All of lore.kernel.org
 help / color / mirror / Atom feed
From: Stefan Agner <stefan@agner.ch>
To: Jingchang Lu <jingchang.lu@freescale.com>
Cc: vinod.koul@intel.com, dmaengine@vger.kernel.org,
	dan.j.williams@intel.com, linux-kernel@vger.kernel.org,
	linux-arm-kernel@lists.infradead.org
Subject: Re: [PATCH 1/2] dmaengine: fsl-edma: add dma memcpy support
Date: Fri, 30 Jan 2015 13:39:03 +0100	[thread overview]
Message-ID: <63612a565e0cbf4081b0b8efa0bc3c46@agner.ch> (raw)
In-Reply-To: <1419928907-18053-2-git-send-email-jingchang.lu@freescale.com>

On 2014-12-30 09:41, Jingchang Lu wrote:
> This adds the memory copy capability support, the memcpy
> functionality needs to configure the DMAMUX of the channel
> with the always on slave id.

Hi Jingchang,

I run some tests on v3.19-rc6 with this patches applied on Vybrid SoC
VF500 and VF610.

On VF500 clocked at 400MHz I get this numbers:
# insmod dmatest.ko max_channels=1 iterations=100 run=1
[  616.809594] dmatest: Started 1 threads using dma0chan0
[  617.293498] dmatest: dma0chan0-copy0: summary 100 tests, 0 failures
209 iops 1613 KB/s (0)

In contrast, on a VF610 clocked at 500MHz I get this:
# insmod dmatest.ko max_channels=1 iterations=100 run=1
[  154.203290] dmatest: Started 1 threads using dma0chan4
[  154.614225] dmatest: dma0chan4-copy0: summary 100 tests, 0 failures
246 iops 2002 KB/s (0)

Not that huge numbers, but I guess this is what the hardware can
deliver... Did you run some performance tests?

But generally, looks good.

Tested-by: Stefan Agner <stefan@agner.ch>

> 
> Signed-off-by: Jingchang Lu <jingchang.lu@freescale.com>
> ---
>  drivers/dma/fsl-edma.c | 63 ++++++++++++++++++++++++++++++++++++++++++++++++--
>  1 file changed, 61 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/dma/fsl-edma.c b/drivers/dma/fsl-edma.c
> index 09e2842..e0bd517 100644
> --- a/drivers/dma/fsl-edma.c
> +++ b/drivers/dma/fsl-edma.c
> @@ -110,6 +110,8 @@
>  #define EDMAMUX_CHCFG_ENBL		0x80
>  #define EDMAMUX_CHCFG_SOURCE(n)		((n) & 0x3F)
>  
> +#define SLAVE_ID_ALWAYSON	63 /* the always on slave id */
> +
>  #define DMAMUX_NR	2
>  
>  #define FSL_EDMA_BUSWIDTHS	BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
> @@ -147,6 +149,7 @@ struct fsl_edma_slave_config {
>  struct fsl_edma_chan {
>  	struct virt_dma_chan		vchan;
>  	enum dma_status			status;
> +	u32				slave_id;
>  	struct fsl_edma_engine		*edma;
>  	struct fsl_edma_desc		*edesc;
>  	struct fsl_edma_slave_config	fsc;
> @@ -637,6 +640,45 @@ static struct dma_async_tx_descriptor
> *fsl_edma_prep_slave_sg(
>  	return vchan_tx_prep(&fsl_chan->vchan, &fsl_desc->vdesc, flags);
>  }
>  
> +static struct dma_async_tx_descriptor *
> +fsl_edma_prep_memcpy(struct dma_chan *chan, dma_addr_t dst,
> +			dma_addr_t src, size_t len, unsigned long tx_flags)
> +{
> +	struct fsl_edma_chan *fsl_chan = to_fsl_edma_chan(chan);
> +	struct fsl_edma_desc *fsl_desc;
> +	u16 soff, doff, attr;
> +
> +	/*
> +	 * use 4-bytes data transfer size if all is 4-bytes aligned,
> +	 * else 2-bytes data transfer size of all is 2-bytes aligned,
> +	 * otherwise 1-byte tranfer size.
> +	 */
> +	if (src & 0x1 || dst & 0x1 || len & 0x1) {
> +		attr = EDMA_TCD_ATTR_SSIZE_8BIT | EDMA_TCD_ATTR_DSIZE_8BIT;
> +		soff = 0x1;
> +		doff = 0x1;
> +	} else if (src & 0x2 || dst & 0x2 || len & 0x2) {
> +		attr = EDMA_TCD_ATTR_SSIZE_16BIT | EDMA_TCD_ATTR_DSIZE_16BIT;
> +		soff = 0x2;
> +		doff = 0x2;
> +	} else {
> +		attr = EDMA_TCD_ATTR_SSIZE_32BIT | EDMA_TCD_ATTR_DSIZE_32BIT;
> +		soff = 0x4;
> +		doff = 0x4;
> +	}
> +
> +	fsl_desc = fsl_edma_alloc_desc(fsl_chan, 1);
> +	if (!fsl_desc)
> +		return NULL;
> +	fsl_desc->iscyclic = false;
> +
> +	fsl_edma_fill_tcd(fsl_desc->tcd[0].vtcd, src, dst, attr, soff, len,
> +			  0, 1, 1, doff, 0, true, true, false);
> +
> +	return vchan_tx_prep(&fsl_chan->vchan, &fsl_desc->vdesc, tx_flags);
> +
> +}
> +
>  static void fsl_edma_xfer_desc(struct fsl_edma_chan *fsl_chan)
>  {
>  	struct virt_dma_desc *vdesc;
> @@ -735,6 +777,7 @@ static struct dma_chan *fsl_edma_xlate(struct
> of_phandle_args *dma_spec,
>  {
>  	struct fsl_edma_engine *fsl_edma = ofdma->of_dma_data;
>  	struct dma_chan *chan, *_chan;
> +	struct fsl_edma_chan *fsl_chan;
>  	unsigned long chans_per_mux = fsl_edma->n_chans / DMAMUX_NR;
>  
>  	if (dma_spec->args_count != 2)
> @@ -748,8 +791,10 @@ static struct dma_chan *fsl_edma_xlate(struct
> of_phandle_args *dma_spec,
>  			chan = dma_get_slave_channel(chan);
>  			if (chan) {
>  				chan->device->privatecnt++;
> -				fsl_edma_chan_mux(to_fsl_edma_chan(chan),
> -					dma_spec->args[1], true);
> +				fsl_chan = to_fsl_edma_chan(chan);
> +				fsl_chan->slave_id = dma_spec->args[1];
> +				fsl_edma_chan_mux(fsl_chan, fsl_chan->slave_id,
> +						  true);
>  				mutex_unlock(&fsl_edma->fsl_edma_mutex);
>  				return chan;
>  			}
> @@ -763,6 +808,17 @@ static int fsl_edma_alloc_chan_resources(struct
> dma_chan *chan)
>  {
>  	struct fsl_edma_chan *fsl_chan = to_fsl_edma_chan(chan);
>  
> +	/*
> +	 * If the slave id of the channel DMAMUX is not set yet,
> +	 * this could happy when the channel is requested by the
> +	 * dma_request_channel() for memory copy purpose instead
> +	 * of by dts binding, then configure the DMAMUX with the
> +	 * always on slave id.
> +	 */
> +	if (fsl_chan->slave_id == 0) {
> +		fsl_chan->slave_id = SLAVE_ID_ALWAYSON;
> +		fsl_edma_chan_mux(fsl_chan, fsl_chan->slave_id, true);
> +	}
>  	fsl_chan->tcd_pool = dma_pool_create("tcd_pool", chan->device->dev,
>  				sizeof(struct fsl_edma_hw_tcd),
>  				32, 0);
> @@ -778,6 +834,7 @@ static void fsl_edma_free_chan_resources(struct
> dma_chan *chan)
>  	spin_lock_irqsave(&fsl_chan->vchan.lock, flags);
>  	fsl_edma_disable_request(fsl_chan);
>  	fsl_edma_chan_mux(fsl_chan, 0, false);
> +	fsl_chan->slave_id = 0;
>  	fsl_chan->edesc = NULL;
>  	vchan_get_all_descriptors(&fsl_chan->vchan, &head);
>  	spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
> @@ -903,6 +960,7 @@ static int fsl_edma_probe(struct platform_device *pdev)
>  	dma_cap_set(DMA_PRIVATE, fsl_edma->dma_dev.cap_mask);
>  	dma_cap_set(DMA_SLAVE, fsl_edma->dma_dev.cap_mask);
>  	dma_cap_set(DMA_CYCLIC, fsl_edma->dma_dev.cap_mask);
> +	dma_cap_set(DMA_MEMCPY, fsl_edma->dma_dev.cap_mask);
>  
>  	fsl_edma->dma_dev.dev = &pdev->dev;
>  	fsl_edma->dma_dev.device_alloc_chan_resources
> @@ -912,6 +970,7 @@ static int fsl_edma_probe(struct platform_device *pdev)
>  	fsl_edma->dma_dev.device_tx_status = fsl_edma_tx_status;
>  	fsl_edma->dma_dev.device_prep_slave_sg = fsl_edma_prep_slave_sg;
>  	fsl_edma->dma_dev.device_prep_dma_cyclic = fsl_edma_prep_dma_cyclic;
> +	fsl_edma->dma_dev.device_prep_dma_memcpy = fsl_edma_prep_memcpy;
>  	fsl_edma->dma_dev.device_config = fsl_edma_slave_config;
>  	fsl_edma->dma_dev.device_pause = fsl_edma_pause;
>  	fsl_edma->dma_dev.device_resume = fsl_edma_resume;

WARNING: multiple messages have this Message-ID (diff)
From: stefan@agner.ch (Stefan Agner)
To: linux-arm-kernel@lists.infradead.org
Subject: [PATCH 1/2] dmaengine: fsl-edma: add dma memcpy support
Date: Fri, 30 Jan 2015 13:39:03 +0100	[thread overview]
Message-ID: <63612a565e0cbf4081b0b8efa0bc3c46@agner.ch> (raw)
In-Reply-To: <1419928907-18053-2-git-send-email-jingchang.lu@freescale.com>

On 2014-12-30 09:41, Jingchang Lu wrote:
> This adds the memory copy capability support, the memcpy
> functionality needs to configure the DMAMUX of the channel
> with the always on slave id.

Hi Jingchang,

I run some tests on v3.19-rc6 with this patches applied on Vybrid SoC
VF500 and VF610.

On VF500 clocked at 400MHz I get this numbers:
# insmod dmatest.ko max_channels=1 iterations=100 run=1
[  616.809594] dmatest: Started 1 threads using dma0chan0
[  617.293498] dmatest: dma0chan0-copy0: summary 100 tests, 0 failures
209 iops 1613 KB/s (0)

In contrast, on a VF610 clocked at 500MHz I get this:
# insmod dmatest.ko max_channels=1 iterations=100 run=1
[  154.203290] dmatest: Started 1 threads using dma0chan4
[  154.614225] dmatest: dma0chan4-copy0: summary 100 tests, 0 failures
246 iops 2002 KB/s (0)

Not that huge numbers, but I guess this is what the hardware can
deliver... Did you run some performance tests?

But generally, looks good.

Tested-by: Stefan Agner <stefan@agner.ch>

> 
> Signed-off-by: Jingchang Lu <jingchang.lu@freescale.com>
> ---
>  drivers/dma/fsl-edma.c | 63 ++++++++++++++++++++++++++++++++++++++++++++++++--
>  1 file changed, 61 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/dma/fsl-edma.c b/drivers/dma/fsl-edma.c
> index 09e2842..e0bd517 100644
> --- a/drivers/dma/fsl-edma.c
> +++ b/drivers/dma/fsl-edma.c
> @@ -110,6 +110,8 @@
>  #define EDMAMUX_CHCFG_ENBL		0x80
>  #define EDMAMUX_CHCFG_SOURCE(n)		((n) & 0x3F)
>  
> +#define SLAVE_ID_ALWAYSON	63 /* the always on slave id */
> +
>  #define DMAMUX_NR	2
>  
>  #define FSL_EDMA_BUSWIDTHS	BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
> @@ -147,6 +149,7 @@ struct fsl_edma_slave_config {
>  struct fsl_edma_chan {
>  	struct virt_dma_chan		vchan;
>  	enum dma_status			status;
> +	u32				slave_id;
>  	struct fsl_edma_engine		*edma;
>  	struct fsl_edma_desc		*edesc;
>  	struct fsl_edma_slave_config	fsc;
> @@ -637,6 +640,45 @@ static struct dma_async_tx_descriptor
> *fsl_edma_prep_slave_sg(
>  	return vchan_tx_prep(&fsl_chan->vchan, &fsl_desc->vdesc, flags);
>  }
>  
> +static struct dma_async_tx_descriptor *
> +fsl_edma_prep_memcpy(struct dma_chan *chan, dma_addr_t dst,
> +			dma_addr_t src, size_t len, unsigned long tx_flags)
> +{
> +	struct fsl_edma_chan *fsl_chan = to_fsl_edma_chan(chan);
> +	struct fsl_edma_desc *fsl_desc;
> +	u16 soff, doff, attr;
> +
> +	/*
> +	 * use 4-bytes data transfer size if all is 4-bytes aligned,
> +	 * else 2-bytes data transfer size of all is 2-bytes aligned,
> +	 * otherwise 1-byte tranfer size.
> +	 */
> +	if (src & 0x1 || dst & 0x1 || len & 0x1) {
> +		attr = EDMA_TCD_ATTR_SSIZE_8BIT | EDMA_TCD_ATTR_DSIZE_8BIT;
> +		soff = 0x1;
> +		doff = 0x1;
> +	} else if (src & 0x2 || dst & 0x2 || len & 0x2) {
> +		attr = EDMA_TCD_ATTR_SSIZE_16BIT | EDMA_TCD_ATTR_DSIZE_16BIT;
> +		soff = 0x2;
> +		doff = 0x2;
> +	} else {
> +		attr = EDMA_TCD_ATTR_SSIZE_32BIT | EDMA_TCD_ATTR_DSIZE_32BIT;
> +		soff = 0x4;
> +		doff = 0x4;
> +	}
> +
> +	fsl_desc = fsl_edma_alloc_desc(fsl_chan, 1);
> +	if (!fsl_desc)
> +		return NULL;
> +	fsl_desc->iscyclic = false;
> +
> +	fsl_edma_fill_tcd(fsl_desc->tcd[0].vtcd, src, dst, attr, soff, len,
> +			  0, 1, 1, doff, 0, true, true, false);
> +
> +	return vchan_tx_prep(&fsl_chan->vchan, &fsl_desc->vdesc, tx_flags);
> +
> +}
> +
>  static void fsl_edma_xfer_desc(struct fsl_edma_chan *fsl_chan)
>  {
>  	struct virt_dma_desc *vdesc;
> @@ -735,6 +777,7 @@ static struct dma_chan *fsl_edma_xlate(struct
> of_phandle_args *dma_spec,
>  {
>  	struct fsl_edma_engine *fsl_edma = ofdma->of_dma_data;
>  	struct dma_chan *chan, *_chan;
> +	struct fsl_edma_chan *fsl_chan;
>  	unsigned long chans_per_mux = fsl_edma->n_chans / DMAMUX_NR;
>  
>  	if (dma_spec->args_count != 2)
> @@ -748,8 +791,10 @@ static struct dma_chan *fsl_edma_xlate(struct
> of_phandle_args *dma_spec,
>  			chan = dma_get_slave_channel(chan);
>  			if (chan) {
>  				chan->device->privatecnt++;
> -				fsl_edma_chan_mux(to_fsl_edma_chan(chan),
> -					dma_spec->args[1], true);
> +				fsl_chan = to_fsl_edma_chan(chan);
> +				fsl_chan->slave_id = dma_spec->args[1];
> +				fsl_edma_chan_mux(fsl_chan, fsl_chan->slave_id,
> +						  true);
>  				mutex_unlock(&fsl_edma->fsl_edma_mutex);
>  				return chan;
>  			}
> @@ -763,6 +808,17 @@ static int fsl_edma_alloc_chan_resources(struct
> dma_chan *chan)
>  {
>  	struct fsl_edma_chan *fsl_chan = to_fsl_edma_chan(chan);
>  
> +	/*
> +	 * If the slave id of the channel DMAMUX is not set yet,
> +	 * this could happy when the channel is requested by the
> +	 * dma_request_channel() for memory copy purpose instead
> +	 * of by dts binding, then configure the DMAMUX with the
> +	 * always on slave id.
> +	 */
> +	if (fsl_chan->slave_id == 0) {
> +		fsl_chan->slave_id = SLAVE_ID_ALWAYSON;
> +		fsl_edma_chan_mux(fsl_chan, fsl_chan->slave_id, true);
> +	}
>  	fsl_chan->tcd_pool = dma_pool_create("tcd_pool", chan->device->dev,
>  				sizeof(struct fsl_edma_hw_tcd),
>  				32, 0);
> @@ -778,6 +834,7 @@ static void fsl_edma_free_chan_resources(struct
> dma_chan *chan)
>  	spin_lock_irqsave(&fsl_chan->vchan.lock, flags);
>  	fsl_edma_disable_request(fsl_chan);
>  	fsl_edma_chan_mux(fsl_chan, 0, false);
> +	fsl_chan->slave_id = 0;
>  	fsl_chan->edesc = NULL;
>  	vchan_get_all_descriptors(&fsl_chan->vchan, &head);
>  	spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
> @@ -903,6 +960,7 @@ static int fsl_edma_probe(struct platform_device *pdev)
>  	dma_cap_set(DMA_PRIVATE, fsl_edma->dma_dev.cap_mask);
>  	dma_cap_set(DMA_SLAVE, fsl_edma->dma_dev.cap_mask);
>  	dma_cap_set(DMA_CYCLIC, fsl_edma->dma_dev.cap_mask);
> +	dma_cap_set(DMA_MEMCPY, fsl_edma->dma_dev.cap_mask);
>  
>  	fsl_edma->dma_dev.dev = &pdev->dev;
>  	fsl_edma->dma_dev.device_alloc_chan_resources
> @@ -912,6 +970,7 @@ static int fsl_edma_probe(struct platform_device *pdev)
>  	fsl_edma->dma_dev.device_tx_status = fsl_edma_tx_status;
>  	fsl_edma->dma_dev.device_prep_slave_sg = fsl_edma_prep_slave_sg;
>  	fsl_edma->dma_dev.device_prep_dma_cyclic = fsl_edma_prep_dma_cyclic;
> +	fsl_edma->dma_dev.device_prep_dma_memcpy = fsl_edma_prep_memcpy;
>  	fsl_edma->dma_dev.device_config = fsl_edma_slave_config;
>  	fsl_edma->dma_dev.device_pause = fsl_edma_pause;
>  	fsl_edma->dma_dev.device_resume = fsl_edma_resume;

  parent reply	other threads:[~2015-01-30 12:39 UTC|newest]

Thread overview: 20+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-12-30  8:41 [PATCH 0/2] dmaengine: fsl-edma: add dma memcpy and suspend/resume support Jingchang Lu
2014-12-30  8:41 ` Jingchang Lu
2014-12-30  8:41 ` [PATCH 1/2] dmaengine: fsl-edma: add dma memcpy support Jingchang Lu
2014-12-30  8:41   ` Jingchang Lu
2015-01-14  9:36   ` Jingchang Lu
2015-01-14  9:36     ` Jingchang Lu
2015-01-30  9:00     ` Yao Yuan
2015-01-30  9:00       ` Yao Yuan
2015-01-30 12:39   ` Stefan Agner [this message]
2015-01-30 12:39     ` Stefan Agner
2015-02-11  8:06     ` Yao Yuan
2015-02-11  8:06       ` Yao Yuan
2015-03-05 10:39   ` Vinod Koul
2015-03-05 10:39     ` Vinod Koul
2014-12-30  8:41 ` [PATCH 2/2] dmaengine: fsl-edma: add PM suspend/resume support Jingchang Lu
2014-12-30  8:41   ` Jingchang Lu
2015-01-30 13:05   ` Stefan Agner
2015-01-30 13:05     ` Stefan Agner
2015-03-05 10:41   ` Vinod Koul
2015-03-05 10:41     ` Vinod Koul

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=63612a565e0cbf4081b0b8efa0bc3c46@agner.ch \
    --to=stefan@agner.ch \
    --cc=dan.j.williams@intel.com \
    --cc=dmaengine@vger.kernel.org \
    --cc=jingchang.lu@freescale.com \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=vinod.koul@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.