From mboxrd@z Thu Jan  1 00:00:00 1970
From: Vladislav Bolkhovitin <vst@vlnb.net>
Subject: Re: [PATCH 8/19]: SCST SYSFS interface implementation
Date: Wed, 10 Nov 2010 23:38:41 +0300
Message-ID: <4CDB02D1.2050400@vlnb.net>
References: <20101011213235.GA11489@kroah.com> <4CB4AEB9.30501@vlnb.net> <20101012190345.GA25737@kroah.com> <4CB75E81.7000208@vlnb.net> <20101014200413.GA30831@kroah.com> <4CC1CA4D.1090609@vlnb.net> <20101022175624.GA13640@kroah.com> <4CC1DAA2.7030602@vlnb.net> <20101022185437.GA9103@kroah.com> <4CD8566D.1020202@vlnb.net> <20101109002829.GA22633@kroah.com> <4CD9A9B8.70708@vlnb.net> <4CDA6CD4.3010308@panasas.com> <4CDAFE6E.7050200@vlnb.net> <4CDB00C2.2090200@cisco.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Return-path: <linux-kernel-owner@vger.kernel.org>
In-Reply-To: <4CDB00C2.2090200@cisco.com>
Sender: linux-kernel-owner@vger.kernel.org
To: Joe Eykholt <jeykholt@cisco.com>
Cc: Boaz Harrosh <bharrosh@panasas.com>, Greg KH <greg@kroah.com>, linux-scsi@vger.kernel.org, linux-kernel@vger.kernel.org, scst-devel <scst-devel@lists.sourceforge.net>, James Bottomley <James.Bottomley@HansenPartnership.com>, Andrew Morton <akpm@linux-foundation.org>, FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>, Mike Christie <michaelc@cs.wisc.edu>, Vu Pham <vuhuong@mellanox.com>, Bart Van Assche <bart.vanassche@gmail.com>, James Smart <James.Smart@Emulex.Com>, Andy Yan <ayan@marvell.com>, Chetan Loke <generationgnu@yahoo.com>, Dmitry Torokhov <dmitry.torokhov@gmail.com>, Hannes Reinecke <hare@suse.de>, Richard Sharpe <realrichardsharpe@gmail.com>, Daniel Henrique Debonzi <debonzi@linux.vnet.ibm.com>
List-Id: linux-scsi@vger.kernel.org

Joe Eykholt, on 11/10/2010 11:29 PM wrote:
>> Thanks for sharing your thoughts with us. But the question isn't about
>> if it's possible to implement what we need locklessly. The question is
>> in two approaches how to synchronously delete objects with entries on SYSFS:
>>
>> 1. struct object_x {
>> 	...
>> 	struct kobject kobj;
>> 	struct completion *release_completion;
>> };
>>
>> static void x_release(struct kobject *kobj)
>> {
>> 	struct object_x *x;
>> 	struct completion *c;
>>
>> 	x = container_of(kobj, struct object_x, kobj);
>> 	c = x->release_completion;
>> 	kfree(x);
>> 	complete_all(c);
>> }
>>
>> void del_object(struct object_x *x)
>> {
>> 	DECLARE_COMPLETION_ONSTACK(completion);
>>
>> 	...
>> 	x->release_completion = &completion;
>> 	kobject_put(&x->kobj);
>> 	wait_for_completion(&completion);
>> }
>>
>> and
>>
>> 2. struct object_x {
>> 	...
>> 	struct kobject kobj;
>> 	struct completion release_completion;
>> };
>>
>> static void x_release(struct kobject *kobj)
>> {
>> 	struct object_x *x;
>>
>> 	x = container_of(kobj, struct object_x, kobj);
>> 	complete_all(&x->release_completion);
>> }
>>
>> void del_object(struct object_x *x)
>> {
>> 	...
>> 	kobject_put(&x->kobj);
>> 	wait_for_completion(&completion);
>> 	...
>> 	kfree(x);
>> }
> 
> I'll admit I don't understand this all that well, but
> why not just have x_release() (based on (2))
> do free(x), and have del_object
> do the kobject_put(&x->kobj) as its very last thing?
> Then you don't need the completion.

We are discussing _synchronous_ delete of x, so need to wait until
x->kobj released, hence the completion is needed in both cases.

For instance, the sync delete is needed for targets to let the
corresponding target driver be safely unloaded after its target
unregistered.

Vlad