Re: [PATCH] usb: typec: ucsi: Fix NULL pointer dereference in power level debugfs
From: Bertrand Jacquin
Date: Sat Nov 21 2026 - 11:08:25 EST
Hi Greg,
On Friday, November 20 2026 at 13:29:13 +0100, Greg KH wrote:
> On Sat, Oct 10, 2026 at 01:03:57PM +0100, Bertrand Jacquin wrote:
> > ucsi_register() creates the debugfs entries right away, but
> > ucsi->connector is only set by ucsi_init(), which runs later from
> > ucsi->work and only publishes the connectors once the PPM has been
> > reset and all ports are registered. Until then, or for good if init
> > fails, ucsi->connector is NULL, yet the peak_current, avg_current and
> > vbus_voltage show callbacks dereference it unconditionally.
> >
> > On an Intel NUC13ANKi5 whose PPM never answers when CONFIG_ACPI_EC=n,
> > reading those files after the failed init oopses:
> >
> > ucsi_acpi USBC000:00: error -ETIMEDOUT: PPM init failed
> > BUG: kernel NULL pointer dereference, address: 000000000000045c
> > #PF: supervisor read access in kernel mode
> > RIP: 0010:ucsi_debugfs_register+0x291/0x420 [typec_ucsi]
> > Call Trace:
> > <TASK>
> > seq_read_iter+0x1e4/0x4c0
> > seq_read+0x12b/0x160
> > full_proxy_read+0x9a/0x1f0
> > vfs_read+0xba/0x320
> > ksys_read+0x74/0x110
> > do_syscall_64+0x79/0x240
> > entry_SYSCALL_64_after_hwframe+0x71/0x79
> >
> > Return -ENODEV from those callbacks while no connector is available.
> >
> > Fixes: c851b71fd6cd ("usb: typec: ucsi: Add support for READ_POWER_LEVEL command")
> > Cc: stable@xxxxxxxxxxxxxxx
> > Signed-off-by: Bertrand Jacquin <bertrand@xxxxxxxxxxx>
> > ---
> > drivers/usb/typec/ucsi/debugfs.c | 9 +++++++++
> > 1 file changed, 9 insertions(+)
> >
> > diff --git a/drivers/usb/typec/ucsi/debugfs.c b/drivers/usb/typec/ucsi/debugfs.c
> > index 77a0dd75edd3..911a151ac16a 100644
> > --- a/drivers/usb/typec/ucsi/debugfs.c
> > +++ b/drivers/usb/typec/ucsi/debugfs.c
> > @@ -92,6 +92,9 @@ static int ucsi_peak_curr_show(struct seq_file *m, void *v)
> > {
> > struct ucsi *ucsi = m->private;
> >
> > + if (!ucsi->connector)
> > + return -ENODEV;
>
> What keeps connector from changing to NULL right after you read from it?
ucsi->connector never transitions back to NULL. only ucsi_init() writes to it
through ucsi->work where it's assigned only once, after all ports are
registered. Every failure path before that assignment leaves it NULL and never
cleared afterwards.
The only end of its lifetime is ucsi_unregister(), which calls
ucsi_debugfs_unregister() before kfree(ucsi->connector).
debugfs_remove_recursive() waits for in-flight readers to drain (the
full_proxy fops hold a debugfs_file_get() reference and
__debugfs_file_removed() waits on active_users_drained). Any read
after that fails with -EIO. So no show callback can run once the
array is freed.
A reader can therefore only see NULL (init still pending, or failed
for good, as in the reported oops) or a valid array, and the check
handles the NULL case.
Cheers,
--
Bertrand
Attachment:
signature.asc
Description: PGP signature