From mboxrd@z Thu Jan 1 00:00:00 1970 From: =?UTF-8?B?xYF1a2FzeiBDemVyd2nFhHNraQ==?= Subject: Re: [RFC PATCH 6/6] spi: spi-s3c64xx: Move DMA initialization Date: Wed, 11 Sep 2013 09:45:58 +0200 Message-ID: <52301FB6.6060601@samsung.com> References: <1378735766-12330-1-git-send-email-l.czerwinski@samsung.com> <1378735766-12330-7-git-send-email-l.czerwinski@samsung.com> <20130909145351.GD29403@sirena.org.uk> <522F0151.50905@samsung.com> <20130910120358.GG29403@sirena.org.uk> <522F1BCB.1000400@samsung.com> <20130910163248.GO29403@sirena.org.uk> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: Received: from mailout1.w1.samsung.com ([210.118.77.11]:32558 "EHLO mailout1.w1.samsung.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751376Ab3IKHqB (ORCPT ); Wed, 11 Sep 2013 03:46:01 -0400 In-reply-to: <20130910163248.GO29403@sirena.org.uk> Sender: linux-samsung-soc-owner@vger.kernel.org List-Id: linux-samsung-soc@vger.kernel.org To: Mark Brown Cc: s.nawrocki@samsung.com, linux-samsung-soc@vger.kernel.org, linux-spi@vger.kernel.org On 09/10/2013 06:32 PM, Mark Brown wrote: > On Tue, Sep 10, 2013 at 03:16:59PM +0200, =C5=81ukasz Czerwi=C5=84ski= wrote: >> On 09/10/2013 02:03 PM, Mark Brown wrote: >>> On Tue, Sep 10, 2013 at 01:24:01PM +0200, =C5=81ukasz Czerwi=C5=84s= ki wrote: > >>>> For 16kB buffer upload time is reduced from ~90ms to ~80ms (I test= ed >>>> with S5C73M3 355560B transfer). If you think that performance >>>> increase isn't valuable we should skip that patch. > >>> That does seem worthwhile. Is that just a single transfer in isola= tion? > >> I gave you summary time when firmware is transferred via 22x16kB blo= cks. >> Each transfer was isolated. > > OK, so that's the per-transfer number? I'm just wondering if moving = it > to prepare/unprepare would achieve the same effect here. > Yes You are right. S5C73M3 splits firmware into several spi_messages with single SPI=20 transfer. It caused for each single spi_message transfer=20 prepare/unprepare. I tried mistakenly optimize through moving=20 dma_channel_request(in prepare) from prepare to probe(do it once). I based my patch on the spi-tegra114.c driver. There is=20 dma_chanel_request at the beginning in the probe function. I really should correct S5C73M3 firmware upload function. When I added small modification into S5C73M3 driver I achieved the same= =20 effect. Sorry for wasting your time. Thanks Lukasz