linux-kernel - Recursive/circular locking in serial8250_console_write/serial8250_do

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 12 Aug 2020 08:48:13 -0700
From:   Guenter Roeck <linux@...ck-us.net>
To:     gregkh@...uxfoundation.org
Cc:     linux-serial@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Recursive/circular locking in
 serial8250_console_write/serial8250_do_startup

Hi,

crbug.com/1114800 reports a hard lockup due to circular locking in the
8250 console driver. This is seen if CONFIG_PROVE_LOCKING is enabled.

Problem is as follows:
- serial8250_do_startup() locks the serial (console) port.
- serial8250_do_startup() then disables interrupts if interrupts are
  shared, by calling disable_irq_nosync().
- disable_irq_nosync() calls __irq_get_desc_lock() to lock the interrupt
  descriptor.
- __irq_get_desc_lock() calls lock_acquire()
- If CONFIG_PROVE_LOCKING is enabled, validate_chain() and check_noncircular()
  are called and identify a potential locking error.
- This locking error is reported via printk, which ultimately calls
  serial8250_console_write().
- serial8250_console_write() tries to lock the serial console port.
  Since it is already locked, the system hangs and ultimately reports
  a hard lockup.

I understand we'll need to figure out and fix what lockdep complains about,
and I am working on that. However, even if that is fixed, we'll need a
solution for the recursive lock: Fixing the lockdep problem doesn't
guarantee that a similar problem (or some other log message) won't be
detected and reported sometime in the future while serial8250_do_startup()
holds the console port lock.

Ideas, anyone ? Everything I came up with so far seems clumsy and hackish.

Thanks,
Guenter