diff options
Diffstat (limited to 'drivers/thunderbolt/tb.c')
| -rw-r--r-- | drivers/thunderbolt/tb.c | 1968 |
1 files changed, 1456 insertions, 512 deletions
diff --git a/drivers/thunderbolt/tb.c b/drivers/thunderbolt/tb.c index 3fb4553a6442..4f5f1dfc0fbf 100644 --- a/drivers/thunderbolt/tb.c +++ b/drivers/thunderbolt/tb.c @@ -16,8 +16,38 @@ #include "tb_regs.h" #include "tunnel.h" -#define TB_TIMEOUT 100 /* ms */ -#define MAX_GROUPS 7 /* max Group_ID is 7 */ +#define TB_TIMEOUT 100 /* ms */ +#define TB_RELEASE_BW_TIMEOUT 10000 /* ms */ + +/* + * How many time bandwidth allocation request from graphics driver is + * retried if the DP tunnel is still activating. + */ +#define TB_BW_ALLOC_RETRIES 3 + +/* + * Minimum bandwidth (in Mb/s) that is needed in the single transmitter/receiver + * direction. This is 40G - 10% guard band bandwidth. + */ +#define TB_ASYM_MIN (40000 * 90 / 100) + +/* + * Threshold bandwidth (in Mb/s) that is used to switch the links to + * asymmetric and back. This is selected as 45G which means when the + * request is higher than this, we switch the link to asymmetric, and + * when it is less than this we switch it back. The 45G is selected so + * that we still have 27G (of the total 72G) for bulk PCIe traffic when + * switching back to symmetric. + */ +#define TB_ASYM_THRESHOLD 45000 + +#define MAX_GROUPS 7 /* max Group_ID is 7 */ + +static unsigned int asym_threshold = TB_ASYM_THRESHOLD; +module_param_named(asym_threshold, asym_threshold, uint, 0444); +MODULE_PARM_DESC(asym_threshold, + "threshold (Mb/s) when to Gen 4 switch link symmetry. 0 disables. (default: " + __MODULE_STRING(TB_ASYM_THRESHOLD) ")"); /** * struct tb_cm - Simple Thunderbolt connection manager @@ -45,120 +75,20 @@ static inline struct tb *tcm_to_tb(struct tb_cm *tcm) } struct tb_hotplug_event { - struct work_struct work; + struct delayed_work work; struct tb *tb; u64 route; u8 port; bool unplug; + int retry; }; -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_scan_port(struct tb_port *port); static void tb_handle_hotplug(struct work_struct *work); +static void tb_dp_resource_unavailable(struct tb *tb, struct tb_port *port, + const char *reason); +static void tb_queue_dp_bandwidth_request(struct tb *tb, u64 route, u8 port, + int retry, unsigned long delay); static void tb_queue_hotplug(struct tb *tb, u64 route, u8 port, bool unplug) { @@ -172,8 +102,8 @@ static void tb_queue_hotplug(struct tb *tb, u64 route, u8 port, bool unplug) ev->route = route; ev->port = port; ev->unplug = unplug; - INIT_WORK(&ev->work, tb_handle_hotplug); - queue_work(tb->wq, &ev->work); + INIT_DELAYED_WORK(&ev->work, tb_handle_hotplug); + queue_delayed_work(tb->wq, &ev->work, 0); } /* enumeration & hot plug handling */ @@ -190,7 +120,17 @@ static void tb_add_dp_resources(struct tb_switch *sw) if (!tb_switch_query_dp_resource(sw, port)) continue; - list_add_tail(&port->list, &tcm->dp_resources); + /* + * If DP IN on device router exist, position it at the + * beginning of the DP resources list, so that it is used + * before DP IN of the host router. This way external GPU(s) + * will be prioritized when pairing DP IN to a DP OUT. + */ + if (tb_route(sw)) + list_add(&port->list, &tcm->dp_resources); + else + list_add_tail(&port->list, &tcm->dp_resources); + tb_port_dbg(port, "DP IN resource available\n"); } } @@ -255,13 +195,13 @@ static int tb_enable_clx(struct tb_switch *sw) * this in the future to cover the whole topology if it turns * out to be beneficial. */ - while (sw && sw->config.depth > 1) + while (sw && tb_switch_depth(sw) > 1) sw = tb_switch_parent(sw); if (!sw) return 0; - if (sw->config.depth != 1) + if (tb_switch_depth(sw) != 1) return 0; /* @@ -285,14 +225,30 @@ static int tb_enable_clx(struct tb_switch *sw) return ret == -EOPNOTSUPP ? 0 : ret; } -/* Disables CL states up to the host router */ -static void tb_disable_clx(struct tb_switch *sw) +/* + * Disables CL states from @sw up to the host router. + * + * This can be used to figure out whether the link was setup by us or the + * boot firmware so we don't accidentally enable them if they were not + * enabled during discovery. + */ +static bool tb_disable_clx(struct tb_switch *sw) { + bool disabled = false; + do { - if (tb_switch_clx_disable(sw) < 0) + int ret; + + ret = tb_switch_clx_disable(sw); + if (ret > 0) + disabled = true; + else if (ret < 0) tb_sw_warn(sw, "failed to disable CL states\n"); + sw = tb_switch_parent(sw); } while (sw); + + return disabled; } static int tb_increase_switch_tmu_accuracy(struct device *dev, void *data) @@ -342,13 +298,31 @@ static void tb_increase_tmu_accuracy(struct tb_tunnel *tunnel) device_for_each_child(&sw->dev, NULL, tb_increase_switch_tmu_accuracy); } +static int tb_switch_tmu_hifi_uni_required(struct device *dev, void *not_used) +{ + struct tb_switch *sw = tb_to_switch(dev); + + if (sw && tb_switch_tmu_is_enabled(sw) && + tb_switch_tmu_is_configured(sw, TB_SWITCH_TMU_MODE_HIFI_UNI)) + return 1; + + return device_for_each_child(dev, NULL, + tb_switch_tmu_hifi_uni_required); +} + +static bool tb_tmu_hifi_uni_required(struct tb *tb) +{ + return device_for_each_child(&tb->dev, NULL, + tb_switch_tmu_hifi_uni_required) == 1; +} + static int tb_enable_tmu(struct tb_switch *sw) { int ret; /* * If both routers at the end of the link are v2 we simply - * enable the enhanched uni-directional mode. That covers all + * enable the enhanced uni-directional mode. That covers all * the CL states. For v1 and before we need to use the normal * rate to allow CL1 (when supported). Otherwise we keep the TMU * running at the highest accuracy. @@ -356,12 +330,30 @@ static int tb_enable_tmu(struct tb_switch *sw) ret = tb_switch_tmu_configure(sw, TB_SWITCH_TMU_MODE_MEDRES_ENHANCED_UNI); if (ret == -EOPNOTSUPP) { - if (tb_switch_clx_is_enabled(sw, TB_CL1)) - ret = tb_switch_tmu_configure(sw, - TB_SWITCH_TMU_MODE_LOWRES); - else - ret = tb_switch_tmu_configure(sw, - TB_SWITCH_TMU_MODE_HIFI_BI); + if (tb_switch_clx_is_enabled(sw, TB_CL1)) { + /* + * Figure out uni-directional HiFi TMU requirements + * currently in the domain. If there are no + * uni-directional HiFi requirements we can put the TMU + * into LowRes mode. + * + * Deliberately skip bi-directional HiFi links + * as these work independently of other links + * (and they do not allow any CL states anyway). + */ + if (tb_tmu_hifi_uni_required(sw->tb)) + ret = tb_switch_tmu_configure(sw, + TB_SWITCH_TMU_MODE_HIFI_UNI); + else + ret = tb_switch_tmu_configure(sw, + TB_SWITCH_TMU_MODE_LOWRES); + } else { + ret = tb_switch_tmu_configure(sw, TB_SWITCH_TMU_MODE_HIFI_BI); + } + + /* If not supported, fallback to bi-directional HiFi */ + if (ret == -EOPNOTSUPP) + ret = tb_switch_tmu_configure(sw, TB_SWITCH_TMU_MODE_HIFI_BI); } if (ret) return ret; @@ -421,34 +413,6 @@ 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)) @@ -462,8 +426,6 @@ static void tb_port_unconfigure_xdomain(struct tb_port *port) usb4_port_unconfigure_xdomain(port); else tb_lc_unconfigure_xdomain(port); - - tb_port_enable(port->dual_link_port); } static void tb_scan_xdomain(struct tb_port *port) @@ -492,10 +454,8 @@ static void tb_scan_xdomain(struct tb_port *port) } } -/** - * tb_find_unused_port() - return the first inactive port on @sw - * @sw: Switch to find the port on - * @type: Port type to look for +/* + * Returns the first inactive port on @sw. */ static struct tb_port *tb_find_unused_port(struct tb_switch *sw, enum tb_port_type type) @@ -553,7 +513,7 @@ static struct tb_tunnel *tb_find_first_usb3_tunnel(struct tb *tb, struct tb_switch *sw; /* Pick the router that is deepest in the topology */ - if (dst_port->sw->config.depth > src_port->sw->config.depth) + if (tb_port_path_direction_downstream(src_port, dst_port)) sw = dst_port->sw; else sw = src_port->sw; @@ -572,133 +532,327 @@ static struct tb_tunnel *tb_find_first_usb3_tunnel(struct tb *tb, return tb_find_tunnel(tb, TB_TUNNEL_USB3, usb3_down, NULL); } -static int tb_available_bandwidth(struct tb *tb, struct tb_port *src_port, - struct tb_port *dst_port, int *available_up, int *available_down) -{ - int usb3_consumed_up, usb3_consumed_down, ret; - struct tb_cm *tcm = tb_priv(tb); +/** + * tb_consumed_usb3_pcie_bandwidth() - Consumed USB3/PCIe bandwidth over a single link + * @tb: Domain structure + * @src_port: Source protocol adapter + * @dst_port: Destination protocol adapter + * @port: USB4 port the consumed bandwidth is calculated + * @consumed_up: Consumed upstream bandwidth (Mb/s) + * @consumed_down: Consumed downstream bandwidth (Mb/s) + * + * Calculates consumed USB3 and PCIe bandwidth at @port between path + * from @src_port to @dst_port. Does not take USB3 tunnel starting from + * @src_port and ending on @src_port into account because that bandwidth is + * already included in as part of the "first hop" USB3 tunnel. + * + * Return: %0 on success, negative errno otherwise. + */ +static int tb_consumed_usb3_pcie_bandwidth(struct tb *tb, + struct tb_port *src_port, + struct tb_port *dst_port, + struct tb_port *port, + int *consumed_up, + int *consumed_down) +{ + int pci_consumed_up, pci_consumed_down; struct tb_tunnel *tunnel; - struct tb_port *port; - tb_dbg(tb, "calculating available bandwidth between %llx:%u <-> %llx:%u\n", - tb_route(src_port->sw), src_port->port, tb_route(dst_port->sw), - dst_port->port); + *consumed_up = *consumed_down = 0; tunnel = tb_find_first_usb3_tunnel(tb, src_port, dst_port); - if (tunnel && tunnel->src_port != src_port && - tunnel->dst_port != dst_port) { - ret = tb_tunnel_consumed_bandwidth(tunnel, &usb3_consumed_up, - &usb3_consumed_down); + if (tunnel && !tb_port_is_usb3_down(src_port) && + !tb_port_is_usb3_up(dst_port)) { + int ret; + + ret = tb_tunnel_consumed_bandwidth(tunnel, consumed_up, + consumed_down); if (ret) return ret; - } else { - usb3_consumed_up = 0; - usb3_consumed_down = 0; } - /* Maximum possible bandwidth asymmetric Gen 4 link is 120 Gb/s */ - *available_up = *available_down = 120000; + /* + * If there is anything reserved for PCIe bulk traffic take it + * into account here too. + */ + if (tb_tunnel_reserved_pci(port, &pci_consumed_up, &pci_consumed_down)) { + *consumed_up += pci_consumed_up; + *consumed_down += pci_consumed_down; + } - /* Find the minimum available bandwidth over all links */ - tb_for_each_port_on_path(src_port, dst_port, port) { - int link_speed, link_width, up_bw, down_bw; + return 0; +} - if (!tb_port_is_null(port)) +/** + * tb_consumed_dp_bandwidth() - Consumed DP bandwidth over a single link + * @tb: Domain structure + * @src_port: Source protocol adapter + * @dst_port: Destination protocol adapter + * @port: USB4 port the consumed bandwidth is calculated + * @consumed_up: Consumed upstream bandwidth (Mb/s) + * @consumed_down: Consumed downstream bandwidth (Mb/s) + * + * 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. + * + * Return: %0 on success, negative errno otherwise. + */ +static int tb_consumed_dp_bandwidth(struct tb *tb, + struct tb_port *src_port, + struct tb_port *dst_port, + struct tb_port *port, + int *consumed_up, + int *consumed_down) +{ + int group_reserved[MAX_GROUPS] = {}; + struct tb_cm *tcm = tb_priv(tb); + struct tb_tunnel *tunnel; + bool downstream; + int i, ret; + + *consumed_up = *consumed_down = 0; + + /* + * Find all DP tunnels that cross the port and reduce + * 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)) + continue; + + if (!tb_tunnel_is_dp(tunnel)) + continue; + + if (!tb_tunnel_port_on_path(tunnel, port)) continue; - if (tb_is_upstream_port(port)) { - link_speed = port->sw->link_speed; + /* + * 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 + * re-calculating estimated bandwidth. + */ + if (tunnel->src_port == src_port && + tunnel->dst_port == dst_port) + continue; + + ret = tb_tunnel_consumed_bandwidth(tunnel, &dp_consumed_up, + &dp_consumed_down); + if (ret) + return ret; + + *consumed_up += dp_consumed_up; + *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; +} + +static bool tb_asym_supported(struct tb_port *src_port, struct tb_port *dst_port, + struct tb_port *port) +{ + bool downstream = tb_port_path_direction_downstream(src_port, dst_port); + enum tb_link_width width; + + if (tb_is_upstream_port(port)) + width = downstream ? TB_LINK_WIDTH_ASYM_RX : TB_LINK_WIDTH_ASYM_TX; + else + width = downstream ? TB_LINK_WIDTH_ASYM_TX : TB_LINK_WIDTH_ASYM_RX; + + return tb_port_width_supported(port, width); +} + +/** + * tb_maximum_bandwidth() - Maximum bandwidth over a single link + * @tb: Domain structure + * @src_port: Source protocol adapter + * @dst_port: Destination protocol adapter + * @port: USB4 port the total bandwidth is calculated + * @max_up: Maximum upstream bandwidth (Mb/s) + * @max_down: Maximum downstream bandwidth (Mb/s) + * @include_asym: Include bandwidth if the link is switched from + * symmetric to asymmetric + * + * Returns maximum possible bandwidth in @max_up and @max_down over a + * single link at @port. If @include_asym is set then includes the + * additional banwdith if the links are transitioned into asymmetric to + * direction from @src_port to @dst_port. + * + * Return: %0 on success, negative errno otherwise. + */ +static int tb_maximum_bandwidth(struct tb *tb, struct tb_port *src_port, + struct tb_port *dst_port, struct tb_port *port, + int *max_up, int *max_down, bool include_asym) +{ + bool downstream = tb_port_path_direction_downstream(src_port, dst_port); + int link_speed, link_width, up_bw, down_bw; + + /* + * Can include asymmetric, only if it is actually supported by + * the lane adapter. + */ + if (!tb_asym_supported(src_port, dst_port, port)) + include_asym = false; + + if (tb_is_upstream_port(port)) { + link_speed = port->sw->link_speed; + /* + * sw->link_width is from upstream perspective so we use + * the opposite for downstream of the host router. + */ + if (port->sw->link_width == TB_LINK_WIDTH_ASYM_TX) { + up_bw = link_speed * 3 * 1000; + down_bw = link_speed * 1 * 1000; + } else if (port->sw->link_width == TB_LINK_WIDTH_ASYM_RX) { + up_bw = link_speed * 1 * 1000; + down_bw = link_speed * 3 * 1000; + } else if (include_asym) { /* - * sw->link_width is from upstream perspective - * so we use the opposite for downstream of the - * host router. + * The link is symmetric at the moment but we + * can switch it to asymmetric as needed. Report + * this bandwidth as available (even though it + * is not yet enabled). */ - if (port->sw->link_width == TB_LINK_WIDTH_ASYM_TX) { - up_bw = link_speed * 3 * 1000; - down_bw = link_speed * 1 * 1000; - } else if (port->sw->link_width == TB_LINK_WIDTH_ASYM_RX) { + if (downstream) { up_bw = link_speed * 1 * 1000; down_bw = link_speed * 3 * 1000; } else { - up_bw = link_speed * port->sw->link_width * 1000; - down_bw = up_bw; + up_bw = link_speed * 3 * 1000; + down_bw = link_speed * 1 * 1000; } } else { - link_speed = tb_port_get_link_speed(port); - if (link_speed < 0) - return link_speed; - - link_width = tb_port_get_link_width(port); - if (link_width < 0) - return link_width; - - if (link_width == TB_LINK_WIDTH_ASYM_TX) { + up_bw = link_speed * port->sw->link_width * 1000; + down_bw = up_bw; + } + } else { + link_speed = tb_port_get_link_speed(port); + if (link_speed < 0) + return link_speed; + + link_width = tb_port_get_link_width(port); + if (link_width < 0) + return link_width; + + if (link_width == TB_LINK_WIDTH_ASYM_TX) { + up_bw = link_speed * 1 * 1000; + down_bw = link_speed * 3 * 1000; + } else if (link_width == TB_LINK_WIDTH_ASYM_RX) { + up_bw = link_speed * 3 * 1000; + down_bw = link_speed * 1 * 1000; + } else if (include_asym) { + /* + * The link is symmetric at the moment but we + * can switch it to asymmetric as needed. Report + * this bandwidth as available (even though it + * is not yet enabled). + */ + if (downstream) { up_bw = link_speed * 1 * 1000; down_bw = link_speed * 3 * 1000; - } else if (link_width == TB_LINK_WIDTH_ASYM_RX) { + } else { up_bw = link_speed * 3 * 1000; down_bw = link_speed * 1 * 1000; - } else { - up_bw = link_speed * link_width * 1000; - down_bw = up_bw; } + } else { + up_bw = link_speed * link_width * 1000; + down_bw = up_bw; } + } - /* Leave 10% guard band */ - up_bw -= up_bw / 10; - down_bw -= down_bw / 10; - - tb_port_dbg(port, "link total bandwidth %d/%d Mb/s\n", up_bw, - down_bw); + /* Leave 10% guard band */ + *max_up = up_bw - up_bw / 10; + *max_down = down_bw - down_bw / 10; - /* - * Find all DP tunnels that cross the port and reduce - * their consumed bandwidth from the available. - */ - list_for_each_entry(tunnel, &tcm->tunnel_list, list) { - int dp_consumed_up, dp_consumed_down; + tb_port_dbg(port, "link maximum bandwidth %d/%d Mb/s\n", *max_up, *max_down); + return 0; +} - if (tb_tunnel_is_invalid(tunnel)) - continue; +/** + * tb_available_bandwidth() - Available bandwidth for tunneling + * @tb: Domain structure + * @src_port: Source protocol adapter + * @dst_port: Destination protocol adapter + * @available_up: Available bandwidth upstream (Mb/s) + * @available_down: Available bandwidth downstream (Mb/s) + * @include_asym: Include bandwidth if the link is switched from + * symmetric to asymmetric + * + * Calculates maximum available bandwidth for protocol tunneling between + * @src_port and @dst_port at the moment. This is minimum of maximum + * link bandwidth across all links reduced by currently consumed + * bandwidth on that link. + * + * If @include_asym is true then includes also bandwidth that can be + * added when the links are transitioned into asymmetric (but does not + * transition the links). + * + * Return: %0 on success, negative errno otherwise. + */ +static int tb_available_bandwidth(struct tb *tb, struct tb_port *src_port, + struct tb_port *dst_port, int *available_up, + int *available_down, bool include_asym) +{ + struct tb_port *port; + int ret; - if (!tb_tunnel_is_dp(tunnel)) - continue; + /* Maximum possible bandwidth asymmetric Gen 4 link is 120 Gb/s */ + *available_up = *available_down = 120000; - if (!tb_tunnel_port_on_path(tunnel, port)) - continue; + /* Find the minimum available bandwidth over all links */ + tb_for_each_port_on_path(src_port, dst_port, port) { + int max_up, max_down, consumed_up, consumed_down; - /* - * Ignore the DP tunnel between src_port and - * dst_port because it is the same tunnel and we - * may be re-calculating estimated bandwidth. - */ - if (tunnel->src_port == src_port && - tunnel->dst_port == dst_port) - continue; + if (!tb_port_is_null(port)) + continue; - ret = tb_tunnel_consumed_bandwidth(tunnel, - &dp_consumed_up, - &dp_consumed_down); - if (ret) - return ret; + ret = tb_maximum_bandwidth(tb, src_port, dst_port, port, + &max_up, &max_down, include_asym); + if (ret) + return ret; - up_bw -= dp_consumed_up; - down_bw -= dp_consumed_down; - } + ret = tb_consumed_usb3_pcie_bandwidth(tb, src_port, dst_port, + port, &consumed_up, + &consumed_down); + if (ret) + return ret; + max_up -= consumed_up; + max_down -= consumed_down; - /* - * If USB3 is tunneled from the host router down to the - * branch leading to port we need to take USB3 consumed - * bandwidth into account regardless whether it actually - * crosses the port. - */ - up_bw -= usb3_consumed_up; - down_bw -= usb3_consumed_down; + ret = tb_consumed_dp_bandwidth(tb, src_port, dst_port, port, + &consumed_up, &consumed_down); + if (ret) + return ret; + max_up -= consumed_up; + max_down -= consumed_down; - if (up_bw < *available_up) - *available_up = up_bw; - if (down_bw < *available_down) - *available_down = down_bw; + if (max_up < *available_up) + *available_up = max_up; + if (max_down < *available_down) + *available_down = max_down; } if (*available_up < 0) @@ -729,21 +883,21 @@ static void tb_reclaim_usb3_bandwidth(struct tb *tb, struct tb_port *src_port, if (!tunnel) return; - tb_dbg(tb, "reclaiming unused bandwidth for USB3\n"); + tb_tunnel_dbg(tunnel, "reclaiming unused bandwidth\n"); /* * Calculate available bandwidth for the first hop USB3 tunnel. * That determines the whole USB3 bandwidth for this branch. */ ret = tb_available_bandwidth(tb, tunnel->src_port, tunnel->dst_port, - &available_up, &available_down); + &available_up, &available_down, false); if (ret) { - tb_warn(tb, "failed to calculate available bandwidth\n"); + tb_tunnel_warn(tunnel, "failed to calculate available bandwidth\n"); return; } - tb_dbg(tb, "available bandwidth for USB3 %d/%d Mb/s\n", - available_up, available_down); + tb_tunnel_dbg(tunnel, "available bandwidth %d/%d Mb/s\n", available_up, + available_down); tb_tunnel_reclaim_available_bandwidth(tunnel, &available_up, &available_down); } @@ -794,14 +948,23 @@ static int tb_tunnel_usb3(struct tb *tb, struct tb_switch *sw) return ret; } - ret = tb_available_bandwidth(tb, down, up, &available_up, - &available_down); + ret = tb_available_bandwidth(tb, down, up, &available_up, &available_down, + false); if (ret) goto err_reclaim; tb_port_dbg(up, "available bandwidth for new USB3 tunnel %d/%d Mb/s\n", available_up, available_down); + /* + * If the available bandwidth is less than 1.5 Gb/s notify + * userspace that the connected isochronous device may not work + * properly. + */ + if (available_up < 1500 || available_down < 1500) + tb_tunnel_event(tb, TB_TUNNEL_LOW_BANDWIDTH, TB_TUNNEL_USB3, + down, up); + tunnel = tb_tunnel_alloc_usb3(tb, up, down, available_up, available_down); if (!tunnel) { @@ -823,7 +986,7 @@ static int tb_tunnel_usb3(struct tb *tb, struct tb_switch *sw) return 0; err_free: - tb_tunnel_free(tunnel); + tb_tunnel_put(tunnel); err_reclaim: if (tb_route(parent)) tb_reclaim_usb3_bandwidth(tb, down, up); @@ -856,7 +1019,253 @@ static int tb_create_usb3_tunnels(struct tb_switch *sw) return 0; } -static void tb_scan_port(struct tb_port *port); +/** + * tb_configure_asym() - Transition links to asymmetric if needed + * @tb: Domain structure + * @src_port: Source adapter to start the transition + * @dst_port: Destination adapter + * @requested_up: Additional bandwidth (Mb/s) required upstream + * @requested_down: Additional bandwidth (Mb/s) required downstream + * + * Transition links between @src_port and @dst_port into asymmetric, with + * three lanes in the direction from @src_port towards @dst_port and one lane + * in the opposite direction, if the bandwidth requirements + * (requested + currently consumed) on that link exceed @asym_threshold. + * + * Must be called with available >= requested over all links. + * + * Return: %0 on success, negative errno otherwise. + */ +static int tb_configure_asym(struct tb *tb, struct tb_port *src_port, + struct tb_port *dst_port, int requested_up, + int requested_down) +{ + bool clx = false, clx_disabled = false, downstream; + struct tb_switch *sw; + struct tb_port *up; + int ret = 0; + + if (!asym_threshold) + return 0; + + downstream = tb_port_path_direction_downstream(src_port, dst_port); + /* Pick up router deepest in the hierarchy */ + if (downstream) + sw = dst_port->sw; + else + sw = src_port->sw; + + tb_for_each_upstream_port_on_path(src_port, dst_port, up) { + struct tb_port *down = tb_switch_downstream_port(up->sw); + enum tb_link_width width_up, width_down; + int consumed_up, consumed_down; + + ret = tb_consumed_dp_bandwidth(tb, src_port, dst_port, up, + &consumed_up, &consumed_down); + if (ret) + break; + + if (downstream) { + /* + * Downstream so make sure upstream is within the 36G + * (40G - guard band 10%), and the requested is above + * what the threshold is. + */ + if (consumed_up + requested_up >= TB_ASYM_MIN) { + ret = -ENOBUFS; + break; + } + /* Does consumed + requested exceed the threshold */ + if (consumed_down + requested_down < asym_threshold) + continue; + + width_up = TB_LINK_WIDTH_ASYM_RX; + width_down = TB_LINK_WIDTH_ASYM_TX; + } else { + /* Upstream, the opposite of above */ + if (consumed_down + requested_down >= TB_ASYM_MIN) { + ret = -ENOBUFS; + break; + } + if (consumed_up + requested_up < asym_threshold) + continue; + + width_up = TB_LINK_WIDTH_ASYM_TX; + width_down = TB_LINK_WIDTH_ASYM_RX; + } + + if (up->sw->link_width == width_up) + continue; + + if (!tb_port_width_supported(up, width_up) || + !tb_port_width_supported(down, width_down)) + continue; + + /* + * Disable CL states before doing any transitions. We + * delayed it until now that we know there is a real + * transition taking place. + */ + if (!clx_disabled) { + clx = tb_disable_clx(sw); + clx_disabled = true; + } + + tb_sw_dbg(up->sw, "configuring asymmetric link\n"); + + /* + * Here requested + consumed > threshold so we need to + * transition the link into asymmetric now. + */ + ret = tb_switch_set_link_width(up->sw, width_up); + if (ret) { + tb_sw_warn(up->sw, "failed to set link width\n"); + break; + } + } + + /* Re-enable CL states if they were previosly enabled */ + if (clx) + tb_enable_clx(sw); + + return ret; +} + +/** + * tb_configure_sym() - Transition links to symmetric if possible + * @tb: Domain structure + * @src_port: Source adapter to start the transition + * @dst_port: Destination adapter + * @keep_asym: Keep asymmetric link if preferred + * + * Goes over each link from @src_port to @dst_port and tries to + * transition the link to symmetric if the currently consumed bandwidth + * allows and link asymmetric preference is ignored (if @keep_asym is %false). + * + * Return: %0 on success, negative errno otherwise. + */ +static int tb_configure_sym(struct tb *tb, struct tb_port *src_port, + struct tb_port *dst_port, bool keep_asym) +{ + bool clx = false, clx_disabled = false, downstream; + struct tb_switch *sw; + struct tb_port *up; + int ret = 0; + + if (!asym_threshold) + return 0; + + downstream = tb_port_path_direction_downstream(src_port, dst_port); + /* Pick up router deepest in the hierarchy */ + if (downstream) + sw = dst_port->sw; + else + sw = src_port->sw; + + tb_for_each_upstream_port_on_path(src_port, dst_port, up) { + int consumed_up, consumed_down; + + /* Already symmetric */ + if (up->sw->link_width <= TB_LINK_WIDTH_DUAL) + continue; + /* Unplugged, no need to switch */ + if (up->sw->is_unplugged) + continue; + + ret = tb_consumed_dp_bandwidth(tb, src_port, dst_port, up, + &consumed_up, &consumed_down); + if (ret) + break; + + if (downstream) { + /* + * Downstream so we want the consumed_down < threshold. + * Upstream traffic should be less than 36G (40G + * guard band 10%) as the link was configured asymmetric + * already. + */ + if (consumed_down >= asym_threshold) + continue; + } else { + if (consumed_up >= asym_threshold) + continue; + } + + if (up->sw->link_width == TB_LINK_WIDTH_DUAL) + continue; + + /* + * Here consumed < threshold so we can transition the + * link to symmetric. + * + * However, if the router prefers asymmetric link we + * honor that (unless @keep_asym is %false). + */ + if (keep_asym && + up->sw->preferred_link_width > TB_LINK_WIDTH_DUAL) { + tb_sw_dbg(up->sw, "keeping preferred asymmetric link\n"); + continue; + } + + /* Disable CL states before doing any transitions */ + if (!clx_disabled) { + clx = tb_disable_clx(sw); + clx_disabled = true; + } + + tb_sw_dbg(up->sw, "configuring symmetric link\n"); + + ret = tb_switch_set_link_width(up->sw, TB_LINK_WIDTH_DUAL); + if (ret) { + tb_sw_warn(up->sw, "failed to set link width\n"); + break; + } + } + + /* Re-enable CL states if they were previosly enabled */ + if (clx) + tb_enable_clx(sw); + + return ret; +} + +static void tb_configure_link(struct tb_port *down, struct tb_port *up, + struct tb_switch *sw) +{ + struct tb *tb = sw->tb; + + /* Link the routers using both links if available */ + down->remote = up; + up->remote = down; + if (down->dual_link_port && up->dual_link_port) { + down->dual_link_port->remote = up->dual_link_port; + up->dual_link_port->remote = down->dual_link_port; + } + + /* + * Enable lane bonding if the link is currently two single lane + * links. + */ + if (sw->link_width < TB_LINK_WIDTH_DUAL) + tb_switch_set_link_width(sw, TB_LINK_WIDTH_DUAL); + + /* + * Device router that comes up as symmetric link is + * connected deeper in the hierarchy, we transition the links + * above into symmetric if bandwidth allows. + */ + if (tb_switch_depth(sw) > 1 && + tb_port_get_link_generation(up) >= 4 && + up->sw->link_width == TB_LINK_WIDTH_DUAL) { + struct tb_port *host_port; + + host_port = tb_port_at(tb_route(sw), tb->root_switch); + tb_configure_sym(tb, host_port, up, false); + } + + /* Set the link configured */ + tb_switch_configure_link(sw); +} /* * tb_scan_switch() - scan for and initialize downstream switches @@ -913,12 +1322,16 @@ static void tb_scan_port(struct tb_port *port) goto out_rpm_put; } - tb_retimer_scan(port, true); - sw = tb_switch_alloc(port->sw->tb, &port->sw->dev, tb_downstream_route(port)); if (IS_ERR(sw)) { /* + * Make the downstream retimers available even if there + * is no router connected. + */ + tb_retimer_scan(port, true); + + /* * If there is an error accessing the connected switch * it may be connected to another domain. Also we allow * the other domain to be connected to a max depth switch. @@ -964,19 +1377,17 @@ static void tb_scan_port(struct tb_port *port) goto out_rpm_put; } - /* Link the switches using both links if available */ upstream_port = tb_upstream_port(sw); - port->remote = upstream_port; - upstream_port->remote = port; - if (port->dual_link_port && upstream_port->dual_link_port) { - port->dual_link_port->remote = upstream_port->dual_link_port; - upstream_port->dual_link_port->remote = port->dual_link_port; - } + tb_configure_link(port, upstream_port, sw); + + /* + * Scan for downstream retimers. We only scan them after the + * router has been enumerated to avoid issues with certain + * Pluggable devices that expect the host to enumerate them + * within certain timeout. + */ + tb_retimer_scan(port, true); - /* Enable lane bonding if supported */ - tb_switch_lane_bonding_enable(sw); - /* Set the link configured */ - tb_switch_configure_link(sw); /* * CL0s and CL1 are enabled and supported together. * Silently ignore CLx enabling in case CLx is not supported. @@ -1017,6 +1428,297 @@ out_rpm_put: } } +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_tunnel_direction_downstream(tunnel)) + estimated_bw = estimated_down; + else + estimated_bw = estimated_up; + + /* + * 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"); + } + + 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 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; + + 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); + INIT_DELAYED_WORK(&group->release_work, + tb_bandwidth_group_release_work); + } +} + +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); + + /* 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); + } + } +} + +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; @@ -1040,6 +1742,11 @@ static void tb_deactivate_and_free_tunnel(struct tb_tunnel *tunnel) * deallocated properly. */ tb_switch_dealloc_dp_resource(src_port->sw, src_port); + /* + * If bandwidth on a link is < asym_threshold + * transition the link to symmetric. + */ + 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); @@ -1059,7 +1766,7 @@ static void tb_deactivate_and_free_tunnel(struct tb_tunnel *tunnel) break; } - tb_tunnel_free(tunnel); + tb_tunnel_put(tunnel); } /* @@ -1092,7 +1799,8 @@ static void tb_free_unplugged_children(struct tb_switch *sw) tb_retimer_remove_all(port); tb_remove_dp_resources(port->remote->sw); tb_switch_unconfigure_link(port->remote->sw); - tb_switch_lane_bonding_disable(port->remote->sw); + tb_switch_set_link_width(port->remote->sw, + TB_LINK_WIDTH_SINGLE); tb_switch_remove(port->remote->sw); port->remote = NULL; if (port->dual_link_port) @@ -1152,99 +1860,6 @@ out: 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_port_warn(in, - "failed to release unused bandwidth\n"); - break; - } - } - - out = tunnel->dst_port; - ret = tb_available_bandwidth(tb, in, out, &estimated_up, - &estimated_down); - if (ret) { - tb_port_warn(in, - "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_port_dbg(in, "re-calculated estimated bandwidth %u/%u Mb/s\n", - estimated_up, estimated_down); - - if (in->sw->config.depth < out->sw->config.depth) - estimated_bw = estimated_down; - else - estimated_bw = estimated_up; - - if (usb4_dp_port_set_estimated_bandwidth(in, estimated_bw)) - tb_port_warn(in, "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; @@ -1262,6 +1877,12 @@ static struct tb_port *tb_find_dp_out(struct tb *tb, struct tb_port *in) continue; } + /* Needs to be on different routers */ + if (in->sw == port->sw) { + tb_port_dbg(port, "skipping DP OUT on same router\n"); + continue; + } + tb_port_dbg(port, "DP OUT available\n"); /* @@ -1282,52 +1903,77 @@ static struct tb_port *tb_find_dp_out(struct tb *tb, struct tb_port *in) return NULL; } -static void tb_tunnel_dp(struct tb *tb) +static void tb_dp_tunnel_active(struct tb_tunnel *tunnel, void *data) { - int available_up, available_down, ret, link_nr; - struct tb_cm *tcm = tb_priv(tb); - struct tb_port *port, *in, *out; - struct tb_tunnel *tunnel; - - if (!tb_acpi_may_tunnel_dp()) { - tb_dbg(tb, "DP tunneling disabled, not creating tunnel\n"); - return; - } + struct tb_port *in = tunnel->src_port; + struct tb_port *out = tunnel->dst_port; + struct tb *tb = data; - /* - * 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"); + mutex_lock(&tb->lock); + if (tb_tunnel_is_active(tunnel)) { + int consumed_up, consumed_down, ret; - in = NULL; - out = NULL; - list_for_each_entry(port, &tcm->dp_resources, list) { - if (!tb_port_is_dpin(port)) - continue; + tb_tunnel_dbg(tunnel, "DPRX capabilities read completed\n"); - if (tb_port_is_enabled(port)) { - tb_port_dbg(port, "DP IN in use\n"); - continue; + /* If fail reading tunnel's consumed bandwidth, tear it down */ + ret = tb_tunnel_consumed_bandwidth(tunnel, &consumed_up, + &consumed_down); + if (ret) { + tb_tunnel_warn(tunnel, + "failed to read consumed bandwidth, tearing down\n"); + tb_deactivate_and_free_tunnel(tunnel); + } else { + tb_reclaim_usb3_bandwidth(tb, in, out); + /* + * Transition the links to asymmetric if the + * consumption exceeds the threshold. + */ + tb_configure_asym(tb, in, out, consumed_up, + consumed_down); + /* + * Update the domain with the new bandwidth + * estimation. + */ + tb_recalc_estimated_bandwidth(tb); + /* + * In case DP tunnel exists, change host + * router's 1st children TMU mode to HiFi for + * CL0s to work. + */ + tb_increase_tmu_accuracy(tunnel); } + } else { + struct tb_port *in = tunnel->src_port; - tb_port_dbg(port, "DP IN available\n"); - - out = tb_find_dp_out(tb, port); - if (out) { - in = port; - break; - } + /* + * This tunnel failed to establish. This means DPRX + * negotiation most likely did not complete which + * happens either because there is no graphics driver + * loaded or not all DP cables where connected to the + * discrete router. + * + * In both cases we remove the DP IN adapter from the + * available resources as it is not usable. This will + * also tear down the tunnel and try to re-use the + * released DP OUT. + * + * It will be added back only if there is hotplug for + * the DP IN again. + */ + tb_tunnel_warn(tunnel, "not active, tearing down\n"); + tb_dp_resource_unavailable(tb, in, "DPRX negotiation failed"); } + mutex_unlock(&tb->lock); - if (!in) { - tb_dbg(tb, "no suitable DP IN adapter available, not tunneling\n"); - return; - } - if (!out) { - tb_dbg(tb, "no suitable DP OUT adapter available, not tunneling\n"); - return; - } + tb_domain_put(tb); +} + +static void 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_tunnel *tunnel; /* * This is only applicable to links that are not bonded (so @@ -1369,42 +2015,40 @@ static void tb_tunnel_dp(struct tb *tb) goto err_detach_group; } - ret = tb_available_bandwidth(tb, in, out, &available_up, &available_down); - if (ret) + ret = tb_available_bandwidth(tb, in, out, &available_up, &available_down, + true); + if (ret) { + tb_tunnel_event(tb, TB_TUNNEL_NO_BANDWIDTH, TB_TUNNEL_DP, in, out); goto err_reclaim_usb; + } tb_dbg(tb, "available bandwidth for new DP tunnel %u/%u Mb/s\n", available_up, available_down); tunnel = tb_tunnel_alloc_dp(tb, in, out, link_nr, available_up, - available_down); + available_down, tb_dp_tunnel_active, + tb_domain_get(tb)); if (!tunnel) { tb_port_dbg(out, "could not allocate DP tunnel\n"); goto err_reclaim_usb; } - if (tb_tunnel_activate(tunnel)) { + list_add_tail(&tunnel->list, &tcm->tunnel_list); + + ret = tb_tunnel_activate(tunnel); + if (ret && ret != -EINPROGRESS) { tb_port_info(out, "DP tunnel activation failed, aborting\n"); + list_del(&tunnel->list); goto err_free; } - list_add_tail(&tunnel->list, &tcm->tunnel_list); - tb_reclaim_usb3_bandwidth(tb, in, out); - - /* Update the domain with the new bandwidth estimation */ - tb_recalc_estimated_bandwidth(tb); - - /* - * In case of DP tunnel exists, change host router's 1st children - * TMU mode to HiFi for CL0s to work. - */ - tb_increase_tmu_accuracy(tunnel); return; err_free: - tb_tunnel_free(tunnel); + tb_tunnel_put(tunnel); err_reclaim_usb: tb_reclaim_usb3_bandwidth(tb, in, out); + tb_domain_put(tb); err_detach_group: tb_detach_bandwidth_group(in); err_dealloc_dp: @@ -1416,23 +2060,142 @@ err_rpm_put: pm_runtime_put_autosuspend(&in->sw->dev); } -static void tb_dp_resource_unavailable(struct tb *tb, struct tb_port *port) +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; + } + + /* + * 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_enter_redrive(struct tb_port *port) +{ + struct tb_switch *sw = port->sw; + + if (!(sw->quirks & QUIRK_KEEP_POWER_IN_DP_REDRIVE)) + return; + + /* + * If we get hot-unplug for the DP IN port of the host router + * and the DP resource is not available anymore it means there + * is a monitor connected directly to the Type-C port and we are + * in "redrive" mode. For this to work we cannot enter RTD3 so + * we bump up the runtime PM reference count here. + */ + if (!tb_port_is_dpin(port)) + return; + if (tb_route(sw)) + return; + if (!tb_switch_query_dp_resource(sw, port)) { + port->redrive = true; + pm_runtime_get(&sw->dev); + tb_port_dbg(port, "enter redrive mode, keeping powered\n"); + } +} + +static void tb_exit_redrive(struct tb_port *port) +{ + struct tb_switch *sw = port->sw; + + if (!(sw->quirks & QUIRK_KEEP_POWER_IN_DP_REDRIVE)) + return; + + if (!tb_port_is_dpin(port)) + return; + if (tb_route(sw)) + return; + if (port->redrive && tb_switch_query_dp_resource(sw, port)) { + port->redrive = false; + pm_runtime_put(&sw->dev); + tb_port_dbg(port, "exit redrive mode\n"); + } +} + +static void tb_switch_enter_redrive(struct tb_switch *sw) +{ + struct tb_port *port; + + tb_switch_for_each_port(sw, port) + tb_enter_redrive(port); +} + +/* + * Called during system and runtime suspend to forcefully exit redrive + * mode without querying whether the resource is available. + */ +static void tb_switch_exit_redrive(struct tb_switch *sw) +{ + struct tb_port *port; + + if (!(sw->quirks & QUIRK_KEEP_POWER_IN_DP_REDRIVE)) + return; + + tb_switch_for_each_port(sw, port) { + if (!tb_port_is_dpin(port)) + continue; + + if (port->redrive) { + port->redrive = false; + pm_runtime_put(&sw->dev); + tb_port_dbg(port, "exit redrive mode\n"); + } + } +} + +static void tb_dp_resource_unavailable(struct tb *tb, struct tb_port *port, + const char *reason) { struct tb_port *in, *out; struct tb_tunnel *tunnel; if (tb_port_is_dpin(port)) { - tb_port_dbg(port, "DP IN resource unavailable\n"); + tb_port_dbg(port, "DP IN resource unavailable: %s\n", reason); in = port; out = NULL; } else { - tb_port_dbg(port, "DP OUT resource unavailable\n"); + tb_port_dbg(port, "DP OUT resource unavailable: %s\n", reason); in = NULL; out = port; } tunnel = tb_find_tunnel(tb, TB_TUNNEL_DP, in, out); - tb_deactivate_and_free_tunnel(tunnel); + if (tunnel) + tb_deactivate_and_free_tunnel(tunnel); + else + tb_enter_redrive(port); list_del_init(&port->list); /* @@ -1456,9 +2219,10 @@ static void tb_dp_resource_available(struct tb *tb, struct tb_port *port) return; } - tb_port_dbg(port, "DP %s resource available\n", + tb_port_dbg(port, "DP %s resource available after hotplug\n", tb_port_is_dpin(port) ? "IN" : "OUT"); list_add_tail(&port->list, &tcm->dp_resources); + tb_exit_redrive(port); /* Look for suitable DP IN <-> DP OUT pairs now */ tb_tunnel_dp(tb); @@ -1504,7 +2268,7 @@ static int tb_disconnect_pci(struct tb *tb, struct tb_switch *sw) tb_tunnel_deactivate(tunnel); list_del(&tunnel->list); - tb_tunnel_free(tunnel); + tb_tunnel_put(tunnel); return 0; } @@ -1534,7 +2298,7 @@ static int tb_tunnel_pci(struct tb *tb, struct tb_switch *sw) if (tb_tunnel_activate(tunnel)) { tb_port_info(up, "PCIe tunnel activation failed, aborting\n"); - tb_tunnel_free(tunnel); + tb_tunnel_put(tunnel); return -EIO; } @@ -1593,7 +2357,7 @@ static int tb_approve_xdomain_paths(struct tb *tb, struct tb_xdomain *xd, return 0; err_free: - tb_tunnel_free(tunnel); + tb_tunnel_put(tunnel); err_clx: tb_enable_clx(sw); mutex_unlock(&tb->lock); @@ -1656,7 +2420,7 @@ static int tb_disconnect_xdomain_paths(struct tb *tb, struct tb_xdomain *xd, */ static void tb_handle_hotplug(struct work_struct *work) { - struct tb_hotplug_event *ev = container_of(work, typeof(*ev), work); + struct tb_hotplug_event *ev = container_of(work, typeof(*ev), work.work); struct tb *tb = ev->tb; struct tb_cm *tcm = tb_priv(tb); struct tb_switch *sw; @@ -1701,7 +2465,8 @@ static void tb_handle_hotplug(struct work_struct *work) tb_remove_dp_resources(port->remote->sw); tb_switch_tmu_disable(port->remote->sw); tb_switch_unconfigure_link(port->remote->sw); - tb_switch_lane_bonding_disable(port->remote->sw); + tb_switch_set_link_width(port->remote->sw, + TB_LINK_WIDTH_SINGLE); tb_switch_remove(port->remote->sw); port->remote = NULL; if (port->dual_link_port) @@ -1727,7 +2492,7 @@ static void tb_handle_hotplug(struct work_struct *work) tb_xdomain_put(xd); tb_port_unconfigure_xdomain(port); } else if (tb_port_is_dpout(port) || tb_port_is_dpin(port)) { - tb_dp_resource_unavailable(tb, port); + tb_dp_resource_unavailable(tb, port, "adapter unplug"); } else if (!port->port) { tb_sw_dbg(sw, "xHCI disconnect request\n"); tb_switch_xhci_disconnect(sw); @@ -1771,8 +2536,10 @@ 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) @@ -1781,8 +2548,8 @@ static int tb_alloc_dp_bandwidth(struct tb_tunnel *tunnel, int *requested_up, in = tunnel->src_port; out = tunnel->dst_port; - tb_port_dbg(in, "bandwidth allocated currently %d/%d Mb/s\n", - allocated_up, allocated_down); + tb_tunnel_dbg(tunnel, "bandwidth allocated currently %d/%d Mb/s\n", + allocated_up, allocated_down); /* * If we get rounded up request from graphics side, say HBR2 x 4 @@ -1798,11 +2565,11 @@ 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); @@ -1823,29 +2590,63 @@ static int tb_alloc_dp_bandwidth(struct tb_tunnel *tunnel, int *requested_up, else if (requested_down_corrected < 0) requested_down_corrected = 0; - tb_port_dbg(in, "corrected bandwidth request %d/%d Mb/s\n", - requested_up_corrected, requested_down_corrected); + tb_tunnel_dbg(tunnel, "corrected bandwidth request %d/%d Mb/s\n", + requested_up_corrected, requested_down_corrected); if ((*requested_up >= 0 && requested_up_corrected > max_up_rounded) || (*requested_down >= 0 && requested_down_corrected > max_down_rounded)) { - tb_port_dbg(in, "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; + tb_tunnel_dbg(tunnel, + "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); + ret = -ENOBUFS; + 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 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. - */ - return tb_tunnel_alloc_bandwidth(tunnel, requested_up, - requested_down); + 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_percpu_wq, &group->release_work, + msecs_to_jiffies(TB_RELEASE_BW_TIMEOUT)); + } + } + + ret = tb_tunnel_alloc_bandwidth(tunnel, requested_up, + requested_down); + if (ret) + goto fail; + + return 0; } /* @@ -1854,42 +2655,88 @@ 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 * are also in the same group but we use the same function here * that we use with the normal bandwidth allocation). */ - ret = tb_available_bandwidth(tb, in, out, &available_up, &available_down); + ret = tb_available_bandwidth(tb, in, out, &available_up, &available_down, + true); if (ret) goto reclaim; - tb_port_dbg(in, "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 bandwidth on a link is >= asym_threshold + * transition the link to asymmetric. + */ + ret = tb_configure_asym(tb, in, out, *requested_up, + *requested_down); + if (ret) { + tb_configure_sym(tb, in, out, true); + goto fail; + } - if ((*requested_up >= 0 && available_up >= requested_up_corrected) || - (*requested_down >= 0 && available_down >= requested_down_corrected)) { ret = tb_tunnel_alloc_bandwidth(tunnel, requested_up, requested_down); + if (ret) { + tb_tunnel_warn(tunnel, "failed to allocate bandwidth\n"); + 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; } 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); + tb_tunnel_event(tb, TB_TUNNEL_NO_BANDWIDTH, TB_TUNNEL_DP, in, out); + } + return ret; } static void tb_handle_dp_bandwidth_request(struct work_struct *work) { - struct tb_hotplug_event *ev = container_of(work, typeof(*ev), work); + struct tb_hotplug_event *ev = container_of(work, typeof(*ev), work.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); @@ -1907,37 +2754,53 @@ static void tb_handle_dp_bandwidth_request(struct work_struct *work) in = &sw->ports[ev->port]; if (!tb_port_is_dpin(in)) { tb_port_warn(in, "bandwidth request to non-DP IN adapter\n"); - goto unlock; + goto put_sw; } - tb_port_dbg(in, "handling bandwidth allocation request\n"); + tb_port_dbg(in, "handling bandwidth allocation request, retry %d\n", ev->retry); + + 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"); - goto unlock; + 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; } 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 DPRX capabilities so we + * can offer better bandwidth estimation. + */ + 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 unlock; + } + goto put_sw; } requested_bw = ret; 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 unlock; - } - - out = tunnel->dst_port; - - if (in->sw->config.depth < out->sw->config.depth) { + if (tb_tunnel_direction_downstream(tunnel)) { requested_up = -1; requested_down = requested_bw; } else { @@ -1947,18 +2810,44 @@ static void tb_handle_dp_bandwidth_request(struct work_struct *work) ret = tb_alloc_dp_bandwidth(tunnel, &requested_up, &requested_down); if (ret) { - if (ret == -ENOBUFS) - tb_port_warn(in, "not enough bandwidth available\n"); - else - tb_port_warn(in, "failed to change bandwidth allocation\n"); + if (ret == -ENOBUFS) { + tb_tunnel_warn(tunnel, + "not enough bandwidth available\n"); + } else if (ret == -ENOTCONN) { + tb_tunnel_dbg(tunnel, "not active yet\n"); + /* + * We got bandwidth allocation request but the + * tunnel is not yet active. This means that + * tb_dp_tunnel_active() is not yet called for + * this tunnel. Allow it some time and retry + * this request a couple of times. + */ + if (ev->retry < TB_BW_ALLOC_RETRIES) { + tb_tunnel_dbg(tunnel, + "retrying bandwidth allocation request\n"); + tb_queue_dp_bandwidth_request(tb, ev->route, + ev->port, + ev->retry + 1, + msecs_to_jiffies(50)); + } else { + tb_tunnel_dbg(tunnel, + "run out of retries, failing the request"); + } + } else { + tb_tunnel_warn(tunnel, + "failed to change bandwidth allocation\n"); + } } else { - tb_port_dbg(in, "bandwidth allocation changed to %d/%d Mb/s\n", - requested_up, requested_down); + tb_tunnel_dbg(tunnel, + "bandwidth allocation changed to %d/%d Mb/s\n", + requested_up, requested_down); /* Update other clients about the allocation change */ tb_recalc_estimated_bandwidth(tb); } +put_sw: + tb_switch_put(sw); unlock: mutex_unlock(&tb->lock); @@ -1968,7 +2857,8 @@ unlock: kfree(ev); } -static void tb_queue_dp_bandwidth_request(struct tb *tb, u64 route, u8 port) +static void tb_queue_dp_bandwidth_request(struct tb *tb, u64 route, u8 port, + int retry, unsigned long delay) { struct tb_hotplug_event *ev; @@ -1979,8 +2869,9 @@ static void tb_queue_dp_bandwidth_request(struct tb *tb, u64 route, u8 port) ev->tb = tb; ev->route = route; ev->port = port; - INIT_WORK(&ev->work, tb_handle_dp_bandwidth_request); - queue_work(tb->wq, &ev->work); + ev->retry = retry; + INIT_DELAYED_WORK(&ev->work, tb_handle_dp_bandwidth_request); + queue_delayed_work(tb->wq, &ev->work, delay); } static void tb_handle_notification(struct tb *tb, u64 route, @@ -2000,7 +2891,7 @@ static void tb_handle_notification(struct tb *tb, u64 route, if (tb_cfg_ack_notification(tb->ctl, route, error)) tb_warn(tb, "could not ack notification on %llx\n", route); - tb_queue_dp_bandwidth_request(tb, route, error->port); + tb_queue_dp_bandwidth_request(tb, route, error->port, 0, 0); break; default: @@ -2055,12 +2946,22 @@ static void tb_stop(struct tb *tb) */ if (tb_tunnel_is_dma(tunnel)) tb_tunnel_deactivate(tunnel); - tb_tunnel_free(tunnel); + tb_tunnel_put(tunnel); } tb_switch_remove(tb->root_switch); 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)) { @@ -2082,9 +2983,10 @@ static int tb_scan_finalize_switch(struct device *dev, void *data) return 0; } -static int tb_start(struct tb *tb) +static int tb_start(struct tb *tb, bool reset) { struct tb_cm *tcm = tb_priv(tb); + bool discover = true; int ret; tb->root_switch = tb_switch_alloc(tb, &tb->dev, 0); @@ -2123,12 +3025,28 @@ static int tb_start(struct tb *tb) tb_switch_tmu_configure(tb->root_switch, TB_SWITCH_TMU_MODE_LOWRES); /* Enable TMU if it is off */ tb_switch_tmu_enable(tb->root_switch); - /* Full scan to discover devices added before the driver was loaded. */ - tb_scan_switch(tb->root_switch); - /* Find out tunnels created by the boot firmware */ - tb_discover_tunnels(tb); - /* Add DP resources from the DP tunnels created by the boot firmware */ - tb_discover_dp_resources(tb); + + /* + * Boot firmware might have created tunnels of its own. Since we + * cannot be sure they are usable for us, tear them down and + * reset the ports to handle it as new hotplug for USB4 v1 + * routers (for USB4 v2 and beyond we already do host reset). + */ + if (reset && tb_switch_is_usb4(tb->root_switch)) { + discover = false; + if (usb4_switch_version(tb->root_switch) == 1) + tb_switch_reset(tb->root_switch); + } + + if (discover) { + /* Full scan to discover devices added before the driver was loaded. */ + tb_scan_switch(tb->root_switch); + /* Find out tunnels created by the boot firmware */ + tb_discover_tunnels(tb); + /* Add DP resources from the DP tunnels created by the boot firmware */ + tb_discover_dp_resources(tb); + } + /* * If the boot firmware did not create USB 3.x tunnels create them * now for the whole topology. @@ -2136,6 +3054,7 @@ static int tb_start(struct tb *tb) tb_create_usb3_tunnels(tb->root_switch); /* Add DP IN resources for the root switch */ tb_add_dp_resources(tb->root_switch); + tb_switch_enter_redrive(tb->root_switch); /* Make the discovered switches available to the userspace */ device_for_each_child(&tb->root_switch->dev, NULL, tb_scan_finalize_switch); @@ -2151,6 +3070,7 @@ static int tb_suspend_noirq(struct tb *tb) tb_dbg(tb, "suspending...\n"); tb_disconnect_and_release_dp(tb); + tb_switch_exit_redrive(tb->root_switch); tb_switch_suspend(tb->root_switch, false); tcm->hotplug_active = false; /* signal tb_handle_hotplug to quit */ tb_dbg(tb, "suspend finished\n"); @@ -2179,7 +3099,8 @@ static void tb_restore_children(struct tb_switch *sw) continue; if (port->remote) { - tb_switch_lane_bonding_enable(port->remote->sw); + tb_switch_set_link_width(port->remote->sw, + port->remote->sw->link_width); tb_switch_configure_link(port->remote->sw); tb_restore_children(port->remote->sw); @@ -2198,10 +3119,14 @@ static int tb_resume_noirq(struct tb *tb) tb_dbg(tb, "resuming...\n"); - /* remove any pci devices the firmware might have setup */ - tb_switch_reset(tb->root_switch); + /* + * For non-USB4 hosts (Apple systems) remove any PCIe devices + * the firmware might have setup. + */ + if (!tb_switch_is_usb4(tb->root_switch)) + tb_switch_reset(tb->root_switch); - tb_switch_resume(tb->root_switch); + tb_switch_resume(tb->root_switch, false); tb_free_invalid_tunnels(tb); tb_free_unplugged_children(tb->root_switch); tb_restore_children(tb->root_switch); @@ -2217,7 +3142,7 @@ static int tb_resume_noirq(struct tb *tb) if (tb_tunnel_is_usb3(tunnel)) usb3_delay = 500; tb_tunnel_deactivate(tunnel); - tb_tunnel_free(tunnel); + tb_tunnel_put(tunnel); } /* Re-create our tunnels now */ @@ -2228,7 +3153,7 @@ static int tb_resume_noirq(struct tb *tb) /* Only need to do it once */ usb3_delay = 0; } - tb_tunnel_restart(tunnel); + tb_tunnel_activate(tunnel); } if (!list_empty(&tcm->tunnel_list)) { /* @@ -2238,6 +3163,7 @@ static int tb_resume_noirq(struct tb *tb) tb_dbg(tb, "tunnels restarted, sleeping for 100ms\n"); msleep(100); } + tb_switch_enter_redrive(tb->root_switch); /* Allow tb_handle_hotplug to progress events */ tcm->hotplug_active = true; tb_dbg(tb, "resume finished\n"); @@ -2301,6 +3227,12 @@ static int tb_runtime_suspend(struct tb *tb) struct tb_cm *tcm = tb_priv(tb); mutex_lock(&tb->lock); + /* + * The below call only releases DP resources to allow exiting and + * re-entering redrive mode. + */ + tb_disconnect_and_release_dp(tb); + tb_switch_exit_redrive(tb->root_switch); tb_switch_suspend(tb->root_switch, true); tcm->hotplug_active = false; mutex_unlock(&tb->lock); @@ -2327,11 +3259,12 @@ static int tb_runtime_resume(struct tb *tb) struct tb_tunnel *tunnel, *n; mutex_lock(&tb->lock); - tb_switch_resume(tb->root_switch); + tb_switch_resume(tb->root_switch, true); tb_free_invalid_tunnels(tb); tb_restore_children(tb->root_switch); list_for_each_entry_safe(tunnel, n, &tcm->tunnel_list, list) - tb_tunnel_restart(tunnel); + tb_tunnel_activate(tunnel); + tb_switch_enter_redrive(tb->root_switch); tcm->hotplug_active = true; mutex_unlock(&tb->lock); @@ -2347,6 +3280,7 @@ 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, @@ -2368,12 +3302,13 @@ static const struct tb_cm_ops tb_cm_ops = { * downstream ports and the NHI so that the device core will make sure * NHI is resumed first before the rest. */ -static void tb_apple_add_links(struct tb_nhi *nhi) +static bool tb_apple_add_links(struct tb_nhi *nhi) { struct pci_dev *upstream, *pdev; + bool ret; if (!x86_apple_machine) - return; + return false; switch (nhi->pdev->device) { case PCI_DEVICE_ID_INTEL_LIGHT_RIDGE: @@ -2382,33 +3317,34 @@ static void tb_apple_add_links(struct tb_nhi *nhi) case PCI_DEVICE_ID_INTEL_FALCON_RIDGE_4C_NHI: break; default: - return; + return false; } upstream = pci_upstream_bridge(nhi->pdev); while (upstream) { if (!pci_is_pcie(upstream)) - return; + return false; if (pci_pcie_type(upstream) == PCI_EXP_TYPE_UPSTREAM) break; upstream = pci_upstream_bridge(upstream); } if (!upstream) - return; + return false; /* * For each hotplug downstream port, create add device link * back to NHI so that PCIe tunnels can be re-established after * sleep. */ + ret = false; for_each_pci_bridge(pdev, upstream->subordinate) { const struct device_link *link; if (!pci_is_pcie(pdev)) continue; if (pci_pcie_type(pdev) != PCI_EXP_TYPE_DOWNSTREAM || - !pdev->is_hotplug_bridge) + !pdev->is_pciehp) continue; link = device_link_add(&pdev->dev, &nhi->pdev->dev, @@ -2417,11 +3353,14 @@ static void tb_apple_add_links(struct tb_nhi *nhi) if (link) { dev_dbg(&nhi->pdev->dev, "created link from %s\n", dev_name(&pdev->dev)); + ret = true; } else { dev_warn(&nhi->pdev->dev, "device link creation from %s failed\n", dev_name(&pdev->dev)); } } + + return ret; } struct tb *tb_probe(struct tb_nhi *nhi) @@ -2448,8 +3387,13 @@ struct tb *tb_probe(struct tb_nhi *nhi) tb_dbg(tb, "using software connection manager\n"); - tb_apple_add_links(nhi); - tb_acpi_add_links(nhi); + /* + * Device links are needed to make sure we establish tunnels + * before the PCIe/USB stack is resumed so complain here if we + * found them missing. + */ + if (!tb_apple_add_links(nhi) && !tb_acpi_add_links(nhi)) + tb_warn(tb, "device links to tunneled native ports are missing!\n"); return tb; } |