[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20140715162330.f9b18b79c0f32b84cb7a9071@linux-foundation.org>
Date: Tue, 15 Jul 2014 16:23:30 -0700
From: Andrew Morton <akpm@...ux-foundation.org>
To: Mathias Krause <minipli@...glemail.com>
Cc: Thomas Gleixner <tglx@...utronix.de>,
Ingo Molnar <mingo@...hat.com>,
"H. Peter Anvin" <hpa@...or.com>, Joe Perches <joe@...ches.com>,
Rasmus Villemoes <linux@...musvillemoes.dk>,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2 2/8] printk: Provide pi_<level> / pe_<level> macros
for __init / __exit code
On Sat, 12 Jul 2014 16:43:26 +0200 Mathias Krause <minipli@...glemail.com> wrote:
> The memory used for functions marked with __init will be released after
> initialization, albeit static data referenced by such code will not, if
> not explicitly marked this way, too. This is especially true for format
> strings used in messages printed by such code. Those are not marked and
> therefore will survive the __init cleanup -- dangling in memory without
> ever being referenced again.
>
> Ideally we would like the compiler to automatically recognise such
> constructs but it does not and it's not as simple as it might sound, as
> strings used in initialization code might latter still be used, e.g. as
> a slab cache name. Therefore we need to explicitly mark the strings we
> want to release together with the function itself.
>
> For a seamless integration of the necessary __init_str() / __exit_str()
> macros to mark the format strings, this patch provides new variants of
> the well known pr_<level>() macros as pi_<level>() for __init code and
> pe_<level>() for __exit code. Changing existing code should thereby be
> as simple as changing a single letter.
>
> For code that cannot be changed to use the pi_<level>() / pe_<level>()
> macros printk_init() and printk_exit() macros are provided, too.
>
> One remark, though: We cannot provide appropriate p[ie]_debug() macros
> for the CONFIG_DYNAMIC_DEBUG case as there is (currently) no way to
> remove entries from dyndbg after initialization. But supporting that
> scenario would require more work (and code), therefore not necessarily
> justifying the memory savings.
I assume that if a programmer gets this wrong,
CONFIG_DEBUG_SECTION_MISMATCH will detect and report the error?
Please thoroughly test this if you have not done so.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists