[PATCH 00/10] thunderbolt: DisplayPort and bandwidth allocation mode improvements

[PATCH 00/10] thunderbolt: DisplayPort and bandwidth allocation mode improvements

[Mika Westerberg]

Hi all,

This series improves bandwidth allocation mode support to follow the
USB4 spec more closely. In addition to that DisplayPort tunneling is
made more robust so that if a tunnel does not come up properly (for
example on a discrete controller the DisplayPort source from the
graphics is not connected to the DisplayPort sink of the controller) we
tear it down and continue pairing.

This series improves bandwidth allocation mode support and DisplayPort
tunneling to follow the USB4 spec more closely.

Gil Fine (2):
  thunderbolt: Calculate DisplayPort tunnel bandwidth after DPRX capabilities read
  thunderbolt: Improve DisplayPort tunnel setup process to be more robust

Mika Westerberg (8):
  thunderbolt: Use DP_LOCAL_CAP for maximum bandwidth calculation
  thunderbolt: Re-calculate estimated bandwidth when allocation mode is enabled
  thunderbolt: Handle bandwidth allocation mode disable request
  thunderbolt: Log an error if DPTX request is not cleared
  thunderbolt: Fail the failed bandwidth request properly
  thunderbolt: Re-order bandwidth group functions
  thunderbolt: Introduce tb_tunnel_direction_downstream()
  thunderbolt: Reserve released DisplayPort bandwidth for a group for 10 seconds

 drivers/thunderbolt/domain.c |   4 +
 drivers/thunderbolt/tb.c     | 898 +++++++++++++++++++++--------------
 drivers/thunderbolt/tb.h     |  10 +
 drivers/thunderbolt/tunnel.c |  96 ++--
 drivers/thunderbolt/tunnel.h |   6 +
 drivers/thunderbolt/usb4.c   |   4 +-
 6 files changed, 596 insertions(+), 422 deletions(-)

-- 
2.43.0

[PATCH 01/10] thunderbolt: Use DP_LOCAL_CAP for maximum bandwidth calculation

[Mika Westerberg]

The DisplayPort IN adapter DP_LOCAL_CAP holds the aggregated
capabilities and gets updated after graphics side does the DPRX
capabilities read so we should use this to figure out the maximum
possible bandwidth for the DisplayPort tunnel.

While there make the variable name to match better what it is used for
and add kernel-doc comment to the function.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
---
 drivers/thunderbolt/tunnel.c | 57 ++++++++++++++++--------------------
 1 file changed, 25 insertions(+), 32 deletions(-)

diff --git a/drivers/thunderbolt/tunnel.c b/drivers/thunderbolt/tunnel.c
index 6fffb2c82d3d..a766ab297064 100644
--- a/drivers/thunderbolt/tunnel.c
+++ b/drivers/thunderbolt/tunnel.c
@@ -926,12 +926,18 @@ static int tb_dp_activate(struct tb_tunnel *tunnel, bool active)
 	return 0;
 }
 
-/* max_bw is rounded up to next granularity */
+/**
+ * tb_dp_bandwidth_mode_maximum_bandwidth() - Maximum possible bandwidth
+ * @tunnel: DP tunnel to check
+ * @max_bw_rounded: Maximum bandwidth in Mb/s rounded up to the next granularity
+ *
+ * Returns maximum possible bandwidth for this tunnel in Mb/s.
+ */
 static int tb_dp_bandwidth_mode_maximum_bandwidth(struct tb_tunnel *tunnel,
-						  int *max_bw)
+						  int *max_bw_rounded)
 {
 	struct tb_port *in = tunnel->src_port;
-	int ret, rate, lanes, nrd_bw;
+	int ret, rate, lanes, max_bw;
 	u32 cap;
 
 	/*
@@ -947,32 +953,18 @@ static int tb_dp_bandwidth_mode_maximum_bandwidth(struct tb_tunnel *tunnel,
 		return ret;
 
 	rate = tb_dp_cap_get_rate_ext(cap);
-	if (tb_dp_is_uhbr_rate(rate)) {
-		/*
-		 * When UHBR is used there is no reduction in lanes so
-		 * we can use this directly.
-		 */
-		lanes = tb_dp_cap_get_lanes(cap);
-	} else {
-		/*
-		 * If there is no UHBR supported then check the
-		 * non-reduced rate and lanes.
-		 */
-		ret = usb4_dp_port_nrd(in, &rate, &lanes);
-		if (ret)
-			return ret;
-	}
+	lanes = tb_dp_cap_get_lanes(cap);
 
-	nrd_bw = tb_dp_bandwidth(rate, lanes);
+	max_bw = tb_dp_bandwidth(rate, lanes);
 
-	if (max_bw) {
+	if (max_bw_rounded) {
 		ret = usb4_dp_port_granularity(in);
 		if (ret < 0)
 			return ret;
-		*max_bw = roundup(nrd_bw, ret);
+		*max_bw_rounded = roundup(max_bw, ret);
 	}
 
-	return nrd_bw;
+	return max_bw;
 }
 
 static int tb_dp_bandwidth_mode_consumed_bandwidth(struct tb_tunnel *tunnel,
@@ -981,7 +973,7 @@ static int tb_dp_bandwidth_mode_consumed_bandwidth(struct tb_tunnel *tunnel,
 {
 	struct tb_port *out = tunnel->dst_port;
 	struct tb_port *in = tunnel->src_port;
-	int ret, allocated_bw, max_bw;
+	int ret, allocated_bw, max_bw_rounded;
 
 	if (!usb4_dp_port_bandwidth_mode_enabled(in))
 		return -EOPNOTSUPP;
@@ -995,10 +987,10 @@ static int tb_dp_bandwidth_mode_consumed_bandwidth(struct tb_tunnel *tunnel,
 		return ret;
 	allocated_bw = ret;
 
-	ret = tb_dp_bandwidth_mode_maximum_bandwidth(tunnel, &max_bw);
+	ret = tb_dp_bandwidth_mode_maximum_bandwidth(tunnel, &max_bw_rounded);
 	if (ret < 0)
 		return ret;
-	if (allocated_bw == max_bw)
+	if (allocated_bw == max_bw_rounded)
 		allocated_bw = ret;
 
 	if (tb_port_path_direction_downstream(in, out)) {
@@ -1023,17 +1015,18 @@ static int tb_dp_allocated_bandwidth(struct tb_tunnel *tunnel, int *allocated_up
 	 * Otherwise we read it from the DPRX.
 	 */
 	if (usb4_dp_port_bandwidth_mode_enabled(in) && tunnel->bw_mode) {
-		int ret, allocated_bw, max_bw;
+		int ret, allocated_bw, max_bw_rounded;
 
 		ret = usb4_dp_port_allocated_bandwidth(in);
 		if (ret < 0)
 			return ret;
 		allocated_bw = ret;
 
-		ret = tb_dp_bandwidth_mode_maximum_bandwidth(tunnel, &max_bw);
+		ret = tb_dp_bandwidth_mode_maximum_bandwidth(tunnel,
+							     &max_bw_rounded);
 		if (ret < 0)
 			return ret;
-		if (allocated_bw == max_bw)
+		if (allocated_bw == max_bw_rounded)
 			allocated_bw = ret;
 
 		if (tb_port_path_direction_downstream(in, out)) {
@@ -1055,24 +1048,24 @@ static int tb_dp_alloc_bandwidth(struct tb_tunnel *tunnel, int *alloc_up,
 {
 	struct tb_port *out = tunnel->dst_port;
 	struct tb_port *in = tunnel->src_port;
-	int max_bw, ret, tmp;
+	int max_bw_rounded, ret, tmp;
 
 	if (!usb4_dp_port_bandwidth_mode_enabled(in))
 		return -EOPNOTSUPP;
 
-	ret = tb_dp_bandwidth_mode_maximum_bandwidth(tunnel, &max_bw);
+	ret = tb_dp_bandwidth_mode_maximum_bandwidth(tunnel, &max_bw_rounded);
 	if (ret < 0)
 		return ret;
 
 	if (tb_port_path_direction_downstream(in, out)) {
-		tmp = min(*alloc_down, max_bw);
+		tmp = min(*alloc_down, max_bw_rounded);
 		ret = usb4_dp_port_allocate_bandwidth(in, tmp);
 		if (ret)
 			return ret;
 		*alloc_down = tmp;
 		*alloc_up = 0;
 	} else {
-		tmp = min(*alloc_up, max_bw);
+		tmp = min(*alloc_up, max_bw_rounded);
 		ret = usb4_dp_port_allocate_bandwidth(in, tmp);
 		if (ret)
 			return ret;
-- 
2.43.0

[PATCH 02/10] thunderbolt: Re-calculate estimated bandwidth when allocation mode is enabled

[Mika Westerberg]

