[RFC PATCH] ARM: OMAP2+: omap-device: Do not overwrite resources allocated by OF layer

[Benoit Cousson]

Hi Vaibhav,

On 08/29/2012 11:48 AM, Vaibhav Hiremath wrote:
> With the new devices (like, AM33XX and OMAP5) we now only support
> DT boot mode of operation and now it is the time to start killing
> slowly the dependency on hwmod, so with this patch, we are starting
> with device resources.

Thanks for working on that.

> The idea here is implemented considering to both boot modes -
>   - DT boot mode
>     OF framework will construct the resource structure (currently
>     does for MEM & IRQ resource) and we should respect/use these
>     resources, killing hwmod dependency.
>     If pdev->num_resources > 0, we assume that MEM & IRQ resources
>     have been allocated by OF layer already (through DTB).
> 
>     Once DMA resource is available from OF layer, we should
>     kill filling any resources from hwmod.

Yeap, I did a similar patch some months ago and decided to drop it
because I was expected the DMA binding to be there and wanted to avoid
adding more code that we are going to remove later.

The other potential issue is that when the binding will be there, we
will have to update all the drivers and DTS first before being able to
change the hwmod code from hwmod DMA to DTS DMA.
I was thinking of something smoother that will check if DMA is in DTS
and fall back to hwmod if not to avoid that.
But again, I'm not sure it worth the effort.

>   - Non-DT boot mode
>     Here, pdev->num_resources = 0, and we should get all the
>     resources from hwmod (following existing steps)
> 
> Signed-off-by: Vaibhav Hiremath <hvaibhav at ti.com>
> Cc: Benoit Cousson <b-cousson at ti.com>
> Cc: Tony Lindgren <tony at atomide.com>
> Cc: Paul Walmsley <paul at pwsan.com>
> Cc: Kevin Hilman <khilman at ti.com>
> ---
> This patch is tested on BeagleBone and AM37xEVM.

I'll try to do more testing on Panda next week.

> Why RFC?
> Still we have function duplication omap_device_fill_resources() and
> omap_device_fill_dma_resources(), we can actually split the function
> into 3 resources and avoid duplication -
>   - omap_device_fill_dma_resources()
>   - omap_device_fill_mem_resources()
>   - omap_device_fill_irq_resources()
> 
> Actually I wanted to clean it further but thought of getting
> feedback first and then proceed further.

Well, that's anyway temporary code that should be gone in 6 months from
now (ideally). So that looks pretty good to me.

>  arch/arm/mach-omap2/omap_hwmod.c             |   27 ++++++++++
>  arch/arm/plat-omap/include/plat/omap_hwmod.h |    1 +
>  arch/arm/plat-omap/omap_device.c             |   72 +++++++++++++++++++++----
>  3 files changed, 88 insertions(+), 12 deletions(-)
> 
> diff --git a/arch/arm/mach-omap2/omap_hwmod.c b/arch/arm/mach-omap2/omap_hwmod.c
> index 31ec283..edabfb3 100644
> --- a/arch/arm/mach-omap2/omap_hwmod.c
> +++ b/arch/arm/mach-omap2/omap_hwmod.c
> @@ -3330,6 +3330,33 @@ int omap_hwmod_fill_resources(struct omap_hwmod *oh, struct resource *res)
>  }
> 
>  /**
> + * omap_hwmod_fill_dma_resources - fill struct resource array with dma data
> + * @oh: struct omap_hwmod *
> + * @res: pointer to the array of struct resource to fill
> + *
> + * Fill the struct resource array @res with dma resource data from the
> + * omap_hwmod @oh.  Intended to be called by code that registers
> + * omap_devices.  See also omap_hwmod_count_resources().  Returns the
> + * number of array elements filled.
> + */
> +int omap_hwmod_fill_dma_resources(struct omap_hwmod *oh, struct resource *res)
> +{
> +	int i, sdma_reqs_cnt;
> +	int r = 0;
> +
> +	sdma_reqs_cnt = _count_sdma_reqs(oh);
> +	for (i = 0; i < sdma_reqs_cnt; i++) {
> +		(res + r)->name = (oh->sdma_reqs + i)->name;
> +		(res + r)->start = (oh->sdma_reqs + i)->dma_req;
> +		(res + r)->end = (oh->sdma_reqs + i)->dma_req;
> +		(res + r)->flags = IORESOURCE_DMA;
> +		r++;
> +	}
> +
> +	return r;
> +}
> +
> +/**
>   * omap_hwmod_get_resource_byname - fetch IP block integration data by name
>   * @oh: struct omap_hwmod * to operate on
>   * @type: one of the IORESOURCE_* constants from include/linux/ioport.h
> diff --git a/arch/arm/plat-omap/include/plat/omap_hwmod.h b/arch/arm/plat-omap/include/plat/omap_hwmod.h
> index 9b9646c..0533073 100644
> --- a/arch/arm/plat-omap/include/plat/omap_hwmod.h
> +++ b/arch/arm/plat-omap/include/plat/omap_hwmod.h
> @@ -615,6 +615,7 @@ int omap_hwmod_softreset(struct omap_hwmod *oh);
> 
>  int omap_hwmod_count_resources(struct omap_hwmod *oh);
>  int omap_hwmod_fill_resources(struct omap_hwmod *oh, struct resource *res);
> +int omap_hwmod_fill_dma_resources(struct omap_hwmod *oh, struct resource *res);
>  int omap_hwmod_get_resource_byname(struct omap_hwmod *oh, unsigned int type,
>  				   const char *name, struct resource *res);
> 
> diff --git a/arch/arm/plat-omap/omap_device.c b/arch/arm/plat-omap/omap_device.c
> index c490240..fd15a3a 100644
> --- a/arch/arm/plat-omap/omap_device.c
> +++ b/arch/arm/plat-omap/omap_device.c
> @@ -486,6 +486,33 @@ static int omap_device_fill_resources(struct omap_device *od,
>  }
> 
>  /**
> + * omap_device_fill_dma_resources - fill in array of struct resource with dma resources
> + * @od: struct omap_device *
> + * @res: pointer to an array of struct resource to be filled in
> + *
> + * Populate one or more empty struct resource pointed to by @res with
> + * the dma resource data for this omap_device @od.  Used by
> + * omap_device_alloc() after calling omap_device_count_resources().
> + *
> + * Ideally this function would not be needed at all.  If we have
> + * mechanism to get dma resources from DT.
> + *
> + * Returns 0.
> + */
> +static int omap_device_fill_dma_resources(struct omap_device *od,
> +				      struct resource *res)
> +{
> +	int i, r;
> +
> +	for (i = 0; i < od->hwmods_cnt; i++) {
> +		r = omap_hwmod_fill_dma_resources(od->hwmods[i], res);
> +		res += r;
> +	}
> +
> +	return 0;
> +}
> +
> +/**
>   * omap_device_alloc - allocate an omap_device
>   * @pdev: platform_device that will be included in this omap_device
>   * @oh: ptr to the single omap_hwmod that backs this omap_device
> @@ -524,24 +551,45 @@ struct omap_device *omap_device_alloc(struct platform_device *pdev,
>  	od->hwmods = hwmods;
>  	od->pdev = pdev;
> 
> +	res_count = omap_device_count_resources(od);
>  	/*
> -	 * HACK: Ideally the resources from DT should match, and hwmod
> -	 * should just add the missing ones. Since the name is not
> -	 * properly populated by DT, stick to hwmod resources only.
> +	 * DT Boot:
> +	 *   OF framework will construct the resource structure (currently
> +	 *   does for MEM & IRQ resource) and we should respect/use these
> +	 *   resources, killing hwmod dependency.
> +	 *   If pdev->num_resources > 0, we assume that MEM & IRQ resources
> +	 *   have been allocated by OF layer already (through DTB).
> +	 *
> +	 * Non-DT Boot:
> +	 *   Here, pdev->num_resources = 0, and we should get all the
> +	 *   resources from hwmod.
> +	 *
> +	 * TODO: Once DMA resource is available from OF layer, we should
> +	 *   kill filling any resources from hwmod.
>  	 */
> -	if (pdev->num_resources && pdev->resource)
> -		dev_warn(&pdev->dev, "%s(): resources already allocated %d\n",
> -			__func__, pdev->num_resources);
> -
> -	res_count = omap_device_count_resources(od);
> -	if (res_count > 0) {
> -		dev_dbg(&pdev->dev, "%s(): resources allocated from hwmod %d\n",
> -			__func__, res_count);
> +	if (res_count > pdev->num_resources) {
> +		/* Allocate resources memory to account for new resources */
>  		res = kzalloc(sizeof(struct resource) * res_count, GFP_KERNEL);
>  		if (!res)
>  			goto oda_exit3;
> 
> -		omap_device_fill_resources(od, res);
> +		/*
> +		 * If pdev->num_resources > 0, then assume that,
> +		 * MEM and IRQ resources will only come from DT and only
> +		 * fill DMA resource from hwmod layer.
> +		 */
> +		if (pdev->num_resources && pdev->resource) {
> +			dev_dbg(&pdev->dev, "%s(): resources already allocated %d\n",
> +					__func__, res_count);
> +			memcpy(res, pdev->resource,
> +				sizeof(struct resource) * pdev->num_resources);
> +			omap_device_fill_dma_resources(od,
> +					&res[pdev->num_resources]);
> +		} else {
> +			dev_dbg(&pdev->dev, "%s(): using resources from hwmod %d\n",
> +					__func__, res_count);
> +			omap_device_fill_resources(od, res);
> +		}

Considering the temporary aspect of that, I'm more than fine with that
approach.

BTW, did you attend the discussion about the DMA binding during PLC?
Is this going to be finalized soon?

Thanks,
Benoit