Re: [PATCH v2 10/10] spi: atmel-quadspi: add support for sam9x60 qspi controller

[Tudor.Ambarus]

On 01/31/2019 06:32 PM, Boris Brezillon wrote:
> On Thu, 31 Jan 2019 16:15:51 +0000
> <Tudor.Ambarus@xxxxxxxxxxxxx> wrote:
> 
>> From: Tudor Ambarus <tudor.ambarus@xxxxxxxxxxxxx>
>>
>> The sam9x60 qspi controller uses 2 clocks, one for the peripheral register
>> access, the other for the qspi core and phy. Both are mandatory. It uses
>> dedicated register for Read Instruction Code Register (RICR) and
>> Write Instruction Code Register (WICR). ICR/RICR/WICR have identical
>> fields.
>>
>> Tested with sst26vf064b jedec,spi-nor flash. Backward compatibility test
>> done on sama5d2 qspi controller and mx25l25635e jedec,spi-nor flash.
>>
>> Signed-off-by: Tudor Ambarus <tudor.ambarus@xxxxxxxxxxxxx>
>> ---
>> v2:
>> - rework clock handling
>> - reorder setting of register values in set_cfg() calls -> move functions
>>   that can fail in the upper part of the function body.
>>
>>  drivers/spi/atmel-quadspi.c | 296 +++++++++++++++++++++++++++++++++++---------
>>  1 file changed, 239 insertions(+), 57 deletions(-)
>>
>> diff --git a/drivers/spi/atmel-quadspi.c b/drivers/spi/atmel-quadspi.c
>> index d3e76acf8517..80c934f3e479 100644
>> --- a/drivers/spi/atmel-quadspi.c
>> +++ b/drivers/spi/atmel-quadspi.c
>> @@ -19,6 +19,7 @@
>>  #include <linux/kernel.h>
>>  #include <linux/module.h>
>>  #include <linux/of.h>
>> +#include <linux/of_platform.h>
>>  #include <linux/platform_device.h>
>>  #include <linux/spi/spi-mem.h>
>>  
>> @@ -35,7 +36,9 @@
>>  
>>  #define QSPI_IAR     0x0030  /* Instruction Address Register */
>>  #define QSPI_ICR     0x0034  /* Instruction Code Register */
>> +#define QSPI_WICR    0x0034  /* Write Instruction Code Register */
>>  #define QSPI_IFR     0x0038  /* Instruction Frame Register */
>> +#define QSPI_RICR    0x003C  /* Read Instruction Code Register */
>>  
>>  #define QSPI_SMR     0x0040  /* Scrambling Mode Register */
>>  #define QSPI_SKR     0x0044  /* Scrambling Key Register */
>> @@ -88,7 +91,7 @@
>>  #define QSPI_SCR_DLYBS_MASK             GENMASK(23, 16)
>>  #define QSPI_SCR_DLYBS(n)               (((n) << 16) & QSPI_SCR_DLYBS_MASK)
>>  
>> -/* Bitfields in QSPI_ICR (Instruction Code Register) */
>> +/* Bitfields in QSPI_ICR (Read/Write Instruction Code Register) */
>>  #define QSPI_ICR_INST_MASK              GENMASK(7, 0)
>>  #define QSPI_ICR_INST(inst)             (((inst) << 0) & QSPI_ICR_INST_MASK)
>>  #define QSPI_ICR_OPT_MASK               GENMASK(23, 16)
>> @@ -113,6 +116,8 @@
>>  #define QSPI_IFR_OPTL_4BIT              (2 << 8)
>>  #define QSPI_IFR_OPTL_8BIT              (3 << 8)
>>  #define QSPI_IFR_ADDRL                  BIT(10)
>> +#define QSPI_IFR_TFRTYP_TRSFR_MEM	BIT(12)
>> +#define QSPI_IFR_TFRTYP_TRSFR_REG	(0 << 12)
> 
> You don't need to define TRSFR_REG, just set QSPI_IFR_TFRTYP_TRSFR_MEM
> when you do a mem transfer and do nothing when this is a regular
> transfer.


