Re: [RFC PATCH 1/3] proc/meminfo: introduce extra meminfo

[Jaewon Kim <jaewon31.kim@samsung.com>]

On 2020년 03월 12일 02:35, Alexey Dobriyan wrote:
> On Wed, Mar 11, 2020 at 12:44:39PM +0900, Jaewon Kim wrote:
>> Provide APIs to drivers so that they can show its memory usage on
>> /proc/meminfo.
>>
>> int register_extra_meminfo(atomic_long_t *val, int shift,
>> 			   const char *name);
>> int unregister_extra_meminfo(atomic_long_t *val);
>> +			show_val_kb(m, memtemp->name_pad, nr_page);
> I have 3 issues.
>
> Can this be printed without "KB" piece and without useless whitespace,
> like /proc/vmstat does?
>
> I don't know how do you parse /proc/meminfo.
> Do you search for specific string or do you use some kind of map[k] = v
> interface?
If need, I can remove KB. but show_free_areas also seems to print KB for some stats.
And I intentionally added : and space to be same format like others in /proc/meminfo.
show_val_kb(m, "MemTotal:       ", i.totalram);
>
> 2) zsmalloc can create top-level symlink and resolve it to necessary value.
> It will be only 1 readlink(2) system call to fetch it.
Yes it could be done by userspace readlink systemcall. But I wanted to see
all huge memory stats on one node of /proc/meminfo.
>
> 3) android can do the same
>
> For simple values there is no need to register stuff and create
> mini subsystems.
>
> 	/proc/alexey
OK as Leon Romanovsky said, I may be able to move other node of /proc/meminfo_extra.
Please let me know if it still have a point to be reviewed.

Thank you
>
>