Re: [PATCH v4 10/16] mtd: onenand: omap2: Convert to use dmaengine for memcpy

[Ladislav Michl <ladis@linux-mips.org>]

On Wed, Nov 15, 2017 at 10:57:20AM +0200, Roger Quadros wrote:
> On 11/11/17 23:23, Ladislav Michl wrote:
> > From: Peter Ujfalusi <peter.ujfalusi@ti.com>
> > 
> > Do not use the legacy and deprecated omap-dma interface for setting up the
> > memcpy.
> > 
> > Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
> > Signed-off-by: Ladislav Michl <ladis@linux-mips.org>
> 
> Acked-by: Roger Quadros <rogerq@ti.com>
> 
> Please see one comment below.
> 
> > ---
> >  Changes:
> >  -v4: new patch
> > 
> >  drivers/mtd/onenand/omap2.c | 80 +++++++++++++++++++++------------------------
> >  1 file changed, 38 insertions(+), 42 deletions(-)
> > 
> > diff --git a/drivers/mtd/onenand/omap2.c b/drivers/mtd/onenand/omap2.c
> > index 36314124488d..c9ff67100ef4 100644
> > --- a/drivers/mtd/onenand/omap2.c
> > +++ b/drivers/mtd/onenand/omap2.c
> > @@ -32,6 +32,7 @@
> >  #include <linux/interrupt.h>
> >  #include <linux/delay.h>
> >  #include <linux/dma-mapping.h>
> > +#include <linux/dmaengine.h>
> >  #include <linux/io.h>
> >  #include <linux/slab.h>
> >  #include <linux/gpio.h>
> > @@ -39,8 +40,6 @@
> >  #include <asm/mach/flash.h>
> >  #include <linux/platform_data/mtd-onenand-omap2.h>
> >  
> > -#include <linux/omap-dma.h>
> > -
> >  #define DRIVER_NAME "omap2-onenand"
> >  
> >  #define ONENAND_BUFRAM_SIZE	(1024 * 5)
> > @@ -55,17 +54,15 @@ struct omap2_onenand {
> >  	struct onenand_chip onenand;
> >  	struct completion irq_done;
> >  	struct completion dma_done;
> > -	int dma_channel;
> > +	struct dma_chan *dma_chan;
> >  	int freq;
> >  	int (*setup)(void __iomem *base, int *freq_ptr);
> >  	u8 flags;
> >  };
> >  
> > -static void omap2_onenand_dma_cb(int lch, u16 ch_status, void *data)
> > +static void omap2_onenand_dma_complete_func(void *completion)
> >  {
> > -	struct omap2_onenand *c = data;
> > -
> > -	complete(&c->dma_done);
> > +	complete(completion);
> >  }
> >  
> >  static irqreturn_t omap2_onenand_interrupt(int irq, void *dev_id)
> > @@ -292,23 +289,31 @@ static inline int omap2_onenand_dma_transfer(struct omap2_onenand *c,
> >  					     dma_addr_t src, dma_addr_t dst,
> >  					     size_t count)
> >  {
> > -	int data_type = __ffs((src | dst | count));
> > +	struct dma_async_tx_descriptor *tx;
> > +	dma_cookie_t cookie;
> >  
> > -	if (data_type > OMAP_DMA_DATA_TYPE_S32)
> > -		data_type = OMAP_DMA_DATA_TYPE_S32;
> > -
> > -	omap_set_dma_transfer_params(c->dma_channel, data_type,
> > -				     count / BIT(data_type), 1, 0, 0, 0);
> > -	omap_set_dma_src_params(c->dma_channel, 0, OMAP_DMA_AMODE_POST_INC,
> > -				src, 0, 0);
> > -	omap_set_dma_dest_params(c->dma_channel, 0, OMAP_DMA_AMODE_POST_INC,
> > -				 dst, 0, 0);
> > +	tx = dmaengine_prep_dma_memcpy(c->dma_chan, dst, src, count, 0);
> > +	if (!tx) {
> > +		dev_err(&c->pdev->dev, "Failed to prepare DMA memcpy\n");
> > +		return -EIO;
> > +	}
> >  
> >  	reinit_completion(&c->dma_done);
> > -	omap_start_dma(c->dma_channel);
> > +
> > +	tx->callback = omap2_onenand_dma_complete_func;
> > +	tx->callback_param = &c->dma_done;
> > +
> > +	cookie = tx->tx_submit(tx);
> > +	if (dma_submit_error(cookie)) {
> > +		dev_err(&c->pdev->dev, "Failed to do DMA tx_submit\n");
> > +		return -EIO;
> > +	}
> > +
> > +	dma_async_issue_pending(c->dma_chan);
> > +
> >  	if (!wait_for_completion_io_timeout(&c->dma_done,
> >  					    msecs_to_jiffies(20))) {
> > -		omap_stop_dma(c->dma_channel);
> > +		dmaengine_terminate_sync(c->dma_chan);
> >  		return -ETIMEDOUT;
> >  	}
> >  
> > @@ -468,8 +473,7 @@ static int omap2_onenand_probe(struct platform_device *pdev)
> >  	c->flags = pdata->flags;
> >  	c->gpmc_cs = pdata->cs;
> >  	c->gpio_irq = pdata->gpio_irq;
> > -	c->dma_channel = pdata->dma_channel;
> > -	if (c->dma_channel < 0) {
> > +	if (pdata->dma_channel < 0) {
> >  		/* if -1, don't use DMA */
> >  		c->gpio_irq = 0;
> 
> Why are we ignoring gpio_irq if DMA channel is not available?
> The wait mechanism could still benefit from the Interrupt even with no DMA right?

That's what original driver does and logic is preserved here.

> We could fix this in a separate patch of course.

And indeed we did. See "mtd: onenand: omap2: Enable DMA by default" (*) and
"mtd: onenand: omap2: Configure driver from DT".

(*) Now looking at that patch is should be renamed, as enabling DMA still depends
on pdata->dma_chan. It is enabled unconditionally in later patch. Will fix.

> >  	}
> > @@ -521,25 +525,17 @@ static int omap2_onenand_probe(struct platform_device *pdev)
> >  		goto err_release_gpio;
> >  	}
> >  
> > -	if (c->dma_channel >= 0) {
> > -		r = omap_request_dma(0, pdev->dev.driver->name,
> > -				     omap2_onenand_dma_cb, (void *) c,
> > -				     &c->dma_channel);
> > -		if (r == 0) {
> > -			omap_set_dma_write_mode(c->dma_channel,
> > -						OMAP_DMA_WRITE_NON_POSTED);
> > -			omap_set_dma_src_data_pack(c->dma_channel, 1);
> > -			omap_set_dma_src_burst_mode(c->dma_channel,
> > -						    OMAP_DMA_DATA_BURST_8);
> > -			omap_set_dma_dest_data_pack(c->dma_channel, 1);
> > -			omap_set_dma_dest_burst_mode(c->dma_channel,
> > -						     OMAP_DMA_DATA_BURST_8);
> > -		} else {
> > +	if (pdata->dma_channel >= 0) {
> > +		dma_cap_mask_t mask;
> > +
> > +		dma_cap_zero(mask);
> > +		dma_cap_set(DMA_MEMCPY, mask);
> > +
> > +		c->dma_chan = dma_request_channel(mask, NULL, NULL);
> > +		if (!c->dma_chan)
> >  			dev_info(&pdev->dev,
> >  				 "failed to allocate DMA for OneNAND, "
> >  				 "using PIO instead\n");
> > -			c->dma_channel = -1;
> > -		}
> >  	}
> >  
> >  	dev_info(&pdev->dev, "initializing on CS%d, phys base 0x%08lx, virtual "
> > @@ -553,7 +549,7 @@ static int omap2_onenand_probe(struct platform_device *pdev)
> >  	mtd_set_of_node(&c->mtd, pdata->of_node);
> >  
> >  	this = &c->onenand;
> > -	if (c->dma_channel >= 0) {
> > +	if (c->dma_chan) {
> >  		this->wait = omap2_onenand_wait;
> >  		this->read_bufferram = omap2_onenand_read_bufferram;
> >  		this->write_bufferram = omap2_onenand_write_bufferram;
> > @@ -573,8 +569,8 @@ static int omap2_onenand_probe(struct platform_device *pdev)
> >  err_release_onenand:
> >  	onenand_release(&c->mtd);
> >  err_release_dma:
> > -	if (c->dma_channel != -1)
> > -		omap_free_dma(c->dma_channel);
> > +	if (c->dma_chan)
> > +		dma_release_channel(c->dma_chan);
> >  	if (c->gpio_irq)
> >  		free_irq(gpio_to_irq(c->gpio_irq), c);
> >  err_release_gpio:
> > @@ -595,8 +591,8 @@ static int omap2_onenand_remove(struct platform_device *pdev)
> >  	struct omap2_onenand *c = dev_get_drvdata(&pdev->dev);
> >  
> >  	onenand_release(&c->mtd);
> > -	if (c->dma_channel != -1)
> > -		omap_free_dma(c->dma_channel);
> > +	if (c->dma_chan)
> > +		dma_release_channel(c->dma_chan);
> >  	omap2_onenand_shutdown(pdev);
> >  	if (c->gpio_irq) {
> >  		free_irq(gpio_to_irq(c->gpio_irq), c);
> > 
> 
> -- 
> cheers,
> -roger
> 
> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki