[Skiboot] [PATCH v7 18/26] fsp: endian conversions

[Nicholas Piggin]

Vasant Hegde's on November 12, 2019 8:42 pm:
> On 11/11/19 11:19 AM, Nicholas Piggin wrote:
>> Reviewed-by: Vasant Hegde <hegdevasant at linux.vnet.ibm.com>
>> Signed-off-by: Nicholas Piggin <npiggin at gmail.com>
>> ---
>>   hw/fsp/fsp-codeupdate.c |  4 +-
>>   hw/fsp/fsp-codeupdate.h | 86 ++++++++++++++++++++---------------------
>>   hw/fsp/fsp-sysdump.c    |  2 +-
>>   hw/fsp/fsp-sysparam.c   |  2 +-
>>   hw/fsp/fsp.c            | 16 ++++----
> 
> 
> I tried to look into couple of drivers under hw/fsp/* (not yet looked into all 
> files).
> Here are few places we need fixes.
> 
> hw/fsp/fsp-attn.c
> 
> 109         ti_attn->msg_len = VERSION_LEN + BT_FRAME_LEN +
> 110                                    strlen(ti_attn->msg.file_info);
> 111 }
> 
> 
> hw/fsp/fsp-dump.c
> 
> 376         *dump_id = record->id;
> 377         *dump_size = record->size;
> 378         *dump_type = record->type;
> 
> Data returned by fsp_get_sys_param() needs to be converted (we do TCE mapping of 
> buffer and FSP
> will fill BE data).

Okay, I can do another pass over it and take another look. If we give
the OPAL APIs and hardware / in-memory structures the right endian
annotated types, then it mostly becomes a mechanical process of 
following sparse.

> I tried to boot ZZ with this patch series (both LE and BE mode). its not booting 
> for me. I suspect HDAT related changes causing some issue (haven't looked into 
> it in detail).

Thanks, I'll see if I can fix it. We can't be breaking BE builds.

> 
> fsp_mkmsg() usage needs careful audit .. as fsp_wreg() -> out_be32() converts 
> data from cpu_to_be().

I don't expect everything to work with LE after this series, but basic
witherspoon is working so it gives a starting point. I'll have a bit 
more look over the FSP conversions. If it becomes tricky and you'd like
more time to go over it, I can take it out of the initial series.

Thanks,
Nick