Re: [PATCH v3 3/6] spi: spi-mtk-nor: support 7 bytes transfer of generic spi

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, Sep 25, 2020 at 3:47 PM Chuanhong Guo <gch981213@xxxxxxxxx> wrote:
>
> Hi!
>
> On Fri, Sep 25, 2020 at 2:55 PM Ikjoon Jang <ikjn@xxxxxxxxxxxx> wrote:
> >
> > When mtk-nor fallbacks to generic spi transfers, it can actually
> > transfer up to 7 bytes.
>
> generic transfer_one_message should support full-duplex transfers,
> not transfers with special format requirements. (e.g. here the last
> byte is rx only.) These transfers with format requirements should
> be implemented with spi-mem interface instead.

yep, that's correct.

>
> >
> > This patch fixes adjust_op_size() and supports_op() to explicitly
> > check 7 bytes range and also fixes possible under/overflow conditions
> > in register offsets calculation.
> >
> > Signed-off-by: Ikjoon Jang <ikjn@xxxxxxxxxxxx>
>
> I was notified by Bayi about your discussion and sent some
> patches yesterday for the same purpose. Whoops...
> As transfer_one_message isn't the proper place to implement
> this, maybe we could work on my version instead?
>

I didn't noticed that before,
Sure, please go ahead, I'll follow up with your patch in v4.

> > ---
> >
> > (no changes since v1)
>
> This should be "new patch" not "no changes" :P

oops, it seems my script did something wrong.

>
>
> >
> >  drivers/spi/spi-mtk-nor.c | 102 ++++++++++++++++++++++++++++----------
> >  1 file changed, 76 insertions(+), 26 deletions(-)
> >
> > diff --git a/drivers/spi/spi-mtk-nor.c b/drivers/spi/spi-mtk-nor.c
> > index 0f7d4ec68730..e7719d249095 100644
> > --- a/drivers/spi/spi-mtk-nor.c
> > +++ b/drivers/spi/spi-mtk-nor.c
> > @@ -79,7 +79,11 @@
> >  #define MTK_NOR_REG_DMA_DADR           0x720
> >  #define MTK_NOR_REG_DMA_END_DADR       0x724
> >
> > +/* maximum bytes of TX in PRG mode */
> >  #define MTK_NOR_PRG_MAX_SIZE           6
> > +/* maximum bytes of TX + RX is 7, last 1 byte is always being sent as zero */
> > +#define MTK_NOR_PRG_MAX_CYCLES         7
> > +
> >  // Reading DMA src/dst addresses have to be 16-byte aligned
> >  #define MTK_NOR_DMA_ALIGN              16
> >  #define MTK_NOR_DMA_ALIGN_MASK         (MTK_NOR_DMA_ALIGN - 1)
> > @@ -167,6 +171,24 @@ static bool mtk_nor_match_read(const struct spi_mem_op *op)
> >         return false;
> >  }
> >
> > +static bool mtk_nor_check_prg(const struct spi_mem_op *op)
> > +{
> > +       size_t len = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes;
> > +
> > +       if (len > MTK_NOR_PRG_MAX_SIZE)
> > +               return false;
> > +
> > +       if (!op->data.nbytes)
> > +               return true;
> > +
> > +       if (op->data.dir == SPI_MEM_DATA_OUT)
> > +               return ((len + op->data.nbytes) <= MTK_NOR_PRG_MAX_SIZE);
> > +       else if (op->data.dir == SPI_MEM_DATA_IN)
> > +               return ((len + op->data.nbytes) <= MTK_NOR_PRG_MAX_CYCLES);
> > +       else
> > +               return true;
> > +}
> > +
> >  static int mtk_nor_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
> >  {
> >         size_t len;
> > @@ -195,10 +217,22 @@ static int mtk_nor_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
> >                 }
> >         }
> >
> > -       len = MTK_NOR_PRG_MAX_SIZE - op->cmd.nbytes - op->addr.nbytes -
> > -             op->dummy.nbytes;
> > -       if (op->data.nbytes > len)
> > -               op->data.nbytes = len;
> > +       if (mtk_nor_check_prg(op))
> > +               return 0;
> > +
> > +       len = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes;
> > +
> > +       if (op->data.dir == SPI_MEM_DATA_OUT) {
> > +               if (len == MTK_NOR_PRG_MAX_SIZE)
> > +                       return -EINVAL;
> > +               op->data.nbytes = min_t(unsigned int, op->data.nbytes,
> > +                                       MTK_NOR_PRG_MAX_SIZE - len);
> > +       } else  {
> > +               if (len == MTK_NOR_PRG_MAX_CYCLES)
> > +                       return -EINVAL;
> > +               op->data.nbytes = min_t(unsigned int, op->data.nbytes,
> > +                                       MTK_NOR_PRG_MAX_CYCLES - len);
> > +       }
> >
> >         return 0;
> >  }
> > @@ -206,8 +240,6 @@ static int mtk_nor_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
> >  static bool mtk_nor_supports_op(struct spi_mem *mem,
> >                                 const struct spi_mem_op *op)
> >  {
> > -       size_t len;
> > -
> >         if (op->cmd.buswidth != 1)
> >                 return false;
> >
> > @@ -223,12 +255,11 @@ static bool mtk_nor_supports_op(struct spi_mem *mem,
> >                                (op->data.buswidth == 1);
> >         }
> >
> > -       len = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes;
> > -       if ((len > MTK_NOR_PRG_MAX_SIZE) ||
> > -           ((op->data.nbytes) && (len == MTK_NOR_PRG_MAX_SIZE)))
> > +       /* fallback to generic spi xfer */
> > +       if (op->cmd.buswidth > 1 || op->addr.buswidth > 1 || op->data.buswidth > 1)
> >                 return false;
>
> Rejecting an op in supports_op doesn't tell it to fall back to generic
> spi transfer.
> It instead tells caller to abort this transfer completely.
> A fallback only happens when exec_op returns -ENOTSUPP.

yep but I think that case always going PRG mode in exec_op() with the
same condition?

