On 8/14/2025 9:20 AM, Konrad Dybcio wrote:
> On 8/14/25 1:19 AM, Amirreza Zarrabi wrote:
>>
>>
>> On 8/14/2025 8:49 AM, Konrad Dybcio wrote:
>>> On 8/14/25 12:24 AM, Amirreza Zarrabi wrote:
>>>>
>>>>
>>>> On 8/13/2025 8:00 PM, Konrad Dybcio wrote:
>>>>> On 8/13/25 2:35 AM, Amirreza Zarrabi wrote:
>>>>>> Enable userspace to allocate shared memory with QTEE. Since
>>>>>> QTEE handles shared memory as object, a wrapper is implemented
>>>>>> to represent tee_shm as an object. The shared memory identifier,
>>>>>> obtained through TEE_IOC_SHM_ALLOC, is transferred to the driver using
>>>>>> TEE_IOCTL_PARAM_ATTR_TYPE_OBJREF_INPUT/OUTPUT.
>>>>>>
>>>>>> Tested-by: Neil Armstrong <neil.armstrong@xxxxxxxxxx>
>>>>>> Acked-by: Sumit Garg <sumit.garg@xxxxxxxxxxxxxxxx>
>>>>>> Tested-by: Harshal Dev <quic_hdev@xxxxxxxxxxx>
>>>>>> Signed-off-by: Amirreza Zarrabi <amirreza.zarrabi@xxxxxxxxxxxxxxxx>
>>>>>> ---
>>>>>
>>>>> [...]
>>>>>
>>>>>> +/* Mapping information format as expected by QTEE. */
>>>>>> +struct qcomtee_mapping_info {
>>>>>> + u64 paddr;
>>>>>> + u64 len;
>>>>>> + u32 perms;
>>>>>> +} __packed;
>>>>>
>>>>> Please use types with explicit endianness, e.g. __le32. I'm assuming
>>>>> TZ will always be little-endian, regardless of the host OS
>>>>>
>>>>
>>>> I'm not entirely sure how this point is relevant. As I understand it,
>>>> the core that populates this struct is the same one that accesses it in TZ.
>>>> Your argument would absolutely make sense if the host and TZ were operating
>>>> on different cores with distinct architectures -- such as one being
>>>> little-endian and the other big-endian, which is not the case.
>>>
>>> CONFIG_CPU_BIG_ENDIAN=y exists on arm64
>>>
>>
>> Or, you are saying we may have a configuration where host is big-endian
>> but TZ is little-endian?
>
> I was indeed about to say that.. I believe our tz is always little-endian
> but you can run the HLOS of either endianness
>
Okay, I’ve thought about this point and also discussed it internally.
I can add an entry to Kconfig to ensure the driver is unavailable when
CONFIG_CPU_BIG_ENDIAN is selected -- for now. However, without a clear
usecase, adding all the le32_to_cpu/cpu_to_le32 conversions just makes
the driver messy.
Regards,
Amir
> Konrad