I chose to introduce macros with zero value for better code readability. I would
expect that the NOP operations to be optimized at compile time. I will remove them
if you prefer, it will result in fewer lines of code.

> 
>>  #define QSPI_IFR_TFRTYP_MASK            GENMASK(13, 12)
>>  #define QSPI_IFR_TFRTYP_TRSFR_READ      (0 << 12)
>>  #define QSPI_IFR_TFRTYP_TRSFR_READ_MEM  (1 << 12)
> 
> Looks like the read/write flag is on bit 13. Can we just add

for sama5d2 only

> 
> #define QSPI_IFR_TFRTYP_TRSFR_WRITE	BIT(13)
> 
> and drop all others def? This way the implementation is consistent
> between sam9x60 and sama5d2.

BIT(13) has no meaning for sam9x60. I can drop the macros with zero value for
sama5d2 in a separate patch.

> 
>> @@ -121,6 +126,8 @@
>>  #define QSPI_IFR_CRM                    BIT(14)
>>  #define QSPI_IFR_NBDUM_MASK             GENMASK(20, 16)
>>  #define QSPI_IFR_NBDUM(n)               (((n) << 16) & QSPI_IFR_NBDUM_MASK)
>> +#define QSPI_IFR_APBTFRTYP_WRITE	(0 << 24)
> 
> As for the other defs, I don't think you need to define _WRITE.

understood

> 
>> +#define QSPI_IFR_APBTFRTYP_READ		BIT(24)
>>  
>>  /* Bitfields in QSPI_SMR (Scrambling Mode Register) */
>>  #define QSPI_SMR_SCREN                  BIT(0)
>> @@ -137,16 +144,37 @@
>>  #define QSPI_WPSR_WPVSRC(src)           (((src) << 8) & QSPI_WPSR_WPVSRC)
>>  
>>  
>> +/* Describes register values. */
>> +struct atmel_qspi_cfg {
>> +	u32 icr;
>> +	u32 iar;
>> +	u32 ifr;
>> +};
>> +
>> +struct atmel_qspi_caps;
>> +
>>  struct atmel_qspi {
>>  	void __iomem		*regs;
>>  	void __iomem		*mem;
>>  	struct clk		*clk;
> 
> Can we rename that on pclk?

will rename it, together with the support for unnamed clock of sama5d2 in a separate
patch. The dt-bindings patch that imposes "pclk" for sama5d2 should be separated too.

> 
>> +	struct clk		*qspick;
>>  	struct platform_device	*pdev;
>> +	const struct atmel_qspi_caps *caps;
>>  	u32			pending;
>>  	u32			mr;
>>  	struct completion	cmd_completion;
>>  };
>>  
> 
> ...
> 
>> +
>> +static int atmel_sam9x60_qspi_set_cfg(void __iomem *base,
>> +				      const struct spi_mem_op *op,
>> +				      struct atmel_qspi_cfg *cfg)
>> +{
>> +	int ret = atmel_qspi_set_mode(cfg, op);
>> +
>> +	if (ret)
>> +		return ret;
>> +
>> +	ret = atmel_qspi_set_address_mode(cfg, op);
>> +	if (ret)
>> +		return ret;
>> +
>> +	cfg->ifr |= QSPI_IFR_INSTEN;
>> +	cfg->icr |= QSPI_ICR_INST(op->cmd.opcode);
>> +
>> +	/* Set data enable */
>> +	if (op->data.nbytes)
>> +		cfg->ifr |= QSPI_IFR_DATAEN;
>> +
>> +	if (!op->addr.nbytes) {
>> +		cfg->ifr |= QSPI_IFR_TFRTYP_TRSFR_REG;
>> +		if (op->data.dir == SPI_MEM_DATA_OUT)
>> +			cfg->ifr |= QSPI_IFR_APBTFRTYP_WRITE;
>> +		else
>> +			cfg->ifr |= QSPI_IFR_APBTFRTYP_READ;
>> +	} else {
>> +		cfg->ifr |= QSPI_IFR_TFRTYP_TRSFR_MEM;
> 
> Can you try doing only regular transfers and let me know if it still
> works. Support for mem transfers can then be added along with dirmap
> support.

should work. Will try and let you know.

