[PATCH 0/2] dmaengine: rcar-dmac: Add dma-channel-mask property support

[PATCH 0/2] dmaengine: rcar-dmac: Add dma-channel-mask property support

[Yoshihiro Shimoda]

This patch series is based on
 - renesas-drivers.git / renesas-drivers-2019-08-13-v5.3-rc4 tag
 - and the following patch series:
https://patchwork.kernel.org/project/linux-renesas-soc/list/?series=165881

The commit e2d896c08ca3 ("Documentation: bindings: dma: Add binding for
dma-channel-mask") adds the generic property and R-Car also has such
use cases so that I made this patch series. Before adding the property
support, I made a clean-up patch as the patch 1/2.

Yoshihiro Shimoda (2):
  dmaengine: rcar-dmac: Don't set DMACHCLR bit 0 to 1 if iommu is mapped
  dmaengine: rcar-dmac: Add dma-channel-mask property support

 drivers/dma/sh/rcar-dmac.c | 32 ++++++++++++++++++++++++--------
 1 file changed, 24 insertions(+), 8 deletions(-)

-- 
2.7.4

[PATCH 1/2] dmaengine: rcar-dmac: Don't set DMACHCLR bit 0 to 1 if iommu is mapped

[Yoshihiro Shimoda]

The commit 20c169aceb45 ("dmaengine: rcar-dmac: clear pertinence
number of channels") always set the DMACHCLR bit 0 to 1, but if
iommu is mapped to the device, this driver doesn't need to clear it.
So, this patch takes care of it by using "channels_mask" bitfield.

Note that, this patch doesn't have a "Fixes:" tag because the driver
doesn't manage the channel 0 anyway so that the behavior of
the channel is not changed.

Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
---
 drivers/dma/sh/rcar-dmac.c | 22 ++++++++++++++--------
 1 file changed, 14 insertions(+), 8 deletions(-)

diff --git a/drivers/dma/sh/rcar-dmac.c b/drivers/dma/sh/rcar-dmac.c
index 779b715..204160e 100644
--- a/drivers/dma/sh/rcar-dmac.c
+++ b/drivers/dma/sh/rcar-dmac.c
@@ -192,6 +192,7 @@ struct rcar_dmac_chan {
  * @iomem: remapped I/O memory base
  * @n_channels: number of available channels
  * @channels: array of DMAC channels
+ * @channels_mask: bitfield of which DMA channels are managed by this driver
  * @modules: bitmask of client modules in use
  */
 struct rcar_dmac {
@@ -202,6 +203,7 @@ struct rcar_dmac {
 
 	unsigned int n_channels;
 	struct rcar_dmac_chan *channels;
+	unsigned int channels_mask;
 
 	DECLARE_BITMAP(modules, 256);
 };
@@ -446,7 +448,7 @@ static int rcar_dmac_init(struct rcar_dmac *dmac)
 	u16 dmaor;
 
 	/* Clear all channels and enable the DMAC globally. */
-	rcar_dmac_write(dmac, RCAR_DMACHCLR, GENMASK(dmac->n_channels - 1, 0));
+	rcar_dmac_write(dmac, RCAR_DMACHCLR, dmac->channels_mask);
 	rcar_dmac_write(dmac, RCAR_DMAOR,
 			RCAR_DMAOR_PRI_FIXED | RCAR_DMAOR_DME);
 
@@ -822,6 +824,9 @@ static void rcar_dmac_stop_all_chan(struct rcar_dmac *dmac)
 	for (i = 0; i < dmac->n_channels; ++i) {
 		struct rcar_dmac_chan *chan = &dmac->channels[i];
 
+		if (!(dmac->channels_mask & BIT(i)))
+			continue;
+
 		/* Stop and reinitialize the channel. */
 		spin_lock_irq(&chan->lock);
 		rcar_dmac_chan_halt(chan);
@@ -1801,6 +1806,8 @@ static int rcar_dmac_parse_of(struct device *dev, struct rcar_dmac *dmac)
 		return -EINVAL;
 	}
 
