From mboxrd@z Thu Jan  1 00:00:00 1970
From: m.szyprowski@samsung.com (Marek Szyprowski)
Date: Thu, 25 Sep 2014 12:43:35 +0200
Subject: [PATCH v2 01/18] arm: dma-mapping: arm_iommu_attach_device:
 automatically set max_seg_size
In-Reply-To: <20140924170627.GK16244@arm.com>
References: <1410868485-4143-1-git-send-email-m.szyprowski@samsung.com>
 <1410868485-4143-2-git-send-email-m.szyprowski@samsung.com>
 <20140924170627.GK16244@arm.com>
Message-ID: <5423F1D7.3060400@samsung.com>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

Hello,

On 2014-09-24 19:06, Will Deacon wrote:
> Hi Marek,
>
> On Tue, Sep 16, 2014 at 12:54:28PM +0100, Marek Szyprowski wrote:
>> If device has no max_seg_size set, we assume that there is no limit and
>> force it to DMA_BIT_MASK(32) to always use contiguous mappings in DMA
>> address space.
>>
>> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
>> ---
>>   arch/arm/mm/dma-mapping.c | 16 ++++++++++++++++
>>   1 file changed, 16 insertions(+)
>>
>> diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
>> index bcd5f836f27e..84705e24571b 100644
>> --- a/arch/arm/mm/dma-mapping.c
>> +++ b/arch/arm/mm/dma-mapping.c
>> @@ -2050,6 +2050,22 @@ int arm_iommu_attach_device(struct device *dev,
>>   {
>>   	int err;
>>   
>> +	/*
>> +	 * if device has no max_seg_size set, we assume that there is no limit
>> +	 * and force it to DMA_BIT_MASK(32) to always use contiguous mappings
>> +	 * in DMA address space
>> +	 */
>> +	if (!dev->dma_parms) {
>> +		dev->dma_parms = kzalloc(sizeof(*dev->dma_parms), GFP_KERNEL);
>> +		if (!dev->dma_parms)
>> +			return -ENOMEM;
>> +	}
>> +	if (!dev->dma_parms->max_segment_size) {
>> +		err = dma_set_max_seg_size(dev, DMA_BIT_MASK(32));
> Would it make more sense to base this default value off the dma_mask?
> In my IOMMU series, of_dma_configure passes back a size parameter to
> arch_setup_dma_ops which is calculated from the dma-ranges property or the
> coherent dma mask if the ranges property is absent, so maybe we should set
> this there too?

Right, good idea. This patch predates your arch_setup_dma_ops changes, so I
had to use something. The value taken from dma_mask is much better than 
hardcoded
DMA_BIT_MASK(32). Do you want to include an updated patch in next 
version of your
arch_setup_dma_ops patchset?

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland