Re: [PATCH] firmware: qcom_scm: use correct parameter type for dma_alloc_coherent

2018-02-23 Thread Benjamin Gaignard 
2018-02-08 15:26 GMT+01:00 Benjamin Gaignard :
> 2018-02-08 15:09 GMT+01:00 Arnd Bergmann :
>> On Thu, Feb 8, 2018 at 1:21 PM, Benjamin Gaignard
>>  wrote:
>>> dma_alloc_coherent expects it third argument type to be dma_addr_t and
>>> not phys_addr_t.
>>> When phys_addr_t is defined as u32 and dma_addr_t as u64 that generate
>>> a compilation issue.
>>> Change the variable name because dma_alloc_coherent returns a bus
>>> address not a physical address.
>>>
>>
>>> @@ -480,14 +480,14 @@ int qcom_scm_assign_mem(phys_addr_t mem_addr, size_t 
>>> mem_sz,
>>>
>>> /* Fill details of mem buff to map */
>>> mem_to_map = ptr + ALIGN(src_sz, SZ_64);
>>> -   mem_to_map_phys = ptr_phys + ALIGN(src_sz, SZ_64);
>>> +   mem_to_map_phys = handle + ALIGN(src_sz, SZ_64);
>>> mem_to_map[0].mem_addr = cpu_to_le64(mem_addr);
>>> mem_to_map[0].mem_size = cpu_to_le64(mem_sz);
>>>
>>> next_vm = 0;
>>> /* Fill details of next vmid detail */
>>> destvm = ptr + ALIGN(mem_to_map_sz, SZ_64) + ALIGN(src_sz, SZ_64);
>>> -   dest_phys = ptr_phys + ALIGN(mem_to_map_sz, SZ_64) + ALIGN(src_sz, 
>>> SZ_64);
>>> +   dest_phys = handle + ALIGN(mem_to_map_sz, SZ_64) + ALIGN(src_sz, 
>>> SZ_64);
>>
>> This still assigns from a 'dma' address to a 'phys' address without a
>> long comment
>> explaining why you do it. There is also another instance of the naming 
>> confusion
>> in qcom_scm_pas_init_image and __qcom_scm_pas_init_image, so maybe they
>> should be addressed at the same time.
>
> This patch just fix a compilation issue, nothing else.

Gentle ping on this patch since the problem still exist on 4.16-rc2

>
>>
>>Arnd

Re: [PATCH] firmware: qcom_scm: use correct parameter type for dma_alloc_coherent

2018-02-23 Thread Benjamin Gaignard 
2018-02-08 15:26 GMT+01:00 Benjamin Gaignard :
> 2018-02-08 15:09 GMT+01:00 Arnd Bergmann :
>> On Thu, Feb 8, 2018 at 1:21 PM, Benjamin Gaignard
>>  wrote:
>>> dma_alloc_coherent expects it third argument type to be dma_addr_t and
>>> not phys_addr_t.
>>> When phys_addr_t is defined as u32 and dma_addr_t as u64 that generate
>>> a compilation issue.
>>> Change the variable name because dma_alloc_coherent returns a bus
>>> address not a physical address.
>>>
>>
>>> @@ -480,14 +480,14 @@ int qcom_scm_assign_mem(phys_addr_t mem_addr, size_t 
>>> mem_sz,
>>>
>>> /* Fill details of mem buff to map */
>>> mem_to_map = ptr + ALIGN(src_sz, SZ_64);
>>> -   mem_to_map_phys = ptr_phys + ALIGN(src_sz, SZ_64);
>>> +   mem_to_map_phys = handle + ALIGN(src_sz, SZ_64);
>>> mem_to_map[0].mem_addr = cpu_to_le64(mem_addr);
>>> mem_to_map[0].mem_size = cpu_to_le64(mem_sz);
>>>
>>> next_vm = 0;
>>> /* Fill details of next vmid detail */
>>> destvm = ptr + ALIGN(mem_to_map_sz, SZ_64) + ALIGN(src_sz, SZ_64);
>>> -   dest_phys = ptr_phys + ALIGN(mem_to_map_sz, SZ_64) + ALIGN(src_sz, 
>>> SZ_64);
>>> +   dest_phys = handle + ALIGN(mem_to_map_sz, SZ_64) + ALIGN(src_sz, 
>>> SZ_64);
>>
>> This still assigns from a 'dma' address to a 'phys' address without a
>> long comment
>> explaining why you do it. There is also another instance of the naming 
>> confusion
>> in qcom_scm_pas_init_image and __qcom_scm_pas_init_image, so maybe they
>> should be addressed at the same time.
>
> This patch just fix a compilation issue, nothing else.

Gentle ping on this patch since the problem still exist on 4.16-rc2

>
>>
>>Arnd