if (c_mask & m) + ehci->bandwidth[i+j] += c_usecs; + } + } + + tt = find_tt(stream->ps.udev); + if (sign > 0) + list_add_tail(&stream->ps.ps_list, &tt->ps_list); + else + list_del(&stream->ps.ps_list); + + for (i = uframe >> 3; i < EHCI_BANDWIDTH_FRAMES; + i += stream->ps.bw_period) + tt->bandwidth[i] += tt_usecs; + } +} + static inline int itd_slot_ok ( struct ehci_hcd *ehci, - u32 mod, - u32 uframe, - u8 usecs, - u32 period + struct ehci_iso_stream *stream, + unsigned uframe ) { - uframe %= period; - do { - /* can't commit more than 80% periodic == 100 usec */ - if (periodic_usecs (ehci, uframe >> 3, uframe & 0x7) - > (100 - usecs)) - return 0; + unsigned usecs; + + /* convert "usecs we need" to "max already claimed" */ + usecs = ehci->uframe_periodic_max - stream->ps.usecs; - /* we know urb->interval is 2^N uframes */ - uframe += period; - } while (uframe < mod); + for (uframe &= stream->ps.bw_uperiod - 1; uframe < EHCI_BANDWIDTH_SIZE; + uframe += stream->ps.bw_uperiod) { + if (ehci->bandwidth[uframe] > usecs) + return 0; + } return 1; } static inline int sitd_slot_ok ( struct ehci_hcd *ehci, - u32 mod, struct ehci_iso_stream *stream, - u32 uframe, + unsigned uframe, struct ehci_iso_sched *sched, - u32 period_uframes + struct ehci_tt *tt ) { - u32 mask, tmp; - u32 frame, uf; + unsigned mask, tmp; + unsigned frame, uf; + + mask = stream->ps.cs_mask << (uframe & 7); - mask = stream->raw_mask << (uframe & 7); + /* for OUT, don't wrap SSPLIT into H-microframe 7 */ + if (((stream->ps.cs_mask & 0xff) << (uframe & 7)) >= (1 << 7)) + return 0; /* for IN, don't wrap CSPLIT into the next frame */ if (mask & ~0xffff) return 0; - /* this multi-pass logic is simple, but performance may - * suffer when the schedule data isn't cached. - */ - /* check bandwidth */ - uframe %= period_uframes; - do { - u32 max_used; - - frame = uframe >> 3; - uf = uframe & 7; + uframe &= stream->ps.bw_uperiod - 1; + frame = uframe >> 3; #ifdef CONFIG_USB_EHCI_TT_NEWSCHED - /* The tt's fullspeed bus bandwidth must be available. - * tt_available scheduling guarantees 10+% for control/bulk. - */ - if (!tt_available (ehci, period_uframes << 3, - stream->udev, frame, uf, stream->tt_usecs)) - return 0; + /* The tt's fullspeed bus bandwidth must be available. + * tt_available scheduling guarantees 10+% for control/bulk. + */ + uf = uframe & 7; + if (!tt_available(ehci, &stream->ps, tt, frame, uf)) + return 0; #else - /* tt must be idle for start(s), any gap, and csplit. - * assume scheduling slop leaves 10+% for control/bulk. - */ - if (!tt_no_collision (ehci, period_uframes << 3, - stream->udev, frame, mask)) - return 0; + /* tt must be idle for start(s), any gap, and csplit. + * assume scheduling slop leaves 10+% for control/bulk. + */ + if (!tt_no_collision(ehci, stream->ps.bw_period, + stream->ps.udev, frame, mask)) + return 0; #endif + do { + unsigned max_used; + unsigned i; + /* check starts (OUT uses more than one) */ - max_used = 100 - stream->usecs; - for (tmp = stream->raw_mask & 0xff; tmp; tmp >>= 1, uf++) { - if (periodic_usecs (ehci, frame, uf) > max_used) + uf = uframe; + max_used = ehci->uframe_periodic_max - stream->ps.usecs; + for (tmp = stream->ps.cs_mask & 0xff; tmp; tmp >>= 1, uf++) { + if (ehci->bandwidth[uf] > max_used) return 0; } /* for IN, check CSPLIT */ - if (stream->c_usecs) { - uf = uframe & 7; - max_used = 100 - stream->c_usecs; - do { - tmp = 1 << uf; - tmp <<= 8; - if ((stream->raw_mask & tmp) == 0) + if (stream->ps.c_usecs) { + max_used = ehci->uframe_periodic_max - + stream->ps.c_usecs; + uf = uframe & ~7; + tmp = 1 << (2+8); + for (i = (uframe & 7) + 2; i < 8; (++i, tmp <<= 1)) { + if ((stream->ps.cs_mask & tmp) == 0) continue; - if (periodic_usecs (ehci, frame, uf) - > max_used) + if (ehci->bandwidth[uf+i] > max_used) return 0; - } while (++uf < 8); + } } - /* we know urb->interval is 2^N uframes */ - uframe += period_uframes; - } while (uframe < mod); + uframe += stream->ps.bw_uperiod; + } while (uframe < EHCI_BANDWIDTH_SIZE); - stream->splits = cpu_to_hc32(ehci, stream->raw_mask << (uframe & 7)); + stream->ps.cs_mask <<= uframe & 7; + stream->splits = cpu_to_hc32(ehci, stream->ps.cs_mask); return 1; } @@ -1322,8 +1491,6 @@ sitd_slot_ok ( * given EHCI_TUNE_FLS and the slop). Or, write a smarter scheduler! */ -#define SCHEDULE_SLOP 10 /* frames */ - static int iso_stream_schedule ( struct ehci_hcd *ehci, @@ -1331,98 +1498,189 @@ iso_stream_schedule ( struct ehci_iso_stream *stream ) { - u32 now, start, max, period; - int status; + u32 now, base, next, start, period, span, now2; + u32 wrap = 0, skip = 0; + int status = 0; unsigned mod = ehci->periodic_size << 3; struct ehci_iso_sched *sched = urb->hcpriv; + bool empty = list_empty(&stream->td_list); + bool new_stream = false; - if (sched->span > (mod - 8 * SCHEDULE_SLOP)) { - ehci_dbg (ehci, "iso request %p too long\n", urb); - status = -EFBIG; - goto fail; - } + period = stream->uperiod; + span = sched->span; + if (!stream->highspeed) + span <<= 3; - if ((stream->depth + sched->span) > mod) { - ehci_dbg (ehci, "request %p would overflow (%d+%d>%d)\n", - urb, stream->depth, sched->span, mod); - status = -EFBIG; - goto fail; + /* Start a new isochronous stream? */ + if (unlikely(empty && !hcd_periodic_completion_in_progress( + ehci_to_hcd(ehci), urb->ep))) { + + /* Schedule the endpoint */ + if (stream->ps.phase == NO_FRAME) { + int done = 0; + struct ehci_tt *tt = find_tt(stream->ps.udev); + + if (IS_ERR(tt)) { + status = PTR_ERR(tt); + goto fail; + } + compute_tt_budget(ehci->tt_budget, tt); + + start = ((-(++ehci->random_frame)) << 3) & (period - 1); + + /* find a uframe slot with enough bandwidth. + * Early uframes are more precious because full-speed + * iso IN transfers can't use late uframes, + * and therefore they should be allocated last. + */ + next = start; + start += period; + do { + start--; + /* check schedule: enough space? */ + if (stream->highspeed) { + if (itd_slot_ok(ehci, stream, start)) + done = 1; + } else { + if ((start % 8) >= 6) + continue; + if (sitd_slot_ok(ehci, stream, start, + sched, tt)) + done = 1; + } + } while (start > next && !done); + + /* no room in the schedule */ + if (!done) { + ehci_dbg(ehci, "iso sched full %p", urb); + status = -ENOSPC; + goto fail; + } + stream->ps.phase = (start >> 3) & + (stream->ps.period - 1); + stream->ps.bw_phase = stream->ps.phase & + (stream->ps.bw_period - 1); + stream->ps.phase_uf = start & 7; + reserve_release_iso_bandwidth(ehci, stream, 1); + } + + /* New stream is already scheduled; use the upcoming slot */ + else { + start = (stream->ps.phase << 3) + stream->ps.phase_uf; + } + + stream->next_uframe = start; + new_stream = true; } - now = ehci_readl(ehci, &ehci->regs->frame_index) % mod; + now = ehci_read_frame_index(ehci) & (mod - 1); - /* when's the last uframe this urb could start? */ - max = now + mod; + /* Take the isochronous scheduling threshold into account */ + if (ehci->i_thresh) + next = now + ehci->i_thresh; /* uframe cache */ + else + next = (now + 2 + 7) & ~0x07; /* full frame cache */ - /* typical case: reuse current schedule. stream is still active, - * and no gaps from host falling behind (irq delays etc) + /* + * Use ehci->last_iso_frame as the base. There can't be any + * TDs scheduled for earlier than that. */ - if (likely (!list_empty (&stream->td_list))) { - start = stream->next_uframe; - if (start < now) - start += mod; - if (likely ((start + sched->span) < max)) - goto ready; - /* else fell behind; someday, try to reschedule */ - status = -EL2NSYNC; - goto fail; - } + base = ehci->last_iso_frame << 3; + next = (next - base) & (mod - 1); + start = (stream->next_uframe - base) & (mod - 1); + + if (unlikely(new_stream)) + goto do_ASAP; - /* need to schedule; when's the next (u)frame we could start? - * this is bigger than ehci->i_thresh allows; scheduling itself - * isn't free, the slop should handle reasonably slow cpus. it - * can also help high bandwidth if the dma and irq loads don't - * jump until after the queue is primed. + /* + * Typical case: reuse current schedule, stream may still be active. + * Hopefully there are no gaps from the host falling behind + * (irq delays etc). If there are, the behavior depends on + * whether URB_ISO_ASAP