diff options
| -rw-r--r-- | drivers/usb/dwc2/core.h | 85 | ||||
| -rw-r--r-- | drivers/usb/dwc2/hcd.c | 87 | ||||
| -rw-r--r-- | drivers/usb/dwc2/hcd.h | 79 | ||||
| -rw-r--r-- | drivers/usb/dwc2/hcd_queue.c | 1170 |
4 files changed, 1250 insertions, 171 deletions
diff --git a/drivers/usb/dwc2/core.h b/drivers/usb/dwc2/core.h index 52cbea28d0e9..115925909390 100644 --- a/drivers/usb/dwc2/core.h +++ b/drivers/usb/dwc2/core.h @@ -592,6 +592,84 @@ struct dwc2_hregs_backup { bool valid; }; +/* + * Constants related to high speed periodic scheduling + * + * We have a periodic schedule that is DWC2_HS_SCHEDULE_UFRAMES long. From a + * reservation point of view it's assumed that the schedule goes right back to + * the beginning after the end of the schedule. + * + * What does that mean for scheduling things with a long interval? It means + * we'll reserve time for them in every possible microframe that they could + * ever be scheduled in. ...but we'll still only actually schedule them as + * often as they were requested. + * + * We keep our schedule in a "bitmap" structure. This simplifies having + * to keep track of and merge intervals: we just let the bitmap code do most + * of the heavy lifting. In a way scheduling is much like memory allocation. + * + * We schedule 100us per uframe or 80% of 125us (the maximum amount you're + * supposed to schedule for periodic transfers). That's according to spec. + * + * Note that though we only schedule 80% of each microframe, the bitmap that we + * keep the schedule in is tightly packed (AKA it doesn't have 100us worth of + * space for each uFrame). + * + * Requirements: + * - DWC2_HS_SCHEDULE_UFRAMES must even divide 0x4000 (HFNUM_MAX_FRNUM + 1) + * - DWC2_HS_SCHEDULE_UFRAMES must be 8 times DWC2_LS_SCHEDULE_FRAMES (probably + * could be any multiple of 8 times DWC2_LS_SCHEDULE_FRAMES, but there might + * be bugs). The 8 comes from the USB spec: number of microframes per frame. + */ +#define DWC2_US_PER_UFRAME 125 +#define DWC2_HS_PERIODIC_US_PER_UFRAME 100 + +#define DWC2_HS_SCHEDULE_UFRAMES 8 +#define DWC2_HS_SCHEDULE_US (DWC2_HS_SCHEDULE_UFRAMES * \ + DWC2_HS_PERIODIC_US_PER_UFRAME) + +/* + * Constants related to low speed scheduling + * + * For high speed we schedule every 1us. For low speed that's a bit overkill, + * so we make up a unit called a "slice" that's worth 25us. There are 40 + * slices in a full frame and we can schedule 36 of those (90%) for periodic + * transfers. + * + * Our low speed schedule can be as short as 1 frame or could be longer. When + * we only schedule 1 frame it means that we'll need to reserve a time every + * frame even for things that only transfer very rarely, so something that runs + * every 2048 frames will get time reserved in every frame. Our low speed + * schedule can be longer and we'll be able to handle more overlap, but that + * will come at increased memory cost and increased time to schedule. + * + * Note: one other advantage of a short low speed schedule is that if we mess + * up and miss scheduling we can jump in and use any of the slots that we + * happened to reserve. + * + * With 25 us per slice and 1 frame in the schedule, we only need 4 bytes for + * the schedule. There will be one schedule per TT. + * + * Requirements: + * - DWC2_US_PER_SLICE must evenly divide DWC2_LS_PERIODIC_US_PER_FRAME. + */ +#define DWC2_US_PER_SLICE 25 +#define DWC2_SLICES_PER_UFRAME (DWC2_US_PER_UFRAME / DWC2_US_PER_SLICE) + +#define DWC2_ROUND_US_TO_SLICE(us) \ + (DIV_ROUND_UP((us), DWC2_US_PER_SLICE) * \ + DWC2_US_PER_SLICE) + +#define DWC2_LS_PERIODIC_US_PER_FRAME \ + 900 +#define DWC2_LS_PERIODIC_SLICES_PER_FRAME \ + (DWC2_LS_PERIODIC_US_PER_FRAME / \ + DWC2_US_PER_SLICE) + +#define DWC2_LS_SCHEDULE_FRAMES 1 +#define DWC2_LS_SCHEDULE_SLICES (DWC2_LS_SCHEDULE_FRAMES * \ + DWC2_LS_PERIODIC_SLICES_PER_FRAME) + /** * struct dwc2_hsotg - Holds the state of the driver, including the non-periodic * and periodic schedules @@ -682,7 +760,9 @@ struct dwc2_hregs_backup { * This value is in microseconds per (micro)frame. The * assumption is that all periodic transfers may occur in * the same (micro)frame. - * @frame_usecs: Internal variable used by the microframe scheduler + * @hs_periodic_bitmap: Bitmap used by the microframe scheduler any time the + * host is in high speed mode; low speed schedules are + * stored elsewhere since we need one per TT. * @frame_number: Frame number read from the core at SOF. The value ranges * from 0 to HFNUM_MAX_FRNUM. * @periodic_qh_count: Count of periodic QHs, if using several eps. Used for @@ -803,7 +883,8 @@ struct dwc2_hsotg { struct list_head periodic_sched_queued; struct list_head split_order; u16 periodic_usecs; - u16 frame_usecs[8]; + unsigned long hs_periodic_bitmap[ + DIV_ROUND_UP(DWC2_HS_SCHEDULE_US, BITS_PER_LONG)]; u16 frame_number; u16 periodic_qh_count; bool bus_suspended; diff --git a/drivers/usb/dwc2/hcd.c b/drivers/usb/dwc2/hcd.c index 8edd0b45f41c..2b5a706e7c32 100644 --- a/drivers/usb/dwc2/hcd.c +++ b/drivers/usb/dwc2/hcd.c @@ -2252,6 +2252,90 @@ void dwc2_host_hub_info(struct dwc2_hsotg *hsotg, void *context, int *hub_addr, *hub_port = urb->dev->ttport; } +/** + * dwc2_host_get_tt_info() - Get the dwc2_tt associated with context + * + * This will get the dwc2_tt structure (and ttport) associated with the given + * context (which is really just a struct urb pointer). + * + * The first time this is called for a given TT we allocate memory for our + * structure. When everyone is done and has called dwc2_host_put_tt_info() + * then the refcount for the structure will go to 0 and we'll free it. + * + * @hsotg: The HCD state structure for the DWC OTG controller. + * @qh: The QH structure. + * @context: The priv pointer from a struct dwc2_hcd_urb. + * @mem_flags: Flags for allocating memory. + * @ttport: We'll return this device's port number here. That's used to + * reference into the bitmap if we're on a multi_tt hub. + * + * Return: a pointer to a struct dwc2_tt. Don't forget to call + * dwc2_host_put_tt_info()! Returns NULL upon memory alloc failure. + */ + +struct dwc2_tt *dwc2_host_get_tt_info(struct dwc2_hsotg *hsotg, void *context, + gfp_t mem_flags, int *ttport) +{ + struct urb *urb = context; + struct dwc2_tt *dwc_tt = NULL; + + if (urb->dev->tt) { + *ttport = urb->dev->ttport; + + dwc_tt = urb->dev->tt->hcpriv; + if (dwc_tt == NULL) { + size_t bitmap_size; + + /* + * For single_tt we need one schedule. For multi_tt + * we need one per port. + */ + bitmap_size = DWC2_ELEMENTS_PER_LS_BITMAP * + sizeof(dwc_tt->periodic_bitmaps[0]); + if (urb->dev->tt->multi) + bitmap_size *= urb->dev->tt->hub->maxchild; + + dwc_tt = kzalloc(sizeof(*dwc_tt) + bitmap_size, + mem_flags); + if (dwc_tt == NULL) + return NULL; + + dwc_tt->usb_tt = urb->dev->tt; + dwc_tt->usb_tt->hcpriv = dwc_tt; + } + + dwc_tt->refcount++; + } + + return dwc_tt; +} + +/** + * dwc2_host_put_tt_info() - Put the dwc2_tt from dwc2_host_get_tt_info() + * + * Frees resources allocated by dwc2_host_get_tt_info() if