When we program the initial bandwidth estimation the DPTX (graphics
driver) has not yet read the capabilities of the monitor so the values
used are the highest possible of the involved DisplayPort IN and OUT
adapters, not the actual monitor capabilities. To allow the graphics
more accurate bandwidth estimation re-calculate it once we receive the
bandwidth allocation mode enabled notification.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
---
 drivers/thunderbolt/tb.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)

diff --git a/drivers/thunderbolt/tb.c b/drivers/thunderbolt/tb.c
index 64dd22e1f5b2..5b0434c140f9 100644
--- a/drivers/thunderbolt/tb.c
+++ b/drivers/thunderbolt/tb.c
@@ -2413,10 +2413,19 @@ static void tb_handle_dp_bandwidth_request(struct work_struct *work)
 
 	ret = usb4_dp_port_requested_bandwidth(in);
 	if (ret < 0) {
-		if (ret == -ENODATA)
-			tb_port_dbg(in, "no bandwidth request active\n");
-		else
+		if (ret == -ENODATA) {
+			/*
+			 * There is no request active so this means the
+			 * BW allocation mode was enabled from graphics
+			 * side. At this point we know that the graphics
+			 * driver has read the DRPX capabilities so we
+			 * can offer an better bandwidth estimatation.
+			 */
+			tb_port_dbg(in, "DPTX enabled bandwidth allocation mode, updating estimated bandwidth\n");
+			tb_recalc_estimated_bandwidth(tb);
+		} else {
 			tb_port_warn(in, "failed to read requested bandwidth\n");
+		}
 		goto put_sw;
 	}
 	requested_bw = ret;
-- 
2.43.0

[PATCH 03/10] thunderbolt: Handle bandwidth allocation mode disable request

[Mika Westerberg]

Graphics can disable DisplayPort bandwidth allocation mode as well so if
this make sure to reset the tunnel state accordingly.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
---
 drivers/thunderbolt/tb.c | 22 ++++++++++++++++------
 1 file changed, 16 insertions(+), 6 deletions(-)

diff --git a/drivers/thunderbolt/tb.c b/drivers/thunderbolt/tb.c
index 5b0434c140f9..abd86fd8d71f 100644
--- a/drivers/thunderbolt/tb.c
+++ b/drivers/thunderbolt/tb.c
@@ -2406,8 +2406,23 @@ static void tb_handle_dp_bandwidth_request(struct work_struct *work)
 
 	tb_port_dbg(in, "handling bandwidth allocation request\n");
 
+	tunnel = tb_find_tunnel(tb, TB_TUNNEL_DP, in, NULL);
+	if (!tunnel) {
+		tb_port_warn(in, "failed to find tunnel\n");
+		goto put_sw;
+	}
+
 	if (!usb4_dp_port_bandwidth_mode_enabled(in)) {
-		tb_port_warn(in, "bandwidth allocation mode not enabled\n");
+		if (tunnel->bw_mode) {
+			/*
+			 * Reset the tunnel back to use the legacy
+			 * allocation.
+			 */
+			tunnel->bw_mode = false;
+			tb_port_dbg(in, "DPTX disabled bandwidth allocation mode\n");
+		} else {
+			tb_port_warn(in, "bandwidth allocation mode not enabled\n");
+		}
 		goto put_sw;
 	}
 
@@ -2432,11 +2447,6 @@ static void tb_handle_dp_bandwidth_request(struct work_struct *work)
 
 	tb_port_dbg(in, "requested bandwidth %d Mb/s\n", requested_bw);
 
-	tunnel = tb_find_tunnel(tb, TB_TUNNEL_DP, in, NULL);
-	if (!tunnel) {
-		tb_port_warn(in, "failed to find tunnel\n");
-		goto put_sw;
-	}
 
 	out = tunnel->dst_port;
 
-- 
2.43.0

[PATCH 04/10] thunderbolt: Log an error if DPTX request is not cleared

[Mika Westerberg]

This helps debugging issues around DisplayPort bandwidth allocation
mode.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
---
 drivers/thunderbolt/usb4.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/thunderbolt/usb4.c b/drivers/thunderbolt/usb4.c
index 4b35898aa216..f4fba144105d 100644
--- a/drivers/thunderbolt/usb4.c
+++ b/drivers/thunderbolt/usb4.c
@@ -2858,8 +2858,10 @@ static int usb4_dp_port_wait_and_clear_cm_ack(struct tb_port *port,
 		usleep_range(50, 100);
 	} while (ktime_before(ktime_get(), end));
 
-	if (val & ADP_DP_CS_8_DR)
+	if (val & ADP_DP_CS_8_DR) {
+		tb_port_warn(port, "timeout waiting for DPTX request to clear\n");
 		return -ETIMEDOUT;
+	}
 
 	ret = tb_port_read(port, &val, TB_CFG_PORT,
 			   port->cap_adap + ADP_DP_CS_2, 1);
-- 
2.43.0

[PATCH 05/10] thunderbolt: Fail the failed bandwidth request properly

[Mika Westerberg]

The USB4 spec says that if the Connection Manager writes Allocated_BW
that is smaller than Requested_BW, the DisplayPort IN adapter signals
this failure back to the DPTX (graphics driver). Implement this by
rewriting the same allocated bandwidth values back.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
---
 drivers/thunderbolt/tb.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)

diff --git a/drivers/thunderbolt/tb.c b/drivers/thunderbolt/tb.c
index abd86fd8d71f..9dbdf2770f0b 100644
--- a/drivers/thunderbolt/tb.c
+++ b/drivers/thunderbolt/tb.c
@@ -2270,11 +2270,11 @@ static int tb_alloc_dp_bandwidth(struct tb_tunnel *tunnel, int *requested_up,
 	 */
 	ret = tb_tunnel_maximum_bandwidth(tunnel, &max_up, &max_down);
 	if (ret)
-		return ret;
+		goto fail;
 
 	ret = usb4_dp_port_granularity(in);
 	if (ret < 0)
-		return ret;
+		goto fail;
 	granularity = ret;
 
 	max_up_rounded = roundup(max_up, granularity);
@@ -2304,7 +2304,8 @@ static int tb_alloc_dp_bandwidth(struct tb_tunnel *tunnel, int *requested_up,
 			      "bandwidth request too high (%d/%d Mb/s > %d/%d Mb/s)\n",
 			      requested_up_corrected, requested_down_corrected,
 			      max_up_rounded, max_down_rounded);
-		return -ENOBUFS;
+		ret = -ENOBUFS;
+		goto fail;
 	}
 
 	if ((*requested_up >= 0 && requested_up_corrected <= allocated_up) ||
@@ -2332,7 +2333,7 @@ static int tb_alloc_dp_bandwidth(struct tb_tunnel *tunnel, int *requested_up,
 	 */
 	ret = tb_release_unused_usb3_bandwidth(tb, in, out);
 	if (ret)
-		return ret;
+		goto fail;
 
 	/*
 	 * Then go over all tunnels that cross the same USB4 ports (they
@@ -2357,7 +2358,7 @@ static int tb_alloc_dp_bandwidth(struct tb_tunnel *tunnel, int *requested_up,
 					*requested_down);
 		if (ret) {
 			tb_configure_sym(tb, in, out, 0, 0, true);
-			return ret;
+			goto fail;
 		}
 
 		ret = tb_tunnel_alloc_bandwidth(tunnel, requested_up,
@@ -2372,6 +2373,18 @@ static int tb_alloc_dp_bandwidth(struct tb_tunnel *tunnel, int *requested_up,
 
 reclaim:
 	tb_reclaim_usb3_bandwidth(tb, in, out);
+fail:
+	if (ret && ret != -ENODEV) {
+		/*
+		 * Write back the same allocated (so no change), this
+		 * makes the DPTX request fail on graphics side.
+		 */
+		tb_tunnel_dbg(tunnel,
+			      "failing the request by rewriting allocated %d/%d Mb/s\n",
+			      allocated_up, allocated_down);
+		tb_tunnel_alloc_bandwidth(tunnel, &allocated_up, &allocated_down);
+	}
+
 	return ret;
 }
 
-- 
2.43.0

[PATCH 06/10] thunderbolt: Re-order bandwidth group functions

[Mika Westerberg]

This is needed by the following patches so that we do not have to add
forward declaratations for any of these. Separating the move and the
actual changes also makes it easier to review the code.

No functional changes.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
---
 drivers/thunderbolt/tb.c | 454 +++++++++++++++++++--------------------
 1 file changed, 225 insertions(+), 229 deletions(-)

diff --git a/drivers/thunderbolt/tb.c b/drivers/thunderbolt/tb.c
index 9dbdf2770f0b..d23a80339a8d 100644
--- a/drivers/thunderbolt/tb.c
+++ b/drivers/thunderbolt/tb.c
@@ -75,112 +75,6 @@ struct tb_hotplug_event {
 	bool unplug;
 };
 
-static void tb_init_bandwidth_groups(struct tb_cm *tcm)
-{
-	int i;
-
-	for (i = 0; i < ARRAY_SIZE(tcm->groups); i++) {
-		struct tb_bandwidth_group *group = &tcm->groups[i];
-
-		group->tb = tcm_to_tb(tcm);
-		group->index = i + 1;
-		INIT_LIST_HEAD(&group->ports);
-	}
-}
-
-static void tb_bandwidth_group_attach_port(struct tb_bandwidth_group *group,
-					   struct tb_port *in)
-{
-	if (!group || WARN_ON(in->group))
-		return;
-
-	in->group = group;
-	list_add_tail(&in->group_list, &group->ports);
-
-	tb_port_dbg(in, "attached to bandwidth group %d\n", group->index);
-}
-
-static struct tb_bandwidth_group *tb_find_free_bandwidth_group(struct tb_cm *tcm)
-{
-	int i;
-
-	for (i = 0; i < ARRAY_SIZE(tcm->groups); i++) {
-		struct tb_bandwidth_group *group = &tcm->groups[i];
-
-		if (list_empty(&group->ports))
-			return group;
-	}
-
-	return NULL;
-}
-
-static struct tb_bandwidth_group *
-tb_attach_bandwidth_group(struct tb_cm *tcm, struct tb_port *in,
-			  struct tb_port *out)
-{
-	struct tb_bandwidth_group *group;
-	struct tb_tunnel *tunnel;
-
-	/*
-	 * Find all DP tunnels that go through all the same USB4 links
-	 * as this one. Because we always setup tunnels the same way we
-	 * can just check for the routers at both ends of the tunnels
-	 * and if they are the same we have a match.
-	 */
-	list_for_each_entry(tunnel, &tcm->tunnel_list, list) {
-		if (!tb_tunnel_is_dp(tunnel))
-			continue;
-
-		if (tunnel->src_port->sw == in->sw &&
-		    tunnel->dst_port->sw == out->sw) {
-			group = tunnel->src_port->group;
-			if (group) {
-				tb_bandwidth_group_attach_port(group, in);
-				return group;
-			}
-		}
-	}
-
-	/* Pick up next available group then */
-	group = tb_find_free_bandwidth_group(tcm);
-	if (group)
-		tb_bandwidth_group_attach_port(group, in);
-	else
-		tb_port_warn(in, "no available bandwidth groups\n");
-
-	return group;
-}
-
-static void tb_discover_bandwidth_group(struct tb_cm *tcm, struct tb_port *in,
-					struct tb_port *out)
-{
-	if (usb4_dp_port_bandwidth_mode_enabled(in)) {
-		int index, i;
-
-		index = usb4_dp_port_group_id(in);
-		for (i = 0; i < ARRAY_SIZE(tcm->groups); i++) {
-			if (tcm->groups[i].index == index) {
-				tb_bandwidth_group_attach_port(&tcm->groups[i], in);
-				return;
-			}
-		}
-	}
-
-	tb_attach_bandwidth_group(tcm, in, out);
-}
-
-static void tb_detach_bandwidth_group(struct tb_port *in)
-{
-	struct tb_bandwidth_group *group = in->group;
-
-	if (group) {
-		in->group = NULL;
-		list_del_init(&in->group_list);
-
-		tb_port_dbg(in, "detached from bandwidth group %d\n", group->index);
-	}
-}
-
 static void tb_handle_hotplug(struct work_struct *work);
 
 static void tb_queue_hotplug(struct tb *tb, u64 route, u8 port, bool unplug)
@@ -472,34 +366,6 @@ static void tb_switch_discover_tunnels(struct tb_switch *sw,
 	}
 }
 
-static void tb_discover_tunnels(struct tb *tb)
-{
-	struct tb_cm *tcm = tb_priv(tb);
-	struct tb_tunnel *tunnel;
-
-	tb_switch_discover_tunnels(tb->root_switch, &tcm->tunnel_list, true);
-
-	list_for_each_entry(tunnel, &tcm->tunnel_list, list) {
-		if (tb_tunnel_is_pci(tunnel)) {
-			struct tb_switch *parent = tunnel->dst_port->sw;
-
-			while (parent != tunnel->src_port->sw) {
-				parent->boot = true;
-				parent = tb_switch_parent(parent);
-			}
-		} else if (tb_tunnel_is_dp(tunnel)) {
-			struct tb_port *in = tunnel->src_port;
-			struct tb_port *out = tunnel->dst_port;
-
-			/* Keep the domain from powering down */
-			pm_runtime_get_sync(&in->sw->dev);
-			pm_runtime_get_sync(&out->sw->dev);
-
-			tb_discover_bandwidth_group(tcm, in, out);
-		}
-	}
-}
-
 static int tb_port_configure_xdomain(struct tb_port *port, struct tb_xdomain *xd)
 {
 	if (tb_switch_is_usb4(port->sw))
@@ -1464,6 +1330,231 @@ static void tb_scan_port(struct tb_port *port)
 	}
 }
 
+static void
+tb_recalc_estimated_bandwidth_for_group(struct tb_bandwidth_group *group)
+{
+	struct tb_tunnel *first_tunnel;
+	struct tb *tb = group->tb;
+	struct tb_port *in;
+	int ret;
+
+	tb_dbg(tb, "re-calculating bandwidth estimation for group %u\n",
+	       group->index);
+
+	first_tunnel = NULL;
+	list_for_each_entry(in, &group->ports, group_list) {
+		int estimated_bw, estimated_up, estimated_down;
+		struct tb_tunnel *tunnel;
+		struct tb_port *out;
+
+		if (!usb4_dp_port_bandwidth_mode_enabled(in))
+			continue;
+
+		tunnel = tb_find_tunnel(tb, TB_TUNNEL_DP, in, NULL);
+		if (WARN_ON(!tunnel))
+			break;
+
+		if (!first_tunnel) {
+			/*
+			 * Since USB3 bandwidth is shared by all DP
+			 * tunnels under the host router USB4 port, even
+			 * if they do not begin from the host router, we
+			 * can release USB3 bandwidth just once and not
+			 * for each tunnel separately.
+			 */
+			first_tunnel = tunnel;
+			ret = tb_release_unused_usb3_bandwidth(tb,
+				first_tunnel->src_port, first_tunnel->dst_port);
+			if (ret) {
+				tb_tunnel_warn(tunnel,
+					"failed to release unused bandwidth\n");
+				break;
+			}
+		}
+
+		out = tunnel->dst_port;
+		ret = tb_available_bandwidth(tb, in, out, &estimated_up,
+					     &estimated_down, true);
+		if (ret) {
+			tb_tunnel_warn(tunnel,
+				"failed to re-calculate estimated bandwidth\n");
+			break;
+		}
+
+		/*
+		 * Estimated bandwidth includes:
+		 *  - already allocated bandwidth for the DP tunnel
+		 *  - available bandwidth along the path
+		 *  - bandwidth allocated for USB 3.x but not used.
+		 */
+		if (tb_port_path_direction_downstream(in, out))
+			estimated_bw = estimated_down;
+		else
+			estimated_bw = estimated_up;
+
+		if (usb4_dp_port_set_estimated_bandwidth(in, estimated_bw))
+			tb_tunnel_warn(tunnel,
+				       "failed to update estimated bandwidth\n");
+	}
+
+	if (first_tunnel)
+		tb_reclaim_usb3_bandwidth(tb, first_tunnel->src_port,
+					  first_tunnel->dst_port);
+
+	tb_dbg(tb, "bandwidth estimation for group %u done\n", group->index);
+}
+
+static void tb_recalc_estimated_bandwidth(struct tb *tb)
+{
+	struct tb_cm *tcm = tb_priv(tb);
+	int i;
+
+	tb_dbg(tb, "bandwidth consumption changed, re-calculating estimated bandwidth\n");
+
+	for (i = 0; i < ARRAY_SIZE(tcm->groups); i++) {
+		struct tb_bandwidth_group *group = &tcm->groups[i];
+
+		if (!list_empty(&group->ports))
+			tb_recalc_estimated_bandwidth_for_group(group);
+	}
+
+	tb_dbg(tb, "bandwidth re-calculation done\n");
+}
+
+static void tb_init_bandwidth_groups(struct tb_cm *tcm)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(tcm->groups); i++) {
+		struct tb_bandwidth_group *group = &tcm->groups[i];
+
+		group->tb = tcm_to_tb(tcm);
+		group->index = i + 1;
+		INIT_LIST_HEAD(&group->ports);
+	}
+}
+
+static void tb_bandwidth_group_attach_port(struct tb_bandwidth_group *group,
+					   struct tb_port *in)
+{
+	if (!group || WARN_ON(in->group))
+		return;
+
+	in->group = group;
+	list_add_tail(&in->group_list, &group->ports);
+
+	tb_port_dbg(in, "attached to bandwidth group %d\n", group->index);
+}
+
+static struct tb_bandwidth_group *tb_find_free_bandwidth_group(struct tb_cm *tcm)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(tcm->groups); i++) {
+		struct tb_bandwidth_group *group = &tcm->groups[i];
+
+		if (list_empty(&group->ports))
+			return group;
+	}
+
+	return NULL;
+}
+
+static struct tb_bandwidth_group *
+tb_attach_bandwidth_group(struct tb_cm *tcm, struct tb_port *in,
+			  struct tb_port *out)
+{
+	struct tb_bandwidth_group *group;
+	struct tb_tunnel *tunnel;
+
+	/*
+	 * Find all DP tunnels that go through all the same USB4 links
+	 * as this one. Because we always setup tunnels the same way we
+	 * can just check for the routers at both ends of the tunnels
+	 * and if they are the same we have a match.
+	 */
+	list_for_each_entry(tunnel, &tcm->tunnel_list, list) {
+		if (!tb_tunnel_is_dp(tunnel))
+			continue;
+
+		if (tunnel->src_port->sw == in->sw &&
+		    tunnel->dst_port->sw == out->sw) {
+			group = tunnel->src_port->group;
+			if (group) {
+				tb_bandwidth_group_attach_port(group, in);
+				return group;
+			}
+		}
+	}
+
+	/* Pick up next available group then */
+	group = tb_find_free_bandwidth_group(tcm);
+	if (group)
+		tb_bandwidth_group_attach_port(group, in);
+	else
+		tb_port_warn(in, "no available bandwidth groups\n");
+
+	return group;
+}
+
+static void tb_discover_bandwidth_group(struct tb_cm *tcm, struct tb_port *in,
+					struct tb_port *out)
+{
+	if (usb4_dp_port_bandwidth_mode_enabled(in)) {
+		int index, i;
+
+		index = usb4_dp_port_group_id(in);
+		for (i = 0; i < ARRAY_SIZE(tcm->groups); i++) {
+			if (tcm->groups[i].index == index) {
+				tb_bandwidth_group_attach_port(&tcm->groups[i], in);
+				return;
+			}
+		}
+	}
+
+	tb_attach_bandwidth_group(tcm, in, out);
+}
+
+static void tb_detach_bandwidth_group(struct tb_port *in)
+{
+	struct tb_bandwidth_group *group = in->group;
+
+	if (group) {
+		in->group = NULL;
+		list_del_init(&in->group_list);
+
+		tb_port_dbg(in, "detached from bandwidth group %d\n", group->index);
+	}
+}
+
+static void tb_discover_tunnels(struct tb *tb)
+{
+	struct tb_cm *tcm = tb_priv(tb);
+	struct tb_tunnel *tunnel;
+
+	tb_switch_discover_tunnels(tb->root_switch, &tcm->tunnel_list, true);
+
+	list_for_each_entry(tunnel, &tcm->tunnel_list, list) {
+		if (tb_tunnel_is_pci(tunnel)) {
+			struct tb_switch *parent = tunnel->dst_port->sw;
+
+			while (parent != tunnel->src_port->sw) {
+				parent->boot = true;
+				parent = tb_switch_parent(parent);
+			}
+		} else if (tb_tunnel_is_dp(tunnel)) {
+			struct tb_port *in = tunnel->src_port;
+			struct tb_port *out = tunnel->dst_port;
+
+			/* Keep the domain from powering down */
+			pm_runtime_get_sync(&in->sw->dev);
+			pm_runtime_get_sync(&out->sw->dev);
+
+			tb_discover_bandwidth_group(tcm, in, out);
+		}
+	}
+}
+
 static void tb_deactivate_and_free_tunnel(struct tb_tunnel *tunnel)
 {
 	struct tb_port *src_port, *dst_port;
@@ -1605,101 +1696,6 @@ static struct tb_port *tb_find_pcie_down(struct tb_switch *sw,
 	return tb_find_unused_port(sw, TB_TYPE_PCIE_DOWN);
 }
 
