Re: [dpdk-dev] [PATCH v2] Change alarm cancel function to thread-safe:

[Neil Horman <nhorman@tuxdriver.com>]

On Tue, Sep 30, 2014 at 12:30:08PM +0000, Ananyev, Konstantin wrote:
> 
> 
> > -----Original Message-----
> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Wodkowski, PawelX
> > Sent: Tuesday, September 30, 2014 1:05 PM
> > To: Wodkowski, PawelX; Richardson, Bruce
> > Cc: dev@dpdk.org
> > Subject: Re: [dpdk-dev] [PATCH v2] Change alarm cancel function to thread-safe:
> > 
> > > -----Original Message-----
> > > Paweł
> > >
> > > > On Mon, Sep 29, 2014 at 10:11:38AM +0000, Wodkowski, PawelX wrote:
> > > > > > >
> > > > > > > Image how you will be damned by someone that not even notice you
> > > > change
> > > > > > > and he Is managing some kind of resource based on returned number of
> > > > > > > set/canceled timers. If you suddenly start returning negative values how
> > > > those
> > > > > > > application will behave? Silently changing returned value domain is evil in
> > > > its
> > > > > > > pure form.
> > > > > >
> > > > > > As I can see the impact is very limited.
> > > > >
> > > > > It is small impact to DPDK but can be huge to user application:
> > > >
> > > > This is why we traditionally have in the release-notes for each release a
> > > > section dedicated to calling out changes from one release to another. [See
> > > > http://dpdk.org/doc/intel/dpdk-release-notes-1.7.0.pdf section 5]. Since
> > > > from release-to-release there are generally only a couple of changes -
> > > > though our next release may be a little different - the actual changes are
> > > > clear enough to read about without wading through pages of documentation.
> > > I
> > > > thinking calling out the change in both the release notes and the API docs
> > > > is sufficient even for a change like this.
> > > >
> > > > Basically, I wouldn't let API stability factor in too much in trying to get
> > > > a proper fix for this issue.
> > > >
> > > > /Bruce
> > > >
> > >
> > > Summarizing all proposed solutions and to be able to produce final patch - what
> > > Is desired behavior after fix?
> > >
> > > 1. do we leave as is in Patch v2:
> > > 1.1 if canceling from other thread - if one of the alarms is executing, wait to
> > >   finish its execution and then cancel any rearmed alarms.
> > > 1.2 if canceling from alarm handler and one of the alarms to cancel is this
> > >   executing callback do no wait for it to finish and cancel anything else.
> > >
> > >  in both cases return number of canceled callbacks.
> > >
> > > 2. Do exactly like in 1. but return -EINPROGRESS instead of canceled alarms
> > >   if one of the alarms to cancel is currently executing callback from alarm thread
> > >   (information about number of canceled alarms will be lost).
> > 
> > Or instead of returning -EINPROGRESS set errno to EINPROGRESS (replace
> > returning error value by setting errno and function can always return number
> > of canceled callbacks - in error condition 0)?
> 
> Yes that's looks like a better option. 
> As I remember, inside DPDK we have our own rte_errno, that we can probably use here.
> My vote would be for that approach.
> 
You'll want to document that interface to make sure callers understand that a
non-zero return code doesn't automatically mean complete success, but yes, in
this case that seems like a pretty reasonable solution.
Neil

> Konstantin
> 
> > 
> > >
> > > 3. refuse to cancel anything if canceling currently executing alarm from alarm
> > >   callback and return -EINPROGRESS otherwise do like in 1.1.
> > >
> > > 4. Implement behaviour 1/2/3 (which?) and add API call to interrogate list of
> > >   alarms and retrun state of given alarms(s).
> > >
> > > 5. other solutions?
> > >
> > > Pawel
>