all current holders + * of the structure are done. + * + * It's OK to call this with NULL. + * + * @hsotg: The HCD state structure for the DWC OTG controller. + * @dwc_tt: The pointer returned by dwc2_host_get_tt_info. + */ +void dwc2_host_put_tt_info(struct dwc2_hsotg *hsotg, struct dwc2_tt *dwc_tt) +{ + /* Model kfree and make put of NULL a no-op */ + if (dwc_tt == NULL) + return; + + WARN_ON(dwc_tt->refcount < 1); + + dwc_tt->refcount--; + if (!dwc_tt->refcount) { + dwc_tt->usb_tt->hcpriv = NULL; + kfree(dwc_tt); + } +} + int dwc2_host_get_speed(struct dwc2_hsotg *hsotg, void *context) { struct urb *urb = context; @@ -3197,9 +3281,6 @@ int dwc2_hcd_init(struct dwc2_hsotg *hsotg, int irq) hsotg->hc_ptr_array[i] = channel; } - if (hsotg->core_params->uframe_sched > 0) - dwc2_hcd_init_usecs(hsotg); - /* Initialize hsotg start work */ INIT_DELAYED_WORK(&hsotg->start_work, dwc2_hcd_start_func); diff --git a/drivers/usb/dwc2/hcd.h b/drivers/usb/dwc2/hcd.h index fd266ac53a28..140b1511a131 100644 --- a/drivers/usb/dwc2/hcd.h +++ b/drivers/usb/dwc2/hcd.h @@ -212,6 +212,43 @@ enum dwc2_transaction_type { DWC2_TRANSACTION_ALL, }; +/* The number of elements per LS bitmap (per port on multi_tt) */ +#define DWC2_ELEMENTS_PER_LS_BITMAP DIV_ROUND_UP(DWC2_LS_SCHEDULE_SLICES, \ + BITS_PER_LONG) + +/** + * struct dwc2_tt - dwc2 data associated with a usb_tt + * + * @refcount: Number of Queue Heads (QHs) holding a reference. + * @usb_tt: Pointer back to the official usb_tt. + * @periodic_bitmaps: Bitmap for which parts of the 1ms frame are accounted + * for already. Each is DWC2_ELEMENTS_PER_LS_BITMAP + * elements (so sizeof(long) times that in bytes). + * + * This structure is stored in the hcpriv of the official usb_tt. + */ +struct dwc2_tt { + int refcount; + struct usb_tt *usb_tt; + unsigned long periodic_bitmaps[]; +}; + +/** + * struct dwc2_hs_transfer_time - Info about a transfer on the high speed bus. + * + * @start_schedule_usecs: The start time on the main bus schedule. Note that + * the main bus schedule is tightly packed and this + * time should be interpreted as tightly packed (so + * uFrame 0 starts at 0 us, uFrame 1 starts at 100 us + * instead of 125 us). + * @duration_us: How long this transfer goes. + */ + +struct dwc2_hs_transfer_time { + u32 start_schedule_us; + u16 duration_us; +}; + /** * struct dwc2_qh - Software queue head structure * @@ -237,18 +274,33 @@ enum dwc2_transaction_type { * @td_first: Index of first activated isochronous transfer descriptor * @td_last: Index of last activated isochronous transfer descriptor * @host_us: Bandwidth in microseconds per transfer as seen by host + * @device_us: Bandwidth in microseconds per transfer as seen by device * @host_interval: Interval between transfers as seen by the host. If * the host is high speed and the device is low speed this * will be 8 times device interval. - * @next_active_frame: (Micro)frame before we next need to put something on + * @device_interval: Interval between transfers as seen by the device. + * interval. + * @next_active_frame: (Micro)frame _before_ we next need to put something on * the bus. We'll move the qh to active here. If the * host is in high speed mode this will be a uframe. If * the host is in low speed mode this will be a full frame. * @start_active_frame: If we are partway through a split transfer, this will be * what next_active_frame was when we started. Otherwise * it should always be the same as next_active_frame. - * @assigned_uframe: The uframe (0 -7) assigned by dwc2_find_uframe(). - * @frame_usecs: Internal variable used by the microframe scheduler + * @num_hs_transfers: Number of transfers in hs_transfers. + * Normally this is 1 but can be more than one for splits. + * Always >= 1 unless the host is in low/full speed mode. + * @hs_transfers: Transfers that are scheduled as seen by the high speed + * bus. Not used if host is in low or full speed mode (but + * note that it IS USED if the device is low or full speed + * as long as the HOST is in high speed mode). + * @ls_start_schedule_slice: Start time (in slices) on the low speed bus + * schedule that's being used by this device. This + * will be on the periodic_bitmap in a + * "struct dwc2_tt". Not used if this device is high + * speed. Note that this is in "schedule slice" which + * is tightly packed. + * @ls_duration_us: Duration on the low speed bus schedule. * @ntd: Actual number of transfer descriptors in a list * @qtd_list: List of QTDs for this QH * @channel: Host channel currently processing transfers for this QH @@ -261,8 +313,12 @@ enum dwc2_transaction_type { * descriptor and indicates original XferSize value for the * descriptor * @unreserve_timer: Timer for releasing periodic reservation. + * @dwc2_tt: Pointer to our tt info (or NULL if no tt). + * @ttport: Port number within our tt. * @tt_buffer_dirty True if clear_tt_buffer_complete is pending * @unreserve_pending: True if we planned to unreserve but haven't yet. + * @schedule_low_speed: True if we have a low/full speed component (either the + * host is in low/full speed mode or do_split). * * A Queue Head (QH) holds the static characteristics of an endpoint and * maintains a list of transfers (QTDs) for that endpoint. A QH structure may @@ -280,11 +336,14 @@ struct dwc2_qh { u8 td_first; u8 td_last; u16 host_us; + u16 device_us; u16 host_interval; + u16 device_interval; u16 next_active_frame; u16 start_active_frame; - u16 assigned_uframe; - u16 frame_usecs[8]; + s16 num_hs_transfers; + struct dwc2_hs_transfer_time hs_transfers[DWC2_HS_SCHEDULE_UFRAMES]; + u32 ls_start_schedule_slice; u16 ntd; struct list_head qtd_list; struct dwc2_host_chan *channel; @@ -294,8 +353,11 @@ struct dwc2_qh { u32 desc_list_sz; u32 *n_bytes; struct timer_list unreserve_timer; + struct dwc2_tt *dwc_tt; + int ttport; unsigned tt_buffer_dirty:1; unsigned unreserve_pending:1; + unsigned schedule_low_speed:1; }; /** @@ -462,7 +524,6 @@ extern void dwc2_hcd_queue_transactions(struct dwc2_hsotg *hsotg, /* Schedule Queue Functions */ /* Implemented in hcd_queue.c */ -extern void dwc2_hcd_init_usecs(struct dwc2_hsotg *hsotg); extern struct dwc2_qh *dwc2_hcd_qh_create(struct dwc2_hsotg *hsotg, struct dwc2_hcd_urb *urb, gfp_t mem_flags); @@ -728,6 +789,12 @@ extern void dwc2_host_start(struct dwc2_hsotg *hsotg); extern void dwc2_host_disconnect(struct dwc2_hsotg *hsotg); extern void dwc2_host_hub_info(struct dwc2_hsotg *hsotg, void *context, int *hub_addr, int *hub_port); +extern struct dwc2_tt *dwc2_host_get_tt_info(struct dwc2_hsotg *hsotg, + void *context, gfp_t mem_flags, + int *ttport); + +extern void dwc2_host_put_tt_info(struct dwc2_hsotg *hsotg, + struct dwc2_tt *dwc_tt); extern int dwc2_host_get_speed(struct dwc2_hsotg *hsotg, void *context); extern void dwc2_host_complete(struct dwc2_hsotg *hsotg, struct dwc2_qtd *qtd, int status); diff --git a/drivers/usb/dwc2/hcd_queue.c b/drivers/usb/dwc2/hcd_queue.c index 76f3c6596bef..7f634fd771c7 100644 --- a/drivers/usb/dwc2/hcd_queue.c +++ b/drivers/usb/dwc2/hcd_queue.c @@ -136,116 +136,933 @@ static int dwc2_check_periodic_bandwidth(struct dwc2_hsotg *hsotg, } /** - * Microframe scheduler - * track the total use in hsotg->frame_usecs - * keep each qh use in qh->frame_usecs - * when surrendering the qh then donate the time back + * pmap_schedule() - Schedule time in a periodic bitmap (pmap). + * + * @map: The bitmap representing the schedule; will be updated + * upon success. + * @bits_per_period: The schedule represents several periods. This is how many + * bits are in each period. It's assumed that the beginning + * of the schedule will repeat after its end. + * @periods_in_map: The number of periods in the schedule. + * @num_bits: The number of bits we need per period we want to reserve + * in this function call. + * @interval: How often we need to be scheduled for the reservation this + * time. 1 means every period. 2 means every other period. + * ...you get the picture? + * @start: The bit number to start at. Normally 0. Must be within + * the interval or we return failure right away. + * @only_one_period: Normally we'll allow picking a start anywhere within the + * first interval, since we can still make all repetition + * requirements by doing that. However, if you pass true + * here then we'll return failure if we can't fit within + * the period that "start" is in. + * + * The idea here is that we want to schedule time for repeating events that all + * want the same resource. The resource is divided into fixed-sized periods + * and the events want to repeat every "interval" periods. The schedule + * granularity is one bit. + * + * To keep things "simple", we'll represent our schedule with a bitmap that + * contains a fixed number of periods. This gets rid of a lot of complexity + * but does mean that we need to handle things specially (and non-ideally) if + * the number of the periods in the schedule doesn't match well with the + * intervals that we're trying to schedule. + * + * Here's an explanation of the scheme we'll implement, assuming 8 periods. + * - If interval is 1, we need to take up space in each of the 8 + * periods we're scheduling. Easy. + * - If interval is 2, we need to take up space in half of the + * periods. Again, easy. + * - If interval is 3, we actually need to fall back to interval 1. + * Why? Because we might need time in any period. AKA for the + * first 8 periods, we'll be in slot 0, 3, 6. Then we'll be + * in slot 1, 4, 7. Then we'll be in 2, 5. Then we'll be back to + * 0, 3, and 6. Since we could be in any frame we need to reserve + * for all of them. Sucks, but that's what you gotta do. Note that + * if we were instead scheduling 8 * 3 = 24 we'd do much better, but + * then we need more memory and time to do scheduling. + * - If interval is 4, easy. + * - If interval is 5, we again need interval 1. The schedule will be + * 0, 5, 2, 7, 4, 1, 6, 3, 0 + * - If interval is 6, we need interval 2. 0, 6, 4, 2. + * - If interval is 7, we need interval 1. + * - If interval is 8, we need interval 8. + * + * If you do the math, you'll see that we need to pretend that interval is + * equal to the greatest_common_divisor(interval, periods_in_map). + * + * Note that at the moment this function tends to front-pack the schedule. + * In some cases that's really non-ideal (it's hard to schedule things that + * need to repeat every period). In other cases it's perfect (you can easily + * schedule bigger, less often repeating things). + * + * Here's the algorithm in action (8 periods, 5 bits per period): + * |** | |** | |** | |** | | OK 2 bits, intv 2 at 0 + * |*****| ***|*****| ***|*****| ***|*****| ***| OK 3 bits, intv 3 at 2 + * |*****|* ***|*****| ***|*****|* ***|*****| ***| OK 1 bits, intv 4 at 5 + * |** |* |** | |** |* |** | | Remv 3 bits, intv 3 at 2 + * |*** |* |*** | |*** |* |*** | | OK 1 bits, intv 6 at 2 + * |**** |* * |**** | * |**** |* * |**** | * | OK 1 bits, intv 1 at 3 + * |**** |**** |**** | *** |**** |**** |**** | *** | OK 2 bits, intv 2 at 6 + * |*****|*****|*****| ****|*****|*****|*****| ****| OK 1 bits, intv 1 at 4 + * |*****|*****|*****| ****|*****|*****|*****| ****| FAIL 1 bits, intv 1 + * | ***|*****| ***| ****| ***|*****| ***| ****| Remv 2 bits, intv 2 at 0 + * | ***| ****| ***| ****| ***| ****| ***| ****| Remv 1 bits, intv 4 at 5 + * | **| ****| **| ****| **| ****| **| ****| Remv 1 bits, intv 6 at 2 + * | *| ** *| *| ** *| *| ** *| *| ** *| Remv 1 bits, intv 1 at 3 + * | *| *| *| *| *| *| *| *| Remv 2 bits, intv 2 at 6 + * | | | | | | | | | Remv 1 bits, intv 1 at 4 + * |** | |** | |** | |** | | OK 2 bits, intv 2 at 0 + * |*** | |** | |*** | |** | | OK 1 bits, intv 4 at 2 + * |*****| |** **| |*****| |** **| | OK 2 bits, intv 2 at 3 + * |*****|* |** **| |*****|* |** **| | OK 1 bits, intv 4 at 5 + * |*****|*** |** **| ** |*****|*** |** **| ** | OK 2 bits, intv 2 at 6 + * |*****|*****|** **| ****|*****|*****|** **| ****| OK 2 bits, intv 2 at 8 + * |*****|*****|*****| ****|*****|*****|*****| ****| OK 1 bits, intv 4 at 12 + * + * This function is pretty generic and could be easily abstracted if anything + * needed similar scheduling. + * + * Returns either -ENOSPC or a >= 0 start bit which should be passed to the + * unschedule routine. The map bitmap will be updated on a non-error result. */ -static const unsigned short max_uframe_usecs[] = { - 100, 100, 100, 100, 100, 100, 30, 0 -}; +static int pmap_schedule(unsigned long *map, int bits_per_period, + int periods_in_map, int num_bits, + int interval, int start, bool only_one_period) +{ + int interval_bits; + int to_reserve; + int first_end; + int i; + + if (num_bits > bits_per_period) + return -ENOSPC; + + /* Adjust interval as per description */ + interval = gcd(interval, periods_in_map); + + interval_bits = bits_per_period * interval; + to_reserve = periods_in_map / interval; + + /* If start has gotten us past interval then we can't schedule */ + if (start >= interval_bits) + return -ENOSPC; + + if (only_one_period) + /* Must fit within same period as start; end at begin of next */ + first_end = (start / bits_per_period + 1) * bits_per_period; + else + /* Can fit anywhere in the first interval */ + first_end = interval_bits; + + /* + * We'll try to pick the first repetition, then see if that time + * is free for each of the subsequent repetitions. If it's not + * we'll adjust the start time for the next search of the first + * repetition. + */ + while (start + num_bits <= first_end) { + int end; + + /* Need to stay within this period */ + end = (start / bits_per_period + 1) * bits_per_period; + + /* Look for num_bits us in this microframe starting at start */ + start = bitmap_find_next_zero_area(map, end, start, num_bits, + 0); + + /* + * We should get start >= end if we fail. We might be + * able to check the next microframe depending on the + * interval, so continue on (start already updated). + */ + if (start >= end) { + start = end; + continue; + } + + /* At this point we have a valid point for first one */ + for (i = 1; i < to_reserve; i++) { + int ith_start = start + interval_bits * i; + int ith_end = end + interval_bits * i; + int ret; + + /* Use this as a dumb "check if bits are 0" */ + ret = bitmap_find_next_zero_area( + map, ith_start + num_bits, ith_start, num_bits, + 0); + + /* We got the right place, continue checking */ + if (ret == ith_start) + continue; + + /* Move start up for next time and exit for loop */ + ith_start = bitmap_find_next_zero_area( + map, ith_end, ith_start, num_bits, 0); + if (ith_start >= ith_end) + /* Need a while new period next time */ + start = end; + else + start = ith_start - interval_bits * i; + break; + } + + /* If didn't exit the for loop with a break, we have success */ + if (i == to_reserve) + break; + } + + if (start + num_bits > first_end) + return -ENOSPC; -void dwc2_hcd_init_usecs(struct dwc2_hsotg *hsotg) + for (i = 0; i < to_reserve; i++) { + int ith_start = start + interval_bits * i; + + bitmap_set(map, ith_start, num_bits); + } + + return start; +} + +/** + * pmap_unschedule() - Undo work done by pmap_schedule() + * + * @map: See pmap_schedule(). + * @bits_per_period: See pmap_schedule(). + * @periods_in_map: See pmap_schedule(). + * @num_bits: The number of bits that was passed to schedule. + * @interval: The interval that was passed to schedule. + * @start: The return value from pmap_schedule(). + */ +static void pmap_unschedule(unsigned long *map, int bits_per_period, + int periods_in_map, int num_bits, + int interval, int start) { + int interval_bits; + int to_release; int i; - for (i = 0; i < 8; i++) - hsotg->frame_usecs[i] = max_uframe_usecs[i]; + /* Adjust interval as per description in pmap_schedule() */ + interval = gcd(interval, periods_in_map); + + interval_bits = bits_per_period * interval; + to_release = periods_in_map / interval; + + for (i = 0; i < to_release; i++) { + int ith_start = start + interval_bits * i; + + bitmap_clear(map, ith_start, num_bits); + } } -static int dwc2_find_single_uframe(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh) +/* + * cat_printf() - A printf() + strcat() helper + * + * This is useful for concatenating a bunch of strings where each string is + * constructed using printf. + * + * @buf: The destination buffer; will be updated to point after the printed + * data. + * @size: The number of bytes in the buffer (includes space for '\0'). + * @fmt: The format for printf. + * @...: The args for printf. + */ +static void cat_printf(char **buf, size_t *size, const char *fmt, ...) { - unsigned short utime = qh->host_us; + va_list args; int i; - for (i = 0; i < 8; i++) { - /* At the start hsotg->frame_usecs[i] = max_uframe_usecs[i] */ - if (utime <= hsotg->frame_usecs[i]) { - hsotg->frame_usecs[i] -= utime; - qh->frame_usecs[i] += utime; - return i; - } + if (*size == 0) + return; + + va_start(args, fmt); + i = vsnprintf(*buf, *size, fmt, args); + va_end(args); + + if (i >= *size) { + (*buf)[*size - 1] = '\0'; + *buf += *size; + *size = 0; + } else { + *buf += i; + *size -= i; } - return -ENOSPC; } /* - * use this for FS apps that can span multiple uframes + * pmap_print() - Print the given periodic map + * + * Will attempt to print out the periodic schedule. + * + * @map: See pmap_schedule(). + * @bits_per_period: See pmap_schedule(). + * @periods_in_map: See pmap_schedule(). + * @period_name: The name of 1 period, like "uFrame" + * @units: The name of the units, like "us". + * @print_fn: The function to call for printing. + * @print_data: Opaque data to pass to the print function. */ -static int dwc2_find_multi_uframe(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh) +static void pmap_print(unsigned long *map, int bits_per_period, + int periods_in_map, const char *period_name, + const char *units, + void (*print_fn)(const char *str, void *data), + void *print_data) { - unsigned short utime = qh->host_us; - unsigned short xtime; - int t_left; + int period; + + for (period = 0; period < periods_in_map; period++) { + char tmp[64]; + char *buf = tmp; + size_t buf_size = sizeof(tmp); + int period_start = period * bits_per_period; + int period_end = period_start + bits_per_period; + int start = 0; + int count = 0; + bool printed = false; + int i; + + for (i = period_start; i < period_end + 1; i++) { + /* Handle case when ith bit is set */ + if (i < period_end && + bitmap_find_next_zero_area(map, i + 1, + i, 1, 0) != i) { + if (count == 0) + start = i - period_start; + count++; + continue; + } + + /* ith bit isn't set; don't care if count == 0 */ + if (count == 0) + continue; + + if (!printed) + cat_printf(&buf, &buf_size, "%s %d: ", + period_name, period); + else + cat_printf(&buf, &buf_size, ", "); + printed = true; + + cat_printf(&buf, &buf_size, "%d %s -%3d %s", start, + units, start + count - 1, units); + count = 0; + } + + if (printed) + print_fn(tmp, print_data); + } +} + +/** + * dwc2_get_ls_map() - Get the map used for the given qh + * + * @hsotg: The HCD state structure for the DWC OTG controller. + * @qh: QH for the periodic transfer. + * + * We'll always get the periodic map out of our TT. Note that even if we're + * running the host straight in low speed / full speed mode it appears as if + * a TT is allocated for us, so we'll use it. If that ever changes we can + * add logic here to get a map out of "hsotg" if !qh->do_split. + * + * Returns: the map or NULL if a map couldn't be found. + */ +static unsigned long *dwc2_get_ls_map(struct dwc2_hsotg *hsotg, + struct dwc2_qh *qh) +{ + unsigned long *map; + + /* Don't expect to be missing a TT and be doing low speed scheduling */ + if (WARN_ON(!qh->dwc_tt)) + return NULL; + + /* Get the map and adjust if this is a multi_tt hub */ + map = qh->dwc_tt->periodic_bitmaps; + if (qh->dwc_tt->usb_tt->multi) + map += DWC2_ELEMENTS_PER_LS_BITMAP * qh->ttport; + + return map; +} + +struct dwc2_qh_print_data { + struct dwc2_hsotg *hsotg; + struct dwc2_qh *qh; +}; + +/** + * dwc2_qh_print() - Helper function for dwc2_qh_schedule_print() + * + * @str: The string to print + * @data: A pointer to a struct dwc2_qh_print_data + */ +static void dwc2_qh_print(const char *str, void *data) +{ + struct dwc2_qh_print_data *print_data = data; + + dwc2_sch_dbg(print_data->hsotg, "QH=%p ...