Re: [PATCH v2 06/16] block: treat DMPG and SWAPIN requests as special

From: Jeff Moyer <jmoyer@redhat.com>
To: Venkatraman S <svenkatr@ti.com>
Cc: <linux-mmc@vger.kernel.org>, <cjb@laptop.org>,
	<linux-mm@kvack.org>, <linux-fsdevel@vger.kernel.org>,
	<linux-omap@vger.kernel.org>, <linux-kernel@vger.kernel.org>,
	<arnd.bergmann@linaro.org>, <alex.lemberg@sandisk.com>,
	<ilan.smith@sandisk.com>, <lporzio@micron.com>,
	<rmk+kernel@arm.linux.org.uk>
Subject: Re: [PATCH v2 06/16] block: treat DMPG and SWAPIN requests as special
Date: Thu, 03 May 2012 10:38:10 -0400	[thread overview]
Message-ID: <x49fwbhl48d.fsf@segfault.boston.devel.redhat.com> (raw)
In-Reply-To: <1336054995-22988-7-git-send-email-svenkatr@ti.com> (Venkatraman S.'s message of "Thu, 3 May 2012 19:53:05 +0530")

Venkatraman S <svenkatr@ti.com> writes:

> From: Ilan Smith <ilan.smith@sandisk.com>
>
> When exp_swapin and exp_dmpg are set, treat read requests
> marked with DMPG and SWAPIN as high priority and move to
> the front of the queue.
>
[...]
> +	if (bio_swapin(bio) && blk_queue_exp_swapin(q)) {
> +		spin_lock_irq(q->queue_lock);
> +		where = ELEVATOR_INSERT_FLUSH;
> +		goto get_rq;
> +	}
> +
> +	if (bio_dmpg(bio) && blk_queue_exp_dmpg(q)) {
> +		spin_lock_irq(q->queue_lock);
> +		where = ELEVATOR_INSERT_FLUSH;
> +		goto get_rq;

Is ELEVATOR_INSERT_FRONT not good enough?  It seems wrong to use _FLUSH,
here.  If the semantics of ELEVATOR_INSERT_FLUSH are really what is
required, then perhaps we need to have another think about the naming of
these flags.

Cheers,
Jeff