xhci: Implement HS/FS/LS bandwidth checking.

Now that we have a bandwidth interval table per root port or TT that
describes the endpoint bandwidth information, we can finally use it to
check whether the bus bandwidth is oversubscribed for a new device
configuration/alternate interface setting.

The complication for this algorithm is that the bit of hardware logic that
creates the bus schedule is only 12-bit logic.  In order to make sure it
can represent the maximum bus bandwidth in 12 bits, it has to convert the
endpoint max packet size and max esit payload into "blocks" (basically a
less-precise representation).  The block size for each speed of device is
different, aside from low speed and full speed.  In order to make sure we
don't allow a setup where the scheduler might fail, we also have to do the
bandwidth checking in blocks.

After checking that the endpoints fit in the schedule, we store the
bandwidth used for this root port or TT.  If this is a FS/LS device under
an external HS hub, we also update the TT bandwidth and the root port
bandwidth (if this is a newly activated or deactivated TT).

I won't go into the details of the algorithm, as it's pretty well
documented in the comments.
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

drivers/usb/host/xhci.c

@@ -1747,13 +1747,275 @@ static void xhci_finish_resource_reservation(struct xhci_hcd *xhci,
 				xhci->num_active_eps);
 }
 
-/* Run the algorithm on the bandwidth table.  If this table is part of a
- * TT, see if we need to update the number of active TTs.
+unsigned int xhci_get_block_size(struct usb_device *udev)
+{
+	switch (udev->speed) {
+	case USB_SPEED_LOW:
+	case USB_SPEED_FULL:
+		return FS_BLOCK;
+	case USB_SPEED_HIGH:
+		return HS_BLOCK;
+	case USB_SPEED_SUPER:
+		return SS_BLOCK;
+	case USB_SPEED_UNKNOWN:
+	case USB_SPEED_WIRELESS:
+	default:
+		/* Should never happen */
+		return 1;
+	}
+}
+
+unsigned int xhci_get_largest_overhead(struct xhci_interval_bw *interval_bw)
+{
+	if (interval_bw->overhead[LS_OVERHEAD_TYPE])
+		return LS_OVERHEAD;
+	if (interval_bw->overhead[FS_OVERHEAD_TYPE])
+		return FS_OVERHEAD;
+	return HS_OVERHEAD;
+}
+
+/* If we are changing a LS/FS device under a HS hub,
+ * make sure (if we are activating a new TT) that the HS bus has enough
+ * bandwidth for this new TT.
+ */
+static int xhci_check_tt_bw_table(struct xhci_hcd *xhci,
+		struct xhci_virt_device *virt_dev,
+		int old_active_eps)
+{
+	struct xhci_interval_bw_table *bw_table;
+	struct xhci_tt_bw_info *tt_info;
+
+	/* Find the bandwidth table for the root port this TT is attached to. */
+	bw_table = &xhci->rh_bw[virt_dev->real_port - 1].bw_table;
+	tt_info = virt_dev->tt_info;
+	/* If this TT already had active endpoints, the bandwidth for this TT
+	 * has already been added.  Removing all periodic endpoints (and thus
+	 * making the TT enactive) will only decrease the bandwidth used.
+	 */
+	if (old_active_eps)
+		return 0;
+	if (old_active_eps == 0 && tt_info->active_eps != 0) {
+		if (bw_table->bw_used + TT_HS_OVERHEAD > HS_BW_LIMIT)
+			return -ENOMEM;
+		return 0;
+	}
+	/* Not sure why we would have no new active endpoints...
+	 *
+	 * Maybe because of an Evaluate Context change for a hub update or a
+	 * control endpoint 0 max packet size change?
+	 * FIXME: skip the bandwidth calculation in that case.
+	 */
+	return 0;
+}
+
+/*
+ * This algorithm is a very conservative estimate of the worst-case scheduling
+ * scenario for any one interval.  The hardware dynamically schedules the
+ * packets, so we can't tell which microframe could be the limiting factor in
+ * the bandwidth scheduling.  This only takes into account periodic endpoints.
+ *
+ * Obviously, we can't solve an NP complete problem to find the minimum worst
+ * case scenario.  Instead, we come up with an estimate that is no less than
+ * the worst case bandwidth used for any one microframe, but may be an
+ * over-estimate.
+ *
+ * We walk the requirements for each endpoint by interval, starting with the
+ * smallest interval, and place packets in the schedule where there is only one
+ * possible way to schedule packets for that interval.  In order to simplify
+ * this algorithm, we record the largest max packet size for each interval, and
+ * assume all packets will be that size.
+ *
+ * For interval 0, we obviously must schedule all packets for each interval.
+ * The bandwidth for interval 0 is just the amount of data to be transmitted
+ * (the sum of all max ESIT payload sizes, plus any overhead per packet times
+ * the number of packets).
+ *
+ * For interval 1, we have two possible microframes to schedule those packets
+ * in.  For this algorithm, if we can schedule the same number of packets for
+ * each possible scheduling opportunity (each microframe), we will do so.  The
+ * remaining number of packets will be saved to be transmitted in the gaps in
+ * the next interval's scheduling sequence.
+ *
+ * As we move those remaining packets to be scheduled with interval 2 packets,
+ * we have to double the number of remaining packets to transmit.  This is
+ * because the intervals are actually powers of 2, and we would be transmitting
+ * the previous interval's packets twice in this interval.  We also have to be
+ * sure that when we look at the largest max packet size for this interval, we
+ * also look at the largest max packet size for the remaining packets and take
+ * the greater of the two.
+ *
+ * The algorithm continues to evenly distribute packets in each scheduling
+ * opportunity, and push the remaining packets out, until we get to the last
+ * interval.  Then those packets and their associated overhead are just added
+ * to the bandwidth used.
  */
 static int xhci_check_bw_table(struct xhci_hcd *xhci,
 		struct xhci_virt_device *virt_dev,
 		int old_active_eps)
 {
+	unsigned int bw_reserved;
+	unsigned int max_bandwidth;
+	unsigned int bw_used;
+	unsigned int block_size;
+	struct xhci_interval_bw_table *bw_table;
+	unsigned int packet_size = 0;
+	unsigned int overhead = 0;
+	unsigned int packets_transmitted = 0;
+	unsigned int packets_remaining = 0;
+	unsigned int i;
+
+	if (virt_dev->udev->speed == USB_SPEED_HIGH) {
+		max_bandwidth = HS_BW_LIMIT;
+		/* Convert percent of bus BW reserved to blocks reserved */
+		bw_reserved = DIV_ROUND_UP(HS_BW_RESERVED * max_bandwidth, 100);
+	} else {
+		max_bandwidth = FS_BW_LIMIT;
+		bw_reserved = DIV_ROUND_UP(FS_BW_RESERVED * max_bandwidth, 100);
+	}
+
+	bw_table = virt_dev->bw_table;
+	/* We need to translate the max packet size and max ESIT payloads into
+	 * the units the hardware uses.
+	 */
+	block_size = xhci_get_block_size(virt_dev->udev);
+
+	/* If we are manipulating a LS/FS device under a HS hub, double check
+	 * that the HS bus has enough bandwidth if we are activing a new TT.
+	 */
+	if (virt_dev->tt_info) {
+		xhci_dbg(xhci, "Recalculating BW for rootport %u\n",
+				virt_dev->real_port);
+		if (xhci_check_tt_bw_table(xhci, virt_dev, old_active_eps)) {
+			xhci_warn(xhci, "Not enough bandwidth on HS bus for "
+					"newly activated TT.\n");
+			return -ENOMEM;
+		}
+		xhci_dbg(xhci, "Recalculating BW for TT slot %u port %u\n",
+				virt_dev->tt_info->slot_id,
+				virt_dev->tt_info->ttport);
+	} else {
+		xhci_dbg(xhci, "Recalculating BW for rootport %u\n",
+				virt_dev->real_port);
+	}
+
+	/* Add in how much bandwidth will be used for interval zero, or the
+	 * rounded max ESIT payload + number of packets * largest overhead.
+	 */
+	bw_used = DIV_ROUND_UP(bw_table->interval0_esit_payload, block_size) +
+		bw_table->interval_bw[0].num_packets *
+		xhci_get_largest_overhead(&bw_table->interval_bw[0]);
+
+	for (i = 1; i < XHCI_MAX_INTERVAL; i++) {
+		unsigned int bw_added;
+		unsigned int largest_mps;
+		unsigned int interval_overhead;
+
+		/*
+		 * How many packets could we transmit in this interval?
+		 * If packets didn't fit in the previous interval, we will need
+		 * to transmit that many packets twice within this interval.
+		 */
+		packets_remaining = 2 * packets_remaining +
+			bw_table->interval_bw[i].num_packets;
+
+		/* Find the largest max packet size of this or the previous
+		 * interval.
+		 */
+		if (list_empty(&bw_table->interval_bw[i].endpoints))
+			largest_mps = 0;
+		else {
+			struct xhci_virt_ep *virt_ep;
+			struct list_head *ep_entry;
+
+			ep_entry = bw_table->interval_bw[i].endpoints.next;
+			virt_ep = list_entry(ep_entry,
+					struct xhci_virt_ep, bw_endpoint_list);
+			/* Convert to blocks, rounding up */
+			largest_mps = DIV_ROUND_UP(
+					virt_ep->bw_info.max_packet_size,
+					block_size);
+		}
+		if (largest_mps > packet_size)
+			packet_size = largest_mps;
+
+		/* Use the larger overhead of this or the previous interval. */
+		interval_overhead = xhci_get_largest_overhead(
+				&bw_table->interval_bw[i]);
+		if (interval_overhead > overhead)
+			overhead = interval_overhead;
+
+		/* How many packets can we evenly distribute across
+		 * (1 << (i + 1)) possible scheduling opportunities?
+		 */
+		packets_transmitted = packets_remaining >> (i + 1);
+
+		/* Add in the bandwidth used for those scheduled packets */
+		bw_added = packets_transmitted * (overhead + packet_size);
+
+		/* How many packets do we have remaining to transmit? */
+		packets_remaining = packets_remaining % (1 << (i + 1));
+
+		/* What largest max packet size should those packets have? */
+		/* If we've transmitted all packets, don't carry over the
+		 * largest packet size.
+		 */
+		if (packets_remaining == 0) {
+			packet_size = 0;
+			overhead = 0;
+		} else if (packets_transmitted > 0) {
+			/* Otherwise if we do have remaining packets, and we've
+			 * scheduled some packets in this interval, take the
+			 * largest max packet size from endpoints with this
+			 * interval.
+			 */
+			packet_size = largest_mps;
+			overhead = interval_overhead;
+		}
+		/* Otherwise carry over packet_size and overhead from the last
+		 * time we had a remainder.
+		 */
+		bw_used += bw_added;
+		if (bw_used > max_bandwidth) {
+			xhci_warn(xhci, "Not enough bandwidth. "
+					"Proposed: %u, Max: %u\n",
+				bw_used, max_bandwidth);
+			return -ENOMEM;
+		}
+	}
+	/*
+	 * Ok, we know we have some packets left over after even-handedly
+	 * scheduling interval 15.  We don't know which microframes they will
+	 * fit into, so we over-schedule and say they will be scheduled every
+	 * microframe.
+	 */
+	if (packets_remaining > 0)
+		bw_used += overhead + packet_size;
+
+	if (!virt_dev->tt_info && virt_dev->udev->speed == USB_SPEED_HIGH) {
+		unsigned int port_index = virt_dev->real_port - 1;
+
+		/* OK, we're manipulating a HS device attached to a
+		 * root port bandwidth domain.  Include the number of active TTs
+		 * in the bandwidth used.
+		 */
+		bw_used += TT_HS_OVERHEAD *
+			xhci->rh_bw[port_index].num_active_tts;
+	}
+
+	xhci_dbg(xhci, "Final bandwidth: %u, Limit: %u, Reserved: %u, "
+		"Available: %u " "percent\n",
+		bw_used, max_bandwidth, bw_reserved,
+		(max_bandwidth - bw_used - bw_reserved) * 100 /
+		max_bandwidth);
+
+	bw_used += bw_reserved;
+	if (bw_used > max_bandwidth) {
+		xhci_warn(xhci, "Not enough bandwidth. Proposed: %u, Max: %u\n",
+				bw_used, max_bandwidth);
+		return -ENOMEM;
+	}
+
+	bw_table->bw_used = bw_used;
 	return 0;
 }
 
@@ -1888,9 +2150,11 @@ void xhci_update_tt_active_eps(struct xhci_hcd *xhci,
 	if (old_active_eps == 0 &&
 				virt_dev->tt_info->active_eps != 0) {
 		rh_bw_info->num_active_tts += 1;
+		rh_bw_info->bw_table.bw_used += TT_HS_OVERHEAD;
 	} else if (old_active_eps != 0 &&
 				virt_dev->tt_info->active_eps == 0) {
 		rh_bw_info->num_active_tts -= 1;
+		rh_bw_info->bw_table.bw_used -= TT_HS_OVERHEAD;
 	}
 }
 

drivers/usb/host/xhci.h

@@ -756,6 +756,49 @@ struct xhci_bw_info {
 	unsigned int		type;
 };
 
+/* "Block" sizes in bytes the hardware uses for different device speeds.
+ * The logic in this part of the hardware limits the number of bits the hardware
+ * can use, so must represent bandwidth in a less precise manner to mimic what
+ * the scheduler hardware computes.
+ */
+#define	FS_BLOCK	1
+#define	HS_BLOCK	4
+#define	SS_BLOCK	16
+#define	DMI_BLOCK	32
+
+/* Each device speed has a protocol overhead (CRC, bit stuffing, etc) associated
+ * with each byte transferred.  SuperSpeed devices have an initial overhead to
+ * set up bursts.  These are in blocks, see above.  LS overhead has already been
+ * translated into FS blocks.
+ */
+#define DMI_OVERHEAD 8
+#define DMI_OVERHEAD_BURST 4
+#define SS_OVERHEAD 8
+#define SS_OVERHEAD_BURST 32
+#define HS_OVERHEAD 26
+#define FS_OVERHEAD 20
+#define LS_OVERHEAD 128
+/* The TTs need to claim roughly twice as much bandwidth (94 bytes per
+ * microframe ~= 24Mbps) of the HS bus as the devices can actually use because
+ * of overhead associated with split transfers crossing microframe boundaries.
+ * 31 blocks is pure protocol overhead.
+ */
+#define TT_HS_OVERHEAD (31 + 94)
+#define TT_DMI_OVERHEAD (25 + 12)
+
+/* Bandwidth limits in blocks */
+#define FS_BW_LIMIT		1285
+#define TT_BW_LIMIT		1320
+#define HS_BW_LIMIT		1607
+#define SS_BW_LIMIT_IN		3906
+#define DMI_BW_LIMIT_IN		3906
+#define SS_BW_LIMIT_OUT		3906
+#define DMI_BW_LIMIT_OUT	3906
+
+/* Percentage of bus bandwidth reserved for non-periodic transfers */
+#define FS_BW_RESERVED		10
+#define HS_BW_RESERVED		20
+
 struct xhci_virt_ep {
 	struct xhci_ring		*ring;
 	/* Related to endpoints that are configured to use stream IDs only */
@@ -823,6 +866,8 @@ struct xhci_interval_bw {
 struct xhci_interval_bw_table {
 	unsigned int		interval0_esit_payload;
 	struct xhci_interval_bw	interval_bw[XHCI_MAX_INTERVAL];
+	/* Includes reserved bandwidth for async endpoints */
+	unsigned int		bw_used;
 };
 
 
@@ -1397,6 +1442,7 @@ struct xhci_hcd {
 #define XHCI_EP_LIMIT_QUIRK	(1 << 5)
 #define XHCI_BROKEN_MSI		(1 << 6)
 #define XHCI_RESET_ON_RESUME	(1 << 7)
+#define	XHCI_SW_BW_CHECKING	(1 << 8)
 	unsigned int		num_active_eps;
 	unsigned int		limit_active_eps;
 	/* There are two roothubs to keep track of bus suspend info for */