> This comment is incorrect. I'd put this buswidth checking in mtk_nor_check_prg
> instead because mtk_nor_check_prg is checking whether an op is supported
> by prg mode, thus it should reject ops with buswidth > 1.
>
> >
> > -       return true;
> > +       return mtk_nor_check_prg(op);
> >  }
> >
> >  static void mtk_nor_setup_bus(struct mtk_nor *sp, const struct spi_mem_op *op)
> > @@ -459,22 +490,36 @@ static int mtk_nor_transfer_one_message(struct spi_controller *master,
> >         int stat = 0;
> >         int reg_offset = MTK_NOR_REG_PRGDATA_MAX;
> >         void __iomem *reg;
> > -       const u8 *txbuf;
> > -       u8 *rxbuf;
> > -       int i;
> > +       int i, tx_len = 0, rx_len = 0;
> >
> >         list_for_each_entry(t, &m->transfers, transfer_list) {
> > -               txbuf = t->tx_buf;
> > -               for (i = 0; i < t->len; i++, reg_offset--) {
> > +               const u8 *txbuf = t->tx_buf;
> > +
> > +               if (!txbuf) {
> > +                       rx_len += t->len;
> > +                       continue;
> > +               }
> > +
> > +               if (rx_len) {
> > +                       stat = -EPROTO;
> > +                       goto msg_done;
> > +               }
>
> NACK. you are unnecessarily rejecting possible transfers.

yep, ditto

>
> > +
> > +               for (i = 0; i < t->len && reg_offset >= 0; i++, reg_offset--) {
> >                         reg = sp->base + MTK_NOR_REG_PRGDATA(reg_offset);
> > -                       if (txbuf)
> > -                               writeb(txbuf[i], reg);
> > -                       else
> > -                               writeb(0, reg);
> > +                       writeb(txbuf[i], reg);
> > +                       tx_len++;
>
> According to SPI standard, during a rx transfer, tx should be kept low.
> These PROGDATA registers doesn't clear itself so it'll keep sending
> data from last transfer, which violates this rule. That's
> why the original code writes 0 to PRGDATA for rx bytes.

following lines with while() will set 0s to the rest of registers.

>
> >                 }
> > -               trx_len += t->len;
> >         }
> >
> > +       while (reg_offset >= 0) {
> > +               writeb(0, sp->base + MTK_NOR_REG_PRGDATA(reg_offset));
> > +               reg_offset--;
> > +       }
> > +
> > +       rx_len = min_t(unsigned long, MTK_NOR_PRG_MAX_CYCLES - tx_len, rx_len);
> > +       trx_len = tx_len + rx_len;
> > +
> >         writel(trx_len * BITS_PER_BYTE, sp->base + MTK_NOR_REG_PRG_CNT);
> >
> >         stat = mtk_nor_cmd_exec(sp, MTK_NOR_CMD_PROGRAM,
> > @@ -482,13 +527,18 @@ static int mtk_nor_transfer_one_message(struct spi_controller *master,
> >         if (stat < 0)
> >                 goto msg_done;
> >
> > -       reg_offset = trx_len - 1;
> > -       list_for_each_entry(t, &m->transfers, transfer_list) {
> > -               rxbuf = t->rx_buf;
> > -               for (i = 0; i < t->len; i++, reg_offset--) {
> > -                       reg = sp->base + MTK_NOR_REG_SHIFT(reg_offset);
> > -                       if (rxbuf)
> > +       if (rx_len > 0) {
> > +               reg_offset = rx_len - 1;
> > +               list_for_each_entry(t, &m->transfers, transfer_list) {
> > +                       u8 *rxbuf = t->rx_buf;
> > +
> > +                       if (!rxbuf)
> > +                               continue;
> > +
> > +                       for (i = 0; i < t->len && reg_offset >= 0; i++, reg_offset--) {
> > +                               reg = sp->base + MTK_NOR_REG_SHIFT(reg_offset);
> >                                 rxbuf[i] = readb(reg);
> > +                       }
>
> I think this is replacing original code with some equivalent ones, which
> seems unnecessary.

This patch addressed the issue with 1+6 bytes transfer (e.g JEDEC ID)
can have negative reg_offset.
And there's skipping the loop if (rx_len < 0)
anyway I'd like to follow with your new patch. :-)

Thanks!

>
> >                 }
> >         }
> >
> --
> Regards,
> Chuanhong Guo



[Index of Archives]     [Linux Kernel]     [Linux ARM (vger)]     [Linux ARM MSM]     [Linux Omap]     [Linux Arm]     [Linux Tegra]     [Fedora ARM]     [Linux for Samsung SOC]     [eCos]     [Linux Fastboot]     [Gcc Help]     [Git]     [DCCP]     [IETF Announce]     [Security]     [Linux MIPS]     [Yosemite Campsites]

  Powered by Linux