* [PATCH] thunderbolt: Re-add host router DP IN adapters after DPRX failure
@ 2026-11-21 7:08 Alfonso Villalobos Ulate
2026-11-21 7:25 ` sashiko-bot
0 siblings, 1 reply; 2+ messages in thread
From: Alfonso Villalobos Ulate @ 2026-11-21 7:08 UTC (permalink / raw)
To: westeri
Cc: andreas.noever, YehezkelShB, linux-usb, Alfonso Villalobos Ulate,
stable
When DPRX negotiation fails, the DP IN adapter is removed from the DP
resources and is only added back on a hotplug event for that adapter.
Host router DP IN adapters never get such events, so after one failure
(for example when the GPU feeding it is runtime suspended) the adapter
stays unused until the driver is reloaded.
Re-check the host router DP IN adapters when a DP OUT adapter appears
or a device router is plugged in, and skip adapters already on the
list. Keep host DP IN adapters in adapter order so a re-added one is
paired with the same DP OUT as before, instead of swapping monitors
with the other host DP IN.
Tested on an ASM4242 host with a CalDigit TS5 Plus dock.
Written with assistance from Claude Code (Claude Opus 5.5), which
proposed the code changes and drafted this message.
Fixes: d6d458d42e1e ("thunderbolt: Handle DisplayPort tunnel activation asynchronously")
Cc: stable@vger.kernel.org
Signed-off-by: Alfonso Villalobos Ulate <alfonsocvu@gmail.com>
---
drivers/thunderbolt/tb.c | 52 +++++++++++++++++++++++++++++++++++++---
1 file changed, 49 insertions(+), 3 deletions(-)
diff --git a/drivers/thunderbolt/tb.c b/drivers/thunderbolt/tb.c
index 57c11ba13459..4a43da6817a2 100644
--- a/drivers/thunderbolt/tb.c
+++ b/drivers/thunderbolt/tb.c
@@ -109,6 +109,29 @@ static void tb_queue_hotplug(struct tb *tb, u64 route, u8 port, bool unplug)
/* enumeration & hot plug handling */
+/*
+ * Insert a host router DP IN adapter so that host DP IN adapters stay
+ * in adapter number order. A DP IN that is re-added after it was
+ * dropped (for example because DPRX negotiation failed) then gets
+ * paired with the same DP OUT as before, instead of swapping DP OUT
+ * adapters (and therefore monitors) with the other host DP IN.
+ */
+static void tb_add_host_dp_in(struct tb_cm *tcm, struct tb_port *port)
+{
+ struct tb_port *p;
+
+ list_for_each_entry(p, &tcm->dp_resources, list) {
+ if (tb_port_is_dpin(p) && !tb_route(p->sw) &&
+ p->port > port->port) {
+ /* Insert before the first higher numbered one */
+ list_add_tail(&port->list, &p->list);
+ return;
+ }
+ }
+
+ list_add_tail(&port->list, &tcm->dp_resources);
+}
+
static void tb_add_dp_resources(struct tb_switch *sw)
{
struct tb_cm *tcm = tb_priv(sw->tb);
@@ -118,6 +141,9 @@ static void tb_add_dp_resources(struct tb_switch *sw)
if (!tb_port_is_dpin(port))
continue;
+ if (!list_empty(&port->list))
+ continue;
+
if (!tb_switch_query_dp_resource(sw, port))
continue;
@@ -130,7 +156,7 @@ static void tb_add_dp_resources(struct tb_switch *sw)
if (tb_route(sw))
list_add(&port->list, &tcm->dp_resources);
else
- list_add_tail(&port->list, &tcm->dp_resources);
+ tb_add_host_dp_in(tcm, port);
tb_port_dbg(port, "DP IN resource available\n");
}
@@ -1421,6 +1447,16 @@ static void tb_scan_port(struct tb_port *port)
tb_sw_warn(sw, "USB3 tunnel creation failed\n");
tb_add_dp_resources(sw);
+
+ /*
+ * Host router DP IN adapters may have been dropped earlier
+ * (for example because DPRX negotiation failed) and they do
+ * not get hotplug events of their own, so re-check them now
+ * that there is potentially a new DP OUT to pair with.
+ */
+ if (tcm->hotplug_active)
+ tb_add_dp_resources(sw->tb->root_switch);
+
tb_scan_switch(sw);
out_rpm_put:
@@ -1971,8 +2007,10 @@ static void tb_dp_tunnel_active(struct tb_tunnel *tunnel, void *data)
* 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.
+ * It will be added back if there is hotplug for the
+ * DP IN again. Host router DP IN adapters do not get
+ * such events, so they are re-checked whenever a new
+ * DP OUT or device router appears instead.
*/
tb_tunnel_warn(tunnel, "not active, tearing down\n");
tb_dp_resource_unavailable(tb, in, "DPRX negotiation failed");
@@ -2234,6 +2272,14 @@ static void tb_dp_resource_available(struct tb *tb, struct tb_port *port)
list_add_tail(&port->list, &tcm->dp_resources);
tb_exit_redrive(port);
+ /*
+ * Host router DP IN adapters removed because DPRX negotiation
+ * failed never get a hotplug event of their own. Give them
+ * another chance now that a DP OUT became available.
+ */
+ if (tb_port_is_dpout(port))
+ tb_add_dp_resources(tb->root_switch);
+
/* Look for suitable DP IN <-> DP OUT pairs now */
tb_tunnel_dp(tb);
}
--
2.53.0
^ permalink raw reply related [flat|nested] 2+ messages in thread
* Re: [PATCH] thunderbolt: Re-add host router DP IN adapters after DPRX failure
2026-11-21 7:08 [PATCH] thunderbolt: Re-add host router DP IN adapters after DPRX failure Alfonso Villalobos Ulate
@ 2026-11-21 7:25 ` sashiko-bot
0 siblings, 0 replies; 2+ messages in thread
From: sashiko-bot @ 2026-11-21 7:25 UTC (permalink / raw)
To: Alfonso Villalobos Ulate; +Cc: linux-usb
Thank you for your contribution! Sashiko AI review found 2 potential
issues to consider:
- [Medium] thunderbolt: missing tb_exit_redrive() in
tb_add_dp_resources() leaks RPM ref
- [Medium] thunderbolt: list_add_tail() in tb_dp_resource_available()
breaks DP IN order
--
commit 4eedd8886dc180b45204d935f18725051dd0e6a0
Author: Alfonso Villalobos Ulate <alfonsocvu@gmail.com>
thunderbolt: Re-add host router DP IN adapters after DPRX failure
This commit re-checks host router DP IN adapters when a new DP OUT adapter
becomes available or a device router is scanned, keeping host DP IN
adapters ordered by adapter number on tcm->dp_resources.
> diff --git a/drivers/thunderbolt/tb.c b/drivers/thunderbolt/tb.c
> index 418ad0a61a5de..1f3dea836067e 100644
> --- a/drivers/thunderbolt/tb.c
> +++ b/drivers/thunderbolt/tb.c
> @@ -109,6 +109,29 @@ static void tb_queue_hotplug(struct tb *tb, u64 route, u8 port, bool unplug)
>
> /* enumeration & hot plug handling */
>
> +/*
> + * Insert a host router DP IN adapter so that host DP IN adapters stay
> + * in adapter number order. A DP IN that is re-added after it was
> + * dropped (for example because DPRX negotiation failed) then gets
> + * paired with the same DP OUT as before, instead of swapping DP OUT
> + * adapters (and therefore monitors) with the other host DP IN.
> + */
> +static void tb_add_host_dp_in(struct tb_cm *tcm, struct tb_port *port)
> +{
> + struct tb_port *p;
> +
> + list_for_each_entry(p, &tcm->dp_resources, list) {
> + if (tb_port_is_dpin(p) && !tb_route(p->sw) &&
> + p->port > port->port) {
> + /* Insert before the first higher numbered one */
> + list_add_tail(&port->list, &p->list);
> + return;
> + }
> + }
> +
> + list_add_tail(&port->list, &tcm->dp_resources);
> +}
> +
> static void tb_add_dp_resources(struct tb_switch *sw)
> {
> struct tb_cm *tcm = tb_priv(sw->tb);
> @@ -118,6 +141,9 @@ static void tb_add_dp_resources(struct tb_switch *sw)
> if (!tb_port_is_dpin(port))
> continue;
>
> + if (!list_empty(&port->list))
> + continue;
> +
> if (!tb_switch_query_dp_resource(sw, port))
> continue;
>
> @@ -130,7 +156,7 @@ static void tb_add_dp_resources(struct tb_switch *sw)
> if (tb_route(sw))
> list_add(&port->list, &tcm->dp_resources);
> else
> - list_add_tail(&port->list, &tcm->dp_resources);
> + tb_add_host_dp_in(tcm, port);
>
> tb_port_dbg(port, "DP IN resource available\n");
> }
[Severity: Medium]
Does re-adding a host router DP IN port through tb_add_dp_resources() leak
the runtime PM reference on sw->dev when the port was in redrive mode?
On host routers with QUIRK_KEEP_POWER_IN_DP_REDRIVE, tb_enter_redrive()
sets port->redrive and calls pm_runtime_get(&sw->dev). When
tb_add_dp_resources(tb->root_switch) now runs at runtime from
tb_scan_port() or tb_dp_resource_available(), it adds the host DP IN port
to tcm->dp_resources without calling tb_exit_redrive(port). Any later
tb_dp_resource_available() call for that DP IN port then returns early
because the port is already on tcm->dp_resources (or already enabled by
tb_tunnel_dp()), leaving port->redrive set and preventing the host router
from suspending.
[ ... ]
> @@ -2234,6 +2272,14 @@ static void tb_dp_resource_available(struct tb *tb, struct tb_port *port)
> list_add_tail(&port->list, &tcm->dp_resources);
> tb_exit_redrive(port);
>
> + /*
> + * Host router DP IN adapters removed because DPRX negotiation
> + * failed never get a hotplug event of their own. Give them
> + * another chance now that a DP OUT became available.
> + */
> + if (tb_port_is_dpout(port))
> + tb_add_dp_resources(tb->root_switch);
> +
> /* Look for suitable DP IN <-> DP OUT pairs now */
> tb_tunnel_dp(tb);
> }
[Severity: Medium]
Should tb_dp_resource_available() also use tb_add_host_dp_in() when adding
a host router DP IN port?
When a lower-numbered host DP IN port becomes available via hotplug in
tb_dp_resource_available() (such as when exiting redrive mode or after
resume) while a higher-numbered host DP IN port is already on
tcm->dp_resources, list_add_tail() places the lower-numbered port after
the higher-numbered one, breaking the adapter ordering that
tb_add_host_dp_in() maintains.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261121070804.26185-1-alfonsocvu@gmail.com?part=1
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2026-11-21 7:25 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-21 7:08 [PATCH] thunderbolt: Re-add host router DP IN adapters after DPRX failure Alfonso Villalobos Ulate
2026-11-21 7:25 ` sashiko-bot
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox