Re: [PATCH] serial: sh-sci: Initialize spinlock for uart console

From: Lad, Prabhakar
Date: Thu Jul 02 2020 - 14:32:44 EST


Hi Geert,

On Thu, Jul 2, 2020 at 3:39 PM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
>
> Hi Prabhakar,
>
> On Thu, Jul 2, 2020 at 4:05 PM Lad, Prabhakar
> <prabhakar.csengg@xxxxxxxxx> wrote:
>
> > Hi Geert,
> >
> > On Thu, Jul 2, 2020 at 1:52 PM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > >
> > > Hi Prabhakar,
> > >
> > > On Thu, Jul 2, 2020 at 1:42 PM Lad, Prabhakar
> > > <prabhakar.csengg@xxxxxxxxx> wrote:
> > > > On Thu, Jul 2, 2020 at 11:49 AM Lad, Prabhakar
> > > > <prabhakar.csengg@xxxxxxxxx> wrote:
> > > > > On Thu, Jul 2, 2020 at 10:23 AM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > > > > > On Wed, Jul 1, 2020 at 7:28 PM Lad, Prabhakar
> > > > > > <prabhakar.csengg@xxxxxxxxx> wrote:
> > > > > > > On Wed, Jul 1, 2020 at 6:17 PM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > > > > > > > On Wed, Jul 1, 2020 at 5:42 PM Lad Prabhakar
> > > > > > > > <prabhakar.mahadev-lad.rj@xxxxxxxxxxxxxx> wrote:
> > > > > > > > > serial core expects the spinlock to be initialized by the controller
> > > > > > > > > driver for serial console, this patch makes sure the spinlock is
> > > > > > > > > initialized, fixing the below issue:
> > > > > > > > >
> > > > > > > > > [ 0.865928] BUG: spinlock bad magic on CPU#0, swapper/0/1
> > > > > > > > > [ 0.865945] lock: sci_ports+0x0/0x4c80, .magic: 00000000, .owner: <none>/-1, .owner_cpu: 0
> > > > > > > > > [ 0.865955] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.8.0-rc1+ #112
> > > > > > > > > [ 0.865961] Hardware name: HopeRun HiHope RZ/G2H with sub board (DT)
> > > > > > > > > [ 0.865968] Call trace:
> > > > > > > > > [ 0.865979] dump_backtrace+0x0/0x1d8
> > > > > > > > > [ 0.865985] show_stack+0x14/0x20
> > > > > > > > > [ 0.865996] dump_stack+0xe8/0x130
> > > > > > > > > [ 0.866006] spin_dump+0x6c/0x88
> > > > > > > > > [ 0.866012] do_raw_spin_lock+0xb0/0xf8
> > > > > > > > > [ 0.866023] _raw_spin_lock_irqsave+0x80/0xa0
> > > > > > > > > [ 0.866032] uart_add_one_port+0x3a4/0x4e0
> > > > > > > > > [ 0.866039] sci_probe+0x504/0x7c8
> > > > > > > > > [ 0.866048] platform_drv_probe+0x50/0xa0
> > > > > > > > > [ 0.866059] really_probe+0xdc/0x330
> > > > > > > > > [ 0.866066] driver_probe_device+0x58/0xb8
> > > > > > > > > [ 0.866072] device_driver_attach+0x6c/0x90
> > > > > > > > > [ 0.866078] __driver_attach+0x88/0xd0
> > > > > > > > > [ 0.866085] bus_for_each_dev+0x74/0xc8
> > > > > > > > > [ 0.866091] driver_attach+0x20/0x28
> > > > > > > > > [ 0.866098] bus_add_driver+0x14c/0x1f8
> > > > > > > > > [ 0.866104] driver_register+0x60/0x110
> > > > > > > > > [ 0.866109] __platform_driver_register+0x40/0x48
> > > > > > > > > [ 0.866119] sci_init+0x2c/0x34
> > > > > > > > > [ 0.866127] do_one_initcall+0x88/0x428
> > > > > > > > > [ 0.866137] kernel_init_freeable+0x2c0/0x328
> > > > > > > > > [ 0.866143] kernel_init+0x10/0x108
> > > > > > > > > [ 0.866150] ret_from_fork+0x10/0x18
> > > > > > > >
> > > > > > > > Interesting...
> > > > > > > >
> > > > > > > > How can I reproduce that? I do have CONFIG_DEBUG_SPINLOCK=y.
> > > > > > > > I'm wondering why haven't we seen this before...
> > > > > > > >
> > > > > > > I have attached .config for your reference.
> > > > > >
> > > > > > Thank you!
> > > > > >
> > > > > > I gave it a try with v5.8-rc1 on Salvator-XS with R-Car H3 ES2.0.
> > > > > > However, I couldn't reproduce the issue.
> > > > > > Does it happen on that specific board only? Is this serdev-related?
> > > > > > Note that I had to disable CONFIG_EXTRA_FIRMWARE, as I don't have the
> > > > > > firmware blobs it referenced. Do I need them to trigger the issue?
> > > > > > As the .config has a few non-upstream options, do you have any patches
> > > > > > applied that might impact the issue?
> > > > > >
> > > > > Can't think of any patches that might cause an issue, most of it are
> > > > > just the DT's and config additions. Nor do firmware blobs should
> > > > > affect it. I'll try and reproduce it on M3N and get back to you.
> > > > >
> > > > I did manage to replicate this issue on M3N (v5.8-rc3 tag with no
> > > > modifications), I have attached the config file and also the boot log
> > > > without this patch for your reference, after applying this patch I no
> > > > more see this issue.
> > >
> > > Thanks, the boot log finally gave me a clue, and allowed me to reproduce.
> > > The issue happens only when adding:
> > >
> > > console=ttySC0,115200n8
> > >
> > Ack tested it on G2H.
> >
> > > to the kernel command line. Which is something we never did on R-Car
> > > Gen3, as the console= parameter had been deprecated by chosen/stdout-path
> > > on DT systems long before.
> > >
> > > As we did use console= before on arm32, and drivers/tty/serial/sh-sci.c
> > > never called spinlock_init(), I'm wondering if this spinlock bug is
> > > actually a regression in serial_core.c?
> > >
> > Not sure if it's a regression in serial_core.c as I see some drivers
> > calling spin_lock_init().
>
> Bisected to commit a3cb39d258efef83 ("serial: core: Allow detach and
> attach serial device for console").
> The first change to drivers/tty/serial/serial_core.c is the culprit:
>
> static inline void uart_port_spin_lock_init(struct uart_port *port)
> {
> - if (uart_console_enabled(port))
> + if (uart_console(port))
> return;
>
> spin_lock_init(&port->lock);
>
> as it now skips the spinlock initialization if a console= parameter
> is specified.
>
> Apparently we're not the only one bitten by that...
>
Thank you for bisecting through, that explains the culprit.

Cheers,
--Prabhakar

> Gr{oetje,eeting}s,
>
> Geert
>
> --
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
>
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
> -- Linus Torvalds