Re: [PATCH] mm: change "char *bdi_unknown_name" to "char bdi_unknown_name[]"

[liqiong <liqiong@nfschina.com>]

在 2022年05月13日 04:00, Andrew Morton 写道:
> On Thu, 12 May 2022 16:26:37 +0800 liqiong <liqiong@nfschina.com> wrote:
>
>> "char bdi_unknown_nam[]" string form declares a single variable.
>> It is better then "char *bdi_unknown_name" which creates two
>> variables.
>>
>> ...
>>
>> --- a/mm/backing-dev.c
>> +++ b/mm/backing-dev.c
>> @@ -20,7 +20,7 @@ struct backing_dev_info noop_backing_dev_info;
>>  EXPORT_SYMBOL_GPL(noop_backing_dev_info);
>>  
>>  static struct class *bdi_class;
>> -static const char *bdi_unknown_name = "(unknown)";
>> +static const char bdi_unknown_name[] = "(unknown)";
>>  
> heh, fun patch.  We actually do this quite a lot.  
>
> 	grep -r "^[a-z].*char \*[a-z].*= \"" .
>
> is a pathetic pattern which catches a lot of them.
>
>
> However.  I expected your patch to shrink the kernel a bit, but it has
> the opposite effect:
>
> hp2:/usr/src/25> size mm/backing-dev.o
>    text	   data	    bss	    dec	    hex	filename
>   21288	   9396	   3808	  34492	   86bc	mm/backing-dev.o-before
>   21300	   9428	   3808	  34536	   86e8	mm/backing-dev.o-after
>
> Even .data became larger.  I didn't investigate why.
Hi,
It seems  the patch creates a new section:
0000000000000000 l       ___ksymtab_gpl+bdi_dev_name    0000000000000000 __ksymtab_bdi_dev_name
0000000000000f60 l     O .rodata        000000000000000a bdi_unknown_name
And  put  "bdi_unknown_name"  at  .rodata.str1.1:

This work was published in "KernelJanitors/Todo". It says:

"foo []" is better because it declares a single variable,   For variables marked __initdata, the "*foo" form causes
only the pointer, not the string itself, to be dropped from the kernel image, which is a bug.
Using the "foo[]" form with regular 'ole local variables also makes the assembly shorter.

I thought it make sense,  so i  searched the mm tree by "grep  -nHre char.*\*.*=.*\"",  and checked all the "char *foo"
string form,  It seems only this one should be fixed.