%s\n", print_data->qh, str); +} + +/** + * dwc2_qh_schedule_print() - Print the periodic schedule + * + * @hsotg: The HCD state structure for the DWC OTG controller. + * @qh: QH to print. + */ +static void dwc2_qh_schedule_print(struct dwc2_hsotg *hsotg, + struct dwc2_qh *qh) +{ + struct dwc2_qh_print_data print_data = { hsotg, qh }; int i; - int j; - int k; - for (i = 0; i < 8; i++) { - if (hsotg->frame_usecs[i] <= 0) + /* + * The printing functions are quite slow and inefficient. + * If we don't have tracing turned on, don't run unless the special + * define is turned on. + */ +#ifndef DWC2_PRINT_SCHEDULE + return; +#endif + + if (qh->schedule_low_speed) { + unsigned long *map = dwc2_get_ls_map(hsotg, qh); + + dwc2_sch_dbg(hsotg, "QH=%p LS/FS trans: %d=>%d us @ %d us", + qh, qh->device_us, + DWC2_ROUND_US_TO_SLICE(qh->device_us), + DWC2_US_PER_SLICE * qh->ls_start_schedule_slice); + + if (map) { + dwc2_sch_dbg(hsotg, + "QH=%p Whole low/full speed map %p now:\n", + qh, map); + pmap_print(map, DWC2_LS_PERIODIC_SLICES_PER_FRAME, + DWC2_LS_SCHEDULE_FRAMES, "Frame ", "slices", + dwc2_qh_print, &print_data); + } + } + + for (i = 0; i < qh->num_hs_transfers; i++) { + struct dwc2_hs_transfer_time *trans_time = qh->hs_transfers + i; + int uframe = trans_time->start_schedule_us / + DWC2_HS_PERIODIC_US_PER_UFRAME; + int rel_us = trans_time->start_schedule_us % + DWC2_HS_PERIODIC_US_PER_UFRAME; + + dwc2_sch_dbg(hsotg, + "QH=%p HS trans #%d: %d us @ uFrame %d + %d us\n", + qh, i, trans_time->duration_us, uframe, rel_us); + } + if (qh->num_hs_transfers) { + dwc2_sch_dbg(hsotg, "QH=%p Whole high speed map now:\n", qh); + pmap_print(hsotg->hs_periodic_bitmap, + DWC2_HS_PERIODIC_US_PER_UFRAME, + DWC2_HS_SCHEDULE_UFRAMES, "uFrame", "us", + dwc2_qh_print, &print_data); + } + +} + +/** + * dwc2_ls_pmap_schedule() - Schedule a low speed QH + * + * @hsotg: The HCD state structure for the DWC OTG controller. + * @qh: QH for the periodic transfer. + * @search_slice: We'll start trying to schedule at the passed slice. + * Remember that slices are the units of the low speed + * schedule (think 25us or so). + * + * Wraps pmap_schedule() with the right parameters for low speed scheduling. + * + * Normally we schedule low speed devices on the map associated with the TT. + * + * Returns: 0 for success or an error code. + */ +static int dwc2_ls_pmap_schedule(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh, + int search_slice) +{ + int slices = DIV_ROUND_UP(qh->device_us, DWC2_US_PER_SLICE); + unsigned long *map = dwc2_get_ls_map(hsotg, qh); + int slice; + + if (map == NULL) + return -EINVAL; + + /* + * Schedule on the proper low speed map with our low speed scheduling + * parameters. Note that we use the "device_interval" here since + * we want the low speed interval and the only way we'd be in this + * function is if the device is low speed. + * + * If we happen to be doing low speed and high speed scheduling for the + * same transaction (AKA we have a split) we always do low speed first. + * That means we can always pass "false" for only_one_period (that + * parameters is only useful when we're trying to get one schedule to + * match what we already planned in the other schedule). + */ + slice = pmap_schedule(map, DWC2_LS_PERIODIC_SLICES_PER_FRAME, + DWC2_LS_SCHEDULE_FRAMES, slices, + qh->device_interval, search_slice, false); + + if (slice < 0) + return slice; + + qh->ls_start_schedule_slice = slice; + return 0; +} + +/** + * dwc2_ls_pmap_unschedule() - Undo work done by dwc2_ls_pmap_schedule() + * + * @hsotg: The HCD state structure for the DWC OTG controller. + * @qh: QH for the periodic transfer. + */ +static void dwc2_ls_pmap_unschedule(struct dwc2_hsotg *hsotg, + struct dwc2_qh *qh) +{ + int slices = DIV_ROUND_UP(qh->device_us, DWC2_US_PER_SLICE); + unsigned long *map = dwc2_get_ls_map(hsotg, qh); + + /* Schedule should have failed, so no worries about no error code */ + if (map == NULL) + return; + + pmap_unschedule(map, DWC2_LS_PERIODIC_SLICES_PER_FRAME, + DWC2_LS_SCHEDULE_FRAMES, slices, qh->device_interval, + qh->ls_start_schedule_slice); +} + +/** + * dwc2_hs_pmap_schedule - Schedule in the main high speed schedule + * + * This will schedule something on the main dwc2 schedule. + * + * We'll start looking in qh->hs_transfers[index].start_schedule_us. We'll + * update this with the result upon success. We also use the duration from + * the same structure. + * + * @hsotg: The HCD state structure for the DWC OTG controller. + * @qh: QH for the periodic transfer. + * @only_one_period: If true we will limit ourselves to just looking at + * one period (aka one 100us chunk). This is used if we have + * already scheduled something on the low speed schedule and + * need to find something that matches on the high speed one. + * @index: The index into qh->hs_transfers that we're working with. + * + * Returns: 0 for success or an error code. Upon success the + * dwc2_hs_transfer_time specified by "index" will be updated. + */ +static int dwc2_hs_pmap_schedule(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh, + bool only_one_period, int index) +{ + struct dwc2_hs_transfer_time *trans_time = qh->hs_transfers + index; + int us; + + us = pmap_schedule(hsotg->hs_periodic_bitmap, + DWC2_HS_PERIODIC_US_PER_UFRAME, + DWC2_HS_SCHEDULE_UFRAMES, trans_time->duration_us, + qh->host_interval, trans_time->start_schedule_us, + only_one_period); + + if (us < 0) + return us; + + trans_time->start_schedule_us = us; + return 0; +} + +/** + * dwc2_ls_pmap_unschedule() - Undo work done by dwc2_hs_pmap_schedule() + * + * @hsotg: The HCD state structure for the DWC OTG controller. + * @qh: QH for the periodic transfer. + */ +static void dwc2_hs_pmap_unschedule(struct dwc2_hsotg *hsotg, + struct dwc2_qh *qh, int index) +{ + struct dwc2_hs_transfer_time *trans_time = qh->hs_transfers + index; + + pmap_unschedule(hsotg->hs_periodic_bitmap, + DWC2_HS_PERIODIC_US_PER_UFRAME, + DWC2_HS_SCHEDULE_UFRAMES, trans_time->duration_us, + qh->host_interval, trans_time->start_schedule_us); +} + +/** + * dwc2_uframe_schedule_split - Schedule a QH for a periodic split xfer. + * + * This is the most complicated thing in USB. We have to find matching time + * in both the global high speed schedule for the port and the low speed + * schedule for the TT associated with the given device. + * + * Being here means that the host must be running in high speed mode and the + * device is in low or full speed mode (and behind a hub). + * + * @hsotg: The HCD state structure for the DWC OTG controller. + * @qh: QH for the periodic transfer. + */ +static int dwc2_uframe_schedule_split(struct dwc2_hsotg *hsotg, + struct dwc2_qh *qh) +{ + int bytecount = dwc2_hb_mult(qh->maxp) * dwc2_max_packet(qh->maxp); + int ls_search_slice; + int err = 0; + int host_interval_in_sched; + + /* + * The interval (how often to repeat) in the actual host schedule. + * See pmap_schedule() for gcd() explanation. + */ + host_interval_in_sched = gcd(qh->host_interval, + DWC2_HS_SCHEDULE_UFRAMES); + + /* + * We always try to find space in the low speed schedule first, then + * try to find high speed time that matches. If we don't, we'll bump + * up the place we start searching in the low speed schedule and try + * again. To start we'll look right at the beginning of the low speed + * schedule. + * + * Note that this will tend to front-load the high speed schedule. + * We may eventually want to try to avoid this by either considering + * both schedules together or doing some sort of round robin. + */ + ls_search_slice = 0; + + while (ls_search_slice < DWC2_LS_SCHEDULE_SLICES) { + int start_s_uframe; + int ssplit_s_uframe; + int second_s_uframe; + int rel_uframe; + int first_count; + int middle_count; + int end_count; + int first_data_bytes; + int other_data_bytes; + int i; + + if (qh->schedule_low_speed) { + err = dwc2_ls_pmap_schedule(hsotg, qh, ls_search_slice); + + /* + * If we got an error here there's no other magic we + * can do, so bail. All the looping above is only + * helpful to redo things if we got a low speed slot + * and then couldn't find a matching high speed slot. + */ + if (err) + return err; + } else { + /* Must be missing the tt structure? Why? */ + WARN_ON_ONCE(1); + } + + /* + * This will give us a number 0 - 7 if + * DWC2_LS_SCHEDULE_FRAMES == 1, or 0 - 15 if == 2, or ... + */ + start_s_uframe = qh->ls_start_schedule_slice / + DWC2_SLICES_PER_UFRAME; + + /* Get a number that's always 0 - 7 */ + rel_uframe = (start_s_uframe % 8); + + /* + * If we were going to start in uframe 7 then we would need to + * issue a start split in uframe 6, which spec says is not OK. + * Move on to the next full frame (assuming there is one). + * + * See 11.18.4 Host Split Transaction Scheduling Requirements + * bullet 1. + */ + if (rel_uframe == 7) { + if (qh->schedule_low_speed) + dwc2_ls_pmap_unschedule(hsotg, qh); + ls_search_slice = + (qh->ls_start_schedule_slice / + DWC2_LS_PERIODIC_SLICES_PER_FRAME + 1) * + DWC2_LS_PERIODIC_SLICES_PER_FRAME; continue; + } /* - * we need n consecutive slots so use j as a start slot - * j plus j+1 must be enough time (for now) + * For ISOC in: + * - start split (frame -1) + * - complete split w/ data (frame +1) + * - complete split w/ data (frame +2) + * - ... + * - complete split w/ data (frame +num_data_packets) + * - complete split w/ data (frame +num_data_packets+1) + * - complete split w/ data (frame +num_data_packets+2, max 8) + * ...though if frame was "0" then max is 7... + * + * For ISOC out we might need to do: + * - start split w/ data (frame -1) + * - start split w/ data (frame +0) + * - ... + * - start split w/ data (frame +num_data_packets-2) + * + * For INTERRUPT in we might need to do: + * - start split (frame -1) + * - complete split w/ data (frame +1) + * - complete split w/ data (frame +2) + * - complete split w/ data (frame +3, max 8) + * + * For INTERRUPT out we might need to do: + * - start split w/ data (frame -1) + * - complete split (frame +1) + * - complete split (frame +2) + * - complete split (frame +3, max 8) + * + * Start adjusting! */ - xtime = hsotg->frame_usecs[i]; - for (j = i + 1; j < 8; j++) { - /* - * if we add this frame remaining time to xtime we may - * be OK, if not we need to test j for a complete frame - */ - if (xtime + hsotg->frame_usecs[j] < utime) { - if (hsotg->frame_usecs[j] < - max_uframe_usecs[j]) - continue; + ssplit_s_uframe = (start_s_uframe + + host_interval_in_sched - 1) % + host_interval_in_sched; + if (qh->ep_type == USB_ENDPOINT_XFER_ISOC && !qh->ep_is_in) + second_s_uframe = start_s_uframe; + else + second_s_uframe = start_s_uframe + 1; + + /* First data transfer might not be all 188 bytes. */ + first_data_bytes = 188 - + DIV_ROUND_UP(188 * (qh->ls_start_schedule_slice % + DWC2_SLICES_PER_UFRAME), + DWC2_SLICES_PER_UFRAME); + if (first_data_bytes > bytecount) + first_data_bytes = bytecount; + other_data_bytes = bytecount - first_data_bytes; + + /* + * For now, skip OUT xfers where first xfer is partial + * + * Main dwc2 code assumes: + * - INT transfers never get split in two. + * - ISOC transfers can always transfer 188 bytes the first + * time. + * + * Until that code is fixed, try again if the first transfer + * couldn't transfer everything. + * + * This code can be removed if/when the rest of dwc2 handles + * the above cases. Until it's fixed we just won't be able + * to schedule quite as tightly. + */ + if (!qh->ep_is_in && + (first_data_bytes != min_t(int, 188, bytecount))) { + dwc2_sch_dbg(hsotg, + "QH=%p avoiding broken 1st xfer (%d, %d)\n", + qh, first_data_bytes, bytecount); + if (qh->schedule_low_speed) + dwc2_ls_pmap_unschedule(hsotg, qh); + ls_search_slice = (start_s_uframe + 1) * + DWC2_SLICES_PER_UFRAME; + continue; + } + + /* Start by assuming transfers for the bytes */ + qh->num_hs_transfers = 1 + DIV_ROUND_UP(other_data_bytes, 188); + + /* + * Everything except ISOC OUT has extra transfers. Rules are + * complicated. See 11.18.4 Host Split Transaction Scheduling + * Requirements bullet 3. + */ + if (qh->ep_type == USB_ENDPOINT_XFER_INT) { + if (rel_uframe == 6) + qh->num_hs_transfers += 2; + else + qh->num_hs_transfers += 3; + + if (qh->ep_is_in) { + /* + * First is start split, middle/end is data. + * Allocate full data bytes for all data. + */ + first_count = 4; + middle_count = bytecount; + end_count = bytecount; + } else { + /* + * First is data, middle/end is complete. + * First transfer and second can have data. + * Rest should just have complete split. + */ + first_count = first_data_bytes; + middle_count = max_t(int, 4, other_data_bytes); + end_count = 4; } - if (xtime >= utime) { - t_left = utime; - for (k = i; k < 8; k++) { - t_left -= hsotg->frame_usecs[k]; - if (t_left <= 0) { - qh->frame_usecs[k] += - hsotg->frame_usecs[k] - + t_left; - hsotg->frame_usecs[k] = -t_left; - return i; - } else { - qh->frame_usecs[k] += - hsotg->frame_usecs[k]; - hsotg->frame_usecs[k] = 0; - } - } + } else { + if (qh->ep_is_in) { + int last; + + /* Account for the start split */ + qh->num_hs_transfers++; + + /* Calculate "L" value from spec */ + last = rel_uframe + qh->num_hs_transfers + 1; + + /* Start with basic case */ + if (last <= 6) + qh->num_hs_transfers += 2; + else + qh->num_hs_transfers += 1; + + /* Adjust downwards */ + if (last >= 6 && rel_uframe == 0) + qh->num_hs_transfers--; + + /* 1st = start; rest can contain data */ + first_count = 4; + middle_count = min_t(int, 188, bytecount); + end_count = middle_count; + } else { + /* All contain data, last might be smaller */ + first_count = first_data_bytes; + middle_count = min_t(int, 188, + other_data_bytes); + end_count = other_data_bytes % 188; } - /* add the frame time to x time */ - xtime += hsotg->frame_usecs[j]; - /* we must have a fully available next frame or break */ - if (xtime < utime && - hsotg->frame_usecs[j] == max_uframe_usecs[j]) - continue; } + + /* Assign durations per uFrame */ + qh->hs_transfers[0].duration_us = HS_USECS_ISO(first_count); + for (i = 1; i < qh->num_hs_transfers - 1; i++) + qh->hs_transfers[i].duration_us = + HS_USECS_ISO(middle_count); + if (qh->num_hs_transfers > 1) + qh->hs_transfers[qh->num_hs_transfers - 1].duration_us = + HS_USECS_ISO(end_count); + + /* + * Assign start us. The call below to dwc2_hs_pmap_schedule() + * will start with these numbers but may adjust within the same + * microframe. + */ + qh->hs_transfers[0].start_schedule_us = + ssplit_s_uframe * DWC2_HS_PERIODIC_US_PER_UFRAME; + for (i = 1; i < qh->num_hs_transfers; i++) + qh->hs_transfers[i].start_schedule_us = + ((second_s_uframe + i - 1) % + DWC2_HS_SCHEDULE_UFRAMES) * + DWC2_HS_PERIODIC_US_PER_UFRAME; + + /* Try to schedule with filled in hs_transfers above */ + for (i = 0; i < qh->num_hs_transfers; i++) { + err = dwc2_hs_pmap_schedule(hsotg, qh, true, i); + if (err) + break; + } + + /* If we scheduled all w/out breaking out then we're all good */ + if (i == qh->num_hs_transfers) + break; + + for (; i >= 0; i--) + dwc2_hs_pmap_unschedule(hsotg, qh, i); + + if (qh->schedule_low_speed) + dwc2_ls_pmap_unschedule(hsotg, qh); + + /* Try again starting in the next microframe */ + ls_search_slice = (start_s_uframe + 1) * DWC2_SLICES_PER_UFRAME; } - return -ENOSPC; + + if (ls_search_slice >= DWC2_LS_SCHEDULE_SLICES) + return -ENOSPC; + + return 0; } -static int dwc2_find_uframe(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh) +/** + * dwc2_uframe_schedule_hs - Schedule a QH for a periodic high speed xfer. + * + * Basically this just wraps dwc2_hs_pmap_schedule() to provide a clean + * interface. + * + * @hsotg: The HCD state structure for the DWC OTG controller. + * @qh: QH for the periodic transfer. + */ +static int dwc2_uframe_schedule_hs(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh) +{ + /* In non-split host and device time are the same */ + WARN_ON(qh->host_us != qh->device_us); + WARN_ON(qh->host_interval != qh->device_interval); + WARN_ON(qh->num_hs_transfers != 1); + + /* We'll have one transfer; init start to 0 before calling scheduler */ + qh->hs_transfers[0].start_schedule_us = 0; + qh->hs_transfers[0].duration_us = qh->host_us; + + return dwc2_hs_pmap_schedule(hsotg, qh, false, 0); +} + +/** + * dwc2_uframe_schedule_ls - Schedule a QH for a periodic low/full speed xfer. + * + * Basically this just wraps dwc2_ls_pmap_schedule() to provide a clean + * interface. + * + * @hsotg: The HCD state structure for the DWC OTG controller. + * @qh: QH for the periodic transfer. + */ +static int dwc2_uframe_schedule_ls(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh) +{ + /* In non-split host and device time are the same */ + WARN_ON(qh->host_us != qh->device_us); + WARN_ON(qh->host_interval != qh->device_interval); + WARN_ON(!qh->schedule_low_speed); + + /* Run on the main low speed schedule (no split = no hub = no TT) */ + return dwc2_ls_pmap_schedule(hsotg, qh, 0); +} + +/** + * dwc2_uframe_schedule - Schedule a QH for a periodic xfer. + * + * Calls one of the 3 sub-function depending on what type of transfer this QH + * is for. Also adds some printing. + * + * @hsotg: The HCD state structure for the DWC OTG controller. + * @qh: QH for the periodic transfer. + */ +static int dwc2_uframe_schedule(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh) { int ret; - if (qh->dev_speed == USB_SPEED_HIGH) { - /* if this is a hs transaction we need a full frame */ - ret = dwc2_find_single_uframe(hsotg, qh); - } else { - /* - * if this is a fs transaction we may need a sequence - * of frames - */ - ret = dwc2_find_multi_uframe(hsotg, qh); - } + if (qh->dev_speed == USB_SPEED_HIGH) + ret = dwc2_uframe_schedule_hs(hsotg, qh); + else if (!qh->do_split) + ret = dwc2_uframe_schedule_ls(hsotg, qh); + else + ret = dwc2_uframe_schedule_split(hsotg, qh); + + if (ret) + dwc2_sch_dbg(hsotg, "QH=%p Failed to schedule %d\n", qh, ret); + else + dwc2_qh_schedule_print(hsotg, qh); + return ret; } /** + * dwc2_uframe_unschedule - Undoes dwc2_uframe_schedule(). + * + * @hsotg: The HCD state structure for the DWC OTG controller. + * @qh: QH for the periodic transfer. + */ +static void dwc2_uframe_unschedule(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh) +{ + int i; + + for (i = 0; i < qh->num_hs_transfers; i++) + dwc2_hs_pmap_unschedule(hsotg, qh, i); + + if (qh->schedule_low_speed) + dwc2_ls_pmap_unschedule(hsotg, qh); + + dwc2_sch_dbg(hsotg, "QH=%p Unscheduled\n", qh); +} + +/** * dwc2_pick_first_frame() - Choose 1st frame for qh that's already scheduled * * Takes a qh that has already been scheduled (which means we know we have the @@ -265,6 +1082,7 @@ static void dwc2_pick_first_frame(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh) u16 frame_number; u16 earliest_frame; u16 next_active_frame; + u16 relative_frame; u16 interval; /* @@ -292,8 +1110,36 @@ static void dwc2_pick_first_frame(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh) goto exit; } - /* Adjust interval as per high speed schedule which has 8 uFrame */ - interval = gcd(qh->host_interval, 8); + if (qh->dev_speed == USB_SPEED_HIGH || qh->do_split) { + /* + * We're either at high speed or we're doing a split (which + * means we're talking high speed to a hub). In any case + * the first frame should be based on when the first scheduled + * event is. + */ + WARN_ON(qh->num_hs_transfers < 1); + + relative_frame = qh->hs_transfers[0].start_schedule_us / + DWC2_HS_PERIODIC_US_PER_UFRAME; + + /* Adjust interval as per high speed schedule */ + interval = gcd(qh->host_interval, DWC2_HS_SCHEDULE_UFRAMES); + + } else { + /* + * Low or full speed directly on dwc2. Just about the same + * as high speed but on a different schedule and with slightly + * different adjustments. Note that this works because when + * the host and device are both low speed then frames in the + * controller tick at low speed. + */ + relative_frame = qh->ls_start_schedule_slice / + DWC2_LS_PERIODIC_SLICES_PER_FRAME; + interval = gcd(qh->host_interval, DWC2_LS_SCHEDULE_FRAMES); + } + + /* Scheduler messed up if frame is past interval */ + WARN_ON(relative_frame >= interval); /* * We know interval must divide (HFNUM_MAX_FRNUM + 1) now that we've @@ -310,7 +1156,7 @@ static void dwc2_pick_first_frame(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh) * scheduled for. */ next_active_frame = dwc2_frame_num_inc(next_active_frame, - qh->assigned_uframe); + relative_frame); /* * We actually need 1 frame before since the next_active_frame is @@ -351,9 +1197,7 @@ static int dwc2_do_reserve(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh) int status; if (hsotg->core_params->uframe_sched > 0) { - status = dwc2_find_uframe(hsotg, qh); - if (status >= 0) - qh->assigned_uframe = status; + status = dwc2_uframe_schedule(hsotg, qh); } else { status = dwc2_periodic_channel_available(hsotg); if (status) { @@ -410,12 +1254,7 @@ static void dwc2_do_unreserve(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh) hsotg->periodic_usecs -= qh->host_us; if (hsotg->core_params->uframe_sched > 0) { - int i; - - for (i = 0; i < 8; i++) { - hsotg->frame_usecs[i] += qh->frame_usecs[i]; - qh->frame_usecs[i] = 0; - } + dwc2_uframe_unschedule(hsotg, qh); } else { /* Release periodic channel reservation */ hsotg->periodic_channels--; @@ -606,88 +1445,81 @@ static void dwc2_deschedule_periodic(struct dwc2_hsotg *hsotg, * @qh: The QH to init * @urb: Holds the information about the device/endpoint needed to initialize * the QH + * @mem_flags: Flags for allocating memory. */ static void dwc2_qh_init(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh, - struct dwc2_hcd_urb *urb) + struct dwc2_hcd_urb *urb, gfp_t mem_flags) { - int dev_speed, hub_addr, hub_port; + int dev_speed = dwc2_host_get_speed(hsotg, urb->priv); + u8 ep_type = dwc2_hcd_get_pipe_type(&urb->pipe_info); + bool ep_is_in = !!dwc2_hcd_is_pipe_in(&urb->pipe_info); + bool ep_is_isoc = (ep_type == USB_ENDPOINT_XFER_ISOC); + bool ep_is_int = (ep_type == USB_ENDPOINT_XFER_INT); + u32 hprt = dwc2_readl(hsotg->regs + HPRT0); + u32 prtspd = (hprt & HPRT0_SPD_MASK) >> HPRT0_SPD_SHIFT; + bool do_split = (prtspd == HPRT0_SPD_HIGH_SPEED && + dev_speed != USB_SPEED_HIGH); + int maxp = dwc2_hcd_get_mps(&urb->pipe_info); + int bytecount = dwc2_hb_mult(maxp) * dwc2_max_packet(maxp); char *speed, *type; - dev_vdbg(hsotg->dev, "%s()\n", __func__); - /* Initialize QH */ qh->hsotg = hsotg; setup_timer(&qh->unreserve_timer, dwc2_unreserve_timer_fn, (unsigned long)qh); - qh->ep_type = dwc2_hcd_get_pipe_type(&urb->pipe_info); - qh->ep_is_in = dwc2_hcd_is_pipe_in(&urb->pipe_info) ? 