> 
>> +	}
>>  
>>  	/* Clear pending interrupts */
>>  	(void)readl_relaxed(base + QSPI_SR);
>>  
>>  	/* Set QSPI Instruction Frame registers */
>> -	writel_relaxed(iar, base + QSPI_IAR);
>> -	writel_relaxed(icr, base + QSPI_ICR);
>> -	writel_relaxed(ifr, base + QSPI_IFR);
>> +	writel_relaxed(cfg->iar, base + QSPI_IAR);
>> +	if (op->data.dir == SPI_MEM_DATA_OUT)
>> +		writel_relaxed(cfg->icr, base + QSPI_ICR);
>> +	else
>> +		writel_relaxed(cfg->icr, base + QSPI_RICR);
>> +	writel_relaxed(cfg->ifr, base + QSPI_IFR);
>> +
>> +	return 0;
>> +}
>> +
> 
> ...
> 
>> @@ -443,32 +578,52 @@ static int atmel_qspi_probe(struct platform_device *pdev)
>>  	/* Enable the peripheral clock */
>>  	err = clk_prepare_enable(aq->clk);
>>  	if (err) {
>> -		dev_err(&pdev->dev, "failed to enable the peripheral clock\n");
>> +		dev_err(dev, "failed to enable the peripheral clock\n");
>>  		goto exit;
>>  	}
>>  
>> +	if (caps->has_qspick) {
>> +		/* Get the QSPI system clock */
>> +		aq->qspick = devm_clk_get(dev, "qspick");
>> +		if (IS_ERR(aq->qspick)) {
>> +			dev_err(dev, "missing system clock\n");
>> +			err = PTR_ERR(aq->qspick);
>> +			goto disable_clk;
>> +		}
>> +
>> +		/* Enable the QSPI system clock */
>> +		err = clk_prepare_enable(aq->qspick);
>> +		if (err) {
>> +			dev_err(dev,
>> +				"failed to enable the QSPI system clock\n");
>> +			goto disable_clk;
>> +		}
>> +	}
>> +
>>  	/* Request the IRQ */
>>  	irq = platform_get_irq(pdev, 0);
>>  	if (irq < 0) {
>> -		dev_err(&pdev->dev, "missing IRQ\n");
>> +		dev_err(dev, "missing IRQ\n");
>>  		err = irq;
>> -		goto disable_clk;
>> +		goto disable_qspick;
>>  	}
>> -	err = devm_request_irq(&pdev->dev, irq, atmel_qspi_interrupt,
>> -			       0, dev_name(&pdev->dev), aq);
>> +	err = devm_request_irq(dev, irq, atmel_qspi_interrupt, 0,
>> +			       dev_name(dev), aq);
>>  	if (err)
>> -		goto disable_clk;
>> +		goto disable_qspick;
>>  
>>  	err = atmel_qspi_init(aq);
>>  	if (err)
>> -		goto disable_clk;
>> +		goto disable_qspick;
>>  
>>  	err = spi_register_controller(ctrl);
>>  	if (err)
>> -		goto disable_clk;
>> +		goto disable_qspick;
>>  
>>  	return 0;
>>  
>> +disable_qspick:
>> +	clk_disable_unprepare(aq->qspick);
>>  disable_clk:
> 
> We should probably rename this label disable_pclk.

sure. Thanks, Boris!

> 
>>  	clk_disable_unprepare(aq->clk);
>>  exit:
>