Re: [PATCH v17 1/7] firmware: arm_rmm: Add SMC definitions for calling the RMM

[Gavin Shan <gshan@redhat.com>]

On 9/9/26 6:39 PM, Suzuki K Poulose wrote:
> On 08/09/2026 23:41, Gavin Shan wrote:
>> On 9/8/26 8:37 PM, Suzuki K Poulose wrote:
>>> On 08/09/2026 07:19, Gavin Shan wrote:
>>>> On 9/7/26 7:59 PM, Suzuki K Poulose wrote:
>>>>> From: Steven Price <steven.price@arm.com>
>>>>>
>>>>> The RMM (Realm Management Monitor) provides functionality that can be
>>>>> accessed by SMC calls from the host.
>>>>>
>>>>> The SMC definitions are based on DEN0137[1] version 2.0-bet3
>>>>>
>>>>> [1] https://developer.arm.com/documentation/den0137/2-0bet3/
>>>>>
>>>>> Signed-off-by: Steven Price <steven.price@arm.com>
>>>>> Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
>>>
>>> ...
>>>
>>>>> ---
>>>>>   include/linux/arm-smccc-rmi.h | 494 ++++++++++++++++++++++++++++++ ++++
>>>>>   1 file changed, 494 insertions(+)
>>>>>   create mode 100644 include/linux/arm-smccc-rmi.h
>>>>>
>>>>
>>>> Some nitpicks below, please pick up the commits that look reasonable :)
>>>>
>>>
>>>
>>>>> +
>>>>> +#define RMI_RETURN_STATUS_MASK        (0xFFUL)
>>>>> +#define RMI_RETURN_INDEX_MASK        (0xFFUL << 8)
>>>>> +#define RMI_RETURN_MEMREQ_MASK        (0x3UL << 8)
>>>>> +#define RMI_RETURN_CAN_CANCEL_MASK    (0x1UL << 10)
>>>>> +
>>>>> +#define RMI_RETURN_STATUS(ret) FIELD_GET(RMI_RETURN_STATUS_MASK, ret)
>>>>> +#define RMI_RETURN_INDEX(ret) FIELD_GET(RMI_RETURN_INDEX_MASK, ret)
>>>>> +#define RMI_RETURN_MEMREQ(ret) FIELD_GET(RMI_RETURN_MEMREQ_MASK, ret)
>>>>> +#define RMI_RETURN_CAN_CANCEL(ret) FIELD_GET(RMI_RETURN_CAN_CANCEL_MASK, ret)
>>>>> +
>>>>
>>>> If I'm correct enough, RMI_RETURN_{STATUS, INDEX, MEMREQ, CAN_CANEL} _MASK are used
>>>> for once in this header file. So we needn't explicitly expose them and combine their
>>>> definitions with users to fetch the corresponding fields, as below.
>>>>
>>>> #define RMI_RETURN_STATUS(ret)        FIELD_GET(GENMASK(7, 0), ret)
>>>> #define RMI_RETURN_INDEX(ret)        FIELD_GET(GENMASK(15, 8), ret)
>>>> #define RMI_RETURN_MEMREQ(ret)        FIELD_GET(GENMASK(10, 8), ret)
>>>> #define RMI_RETURN_CAN_CANCEL(ret)    FIELD_GEt(GENMASK(10, 10), ret)
>>>
>>> I belive the header files were generated from a tool (by Steven) and
>>> these would make it easier to compare with the generated header file for
>>> any deviation from the spec. So, I am a bit reluctant make these
>>> changes.
>>>
>>> That said, I am happy to convert the mask definitions to GENMASK() for
>>> consistency and readability. i.e., something like:
>>>
>>> diff --git a/include/linux/arm-smccc-rmi.h b/include/linux/arm-smccc- rmi.h
>>> index 3eb88caf40964..078197284415e 100644
>>> --- a/include/linux/arm-smccc-rmi.h
>>> +++ b/include/linux/arm-smccc-rmi.h
>>> @@ -139,10 +139,10 @@
>>>   #define RMI_ABI_VERSION_GET_MINOR(version) ((version) & 0xFFFF)
>>>   #define RMI_ABI_VERSION(major, minor)      (((major) << 16) | (minor))
>>>
>>> -#define RMI_RETURN_STATUS_MASK         (0xFFUL)
>>> -#define RMI_RETURN_INDEX_MASK          (0xFFUL << 8)
>>> -#define RMI_RETURN_MEMREQ_MASK         (0x3UL << 8)
>>> -#define RMI_RETURN_CAN_CANCEL_MASK     (0x1UL << 10)
>>> +#define RMI_RETURN_STATUS_MASK         GENMASK(7, 0)
>>> +#define RMI_RETURN_INDEX_MASK          GENMASK(15, 8)
>>> +#define RMI_RETURN_MEMREQ_MASK         GENMASK(10, 8)
>>> +#define RMI_RETURN_CAN_CANCEL_MASK     BIT(10)
>>>
>>
>> Ok.
>>
>>>>
>>>>> +#define RMI_SUCCESS            0
>>>>> +#define RMI_ERROR_INPUT            1
>>>>> +#define RMI_ERROR_REALM            2
>>>>> +#define RMI_ERROR_REC            3
>>>>> +#define RMI_ERROR_RTT            4
>>>>> +#define RMI_ERROR_NOT_SUPPORTED        5
>>>>> +#define RMI_ERROR_DEVICE        6
>>>>> +#define RMI_ERROR_RTT_AUX        7
>>>>> +#define RMI_ERROR_PSMMU_ST        8
>>>>> +#define RMI_ERROR_DPT            9
>>>>> +#define RMI_BUSY            10
>>>>> +#define RMI_ERROR_GLOBAL        11
>>>>> +#define RMI_ERROR_TRACKING        12
>>>>> +#define RMI_INCOMPLETE            13
>>>>> +#define RMI_BLOCKED            14
>>>>> +#define RMI_ERROR_GPT            15
>>>>> +#define RMI_ERROR_GRANULE        16
>>>>> +
>>>>> +#define RMI_CONTINUE_KEEP_GOING        0
>>>>> +#define RMI_CONTINUE_STOP        1
>>>>> +
>>>>> +#define RMI_OP_MEM_REQ_NONE        0
>>>>> +#define RMI_OP_MEM_REQ_DONATE        1
>>>>> +#define RMI_OP_MEM_REQ_RECLAIM        2
>>>>> +
>>>>> +#define RMI_DONATE_SIZE_MASK        3UL
>>>>> +#define RMI_DONATE_COUNT_MASK        GENMASK(15, 2)
>>>>> +#define RMI_DONATE_CONTIG_MASK        BIT(16)
>>>>> +#define RMI_DONATE_STATE_MASK        GENMASK(18, 17)
>>>>> +
>>>>> +#define RMI_DONATE_SIZE(req)        FIELD_GET(RMI_DONATE_SIZE_MASK, req)
>>>>> +#define RMI_DONATE_COUNT(req) FIELD_GET(RMI_DONATE_COUNT_MASK, req)
>>>>> +#define RMI_DONATE_CONTIG(req) FIELD_GET(RMI_DONATE_CONTIG_MASK, req)
>>>>> +#define RMI_DONATE_STATE(req) FIELD_GET(RMI_DONATE_STATE_MASK, req)
>>>>> +
>>>>
>>>> As above, RMI_DONATE_{SIZE, COUNT, CONTIG, STATE}_MASK are used for once in this
>>>> header file. So their definitions can be dropped by modifying the followup macros,
>>>> as below. Some enhancements are also applicable: (a) Use 'BLOCK' to indicate the
>>>> unit encoded in the request. (b) The macros are put into order from MSB to LSB.
>>>>
>>>> #define RMI_DONATE_STATE(req)        FIELD_GET(GENMASK(18, 17), req)
>>>> #define RMI_DONATE_CONTIG(req)        FIELD_GET(GENMASK(16, 16), req)
>>>> #define RMI_DONATE_BLOCK_COUNT(req)    FIELD_GET(GENMASK(15, 2), req)
>>>> #define RMI_DONATE_BLOCK_SIZE(req)    FIELD_GET(GENMASK(1, 0), req)
>>>
>>> Same as above.
>>>
>>
>> Please rename RMI_DONATE_{SIZE, COUNT} to RMI_DONATE_BLOCK_{SIZE, COUNT} if possible.
> 
> Sorry, forgot to add that, I have renamed RMI_{DONATE,ADDR_RANGE}_SIZE
> => RMI_{*}_BLOCK_SIZE, left  the _COUNT as it is.
> 

