Re: [Qemu-devel] [PATCH v9 08/13] xilinx_spips: Make tx/rx_data_bytes more generic and reusable

[francisco iglesias <frasse.iglesias@gmail.com>]

On 27 November 2017 at 04:52, Philippe Mathieu-Daudé <f4bug@amsat.org>
wrote:

> Hi Francisco,
>
> On 11/26/2017 08:16 PM, Francisco Iglesias wrote:
> > Make tx/rx_data_bytes more generic so they can be reused (when adding
> > support for the Zynqmp Generic QSPI).
> >
> > Signed-off-by: Francisco Iglesias <frasse.iglesias@gmail.com>
> > Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
> > Tested-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
> > ---
> >  hw/ssi/xilinx_spips.c | 64 +++++++++++++++++++++++++++++-
> ---------------------
> >  1 file changed, 37 insertions(+), 27 deletions(-)
> >
> > diff --git a/hw/ssi/xilinx_spips.c b/hw/ssi/xilinx_spips.c
> > index 691d48d..4621dbb 100644
> > --- a/hw/ssi/xilinx_spips.c
> > +++ b/hw/ssi/xilinx_spips.c
> > @@ -47,7 +47,7 @@
> >  /* config register */
> >  #define R_CONFIG            (0x00 / 4)
> >  #define IFMODE              (1U << 31)
> > -#define ENDIAN              (1 << 26)
> > +#define R_CONFIG_ENDIAN     (1 << 26)
> >  #define MODEFAIL_GEN_EN     (1 << 17)
> >  #define MAN_START_COM       (1 << 16)
> >  #define MAN_START_EN        (1 << 15)
> > @@ -450,13 +450,28 @@ static void xilinx_spips_flush_txfifo(XilinxSPIPS
> *s)
> >      }
> >  }
> >
> > -static inline void rx_data_bytes(XilinxSPIPS *s, uint8_t *value, int
> max)
> > +static inline void tx_data_bytes(Fifo8 *fifo, uint32_t value, int num,
> bool be)
> >  {
> >      int i;
> > +    for (i = 0; i < num && !fifo8_is_full(fifo); ++i) {
> > +        if (be) {
> > +            fifo8_push(fifo, (uint8_t)(value >> 24));
> > +            value <<= 8;
> > +        } else {
> > +            fifo8_push(fifo, (uint8_t)value);
> > +            value >>= 8;
> > +        }
> > +    }
> > +}
> >
> > -    for (i = 0; i < max && !fifo8_is_empty(&s->rx_fifo); ++i) {
> > -        value[i] = fifo8_pop(&s->rx_fifo);
> > +static inline int rx_data_bytes(Fifo8 *fifo, uint8_t *value, int max)
> > +{
> > +    int i;
> > +
> > +    for (i = 0; i < max && !fifo8_is_empty(fifo); ++i) {
> > +        value[i] = fifo8_pop(fifo);
> >      }
> > +    return max - i;
> >  }
> >
> >  static uint64_t xilinx_spips_read(void *opaque, hwaddr addr,
> > @@ -466,6 +481,7 @@ static uint64_t xilinx_spips_read(void *opaque,
> hwaddr addr,
> >      uint32_t mask = ~0;
> >      uint32_t ret;
> >      uint8_t rx_buf[4];
> > +    int shortfall;
> >
> >      addr >>= 2;
> >      switch (addr) {
> > @@ -496,9 +512,13 @@ static uint64_t xilinx_spips_read(void *opaque,
> hwaddr addr,
> >          break;
> >      case R_RX_DATA:
> >          memset(rx_buf, 0, sizeof(rx_buf));
> > -        rx_data_bytes(s, rx_buf, s->num_txrx_bytes);
> > -        ret = s->regs[R_CONFIG] & ENDIAN ? cpu_to_be32(*(uint32_t
> *)rx_buf)
> > -                        : cpu_to_le32(*(uint32_t *)rx_buf);
> > +        shortfall = rx_data_bytes(&s->rx_fifo, rx_buf,
> s->num_txrx_bytes);
>
> About this part,
>
> > +        ret = s->regs[R_CONFIG] & R_CONFIG_ENDIAN ?
> > +                        cpu_to_be32(*(uint32_t *)rx_buf) :
> > +                        cpu_to_le32(*(uint32_t *)rx_buf);
> > +        if (!(s->regs[R_CONFIG] & R_CONFIG_ENDIAN)) {
> > +            ret <<= 8 * shortfall;
> > +        }
>
> The following looks easier to read to me, probably matter of taste, but
> you don't have to wonder the cpu <-> device direction and it remove the
> casts:
>
>            if (s->regs[R_CONFIG] & R_CONFIG_ENDIAN) {
>                ret = ldl_be_p(rx_buf);
>            } else {
>                ret = ldl_le_p(rx_buf) << (8 * shortfall);
>            }
>
> What do you think? (before respining)
>
>
Dear Philippe,

First of all, thank you very much for reviewing again! After reading the
code above I agree that it improves readability in this patch but after
thinking a little about it I'm not sure if the change is maybe better of in
separate patch done later on, this basically since cpu_to_le is also used
in an other location in the file and that patch could then update that one
aswell for keeping the file consistent. What do you think, would you prefer
that I do the change and respin or should we wait and do it in a separate
patch later on? (I'm leaning on another patch myself for not breaking the
consistency in the file)

Best regards,
Francisco Iglesias



> >          DB_PRINT_L(0, "addr=" TARGET_FMT_plx " = %x\n", addr * 4, ret);
> >          xilinx_spips_update_ixr(s);
> >          return ret;
> > @@ -509,20 +529,6 @@ static uint64_t xilinx_spips_read(void *opaque,
> hwaddr addr,
> >
> >  }
> >
> > -static inline void tx_data_bytes(XilinxSPIPS *s, uint32_t value, int
> num)
> > -{
> > -    int i;
> > -    for (i = 0; i < num && !fifo8_is_full(&s->tx_fifo); ++i) {
> > -        if (s->regs[R_CONFIG] & ENDIAN) {
> > -            fifo8_push(&s->tx_fifo, (uint8_t)(value >> 24));
> > -            value <<= 8;
> > -        } else {
> > -            fifo8_push(&s->tx_fifo, (uint8_t)value);
> > -            value >>= 8;
> > -        }
> > -    }
> > -}
> > -
> >  static void xilinx_spips_write(void *opaque, hwaddr addr,
> >                                          uint64_t value, unsigned size)
> >  {
> > @@ -563,16 +569,20 @@ static void xilinx_spips_write(void *opaque,
> hwaddr addr,
> >          mask = 0;
> >          break;
> >      case R_TX_DATA:
> > -        tx_data_bytes(s, (uint32_t)value, s->num_txrx_bytes);
> > +        tx_data_bytes(&s->tx_fifo, (uint32_t)value, s->num_txrx_bytes,
> > +                      s->regs[R_CONFIG] & R_CONFIG_ENDIAN);
> >          goto no_reg_update;
> >      case R_TXD1:
> > -        tx_data_bytes(s, (uint32_t)value, 1);
> > +        tx_data_bytes(&s->tx_fifo, (uint32_t)value, 1,
> > +                      s->regs[R_CONFIG] & R_CONFIG_ENDIAN);
> >          goto no_reg_update;
> >      case R_TXD2:
> > -        tx_data_bytes(s, (uint32_t)value, 2);
> > +        tx_data_bytes(&s->tx_fifo, (uint32_t)value, 2,
> > +                      s->regs[R_CONFIG] & R_CONFIG_ENDIAN);
> >          goto no_reg_update;
> >      case R_TXD3:
> > -        tx_data_bytes(s, (uint32_t)value, 3);
> > +        tx_data_bytes(&s->tx_fifo, (uint32_t)value, 3,
> > +                      s->regs[R_CONFIG] & R_CONFIG_ENDIAN);
> >          goto no_reg_update;
> >      }
> >      s->regs[addr] = (s->regs[addr] & ~mask) | (value & mask);
> > @@ -682,11 +692,11 @@ static void lqspi_load_cache(void *opaque, hwaddr
> addr)
> >
> >          while (cache_entry < LQSPI_CACHE_SIZE) {
> >              for (i = 0; i < 64; ++i) {
> > -                tx_data_bytes(s, 0, 1);
> > +                tx_data_bytes(&s->tx_fifo, 0, 1, false);
> >              }
> >              xilinx_spips_flush_txfifo(s);
> >              for (i = 0; i < 64; ++i) {
> > -                rx_data_bytes(s, &q->lqspi_buf[cache_entry++], 1);
> > +                rx_data_bytes(&s->rx_fifo,
> &q->lqspi_buf[cache_entry++], 1);
> >              }
> >          }
> >
> >
>