Re: [PATCH 1/1] tty: serial: owl: Initialize lock before registering port

[Cristian Ciocaltea <cristian.ciocaltea@gmail.com>]

On Fri, Jul 10, 2020 at 03:35:42PM +0200, Greg Kroah-Hartman wrote:
> On Fri, Jul 10, 2020 at 03:05:49PM +0300, Cristian Ciocaltea wrote:
> > On Fri, May 29, 2020 at 01:34:19PM +0200, Greg Kroah-Hartman wrote:
> > > On Fri, May 29, 2020 at 02:06:47PM +0300, Cristian Ciocaltea wrote:
> > > > Running a lockdep-enabled kernel leads to the following splat when
> > > > probing the owl-uart driver:
> > > > 
> > > > [    1.271784] b0124000.serial: ttyOWL2 at MMIO 0xb0124000 (irq = 22, base_baud = 1500000) is a owl-uart
> > > > [    1.281226] INFO: trying to register non-static key.
> > > > [    1.286179] the code is fine but needs lockdep annotation.
> > > > [    1.291648] turning off the locking correctness validator.
> > > > [    1.297125] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.7.0-rc7+ #70
> > > > [    1.303462] Hardware name: Generic DT based system
> > > > [    1.308268] [<80111d3c>] (unwind_backtrace) from [<8010c9b8>] (show_stack+0x10/0x14)
> > > > [    1.316003] [<8010c9b8>] (show_stack) from [<805016b4>] (dump_stack+0xb4/0xe0)
> > > > [    1.323218] [<805016b4>] (dump_stack) from [<80182dec>] (register_lock_class+0x25c/0x8f4)
> > > > [    1.331391] [<80182dec>] (register_lock_class) from [<8017ee34>] (__lock_acquire+0xb4/0x2ee4)
> > > > [    1.339901] [<8017ee34>] (__lock_acquire) from [<8018291c>] (lock_acquire+0x424/0x4c8)
> > > > [    1.347813] [<8018291c>] (lock_acquire) from [<808597b0>] (_raw_spin_lock_irqsave+0x54/0x68)
> > > > [    1.356242] [<808597b0>] (_raw_spin_lock_irqsave) from [<80582e94>] (uart_add_one_port+0x384/0x510)
> > > > [    1.365276] [<80582e94>] (uart_add_one_port) from [<8058b4d0>] (owl_uart_probe+0x1bc/0x248)
> > > > [    1.373621] [<8058b4d0>] (owl_uart_probe) from [<8059c0e4>] (platform_drv_probe+0x48/0x94)
> > > > [    1.381874] [<8059c0e4>] (platform_drv_probe) from [<805997c4>] (really_probe+0x200/0x470)
> > > > [    1.390123] [<805997c4>] (really_probe) from [<80599c80>] (driver_probe_device+0x16c/0x1bc)
> > > > [    1.398461] [<80599c80>] (driver_probe_device) from [<80599f18>] (device_driver_attach+0x44/0x60)
> > > > [    1.407317] [<80599f18>] (device_driver_attach) from [<8059a078>] (__driver_attach+0x144/0x14c)
> > > > [    1.416000] [<8059a078>] (__driver_attach) from [<805978ac>] (bus_for_each_dev+0x5c/0xb4)
> > > > [    1.424162] [<805978ac>] (bus_for_each_dev) from [<8059896c>] (bus_add_driver+0x118/0x204)
> > > > [    1.432410] [<8059896c>] (bus_add_driver) from [<8059ae6c>] (driver_register+0xbc/0xf8)
> > > > [    1.440405] [<8059ae6c>] (driver_register) from [<80c1fd24>] (owl_uart_init+0x20/0x40)
> > > > [    1.448313] [<80c1fd24>] (owl_uart_init) from [<801021d8>] (do_one_initcall+0x184/0x3a4)
> > > > [    1.456399] [<801021d8>] (do_one_initcall) from [<80c01100>] (kernel_init_freeable+0x264/0x2e4)
> > > > [    1.465085] [<80c01100>] (kernel_init_freeable) from [<80850a88>] (kernel_init+0x8/0x110)
> > > > [    1.473249] [<80850a88>] (kernel_init) from [<80100114>] (ret_from_fork+0x14/0x20)
> > > > [    1.480800] Exception stack(0xee8bdfb0 to 0xee8bdff8)
> > > > [    1.485841] dfa0:                                     00000000 00000000 00000000 00000000
> > > > [    1.494002] dfc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
> > > > [    1.502162] dfe0: 00000000 00000000 00000000 00000000 00000013 00000000
> > > > [    1.508914] printk: console [ttyOWL2] enabled
> > > > 
> > > > The locking issue occurs in uart_configure_port() when trying to
> > > > guard the call to set_mctrl().
> > > > 
> > > > uart_add_one_port() should normally initialize the spinlock via
> > > > uart_port_spin_lock_init(), but it never happens because the port is
> > > > detected as a console and, as a consequence, the spinlock is expected
> > > > to be already initialized.
> > > > 
> > > > The commit a3cb39d258ef
> > > > ("serial: core: Allow detach and attach serial device for console")
> > > > changed the lock initialization logic to assume the spinlock is
> > > > initialized even if the console is not enabled.
> > > > 
> > > > Therefore, initialize the lock explicitly in owl_uart_probe(), before
> > > > attempting to invoke uart_add_one_port().
> > > > 
> > > > Fixes: a3cb39d258ef ("serial: core: Allow detach and attach serial device for console")
> > > > Signed-off-by: Cristian Ciocaltea <cristian.ciocaltea@gmail.com>
> > > > ---
> > > >  drivers/tty/serial/owl-uart.c | 2 ++
> > > >  1 file changed, 2 insertions(+)
> > > > 
> > > > diff --git a/drivers/tty/serial/owl-uart.c b/drivers/tty/serial/owl-uart.c
> > > > index c149f8c30007..c2fa2f15d50a 100644
> > > > --- a/drivers/tty/serial/owl-uart.c
> > > > +++ b/drivers/tty/serial/owl-uart.c
> > > > @@ -705,6 +705,8 @@ static int owl_uart_probe(struct platform_device *pdev)
> > > >  	owl_uart_ports[pdev->id] = owl_port;
> > > >  	platform_set_drvdata(pdev, owl_port);
> > > >  
> > > > +	spin_lock_init(&owl_port->port.lock);
> > > > +
> > > >  	ret = uart_add_one_port(&owl_uart_driver, &owl_port->port);
> > > >  	if (ret)
> > > >  		owl_uart_ports[pdev->id] = NULL;
> > > 
> > > Ugh, another one :(
> > > 
> > > Thanks for this, will queue this up now.
> > > 
> > > greg k-h
> > 
> > Hi Greg,
> > 
> > I think this was mistakenly omitted from the inclusion in v5.8, or
> > has it been postponed for v5.9?
> 
> I can't find it anywhere :(
> 
> But I think we now have a core fix for this, no need to do this in any
> individual driver, right?  Can you test my tty-linus branch to see if
> your change is still needed or not?

I was not aware of the alternative approach, at that time Andy
confirmed the expected behaviour is to have a proper initialization
of the spinlock in the driver:
https://lore.kernel.org/lkml/CAHp75Vcz0a87LSnb6Ubt5_bSF3wUcs21GbP119trXER5KBDxbQ@mail.gmail.com/
https://lore.kernel.org/lkml/CAHp75Vf8xJ2yX-11JsTDnRvZQOK+16aePcB1AUzq=5oO-mFCGQ@mail.gmail.com/

I have just checked the tty-linus tree and the only related change
seems to be f743061a85f5
("serial: core: Initialise spin lock before use in uart_configure_port()")

I tested it on top of 5.8.0-rc4 and I confirm it fixes the splat,
however I'm a bit confused now regarding the recommended approach since
there is at least one more commit in the tty-linux tree that still 
handles a similar issue in the driver specific code:
f38278e9b810 ("serial: sh-sci: Initialize spinlock for uart console")

Thanks,
Cristi

> 
> thanks,
> 
> greg k-h

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel