On Tue, 13 Aug 2013, James Stone wrote: > >> http://www.spinics.net/lists/linux-usb/msg90457.html > > > > This patch is definitely not the way to go. A week ago I posted a > > completely different approach here: > > > > http://marc.info/?l=linux-usb&m=137537866909950&w=2 > > > > So far there has been no feedback on it; I'm still waiting to hear from > > any of the ALSA people (hint). > > > > Even that patch is not totally right. If the ALSA buffer size is > > larger than the minimum hardware requirement, then an entire buffer's > > worth of data should be queued at all times. Regardless, it's > > probably too big for a stable kernel, which means it won't appear > > before the 3.12 release is out. Maybe not even then. > > > > This patch does not seem to be needed at all. Performance is better without it. Like I said, it isn't totally right. But _something_ is definitely needed. On a slightly different note, the big patch below should speed up the bandwidth calculations in ehci-hcd quite a lot. Please try it out and see if you still get those underrun warnings when the camera or whatever is plugged in. Also, see what the irqsoff tracer has to say about the interrupt latency when these events occur. Alan Stern Index: usb-3.11/include/linux/usb/hcd.h =================================================================== --- usb-3.11.orig/include/linux/usb/hcd.h +++ usb-3.11/include/linux/usb/hcd.h @@ -479,6 +479,7 @@ struct usb_tt { struct usb_device *hub; /* upstream highspeed hub */ int multi; /* true means one TT per port */ unsigned think_time; /* think time in ns */ + void *hcpriv; /* HCD private data */ /* for control/bulk error recovery (CLEAR_TT_BUFFER) */ spinlock_t lock; @@ -537,9 +538,8 @@ extern void usb_ep0_reinit(struct usb_de * of (7/6 * 8 * bytecount) = 9.33 * bytecount */ /* bytecount = data payload byte count */ -#define NS_TO_US(ns) ((ns + 500L) / 1000L) - /* convert & round nanoseconds to microseconds */ - +#define NS_TO_US(ns) DIV_ROUND_UP(ns, 1000L) + /* convert nanoseconds to microseconds, rounding up */ /* * Full/low speed bandwidth allocation constants/support. Index: usb-3.11/drivers/usb/host/ehci.h =================================================================== --- usb-3.11.orig/drivers/usb/host/ehci.h +++ usb-3.11/drivers/usb/host/ehci.h @@ -54,6 +54,28 @@ struct ehci_stats { unsigned long unlink; }; +/* + * Scheduling and budgeting information for periodic transfers, for both + * high-speed devices and full/low-speed devices lying behind a TT. + */ +struct ehci_per_sched { + struct usb_device *udev; /* access to the TT */ + struct usb_host_endpoint *ep; + struct list_head ps_list; /* node on ehci_tt's ps_list */ + u16 tt_usecs; /* time on the FS/LS bus */ + u16 cs_mask; /* C-mask and S-mask bytes */ + u16 period; /* actual period in frames */ + u16 phase; /* actual phase, frame part */ + u8 bw_phase; /* same, for bandwidth + reservation */ + u8 phase_uf; /* uframe part of the phase */ + u8 usecs, c_usecs; /* times on the HS bus */ + u8 bw_uperiod; /* period in microframes, for + bandwidth reservation */ + u8 bw_period; /* same, in frames */ +}; +#define NO_FRAME 29999 /* frame not assigned yet */ + /* ehci_hcd->lock guards shared data against other CPUs: * ehci_hcd: async, unlink, periodic (and shadow), ... * usb_host_endpoint: hcpriv @@ -226,6 +248,15 @@ struct ehci_hcd { /* one per controlle struct dentry *debug_dir; #endif + /* bandwidth usage */ +#define EHCI_BANDWIDTH_SIZE 64 +#define EHCI_BANDWIDTH_FRAMES (EHCI_BANDWIDTH_SIZE >> 3) + u8 bandwidth[EHCI_BANDWIDTH_SIZE]; + /* us allocated per uframe */ + u8 tt_budget[EHCI_BANDWIDTH_SIZE]; + /* us budgeted per uframe */ + struct list_head tt_list; + /* platform-specific data -- must come last */ unsigned long priv[0] __aligned(sizeof(s64)); }; @@ -381,6 +412,7 @@ struct ehci_qh { struct list_head intr_node; /* list of intr QHs */ struct ehci_qtd *dummy; struct