Re: [PATCH 04/12] staging: ccree: cleanup lli access macro

[Gilad Ben-Yossef]

On Mon, May 29, 2017 at 5:41 PM, Greg Kroah-Hartman
<gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> On Sun, May 28, 2017 at 05:40:29PM +0300, Gilad Ben-Yossef wrote:
>> The Linked List Item descriptors were being accessed via
>> a baroque set of defines and macro. Re-factor for structs
>> and inline function for readability and sanity.
>>
>> Signed-off-by: Gilad Ben-Yossef <gilad@xxxxxxxxxxxxx>
>> ---
>>  drivers/staging/ccree/cc_lli_defs.h    | 65 +++++++++++++++++++---------------
>>  drivers/staging/ccree/ssi_buffer_mgr.c | 45 +++++------------------
>>  2 files changed, 44 insertions(+), 66 deletions(-)
>>
>> diff --git a/drivers/staging/ccree/cc_lli_defs.h b/drivers/staging/ccree/cc_lli_defs.h
>> index 857b94f..c6b2917 100644
>> --- a/drivers/staging/ccree/cc_lli_defs.h
>> +++ b/drivers/staging/ccree/cc_lli_defs.h
>> @@ -28,36 +28,43 @@
>>
>>  #define CC_MAX_MLLI_ENTRY_SIZE 0x10000
>>
>> -#define LLI_SET_ADDR(__lli_p, __addr) do {                           \
>> -             u32 *lli_p = (u32 *)__lli_p;                            \
>> -             typeof(__addr) addr = __addr;                           \
>> -                                                                     \
>> -             BITFIELD_SET(lli_p[LLI_WORD0_OFFSET],                   \
>> -                     LLI_LADDR_BIT_OFFSET,                           \
>> -                     LLI_LADDR_BIT_SIZE, (addr & U32_MAX));          \
>> -                                                                     \
>> -             BITFIELD_SET(lli_p[LLI_WORD1_OFFSET],                   \
>> -                     LLI_HADDR_BIT_OFFSET,                           \
>> -                     LLI_HADDR_BIT_SIZE, MSB64(addr));               \
>> -     } while (0)
>> -
>> -#define LLI_SET_SIZE(lli_p, size)                                    \
>> -             BITFIELD_SET(((u32 *)(lli_p))[LLI_WORD1_OFFSET],        \
>> -             LLI_SIZE_BIT_OFFSET, LLI_SIZE_BIT_SIZE, size)
>> +#define LLI_MAX_NUM_OF_DATA_ENTRIES 128
>> +#define LLI_MAX_NUM_OF_ASSOC_DATA_ENTRIES 4
>> +#define MLLI_TABLE_MIN_ALIGNMENT 4 /* 32 bit alignment */
>> +#define MAX_NUM_OF_BUFFERS_IN_MLLI 4
>> +#define MAX_NUM_OF_TOTAL_MLLI_ENTRIES (2 * LLI_MAX_NUM_OF_DATA_ENTRIES + \
>> +                                    LLI_MAX_NUM_OF_ASSOC_DATA_ENTRIES)
>> +
>> +struct cc_lli_entry {
>> +#ifndef __LITTLE_ENDIAN__
>> +     u32 addr_lsb;
>> +     u16 size;
>> +     u16 addr_msb;
>> +#else /* __BIG_ENDIAN__ */
>> +     u16 addr_msb;
>> +     u16 size;
>> +     u32 addr_lsb;
>> +#endif
>
> How is the bits within the different variables also not affected by the
> endian issues?  Just moving them around in the 64bits seems really
> strange.

I actually did not overlook this issue. My reasoning was that the new code
is doing exactly what the old code was doing. Either the callers was doing
the right thing, or the HW is magically taking care of it[1] or the
new code is as
broken as the old in a bug to bug compatibility sort of way.

And if it's broken, fixing it in this patch would actually break the
"do one thing"
rule... :-)

[1] I know this sounds strange but the control register description has language
that sounds like it does automagicall swabbing under certain circumstances I am
yet to understand and since I don't have a big endian test system at
my disposable
I chose to leave it as is for now.

>
> Why not just use the "normal" ways to address data in an endian-neutral
> way instead of making your own macros up here?  (i.e. shift and mask.)
>

I personally find it more readable it this way and there are other
example in the
kernel source of doing this) but I will gladly switch if you think it
is preferred.

Thanks,
Gilad

> thanks,
>
> greg k-h



-- 
Gilad Ben-Yossef
Chief Coffee Drinker

"If you take a class in large-scale robotics, can you end up in a
situation where the homework eats your dog?"
 -- Jean-Baptiste Queru