Date: Tue, 1 May 2012 23:39:34 +0200
From: Andi Shyti <andi.shyti@gmail.com>
To: scameron@beardog.cce.hp.com
Cc: james.bottomley@hansenpartnership.com, linux-scsi@vger.kernel.org,
        linux-kernel@vger.kernel.org, stephenmcameron@gmail.com,
        thenzl@redhat.com, akpm@linux-foundation.org, mikem@beardog.cce.hp.com
Subject: Re: [PATCH 07/17] hpsa: do not give up retry of driver cmds after
 only 3 retries
Message-ID: <20120501213934.GD11302@andi>
Mail-Followup-To: scameron@beardog.cce.hp.com,
	james.bottomley@hansenpartnership.com, linux-scsi@vger.kernel.org,
	linux-kernel@vger.kernel.org, stephenmcameron@gmail.com,
	thenzl@redhat.com, akpm@linux-foundation.org,
	mikem@beardog.cce.hp.com
References: <20120501163819.11705.10299.stgit@beardog.cce.hp.com>
 <20120501164240.11705.10308.stgit@beardog.cce.hp.com>
 <20120501172634.GA11302@andi>
 <20120501182011.GS11802@beardog.cce.hp.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20120501182011.GS11802@beardog.cce.hp.com>
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2836
Lines: 69

Hi Steve,

Thanks for the walk-through, but still some doubts...

On Tue, May 01, 2012 at 01:20:11PM -0500, scameron@beardog.cce.hp.com wrote:
> On Tue, May 01, 2012 at 07:26:34PM +0200, Andi Shyti wrote:
> > >  
> > >  	do {
> > >  		memset(c->err_info, 0, sizeof(*c->err_info));
> > >  		hpsa_scsi_do_simple_cmd_core(h, c);
> > >  		retry_count++;
> > > +		if (retry_count > 3) {
> > > +			msleep(backoff_time);
> > 
> > for 10ms isn't it better to avoid using msleep?
> 
> [...] 
> > > +			if (backoff_time < 1000)
> > > +				backoff_time *= 2;
> 
> Eh, maybe.  from Documentation/timers-howto.txt
> 
>                         msleep(1~20) may not do what the caller intends, and
>                         will often sleep longer (~20 ms actual sleep for any
>                         value given in the 1~20ms range). In many cases this
>                         is not the desired behavior.
> 
> Sleeping longer (~20ms instead of 10ms) in this instance is fine, as I don't
> really care too much exactly how long it sleeps, and it backs off to up to
> 1280ms eventually anyway.  The idea is, "wait a bit, and retry, and then if
> that doesn't work, wait twice as long, and retry, etc."  *exactly* how long
> "a bit" is is not super important.  I could change the initial back_off time
> to 20 or 30 to satisfy the letter of the advice in Documentation/timers-howto.txt,
> if doing so is important.

No, you're right, it should not really matter, but here in the
worst case you put the driver on sleep for almost 22 seconds,
that is a huge difference compared to the original
implementation.
 
> This is kind of a corner case of a corner case, I don't expect
> things will ordinarily end up waiting that long, because normally
> one of the 1st 3 retries will succeed.  I just wanted to make it 
> a little more robust and not just give up immediately if the 3
> initial retries don't succeed, the specific number of retries,
> wait times, etc, I just made up.

Premising that I don't know the device, therefore I could be
totally wrong, if you don't expect things to wait so long, why not
to decrease the MAX_DRIVER_CMD_RETRIES and sleep increasingly (as
you did) but for shorter period?

Andi

> It still does eventually give up
> though, and then probably doesn't do anything good after that
> (same as current behavior, just somewhat less likely to get to
> that point.)  I'm not actually aware of any complaints of the 
> retries failing though (apart from the complaint that prompted
> the patch prior to this one, that we didn't retry on getting
> SAM_STAT_BUSY.)
> 
> -- steve
> 
--
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/