list_head unlink_node; + struct ehci_per_sched ps; /* scheduling info */ unsigned unlink_cycle; @@ -394,16 +426,8 @@ struct ehci_qh { u8 xacterrs; /* XactErr retry counter */ #define QH_XACTERR_MAX 32 /* XactErr retry limit */ - /* periodic schedule info */ - u8 usecs; /* intr bandwidth */ u8 gap_uf; /* uframes split/csplit gap */ - u8 c_usecs; /* ... split completion bw */ - u16 tt_usecs; /* tt downstream bandwidth */ - unsigned short period; /* polling interval */ - unsigned short start; /* where polling starts */ -#define NO_FRAME ((unsigned short)~0) /* pick new start */ - struct usb_device *dev; /* access to TT */ unsigned is_out:1; /* bulk or intr OUT */ unsigned clearing_tt:1; /* Clear-TT-Buf in progress */ unsigned dequeue_during_giveback:1; @@ -445,22 +469,17 @@ struct ehci_iso_stream { u8 highspeed; struct list_head td_list; /* queued itds/sitds */ struct list_head free_list; /* list of unused itds/sitds */ - struct usb_device *udev; - struct usb_host_endpoint *ep; /* output of (re)scheduling */ - int next_uframe; + struct ehci_per_sched ps; /* scheduling info */ + unsigned next_uframe; __hc32 splits; /* the rest is derived from the endpoint descriptor, - * trusting urb->interval == f(epdesc->bInterval) and * including the extra info for hw_bufp[0..2] */ - u8 usecs, c_usecs; - u16 interval; - u16 tt_usecs; + u16 uperiod; /* period in uframes */ u16 maxp; - u16 raw_mask; unsigned bandwidth; /* This is used to initialize iTD's hw_bufp fields */ @@ -575,6 +594,35 @@ struct ehci_fstn { /*-------------------------------------------------------------------------*/ +/* + * USB-2.0 Specification Sections 11.14 and 11.18 + * Scheduling and budgeting split transactions using TTs + * + * A hub can have a single TT for all its ports, or multiple TTs (one for each + * port). The bandwidth and budgeting information for the full/low-speed bus + * below each TT is self-contained and independent of the other TT's or the + * high-speed bus. + * + * "Bandwidth" refers to the number of microseconds on the FS/LS bus allocated + * to an interrupt or isochronous endpoint for each frame. "Budget" refers to + * the best-case estimate of the number of full-speed bytes allocated to an + * endpoint for each microframe within an allocated frame. + * + * Removal of an endpoint invalidates a TT's budget. Instead of trying to + * keep an up-to-date record, we recompute the budget when it is needed. + */ + +struct ehci_tt { + u16 bandwidth[EHCI_BANDWIDTH_FRAMES]; + + struct list_head tt_list; /* List of all ehci_tt's */ + struct list_head ps_list; /* Items using this TT */ + struct usb_tt *usb_tt; + int tt_port; /* TT port number */ +}; + +/*-------------------------------------------------------------------------*/ + /* Prepare the PORTSC wakeup flags during controller suspend/resume */ #define ehci_prepare_ports_for_controller_suspend(ehci, do_wakeup) \ Index: usb-3.11/drivers/usb/host/ehci-sched.c =================================================================== --- usb-3.11.orig/drivers/usb/host/ehci-sched.c +++ usb-3.11/drivers/usb/host/ehci-sched.c @@ -106,80 +106,210 @@ static void periodic_unlink (struct ehci *hw_p = 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) -{ - __hc32 *hw_p = &ehci->periodic [frame]; - union ehci_shadow *q = &ehci->pshadow [frame]; - unsigned usecs = 0; - struct ehci_qh_hw *hw; - - while (q->ptr) { - switch (hc32_to_cpu(ehci, Q_NEXT_TYPE(ehci, *hw_p))) { - case Q_TYPE_QH: - hw = q->qh->hw; - /* is it in the S-mask? */ - if (hw->hw_info2 & cpu_to_hc32(ehci, 1 << uframe)) - usecs += q->qh->usecs; - /* ... or C-mask? */ - if (hw->hw_info2 & cpu_to_hc32(ehci, - 1 << (8 + uframe))) - usecs += q->qh->c_usecs; - hw_p = &hw->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); - } +/*-------------------------------------------------------------------------*/ + +/* Bandwidth and TT management */ + +/* Find the TT data structure for this device; create it if necessary */ +static int find_tt(struct usb_device *udev, struct ehci_tt **ptt) +{ + struct usb_tt *utt = udev->tt; + struct ehci_tt *tt, **tt_index, **ptt2; + unsigned port; + bool allocated_index = false; + + if (!utt) { + *ptt = NULL; /* Not below a TT */ + return 0; + } + + /* + * 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 -ENOMEM; + utt->hcpriv = tt_index; + allocated_index = true; + } + port = udev->ttport - 1; + ptt2 = &tt_index[port]; + } else { + port = 0; + ptt2 = (struct ehci_tt **) &utt->hcpriv; + } - /* ... 