-static void
-tb_recalc_estimated_bandwidth_for_group(struct tb_bandwidth_group *group)
-{
-	struct tb_tunnel *first_tunnel;
-	struct tb *tb = group->tb;
-	struct tb_port *in;
-	int ret;
-
-	tb_dbg(tb, "re-calculating bandwidth estimation for group %u\n",
-	       group->index);
-
-	first_tunnel = NULL;
-	list_for_each_entry(in, &group->ports, group_list) {
-		int estimated_bw, estimated_up, estimated_down;
-		struct tb_tunnel *tunnel;
-		struct tb_port *out;
-
-		if (!usb4_dp_port_bandwidth_mode_enabled(in))
-			continue;
-
-		tunnel = tb_find_tunnel(tb, TB_TUNNEL_DP, in, NULL);
-		if (WARN_ON(!tunnel))
-			break;
-
-		if (!first_tunnel) {
-			/*
-			 * Since USB3 bandwidth is shared by all DP
-			 * tunnels under the host router USB4 port, even
-			 * if they do not begin from the host router, we
-			 * can release USB3 bandwidth just once and not
-			 * for each tunnel separately.
-			 */
-			first_tunnel = tunnel;
-			ret = tb_release_unused_usb3_bandwidth(tb,
-				first_tunnel->src_port, first_tunnel->dst_port);
-			if (ret) {
-				tb_tunnel_warn(tunnel,
-					"failed to release unused bandwidth\n");
-				break;
-			}
-		}
-
-		out = tunnel->dst_port;
-		ret = tb_available_bandwidth(tb, in, out, &estimated_up,
-					     &estimated_down, true);
-		if (ret) {
-			tb_tunnel_warn(tunnel,
-				"failed to re-calculate estimated bandwidth\n");
-			break;
-		}
-
-		/*
-		 * Estimated bandwidth includes:
-		 *  - already allocated bandwidth for the DP tunnel
-		 *  - available bandwidth along the path
-		 *  - bandwidth allocated for USB 3.x but not used.
-		 */
-		tb_tunnel_dbg(tunnel,
-			      "re-calculated estimated bandwidth %u/%u Mb/s\n",
-			      estimated_up, estimated_down);
-
-		if (tb_port_path_direction_downstream(in, out))
-			estimated_bw = estimated_down;
-		else
-			estimated_bw = estimated_up;
-
-		if (usb4_dp_port_set_estimated_bandwidth(in, estimated_bw))
-			tb_tunnel_warn(tunnel,
-				       "failed to update estimated bandwidth\n");
-	}
-
-	if (first_tunnel)
-		tb_reclaim_usb3_bandwidth(tb, first_tunnel->src_port,
-					  first_tunnel->dst_port);
-
-	tb_dbg(tb, "bandwidth estimation for group %u done\n", group->index);
-}
-
-static void tb_recalc_estimated_bandwidth(struct tb *tb)
-{
-	struct tb_cm *tcm = tb_priv(tb);
-	int i;
-
-	tb_dbg(tb, "bandwidth consumption changed, re-calculating estimated bandwidth\n");
-
-	for (i = 0; i < ARRAY_SIZE(tcm->groups); i++) {
-		struct tb_bandwidth_group *group = &tcm->groups[i];
-
-		if (!list_empty(&group->ports))
-			tb_recalc_estimated_bandwidth_for_group(group);
-	}
-
-	tb_dbg(tb, "bandwidth re-calculation done\n");
-}
-
 static struct tb_port *tb_find_dp_out(struct tb *tb, struct tb_port *in)
 {
 	struct tb_port *host_port, *port;
-- 
2.43.0

[PATCH 07/10] thunderbolt: Introduce tb_tunnel_direction_downstream()

[Mika Westerberg]

This helper takes tunnel as parameter. Convert existing code to call
this where possible.

No functional changes.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
---
 drivers/thunderbolt/tb.c     |  9 +++------
 drivers/thunderbolt/tunnel.c | 23 +++++++++--------------
 drivers/thunderbolt/tunnel.h |  6 ++++++
 3 files changed, 18 insertions(+), 20 deletions(-)

diff --git a/drivers/thunderbolt/tb.c b/drivers/thunderbolt/tb.c
index d23a80339a8d..e664045ad41c 100644
--- a/drivers/thunderbolt/tb.c
+++ b/drivers/thunderbolt/tb.c
@@ -1387,7 +1387,7 @@ tb_recalc_estimated_bandwidth_for_group(struct tb_bandwidth_group *group)
 		 *  - available bandwidth along the path
 		 *  - bandwidth allocated for USB 3.x but not used.
 		 */
-		if (tb_port_path_direction_downstream(in, out))
+		if (tb_tunnel_direction_downstream(tunnel))
 			estimated_bw = estimated_down;
 		else
 			estimated_bw = estimated_up;
@@ -2388,11 +2388,11 @@ static void tb_handle_dp_bandwidth_request(struct work_struct *work)
 {
 	struct tb_hotplug_event *ev = container_of(work, typeof(*ev), work);
 	int requested_bw, requested_up, requested_down, ret;
-	struct tb_port *in, *out;
 	struct tb_tunnel *tunnel;
 	struct tb *tb = ev->tb;
 	struct tb_cm *tcm = tb_priv(tb);
 	struct tb_switch *sw;
+	struct tb_port *in;
 
 	pm_runtime_get_sync(&tb->dev);
 
