Received-SPF: pass (google.com: domain of linux-kernel+bounces-48384-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
Date: Thu, 1 Feb 2024 17:18:26 +0200
From: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
To: Qu Wenruo <wqu@suse.com>
Cc: Qu Wenruo <quwenruo.btrfs@gmx.com>, linux-btrfs@vger.kernel.org,
	linux-kernel@vger.kernel.org, akpm@linux-foundation.org,
	christophe.jaillet@wanadoo.fr, David.Laight@aculab.com,
	ddiss@suse.de, geert@linux-m68k.org
Subject: Re: [PATCH v3 0/4] kstrtox: introduce memparse_safe()
Message-ID: <Zbu2QqhC2LyK4ttb@smile.fi.intel.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Organization: Intel Finland Oy - BIC 0357606-4 - Westendinkatu 7, 02160 Espoo

On Mon, Jan 15, 2024 at 06:31:05AM +1030, Qu Wenruo wrote:
> On 2024/1/15 00:12, Andy Shevchenko wrote:
> > On Sun, Jan 07, 2024 at 07:28:27AM +1030, Qu Wenruo wrote:
> > > On 2024/1/7 01:04, Andy Shevchenko wrote:
> > > > On Thu, Jan 04, 2024 at 09:57:47AM +1030, Qu Wenruo wrote:

..

> > > > Having test cases is quite good, thanks!
> > > > But as I understood what Alexey wanted, is not using the kstrtox files for this.
> > > > You can introduce it in the cmdline.c, correct? Just include local "kstrtox.h".
> > > 
> > > Not really possible, all the needed parsing helpers are internal inside
> > > kstrtox.c.
> > 
> > I'm not sure I follow. The functions are available to other library (built-in)
> > modules.
> 
> Did I miss something?
> 
> Firstly neither _parse_integer_fixup_radix() nor _parse_integer_limit() is
> exported to modules. (No EXPORT_SYMBOL() call on them).

Should they?

> Secondly _parse_integer_fixup_radix() and _parse_integer_limit have "_"
> prefix, and is only declared in "lib/kstrtox.h", which means they are
> designed only for internal usage.
> If putting memparse_safe() into cmdline.c, at least we would need to include
> local header "kstrtox.h", and I'm not sure if this is any better than
> putting memparse_safe() into kstrtox.[ch].

Yes, this is much better.

> Finally, I just tried putting memparse_safe() into cmdline.c, and it failed
> at linkage stage, even if that offending file has no call to
> memparse_safe():
> 
>   ld: arch/x86/boot/compressed/kaslr.o: in function `memparse_safe':
> kaslr.c:(.text+0xbb1): undefined reference to `_parse_integer_fixup_radix'
>   ld: kaslr.c:(.text+0xbc5): undefined reference to `_parse_integer'
>   ld: arch/x86/boot/compressed/vmlinux: hidden symbol `_parse_integer' isn't
> defined

I don't understand. Implement memparse_safe() in the C file
(i.e.  lib/cmdline.c) and if it's needed to be exported, export it.

> I can try again but I'm not sure if it's possible to move memparse_safe() to
> cmdline.[ch].

Please do. I do not see design issues with proposed approach.

..

> > > Furthermore, this also means memparse() can not be enhanced due to:
> > > 
> > > - Lack of ways to return errors
> > 
> > What does this mean?
> 
> If you want to keep the prototype of memparse() (aka, a drop-in
> enhancement), then there is no good way to indicate the errors like overflow
> at all.

No need to replace one by another in the implementation side, having two is
okay, while moving affected code to the better one one-by-one. Then we may
remove the old one. And then, if there is a wish, we may rename it back.

> > > - Unable to call the parsing helpers inside cmdline.c
> > 
> > ??? (see above)
> > 
> See above.

Got it, but please try again. I do believe it's possible to avoid touching
kstrotox.c.

-- 
With Best Regards,
Andy Shevchenko