Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Fri, 18 Jan 2019 20:53:55 -0500
From:   Joel Fernandes <joel@joelfernandes.org>
To:     Hugo Lefeuvre <hle@owl.eu.com>
Cc:     Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Greg Hartman <ghartman@google.com>,
        Alistair Strachan <strachan@google.com>,
        Arve =?iso-8859-1?B?SGr4bm5lduVn?= <arve@android.com>,
        Todd Kjos <tkjos@android.com>,
        Martijn Coenen <maco@android.com>,
        Christian Brauner <christian@brauner.io>,
        Ingo Molnar <mingo@redhat.com>,
        Peter Zijlstra <peterz@infradead.org>,
        devel@driverdev.osuosl.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] sched/wait: introduce wait_event_freezable_hrtimeout
Message-ID: <20190119015355.GA115342@google.com>
References: <20190117224135.GC8100@hle-laptop.local>
 <20190118151941.GB187589@google.com>
 <20190118170801.GA4537@hle-laptop.local>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190118170801.GA4537@hle-laptop.local>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Jan 18, 2019 at 06:08:01PM +0100, Hugo Lefeuvre wrote:
[...] 
> > > +/*
> > > + * like wait_event_hrtimeout() -- except it uses TASK_INTERRUPTIBLE to avoid
> > > + * increasing load and is freezable.
> > > + */
> > > +#define wait_event_freezable_hrtimeout(wq_head, condition, timeout)		\
> > 
> > You should document the variable names in the header comments.
> 
> Agree. This comment was copy/pasted from wait_event_freezable_timeout,
> should I fix it there as well?
> 
> > Also, this new API appears to conflict with definition of 'freezable' in
> > wait_event_freezable I think,
> > 
> > wait_event_freezable - sleep or freeze until condition is true.
> > 
> > wait_event_freezable_hrtimeout - sleep but make sure freezer is not blocked.
> > (your API)
> > 
> > It seems to me these are 2 different definitions of 'freezing' (or I could be
> > completely confused). But in the first case it calls try_to_freeze after
> > schedule(). In the second case (your new API), it calls freezable_schedule().
> > 
> > So I am wondering why is there this difference in the 'meaning of freezable'.
> > In the very least, any such subtle differences should be documented in the
> > header comments IMO.
> 
> It appears that freezable_schedule() and schedule(); try_to_freeze() are
> almost identical:
> 
>     static inline void freezable_schedule(void)
>     {
>         freezer_do_not_count();
>         schedule();
>         freezer_count();
>     }
> 
> and
> 
>     static inline void freezer_do_not_count(void)
>     {
>         current->flags |= PF_FREEZER_SKIP;
>     }
> 
>     static inline void freezer_count(void)
>     {
>         current->flags &= ~PF_FREEZER_SKIP;
>         /*
>          * If freezing is in progress, the following paired with smp_mb()
>          * in freezer_should_skip() ensures that either we see %true
>          * freezing() or freezer_should_skip() sees !PF_FREEZER_SKIP.
>          */
>         smp_mb();
>         try_to_freeze();
>     }
> 
> as far as I understand this code, freezable_schedule() avoids blocking the
> freezer during the schedule() call, but in the end try_to_freeze() is still
> called so the result is the same, right?
> I wonder why wait_event_freezable is not calling freezable_schedule().

It could be something subtle in my view. freezable_schedule() actually makes
the freezer code not send a wake up to the sleeping task if a freeze happens,
because the PF_FREEZER_SKIP flag is set, as you pointed.

Whereas wait_event_freezable() which uses try_to_freeze() does not seem to have
this behavior and the task will enter the freezer. So I'm a bit skeptical if
your API will behave as expected (or at least consistently with other wait
APIs).

> That being said, I am not sure that the try_to_freeze() call does anything
> in the vsoc case because there is no call to set_freezable() so the thread
> still has PF_NOFREEZE...

I traced this, and PF_NOFREEZE is not set by default for tasks.

thanks,

 - Joel