Skip to contentSkip to conversation

dm: sysfs add empty release function to avoid debug warning

DISCUSSION Greg KH 3 participants 10 messages Dec 16, 2009 – Dec 22, 2009 1 message not in the archive Device mapper

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

Greg KH

author
#1 lore raw
Quoted 11 lines · James Bottomley

commit: d2bb7df8cac647b92f51fb84ae735771e7adbfa7
From: Milan Broz <[email protected]>
Date: Thu, 10 Dec 2009 23:51:53 +0000
Subject: [PATCH] dm: sysfs add empty release function to avoid debug warning

This patch just removes an unnecessary warning:
kobject: 'dm': does not have a release() function, it is broken and must be fixed.

The kobject is embedded in mapped device struct, so code does not need to release memory explicitly here.

Oh no you did not!

Come on people, do you think that the kernel is just randomly spitting out warnings because it fricken feels like it!

No, please revert this patch now before someone does it for you.

Please, this is totally and completly wrong. And if you feel that it is needed, then your design is wrong and it needs to be fixed.

And note, as per the Documentation/kobject.txt file, you were warned about this public mocking, next time it will be harsher.

Heh, and you tried to get this into a -stable kernel release, as if.

Patch left below so that others can mock.

bah,

greg k-h

Quoted 38 lines
---
 drivers/md/dm-sysfs.c |    8 ++++++++
 1 files changed, 8 insertions(+), 0 deletions(-)
drivers/md/dm-sysfs.c +8 -0
index 4b04590..b000de3 100644
@@ -80,12 +80,20 @@ static struct sysfs_ops dm_sysfs_ops = {
 };
 
 /*
+ * The sysfs structure is embedded in md struct, nothing to do here
+ */
+static void dm_sysfs_release(struct kobject *kobj)
+{
+}
+
+/*
  * dm kobject is embedded in mapped_device structure
  * no need to define release function here
  */
 static struct kobj_type dm_ktype = {
 	.sysfs_ops	= &dm_sysfs_ops,
 	.default_attrs	= dm_attrs,
+	.release	= dm_sysfs_release
 };
 
 /*

_______________________________________________
stable mailing list
[email protected]
http://linux.kernel.org/mailman/listinfo/stable

#2 lore raw
Quoted 10 lines · Greg KH
Quoted 9 lines · James Bottomley

commit: d2bb7df8cac647b92f51fb84ae735771e7adbfa7
From: Milan Broz <[email protected]>
Date: Thu, 10 Dec 2009 23:51:53 +0000
Subject: [PATCH] dm: sysfs add empty release function to avoid debug warning

This patch just removes an unnecessary warning:
kobject: 'dm': does not have a release() function, it is broken and must be fixed.

Quoted 6 lines
Quoted 2 lines

The kobject is embedded in mapped device struct, so code does not need to release memory explicitly here.

Please, this is totally and completly wrong. And if you feel that it is needed, then your design is wrong and it needs to be fixed.

There are several places in kernel, where kobject have not defined release method. Yes, something is wrong here.
So quietly ignoring warning is ok? Why is not there BUG_ON(!release) then?

The sysfs attributes here just represents attributes of block device object, this device is always removed before release here is called.

So if there is preferred to do another alloc/free, no problem.

Quoted 4 lines

And note, as per the Documentation/kobject.txt file, you were warned about this public mocking, next time it will be harsher.

Heh, and you tried to get this into a -stable kernel release, as if.

btw I did'n added stable@, someone added that later. But it is not relevant.

Quoted 1 line

Patch left below so that others can mock.

Enjoy;-)

Milan

author
#3 lore raw
Quoted 15 lines · Milan Broz
Quoted 13 lines · Greg KH
Quoted 11 lines · James Bottomley

commit: d2bb7df8cac647b92f51fb84ae735771e7adbfa7
From: Milan Broz <[email protected]>
Date: Thu, 10 Dec 2009 23:51:53 +0000
Subject: [PATCH] dm: sysfs add empty release function to avoid debug warning

This patch just removes an unnecessary warning:
kobject: 'dm': does not have a release() function, it is broken and must be fixed.

The kobject is embedded in mapped device struct, so code does not need to release memory explicitly here.

Quoted 5 lines
Quoted 2 lines

Please, this is totally and completly wrong. And if you feel that it is needed, then your design is wrong and it needs to be fixed.

There are several places in kernel, where kobject have not defined release method. Yes, something is wrong here.

Where are those instances, becides the use of static kobjects, which is being worked on?

That still does not make this change acceptable, it is incorrect.

Quoted 1 line

So quietly ignoring warning is ok? Why is not there BUG_ON(!release) then?

I was trying to be nice and not crash your machine and give you the opportunity to fix it easily.

Quoted 2 lines

The sysfs attributes here just represents attributes of block device object, this device is always removed before release here is called.

Then why use a kobject at all? Your reference counting will be all wrong if you use it incorrectly like this.

Quoted 1 line

So if there is preferred to do another alloc/free, no problem.

Please do.

Also, with the addition of this patch, your comments for the kobject are now incorrect as well.

thanks,

greg k-h

#4 lore raw

Dne 16.12.2009 14:45, Greg KH napsal(a):

Quoted 38 lines · Milan Broz
Quoted 15 lines · Greg KH
Quoted 13 lines · James Bottomley
Quoted 11 lines

commit: d2bb7df8cac647b92f51fb84ae735771e7adbfa7
From: Milan Broz <[email protected]>
Date: Thu, 10 Dec 2009 23:51:53 +0000
Subject: [PATCH] dm: sysfs add empty release function to avoid debug warning

This patch just removes an unnecessary warning:
kobject: 'dm': does not have a release() function, it is broken and must be fixed.

The kobject is embedded in mapped device struct, so code does not need to release memory explicitly here.

Quoted 5 lines
Quoted 2 lines

Please, this is totally and completly wrong. And if you feel that it is needed, then your design is wrong and it needs to be fixed.

There are several places in kernel, where kobject have not defined release method. Yes, something is wrong here.

Where are those instances, becides the use of static kobjects, which is being worked on?

That still does not make this change acceptable, it is incorrect.

Quoted 1 line

So quietly ignoring warning is ok? Why is not there BUG_ON(!release) then?

I was trying to be nice and not crash your machine and give you the opportunity to fix it easily.

Quoted 2 lines

The sysfs attributes here just represents attributes of block device object, this device is always removed before release here is called.

Then why use a kobject at all? Your reference counting will be all wrong if you use it incorrectly like this.

Quoted 7 lines
Quoted 1 line

So if there is preferred to do another alloc/free, no problem.

Please do.

Also, with the addition of this patch, your comments for the kobject are now incorrect as well.

Here is list of just some errors I get on my 2.6.32 linux kernel. (error message itself is removed)
Messages are printed during module removal/reboot.

'aead' (ffffffffa017c050): does not have a release() function, it is broken and must be fixed.
'bridge'
'cdrom'
'cfg80211'
'crypto_hash'
'cryptomgr'
'dm_log'
'dm_mirror'
'dm_mod'
'dm_multipath'
'dm_region_hash'
'ehci_hcd'
'e1000e'
'iptable_filter'
'iptable_nat'
'ip_tables'
'ipt_MASQUERADE'
'ipt_REJECT'
'iTCO_vendor_support'
'iTCO_wdt'
'iwlcore'
'iwl3945'
'iwl3945'
'kvm'
'kvm_intel'
'kvm0'
'llc'
'mac80211'
'microcode'
'mmc_core'
'mmc'
'nf_conntrack'
'nf_conntrack_ipv4'
'nf_defrag_ipv4'
'nf_nat'
'pcompress'
'psmouse'
'rtc_cmos'
'rtc_core'
'rtc_lib'
'scsi_wait_scan'
'sdhci'
'sdhci_pci'
'sdio'
'snd_mixer_oss'
'snd_pcm_oss'
'snd_seq_device'
'snd_seq'
'snd_seq_midi_event'
'snd_seq_oss'
'sr_mod'
'stp'
'x_tables'
'xt_state'
'xt_tcpudp'

Zdenek

author
#5 lore raw
Quoted 47 lines · Zdenek Kabelac

Dne 16.12.2009 14:45, Greg KH napsal(a):

Quoted 38 lines · Milan Broz
Quoted 15 lines · Greg KH
Quoted 13 lines · James Bottomley
Quoted 11 lines

commit: d2bb7df8cac647b92f51fb84ae735771e7adbfa7
From: Milan Broz <[email protected]>
Date: Thu, 10 Dec 2009 23:51:53 +0000
Subject: [PATCH] dm: sysfs add empty release function to avoid debug warning

This patch just removes an unnecessary warning:
kobject: 'dm': does not have a release() function, it is broken and must be fixed.

The kobject is embedded in mapped device struct, so code does not need to release memory explicitly here.

Quoted 5 lines
Quoted 2 lines

Please, this is totally and completly wrong. And if you feel that it is needed, then your design is wrong and it needs to be fixed.

There are several places in kernel, where kobject have not defined release method. Yes, something is wrong here.

Where are those instances, becides the use of static kobjects, which is being worked on?

That still does not make this change acceptable, it is incorrect.

Quoted 1 line

So quietly ignoring warning is ok? Why is not there BUG_ON(!release) then?

I was trying to be nice and not crash your machine and give you the opportunity to fix it easily.

Quoted 2 lines

The sysfs attributes here just represents attributes of block device object, this device is always removed before release here is called.

Then why use a kobject at all? Your reference counting will be all wrong if you use it incorrectly like this.

Quoted 7 lines
Quoted 1 line

So if there is preferred to do another alloc/free, no problem.

Please do.

Also, with the addition of this patch, your comments for the kobject are now incorrect as well.

Quoted 8 lines

Here is list of just some errors I get on my 2.6.32 linux kernel. (error message itself is removed)
Messages are printed during module removal/reboot.

'aead' (ffffffffa017c050): does not have a release() function, it is broken and must be fixed.
'bridge'
'cdrom'

<snip>

You will find that all of these are modules, right? That's being worked on still.

Again, please fix your code not to have an empty release function. Will you revert this change and fix it properly?

thanks,

greg k-h

#6 lore raw
Quoted 2 lines · Greg KH

You will find that all of these are modules, right? That's being worked on still.

probably. I see that this implementation is in kernel since import to git...

What's the right solution in this particular case?
It is exactly the same situation (using embedded kobj like module handle does).

In DM, the whole problem here is that kobject is used for attribute representation of mapped device, mapped device struct is using its own reference counting and can dissappear before the kobject. So it must properly solve this situation.

Quoted 2 lines

Again, please fix your code not to have an empty release function. Will you revert this change and fix it properly?

Sure, working on it. Proper fix will replace this when ready.

Milan

author
#7 lore raw
Quoted 5 lines · Milan Broz
Quoted 2 lines · Greg KH

You will find that all of these are modules, right? That's being worked on still.

probably. I see that this implementation is in kernel since import to git...

And you will also note that I didn't paper over the issue by trying to provide an empty release function :)

Quoted 6 lines

What's the right solution in this particular case?
It is exactly the same situation (using embedded kobj like module handle does).

In DM, the whole problem here is that kobject is used for attribute representation of mapped device, mapped device struct is using its own reference counting and can dissappear before the kobject. So it must properly solve this situation.

Then dynamically create the kobject, as it obviously has a different lifespan.

Or do something else, but as it is now, you can not have two reference counts with different lifespan rules, within the same object, and expect things to work properly.

thanks,

greg k-h

#8 lore raw
Quoted 9 lines · Greg KH
Quoted 5 lines · Milan Broz
Quoted 2 lines · Greg KH

You will find that all of these are modules, right? That's being worked on still.

probably. I see that this implementation is in kernel since import to git...

And you will also note that I didn't paper over the issue by trying to provide an empty release function :)

Alasdair, can we request revert of that last patch for now (which introduced the empty release function)?

Milan

author
#9 lore raw
Quoted 6 lines · Milan Broz
Quoted 5 lines · Greg KH
Quoted 4 lines · Milan Broz
Quoted 2 lines · Greg KH

You will find that all of these are modules, right? That's being worked on still.

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

probably. I see that this implementation is in kernel since import to git...

And you will also note that I didn't paper over the issue by trying to provide an empty release function :)

Alasdair, can we request revert of that last patch for now (which introduced the empty release function)?

Ping? Any word on the revert?

thanks,

greg k-h

#10 lore raw
In reply to Greg KH

Dne 16.12.2009 18:46, Greg KH napsal(a):

Quoted 57 lines · Zdenek Kabelac
Quoted 47 lines

Dne 16.12.2009 14:45, Greg KH napsal(a):

Quoted 38 lines · Milan Broz
Quoted 15 lines · Greg KH
Quoted 13 lines · James Bottomley
Quoted 11 lines

commit: d2bb7df8cac647b92f51fb84ae735771e7adbfa7
From: Milan Broz <[email protected]>
Date: Thu, 10 Dec 2009 23:51:53 +0000
Subject: [PATCH] dm: sysfs add empty release function to avoid debug warning

This patch just removes an unnecessary warning:
kobject: 'dm': does not have a release() function, it is broken and must be fixed.

The kobject is embedded in mapped device struct, so code does not need to release memory explicitly here.

Quoted 5 lines
Quoted 2 lines

Please, this is totally and completly wrong. And if you feel that it is needed, then your design is wrong and it needs to be fixed.

There are several places in kernel, where kobject have not defined release method. Yes, something is wrong here.

Where are those instances, becides the use of static kobjects, which is being worked on?

That still does not make this change acceptable, it is incorrect.

Quoted 1 line

So quietly ignoring warning is ok? Why is not there BUG_ON(!release) then?

I was trying to be nice and not crash your machine and give you the opportunity to fix it easily.

Quoted 2 lines

The sysfs attributes here just represents attributes of block device object, this device is always removed before release here is called.

Then why use a kobject at all? Your reference counting will be all wrong if you use it incorrectly like this.

Quoted 7 lines
Quoted 1 line

So if there is preferred to do another alloc/free, no problem.

Please do.

Also, with the addition of this patch, your comments for the kobject are now incorrect as well.

Quoted 8 lines

Here is list of just some errors I get on my 2.6.32 linux kernel. (error message itself is removed)
Messages are printed during module removal/reboot.

'aead' (ffffffffa017c050): does not have a release() function, it is broken and must be fixed.
'bridge'
'cdrom'

Quoted 5 lines

<snip>

You will find that all of these are modules, right? That's being worked on still.

Yes - true in most cases - here are exceptions:

fill_kobj_path: path = '/devices/system/kvm/kvm0'
fill_kobj_path: path = '/devices/system/kvm'
fill_kobj_path: path = '/bus/sdio'
fill_kobj_path: path = '/bus/mmc'

Please note, these are only modules in use on my system in use - so there is probably more of them...

Zdenek