@@ -2456,10 +2456,7 @@ static void tb_handle_dp_bandwidth_request(struct work_struct *work)
 
 	tb_port_dbg(in, "requested bandwidth %d Mb/s\n", requested_bw);
 
-
-	out = tunnel->dst_port;
-
-	if (tb_port_path_direction_downstream(in, out)) {
+	if (tb_tunnel_direction_downstream(tunnel)) {
 		requested_up = -1;
 		requested_down = requested_bw;
 	} else {
diff --git a/drivers/thunderbolt/tunnel.c b/drivers/thunderbolt/tunnel.c
index a766ab297064..e02b34654d29 100644
--- a/drivers/thunderbolt/tunnel.c
+++ b/drivers/thunderbolt/tunnel.c
@@ -706,7 +706,7 @@ static int tb_dp_xchg_caps(struct tb_tunnel *tunnel)
 		      "DP OUT maximum supported bandwidth %u Mb/s x%u = %u Mb/s\n",
 		      out_rate, out_lanes, bw);
 
-	if (tb_port_path_direction_downstream(in, out))
+	if (tb_tunnel_direction_downstream(tunnel))
 		max_bw = tunnel->max_down;
 	else
 		max_bw = tunnel->max_up;
@@ -831,7 +831,7 @@ static int tb_dp_bandwidth_alloc_mode_enable(struct tb_tunnel *tunnel)
 	 * max_up/down fields. For discovery we just read what the
 	 * estimation was set to.
 	 */
-	if (tb_port_path_direction_downstream(in, out))
+	if (tb_tunnel_direction_downstream(tunnel))
 		estimated_bw = tunnel->max_down;
 	else
 		estimated_bw = tunnel->max_up;
@@ -971,7 +971,6 @@ static int tb_dp_bandwidth_mode_consumed_bandwidth(struct tb_tunnel *tunnel,
 						   int *consumed_up,
 						   int *consumed_down)
 {
-	struct tb_port *out = tunnel->dst_port;
 	struct tb_port *in = tunnel->src_port;
 	int ret, allocated_bw, max_bw_rounded;
 
@@ -993,7 +992,7 @@ static int tb_dp_bandwidth_mode_consumed_bandwidth(struct tb_tunnel *tunnel,
 	if (allocated_bw == max_bw_rounded)
 		allocated_bw = ret;
 
-	if (tb_port_path_direction_downstream(in, out)) {
+	if (tb_tunnel_direction_downstream(tunnel)) {
 		*consumed_up = 0;
 		*consumed_down = allocated_bw;
 	} else {
@@ -1007,7 +1006,6 @@ static int tb_dp_bandwidth_mode_consumed_bandwidth(struct tb_tunnel *tunnel,
 static int tb_dp_allocated_bandwidth(struct tb_tunnel *tunnel, int *allocated_up,
 				     int *allocated_down)
 {
-	struct tb_port *out = tunnel->dst_port;
 	struct tb_port *in = tunnel->src_port;
 
 	/*
@@ -1029,7 +1027,7 @@ static int tb_dp_allocated_bandwidth(struct tb_tunnel *tunnel, int *allocated_up
 		if (allocated_bw == max_bw_rounded)
 			allocated_bw = ret;
 
-		if (tb_port_path_direction_downstream(in, out)) {
+		if (tb_tunnel_direction_downstream(tunnel)) {
 			*allocated_up = 0;
 			*allocated_down = allocated_bw;
 		} else {
@@ -1046,7 +1044,6 @@ static int tb_dp_allocated_bandwidth(struct tb_tunnel *tunnel, int *allocated_up
 static int tb_dp_alloc_bandwidth(struct tb_tunnel *tunnel, int *alloc_up,
 				 int *alloc_down)
 {
-	struct tb_port *out = tunnel->dst_port;
 	struct tb_port *in = tunnel->src_port;
 	int max_bw_rounded, ret, tmp;
 
@@ -1057,7 +1054,7 @@ static int tb_dp_alloc_bandwidth(struct tb_tunnel *tunnel, int *alloc_up,
 	if (ret < 0)
 		return ret;
 
-	if (tb_port_path_direction_downstream(in, out)) {
+	if (tb_tunnel_direction_downstream(tunnel)) {
 		tmp = min(*alloc_down, max_bw_rounded);
 		ret = usb4_dp_port_allocate_bandwidth(in, tmp);
 		if (ret)
@@ -1143,17 +1140,16 @@ static int tb_dp_read_cap(struct tb_tunnel *tunnel, unsigned int cap, u32 *rate,
 static int tb_dp_maximum_bandwidth(struct tb_tunnel *tunnel, int *max_up,
 				   int *max_down)
 {
-	struct tb_port *in = tunnel->src_port;
 	int ret;
 
-	if (!usb4_dp_port_bandwidth_mode_enabled(in))
+	if (!usb4_dp_port_bandwidth_mode_enabled(tunnel->src_port))
 		return -EOPNOTSUPP;
 
 	ret = tb_dp_bandwidth_mode_maximum_bandwidth(tunnel, NULL);
 	if (ret < 0)
 		return ret;
 
-	if (tb_port_path_direction_downstream(in, tunnel->dst_port)) {
+	if (tb_tunnel_direction_downstream(tunnel)) {
 		*max_up = 0;
 		*max_down = ret;
 	} else {
@@ -1167,8 +1163,7 @@ static int tb_dp_maximum_bandwidth(struct tb_tunnel *tunnel, int *max_up,
 static int tb_dp_consumed_bandwidth(struct tb_tunnel *tunnel, int *consumed_up,
 				    int *consumed_down)
 {
-	struct tb_port *in = tunnel->src_port;
-	const struct tb_switch *sw = in->sw;
+	const struct tb_switch *sw = tunnel->src_port->sw;
 	u32 rate = 0, lanes = 0;
 	int ret;
 
@@ -1214,7 +1209,7 @@ static int tb_dp_consumed_bandwidth(struct tb_tunnel *tunnel, int *consumed_up,
 		return 0;
 	}
 
-	if (tb_port_path_direction_downstream(in, tunnel->dst_port)) {
+	if (tb_tunnel_direction_downstream(tunnel)) {
 		*consumed_up = 0;
 		*consumed_down = tb_dp_bandwidth(rate, lanes);
 	} else {
diff --git a/drivers/thunderbolt/tunnel.h b/drivers/thunderbolt/tunnel.h
index b4cff5482112..1a27ccd08b86 100644
--- a/drivers/thunderbolt/tunnel.h
+++ b/drivers/thunderbolt/tunnel.h
@@ -139,6 +139,12 @@ static inline bool tb_tunnel_is_usb3(const struct tb_tunnel *tunnel)
 	return tunnel->type == TB_TUNNEL_USB3;
 }
 
+static inline bool tb_tunnel_direction_downstream(const struct tb_tunnel *tunnel)
+{
+	return tb_port_path_direction_downstream(tunnel->src_port,
+						 tunnel->dst_port);
+}
+
 const char *tb_tunnel_type_name(const struct tb_tunnel *tunnel);
 
 #define __TB_TUNNEL_PRINT(level, tunnel, fmt, arg...)                   \
-- 
2.43.0

[PATCH 08/10] thunderbolt: Reserve released DisplayPort bandwidth for a group for 10 seconds

[Mika Westerberg]

The USB4 spec says that the Connection Manager should reserve the
bandwidth that is released in the same group for 10 seconds before it
can be shared with other groups. Add support for this. We also delay the
symmetric transition by that same 10 seconds to avoid any unnecessary
transitions (i.e if the released bandwidth is used by another
DisplayPort tunnel in the same group the link can stay asymmetric the
whole time).

