Re: [PATCH] dt-bindings: nand: denali: reduce the register space in the example

From: Masahiro Yamada
Date: Thu Sep 21 2017 - 07:11:38 EST


2017-09-21 15:09 GMT+09:00 Oleksij Rempel <ore@xxxxxxxxxxxxxx>:
> Hi,
>
> On 21.09.2017 07:26, Masahiro Yamada wrote:
>>
>> Hi.
>
>
> ......
>
>>> Hm.. according to
>>>
>>> https://www.altera.com/en_US/pdfs/literature/hb/cyclone-v/cyclone5_handbook.pdf
>>> Table 13-18: NAND Controller Module Data Space Address Range
>>>
>>> Module Instance Start Address End Address
>>> NAND_DATA 0xFF900000 0xFF9FFFFF
>>>
>>> So <0xff900000 0x100000> seems to be a proper value.
>>>
>>
>> The Alrera's SOCFPGA document describes so.
>> It is up to each SoC vendor how to describe the register space.
>>
>> I am focusing on the Denali IP
>> because this IP is used among several SoCs.
>>
>>
>>
>> You can see the peripheral region map
>> starting at page 1-18 of the document you referred to:
>>
>> Slave ID Description Base Addr Size
>> ---------------------------------------------------------------
>> L3REGS L3 interconnect GPV 0xFF800000 1 MB
>> NANDDATA NAND flash controller data 0xFF900000 64 KB
>> QSPIDATA Quad SPI flash data 0xFFA00000 1 MB
>>
>> (In the doc, the base is described as 0xFFB900000, but this is
>> apparently a typo.)
>>
>>
>> The rationale of the "End Address 0xFF9FFFFF" of NAND_DATA
>> is the fact that the base address of the next peripheral (QSPIDATA) is
>> 0xFFA00000.
>>
>>
>>
>>
>> One more, if you look at the next page,
>>
>> Slave ID Description Base Addr Size
>> ---------------------------------------------------------------
>> NANDREGS NAND flash controller registers 0xFFB80000 64 KB
>> FPGAMGRDATA FPGA manager configuration data 0xFFB90000 4 KB
>>
>>
>> The size of NAND register space is described as 64KB,
>> but the rationale is just the start of the next peripheral is 0xFFB90000.
>>
>> (0xFFB90000 - 0xFFB80000 = 0x10000 = 64KB)
>>
>>
>>
>> Altera apparently reserved address space just for the purpose
>> of matching the end address to the base address of the next peripheral.
>>
>>
>> That means, this document specifies address region
>> much bigger than the IP actually provides.
>>
>>
>> If you look at page 13-6, there are only two registers
>> in NANDDATA space.
>>
>> Table 13-4: Register Map for Indexed Addressing
>> Control 0x0
>> Data 0x10
>>
>>
>>
>> For NANDREGS, in page 13-106, the following is the last register
>> in the NANDREGS space.
>>
>> lun_status_cmd
>> Offset 0x7A0
>>
>> Obviously, 0x1000 (4KB) is enough for NANDREGS.
>>
>> To conclude this, this binding document was written
>> based on the Altera's SOCFPGA specification.
>>
>> Altera specifies the region size
>> so that end address matches to the base of the next peripheral.
>> This is just a matter of SOCFPGA address mapping.
>>
>> In my opinion, the binding document should not be oriented
>> to a particular SoC, which is not true for other SoCs.
>
> ok. Thank you for detailed response.



BTW, I am not forcing you to change the SOCFPGA DTS.
It is up to you and Dihn whether change it or not.

I do not have any SOCFPGA board.



--
Best Regards
Masahiro Yamada