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 = *ptt2; + if (!tt) { /* Create the ehci_tt */ + struct ehci_hcd *ehci = + hcd_to_ehci(bus_to_hcd(udev->bus)); + + tt = kzalloc(sizeof(*tt), GFP_ATOMIC); + if (!tt) { + if (allocated_index) { + utt->hcpriv = NULL; + kfree(tt_index); } + return -ENOMEM; + } + list_add_tail(&tt->tt_list, &ehci->tt_list); + INIT_LIST_HEAD(&tt->ps_list); + tt->usb_tt = utt; + tt->tt_port = port; + *ptt2 = tt; + } - hw_p = &q->sitd->hw_next; - q = &q->sitd->sitd_next; - break; + *ptt = tt; + return 0; +} + +/* 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 > ehci->uframe_periodic_max) - ehci_err (ehci, "uframe %d sched overrun: %d usecs\n", - frame * 8 + uframe, usecs); -#endif - return usecs; + + /* FS/LS bus bandwidth */ + if (tt_usecs) { + find_tt(qh->ps.udev, &tt); + 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; @@ -227,68 +357,6 @@ static inline void carryover_tt_bandwidt } } -/* 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->hw_info2); - tt_usecs[uf] += q->qh->tt_usecs; - } - hw_p = &q->qh->hw->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 @@ -312,20 +380,29 @@ 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]; + + if (tt->bandwidth[frame] + usecs > 900) + return 0; - periodic_tt_usecs (ehci, dev, frame, tt_usecs); + uf = frame << 3; + for (i = 0; i < 8; (++i, ++uf)) + tt_usecs[i] = ehci->tt_budget[uf]; ehci_vdbg(ehci, "tt frame %d check %d usecs start uframe %d in" " schedule %d/%d/%d/%d/%d/%d/%d/%d\n", @@ -346,7 +423,7 @@ static int tt_available ( */ if (125 < usecs) { int ufs = (usecs / 125); - int i; + for (i = uframe; i < (uframe + ufs) && i < 8; i++) if (0 < tt_usecs[i]) { ehci_vdbg(ehci, @@ -409,7 +486,7 @@ static int tt_no_collision ( continue; case Q_TYPE_QH: hw = here.qh->hw; - if (same_tt (dev, here.qh->dev)) { + if (same_tt(dev, here.qh->ps.udev)) { u32 mask; mask = hc32_to_cpu(ehci, @@ -489,19 +566,19 @@ static void disable_periodic(struct ehci 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->hw_info2) & (QH_CMASK | QH_SMASK), - qh, qh->start, qh->usecs, qh->c_usecs); + 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; @@ -521,7 +598,7 @@ static void qh_link_periodic(struct ehci * 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->hw_next; @@ -541,10 +618,10 @@ static void qh_link_periodic(struct ehci 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); list_add(&qh->intr_node, &ehci->intr_qh_list); @@ -574,22 +651,21 @@ static void qh_unlink_periodic(struct eh */ /* 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, + qh->ps.period, hc32_to_cpup(ehci, &qh->hw->hw_info2) & (QH_CMASK | QH_SMASK), - qh, qh->start, qh->usecs, qh->c_usecs); + qh, qh->ps.phase, qh->ps.usecs, qh->ps.c_usecs); /* qh->qh_next still "live" to HC */ qh->qh_state = QH_STATE_UNLINK; @@ -674,11 +750,9 @@ 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... */ @@ -688,25 +762,10 @@ static int check_period ( /* 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! @@ -717,40 +776,40 @@ 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, + __hc32 *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. @@ -761,15 +820,15 @@ 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; } @@ -785,60 +844,63 @@ static int qh_schedule(struct ehci_hcd * { int status; 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; 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, &hw->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; + status = find_tt(qh->ps.udev, &tt); + if (status) + 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) { - int i; - - for (i = qh->period; status && i > 0; --i) { - frame = ++ehci->random_frame % qh->period; - for (uframe = 0; uframe < 8; uframe++) { - status = check_intr_schedule (ehci, - frame, uframe, qh, - &c_mask); - if (status == 0) - break; - } + /* "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; } - - /* qh->period == 0 means every uframe */ - } else { - frame = 0; - status = check_intr_schedule (ehci, 0, 0, qh, &c_mask); } - if (status) - goto done; - qh->start = frame; - /* reset S-frame and (maybe) C-frame masks */ - hw->hw_info2 &= cpu_to_hc32(ehci, ~(QH_CMASK | QH_SMASK)); - hw->hw_info2 |= qh->period - ? cpu_to_hc32(ehci, 1 << uframe) - : cpu_to_hc32(ehci, QH_SMASK); - hw->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_uperiod - 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); done: return status; @@ -944,7 +1006,7 @@ 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->next_uframe = NO_FRAME; } return stream; } @@ -953,25 +1015,24 @@ 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 { @@ -995,9 +1056,19 @@ 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 /= interval; + 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; @@ -1011,36 +1082,46 @@ 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 /= interval << 3; + 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->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 raw_mask later */ + /* 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; } @@ -1065,9 +1146,7 @@ iso_stream_find (struct ehci_hcd *ehci, stream = iso_stream_alloc(GFP_ATOMIC); if (likely (stream != NULL)) { 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, hw is set */ @@ -1112,7 +1191,7 @@ 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. @@ -1211,7 +1290,7 @@ itd_urb_transaction ( memset (itd, 0, sizeof *itd); itd->itd_dma = itd_dma; - itd->frame = 9999; /* an invalid value */ + itd->frame = NO_FRAME; list_add (&itd->itd_list, &sched->td_list); } spin_unlock_irqrestore (&ehci->lock, flags); @@ -1224,49 +1303,106 @@ 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->next_uframe == 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; + } + } + + find_tt(stream->ps.udev, &tt); + 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 uframe_periodic_max usec */ - if (periodic_usecs (ehci, uframe >> 3, uframe & 0x7) - > (ehci->uframe_periodic_max - usecs)) - return 0; + unsigned usecs; - /* we know urb->interval is 2^N uframes */ - uframe += period; - } while (uframe < mod); + /* convert "usecs we need" to "max already claimed" */ + usecs = ehci->uframe_periodic_max - stream->ps.usecs; + + 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->raw_mask << (uframe & 7); + mask = stream->ps.cs_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; /* check bandwidth */ - uframe %= period_uframes; + uframe &= stream->ps.bw_uperiod - 1; frame = uframe >> 3; #ifdef CONFIG_USB_EHCI_TT_NEWSCHED @@ -1274,54 +1410,48 @@ sitd_slot_ok ( * tt_available scheduling guarantees 10+% for control/bulk. */ uf = uframe & 7; - if (!tt_available(ehci, period_uframes >> 3, - stream->udev, frame, uf, stream->tt_usecs)) + 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)) + if (!tt_no_collision(ehci, stream->ps.bw_period, + stream->ps.udev, frame, mask)) return 0; #endif - /* this multi-pass logic is simple, but performance may - * suffer when the schedule data isn't cached. - */ do { - u32 max_used; - - frame = uframe >> 3; - uf = uframe & 7; + unsigned max_used; + unsigned i; /* check starts (OUT uses more than one) */ - max_used = ehci->uframe_periodic_max - 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 = ehci->uframe_periodic_max - 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; } @@ -1350,21 +1480,20 @@ iso_stream_schedule ( unsigned mod = ehci->periodic_size << 3; struct ehci_iso_sched *sched = urb->hcpriv; - period = urb->interval; + period = stream->uperiod; span = sched->span; - if (!stream->highspeed) { - period <<= 3; + if (!stream->highspeed) span <<= 3; - } now = ehci_read_frame_index(ehci) & (mod - 1); - /* Typical case: reuse current schedule, stream is still active. + /* + * 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. */ - if (likely (!list_empty (&stream->td_list))) { + if (likely(stream->next_uframe != NO_FRAME)) { /* Take the isochronous scheduling threshold into account */ if (ehci->i_thresh) @@ -1372,6 +1501,11 @@ iso_stream_schedule ( else next = (now + 2 + 7) & ~0x07; /* full frame cache */ + /* New stream is already scheduled; use the upcoming slot */ + if (unlikely(list_empty(&stream->td_list))) + stream->next_uframe = next + + ((stream->next_uframe - next) & + (period - 1)); /* * Use ehci->last_iso_frame as the base. There can't be any * TDs scheduled for earlier than that. @@ -1409,6 +1543,7 @@ iso_stream_schedule ( } start += base; + stream->next_uframe = start & (mod - 1); } /* need to schedule; when's the next (u)frame we could start? @@ -1418,7 +1553,13 @@ iso_stream_schedule ( * jump until after the queue is primed. */ else { - int done = 0; + int done = 0; + struct ehci_tt *tt; + + status = find_tt(stream->ps.udev, &tt); + if (status) + goto fail; + compute_tt_budget(ehci->tt_budget, tt); base = now & ~0x07; start = base + SCHEDULING_DELAY; @@ -1434,14 +1575,13 @@ iso_stream_schedule ( start--; /* check schedule: enough space? */ if (stream->highspeed) { - if (itd_slot_ok(ehci, mod, start, - stream->usecs, period)) + if (itd_slot_ok(ehci, stream, start)) done = 1; } else { if ((start % 8) >= 6) continue; - if (sitd_slot_ok(ehci, mod, stream, - start, sched, period)) + if (sitd_slot_ok(ehci, stream, start, + sched, tt)) done = 1; } } while (start > next && !done); @@ -1452,6 +1592,13 @@ iso_stream_schedule ( status = -ENOSPC; goto fail; } + + stream->next_uframe = start & (mod - 1); + 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); } /* Tried to schedule too far into the future? */ @@ -1462,8 +1609,6 @@ iso_stream_schedule ( goto fail; } - stream->next_uframe = start & (mod - 1); - /* report high speed start in uframes; full speed, in frames */ urb->start_frame = stream->next_uframe; if (!stream->highspeed) @@ -1612,7 +1757,7 @@ static void itd_link_urb( itd_patch(ehci, itd, iso_sched, packet, uframe); - next_uframe += stream->interval; + next_uframe += stream->uperiod; next_uframe &= mod - 1; packet++; @@ -1757,9 +1902,9 @@ static int itd_submit (struct ehci_hcd * 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; } @@ -1820,7 +1965,7 @@ 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. @@ -1912,7 +2057,7 @@ sitd_urb_transaction ( memset (sitd, 0, sizeof *sitd); sitd->sitd_dma = sitd_dma; - sitd->frame = 9999; /* an invalid value */ + sitd->frame = NO_FRAME; list_add (&sitd->sitd_list, &iso_sched->td_list); } @@ -1991,7 +2136,7 @@ static void sitd_link_urb( urb->dev->devpath, stream->bEndpointAddress & 0x0f, (stream->bEndpointAddress & USB_DIR_IN) ? "in" : "out", (next_uframe >> 3) & (ehci->periodic_size - 1), - stream->interval, hc32_to_cpu(ehci, stream->splits)); + stream->ps.period, stream->ps.cs_mask); } if (ehci_to_hcd(ehci)->self.bandwidth_isoc_reqs == 0) { @@ -2021,7 +2166,7 @@ static void sitd_link_urb( sitd_link(ehci, (next_uframe >> 3) & (ehci->periodic_size - 1), sitd); - next_uframe += stream->interval << 3; + next_uframe += stream->uperiod; } stream->next_uframe = next_uframe & (mod - 1); @@ -2145,9 +2290,9 @@ static int sitd_submit (struct ehci_hcd 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; } Index: usb-3.11/drivers/usb/host/ehci-dbg.c =================================================================== --- usb-3.11.orig/drivers/usb/host/ehci-dbg.c +++ usb-3.11/drivers/usb/host/ehci-dbg.c @@ -334,9 +334,9 @@ static inline void remove_debug_files (s /* troubleshooting help: expose state in debugfs */ static int debug_async_open(struct inode *, struct file *); +static int debug_bandwidth_open(struct inode *, struct file *); static int debug_periodic_open(struct inode *, struct file *); static int debug_registers_open(struct inode *, struct file *); -static int debug_async_open(struct inode *, struct file *); static ssize_t debug_output(struct file*, char __user*, size_t, loff_t*); static int debug_close(struct inode *, struct file *); @@ -348,6 +348,13 @@ static const struct file_operations debu .release = debug_close, .llseek = default_llseek, }; +static const struct file_operations debug_bandwidth_fops = { + .owner = THIS_MODULE, + .open = debug_bandwidth_open, + .read = debug_output, + .release = debug_close, + .llseek = default_llseek, +}; static const struct file_operations debug_periodic_fops = { .owner = THIS_MODULE, .open = debug_periodic_open, @@ -526,6 +533,89 @@ static ssize_t fill_async_buffer(struct return strlen(buf->output_buf); } +static ssize_t fill_bandwidth_buffer(struct debug_buffer *buf) +{ + struct ehci_hcd *ehci; + struct ehci_tt *tt; + struct ehci_per_sched *ps; + unsigned temp, size; + char *next; + unsigned i; + u8 *bw; + u16 *bf; + u8 budget[EHCI_BANDWIDTH_SIZE]; + + ehci = hcd_to_ehci(bus_to_hcd(buf->bus)); + next = buf->output_buf; + size = buf->alloc_size; + + *next = 0; + + spin_lock_irq(&ehci->lock); + + /* Dump the HS bandwidth table */ + temp = scnprintf(next, size, + "HS bandwidth allocation (us per microframe)\n"); + size -= temp; + next += temp; + for (i = 0; i < EHCI_BANDWIDTH_SIZE; i += 8) { + bw = &ehci->bandwidth[i]; + temp = scnprintf(next, size, + "%2u: %4u%4u%4u%4u%4u%4u%4u%4u\n", + i, bw[0], bw[1], bw[2], bw[3], + bw[4], bw[5], bw[6], bw[7]); + size -= temp; + next += temp; + } + + /* Dump all the FS/LS tables */ + list_for_each_entry(tt, &ehci->tt_list, tt_list) { + temp = scnprintf(next, size, + "\nTT %s port %d FS/LS bandwidth allocation (us per frame)\n", + dev_name(&tt->usb_tt->hub->dev), + tt->tt_port + !!tt->usb_tt->multi); + size -= temp; + next += temp; + + bf = tt->bandwidth; + temp = scnprintf(next, size, + " %5u%5u%5u%5u%5u%5u%5u%5u\n", + bf[0], bf[1], bf[2], bf[3], + bf[4], bf[5], bf[6], bf[7]); + size -= temp; + next += temp; + + temp = scnprintf(next, size, + "FS/LS budget (us per microframe)\n"); + size -= temp; + next += temp; + compute_tt_budget(budget, tt); + for (i = 0; i < EHCI_BANDWIDTH_SIZE; i += 8) { + bw = &budget[i]; + temp = scnprintf(next, size, + "%2u: %4u%4u%4u%4u%4u%4u%4u%4u\n", + i, bw[0], bw[1], bw[2], bw[3], + bw[4], bw[5], bw[6], bw[7]); + size -= temp; + next += temp; + } + list_for_each_entry(ps, &tt->ps_list, ps_list) { + temp = scnprintf(next, size, + "%s ep %02x: %4u @ %2u.%u+%u mask %04x\n", + dev_name(&ps->udev->dev), + ps->ep->desc.bEndpointAddress, + ps->tt_usecs, + ps->bw_phase, ps->phase_uf, + ps->bw_period, ps->cs_mask); + size -= temp; + next += temp; + } + } + spin_unlock_irq(&ehci->lock); + + return next - buf->output_buf; +} + #define DBG_SCHED_LIMIT 64 static ssize_t fill_periodic_buffer(struct debug_buffer *buf) { @@ -572,7 +662,7 @@ static ssize_t fill_periodic_buffer(stru case Q_TYPE_QH: hw = p.qh->hw; temp = scnprintf (next, size, " qh%d-%04x/%p", - p.qh->period, + p.qh->ps.period, hc32_to_cpup(ehci, &hw->hw_info2) /* uframe masks */ @@ -619,7 +709,8 @@ static ssize_t fill_periodic_buffer(stru speed_char (scratch), scratch & 0x007f, (scratch >> 8) & 0x000f, type, - p.qh->usecs, p.qh->c_usecs, + p.qh->ps.usecs, + p.qh->ps.c_usecs, temp, 0x7ff & (scratch >> 16)); @@ -646,7 +737,7 @@ static ssize_t fill_periodic_buffer(stru case Q_TYPE_SITD: temp = scnprintf (next, size, " sitd%d-%04x/%p", - p.sitd->stream->interval, + p.sitd->stream->ps.period, hc32_to_cpup(ehci, &p.sitd->hw_uframe) & 0x0000ffff, p.sitd); @@ -919,6 +1010,7 @@ static int debug_close(struct inode *ino return 0; } + static int debug_async_open(struct inode *inode, struct file *file) { file->private_data = alloc_buffer(inode->i_private, fill_async_buffer); @@ -926,6 +1018,14 @@ static int debug_async_open(struct inode return file->private_data ? 0 : -ENOMEM; } +static int debug_bandwidth_open(struct inode *inode, struct file *file) +{ + file->private_data = alloc_buffer(inode->i_private, + fill_bandwidth_buffer); + + return file->private_data ? 0 : -ENOMEM; +} + static int debug_periodic_open(struct inode *inode, struct file *file) { struct debug_buffer *buf; @@ -958,6 +1058,10 @@ static inline void create_debug_files (s &debug_async_fops)) goto file_error; + if (!debugfs_create_file("bandwidth", S_IRUGO, ehci->debug_dir, bus, + &debug_bandwidth_fops)) + goto file_error; + if (!debugfs_create_file("periodic", S_IRUGO, ehci->debug_dir, bus, &debug_periodic_fops)) goto file_error; Index: usb-3.11/drivers/usb/host/ehci-q.c =================================================================== --- usb-3.11.orig/drivers/usb/host/ehci-q.c +++ usb-3.11/drivers/usb/host/ehci-q.c @@ -105,9 +105,9 @@ qh_update (struct ehci_hcd *ehci, struct is_out = qh->is_out; epnum = (hc32_to_cpup(ehci, &hw->hw_info1) >> 8) & 0x0f; - if (unlikely (!usb_gettoggle (qh->dev, epnum, is_out))) { + if (unlikely(!usb_gettoggle(qh->ps.udev, epnum, is_out))) { hw->hw_token &= ~cpu_to_hc32(ehci, QTD_TOGGLE); - usb_settoggle (qh->dev, epnum, is_out, 1); + usb_settoggle(qh->ps.udev, epnum, is_out, 1); } } @@ -809,26 +809,35 @@ qh_make ( * For control/bulk requests, the HC or TT handles these. */ if (type == PIPE_INTERRUPT) { - qh->usecs = NS_TO_US(usb_calc_bus_time(USB_SPEED_HIGH, + unsigned tmp; + + qh->ps.usecs = NS_TO_US(usb_calc_bus_time(USB_SPEED_HIGH, is_input, 0, hb_mult(maxp) * max_packet(maxp))); - qh->start = NO_FRAME; + qh->ps.phase = NO_FRAME; if (urb->dev->speed == USB_SPEED_HIGH) { - qh->c_usecs = 0; + qh->ps.c_usecs = 0; qh->gap_uf = 0; - qh->period = urb->interval >> 3; - if (qh->period == 0 && urb->interval != 1) { + if (urb->interval > 1 && urb->interval < 8) { /* NOTE interval 2 or 4 uframes could work. * But interval 1 scheduling is simpler, and * includes high bandwidth. */ urb->interval = 1; - } else if (qh->period > ehci->periodic_size) { - qh->period = ehci->periodic_size; - urb->interval = qh->period << 3; + } else if (urb->interval > ehci->periodic_size << 3) { + urb->interval = ehci->periodic_size << 3; } + qh->ps.period = urb->interval >> 3; + + /* period for bandwidth allocation */ + tmp = min_t(unsigned, EHCI_BANDWIDTH_SIZE, + 1 << (urb->ep->desc.bInterval - 1)); + + /* Allow urb->interval to override */ + qh->ps.bw_uperiod = min_t(unsigned, tmp, urb->interval); + qh->ps.bw_period = qh->ps.bw_uperiod >> 3; } else { int think_time; @@ -838,27 +847,35 @@ qh_make ( /* FIXME this just approximates SPLIT/CSPLIT times */ if (is_input) { // SPLIT, gap, CSPLIT+DATA - qh->c_usecs = qh->usecs + HS_USECS (0); - qh->usecs = HS_USECS (1); + qh->ps.c_usecs = qh->ps.usecs + HS_USECS(0); + qh->ps.usecs = HS_USECS(1); } else { // SPLIT+DATA, gap, CSPLIT - qh->usecs += HS_USECS (1); - qh->c_usecs = HS_USECS (0); + qh->ps.usecs += HS_USECS(1); + qh->ps.c_usecs = HS_USECS(0); } think_time = tt ? tt->think_time : 0; - qh->tt_usecs = NS_TO_US (think_time + + qh->ps.tt_usecs = NS_TO_US(think_time + usb_calc_bus_time (urb->dev->speed, is_input, 0, max_packet (maxp))); - qh->period = urb->interval; - if (qh->period > ehci->periodic_size) { - qh->period = ehci->periodic_size; - urb->interval = qh->period; - } + if (urb->interval > ehci->periodic_size) + urb->interval = ehci->periodic_size; + qh->ps.period = urb->interval; + + /* period for bandwidth allocation */ + tmp = min_t(unsigned, EHCI_BANDWIDTH_FRAMES, + urb->ep->desc.bInterval); + tmp = rounddown_pow_of_two(tmp); + + /* Allow urb->interval to override */ + qh->ps.bw_period = min_t(unsigned, tmp, urb->interval); + qh->ps.bw_uperiod = qh->ps.bw_period << 3; } } /* support for tt scheduling, and access to toggles */ - qh->dev = urb->dev; + qh->ps.udev = urb->dev; + qh->ps.ep = urb->ep; /* using TT? */ switch (urb->dev->speed) { Index: usb-3.11/drivers/usb/host/ehci-sysfs.c =================================================================== --- usb-3.11.orig/drivers/usb/host/ehci-sysfs.c +++ usb-3.11/drivers/usb/host/ehci-sysfs.c @@ -97,8 +97,7 @@ static ssize_t store_uframe_periodic_max { struct ehci_hcd *ehci; unsigned uframe_periodic_max; - unsigned frame, uframe; - unsigned short allocated_max; + unsigned uframe; unsigned long flags; ssize_t ret; @@ -122,16 +121,14 @@ static ssize_t store_uframe_periodic_max /* * for request to decrease max periodic bandwidth, we have to check - * every microframe in the schedule to see whether the decrease is - * possible. + * to see whether the decrease is possible. */ if (uframe_periodic_max < ehci->uframe_periodic_max) { - allocated_max = 0; + u8 allocated_max = 0; - for (frame = 0; frame < ehci->periodic_size; ++frame) - for (uframe = 0; uframe < 7; ++uframe) - allocated_max = max(allocated_max, - periodic_usecs (ehci, frame, uframe)); + for (uframe = 0; uframe < EHCI_BANDWIDTH_SIZE; ++uframe) + allocated_max = max(allocated_max, + ehci->bandwidth[uframe]); if (allocated_max > uframe_periodic_max) { ehci_info(ehci, Index: usb-3.11/drivers/usb/host/ehci-hcd.c =================================================================== --- usb-3.11.orig/drivers/usb/host/ehci-hcd.c +++ usb-3.11/drivers/usb/host/ehci-hcd.c @@ -110,6 +110,9 @@ MODULE_PARM_DESC (ignore_oc, "ignore bog #include "ehci.h" #include "pci-quirks.h" +static void compute_tt_budget(u8 budget_table[EHCI_BANDWIDTH_SIZE], + struct ehci_tt *tt); + /* * The MosChip MCS9990 controller updates its microframe counter * a little before the frame counter, and occasionally we will read @@ -491,6 +494,7 @@ static int ehci_init(struct usb_hcd *hcd INIT_LIST_HEAD(&ehci->intr_qh_list); INIT_LIST_HEAD(&ehci->cached_itd_list); INIT_LIST_HEAD(&ehci->cached_sitd_list); + INIT_LIST_HEAD(&ehci->tt_list); if (HCC_PGM_FRAMELISTLEN(hcc_params)) { /* periodic schedule size can be smaller than default */ @@ -963,6 +967,7 @@ rescan: goto idle_timeout; /* BUG_ON(!list_empty(&stream->free_list)); */ + reserve_release_iso_bandwidth(ehci, stream, -1); kfree(stream); goto done; } @@ -993,6 +998,8 @@ idle_timeout: if (qh->clearing_tt) goto idle_timeout; if (list_empty (&qh->qtd_list)) { + if (qh->ps.bw_uperiod) + reserve_release_intr_bandwidth(ehci, qh, -1); qh_destroy(ehci, qh); break; } @@ -1033,7 +1040,7 @@ ehci_endpoint_reset(struct usb_hcd *hcd, * the toggle bit in the QH. */ if (qh) { - usb_settoggle(qh->dev, epnum, is_out, 0); + usb_settoggle(qh->ps.udev, epnum, is_out, 0); if (!list_empty(&qh->qtd_list)) { WARN_ONCE(1, "clear_halt for a busy endpoint\n"); } else { @@ -1059,6 +1066,19 @@ static int ehci_get_frame (struct usb_hc /*-------------------------------------------------------------------------*/ +/* Device addition and removal */ + +static void ehci_remove_device(struct usb_hcd *hcd, struct usb_device *udev) +{ + struct ehci_hcd *ehci = hcd_to_ehci(hcd); + + spin_lock_irq(&ehci->lock); + drop_tt(udev); + spin_unlock_irq(&ehci->lock); +} + +/*-------------------------------------------------------------------------*/ + #ifdef CONFIG_PM /* suspend/resume, section 4.3 */ @@ -1202,6 +1222,11 @@ static const struct hc_driver ehci_hc_dr .bus_resume = ehci_bus_resume, .relinquish_port = ehci_relinquish_port, .port_handed_over = ehci_port_handed_over, + + /* + * device support + */ + .free_dev = ehci_remove_device, }; void ehci_init_driver(struct hc_driver *drv, -- 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