+	dmac->channels_mask = GENMASK(dmac->n_channels - 1, 0);
+
 	return 0;
 }
 
@@ -1810,7 +1817,6 @@ static int rcar_dmac_probe(struct platform_device *pdev)
 		DMA_SLAVE_BUSWIDTH_2_BYTES | DMA_SLAVE_BUSWIDTH_4_BYTES |
 		DMA_SLAVE_BUSWIDTH_8_BYTES | DMA_SLAVE_BUSWIDTH_16_BYTES |
 		DMA_SLAVE_BUSWIDTH_32_BYTES | DMA_SLAVE_BUSWIDTH_64_BYTES;
-	unsigned int channels_offset = 0;
 	struct dma_device *engine;
 	struct rcar_dmac *dmac;
 	const struct rcar_dmac_of_data *data;
@@ -1843,10 +1849,8 @@ static int rcar_dmac_probe(struct platform_device *pdev)
 	 * level we can't disable it selectively, so ignore channel 0 for now if
 	 * the device is part of an IOMMU group.
 	 */
-	if (device_iommu_mapped(&pdev->dev)) {
-		dmac->n_channels--;
-		channels_offset = 1;
-	}
+	if (device_iommu_mapped(&pdev->dev))
+		dmac->channels_mask &= ~BIT(0);
 
 	dmac->channels = devm_kcalloc(&pdev->dev, dmac->n_channels,
 				      sizeof(*dmac->channels), GFP_KERNEL);
@@ -1903,8 +1907,10 @@ static int rcar_dmac_probe(struct platform_device *pdev)
 	INIT_LIST_HEAD(&engine->channels);
 
 	for (i = 0; i < dmac->n_channels; ++i) {
-		ret = rcar_dmac_chan_probe(dmac, &dmac->channels[i], data,
-					   i + channels_offset);
+		if (!(dmac->channels_mask & BIT(i)))
+			continue;
+
+		ret = rcar_dmac_chan_probe(dmac, &dmac->channels[i], data, i);
 		if (ret < 0)
 			goto error;
 	}
-- 
2.7.4

Re: [PATCH 1/2] dmaengine: rcar-dmac: Don't set DMACHCLR bit 0 to 1 if iommu is mapped

[Simon Horman]

On Wed, Aug 28, 2019 at 08:13:54PM +0900, Yoshihiro Shimoda wrote:
> The commit 20c169aceb45 ("dmaengine: rcar-dmac: clear pertinence
> number of channels") always set the DMACHCLR bit 0 to 1, but if
> iommu is mapped to the device, this driver doesn't need to clear it.
> So, this patch takes care of it by using "channels_mask" bitfield.
> 
> Note that, this patch doesn't have a "Fixes:" tag because the driver
> doesn't manage the channel 0 anyway so that the behavior of
> the channel is not changed.
> 
> Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>

Reviewed-by: Simon Horman <horms+renesas@verge.net.au>

Re: [PATCH 1/2] dmaengine: rcar-dmac: Don't set DMACHCLR bit 0 to 1 if iommu is mapped

[Geert Uytterhoeven]

Hi Shimoda-san,

On Wed, Aug 28, 2019 at 1:15 PM Yoshihiro Shimoda
<yoshihiro.shimoda.uh@renesas.com> wrote:
> The commit 20c169aceb45 ("dmaengine: rcar-dmac: clear pertinence
> number of channels") always set the DMACHCLR bit 0 to 1, but if
> iommu is mapped to the device, this driver doesn't need to clear it.
> So, this patch takes care of it by using "channels_mask" bitfield.

Thanks for your patch!

> Note that, this patch doesn't have a "Fixes:" tag because the driver
> doesn't manage the channel 0 anyway so that the behavior of
> the channel is not changed.

This patch does fix a bug, as GENMASK(dmac->n_channels - 1, 0) doesn't
take into account channels_offset.  Hence it not only clears channel 0,
as you mentioned, but also forgets to clear the last channel, which
is a real bug.

So I think this does warrant a
Fixes: 20c169aceb459575 ("dmaengine: rcar-dmac: clear pertinence
number of channels")

Or perhaps the actual bug should be fixed first in a separate patch?

> Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>

Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>

> --- a/drivers/dma/sh/rcar-dmac.c
> +++ b/drivers/dma/sh/rcar-dmac.c

> @@ -446,7 +448,7 @@ static int rcar_dmac_init(struct rcar_dmac *dmac)
>         u16 dmaor;
>
>         /* Clear all channels and enable the DMAC globally. */
> -       rcar_dmac_write(dmac, RCAR_DMACHCLR, GENMASK(dmac->n_channels - 1, 0));
> +       rcar_dmac_write(dmac, RCAR_DMACHCLR, dmac->channels_mask);
>         rcar_dmac_write(dmac, RCAR_DMAOR,
>                         RCAR_DMAOR_PRI_FIXED | RCAR_DMAOR_DME);
>
> @@ -822,6 +824,9 @@ static void rcar_dmac_stop_all_chan(struct rcar_dmac *dmac)
>         for (i = 0; i < dmac->n_channels; ++i) {
>                 struct rcar_dmac_chan *chan = &dmac->channels[i];
>
> +               if (!(dmac->channels_mask & BIT(i)))
> +                       continue;
> +
>                 /* Stop and reinitialize the channel. */
>                 spin_lock_irq(&chan->lock);
>                 rcar_dmac_chan_halt(chan);
> @@ -1801,6 +1806,8 @@ static int rcar_dmac_parse_of(struct device *dev, struct rcar_dmac *dmac)
>                 return -EINVAL;
>         }
>
> +       dmac->channels_mask = GENMASK(dmac->n_channels - 1, 0);

You're aware dmac->n_channels can be 99, as per the check above, jut out of
context? ;-)

Probably that check should be changed to reject >= 32, as the hardware
and driver don't support more than 32 bits in CHCLR anyway.
> +
>         return 0;
>  }
>
> @@ -1810,7 +1817,6 @@ static int rcar_dmac_probe(struct platform_device *pdev)
>                 DMA_SLAVE_BUSWIDTH_2_BYTES | DMA_SLAVE_BUSWIDTH_4_BYTES |
>                 DMA_SLAVE_BUSWIDTH_8_BYTES | DMA_SLAVE_BUSWIDTH_16_BYTES |
>                 DMA_SLAVE_BUSWIDTH_32_BYTES | DMA_SLAVE_BUSWIDTH_64_BYTES;
> -       unsigned int channels_offset = 0;
>         struct dma_device *engine;
>         struct rcar_dmac *dmac;
>         const struct rcar_dmac_of_data *data;
> @@ -1843,10 +1849,8 @@ static int rcar_dmac_probe(struct platform_device *pdev)
>          * level we can't disable it selectively, so ignore channel 0 for now if
>          * the device is part of an IOMMU group.
>          */
> -       if (device_iommu_mapped(&pdev->dev)) {
> -               dmac->n_channels--;
> -               channels_offset = 1;
> -       }
> +       if (device_iommu_mapped(&pdev->dev))
> +               dmac->channels_mask &= ~BIT(0);
>
>         dmac->channels = devm_kcalloc(&pdev->dev, dmac->n_channels,
>                                       sizeof(*dmac->channels), GFP_KERNEL);

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

RE: [PATCH 1/2] dmaengine: rcar-dmac: Don't set DMACHCLR bit 0 to 1 if iommu is mapped

[Yoshihiro Shimoda]

Hi Geert-san,

> From: Geert Uytterhoeven, Sent: Monday, September 2, 2019 5:36 PM
> 
> Hi Shimoda-san,
> 
> On Wed, Aug 28, 2019 at 1:15 PM Yoshihiro Shimoda
> <yoshihiro.shimoda.uh@renesas.com> wrote:
> > The commit 20c169aceb45 ("dmaengine: rcar-dmac: clear pertinence
> > number of channels") always set the DMACHCLR bit 0 to 1, but if
> > iommu is mapped to the device, this driver doesn't need to clear it.
> > So, this patch takes care of it by using "channels_mask" bitfield.
> 
> Thanks for your patch!
> 
> > Note that, this patch doesn't have a "Fixes:" tag because the driver
> > doesn't manage the channel 0 anyway so that the behavior of
> > the channel is not changed.
> 
> This patch does fix a bug, as GENMASK(dmac->n_channels - 1, 0) doesn't
> take into account channels_offset.  Hence it not only clears channel 0,
> as you mentioned, but also forgets to clear the last channel, which
> is a real bug.

Indeed.

> So I think this does warrant a
> Fixes: 20c169aceb459575 ("dmaengine: rcar-dmac: clear pertinence
> number of channels")
> 
> Or perhaps the actual bug should be fixed first in a separate patch?

I think so. So, now I had already submitted two series like below, but
I'll fix this at first.

https://patchwork.kernel.org/project/linux-renesas-soc/list/?series=165881
https://patchwork.kernel.org/project/linux-renesas-soc/list/?series=166457

> > Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
> 
> Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>

Thank you for your review!

> > --- a/drivers/dma/sh/rcar-dmac.c
> > +++ b/drivers/dma/sh/rcar-dmac.c
> 
> > @@ -446,7 +448,7 @@ static int rcar_dmac_init(struct rcar_dmac *dmac)
> >         u16 dmaor;
> >
> >         /* Clear all channels and enable the DMAC globally. */
> > -       rcar_dmac_write(dmac, RCAR_DMACHCLR, GENMASK(dmac->n_channels - 1, 0));
> > +       rcar_dmac_write(dmac, RCAR_DMACHCLR, dmac->channels_mask);
> >         rcar_dmac_write(dmac, RCAR_DMAOR,
> >                         RCAR_DMAOR_PRI_FIXED | RCAR_DMAOR_DME);
> >
> > @@ -822,6 +824,9 @@ static void rcar_dmac_stop_all_chan(struct rcar_dmac *dmac)
> >         for (i = 0; i < dmac->n_channels; ++i) {
> >                 struct rcar_dmac_chan *chan = &dmac->channels[i];
> >
> > +               if (!(dmac->channels_mask & BIT(i)))
> > +                       continue;
> > +
> >                 /* Stop and reinitialize the channel. */
> >                 spin_lock_irq(&chan->lock);
> >                 rcar_dmac_chan_halt(chan);
> > @@ -1801,6 +1806,8 @@ static int rcar_dmac_parse_of(struct device *dev, struct rcar_dmac *dmac)
> >                 return -EINVAL;
> >         }
> >
> > +       dmac->channels_mask = GENMASK(dmac->n_channels - 1, 0);
> 
> You're aware dmac->n_channels can be 99, as per the check above, jut out of
> context? ;-)
> 
> Probably that check should be changed to reject >= 32, as the hardware
> and driver don't support more than 32 bits in CHCLR anyway.

I got it. So, I'll fix the rcar_dmac_parse_of() as one more a separate patch.

Best regards,
Yoshihiro Shimoda

> > +
> >         return 0;
> >  }
> >
> > @@ -1810,7 +1817,6 @@ static int rcar_dmac_probe(struct platform_device *pdev)
> >                 DMA_SLAVE_BUSWIDTH_2_BYTES | DMA_SLAVE_BUSWIDTH_4_BYTES |
> >                 DMA_SLAVE_BUSWIDTH_8_BYTES | DMA_SLAVE_BUSWIDTH_16_BYTES |
> >                 DMA_SLAVE_BUSWIDTH_32_BYTES | DMA_SLAVE_BUSWIDTH_64_BYTES;
> > -       unsigned int channels_offset = 0;
> >         struct dma_device *engine;
> >         struct rcar_dmac *dmac;
> >         const struct rcar_dmac_of_data *data;
> > @@ -1843,10 +1849,8 @@ static int rcar_dmac_probe(struct platform_device *pdev)
> >          * level we can't disable it selectively, so ignore channel 0 for now if
> >          * the device is part of an IOMMU group.
> >          */
> > -       if (device_iommu_mapped(&pdev->dev)) {
> > -               dmac->n_channels--;
> > -               channels_offset = 1;
> > -       }
> > +       if (device_iommu_mapped(&pdev->dev))
> > +               dmac->channels_mask &= ~BIT(0);
> >
> >         dmac->channels = devm_kcalloc(&pdev->dev, dmac->n_channels,
> >                                       sizeof(*dmac->channels), GFP_KERNEL);
> 
> Gr{oetje,eeting}s,
> 
>                         Geert
> 
> --
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
> 
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
>                                 -- Linus Torvalds

Re: [PATCH 1/2] dmaengine: rcar-dmac: Don't set DMACHCLR bit 0 to 1 if iommu is mapped

[Geert Uytterhoeven]

Hi Shimoda-san,

On Wed, Aug 28, 2019 at 1:15 PM Yoshihiro Shimoda
<yoshihiro.shimoda.uh@renesas.com> wrote:
> The commit 20c169aceb45 ("dmaengine: rcar-dmac: clear pertinence
> number of channels") always set the DMACHCLR bit 0 to 1, but if
> iommu is mapped to the device, this driver doesn't need to clear it.
> So, this patch takes care of it by using "channels_mask" bitfield.
>
> Note that, this patch doesn't have a "Fixes:" tag because the driver
> doesn't manage the channel 0 anyway so that the behavior of
> the channel is not changed.
>
> Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
> ---
>  drivers/dma/sh/rcar-dmac.c | 22 ++++++++++++++--------
>  1 file changed, 14 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/dma/sh/rcar-dmac.c b/drivers/dma/sh/rcar-dmac.c
> index 779b715..204160e 100644
> --- a/drivers/dma/sh/rcar-dmac.c
> +++ b/drivers/dma/sh/rcar-dmac.c
> @@ -192,6 +192,7 @@ struct rcar_dmac_chan {
>   * @iomem: remapped I/O memory base
>   * @n_channels: number of available channels
>   * @channels: array of DMAC channels
> + * @channels_mask: bitfield of which DMA channels are managed by this driver
>   * @modules: bitmask of client modules in use
>   */
>  struct rcar_dmac {
> @@ -202,6 +203,7 @@ struct rcar_dmac {
>
>         unsigned int n_channels;
>         struct rcar_dmac_chan *channels;
> +       unsigned int channels_mask;

Given you want to store the output of of_property_read_u32() here in a
subsequent patch, you may want to use u32 instead of unsigned int.

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

RE: [PATCH 1/2] dmaengine: rcar-dmac: Don't set DMACHCLR bit 0 to 1 if iommu is mapped

[Yoshihiro Shimoda]

Hi Geert-san,

> From: Geert Uytterhoeven, Sent: Monday, September 2, 2019 5:52 PM
<snip>
> > @@ -202,6 +203,7 @@ struct rcar_dmac {
> >
> >         unsigned int n_channels;
> >         struct rcar_dmac_chan *channels;
> > +       unsigned int channels_mask;
> 
> Given you want to store the output of of_property_read_u32() here in a
> subsequent patch, you may want to use u32 instead of unsigned int.

I got it. I'll fix it.

Best regards,
Yoshihiro Shimoda

[PATCH 2/2] dmaengine: rcar-dmac: Add dma-channel-mask property support

[Yoshihiro Shimoda]

This patch adds dma-channel-mask property support not to reserve
some DMA channels for some reasons. (for example: a heterogeneous
CPU uses it.)

Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
---
 drivers/dma/sh/rcar-dmac.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/drivers/dma/sh/rcar-dmac.c b/drivers/dma/sh/rcar-dmac.c
index 204160e..bae0fe8 100644
--- a/drivers/dma/sh/rcar-dmac.c
+++ b/drivers/dma/sh/rcar-dmac.c
@@ -1806,7 +1806,17 @@ static int rcar_dmac_parse_of(struct device *dev, struct rcar_dmac *dmac)
 		return -EINVAL;
 	}
 
-	dmac->channels_mask = GENMASK(dmac->n_channels - 1, 0);
+	/*
+	 * If the driver is unable to read dma-channel-mask property,
+	 * the driver assumes that it can use all channels.
+	 */
+	ret = of_property_read_u32(np, "dma-channel-mask",
+				   &dmac->channels_mask);
+	if (ret < 0)
+		dmac->channels_mask = GENMASK(dmac->n_channels - 1, 0);
+
+	/* If the property has out-of-channel mask, this driver clears it */
+	dmac->channels_mask &= GENMASK(dmac->n_channels - 1, 0);
 
 	return 0;
 }
-- 
2.7.4

Re: [PATCH 2/2] dmaengine: rcar-dmac: Add dma-channel-mask property support

[Simon Horman]

On Wed, Aug 28, 2019 at 08:13:55PM +0900, Yoshihiro Shimoda wrote:
> This patch adds dma-channel-mask property support not to reserve
> some DMA channels for some reasons. (for example: a heterogeneous
> CPU uses it.)
> 
> Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>

Reviewed-by: Simon Horman <horms+renesas@verge.net.au>

Re: [PATCH 2/2] dmaengine: rcar-dmac: Add dma-channel-mask property support

[Geert Uytterhoeven]

Hi Shimoda-san,

On Wed, Aug 28, 2019 at 1:15 PM Yoshihiro Shimoda
<yoshihiro.shimoda.uh@renesas.com> wrote:
> This patch adds dma-channel-mask property support not to reserve
> some DMA channels for some reasons. (for example: a heterogeneous
> CPU uses it.)

Thanks for your patch!

> Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>

Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>

One suggestion below.

> --- a/drivers/dma/sh/rcar-dmac.c
> +++ b/drivers/dma/sh/rcar-dmac.c
> @@ -1806,7 +1806,17 @@ static int rcar_dmac_parse_of(struct device *dev, struct rcar_dmac *dmac)
>                 return -EINVAL;
>         }
>
> -       dmac->channels_mask = GENMASK(dmac->n_channels - 1, 0);
> +       /*
> +        * If the driver is unable to read dma-channel-mask property,
> +        * the driver assumes that it can use all channels.
> +        */
> +       ret = of_property_read_u32(np, "dma-channel-mask",
> +                                  &dmac->channels_mask);
> +       if (ret < 0)
> +               dmac->channels_mask = GENMASK(dmac->n_channels - 1, 0);

You could keep the preinitialization, and just ignore the return value:

    dmac->channels_mask = GENMASK(dmac->n_channels - 1, 0);
    of_property_read_u32(np, "dma-channel-mask", &dmac->channels_mask);

>
> +
> +       /* If the property has out-of-channel mask, this driver clears it */
> +       dmac->channels_mask &= GENMASK(dmac->n_channels - 1, 0);
>
>         return 0;
>  }

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds