mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH] usb: cdns3: fix NULL dereference in cdns_host_resume()
@ 2026-11-18 12:08 Guomin Chen
  0 siblings, 0 replies; only message in thread
From: Guomin Chen @ 2026-11-18 12:08 UTC (permalink / raw)
  To: Peter Chen, Pawel Laszczak, Greg Kroah-Hartman
  Cc: Roger Quadros, linux-usb, linux-kernel, Guoming Chen

From: Guoming Chen <chenguoming88s@gmail.com>

cdns_host_resume() assumes that the xhci-hcd child device is bound and
passes platform_get_drvdata() straight to hcd_to_xhci_priv().  That is
not guaranteed: __cdns_host_init() creates the child with
platform_device_add(), but the child is only probed once xhci-plat
registers.  cdns3/ is linked before host/ in drivers/usb/Makefile and
both are module_init(), so at boot the child stays unbound until
xhci-plat's initcall runs; when xhci-plat is a module it stays unbound
until the module is loaded, and it is retried even later when its probe
returns -EPROBE_DEFER.  Platforms that set
CDNS3_DEFAULT_PM_RUNTIME_ALLOW (cdns3-imx is the only in-tree user) let
the controller autosuspend (20 ms), so the parent may already be
runtime-suspended by the time the child's probe actually runs.

When the child is finally probed, driver core resumes its parent before
calling really_probe() (pm_runtime_get_sync(dev->parent) in
__driver_probe_device()).  That runs cdns_resume() ->
cdns_host_resume() while platform_get_drvdata() still returns NULL,
because the drvdata is only set later, by dev_set_drvdata() in
__usb_create_hcd() called from xhci_plat_probe().  hcd_to_xhci_priv(NULL)
then evaluates usb_hcd_is_primary_hcd(NULL), which dereferences
NULL + offsetof(struct usb_hcd, primary_hcd) (the offset depends on the
configuration).  The oops faults in the first instruction of that
helper:

  pc : usb_hcd_is_primary_hcd+0x0/0x1c
  lr : cdns_host_resume+0x24/0x44

Return early when the host child is not available: it has either not
been created yet or not been probed.  cdns->host_dev is also cleared by
cdns_host_exit() while cdns_role_stop() leaves cdns->role pointing at
the host role, so a resume after a role stop reaches this function as
well.  Nothing is lost in either case: the child's probe initializes the
controller from scratch, and xhci_plat_priv.power_lost is only consumed
when an already initialized host resumes.

Fixes: 3a85c1011540 ("usb: host: cdns3: forward lost power information to xhci")
Signed-off-by: Guoming Chen <chenguoming88s@gmail.com>
---
 drivers/usb/cdns3/host.c | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/drivers/usb/cdns3/host.c b/drivers/usb/cdns3/host.c
index e01264dc625d..3456f065111b 100644
--- a/drivers/usb/cdns3/host.c
+++ b/drivers/usb/cdns3/host.c
@@ -140,9 +140,21 @@ static void cdns_host_exit(struct cdns *cdns)
 
 static int cdns_host_resume(struct cdns *cdns, bool power_lost)
 {
-	struct usb_hcd *hcd = platform_get_drvdata(cdns->host_dev);
-	struct xhci_plat_priv *priv = hcd_to_xhci_priv(hcd);
+	struct usb_hcd *hcd;
+	struct xhci_plat_priv *priv;
+
+	/*
+	 * The xhci-hcd child may not be bound yet (xhci-plat can probe it
+	 * after cdns3 created it), or may already be gone after a role stop.
+	 */
+	if (!cdns->host_dev)
+		return 0;
+
+	hcd = platform_get_drvdata(cdns->host_dev);
+	if (!hcd)
+		return 0;
 
+	priv = hcd_to_xhci_priv(hcd);
 	priv->power_lost = power_lost;
 
 	return 0;
-- 
2.43.0

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2026-11-18 12:08 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-18 12:08 [PATCH] usb: cdns3: fix NULL dereference in cdns_host_resume() Guomin Chen

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®