Subject: Re: [PATCH 2/2] scsi: don't use execute_in_process_context()
From: James Bottomley <James.Bottomley@suse.de>
To: Tejun Heo <tj@kernel.org>
Cc: Linux SCSI List <linux-scsi@vger.kernel.org>,
        FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>,
        lkml <linux-kernel@vger.kernel.org>
In-Reply-To: <4D073E9A.3000608@kernel.org>
References: <4CBD95C0.6060302@kernel.org>  <4CBD95DC.8000001@kernel.org>
	 <1292194113.2989.9.camel@mulgrave.site>  <4D073E9A.3000608@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Date: Tue, 14 Dec 2010 14:09:14 +0000
Message-ID: <1292335754.3058.2.camel@mulgrave.site>
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2177
Lines: 48

On Tue, 2010-12-14 at 10:53 +0100, Tejun Heo wrote:
> Hello, James.
> 
> On 12/12/2010 11:48 PM, James Bottomley wrote:
> > The analysis above isn't quite correct, I'm afraid.  We use the
> > execute_in_process_context() not to avoid deadlocks, but to acquire
> > process context if we don't have it because the API allows calling from
> > sites at interrupt context.  The point of using
> > execute_in_process_context() is that we actually want to make use of the
> > user context if we have one ... there's no point using a workqueue in
> > that case, because there's nothing to be gained (except to slow
> > everything down).  We have no ordering constraints (the traditional
> > reason for using workqueues) so this is purely about context.
> 
> Sure, what I tried to say was that the change couldn't introduce
> deadlock no matter how it was used.  Sure execute_in_process_context()
> would be slightly more efficient, but it currently is used a few times
> only in quite cold paths where optimization isn't necessary at all and
> the API is somewhat incomplete in that it doesn't provide ordering or
> synchronization APIs.

That's the point ... it's purely for operations which require user
context which may not have it.  There's no synchronisation by design
(it's a simple API).

> So, unless there's a compelling reason, let's remove it.

The open coding of if (in_atomic()) { do workqueue stuff } else
{ execute function } is rather bug prone (most people tend to do
in_interrupt()).  It's better to encapsulate it in an API.

>   It has been
> there for quite some time now and hasn't grown any other users.  There
> wouldn't be any noticeable difference for the current users either.
> If you really like to keep it in the current users, let's move it into
> SCSI.  I don't see much reason to keep it as a part of generic
> workqueue API in its current form.

It was in SCSI ... I got told to make it generic.

James


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/