Re: [PATCH 02/32] spi: dw: Add support for 32-bits ctrlr0 layout

From: Sean Anderson <seanga2@gmail.com>
To: Serge Semin <fancer.lancer@gmail.com>,
	Damien Le Moal <damien.lemoal@wdc.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>,
	linux-riscv@lists.infradead.org, Rob Herring <robh+dt@kernel.org>,
	Frank Rowand <frowand.list@gmail.com>,
	devicetree@vger.kernel.org, Mark Brown <broonie@kernel.org>,
	linux-spi@vger.kernel.org, Stephen Boyd <sboyd@kernel.org>,
	linux-clk@vger.kernel.org,
	Linus Walleij <linus.walleij@linaro.org>,
	linux-gpio@vger.kernel.org,
	Philipp Zabel <p.zabel@pengutronix.de>
Subject: Re: [PATCH 02/32] spi: dw: Add support for 32-bits ctrlr0 layout
Date: Mon, 9 Nov 2020 09:33:58 -0500	[thread overview]
Message-ID: <6dc6d52e-a39d-c351-5280-71b9e8eafe37@gmail.com> (raw)
In-Reply-To: <20201109142529.tt5ojjyn36vd3ohe@mobilestation>

On 11/9/20 9:25 AM, Serge Semin wrote:
> Hello Damien,
> Thanks for your patches. My comments are below.
> 
> On Sat, Nov 07, 2020 at 05:13:50PM +0900, Damien Le Moal wrote:
>> Synopsis DesignWare DW_apb_ssi version 4 defines a 32-bit layout of
>> the ctrlr0 register for SPI masters. The layout of ctrlr0 is:
>>
>> |   31 .. 23  | 22 .. 21 | 20 .. 16 |
>> | other stuff | spi_frf  |  dfs_32  |
>>
>> |   15 .. 10  | 9 .. 8 | 7 .. 6 | 5 .. 4 | 3 .. 0 |
>> | other stuff |  tmod  |  mode  |  frf   |  dfs   |
>>
> 
>> Th main difference of this layout with the 16-bits version is the data
>     ^
>     |
>     e
> 
>> frame format field which resides in bits 16..20 instead of bits 3..0.
>>
> 
> Are you sure they have been moved from [0, 3] to [16, 20]? I don't have the
> manual for the 4.0x version of the core, but according to this patch:
> https://patchwork.kernel.org/project/spi-devel-general/patch/1575907443-26377-7-git-send-email-wan.ahmad.zainie.wan.mohamad@intel.com/
> it has been ok to use the lowest four bits for DFS setting. Is the commit
> message misleading there?

This commit message is a truncated version of [1]. Importantly, DFS is
valid when SSI_MAX_XFER_SIZE=16. When it =32, then DFS_32 must be used
(since DFS is constant 0xF). Since SSI_MAX_XFER_SIZE is a synthesis
parameter, there exist devices where DFS must be used, and also where
DFS_32 must be used. 

[1] https://patchwork.ozlabs.org/project/uboot/patch/20201016225755.302659-10-seanga2@gmail.com/

--Sean

> 
>> Introduce the DW SPI capability flag DW_SPI_CAP_DFS_32 to let a
>> platform signal that this layout is in use. Modify
>> dw_spi_update_config() to test this capability flag to set the data
>> frame format field at the correct register location.
>>
>> Suggested-by: Sean Anderson <seanga2@gmail.com>
>> Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
>> ---
>>  drivers/spi/spi-dw-core.c | 8 ++++++--
>>  drivers/spi/spi-dw.h      | 9 +++++++++
>>  2 files changed, 15 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/spi/spi-dw-core.c b/drivers/spi/spi-dw-core.c
>> index 2e50cc0a9291..841c85247f01 100644
>> --- a/drivers/spi/spi-dw-core.c
>> +++ b/drivers/spi/spi-dw-core.c
>> @@ -311,8 +311,12 @@ void dw_spi_update_config(struct dw_spi *dws, struct spi_device *spi,
>>  	u32 speed_hz;
>>  	u16 clk_div;
>>  
> 
>> -	/* CTRLR0[ 4/3: 0] Data Frame Size */
>> -	cr0 |= (cfg->dfs - 1);
>> +	if (!(dws->caps & DW_SPI_CAP_DFS_32))
>> +		/* CTRLR0[ 4/3: 0] Data Frame Size */
>> +		cr0 |= (cfg->dfs - 1);
>> +	else
>> +		/* CTRLR0[20: 16] Data Frame Size */
>> +		cr0 |= (cfg->dfs - 1) << DWC_APB_CTRLR0_32_DFS_OFFSET;
> 
> If you extend the dfs field from four to five bits, then
> controller->bits_per_word_mask field should be properly updated too.
> 
> Alas it hasn't been done for the DWC_ssi version of the core. So I suppose it
> should be fixed for the both of them.
> 
> Just for the record. There are very handy macros for setting and getting bit fields
> to/from a variable. This is a good place to use them instead of manually
> shifting and defining the offsets. The macros are defined in linux/bitfield.h .
> Alas this driver hasn't been converted to using them. So I won't insist on using
> them here. But I hope someone will fix it sometime in future...

I second that request.

> -Sergey
> 
>>  
>>  	if (!(dws->caps & DW_SPI_CAP_DWC_SSI))
>>  		/* CTRLR0[ 9:8] Transfer Mode */
>> diff --git a/drivers/spi/spi-dw.h b/drivers/spi/spi-dw.h
>> index faf40cb66498..48a11a51a407 100644
>> --- a/drivers/spi/spi-dw.h
>> +++ b/drivers/spi/spi-dw.h
>> @@ -9,6 +9,7 @@
>>  #include <linux/io.h>
>>  #include <linux/scatterlist.h>
>>  #include <linux/spi/spi-mem.h>
>> +#include <linux/bitfield.h>
>>  
>>  /* Register offsets */
>>  #define DW_SPI_CTRLR0			0x00
>> @@ -72,6 +73,13 @@
>>  #define DWC_SSI_CTRLR0_FRF_OFFSET	6
>>  #define DWC_SSI_CTRLR0_DFS_OFFSET	0
>>  
>> +/*
>> + * Bit fields in CTRLR0 for DWC_apb_ssi v4 32-bits ctrlr0.
>> + * Based on DW_apb_ssi Databook v4.02a.
>> + */
>> +#define DWC_APB_CTRLR0_32_DFS_OFFSET	16
>> +#define DWC_APB_CTRLR0_32_DFS_MASK	GENMASK(20, 16)
>> +
>>  /*
>>   * For Keem Bay, CTRLR0[31] is used to select controller mode.
>>   * 0: SSI is slave
>> @@ -121,6 +129,7 @@ enum dw_ssi_type {
>>  #define DW_SPI_CAP_CS_OVERRIDE		BIT(0)
>>  #define DW_SPI_CAP_KEEMBAY_MST		BIT(1)
>>  #define DW_SPI_CAP_DWC_SSI		BIT(2)
>> +#define DW_SPI_CAP_DFS_32		BIT(3)
>>  
>>  /* Slave spi_transfer/spi_mem_op related */
>>  struct dw_spi_cfg {
>> -- 
>> 2.28.0
>>