On Wed 2021-01-06 15:36:36, Vineet Gupta wrote:
> +CC Buildroot folks
>
> Hi Petr,
>
> On 11/11/20 5:54 AM, Petr Mladek wrote:
> > stdin, stdout, and stderr standard I/O stream are created for the init
> > process. They are not available when there is no console registered
> > for /dev/console. It might lead to a crash when the init process
> > tries to use them, see the commit 48021f98130880dd742 ("printk: handle
> > blank console arguments passed in.").
> >
> > Normally, ttySX and ttyX consoles are used as a fallback when no consoles
> > are defined via the command line, device tree, or SPCR. But there
> > will be no console registered when an invalid console name is configured
> > or when the configured consoles do not exist on the system.
> >
> > Users even try to avoid the console intentionally, for example,
> > by using console="" or console=null. It is used on production
> > systems where the serial port or terminal are not visible to
> > users. Pushing messages to these consoles would just unnecessary
> > slowdown the system.
> >
> > Make sure that stdin, stdout, stderr, and /dev/console are always
> > available by a fallback to the existing ttynull driver. It has
> > been implemented for exactly this purpose but it was used only
> > when explicitly configured.
> >
> > Signed-off-by: Petr Mladek <pmladek@xxxxxxxx>
>
> > --- a/init/main.c
> > +++ b/init/main.c
> > @@ -1470,8 +1470,14 @@ void __init console_on_rootfs(void)
> > struct file *file = filp_open("/dev/console", O_RDWR, 0);
> > if (IS_ERR(file)) {
> > - pr_err("Warning: unable to open an initial console.\n");
> > - return;
> > + pr_err("Warning: unable to open an initial console. Fallback to ttynull.\n");
> > + register_ttynull_console();
> > +
> > + file = filp_open("/dev/console", O_RDWR, 0);
> > + if (IS_ERR(file)) {
> > + pr_err("Warning: Failed to add ttynull console. No stdin, stdout, and stderr for the init process!\n");
> > + return;
> > + }
>
>
> This breaks ARC booting (no output on console).
This is likely the same problem as with kunit and um kernels.
It is being discussed at
https://lore.kernel.org/linux-kselftest/X%2FSRA1P8t+ONZFKb@alley/#t
We have several workarounds. I am still squashing my head about the
right solution. The console registration code is like a vasps' nest.
It is always a pain when we touch it.
I hope that I will send a patchset for review later today.
In the worst case, we will revert the patch in the mainline.
> Our Buildroot based setup has dynamic /dev where /dev/console doesn't exist
> statically and there's a primoridla /init shell script which does following
>
> /bin/mount -t devtmpfs devtmpfs /dev
> exec 0</dev/console
> exec 1>/dev/console
> exec 2>/dev/console
> exec /sbin/init "$@"
>
> Buildroot has had this way of handling missing /dev/console since 2011 [1]
> and [2].
Good to know.
> Please advise what needs to be done to unbork boot. Otherwise this seems
> like a kernel change which breaks user-space and needs to be backed-out (or
> perhaps conditionalize on CONFIG_NULL_TTY. I'm surprised it hasn't been
> reported by any other embedded folks
Two workarounds can be fount at
https://lore.kernel.org/linux-kselftest/X%2FSYhBZyudfnKY1u@alley/
https://lore.kernel.org/linux-kselftest/X%2FW2sl7RMvfaV4Ru@alley/
But I still see them as only a partial solutiuon. I still another sources
of potential problems.
Best Regards,
Petr
