lkml.org 
[lkml]   [2010]   [Dec]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH 2/2] scsi: don't use execute_in_process_context()
From
Date
On Tue, 2010-12-14 at 15:19 +0100, Tejun Heo wrote:
> Hello, James.
>
> On 12/14/2010 03:09 PM, James Bottomley wrote:
> > 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).
>
> Well, the problem is that you do require proper synchornization
> anyway; otherwise, there is no way to make sure that the work is
> finished before the SCSI module is about to be unloaded. Currently,
> the code uses flush_scheduled_work() for this, which is going away
> because the latency can grow arbitrarily large and the behavior is
> dependent on completely unrelated work items. So, either we need to
> add a separate flush interface for ew's, flush the work item inside
> ew's or schedule them to a dedicated workqueue.

Depends what you're doing about the flush problem. The synchronisation
is inherent in the use (we're holding a reference to the module within
the executed code). The flush is to try to speed things up so the user
doesn't get annoyed during rmmod. We don't need a sync, just an
accelerator.

> >> 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.
>
> Compelling reason for it to exist. Why not just use work when you
> need execution context and the caller might or might not have one?

Because it's completely lame to have user context and not use it.

> > It was in SCSI ... I got told to make it generic.
>
> Heh, yeah, that would feel quite silly. Sorry about that. :-)
>
> But, really, let's just remove it. At this point, we either need to
> fortify the interface or remove it and given the current usage, I
> think we're better off with the latter.

I really don't think the open coding is a good idea. It's complex and
error prone; exactly the type of thing that should be in an API.

> If any pressing need arises,
> we can always add a proper API with all the necessary bells and
> whistles later.

James




\
 
 \ /
  Last update: 2010-12-14 15:29    [W:0.075 / U:0.076 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site