[PATCH] mm: bring back /sys/kernel/mm

[PATCH] mm: bring back /sys/kernel/mm

[Hugh Dickins]

Commit da29bd36224b ("mm/mm_init.c: make creation of the mm_kobj happen
earlier than device_initcall") changed to pure_initcall(mm_sysfs_init).

That's too early: mm_sysfs_init() depends on core_initcall(ksysfs_init)
to have made the kernel_kobj directory "kernel" in which to create "mm".

Make it postcore_initcall(mm_sysfs_init).  We could use core_initcall(),
and depend upon Makefile link order kernel/ mm/ fs/ ipc/ security/ ...
as core_initcall(debugfs_init) and core_initcall(securityfs_init) do;
but better not.

Signed-off-by: Hugh Dickins <hughd@google.com>
---

 mm/mm_init.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

--- 3.13.0+/mm/mm_init.c	2014-01-23 21:51:26.004001378 -0800
+++ linux/mm/mm_init.c	2014-01-26 18:06:40.488488209 -0800
@@ -202,4 +202,4 @@ static int __init mm_sysfs_init(void)
 
 	return 0;
 }
-pure_initcall(mm_sysfs_init);
+postcore_initcall(mm_sysfs_init);

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: bring back /sys/kernel/mm

[Paul Gortmaker]

[[PATCH] mm: bring back /sys/kernel/mm] On 26/01/2014 (Sun 18:52) Hugh Dickins wrote:

> Commit da29bd36224b ("mm/mm_init.c: make creation of the mm_kobj happen
> earlier than device_initcall") changed to pure_initcall(mm_sysfs_init).
> 
> That's too early: mm_sysfs_init() depends on core_initcall(ksysfs_init)
> to have made the kernel_kobj directory "kernel" in which to create "mm".
> 
> Make it postcore_initcall(mm_sysfs_init).  We could use core_initcall(),
> and depend upon Makefile link order kernel/ mm/ fs/ ipc/ security/ ...
> as core_initcall(debugfs_init) and core_initcall(securityfs_init) do;
> but better not.

Agreed, N+1 is better than link order.  I guess it silently fails then,
with /sys/kernel/mm missing as the symptom?  I'd booted i386 and ppc
and didn't spot this, unfortunately...  wondering now if there was a
hint in dmesg that I simply failed to notice.

> 
> Signed-off-by: Hugh Dickins <hughd@google.com>

Acked-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Thanks,
Paul.

> ---
> 
>  mm/mm_init.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> --- 3.13.0+/mm/mm_init.c	2014-01-23 21:51:26.004001378 -0800
> +++ linux/mm/mm_init.c	2014-01-26 18:06:40.488488209 -0800
> @@ -202,4 +202,4 @@ static int __init mm_sysfs_init(void)
>  
>  	return 0;
>  }
> -pure_initcall(mm_sysfs_init);
> +postcore_initcall(mm_sysfs_init);

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: bring back /sys/kernel/mm

[Hugh Dickins]

On Sun, 26 Jan 2014, Paul Gortmaker wrote:
> [[PATCH] mm: bring back /sys/kernel/mm] On 26/01/2014 (Sun 18:52) Hugh Dickins wrote:
> 
> > Commit da29bd36224b ("mm/mm_init.c: make creation of the mm_kobj happen
> > earlier than device_initcall") changed to pure_initcall(mm_sysfs_init).
> > 
> > That's too early: mm_sysfs_init() depends on core_initcall(ksysfs_init)
> > to have made the kernel_kobj directory "kernel" in which to create "mm".
> > 
> > Make it postcore_initcall(mm_sysfs_init).  We could use core_initcall(),
> > and depend upon Makefile link order kernel/ mm/ fs/ ipc/ security/ ...
> > as core_initcall(debugfs_init) and core_initcall(securityfs_init) do;
> > but better not.
> 
> Agreed, N+1 is better than link order.  I guess it silently fails then,
> with /sys/kernel/mm missing as the symptom?  I'd booted i386 and ppc
> and didn't spot this, unfortunately...  wondering now if there was a
> hint in dmesg that I simply failed to notice.

No, nothing in dmesg at all: both mm_sysfs_init() and ksm_init()
(it was /sys/kernel/mm/ksm/run that I was looking for) thought
they had succeeded.

Ah, I get it: it's normal to pass NULL parent to kobject_create_and_add(),
that just means create at the root.  And when I look at an unfixed box,
yes, there's /sys/mm with all the right contents.

Given /sys/block and /sys/fs and /sys/kernel, I think /sys/mm is a better
location; but we're a few years too late to be making that change now ;)

Hugh

> 
> > 
> > Signed-off-by: Hugh Dickins <hughd@google.com>
> 
> Acked-by: Paul Gortmaker <paul.gortmaker@windriver.com>
> 
> Thanks,
> Paul.
> 
> > ---
> > 
> >  mm/mm_init.c |    2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > --- 3.13.0+/mm/mm_init.c	2014-01-23 21:51:26.004001378 -0800
> > +++ linux/mm/mm_init.c	2014-01-26 18:06:40.488488209 -0800
> > @@ -202,4 +202,4 @@ static int __init mm_sysfs_init(void)
> >  
> >  	return 0;
> >  }
> > -pure_initcall(mm_sysfs_init);
> > +postcore_initcall(mm_sysfs_init);

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: bring back /sys/kernel/mm

[Paul Gortmaker]

[Re: [PATCH] mm: bring back /sys/kernel/mm] On 26/01/2014 (Sun 21:02) Hugh Dickins wrote:

> On Sun, 26 Jan 2014, Paul Gortmaker wrote:
> > [[PATCH] mm: bring back /sys/kernel/mm] On 26/01/2014 (Sun 18:52) Hugh Dickins wrote:
> > 
> > > Commit da29bd36224b ("mm/mm_init.c: make creation of the mm_kobj happen
> > > earlier than device_initcall") changed to pure_initcall(mm_sysfs_init).
> > > 
> > > That's too early: mm_sysfs_init() depends on core_initcall(ksysfs_init)
> > > to have made the kernel_kobj directory "kernel" in which to create "mm".
> > > 
> > > Make it postcore_initcall(mm_sysfs_init).  We could use core_initcall(),
> > > and depend upon Makefile link order kernel/ mm/ fs/ ipc/ security/ ...
> > > as core_initcall(debugfs_init) and core_initcall(securityfs_init) do;
> > > but better not.
> > 
> > Agreed, N+1 is better than link order.  I guess it silently fails then,
> > with /sys/kernel/mm missing as the symptom?  I'd booted i386 and ppc
> > and didn't spot this, unfortunately...  wondering now if there was a
> > hint in dmesg that I simply failed to notice.
> 
> No, nothing in dmesg at all: both mm_sysfs_init() and ksm_init()
> (it was /sys/kernel/mm/ksm/run that I was looking for) thought
> they had succeeded.
> 
> Ah, I get it: it's normal to pass NULL parent to kobject_create_and_add(),
> that just means create at the root.  And when I look at an unfixed box,
> yes, there's /sys/mm with all the right contents.

Aha - yeah, I would have never seen that; the subtle rename is too
easy to overlook unless one is used to going after some path by habit,
and wonders why tab completion isn't working anymore...

> 
> Given /sys/block and /sys/fs and /sys/kernel, I think /sys/mm is a better
> location; but we're a few years too late to be making that change now ;)

Agreed.

Thanks again for the root cause, I don't feel so bad for missing it now.
Paul.
--

> 
> Hugh
> 
> > 
> > > 
> > > Signed-off-by: Hugh Dickins <hughd@google.com>
> > 
> > Acked-by: Paul Gortmaker <paul.gortmaker@windriver.com>
> > 
> > Thanks,
> > Paul.
> > 
> > > ---
> > > 
> > >  mm/mm_init.c |    2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > --- 3.13.0+/mm/mm_init.c	2014-01-23 21:51:26.004001378 -0800
> > > +++ linux/mm/mm_init.c	2014-01-26 18:06:40.488488209 -0800
> > > @@ -202,4 +202,4 @@ static int __init mm_sysfs_init(void)
> > >  
> > >  	return 0;
> > >  }
> > > -pure_initcall(mm_sysfs_init);
> > > +postcore_initcall(mm_sysfs_init);

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>