It's ok to me. The point is to have "BLOCK", indicating it's related to
a block. In rmi.c, 'unit' is also used and needs to be replaced with 'block',
consistent with the spec.

>>
>>>>
>>>>> +#define RMI_OP_MEM_DELEGATED        0
>>>>> +#define RMI_OP_MEM_UNDELEGATED        1
>>>>> +#define RMI_OP_MEM_CONDITIONAL        2
>>>>> +
>>>>
>>>> B4.6.40 RmiOpMemContig type is missed here and they should be used in the c code.
>>>>
>>>> #define RMI_OP_MEM_NON_CONTIG    0
>>>> #define RMI_OP_MEM_CONTIG    1
>>>>
>>>> In the c code, we shall have:
>>>>
>>>> if (RMI_DONATE_CONTIG(req) == RMI_OP_MEM_CONTIG)) {
>>>>     ...
>>>> } else {
>>>>     ...
>>>> }
>>>>
>>>
>>> Does it make much sense though ? I prefer
>>>
>>> if (RMI_DONATE_CONTIG(req))
>>>
>>> to
>>>
>>>
>>> if (RMI_DONATE_CONTIG(req) == RMI_OP_MEM_CONTIG)) {
>>>
>>>
>>> than having a field for a single bit values. I am happy to change
>>> this if there is strong preference to the other.
>>>
>>
>> The point is RMM-v2.0-bet3 expose RMI_OP_{MEM, MEM_NON}_CONTIG instead of true/false
>> to indicate if the requested memory blocks are physically contiguous or not. So the
>> expression 'if (RMI_DONATE_CONTIG(req) == RMI_OP_MEM_CONTIG)' is consistent to the
>> RMM-v2.0-bet3 spec.
> 
> Agree, but the values are intuitively assigned to indicate whether it is
> 1/0. I personally think if (RMI_DONATE_CONTIG(req)) is good enough. But
> to be consistent with the spec, I will change them.
> 

Ok.

Thanks,
Gavin

> 
> Cheers
> Suzuki
>