Can we please revert commit e4c570c4cb7a95dbfafa3d016d2739bf3fdfe319
"task_struct: make journal_info conditional"
I think I gave a good enough arguments to not merge it.
To iterate:
* patch makes impossible to start using ext3 on EXT3_FS=n kernels
without reboot.
* this is done only for one pointer on task_struct
None of config options which define task_struct are tristate directly
or effectively.
There are other examples where we don't do this:
* quota stuff on superblock aren't ifdeffed because of modular XFS
with quota support.
* exports operations and s_anon for NFS
and so on
On Thu, 17 Dec 2009 10:31:52 +0200
Alexey Dobriyan <[email protected]> wrote:
> Can we please revert commit e4c570c4cb7a95dbfafa3d016d2739bf3fdfe319
> "task_struct: make journal_info conditional"
>
> I think I gave a good enough arguments to not merge it.
> To iterate:
> * patch makes impossible to start using ext3 on EXT3_FS=n kernels
> without reboot.
> * this is done only for one pointer on task_struct
>
> None of config options which define task_struct are tristate directly
> or effectively.
>
> There are other examples where we don't do this:
> * quota stuff on superblock aren't ifdeffed because of modular XFS
> with quota support.
> * exports operations and s_anon for NFS
> and so on
Confused. I "dropped" that patch when you pointed out the problem then
forgot about it.
Maybe I dropped something else by accident. argh.
I agree - I'll queue a revert.
Alexey Dobriyan wrote:
> Can we please revert commit e4c570c4cb7a95dbfafa3d016d2739bf3fdfe319
> "task_struct: make journal_info conditional"
I'm fine to revert it, because it seems merged accidentally.
>
> I think I gave a good enough arguments to not merge it.
> To iterate:
> * patch makes impossible to start using ext3 on EXT3_FS=n kernels
> without reboot.
> * this is done only for one pointer on task_struct
I don't think it's only one pointer.
There might be a lot of "only one pointer".
>
> None of config options which define task_struct are tristate directly
> or effectively.
So we never allow to make memory usage small with providing an option
to remove unused area, right? Of cause, that option should be handled
carefully. If I want to reduce memory usage by this way, should I keep
this kind of patches out of tree?
thanks,
Hiroshi
On 12/18/09, Hiroshi Shimamoto <[email protected]> wrote:
> So we never allow to make memory usage small with providing an option
> to remove unused area, right?
We certainly allow this if it results in zero loss in functionality.
> If I want to reduce memory usage by this way, should I keep
> this kind of patches out of tree?
Certainly nobody can prohibit you from keeping patch out of tree.
But if you want something mainlinable, moving ->journal_info
to fs-specific data structures should do the trick. Or something.
Alexey Dobriyan wrote:
> On 12/18/09, Hiroshi Shimamoto <[email protected]> wrote:
>> So we never allow to make memory usage small with providing an option
>> to remove unused area, right?
>
> We certainly allow this if it results in zero loss in functionality.
Thanks for clarifying this topic.
If you don't mind could you please tell me what zero loss is?
I don't think I could get it exactly.
Is it OK that removing journal_info if !CONFIG_BLOCK?
>
>> If I want to reduce memory usage by this way, should I keep
>> this kind of patches out of tree?
>
> Certainly nobody can prohibit you from keeping patch out of tree.
> But if you want something mainlinable, moving ->journal_info
> to fs-specific data structures should do the trick. Or something.
Thanks for the advice, I'll look at this.
Thanks,
Hiroshi
Alexey Dobriyan <[email protected]> writes:
> On 12/18/09, Hiroshi Shimamoto <[email protected]> wrote:
>> So we never allow to make memory usage small with providing an option
>> to remove unused area, right?
>
> We certainly allow this if it results in zero loss in functionality.
>
>> If I want to reduce memory usage by this way, should I keep
>> this kind of patches out of tree?
>
> Certainly nobody can prohibit you from keeping patch out of tree.
> But if you want something mainlinable, moving ->journal_info
> to fs-specific data structures should do the trick. Or something.
Why doesn't this use EMBEDDED?
I.e.
boot "xxx" if EMBEDDED
default y
style Kconfig.
Only reason which seems reasonable is "this is done only for one pointer
on task_struct". This reason seems something like "You care, but I don't
care it". And, other reasons is *normal* situation on EMBEDDED, right?
Thanks.
--
OGAWA Hirofumi <[email protected]>