[U-Boot] [UBOOT PATCH v3 3/3] spi: xilinx_spi: Added support to read JEDEC-id twice at the boot time

[Michal Simek]

On 27.6.2018 09:35, Jagan Teki wrote:
> On Wed, Jun 27, 2018 at 12:59 PM, Michal Simek <michal.simek at xilinx.com> wrote:
>> Hi Jagan,
>>
>> On 27.6.2018 08:43, Jagan Teki wrote:
>>> On Mon, Jun 25, 2018 at 4:06 PM, Vipul Kumar <vipulk at xilinx.com> wrote:
>>>> Hi Jagan,
>>>>
>>>>> -----Original Message-----
>>>>> From: Jagan Teki [mailto:jagannadh.teki at gmail.com]
>>>>> Sent: Monday, June 25, 2018 3:47 PM
>>>>> To: Vipul Kumar <vipulk at xilinx.com>
>>>>> Cc: U-Boot Mailing List <u-boot at lists.denx.de>; Michal Simek
>>>>> <michals at xilinx.com>
>>>>> Subject: Re: [U-Boot] [UBOOT PATCH v3 3/3] spi: xilinx_spi: Added support to
>>>>> read JEDEC-id twice at the boot time
>>>>>
>>>>> On Thu, Jun 21, 2018 at 2:53 PM, Vipul Kumar <vipul.kumar at xilinx.com>
>>>>> wrote:
>>>>>> This patch is for the startup block issue in the spi controller.
>>>>>> SPI clock is passing through STARTUP block to FLASH. STARTUP block
>>>>>> don't provide clock as soon as QSPI provides command. So, first
>>>>>> command fails.
>>>>>
>>>>> Does this a controller issue?
>>>>
>>>> Yes, this is a controller issue.
>>>>
>>>>>
>>>>>>
>>>>>> This patch added support to read JEDEC id in xilinx_spi_xfer ().
>>>>>>
>>>>>> Signed-off-by: Vipul Kumar <vipul.kumar at xilinx.com>
>>>>>> ---
>>>>>> Changes in v3:
>>>>>> - No change
>>>>>> ---
>>>>>>  drivers/spi/xilinx_spi.c | 41
>>>>>> +++++++++++++++++++++++++++++++++++++++++
>>>>>>  1 file changed, 41 insertions(+)
>>>>>>
>>>>>> diff --git a/drivers/spi/xilinx_spi.c b/drivers/spi/xilinx_spi.c index
>>>>>> 4026540..61301c2 100644
>>>>>> --- a/drivers/spi/xilinx_spi.c
>>>>>> +++ b/drivers/spi/xilinx_spi.c
>>>>>> @@ -204,6 +204,40 @@ static u32 xilinx_spi_read_rxfifo(struct udevice
>>>>> *bus, u8 *rxp, u32 rxbytes)
>>>>>>         return i;
>>>>>>  }
>>>>>>
>>>>>> +static void xilinx_startup_block(struct udevice *dev, unsigned int bytes,
>>>>>> +                                const void *dout, void *din) {
>>>>>> +       struct udevice *bus = dev_get_parent(dev);
>>>>>> +       struct xilinx_spi_priv *priv = dev_get_priv(bus);
>>>>>> +       struct xilinx_spi_regs *regs = priv->regs;
>>>>>> +       struct dm_spi_slave_platdata *slave_plat =
>>>>> dev_get_parent_platdata(dev);
>>>>>> +       const unsigned char *txp = dout;
>>>>>> +       unsigned char *rxp = din;
>>>>>> +       static int startup;
>>>>>> +       u32 reg, count;
>>>>>> +       u32 txbytes = bytes;
>>>>>> +       u32 rxbytes = bytes;
>>>>>> +
>>>>>> +       /*
>>>>>> +        * This loop runs two times. First time to send the command.
>>>>>> +        * Second time to transfer data. After transferring data,
>>>>>> +        * it sets txp to the initial value for the normal operation.
>>>>>> +        */
>>>>>> +       for ( ; startup < 2; startup++) {
>>>>>> +               count = xilinx_spi_fill_txfifo(bus, txp, txbytes);
>>>>>> +               reg = readl(&regs->spicr) & ~SPICR_MASTER_INHIBIT;
>>>>>> +               writel(reg, &regs->spicr);
>>>>>> +               count = xilinx_spi_read_rxfifo(bus, rxp, rxbytes);
>>>>>> +               txp = din;
>>>>>> +
>>>>>> +               if (startup) {
>>>>>> +                       spi_cs_deactivate(dev);
>>>>>> +                       spi_cs_activate(dev, slave_plat->cs);
>>>>>> +                       txp = dout;
>>>>>> +               }
>>>>>> +       }
>>>>>> +}
>>>>>> +
>>>>>>  static int xilinx_spi_xfer(struct udevice *dev, unsigned int bitlen,
>>>>>>                             const void *dout, void *din, unsigned long
>>>>>> flags)  { @@ -236,6 +270,13 @@ static int xilinx_spi_xfer(struct
>>>>>> udevice *dev, unsigned int bitlen,
>>>>>>         if (flags & SPI_XFER_BEGIN)
>>>>>>                 spi_cs_activate(dev, slave_plat->cs);
>>>>>>
>>>>>> +       /*
>>>>>> +        * This is the work around for the startup block issue in
>>>>>> +        * the spi controller. SPI clock is passing through STARTUP
>>>>>> +        * block to FLASH. STARTUP block don't provide clock as soon
>>>>>> +        * as QSPI provides command. So first command fails.
>>>>>> +        */
>>>>>> +       xilinx_startup_block(dev, bytes, dout, din);
>>>>>
>>>>> But not every spi_xfer from flash side is to read JEDEC ID? during probe of sf,
>>>>> the relevant spi_xfer is to read JEDEC ID
>>>>
>>>> Inside xilinx_startup_block() function, we are using static variable, so that code will execute only first time.
>>>
>>> Sorry, IMHO mainline won't accept this kind of hack (particularly
>>> avoiding function execution through static variables) by for the sake
>>> of bug fix. Having a proper condition to check and fix the errata look
>>> fine to me, any idea how Linux handling this?
>>
>> ok - static startup variable should be move to private data structure to
>> be private for every instance.
>>
>> Also please put a link to errata for a record.
>>
>> Jagan: It will be much easier if you can simply write snippet which you
>> are suggesting.
> 
> thought of, I'm still thinking how we can do this. I didn't see the
> Linux code yet, any idea how Linux does?

The issue is present in Linux but it is not fixed there yet.
It means Linux can't be taken as pattern.
It is quite clear that this should be handled in the driver code and it
should be enabled by default for all IPs till this is fixed in hw IP.

Thanks,
Michal