is set. */ - start = SCHEDULE_SLOP * 8 + (now & ~0x07); - start %= mod; - stream->next_uframe = start; + now2 = (now - base) & (mod - 1); - /* NOTE: assumes URB_ISO_ASAP, to limit complexity/bugs */ + /* Is the schedule already full? */ + if (unlikely(!empty && start < period)) { + ehci_dbg(ehci, "iso sched full %p (%u-%u < %u mod %u)\n", + urb, stream->next_uframe, base, period, mod); + status = -ENOSPC; + goto fail; + } - period = urb->interval; - if (!stream->highspeed) - period <<= 3; + /* Is the next packet scheduled after the base time? */ + if (likely(!empty || start <= now2 + period)) { - /* find a uframe slot with enough bandwidth */ - for (; start < (stream->next_uframe + period); start++) { - int enough_space; + /* URB_ISO_ASAP: make sure that start >= next */ + if (unlikely(start < next && + (urb->transfer_flags & URB_ISO_ASAP))) + goto do_ASAP; - /* check schedule: enough space? */ - if (stream->highspeed) - enough_space = itd_slot_ok (ehci, mod, start, - stream->usecs, period); - else { - if ((start % 8) >= 6) - continue; - enough_space = sitd_slot_ok (ehci, mod, stream, - start, sched, period); - } + /* Otherwise use start, if it's not in the past */ + if (likely(start >= now2)) + goto use_start; - /* schedule it here if there's enough bandwidth */ - if (enough_space) { - stream->next_uframe = start % mod; - goto ready; - } + /* Otherwise we got an underrun while the queue was empty */ + } else { + if (urb->transfer_flags & URB_ISO_ASAP) + goto do_ASAP; + wrap = mod; + now2 += mod; } - /* no room in the schedule */ - ehci_dbg (ehci, "iso %ssched full %p (now %d max %d)\n", - list_empty (&stream->td_list) ? "" : "re", - urb, now, max); - status = -ENOSPC; + /* How many uframes and packets do we need to skip? */ + skip = (now2 - start + period - 1) & -period; + if (skip >= span) { /* Entirely in the past? */ + ehci_dbg(ehci, "iso underrun %p (%u+%u < %u) [%u]\n", + urb, start + base, span - period, now2 + base, + base); + + /* Try to keep the last TD intact for scanning later */ + skip = span - period; + + /* Will it come before the current scan position? */ + if (empty) { + skip = span; /* Skip the entire URB */ + status = 1; /* and give it back immediately */ + iso_sched_free(stream, sched); + sched = NULL; + } + } + urb->error_count = skip / period; + if (sched) + sched->first_packet = urb->error_count; + goto use_start; + + do_ASAP: + /* Use the first slot after "next" */ + start = next + ((start - next) & (period - 1)); + + use_start: + /* Tried to schedule too far into the future? */ + if (unlikely(start + span - period >= mod + wrap)) { + ehci_dbg(ehci, "request %p would overflow (%u+%u >= %u)\n", + urb, start, span - period, mod + wrap); + status = -EFBIG; + goto fail; + } -fail: - iso_sched_free (stream, sched); - urb->hcpriv = NULL; - return status; + start += base; + stream->next_uframe = (start + skip) & (mod - 1); -ready: /* report high speed start in uframes; full speed, in frames */ - urb->start_frame = stream->next_uframe; + urb->start_frame = start & (mod - 1); if (!stream->highspeed) urb->start_frame >>= 3; - return 0; + + /* Make sure scan_isoc() sees these */ + if (ehci->isoc_count == 0) + ehci->last_iso_frame = now >> 3; + return status; + + fail: + iso_sched_free(stream, sched); + urb->hcpriv = NULL; + return status; } /*-------------------------------------------------------------------------*/ @@ -1480,18 +1738,31 @@ itd_patch( static inline void itd_link (struct ehci_hcd *ehci, unsigned frame, struct ehci_itd *itd) { - /* always prepend ITD/SITD ... only QH tree is order-sensitive */ - itd->itd_next = ehci->pshadow [frame]; - itd->hw_next = ehci->periodic [frame]; - ehci->pshadow [frame].itd = itd; + union ehci_shadow *prev = &ehci->pshadow[frame]; + __hc32 *hw_p = &ehci->periodic[frame]; + union ehci_shadow here = *prev; + __hc32 type = 0; + + /* skip any iso nodes which might belong to previous microframes */ + while (here.ptr) { + type = Q_NEXT_TYPE(ehci, *hw_p); + if (type == cpu_to_hc32(ehci, Q_TYPE_QH)) + break; + prev = periodic_next_shadow(ehci, prev, type); + hw_p = shadow_next_periodic(ehci, &here, type); + here = *prev; + } + + itd->itd_next = here; + itd->hw_next = *hw_p; + prev->itd = itd; itd->frame = frame; wmb (); - ehci->periodic[frame] = cpu_to_hc32(ehci, itd->itd_dma | Q_TYPE_ITD); + *hw_p = cpu_to_hc32(ehci, itd->itd_dma | Q_TYPE_ITD); } /* fit urb's itds into the selected schedule slot; activate as needed */ -static int -itd_link_urb ( +static void itd_link_urb( struct ehci_hcd *ehci, struct urb *urb, unsigned mod, @@ -1503,23 +1774,22 @@ itd_link_urb ( struct ehci_iso_sched *iso_sched = urb->hcpriv; struct ehci_itd *itd; - next_uframe = stream->next_uframe % mod; + next_uframe = stream->next_uframe & (mod - 1); - if (unlikely (list_empty(&stream->td_list))) { + if (unlikely (list_empty(&stream->td_list))) ehci_to_hcd(ehci)->self.bandwidth_allocated += stream->bandwidth; - ehci_vdbg (ehci, - "schedule devp %s ep%d%s-iso period %d start %d.%d\n", - urb->dev->devpath, stream->bEndpointAddress & 0x0f, - (stream->bEndpointAddress & USB_DIR_IN) ? "in" : "out", - urb->interval, - next_uframe >> 3, next_uframe & 0x7); - stream->start = jiffies; + + if (ehci_to_hcd(ehci)->self.bandwidth_isoc_reqs == 0) { + if (ehci->amd_pll_fix == 1) + usb_amd_quirk_pll_disable(); } + ehci_to_hcd(ehci)->self.bandwidth_isoc_reqs++; /* fill iTDs uframe by uframe */ - for (packet = 0, itd = NULL; packet < urb->number_of_packets; ) { + for (packet = iso_sched->first_packet, itd = NULL; + packet < urb->number_of_packets;) { if (itd == NULL) { /* ASSERT: we have all necessary itds */ // BUG_ON (list_empty (&iso_sched->td_list)); @@ -1529,8 +1799,8 @@ itd_link_urb ( itd = list_entry (iso_sched->td_list.next, struct ehci_itd, itd_list); list_move_tail (&itd->itd_list, &stream->td_list); - itd->stream = iso_stream_get (stream); - itd->urb = usb_get_urb (urb); + itd->stream = stream; + itd->urb = urb; itd_init (ehci, stream, itd); } @@ -1539,15 +1809,14 @@ itd_link_urb ( itd_patch(ehci, itd, iso_sched, packet, uframe); - next_uframe += stream->interval; - stream->depth += stream->interval; - next_uframe %= mod; + next_uframe += stream->uperiod; + next_uframe &= mod - 1; packet++; /* link completed itds into the schedule */ if (((next_uframe >> 3) != frame) || packet == urb->number_of_packets) { - itd_link (ehci, frame % ehci->periodic_size, itd); + itd_link(ehci, frame & (ehci->periodic_size - 1), itd); itd = NULL; } } @@ -1555,12 +1824,10 @@ itd_link_urb ( /* don't need that schedule data any more */ iso_sched_free (stream, iso_sched); - urb->hcpriv = NULL; + urb->hcpriv = stream; - timer_action (ehci, TIMER_IO_WATCHDOG); - if (unlikely (!ehci->periodic_sched++)) - return enable_periodic (ehci); - return 0; + ++ehci->isoc_count; + enable_periodic(ehci); } #define ISO_ERRS (EHCI_ISOC_BUF_ERR | EHCI_ISOC_BABBLE | EHCI_ISOC_XACTERR) @@ -1575,11 +1842,8 @@ itd_link_urb ( * (b) only this endpoint's completions submit URBs. It seems some silicon * corrupts things if you reuse completed descriptors very quickly... */ -static unsigned -itd_complete ( - struct ehci_hcd *ehci, - struct ehci_itd *itd -) { +static bool itd_complete(struct ehci_hcd *ehci, struct ehci_itd *itd) +{ struct urb *urb = itd->urb; struct usb_iso_packet_descriptor *desc; u32 t; @@ -1587,7 +1851,7 @@ itd_complete ( int urb_index = -1; struct ehci_iso_stream *stream = itd->stream; struct usb_device *dev; - unsigned retval = false; + bool retval = false; /* for each uframe with a packet */ for (uframe = 0; uframe < 8; uframe++) { @@ -1598,7 +1862,6 @@ itd_complete ( t = hc32_to_cpup(ehci, &itd->hw_transaction [uframe]); itd->hw_transaction [uframe] = 0; - stream->depth -= stream->interval; /* report transfer status */ if (unlikely (t & ISO_ERRS)) { @@ -1613,11 +1876,17 @@ itd_complete ( desc->status = -EPROTO; /* HC need not update length with this error */ - if (!(t & EHCI_ISOC_BABBLE)) - desc->actual_length = EHCI_ITD_LENGTH (t); + if (!(t & EHCI_ISOC_BABBLE)) { + desc->actual_length = EHCI_ITD_LENGTH(t); + urb->actual_length += desc->actual_length; + } } else if (likely ((t & EHCI_ISOC_ACTIVE) == 0)) { desc->status = 0; - desc->actual_length = EHCI_ITD_LENGTH (t); + desc->actual_length = EHCI_ITD_LENGTH(t); + urb->actual_length += desc->actual_length; + } else { + /* URB was too late */ + urb->error_count++; } } @@ -1635,25 +1904,32 @@ itd_complete ( ehci_urb_done(ehci, urb, 0); retval = true; urb = NULL; - ehci->periodic_sched--; + + --ehci->isoc_count; + disable_periodic(ehci); + ehci_to_hcd(ehci)->self.bandwidth_isoc_reqs--; + if (ehci_to_hcd(ehci)->self.bandwidth_isoc_reqs == 0) { + if (ehci->amd_pll_fix == 1) + usb_amd_quirk_pll_enable(); + } - if (unlikely (list_empty (&stream->td_list))) { + if (unlikely(list_is_singular(&stream->td_list))) ehci_to_hcd(ehci)->self.bandwidth_allocated -= stream->bandwidth; - ehci_vdbg (ehci, - "deschedule devp %s ep%d%s-iso\n", - dev->devpath, stream->bEndpointAddress & 0x0f, - (stream->bEndpointAddress & USB_DIR_IN) ? "in" : "out"); - } - iso_stream_put (ehci, stream); - /* OK to recycle this ITD now that its completion callback ran. */ + done: - usb_put_urb(urb); itd->urb = NULL; - itd->stream = NULL; - list_move(&itd->itd_list, &stream->free_list); - iso_stream_put(ehci, stream); + + /* Add to the end of the free list for later reuse */ + list_move_tail(&itd->itd_list, &stream->free_list); + + /* Recycle the iTDs when the pipeline is empty (ep no longer in use) */ + if (list_empty(&stream->td_list)) { + list_splice_tail_init(&stream->free_list, + &ehci->cached_itd_list); + start_free_itds(ehci); + } return retval; } @@ -1673,16 +1949,16 @@ static int itd_submit (struct ehci_hcd *ehci, struct urb *urb, ehci_dbg (ehci, "can't get iso stream\n"); return -ENOMEM; } - if (unlikely (urb->interval != stream->interval)) { + if (unlikely(urb->interval != stream->uperiod)) { ehci_dbg (ehci, "can't change iso interval %d --> %d\n", - stream->interval, urb->interval); + stream->uperiod, urb->interval); goto done; } #ifdef EHCI_URB_TRACE ehci_dbg (ehci, "%s %s urb %p ep%d%s len %d, %d pkts %d uframes [%p]\n", - __FUNCTION__, urb->dev->devpath, urb, + __func__, urb->dev->devpath, urb, usb_pipeendpoint (urb->pipe), usb_pipein (urb->pipe) ? "in" : "out", urb->transfer_buffer_length, @@ -1699,8 +1975,7 @@ static int itd_submit (struct ehci_hcd *ehci, struct urb *urb, /* schedule ... need to lock */ spin_lock_irqsave (&ehci->lock, flags); - if (unlikely(!test_bit(HCD_FLAG_HW_ACCESSIBLE, - &ehci_to_hcd(ehci)->flags))) { + if (unlikely(!HCD_HW_ACCESSIBLE(ehci_to_hcd(ehci)))) { status = -ESHUTDOWN; goto done_not_linked; } @@ -1708,16 +1983,17 @@ static int itd_submit (struct ehci_hcd *ehci, struct urb *urb, if (unlikely(status)) goto done_not_linked; status = iso_stream_schedule(ehci, urb, stream); - if (likely (status == 0)) + if (likely(status == 0)) { itd_link_urb (ehci, urb, ehci->periodic_size << 3, stream); - else + } else if (status > 0) { + status = 0; + ehci_urb_done(ehci, urb, 0); + } else { usb_hcd_unlink_urb_from_ep(ehci_to_hcd(ehci), urb); -done_not_linked: + } + done_not_linked: spin_unlock_irqrestore (&ehci->lock, flags); - -done: - if (unlikely (status < 0)) - iso_stream_put (ehci, stream); + done: return status; } @@ -1740,7 +2016,7 @@ sitd_sched_init( dma_addr_t dma = urb->transfer_dma; /* how many frames are needed for these transfers */ - iso_sched->span = urb->number_of_packets * stream->interval; + iso_sched->span = urb->number_of_packets * stream->ps.period; /* figure out per-frame sitd fields that we'll need later * when we fit new sitds into the schedule. @@ -1806,33 +2082,33 @@ sitd_urb_transaction ( * means we never need two sitds for full speed packets. */ - /* free_list.next might be cache-hot ... but maybe - * the HC caches it too. avoid that issue for now. + /* + * Use siTDs from the free list, but not siTDs that may + * still be in use by the hardware. */ - - /* prefer previously-allocated sitds */ - if (!list_empty(&stream->free_list)) { - sitd = list_entry (stream->free_list.prev, + if (likely(!list_empty(&stream->free_list))) { + sitd = list_first_entry(&stream->free_list, struct ehci_sitd, sitd_list); + if (sitd->frame == ehci->now_frame) + goto alloc_sitd; list_del (&sitd->sitd_list); sitd_dma = sitd->sitd_dma; - } else - sitd = NULL; - - if (!sitd) { + } else { + alloc_sitd: spin_unlock_irqrestore (&ehci->lock, flags); sitd = dma_pool_alloc (ehci->sitd_pool, mem_flags, &sitd_dma); spin_lock_irqsave (&ehci->lock, flags); + if (!sitd) { + iso_sched_free(stream, iso_sched); + spin_unlock_irqrestore(&ehci->lock, flags); + return -ENOMEM; + } } - if (!sitd) { - iso_sched_free (stream, iso_sched); - spin_unlock_irqrestore (&ehci->lock, flags); - return -ENOMEM; - } memset (sitd, 0, sizeof *sitd); sitd->sitd_dma = sitd_dma; + sitd->frame = NO_FRAME; list_add (&sitd->sitd_list, &iso_sched->td_list); } @@ -1888,8 +2164,7 @@ sitd_link (struct ehci_hcd *ehci, unsigned frame, struct ehci_sitd *sitd) } /* fit urb's sitds into the selected schedule slot; activate as needed */ -static int -sitd_link_urb ( +static void sitd_link_urb( struct ehci_hcd *ehci, struct urb *urb, unsigned mod, @@ -1903,22 +2178,20 @@ sitd_link_urb ( next_uframe = stream->next_uframe; - if (list_empty(&stream->td_list)) { + if (list_empty(&stream->td_list)) /* usbfs ignores TT bandwidth */ ehci_to_hcd(ehci)->self.bandwidth_allocated += stream->bandwidth; - ehci_vdbg (ehci, - "sched devp %s ep%d%s-iso [%d] %dms/%04x\n", - urb->dev->devpath, stream->bEndpointAddress & 0x0f, - (stream->bEndpointAddress & USB_DIR_IN) ? "in" : "out", - (next_uframe >> 3) % ehci->periodic_size, - stream->interval, hc32_to_cpu(ehci, stream->splits)); - stream->start = jiffies; + + if (ehci_to_hcd(ehci)->self.bandwidth_isoc_reqs == 0) { + if (ehci->amd_pll_fix == 1) + usb_amd_quirk_pll_disable(); } + ehci_to_hcd(ehci)->self.bandwidth_isoc_reqs++; /* fill sITDs frame by frame */ - for (packet = 0, sitd = NULL; + for (packet = sched->first_packet, sitd = NULL; packet < urb->number_of_packets; packet++) { @@ -1930,26 +2203,23 @@ sitd_link_urb ( sitd = list_entry (sched->td_list.next, struct ehci_sitd, sitd_list); list_move_tail (&sitd->sitd_list, &stream->td_list); - sitd->stream = iso_stream_get (stream); - sitd->urb = usb_get_urb (urb); + sitd->stream = stream; + sitd->urb = urb; sitd_patch(ehci, stream, sitd, sched, packet); - sitd_link (ehci, (next_uframe >> 3) % ehci->periodic_size, + sitd_link(ehci, (next_uframe >> 3) & (ehci->periodic_size - 1), sitd); - next_uframe += stream->interval << 3; - stream->depth += stream->interval << 3; + next_uframe += stream->uperiod; } - stream->next_uframe = next_uframe % mod; + stream->next_uframe = next_uframe & (mod - 1); /* don't need that schedule data any more */ iso_sched_free (stream, sched); - urb->hcpriv = NULL; + urb->hcpriv = stream; - timer_action (ehci, TIMER_IO_WATCHDOG); - if (!ehci->periodic_sched++) - return enable_periodic (ehci); - return 0; + ++ehci->isoc_count; + enable_periodic(ehci); } /*-------------------------------------------------------------------------*/ @@ -1967,25 +2237,22 @@ sitd_link_urb ( * (b) only this endpoint's completions submit URBs. It seems some silicon * corrupts things if you reuse completed descriptors very quickly... */ -static unsigned -sitd_complete ( - struct ehci_hcd *ehci, - struct ehci_sitd *sitd -) { +static bool sitd_complete(struct ehci_hcd *ehci, struct ehci_sitd *sitd) +{ struct urb *urb = sitd->urb; struct usb_iso_packet_descriptor *desc; u32 t; int urb_index = -1; struct ehci_iso_stream *stream = sitd->stream; struct usb_device *dev; - unsigned retval = false; + bool retval = false; urb_index = sitd->index; desc = &urb->iso_frame_desc [urb_index]; t = hc32_to_cpup(ehci, &sitd->hw_results); /* report transfer status */ - if (t & SITD_ERRS) { + if (unlikely(t & SITD_ERRS)) { urb->error_count++; if (t & SITD_STS_DBE) desc->status = usb_pipein (urb->pipe) @@ -1995,11 +2262,14 @@ sitd_complete ( desc->status = -EOVERFLOW; else /* XACT, MMF, etc */ desc->status = -EPROTO; + } else if (unlikely(t & SITD_STS_ACTIVE)) { + /* URB was too late */ + urb->error_count++; } else { desc->status = 0; - desc->actual_length = desc->length - SITD_LENGTH (t); + desc->actual_length = desc->length - SITD_LENGTH(t); + urb->actual_length += desc->actual_length; } - stream->depth -= stream->interval << 3; /* handle completion now? */ if ((urb_index + 1) != urb->number_of_packets) @@ -2015,25 +2285,32 @@ sitd_complete ( ehci_urb_done(ehci, urb, 0); retval = true; urb = NULL; - ehci->periodic_sched--; + + --ehci->isoc_count; + disable_periodic(ehci); + ehci_to_hcd(ehci)->self.bandwidth_isoc_reqs--; + if (ehci_to_hcd(ehci)->self.bandwidth_isoc_reqs == 0) { + if (ehci->amd_pll_fix == 1) + usb_amd_quirk_pll_enable(); + } - if (list_empty (&stream->td_list)) { + if (list_is_singular(&stream->td_list)) ehci_to_hcd(ehci)->self.bandwidth_allocated -= stream->bandwidth; - ehci_vdbg (ehci, - "deschedule devp %s ep%d%s-iso\n", - dev->devpath, stream->bEndpointAddress & 0x0f, - (stream->bEndpointAddress & USB_DIR_IN) ? "in" : "out"); - } - iso_stream_put (ehci, stream); - /* OK to recycle this SITD now that its completion callback ran. */ + done: - usb_put_urb(urb); sitd->urb = NULL; - sitd->stream = NULL; - list_move(&sitd->sitd_list, &stream->free_list); - iso_stream_put(ehci, stream); + + /* Add to the end of the free list for later reuse */ + list_move_tail(&sitd->sitd_list, &stream->free_list); + + /* Recycle the siTDs when the pipeline is empty (ep no longer in use) */ + if (list_empty(&stream->td_list)) { + list_splice_tail_init(&stream->free_list, + &ehci->cached_sitd_list); + start_free_itds(ehci); + } return retval; } @@ -2052,9 +2329,9 @@ static int sitd_submit (struct ehci_hcd *ehci, struct urb *urb, ehci_dbg (ehci, "can't get iso stream\n"); return -ENOMEM; } - if (urb->interval != stream->interval) { + if (urb->interval != stream->ps.period) { ehci_dbg (ehci, "can't change iso interval %d --> %d\n", - stream->interval, urb->interval); + stream->ps.period, urb->interval); goto done; } @@ -2076,8 +2353,7 @@ static int sitd_submit (struct ehci_hcd *ehci, struct urb *urb, /* schedule ... need to lock */ spin_lock_irqsave (&ehci->lock, flags); - if (unlikely(!test_bit(HCD_FLAG_HW_ACCESSIBLE, - &ehci_to_hcd(ehci)->flags))) { + if (unlikely(!HCD_HW_ACCESSIBLE(ehci_to_hcd(ehci)))) { status = -ESHUTDOWN; goto done_not_linked; } @@ -2085,47 +2361,47 @@ static int sitd_submit (struct ehci_hcd *ehci, struct urb *urb, if (unlikely(status)) goto done_not_linked; status = iso_stream_schedule(ehci, urb, stream); - if (status == 0) + if (likely(status == 0)) { sitd_link_urb (ehci, urb, ehci->periodic_size << 3, stream); - else + } else if (status > 0) { + status = 0; + ehci_urb_done(ehci, urb, 0); + } else { usb_hcd_unlink_urb_from_ep(ehci_to_hcd(ehci), urb); -done_not_linked: + } + done_not_linked: spin_unlock_irqrestore (&ehci->lock, flags); - -done: - if (status < 0) - iso_stream_put (ehci, stream); + done: return status; } /*-------------------------------------------------------------------------*/ -static void -scan_periodic (struct ehci_hcd *ehci) +static void scan_isoc(struct ehci_hcd *ehci) { - unsigned frame, clock, now_uframe, mod; - unsigned modified; - - mod = ehci->periodic_size << 3; + unsigned uf, now_frame, frame; + unsigned fmask = ehci->periodic_size - 1; + bool modified, live; /* * When running, scan from last scan point up to "now" * else clean up by scanning everything that's left. * Touches as few pages as possible: cache-friendly. */ - now_uframe = ehci->next_uframe; - if (HC_IS_RUNNING (ehci_to_hcd(ehci)->state)) - clock = ehci_readl(ehci, &ehci->regs->frame_index); - else - clock = now_uframe + mod - 1; - clock %= mod; + if (ehci->rh_state >= EHCI_RH_RUNNING) { + uf = ehci_read_frame_index(ehci); + now_frame = (uf >> 3) & fmask; + live = true; + } else { + now_frame = (ehci->last_iso_frame - 1) & fmask; + live = false; + } + ehci->now_frame = now_frame; + frame = ehci->last_iso_frame; for (;;) { union ehci_shadow q, *q_p; __hc32 type, *hw_p; - unsigned incomplete = false; - - frame = now_uframe >> 3; restart: /* scan each element in frame's queue for completions */ @@ -2133,54 +2409,32 @@ restart: hw_p = &ehci->periodic [frame]; q.ptr = q_p->ptr; type = Q_NEXT_TYPE(ehci, *hw_p); - modified = 0; + modified = false; while (q.ptr != NULL) { - unsigned uf; - union ehci_shadow temp; - int live; - - live = HC_IS_RUNNING (ehci_to_hcd(ehci)->state); switch (hc32_to_cpu(ehci, type)) { - case Q_TYPE_QH: - /* 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); - break; - case