1 : 0; + qh->ep_type = ep_type; + qh->ep_is_in = ep_is_in; qh->data_toggle = DWC2_HC_PID_DATA0; - qh->maxp = dwc2_hcd_get_mps(&urb->pipe_info); + qh->maxp = maxp; INIT_LIST_HEAD(&qh->qtd_list); INIT_LIST_HEAD(&qh->qh_list_entry); - /* FS/LS Endpoint on HS Hub, NOT virtual root hub */ - dev_speed = dwc2_host_get_speed(hsotg, urb->priv); + qh->do_split = do_split; + qh->dev_speed = dev_speed; + + if (ep_is_int || ep_is_isoc) { + /* Compute scheduling parameters once and save them */ + int host_speed = do_split ? USB_SPEED_HIGH : dev_speed; + struct dwc2_tt *dwc_tt = dwc2_host_get_tt_info(hsotg, urb->priv, + mem_flags, + &qh->ttport); + int device_ns; - dwc2_host_hub_info(hsotg, urb->priv, &hub_addr, &hub_port); + qh->dwc_tt = dwc_tt; - if ((dev_speed == USB_SPEED_LOW || dev_speed == USB_SPEED_FULL) && - hub_addr != 0 && hub_addr != 1) { - dev_vdbg(hsotg->dev, - "QH init: EP %d: TT found at hub addr %d, for port %d\n", - dwc2_hcd_get_ep_num(&urb->pipe_info), hub_addr, - hub_port); - qh->do_split = 1; - } + qh->host_us = NS_TO_US(usb_calc_bus_time(host_speed, ep_is_in, + ep_is_isoc, bytecount)); + device_ns = usb_calc_bus_time(dev_speed, ep_is_in, + ep_is_isoc, bytecount); - if (qh->ep_type == USB_ENDPOINT_XFER_INT || - qh->ep_type == USB_ENDPOINT_XFER_ISOC) { - /* Compute scheduling parameters once and save them */ - u32 hprt, prtspd; - - /* Todo: Account for split transfers in the bus time */ - int bytecount = - dwc2_hb_mult(qh->maxp) * dwc2_max_packet(qh->maxp); - - qh->host_us = NS_TO_US(usb_calc_bus_time(qh->do_split ? - USB_SPEED_HIGH : dev_speed, qh->ep_is_in, - qh->ep_type == USB_ENDPOINT_XFER_ISOC, - bytecount)); - - qh->host_interval = urb->interval; - dwc2_sch_dbg(hsotg, "QH=%p init nxt=%04x, fn=%04x, int=%#x\n", - qh, qh->next_active_frame, hsotg->frame_number, - qh->host_interval); -#if 0 - /* Increase interrupt polling rate for debugging */ - if (qh->ep_type == USB_ENDPOINT_XFER_INT) - qh->host_interval = 8; -#endif - hprt = dwc2_readl(hsotg->regs + HPRT0); - prtspd = (hprt & HPRT0_SPD_MASK) >> HPRT0_SPD_SHIFT; - if (prtspd == HPRT0_SPD_HIGH_SPEED && - (dev_speed == USB_SPEED_LOW || - dev_speed == USB_SPEED_FULL)) { - qh->host_interval *= 8; - dwc2_sch_dbg(hsotg, - "QH=%p init*8 nxt=%04x, fn=%04x, int=%#x\n", - qh, qh->next_active_frame, - hsotg->frame_number, qh->host_interval); + if (do_split && dwc_tt) + device_ns += dwc_tt->usb_tt->think_time; + qh->device_us = NS_TO_US(device_ns); - } - dev_dbg(hsotg->dev, "interval=%d\n", qh->host_interval); - } - dev_vdbg(hsotg->dev, "DWC OTG HCD QH Initialized\n"); - dev_vdbg(hsotg->dev, "DWC OTG HCD QH - qh = %p\n", qh); - dev_vdbg(hsotg->dev, "DWC OTG HCD QH - Device Address = %d\n", - dwc2_hcd_get_dev_addr(&urb->pipe_info)); - dev_vdbg(hsotg->dev, "DWC OTG HCD QH - Endpoint %d, %s\n", - dwc2_hcd_get_ep_num(&urb->pipe_info), - dwc2_hcd_is_pipe_in(&urb->pipe_info) ? "IN" : "OUT"); + qh->device_interval = urb->interval; + qh->host_interval = urb->interval * (do_split ? 8 : 1); - qh->dev_speed = dev_speed; + /* + * Schedule low speed if we're running the host in low or + * full speed OR if we've got a "TT" to deal with to access this + * device. + */ + qh->schedule_low_speed = prtspd != HPRT0_SPD_HIGH_SPEED || + dwc_tt; + + if (do_split) { + /* We won't know num transfers until we schedule */ + qh->num_hs_transfers = -1; + } else if (dev_speed == USB_SPEED_HIGH) { + qh->num_hs_transfers = 1; + } else { + qh->num_hs_transfers = 0; + } + + /* We'll schedule later when we have something to do */ + } switch (dev_speed) { case USB_SPEED_LOW: @@ -703,7 +1535,6 @@ static void dwc2_qh_init(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh, speed = "?"; break; } - dev_vdbg(hsotg->dev, "DWC OTG HCD QH - Speed = %s\n", speed); switch (qh->ep_type) { case USB_ENDPOINT_XFER_ISOC: @@ -723,13 +1554,21 @@ static void dwc2_qh_init(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh, break; } - dev_vdbg(hsotg->dev, "DWC OTG HCD QH - Type = %s\n", type); - - if (qh->ep_type == USB_ENDPOINT_XFER_INT) { - dev_vdbg(hsotg->dev, "DWC OTG HCD QH - usecs = %d\n", - qh->host_us); - dev_vdbg(hsotg->dev, "DWC OTG HCD QH - interval = %d\n", - qh->host_interval); + dwc2_sch_dbg(hsotg, "QH=%p Init %s, %s speed, %d bytes:\n", qh, type, + speed, bytecount); + dwc2_sch_dbg(hsotg, "QH=%p ...addr=%d, ep=%d, %s\n", qh, + dwc2_hcd_get_dev_addr(&urb->pipe_info), + dwc2_hcd_get_ep_num(&urb->pipe_info), + ep_is_in ? "IN" : "OUT"); + if (ep_is_int || ep_is_isoc) { + dwc2_sch_dbg(hsotg, + "QH=%p ...duration: host=%d us, device=%d us\n", + qh, qh->host_us, qh->device_us); + dwc2_sch_dbg(hsotg, "QH=%p ...interval: host=%d, device=%d\n", + qh, qh->host_interval, qh->device_interval); + if (qh->schedule_low_speed) + dwc2_sch_dbg(hsotg, "QH=%p ...low speed schedule=%p\n", + qh, dwc2_get_ls_map(hsotg, qh)); } } @@ -757,7 +1596,7 @@ struct dwc2_qh *dwc2_hcd_qh_create(struct dwc2_hsotg *hsotg, if (!qh) return NULL; - dwc2_qh_init(hsotg, qh, urb); + dwc2_qh_init(hsotg, qh, urb, mem_flags); if (hsotg->core_params->dma_desc_enable > 0 && dwc2_hcd_qh_init_ddma(hsotg, qh, mem_flags) < 0) { @@ -789,6 +1628,7 @@ void dwc2_hcd_qh_free(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh) dwc2_do_unreserve(hsotg, qh); spin_unlock_irqrestore(&hsotg->lock, flags); } + dwc2_host_put_tt_info(hsotg, qh->dwc_tt); if (qh->desc_list) dwc2_hcd_qh_free_ddma(hsotg, qh); @@ -904,6 +1744,8 @@ static int dwc2_next_for_periodic_split(struct dwc2_hsotg *hsotg, u16 incr; /* + * See dwc2_uframe_schedule_split() for split scheduling. + * * Basically: increment 1 normally, but 2 right after the start split * (except for ISOC out). */ @@ -1006,9 +1848,17 @@ static int dwc2_next_periodic_start(struct dwc2_hsotg *hsotg, if (qh->start_active_frame == qh->next_active_frame || dwc2_frame_num_gt(prev_frame_number, qh->start_active_frame)) { u16 ideal_start = qh->start_active_frame; + int periods_in_map; - /* Adjust interval as per gcd with plan length. */ - interval = gcd(interval, 8); + /* + * Adjust interval as per gcd with map size. + * See pmap_schedule() for more details here. + */ + if (qh->do_split || qh->dev_speed == USB_SPEED_HIGH) + periods_in_map = DWC2_HS_SCHEDULE_UFRAMES; + else + periods_in_map = DWC2_LS_SCHEDULE_FRAMES; + interval = gcd(interval, periods_in_map); do { qh->start_active_frame = dwc2_frame_num_inc( |