Re: [PATCH v5 07/16] printk: Ensure consistent name and devname when updating preferred console

[Petr Mladek]

On Thu 2026-10-29 15:09:52, Chris Down wrote:
> Petr Mladek writes:
> >-		strscpy(pc->name, name);
> >-		pc->index = idx;
> >+		if (!pc->name[0]) {
> >+			is_new_entry = true;
> >+			strscpy(pc->name, name);
> >+			pc->index = idx;
> 
> console_setup() can hand us an empty name or devname:
> 
>     console=,115200   -> name "", idx 0, options "115200"
>     console=,foo:bar  -> devname "", idx -1, options "foo:bar"
> 
> For the devname case, __add_preferred_console() matches the first name-only 
> entry (`strcmp(pc->devname, "") == 0`), e.g. the SPCR "pl011"/0 entry, and this 
> hunk now treats it as new.
> 
> Then index becomes -1 and options become "foo:bar", so the SPCR console never 
> matches again. Before the series that command line only moved 
> preferred_console.
> 
> For the `!pc->name[0]` branch below the slot keeps looking unused, which turns 
> into list corruption with 14/16 (I'll put some details over there instead of 
> duplicating it here).

Great catch! I think that even the original code did not expect that
anyone could pass "" as either @name or @devname.

> Rejecting empty strings up front, after the `WARN_ON(!name && !devname)`, fixes 
> both:
> 
>     	if ((name && !name[0]) || (devname && !devname[0])) {
>     		pr_err("Adding a preferred console with an empty name\n");
>     		return -EINVAL;
>     	}

I like this. I am going to add this as a separate patch.
As mentioned above, I think that it was not handled before...

I have the following as the 5th patch: