Re: [RFC PATCH 0/3] Mark literal strings in __init / __exit code

[Mathias Krause]

On 24 June 2014 21:37, Joe Perches <joe@xxxxxxxxxxx> wrote:
> On Tue, 2014-06-24 at 21:13 +0200, Mathias Krause wrote:
>> I don't think it's that easy. You cannot simply put all strings into
>> the .init.rodata section when code currently gets emitted to
>> .init.text. The reason is because strings used in __init code might be
>> referenced later on, too. For example, the name passed to
>> class_create() won't be copied. If that one would go into the
>> .init.rodata section automatically, we would have dangling pointers
>> after the .init.* memory got freed. Therefore a compiler driven
>> approach would need to be implemented as a compiler extension, a gcc
>> plugin to handle such cases -- know when a string can safely be put
>> into the .init.rodata section and when not. But that decision is not
>> as easy as Joe might think it would be. How would the plugin know
>> which strings to put into the .init.rodata section? Would it only
>> handle the ones passed to printk()?
>
> Yes.

Well, I would like to handle the easy ones, too. E.g. strings used in
parameter parsing, i.e. strcmp()s.

>> I still strongly believe it's better to do this manually.
>
> Maybe.
>
> It'd work with any version of the compiler that way too.

That's a much stronger argument, IMHO. It'll work with gcc < 4.5 and
clang, even. And, it does not require us to maintain compatibility to
the repeating gcc plugin API breakage.

> It's a pretty simple transform.

Indeed, it is.

> I believe this will show most all of the __init
> uses of printks:
>
> $ grep-2.5.4 -rP --include=*.[ch] -n '\b__init\b[^\n][^\}]+\n}' * | \
>   grep -P '^[\w\/\.]+:\d+:|\bprintk\b|\bpr_[a-z]+' | \
>   grep -P -B1 '\bprintk\b|\bpr_[a-z]+'
>
> This shows a little more than a 1000 __init printks
> treewide that could be converted.

A simple awk script found additional 5399 pr_<level> calls in __init
code and 188 calls in __exit code. So it's worth it, IMHO.

>
> For example:
> arch/ia64/include/asm/cyclone.h:6:extern void __init cyclone_setup(void);
>         printk(KERN_ERR "Cyclone Counter: System not configured"
> --
> arch/ia64/kernel/acpi.c:66:static unsigned long __init acpi_find_rsdp(void)
>                 printk(KERN_WARNING PREFIX
> --
> arch/ia64/kernel/acpi.c:366:static int __init acpi_parse_madt(struct acpi_table
>         printk(KERN_INFO PREFIX "Local APIC address %p\n", ipi_base_addr);
>

Those might benefit twice from the change by being converted to
pi_<level> calls along the way. So it's a win-win on all sides, no?

> etc...
>
> There are maybe 200 or so __exit ones.
>

Yeah, the __exit ones are used less. Nonetheless, they should be
converted, too. For completeness, at least.

Thanks,
Mathias
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/