Date:   Wed, 11 May 2022 14:46:27 +0300
From:   Ivan Bornyakov <i.bornyakov@...rotek.ru>
To:     Conor.Dooley@...rochip.com
Cc:     mdf@...nel.org, hao.wu@...el.com, yilun.xu@...el.com,
        trix@...hat.com, robh+dt@...nel.org,
        krzysztof.kozlowski+dt@...aro.org, linux-fpga@...r.kernel.org,
        devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
        system@...rotek.ru
Subject: Re: [PATCH v11 2/3] fpga: microchip-spi: add Microchip MPF FPGA
 manager

On Wed, May 11, 2022 at 11:36:33AM +0000, Conor.Dooley@...rochip.com wrote:
> On 11/05/2022 09:15, Ivan Bornyakov wrote:
> > EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
> > 
> > On Tue, May 10, 2022 at 12:29:54PM +0100, Conor Dooley wrote:
> >> On 09/05/2022 19:56, Conor Dooley wrote:
> >>> On 09/05/2022 18:16, Ivan Bornyakov wrote:
> >>>> On Mon, May 09, 2022 at 11:41:18AM +0000, Conor.Dooley@...rochip.com wrote:
> >>>>> Hey Ivan, one comment below.
> >>>>> Thanks,
> >>>>> Conor.
> >>>>>
> >>>>> On 07/05/2022 08:43, Ivan Bornyakov wrote:
> >>>>>> ... snip ...
> >>>>>> +static int mpf_read_status(struct spi_device *spi)
> >>>>>> +{
> >>>>>> +       u8 status, status_command = MPF_SPI_READ_STATUS;
> >>>>>> +       struct spi_transfer xfer = {
> >>>>>> +               .tx_buf = &status_command,
> >>>>>> +               .rx_buf = &status,
> >>>>>> +               .len = 1,
> >>>>>> +       };
> >>>>>> +       int ret = spi_sync_transfer(spi, &xfer, 1);
> >>>>>> +
> >>>>>> +       if ((status & MPF_STATUS_SPI_VIOLATION) ||
> >>>>>> +           (status & MPF_STATUS_SPI_ERROR))
> >>>>>> +               ret = -EIO;
> >>>>>> +
> >>>>>> +       return ret ? : status;
> >>>>>> +}
> >>>>>> +
> >>>>>> ... snip ...
> >>>>>> +
> >>>>>> +static int poll_status_not_busy(struct spi_device *spi, u8 mask)
> >>>>>> +{
> >>>>>> +       int status, timeout = MPF_STATUS_POLL_TIMEOUT;
> >>>>>> +
> >>>>>> +       while (timeout--) {
> >>>>>> +               status = mpf_read_status(spi);
> >>>>>> +               if (status < 0 ||
> >>>>>> +                   (!(status & MPF_STATUS_BUSY) && (!mask || (status & mask))))
> >>>>>> +                       return status;
> >>>>>> +
> >>>>>> +               usleep_range(1000, 2000);
> >>>>>> +       }
> >>>>>> +
> >>>>>> +       return -EBUSY;
> >>>>>> +}
> >>>>>
> >>>>> Is there a reason you changed this from the snippet you sent me
> >>>>> in the responses to version 8:
> >>>>> static int poll_status_not_busy(struct spi_device *spi, u8 mask)
> >>>>> {
> >>>>>          u8 status, status_command = MPF_SPI_READ_STATUS;
> >>>>>          int ret, timeout = MPF_STATUS_POLL_TIMEOUT;
> >>>>>          struct spi_transfer xfer = {
> >>>>>                  .tx_buf = &status_command,
> >>>>>                  .rx_buf = &status,
> >>>>>                  .len = 1,
> >>>>>          };
> >>>>>
> >>>>>          while (timeout--) {
> >>>>>                  ret = spi_sync_transfer(spi, &xfer, 1);
> >>>>>                  if (ret < 0)
> >>>>>                          return ret;
> >>>>>
> >>>>>                  if (!(status & MPF_STATUS_BUSY) && (!mask || (status & mask)))
> >>>>>                          return status;
> >>>>>
> >>>>>                  usleep_range(1000, 2000);
> >>>>>          }
> >>>>>
> >>>>>          return -EBUSY;
> >>>>> }
> >>>>>
> >>>>> With the current version, I hit the "Failed to write bitstream
> >>>>> frame" check in mpf_ops_write at random points in the transfer.
> >>>>> Replacing poll_status_not_busy with the above allows it to run
> >>>>> to completion.
> >>>>
> >>>> In my eyes they are equivalent, aren't they?
> >>>>
> >>>
> >>> I was in a bit of a rush today & didn't have time to do proper
> >>> debugging, I'll put some debug code in tomorrow and try to find
> >>> exactly what is different between the two.
> >>>
> >>> Off the top of my head, since I don't have a board on me to test,
> >>> the only difference I can see is that with the snippet you only
> >>> checked if spi_sync_transfer was negative whereas now you check
> >>> if it has a value at all w/ that ternary operator.
> >>>
> >>> But even that seems like it *shouldn't* be the problem, since ret
> >>> should contain -errno or zero, right?
> >>> Either way, I will do some digging tomorrow.
> >>
> >> I put a printk("status %x, ret %d", status, ret); into the failure
> >> path of mpf_read_status() & it looks like a status 0xA is being
> >> returned - error & ready? That seems like a very odd combo to be
> >> getting back out of it. It shouldn't be dodgy driver/connection
> >> either, b/c that's what I see if I connect my protocol analyser:
> >> https://i.imgur.com/VbjgfCk.png
> >>
> >> That's mosi (hex), ss, sclk, mosi, miso (hex), miso in descending
> >> order.
> >>
> >> I think what was happening was with the snippet you returned one
> >> of the following: -EBUSY, ret (aka -errno) or status. Since status
> >> is positive, the checks in mpf_spi_write.*() saw nothing wrong at
> >> all and programming continued despite there being a problem.
> >>
> >> The new version fixes this by returning -EIO rather than status from
> >> poll_status_not_busy().
> >>
> >> I wish I had a socketable PolarFire so I could investigate further,
> >> but this looks like it might a be hardware issue somewhere on my
> >> end?
> >>
> >> So ye, sorry for the noise and carry on! I'll try tofind what is to
> >> blame for it.
> >>
> >> Thanks,
> >> Conor.
> >>
> > 
> > Hi, Conor.
> > 
> > I've just noticed in SPI-DirectC User Guide [1] ch. 9 SmartFusion2 and
> > IGLOO2 SPI-Slave Programming Waveform Analysis, that hw status checked
> > two times every time. Does MPF family also need double check hw status?
> > Does adding second mpf_read_status() to poll_status_not_busy() routine
> > help with your issue?
> 
> Hey Ivan,
> Tried your suggestion. Previously I was failing quite consistently at
> transfer 34 of 590k, and sometimes making it a further. With your
> suggestion, I was making it significantly further (100k+) but still
> running into some of the 0xA status.
> Decided to move the double check into mpfs_read_status (see the below
> diff) did not run into any the 0xA statuses.
> It's worth pointing out that this is the *first* time I have seen
> Flash Pro Express report that the FPGA array has been enabled after
> programming!

That's good news!

> Seems like at the very least this (hacky) diff is not harmful?
> Please give it a try yourself and check that things still work for
> you.
> 
> diff --git a/drivers/fpga/microchip-spi.c b/drivers/fpga/microchip-spi.c
> index 63b75dff2522..183cdfc05c4a 100644
> --- a/drivers/fpga/microchip-spi.c
> +++ b/drivers/fpga/microchip-spi.c
> @@ -47,18 +47,30 @@ struct mpf_priv {
>   static int mpf_read_status(struct spi_device *spi)
>   {
>          u8 status, status_command = MPF_SPI_READ_STATUS;
> +       u8 status_repeat;
>          struct spi_transfer xfer = {
>                  .tx_buf = &status_command,
>                  .rx_buf = &status,
>                  .len = 1,
>          };
> +       struct spi_transfer xfer_repeat = {
> +               .tx_buf = &status_command,
> +               .rx_buf = &status_repeat,
> +               .len = 1,
> +       };
>          int ret = spi_sync_transfer(spi, &xfer, 1);
> +       int ret_repeat = spi_sync_transfer(spi, &xfer_repeat, 1);
> +
> +       if (ret || ret_repeat)
> +               return -EIO;
>   
> -       if ((status & MPF_STATUS_SPI_VIOLATION) ||
> -           (status & MPF_STATUS_SPI_ERROR))
> +       if (status != status_repeat)
> +               printk("status disagreement %x %x", status, status_repeat);
> +       if ((status_repeat & MPF_STATUS_SPI_VIOLATION) ||
> +           (status_repeat & MPF_STATUS_SPI_ERROR))
>                  ret = -EIO;
>   
> -       return ret ? : status;
> +       return ret ?: status_repeat;
>   }
>   
>   static enum fpga_mgr_states mpf_ops_state(struct fpga_manager *mgr)

I'll check that and send out v12 if it's all rigth in the near future.

Powered by blists - more mailing lists