[PATCH] ARM64: dma-mapping: fix handling of devices registered before arch_initcall

[PATCH] ARM64: dma-mapping: fix handling of devices registered before arch_initcall

[Marek Szyprowski]

This patch ensures that devices, which got registered before arch_initcall
will be handled correctly by IOMMU-based DMA-mapping code.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
 arch/arm64/mm/dma-mapping.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/arch/arm64/mm/dma-mapping.c b/arch/arm64/mm/dma-mapping.c
index 573b1e6..ce6019f 100644
--- a/arch/arm64/mm/dma-mapping.c
+++ b/arch/arm64/mm/dma-mapping.c
@@ -928,6 +928,10 @@ static int __init __iommu_dma_init(void)
 		ret = register_iommu_dma_ops_notifier(&platform_bus_type);
 	if (!ret)
 		ret = register_iommu_dma_ops_notifier(&amba_bustype);
+
+	/* handle devices queued before this arch_initcall */
+	if (!ret)
+		__iommu_attach_notifier(NULL, BUS_NOTIFY_ADD_DEVICE, NULL);
 	return ret;
 }
 arch_initcall(__iommu_dma_init);
-- 
1.9.2

[PATCH] ARM64: dma-mapping: fix handling of devices registered before arch_initcall

[Robin Murphy]

On 16/02/16 14:14, Marek Szyprowski wrote:
> This patch ensures that devices, which got registered before arch_initcall
> will be handled correctly by IOMMU-based DMA-mapping code.
>
> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
> ---
>   arch/arm64/mm/dma-mapping.c | 4 ++++
>   1 file changed, 4 insertions(+)
>
> diff --git a/arch/arm64/mm/dma-mapping.c b/arch/arm64/mm/dma-mapping.c
> index 573b1e6..ce6019f 100644
> --- a/arch/arm64/mm/dma-mapping.c
> +++ b/arch/arm64/mm/dma-mapping.c
> @@ -928,6 +928,10 @@ static int __init __iommu_dma_init(void)
>   		ret = register_iommu_dma_ops_notifier(&platform_bus_type);
>   	if (!ret)
>   		ret = register_iommu_dma_ops_notifier(&amba_bustype);
> +
> +	/* handle devices queued before this arch_initcall */
> +	if (!ret)
> +		__iommu_attach_notifier(NULL, BUS_NOTIFY_ADD_DEVICE, NULL);

Hmm, surely unless *all* the devices on the relevant buses are created 
before arch_initcall level, then the next one is going to kick off the 
notifier to walk the whole list anyway. Have you seen this being a 
problem in practice?

Robin.

>   	return ret;
>   }
>   arch_initcall(__iommu_dma_init);
>

[PATCH] ARM64: dma-mapping: fix handling of devices registered before arch_initcall

[Marek Szyprowski]

Hello,

On 2016-02-16 20:05, Robin Murphy wrote:
> On 16/02/16 14:14, Marek Szyprowski wrote:
>> This patch ensures that devices, which got registered before 
>> arch_initcall
>> will be handled correctly by IOMMU-based DMA-mapping code.
>>
>> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
>> ---
>>   arch/arm64/mm/dma-mapping.c | 4 ++++
>>   1 file changed, 4 insertions(+)
>>
>> diff --git a/arch/arm64/mm/dma-mapping.c b/arch/arm64/mm/dma-mapping.c
>> index 573b1e6..ce6019f 100644
>> --- a/arch/arm64/mm/dma-mapping.c
>> +++ b/arch/arm64/mm/dma-mapping.c
>> @@ -928,6 +928,10 @@ static int __init __iommu_dma_init(void)
>>           ret = register_iommu_dma_ops_notifier(&platform_bus_type);
>>       if (!ret)
>>           ret = register_iommu_dma_ops_notifier(&amba_bustype);
>> +
>> +    /* handle devices queued before this arch_initcall */
>> +    if (!ret)
>> +        __iommu_attach_notifier(NULL, BUS_NOTIFY_ADD_DEVICE, NULL);
>
> Hmm, surely unless *all* the devices on the relevant buses are created 
> before arch_initcall level, then the next one is going to kick off the 
> notifier to walk the whole list anyway. Have you seen this being a 
> problem in practice?

Yes, I've observed such situation, that's why I've posted this patch. I was
really surprised that this code doesn't handle all cases.

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

[PATCH] ARM64: dma-mapping: fix handling of devices registered before arch_initcall

[Robin Murphy]

On 17/02/16 07:06, Marek Szyprowski wrote:
> Hello,
>
> On 2016-02-16 20:05, Robin Murphy wrote:
>> On 16/02/16 14:14, Marek Szyprowski wrote:
>>> This patch ensures that devices, which got registered before
>>> arch_initcall
>>> will be handled correctly by IOMMU-based DMA-mapping code.
>>>
>>> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
>>> ---
>>>   arch/arm64/mm/dma-mapping.c | 4 ++++
>>>   1 file changed, 4 insertions(+)
>>>
>>> diff --git a/arch/arm64/mm/dma-mapping.c b/arch/arm64/mm/dma-mapping.c
>>> index 573b1e6..ce6019f 100644
>>> --- a/arch/arm64/mm/dma-mapping.c
>>> +++ b/arch/arm64/mm/dma-mapping.c
>>> @@ -928,6 +928,10 @@ static int __init __iommu_dma_init(void)
>>>           ret = register_iommu_dma_ops_notifier(&platform_bus_type);
>>>       if (!ret)
>>>           ret = register_iommu_dma_ops_notifier(&amba_bustype);
>>> +
>>> +    /* handle devices queued before this arch_initcall */
>>> +    if (!ret)
>>> +        __iommu_attach_notifier(NULL, BUS_NOTIFY_ADD_DEVICE, NULL);
>>
>> Hmm, surely unless *all* the devices on the relevant buses are created
>> before arch_initcall level, then the next one is going to kick off the
>> notifier to walk the whole list anyway. Have you seen this being a
>> problem in practice?
>
> Yes, I've observed such situation, that's why I've posted this patch. I was
> really surprised that this code doesn't handle all cases.

Ok, fair enough - I guess I just didn't consider that being a likely 
possibility. Since I am a fan of correctness:

Acked-by: Robin Murphy <robin.murphy@arm.com>

Thanks,
Robin.

>
> Best regards