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
---
 drivers/thunderbolt/domain.c |   4 +
 drivers/thunderbolt/tb.c     | 199 ++++++++++++++++++++++++++++++-----
 drivers/thunderbolt/tb.h     |  10 ++
 3 files changed, 184 insertions(+), 29 deletions(-)

diff --git a/drivers/thunderbolt/domain.c b/drivers/thunderbolt/domain.c
index ee8a894bd70d..d7abb8c445aa 100644
--- a/drivers/thunderbolt/domain.c
+++ b/drivers/thunderbolt/domain.c
@@ -506,6 +506,10 @@ void tb_domain_remove(struct tb *tb)
 	mutex_unlock(&tb->lock);
 
 	flush_workqueue(tb->wq);
+
+	if (tb->cm_ops->deinit)
+		tb->cm_ops->deinit(tb);
+
 	device_unregister(&tb->dev);
 }
 
diff --git a/drivers/thunderbolt/tb.c b/drivers/thunderbolt/tb.c
index e664045ad41c..eda53567fa4a 100644
--- a/drivers/thunderbolt/tb.c
+++ b/drivers/thunderbolt/tb.c
@@ -17,6 +17,7 @@
 #include "tunnel.h"
 
 #define TB_TIMEOUT		100	/* ms */
+#define TB_RELEASE_BW_TIMEOUT	10000	/* ms */
 
 /*
  * Minimum bandwidth (in Mb/s) that is needed in the single transmitter/receiver
@@ -547,6 +548,10 @@ static int tb_consumed_usb3_pcie_bandwidth(struct tb *tb,
  * Calculates consumed DP bandwidth at @port between path from @src_port
  * to @dst_port. Does not take tunnel starting from @src_port and ending
  * from @src_port into account.
+ *
+ * If there is bandwidth reserved for any of the groups between
+ * @src_port and @dst_port (but not yet used) that is also taken into
+ * account in the returned consumed bandwidth.
  */
 static int tb_consumed_dp_bandwidth(struct tb *tb,
 				    struct tb_port *src_port,
@@ -555,9 +560,11 @@ static int tb_consumed_dp_bandwidth(struct tb *tb,
 				    int *consumed_up,
 				    int *consumed_down)
 {
+	int group_reserved[MAX_GROUPS] = {};
 	struct tb_cm *tcm = tb_priv(tb);
 	struct tb_tunnel *tunnel;
-	int ret;
+	bool downstream;
+	int i, ret;
 
 	*consumed_up = *consumed_down = 0;
 
@@ -566,6 +573,7 @@ static int tb_consumed_dp_bandwidth(struct tb *tb,
 	 * their consumed bandwidth from the available.
 	 */
 	list_for_each_entry(tunnel, &tcm->tunnel_list, list) {
+		const struct tb_bandwidth_group *group;
 		int dp_consumed_up, dp_consumed_down;
 
 		if (tb_tunnel_is_invalid(tunnel))
@@ -577,6 +585,15 @@ static int tb_consumed_dp_bandwidth(struct tb *tb,
 		if (!tb_tunnel_port_on_path(tunnel, port))
 			continue;
 
+		/*
+		 * Calculate what is reserved for groups crossing the
+		 * same ports only once (as that is reserved for all the
+		 * tunnels in the group).
+		 */
+		group = tunnel->src_port->group;
+		if (group && group->reserved && !group_reserved[group->index])
+			group_reserved[group->index] = group->reserved;
+
 		/*
 		 * Ignore the DP tunnel between src_port and dst_port
 		 * because it is the same tunnel and we may be
@@ -595,6 +612,14 @@ static int tb_consumed_dp_bandwidth(struct tb *tb,
 		*consumed_down += dp_consumed_down;
 	}
 
+	downstream = tb_port_path_direction_downstream(src_port, dst_port);
+	for (i = 0; i < ARRAY_SIZE(group_reserved); i++) {
+		if (downstream)
+			*consumed_down += group_reserved[i];
+		else
+			*consumed_up += group_reserved[i];
+	}
+
 	return 0;
 }
 
@@ -1047,8 +1072,6 @@ static int tb_configure_asym(struct tb *tb, struct tb_port *src_port,
  * @tb: Domain structure
  * @src_port: Source adapter to start the transition
  * @dst_port: Destination adapter
- * @requested_up: New lower bandwidth request upstream (Mb/s)
- * @requested_down: New lower bandwidth request downstream (Mb/s)
  * @keep_asym: Keep asymmetric link if preferred
  *
  * Goes over each link from @src_port to @dst_port and tries to
@@ -1056,8 +1079,7 @@ static int tb_configure_asym(struct tb *tb, struct tb_port *src_port,
  * allows and link asymmetric preference is ignored (if @keep_asym is %false).
  */
 static int tb_configure_sym(struct tb *tb, struct tb_port *src_port,
-			    struct tb_port *dst_port, int requested_up,
-			    int requested_down, bool keep_asym)
+			    struct tb_port *dst_port, bool keep_asym)
 {
 	bool clx = false, clx_disabled = false, downstream;
 	struct tb_switch *sw;
@@ -1096,10 +1118,10 @@ static int tb_configure_sym(struct tb *tb, struct tb_port *src_port,
 			 * guard band 10%) as the link was configured asymmetric
 			 * already.
 			 */
-			if (consumed_down + requested_down >= asym_threshold)
+			if (consumed_down >= asym_threshold)
 				continue;
 		} else {
-			if (consumed_up + requested_up >= asym_threshold)
+			if (consumed_up >= asym_threshold)
 				continue;
 		}
 
@@ -1172,7 +1194,7 @@ static void tb_configure_link(struct tb_port *down, struct tb_port *up,
 		struct tb_port *host_port;
 
 		host_port = tb_port_at(tb_route(sw), tb->root_switch);
-		tb_configure_sym(tb, host_port, up, 0, 0, false);
+		tb_configure_sym(tb, host_port, up, false);
 	}
 
 	/* Set the link configured */
@@ -1392,7 +1414,17 @@ tb_recalc_estimated_bandwidth_for_group(struct tb_bandwidth_group *group)
 		else
 			estimated_bw = estimated_up;
 
-		if (usb4_dp_port_set_estimated_bandwidth(in, estimated_bw))
+		/*
+		 * If there is reserved bandwidth for the group that is
+		 * not yet released we report that too.
+		 */
+		tb_tunnel_dbg(tunnel,
+			      "re-calculated estimated bandwidth %u (+ %u reserved) = %u Mb/s\n",
+			      estimated_bw, group->reserved,
+			      estimated_bw + group->reserved);
+
+		if (usb4_dp_port_set_estimated_bandwidth(in,
+				estimated_bw + group->reserved))
 			tb_tunnel_warn(tunnel,
 				       "failed to update estimated bandwidth\n");
 	}
@@ -1421,6 +1453,54 @@ static void tb_recalc_estimated_bandwidth(struct tb *tb)
 	tb_dbg(tb, "bandwidth re-calculation done\n");
 }
 
+static bool __release_group_bandwidth(struct tb_bandwidth_group *group)
+{
+	if (group->reserved) {
+		tb_dbg(group->tb, "group %d released total %d Mb/s\n", group->index,
+			group->reserved);
+		group->reserved = 0;
+		return true;
+	}
+	return false;
+}
+
+static void __configure_group_sym(struct tb_bandwidth_group *group)
+{
+	struct tb_tunnel *tunnel;
+	struct tb_port *in;
+
+	if (list_empty(&group->ports))
+		return;
+
+	/*
+	 * All the tunnels in the group go through the same USB4 links
+	 * so we find the first one here and pass the IN and OUT
+	 * adapters to tb_configure_sym() which now transitions the
+	 * links back to symmetric if bandwidth requirement < asym_threshold.
+	 *
+	 * We do this here to avoid unnecessary transitions (for example
+	 * if the graphics released bandwidth for other tunnel in the
+	 * same group).
+	 */
+	in = list_first_entry(&group->ports, struct tb_port, group_list);
+	tunnel = tb_find_tunnel(group->tb, TB_TUNNEL_DP, in, NULL);
+	if (tunnel)
+		tb_configure_sym(group->tb, in, tunnel->dst_port, true);
+}
+
+static void tb_bandwidth_group_release_work(struct work_struct *work)
+{
+	struct tb_bandwidth_group *group =
+		container_of(work, typeof(*group), release_work.work);
+	struct tb *tb = group->tb;
+
+	mutex_lock(&tb->lock);
+	if (__release_group_bandwidth(group))
+		tb_recalc_estimated_bandwidth(tb);
+	__configure_group_sym(group);
+	mutex_unlock(&tb->lock);
+}
+
 static void tb_init_bandwidth_groups(struct tb_cm *tcm)
 {
 	int i;
@@ -1431,6 +1511,8 @@ static void tb_init_bandwidth_groups(struct tb_cm *tcm)
 		group->tb = tcm_to_tb(tcm);
 		group->index = i + 1;
 		INIT_LIST_HEAD(&group->ports);
+		INIT_DELAYED_WORK(&group->release_work,
+				  tb_bandwidth_group_release_work);
 	}
 }
 
@@ -1524,6 +1606,12 @@ static void tb_detach_bandwidth_group(struct tb_port *in)
 		list_del_init(&in->group_list);
 
 		tb_port_dbg(in, "detached from bandwidth group %d\n", group->index);
+
+		/* No more tunnels so release the reserved bandwidth if any */
+		if (list_empty(&group->ports)) {
+			cancel_delayed_work(&group->release_work);
+			__release_group_bandwidth(group);
+		}
 	}
 }
 
@@ -1582,7 +1670,7 @@ static void tb_deactivate_and_free_tunnel(struct tb_tunnel *tunnel)
 		 * If bandwidth on a link is < asym_threshold
 		 * transition the link to symmetric.
 		 */
-		tb_configure_sym(tb, src_port, dst_port, 0, 0, true);
+		tb_configure_sym(tb, src_port, dst_port, true);
 		/* Now we can allow the domain to runtime suspend again */
 		pm_runtime_mark_last_busy(&dst_port->sw->dev);
 		pm_runtime_put_autosuspend(&dst_port->sw->dev);
@@ -2239,8 +2327,10 @@ static int tb_alloc_dp_bandwidth(struct tb_tunnel *tunnel, int *requested_up,
 	int allocated_up, allocated_down, available_up, available_down, ret;
 	int requested_up_corrected, requested_down_corrected, granularity;
 	int max_up, max_down, max_up_rounded, max_down_rounded;
+	struct tb_bandwidth_group *group;
 	struct tb *tb = tunnel->tb;
 	struct tb_port *in, *out;
+	bool downstream;
 
 	ret = tb_tunnel_allocated_bandwidth(tunnel, &allocated_up, &allocated_down);
 	if (ret)
@@ -2304,21 +2394,44 @@ static int tb_alloc_dp_bandwidth(struct tb_tunnel *tunnel, int *requested_up,
 		goto fail;
 	}
 
+	downstream = tb_tunnel_direction_downstream(tunnel);
+	group = in->group;
+
 	if ((*requested_up >= 0 && requested_up_corrected <= allocated_up) ||
 	    (*requested_down >= 0 && requested_down_corrected <= allocated_down)) {
-		/*
-		 * If bandwidth on a link is < asym_threshold transition
-		 * the link to symmetric.
-		 */
-		tb_configure_sym(tb, in, out, *requested_up, *requested_down, true);
-		/*
-		 * If requested bandwidth is less or equal than what is
-		 * currently allocated to that tunnel we simply change
-		 * the reservation of the tunnel. Since all the tunnels
-		 * going out from the same USB4 port are in the same
-		 * group the released bandwidth will be taken into
-		 * account for the other tunnels automatically below.
-		 */
+		if (tunnel->bw_mode) {
+			int reserved;
+			/*
+			 * If requested bandwidth is less or equal than
+			 * what is currently allocated to that tunnel we
+			 * simply change the reservation of the tunnel
+			 * and add the released bandwidth for the group
+			 * for the next 10s. Then we release it for
+			 * others to use.
+			 */
+			if (downstream)
+				reserved = allocated_down - *requested_down;
+			else
+				reserved = allocated_up - *requested_up;
+
+			if (reserved > 0) {
+				group->reserved += reserved;
+				tb_dbg(tb, "group %d reserved %d total %d Mb/s\n",
+				       group->index, reserved, group->reserved);
+
+				/*
+				 * If it was not already pending,
+				 * schedule release now. If it is then
+				 * postpone it for the next 10s (unless
+				 * it is already running in which case
+				 * the 10s already expired and we should
+				 * give the reserved back to others).
+				 */
+				mod_delayed_work(system_wq, &group->release_work,
+					msecs_to_jiffies(TB_RELEASE_BW_TIMEOUT));
+			}
+		}
+
 		return tb_tunnel_alloc_bandwidth(tunnel, requested_up,
 						 requested_down);
 	}
@@ -2341,11 +2454,15 @@ static int tb_alloc_dp_bandwidth(struct tb_tunnel *tunnel, int *requested_up,
 	if (ret)
 		goto reclaim;
 
-	tb_tunnel_dbg(tunnel, "bandwidth available for allocation %d/%d Mb/s\n",
-		      available_up, available_down);
+	tb_tunnel_dbg(tunnel, "bandwidth available for allocation %d/%d (+ %u reserved) Mb/s\n",
+		      available_up, available_down, group->reserved);
+
+	if ((*requested_up >= 0 &&
+		available_up + group->reserved >= requested_up_corrected) ||
+	    (*requested_down >= 0 &&
+		available_down + group->reserved >= requested_down_corrected)) {
+		int released = 0;
 
-	if ((*requested_up >= 0 && available_up >= requested_up_corrected) ||
-	    (*requested_down >= 0 && available_down >= requested_down_corrected)) {
 		/*
 		 * If bandwidth on a link is >= asym_threshold
 		 * transition the link to asymmetric.
@@ -2353,7 +2470,7 @@ static int tb_alloc_dp_bandwidth(struct tb_tunnel *tunnel, int *requested_up,
 		ret = tb_configure_asym(tb, in, out, *requested_up,
 					*requested_down);
 		if (ret) {
-			tb_configure_sym(tb, in, out, 0, 0, true);
+			tb_configure_sym(tb, in, out, true);
 			goto fail;
 		}
 
@@ -2361,7 +2478,20 @@ static int tb_alloc_dp_bandwidth(struct tb_tunnel *tunnel, int *requested_up,
 						requested_down);
 		if (ret) {
 			tb_tunnel_warn(tunnel, "failed to allocate bandwidth\n");
-			tb_configure_sym(tb, in, out, 0, 0, true);
+			tb_configure_sym(tb, in, out, true);
+		}
+
+		if (downstream) {
+			if (*requested_down > available_down)
+				released = *requested_down - available_down;
+		} else {
+			if (*requested_up > available_up)
+				released = *requested_up - available_up;
+		}
+		if (released) {
+			group->reserved -= released;
+			tb_dbg(tb, "group %d released %d total %d Mb/s\n",
+			       group->index, released, group->reserved);
 		}
 	} else {
 		ret = -ENOBUFS;
@@ -2585,6 +2715,16 @@ static void tb_stop(struct tb *tb)
 	tcm->hotplug_active = false; /* signal tb_handle_hotplug to quit */
 }
 
+static void tb_deinit(struct tb *tb)
+{
+	struct tb_cm *tcm = tb_priv(tb);
+	int i;
+
+	/* Cancel all the release bandwidth workers */
+	for (i = 0; i < ARRAY_SIZE(tcm->groups); i++)
+		cancel_delayed_work_sync(&tcm->groups[i].release_work);
+}
+
 static int tb_scan_finalize_switch(struct device *dev, void *data)
 {
 	if (tb_is_switch(dev)) {
@@ -2893,6 +3033,7 @@ static int tb_runtime_resume(struct tb *tb)
 static const struct tb_cm_ops tb_cm_ops = {
 	.start = tb_start,
 	.stop = tb_stop,
+	.deinit = tb_deinit,
 	.suspend_noirq = tb_suspend_noirq,
 	.resume_noirq = tb_resume_noirq,
 	.freeze_noirq = tb_freeze_noirq,
diff --git a/drivers/thunderbolt/tb.h b/drivers/thunderbolt/tb.h
index d0dfbf040356..1bbbeb034e0e 100644
--- a/drivers/thunderbolt/tb.h
+++ b/drivers/thunderbolt/tb.h
@@ -217,6 +217,11 @@ struct tb_switch {
  * @tb: Pointer to the domain the group belongs to
  * @index: Index of the group (aka Group_ID). Valid values %1-%7
  * @ports: DP IN adapters belonging to this group are linked here
+ * @reserved: Bandwidth released by one tunnel in the group, available
+ *	      to others. This is reported as part of estimated_bw for
+ *	      the group.
+ * @release_work: Worker to release the @reserved if it is not used by
+ *		  any of the tunnels.
  *
  * Any tunnel that requires isochronous bandwidth (that's DP for now) is
  * attached to a bandwidth group. All tunnels going through the same
@@ -227,6 +232,8 @@ struct tb_bandwidth_group {
 	struct tb *tb;
 	int index;
 	struct list_head ports;
+	int reserved;
+	struct delayed_work release_work;
 };
 
 /**
@@ -452,6 +459,8 @@ struct tb_path {
  *		  ICM to send driver ready message to the firmware.
  * @start: Starts the domain
  * @stop: Stops the domain
+ * @deinit: Perform any cleanup after the domain is stopped but before
+ *	     it is unregistered. Called without @tb->lock taken. Optional.
  * @suspend_noirq: Connection manager specific suspend_noirq
  * @resume_noirq: Connection manager specific resume_noirq
  * @suspend: Connection manager specific suspend
@@ -485,6 +494,7 @@ struct tb_cm_ops {
 	int (*driver_ready)(struct tb *tb);
 	int (*start)(struct tb *tb, bool reset);
 	void (*stop)(struct tb *tb);
+	void (*deinit)(struct tb *tb);
 	int (*suspend_noirq)(struct tb *tb);
 	int (*resume_noirq)(struct tb *tb);
 	int (*suspend)(struct tb *tb);
-- 
2.43.0

[PATCH 09/10] thunderbolt: Calculate DisplayPort tunnel bandwidth after DPRX capabilities read

[Mika Westerberg]

From: Gil Fine <gil.fine@linux.intel.com>

According to USB4 Connection Manager guide, after DisplayPort tunnel was
setup, the DPRX capabilities read is performed by the DPTX. According to
VESA spec, this shall be completed within 5 seconds after the DisplayPort
tunnel was setup. Hence, if the bit: DPRX Capabilities Read Done, was
not set to '1' by this time, we timeout and fail calculating DisplayPort
tunnel consumed bandwidth.

