Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 3 Nov 2022 12:05:06 +0900
From:   Sergey Senozhatsky <senozhatsky@chromium.org>
To:     Minchan Kim <minchan@kernel.org>
Cc:     Sergey Senozhatsky <senozhatsky@chromium.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        Nitin Gupta <ngupta@vflare.org>, linux-kernel@vger.kernel.org,
        linux-mm@kvack.org
Subject: Re: [PATCHv4 2/9] zram: Add recompression algorithm sysfs knob
Message-ID: <Y2Mv4l+V9iCv9EMg@google.com>
References: <20221018045533.2396670-1-senozhatsky@chromium.org>
 <20221018045533.2396670-3-senozhatsky@chromium.org>
 <Y2LP0OWF/WTnkSne@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Y2LP0OWF/WTnkSne@google.com>
Precedence: bulk

On (22/11/02 13:15), Minchan Kim wrote:
[..]
> >  /* Module params (documentation at end) */
> >  static unsigned int num_devices = 1;
> > @@ -1000,31 +1005,37 @@ static ssize_t max_comp_streams_store(struct device *dev,
> >  	return len;
> >  }
> >  
> > -static ssize_t comp_algorithm_show(struct device *dev,
> > -		struct device_attribute *attr, char *buf)
> 
> Do you have any reason to change show and set placement? Otherwise,
> please keep the function order to reduce unnecesssary churns.

I don't change their placement. It's just show and store for primary and
secondary algorithms use the same __store and __show functions, which
are static and are placed ahead of store and show.

[..]
> Just open question(I might be too paranoid?)
> 
> I am thinking someone want to add third comp algorithm in future
> to balance decompression and memory efficiency.
> 
> If it's not too crazy idea, let's think about the interface.
> Maybe, could we make the recomp knobs works like list?
> 
> # A primary comp
> echo "A" > /zram/comp_algo
> 
> # Multiple secondary comps
> echo "B threshold" > /zram/add_recomp_algo
> echo "C threshold" > /zram/add_recomp_algo
> echo "D threshold" > /zram/add_recomp_algo

What is the threshold here? My design approach is that ZRAM doesn't do
recompression on its own, so no magic is happening automatically. It's
the user-space that triggers recompression for selected pages when
user-space thinks it's time to. This allows us to have various flexible
policies and consider things that ZRAM is not even aware of: battery level,
free memory, CPU load average, etc. E.g. no recompression when all CPUs
are busy rendering video game, or when we are draining battery too fast,
etc.

> "cat /zram/recomp_algo" shows the list
> 
> echo "C" > /zram/remove_recomp_algo
> will remove the C algorithm in stack.

What is the use case for removal of a secondary algorithm?

> My point is that we don't need to implement it atm but makes the
> interface to open the possibility for future extension.
> 
> What do you think?

So, as far as I understand, we don't have reason to add remove_recomp_algo
right now. And existing recomp_algo does not enforce any particular format,
it can be extended. Right now we accept "$name" but can do something like
"$name:$priority". The only thing that we probably need to do is rename
recomp_algo to either add_recomp_algo or register_recomp_algo?

> > +static ssize_t recomp_algorithm_store(struct device *dev,
> > +				      struct device_attribute *attr,
> > +				      const char *buf,
> > +				      size_t len)
> > +{
> > +	struct zram *zram = dev_to_zram(dev);
> > +	int ret;
> > +
> > +	ret = __comp_algorithm_store(zram, ZRAM_SECONDARY_ZCOMP, buf);
> > +	return ret ? ret : len;
> > +}
> > +#endif
> > +
> >  static ssize_t compact_store(struct device *dev,
> >  		struct device_attribute *attr, const char *buf, size_t len)
> >  {
> > @@ -1762,7 +1817,11 @@ static void zram_reset_device(struct zram *zram)
> >  	memset(&zram->stats, 0, sizeof(zram->stats));
> >  	reset_bdev(zram);
> >  
> > -	comp_algorithm_set(zram, ZRAM_PRIMARY_ZCOMP, default_compressor);
> > +	comp_algorithm_set(zram, ZRAM_PRIMARY_ZCOMP,
> > +			   default_comp_algs[ZRAM_PRIMARY_ZCOMP]);
> > +	if (IS_ENABLED(CONFIG_ZRAM_MULTI_COMP))
> 
> Dumb question:
> 
> Why do you use IS_ENABLED instead of ifdef?

#ifdef-s are banned in the new C-code, as far as I know. IS_ENABLED is
what we should use.