Q_TYPE_FSTN: - /* for "save place" FSTNs, look at QH entries - * in the previous frame for completions. - */ - if (q.fstn->hw_prev != EHCI_LIST_END(ehci)) { - dbg ("ignoring completions from FSTNs"); - } - type = Q_NEXT_TYPE(ehci, q.fstn->hw_next); - q = q.fstn->fstn_next; - break; case Q_TYPE_ITD: /* If this ITD is still active, leave it for * later processing ... check the next entry. + * No need to check for activity unless the + * frame is current. */ - rmb (); - for (uf = 0; uf < 8 && live; uf++) { - if (0 == (q.itd->hw_transaction [uf] - & ITD_ACTIVE(ehci))) - continue; - incomplete = true; - q_p = &q.itd->itd_next; - hw_p = &q.itd->hw_next; - type = Q_NEXT_TYPE(ehci, + if (frame == now_frame && live) { + rmb(); + for (uf = 0; uf < 8; uf++) { + if (q.itd->hw_transaction[uf] & + ITD_ACTIVE(ehci)) + break; + } + if (uf < 8) { + q_p = &q.itd->itd_next; + hw_p = &q.itd->hw_next; + type = Q_NEXT_TYPE(ehci, q.itd->hw_next); - q = *q_p; - break; + q = *q_p; + break; + } } - if (uf < 8 && live) - break; /* Take finished ITDs out of the schedule * and process them: recycle, maybe report @@ -2188,7 +2442,12 @@ restart: * pointer for much longer, if at all. */ *q_p = q.itd->itd_next; - *hw_p = q.itd->hw_next; + if (!ehci->use_dummy_qh || + q.itd->hw_next != EHCI_LIST_END(ehci)) + *hw_p = q.itd->hw_next; + else + *hw_p = cpu_to_hc32(ehci, + ehci->dummy->qh_dma); type = Q_NEXT_TYPE(ehci, q.itd->hw_next); wmb(); modified = itd_complete (ehci, q.itd); @@ -2197,10 +2456,15 @@ restart: case Q_TYPE_SITD: /* If this SITD is still active, leave it for * later processing ... check the next entry. + * No need to check for activity unless the + * frame is current. */ - if ((q.sitd->hw_results & SITD_ACTIVE(ehci)) - && live) { - incomplete = true; + if (((frame == now_frame) || + (((frame + 1) & fmask) == now_frame)) + && live + && (q.sitd->hw_results & + SITD_ACTIVE(ehci))) { + q_p = &q.sitd->sitd_next; hw_p = &q.sitd->hw_next; type = Q_NEXT_TYPE(ehci, @@ -2214,63 +2478,40 @@ restart: * URB completion. */ *q_p = q.sitd->sitd_next; - *hw_p = q.sitd->hw_next; + if (!ehci->use_dummy_qh || + q.sitd->hw_next != EHCI_LIST_END(ehci)) + *hw_p = q.sitd->hw_next; + else + *hw_p = cpu_to_hc32(ehci, + ehci->dummy->qh_dma); type = Q_NEXT_TYPE(ehci, q.sitd->hw_next); wmb(); modified = sitd_complete (ehci, q.sitd); q = *q_p; break; default: - dbg ("corrupt type %d frame %d shadow %p", + ehci_dbg(ehci, "corrupt type %d frame %d shadow %p\n", type, frame, q.ptr); // BUG (); + /* FALL THROUGH */ + case Q_TYPE_QH: + case Q_TYPE_FSTN: + /* End of the iTDs and siTDs */ q.ptr = NULL; + break; } /* assume completion callbacks modify the queue */ - if (unlikely (modified)) { - if (likely(ehci->periodic_sched > 0)) - goto restart; - /* maybe we can short-circuit this scan! */ - disable_periodic(ehci); - now_uframe = clock; - break; - } + if (unlikely(modified && ehci->isoc_count > 0)) + goto restart; } - /* If we can tell we caught up to the hardware, stop now. - * We can't advance our scan without collecting the ISO - * transfers that are still pending in this frame. - */ - if (incomplete && HC_IS_RUNNING(ehci_to_hcd(ehci)->state)) { - ehci->next_uframe = now_uframe; + /* Stop when we have reached the current frame */ + if (frame == now_frame) break; - } - - // FIXME: this assumes we won't get lapped when - // latencies climb; that should be rare, but... - // detect it, and just go all the way around. - // FLR might help detect this case, so long as latencies - // don't exceed periodic_size msec (default 1.024 sec). - - // FIXME: likewise assumes HC doesn't halt mid-scan - if (now_uframe == clock) { - unsigned now; - - if (!HC_IS_RUNNING (ehci_to_hcd(ehci)->state) - || ehci->periodic_sched == 0) - break; - ehci->next_uframe = now_uframe; - now = ehci_readl(ehci, &ehci->regs->frame_index) % mod; - if (now_uframe == now) - break; - - /* rescan the rest of this frame, then ... */ - clock = now; - } else { - now_uframe++; - now_uframe %= mod; - } + /* The last frame may still have active siTDs */ + ehci->last_iso_frame = frame; + frame = (frame + 1) & fmask; } } |