Signed-off-by: Gil Fine <gil.fine@linux.intel.com>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
---
 drivers/thunderbolt/tunnel.c | 16 ++++++----------
 1 file changed, 6 insertions(+), 10 deletions(-)

diff --git a/drivers/thunderbolt/tunnel.c b/drivers/thunderbolt/tunnel.c
index e02b34654d29..cb6609a56a03 100644
--- a/drivers/thunderbolt/tunnel.c
+++ b/drivers/thunderbolt/tunnel.c
@@ -1184,17 +1184,13 @@ static int tb_dp_consumed_bandwidth(struct tb_tunnel *tunnel, int *consumed_up,
 		/*
 		 * Then see if the DPRX negotiation is ready and if yes
 		 * return that bandwidth (it may be smaller than the
-		 * reduced one). Otherwise return the remote (possibly
-		 * reduced) caps.
+		 * reduced one). According to VESA spec, the DPRX
+		 * negotiation shall compete in 5 seconds after tunnel
+		 * established. We give it 100ms extra just in case.
 		 */
-		ret = tb_dp_wait_dprx(tunnel, 150);
-		if (ret) {
-			if (ret == -ETIMEDOUT)
-				ret = tb_dp_read_cap(tunnel, DP_REMOTE_CAP,
-						     &rate, &lanes);
-			if (ret)
-				return ret;
-		}
+		ret = tb_dp_wait_dprx(tunnel, 5100);
+		if (ret)
+			return ret;
 		ret = tb_dp_read_cap(tunnel, DP_COMMON_CAP, &rate, &lanes);
 		if (ret)
 			return ret;
-- 
2.43.0

[PATCH 10/10] thunderbolt: Improve DisplayPort tunnel setup process to be more robust

[Mika Westerberg]

From: Gil Fine <gil.fine@linux.intel.com>

After DisplayPort tunnel setup, we add verification that the DPRX
capabilities read process completed. Otherwise, we bail out, teardown
the tunnel, and try setup another DisplayPort tunnel using next
available DP IN adapter. We do so till all DP IN adapters tried. This
way, we avoid allocating DP IN adapter and (bandwidth for it) for
unusable tunnel.

Signed-off-by: Gil Fine <gil.fine@linux.intel.com>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
---
 drivers/thunderbolt/tb.c | 84 ++++++++++++++++++++--------------------
 1 file changed, 43 insertions(+), 41 deletions(-)

diff --git a/drivers/thunderbolt/tb.c b/drivers/thunderbolt/tb.c
index eda53567fa4a..306c62c35a05 100644
--- a/drivers/thunderbolt/tb.c
+++ b/drivers/thunderbolt/tb.c
@@ -1821,48 +1821,14 @@ static struct tb_port *tb_find_dp_out(struct tb *tb, struct tb_port *in)
 	return NULL;
 }
 
-static bool tb_tunnel_one_dp(struct tb *tb)
+static bool tb_tunnel_one_dp(struct tb *tb, struct tb_port *in,
+			     struct tb_port *out)
 {
 	int available_up, available_down, ret, link_nr;
 	struct tb_cm *tcm = tb_priv(tb);
-	struct tb_port *port, *in, *out;
 	int consumed_up, consumed_down;
 	struct tb_tunnel *tunnel;
 
-	/*
-	 * Find pair of inactive DP IN and DP OUT adapters and then
-	 * establish a DP tunnel between them.
-	 */
-	tb_dbg(tb, "looking for DP IN <-> DP OUT pairs:\n");
-
-	in = NULL;
-	out = NULL;
-	list_for_each_entry(port, &tcm->dp_resources, list) {
-		if (!tb_port_is_dpin(port))
-			continue;
-
-		if (tb_port_is_enabled(port)) {
-			tb_port_dbg(port, "DP IN in use\n");
-			continue;
-		}
-
-		in = port;
-		tb_port_dbg(in, "DP IN available\n");
-
-		out = tb_find_dp_out(tb, port);
-		if (out)
-			break;
-	}
-
-	if (!in) {
-		tb_dbg(tb, "no suitable DP IN adapter available, not tunneling\n");
-		return false;
-	}
-	if (!out) {
-		tb_dbg(tb, "no suitable DP OUT adapter available, not tunneling\n");
-		return false;
-	}
-
 	/*
 	 * This is only applicable to links that are not bonded (so
 	 * when Thunderbolt 1 hardware is involved somewhere in the
@@ -1923,15 +1889,19 @@ static bool tb_tunnel_one_dp(struct tb *tb)
 		goto err_free;
 	}
 
+	/* If fail reading tunnel's consumed bandwidth, tear it down */
+	ret = tb_tunnel_consumed_bandwidth(tunnel, &consumed_up, &consumed_down);
+	if (ret)
+		goto err_deactivate;
+
 	list_add_tail(&tunnel->list, &tcm->tunnel_list);
-	tb_reclaim_usb3_bandwidth(tb, in, out);
 
+	tb_reclaim_usb3_bandwidth(tb, in, out);
 	/*
 	 * Transition the links to asymmetric if the consumption exceeds
 	 * the threshold.
 	 */
-	if (!tb_tunnel_consumed_bandwidth(tunnel, &consumed_up, &consumed_down))
-		tb_configure_asym(tb, in, out, consumed_up, consumed_down);
+	tb_configure_asym(tb, in, out, consumed_up, consumed_down);
 
 	/* Update the domain with the new bandwidth estimation */
 	tb_recalc_estimated_bandwidth(tb);
@@ -1943,6 +1913,8 @@ static bool tb_tunnel_one_dp(struct tb *tb)
 	tb_increase_tmu_accuracy(tunnel);
 	return true;
 
+err_deactivate:
+	tb_tunnel_deactivate(tunnel);
 err_free:
 	tb_tunnel_free(tunnel);
 err_reclaim_usb:
@@ -1962,13 +1934,43 @@ static bool tb_tunnel_one_dp(struct tb *tb)
 
 static void tb_tunnel_dp(struct tb *tb)
 {
+	struct tb_cm *tcm = tb_priv(tb);
+	struct tb_port *port, *in, *out;
+
 	if (!tb_acpi_may_tunnel_dp()) {
 		tb_dbg(tb, "DP tunneling disabled, not creating tunnel\n");
 		return;
 	}
 
-	while (tb_tunnel_one_dp(tb))
-		;
+	/*
+	 * Find pair of inactive DP IN and DP OUT adapters and then
+	 * establish a DP tunnel between them.
+	 */
+	tb_dbg(tb, "looking for DP IN <-> DP OUT pairs:\n");
+
+	in = NULL;
+	out = NULL;
+	list_for_each_entry(port, &tcm->dp_resources, list) {
+		if (!tb_port_is_dpin(port))
+			continue;
+
+		if (tb_port_is_enabled(port)) {
+			tb_port_dbg(port, "DP IN in use\n");
+			continue;
+		}
+
+		in = port;
+		tb_port_dbg(in, "DP IN available\n");
+
+		out = tb_find_dp_out(tb, port);
+		if (out)
+			tb_tunnel_one_dp(tb, in, out);
+		else
+			tb_port_dbg(in, "no suitable DP OUT adapter available, not tunneling\n");
+	}
+
+	if (!in)
+		tb_dbg(tb, "no suitable DP IN adapter available, not tunneling\n");
 }
 
 static void tb_dp_resource_unavailable(struct tb *tb, struct tb_port *port)
-- 
2.43.0

Re: [PATCH 00/10] thunderbolt: DisplayPort and bandwidth allocation mode improvements

[Mika Westerberg]

On Fri, Feb 09, 2024 at 04:13:25PM +0200, Mika Westerberg wrote:
> Hi all,
> 
> This series improves bandwidth allocation mode support to follow the
> USB4 spec more closely. In addition to that DisplayPort tunneling is
> made more robust so that if a tunnel does not come up properly (for
> example on a discrete controller the DisplayPort source from the
> graphics is not connected to the DisplayPort sink of the controller) we
> tear it down and continue pairing.
> 
> This series improves bandwidth allocation mode support and DisplayPort
> tunneling to follow the USB4 spec more closely.
> 
> Gil Fine (2):
>   thunderbolt: Calculate DisplayPort tunnel bandwidth after DPRX capabilities read
>   thunderbolt: Improve DisplayPort tunnel setup process to be more robust
> 
> Mika Westerberg (8):
>   thunderbolt: Use DP_LOCAL_CAP for maximum bandwidth calculation
>   thunderbolt: Re-calculate estimated bandwidth when allocation mode is enabled
>   thunderbolt: Handle bandwidth allocation mode disable request
>   thunderbolt: Log an error if DPTX request is not cleared
>   thunderbolt: Fail the failed bandwidth request properly
>   thunderbolt: Re-order bandwidth group functions
>   thunderbolt: Introduce tb_tunnel_direction_downstream()
>   thunderbolt: Reserve released DisplayPort bandwidth for a group for 10 seconds

All applied to thunderbolt.git/next.