Re: [PATCH] sysctl: Add a group of macro functions to initcall the sysctl table of each feature

[Xiaoming Ni <nixiaoming@huawei.com>]

On 2021/12/7 9:38, Andrew Morton wrote:
> On Tue, 7 Dec 2021 09:13:20 +0800 Xiaoming Ni <nixiaoming@huawei.com> wrote:
> 
>> To avoid duplicated code, add a set of macro functions to initialize the
>> sysctl table for each feature.
>>
>> The system initialization process is as follows:
>>
>> 	start_kernel () {
>> 		...
>> 		/* init proc and sysctl base,
>> 		 * proc_root_init()-->proc_sys_init()-->sysctl_init_bases()
>> 		 */
>> 		proc_root_init(); /* init proc and sysctl base */
>> 		...
>> 		arch_call_rest_init();
>> 	}
>>
>> 	arch_call_rest_init()-->rest_init()-->kernel_init()
>> 	kernel_init() {
>> 		...
>> 		kernel_init_freeable(); /* do all initcalls */
>> 		...
>> 		do_sysctl_args(); /* Process the sysctl parameter: sysctl.*= */
>> 	}
>>
>> 	kernel_init_freeable()--->do_basic_setup()-->do_initcalls()
>> 	do_initcalls() {
>> 		for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++) {
>> 			do_initcall_level
>> 	}
>>
>> The sysctl interface of each subfeature should be registered after
>> sysctl_init_bases() and before do_sysctl_args(). It seems that the sysctl
>> interface does not depend on initcall_levels. To prevent the sysctl
>> interface from being initialized before the feature itself. The
>> lowest-level late_initcall() is used as the common sysctl interface
>> registration level.
> 
> I'm not normally a fan of wrapping commonly-used code sequences into
> magical macros, but this one does seem to make sense.
> 
> I wonder if it is possible to cook up a checkpatch rule to tell people
> to henceforth use the magic macros rather than to open-code things in
> the old way.  Sounds hard.
> 
>> --- a/fs/coredump.c
>> +++ b/fs/coredump.c
>> @@ -943,12 +943,7 @@ static struct ctl_table coredump_sysctls[] = {
>>   	{ }
>>   };
>>   
>> -static int __init init_fs_coredump_sysctls(void)
>> -{
>> -	register_sysctl_init("kernel", coredump_sysctls);
>> -	return 0;
>> -}
>> -fs_initcall(init_fs_coredump_sysctls);
>> +kernel_sysctl_initcall(coredump_sysctls);
> 
> But this and several like it are functional changes.
> 
>>   #endif /* CONFIG_SYSCTL */
>>
>> ...
>>
>> --- a/fs/inode.c
>> +++ b/fs/inode.c
>> @@ -132,12 +132,7 @@ static struct ctl_table inodes_sysctls[] = {
>>   	{ }
>>   };
>>   
>> -static int __init init_fs_inode_sysctls(void)
>> -{
>> -	register_sysctl_init("fs", inodes_sysctls);
>> -	return 0;
>> -}
>> -early_initcall(init_fs_inode_sysctls);
>> +fs_sysctl_initcall(inodes_sysctls);
>>   #endif
> 
> Here's another, of many.
> 
> Someone made the decision to use early_initcall() here (why?) and this
> patch switches it to late_initcall()!  Worrisome.  Each such stealth
> conversion should be explained and justified, shouldn't it?
> 

static noinline void __init kernel_init_freeable(void)
{
	...
	do_pre_smp_initcalls(); /* do early_initcall */
	lockup_detector_init();

	smp_init();
	sched_init_smp();

	padata_init();
	page_alloc_init_late();
	/* Initialize page ext after all struct pages are initialized. */
	page_ext_init();

	do_basic_setup();  /* do other initcall */
	...
}

Between do_pre_smp_initcalls() and do_basic_setup(), no sysctl interface 
window is configured. In addition, all sysctl data has initial values. 
Delayed configuration does not affect the behavior after startup.
So I think we can change it to late_initcall().


Thanks
Xiaoming Ni