Re: [PATCH v16 07/22] cxl: Support dpa initialization without a mailbox

[Alejandro Lucero Palau <alucerop@amd.com>]

On 5/21/25 19:47, Dan Williams wrote:
> alejandro.lucero-palau@ wrote:
>> From: Alejandro Lucero <alucerop@amd.com>
>>
>> Type3 relies on mailbox CXL_MBOX_OP_IDENTIFY command for initializing
>> memdev state params which end up being used for DMA initialization.
>>
>> Allow a Type2 driver to initialize DPA simply by giving the size of its
>> volatile and/or non-volatile hardware partitions.
>>
>> Export cxl_dpa_setup as well for initializing those added DPA partitions
>> with the proper resources.
>>
>> Signed-off-by: Alejandro Lucero <alucerop@amd.com>
>> Reviewed-by: Ben Cheatham <benjamin.cheatham@amd.com>
>> Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
>> ---
>>   drivers/cxl/core/mbox.c | 26 ++++++++++++++++++++------
>>   drivers/cxl/cxlmem.h    | 13 -------------
>>   include/cxl/cxl.h       | 14 ++++++++++++++
>>   3 files changed, 34 insertions(+), 19 deletions(-)
>>
>> diff --git a/drivers/cxl/core/mbox.c b/drivers/cxl/core/mbox.c
>> index ab994d459f46..b14cfc6e3dba 100644
>> --- a/drivers/cxl/core/mbox.c
>> +++ b/drivers/cxl/core/mbox.c
>> @@ -1284,6 +1284,22 @@ static void add_part(struct cxl_dpa_info *info, u64 start, u64 size, enum cxl_pa
>>   	info->nr_partitions++;
>>   }
>>   
>> +/**
>> + * cxl_mem_dpa_init: initialize dpa by a driver without a mailbox.
>> + *
>> + * @info: pointer to cxl_dpa_info
>> + * @volatile_bytes: device volatile memory size
>> + * @persistent_bytes: device persistent memory size
>> + */
>> +void cxl_mem_dpa_init(struct cxl_dpa_info *info, u64 volatile_bytes,
>> +		      u64 persistent_bytes)
> I struggle to imagine a Type-2 device with PMEM, or needing anything
> more complicated than a single volatile range. No need to pre-enable
> something that may never exist.
>
> Lets just have a cxl_set_capacity() for the simple / common case:
>
> int cxl_set_capacity(struct cxl_dev_state *cxlds, u64 capacity)
> {
> 	struct cxl_dpa_info range_info = { 0 };
>
> 	add_part(info, 0, capacity, CXL_PARTMODE_RAM);
> 	return cxl_dpa_setup(cxlds, &range_info);
> }
>
> ...then there is no need to move 'struct cxl_dpa_info' to a public
> header, or require type-2 drivers to pass in a pointless PMEM capacity.
>
> If more complicated devices show up later the code can always be made
> more sophisticated at that point.


That seems fine to me. The only problem I see is a driver with a mailbox 
will initialize this differently, getting the cxl_dpa_info first, then 
calling cxl_setup_dpa, or all that also hidden in another function. In 
any case, I guess the first driver needing that will have to work it out.