From mboxrd@z Thu Jan 1 00:00:00 1970 From: Heiko =?iso-8859-1?q?St=FCbner?= Subject: Re: [PATCH 6/7] spi/s3c64xx: Add support DMA engine API Date: Mon, 4 Jul 2011 21:51:43 +0200 Message-ID: <201107042151.44075.heiko@sntech.de> References: <1309781915-31549-1-git-send-email-kgene.kim@samsung.com> <201107041859.11548.heiko@sntech.de> <20110704170217.GH28042@ponder.secretlab.ca> Mime-Version: 1.0 Content-Type: Text/Plain; charset=iso-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: Received: from s15407518.onlinehome-server.info ([82.165.136.167]:47508 "EHLO s15407518.onlinehome-server.info" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751585Ab1GDTvu convert rfc822-to-8bit (ORCPT ); Mon, 4 Jul 2011 15:51:50 -0400 In-Reply-To: <20110704170217.GH28042@ponder.secretlab.ca> Sender: linux-samsung-soc-owner@vger.kernel.org List-Id: linux-samsung-soc@vger.kernel.org To: Grant Likely Cc: Kukjin Kim , linux-arm-kernel@lists.infradead.org, linux-samsung-soc@vger.kernel.org, Vinod Koul , Dan Williams , Jassi Brar , Liam Girdwood , Mark Brown , Boojin Kim Am Montag 04 Juli 2011, 19:02:17 schrieb Grant Likely: > On Mon, Jul 04, 2011 at 06:59:11PM +0200, Heiko St=FCbner wrote: > > Am Montag 04 Juli 2011, 18:42:51 schrieb Grant Likely: > > > On Mon, Jul 04, 2011 at 09:18:34PM +0900, Kukjin Kim wrote: > > > > +#if defined(CONFIG_DMADEV_PL330) > > > > + memset(&slave_config, 0, sizeof(slave_config)); > > > > + slave_config.direction =3D DMA_TO_DEVICE; > > > > + slave_config.src_addr =3D xfer->tx_dma; > > > > + slave_config.dst_addr =3D > > > > + sdd->sfr_start + S3C64XX_SPI_TX_DATA; > > > > + slave_config.dst_addr_width =3D sdd->cur_bpw / 8; > > > > + dmaengine_slave_config(sdd->tx_chan, &slave_config); > > > > + > > > > + sg_init_table(&tx_sg, 1); > > > > + sg_set_page(&tx_sg, pfn_to_page(PFN_DOWN(xfer- >tx_dma)), > > > > + xfer->len, offset_in_page(xfer->tx_dma)); > > > > + sg_dma_len(&tx_sg) =3D xfer->len; > > > > + sg_dma_address(&tx_sg) =3D xfer->tx_dma; > > > > + sdd->tx_desc =3D > > > > + sdd->tx_chan->device->device_prep_slave_sg( > > > > + sdd->tx_chan, &tx_sg, 1, DMA_TO_DEVICE, > > > > + DMA_PREP_INTERRUPT); > > > > + sdd->tx_desc->callback =3D s3c64xx_spi_dma_txcb; > > > > + sdd->tx_desc->callback_param =3D sdd; > > > > + dmaengine_submit(sdd->tx_desc); > > > > + dma_async_issue_pending(sdd->tx_chan); > > > > +#else > > > >=20 > > > > s3c2410_dma_config(sdd->tx_dmach, sdd->cur_bpw / 8); > > > > s3c2410_dma_enqueue(sdd->tx_dmach, (void *)sdd, > > > > =09 > > > > xfer->tx_dma, xfer->len); > > > > =09 > > > > s3c2410_dma_ctrl(sdd->tx_dmach, S3C2410_DMAOP_START); > > > >=20 > > > > +#endif > > >=20 > > > Hmmm, this is not pretty. The driver behaviour is entirely diffe= rent > > > depending on if CONFIG_DMADEV_PL330 is enabled? When we get to > > > multiplatform kernels, is this going to break on some hardware? > > >=20 > > > > @@ -802,8 +951,13 @@ static void s3c64xx_spi_work(struct work_s= truct > > > > *work) > > > >=20 > > > > spin_unlock_irqrestore(&sdd->lock, flags); > > > > =09 > > > > /* Free DMA channels */ > > > >=20 > > > > +#if defined(CONFIG_DMADEV_PL330) > > > > + dma_release_channel(sdd->tx_chan); > > > > + dma_release_channel(sdd->rx_chan); > > > > +#else > > > >=20 > > > > s3c2410_dma_free(sdd->tx_dmach, &s3c64xx_spi_dma_client); > > > > s3c2410_dma_free(sdd->rx_dmach, &s3c64xx_spi_dma_client); > > > >=20 > > > > +#endif > > >=20 > > > Wow. A lot of #ifdefs here. It does not look multiplatform frie= ndly > > > at all. Are the s3c2410_dma functions obsolete when DMADEV_PL330= is > > > selected? If so, can they be removed entirely, or are they requi= red > > > to support certain hardware? > >=20 > > The spi_s3c64xx driver can also support the SPI controller of the > > S3C2416/S3C2443 line of SoCs. > > As I'm currently working on a S3C2416 based project, my small wish = from > > the sidelines would be to not break this support with whatever solu= tion > > you will decide on :-) . >=20 > I will not merge a patch that either breaks a platform, or requires > a compile time either/or choice of device support (enabling support > for one device should not break support for another). The patch above seems to contain the support for both SoCs (i.e. s3c241= 0_dma_*=20 for S3C2416 etc), so it wouldn't break the support. But this form will definitly break future multiplatform kernels when th= e 2416=20 and some variant using the DMADEV_PL330 are selected at the same time. The 2416/2443 seem to be the first Samsung-SoCs to have a SPI-controlle= r of=20 this type. Implementing the DMA engine stuff for these SoCs would obvio= usly=20 solve the ifdef-problem but I don't know if this is possible to do. Heiko From mboxrd@z Thu Jan 1 00:00:00 1970 From: heiko@sntech.de (Heiko =?iso-8859-1?q?St=FCbner?=) Date: Mon, 4 Jul 2011 21:51:43 +0200 Subject: [PATCH 6/7] spi/s3c64xx: Add support DMA engine API In-Reply-To: <20110704170217.GH28042@ponder.secretlab.ca> References: <1309781915-31549-1-git-send-email-kgene.kim@samsung.com> <201107041859.11548.heiko@sntech.de> <20110704170217.GH28042@ponder.secretlab.ca> Message-ID: <201107042151.44075.heiko@sntech.de> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org Am Montag 04 Juli 2011, 19:02:17 schrieb Grant Likely: > On Mon, Jul 04, 2011 at 06:59:11PM +0200, Heiko St?bner wrote: > > Am Montag 04 Juli 2011, 18:42:51 schrieb Grant Likely: > > > On Mon, Jul 04, 2011 at 09:18:34PM +0900, Kukjin Kim wrote: > > > > +#if defined(CONFIG_DMADEV_PL330) > > > > + memset(&slave_config, 0, sizeof(slave_config)); > > > > + slave_config.direction = DMA_TO_DEVICE; > > > > + slave_config.src_addr = xfer->tx_dma; > > > > + slave_config.dst_addr = > > > > + sdd->sfr_start + S3C64XX_SPI_TX_DATA; > > > > + slave_config.dst_addr_width = sdd->cur_bpw / 8; > > > > + dmaengine_slave_config(sdd->tx_chan, &slave_config); > > > > + > > > > + sg_init_table(&tx_sg, 1); > > > > + sg_set_page(&tx_sg, pfn_to_page(PFN_DOWN(xfer- >tx_dma)), > > > > + xfer->len, offset_in_page(xfer->tx_dma)); > > > > + sg_dma_len(&tx_sg) = xfer->len; > > > > + sg_dma_address(&tx_sg) = xfer->tx_dma; > > > > + sdd->tx_desc = > > > > + sdd->tx_chan->device->device_prep_slave_sg( > > > > + sdd->tx_chan, &tx_sg, 1, DMA_TO_DEVICE, > > > > + DMA_PREP_INTERRUPT); > > > > + sdd->tx_desc->callback = s3c64xx_spi_dma_txcb; > > > > + sdd->tx_desc->callback_param = sdd; > > > > + dmaengine_submit(sdd->tx_desc); > > > > + dma_async_issue_pending(sdd->tx_chan); > > > > +#else > > > > > > > > s3c2410_dma_config(sdd->tx_dmach, sdd->cur_bpw / 8); > > > > s3c2410_dma_enqueue(sdd->tx_dmach, (void *)sdd, > > > > > > > > xfer->tx_dma, xfer->len); > > > > > > > > s3c2410_dma_ctrl(sdd->tx_dmach, S3C2410_DMAOP_START); > > > > > > > > +#endif > > > > > > Hmmm, this is not pretty. The driver behaviour is entirely different > > > depending on if CONFIG_DMADEV_PL330 is enabled? When we get to > > > multiplatform kernels, is this going to break on some hardware? > > > > > > > @@ -802,8 +951,13 @@ static void s3c64xx_spi_work(struct work_struct > > > > *work) > > > > > > > > spin_unlock_irqrestore(&sdd->lock, flags); > > > > > > > > /* Free DMA channels */ > > > > > > > > +#if defined(CONFIG_DMADEV_PL330) > > > > + dma_release_channel(sdd->tx_chan); > > > > + dma_release_channel(sdd->rx_chan); > > > > +#else > > > > > > > > s3c2410_dma_free(sdd->tx_dmach, &s3c64xx_spi_dma_client); > > > > s3c2410_dma_free(sdd->rx_dmach, &s3c64xx_spi_dma_client); > > > > > > > > +#endif > > > > > > Wow. A lot of #ifdefs here. It does not look multiplatform friendly > > > at all. Are the s3c2410_dma functions obsolete when DMADEV_PL330 is > > > selected? If so, can they be removed entirely, or are they required > > > to support certain hardware? > > > > The spi_s3c64xx driver can also support the SPI controller of the > > S3C2416/S3C2443 line of SoCs. > > As I'm currently working on a S3C2416 based project, my small wish from > > the sidelines would be to not break this support with whatever solution > > you will decide on :-) . > > I will not merge a patch that either breaks a platform, or requires > a compile time either/or choice of device support (enabling support > for one device should not break support for another). The patch above seems to contain the support for both SoCs (i.e. s3c2410_dma_* for S3C2416 etc), so it wouldn't break the support. But this form will definitly break future multiplatform kernels when the 2416 and some variant using the DMADEV_PL330 are selected at the same time. The 2416/2443 seem to be the first Samsung-SoCs to have a SPI-controller of this type. Implementing the DMA engine stuff for these SoCs would obviously solve the ifdef-problem but I don't know if this is possible to do. Heiko