Re: [PATCH 2/2] i2c: exynos5: configure fifo_depth based on HSI2C module version

[Tomasz Figa <t.figa@samsung.com>]

On Tuesday 10 of December 2013 10:26:40 Naveen Krishna Ch wrote:
> Hello Tomasz,
> 
> 
> On 9 December 2013 22:01, Tomasz Figa <t.figa@samsung.com> wrote:
> >
> > Hi Naveen,
> >
> > On Friday 22 of November 2013 11:44:11 Naveen Krishna Chatradhi wrote:
> > > fifo_depth of the HSI2C is not constant
> > > Exynos5420 and Exynos5250 supports fifo_depth of 64bytes
> > > Exynos5260 supports fifo_depth of 16bytes
> > >
> > > This patch configures the fifo_depth based on HSI2C modules version.
> > > Signed-off-by: Naveen Krishna Chatradhi <ch.naveen@samsung.com>
> > > ---
> > >  drivers/i2c/busses/i2c-exynos5.c |   29 ++++++++++++++++++-----------
> > >  1 file changed, 18 insertions(+), 11 deletions(-)
> > >
> > > diff --git a/drivers/i2c/busses/i2c-exynos5.c b/drivers/i2c/busses/i2c-exynos5.c
> > > index cbb49e2..19277d8 100644
> > > --- a/drivers/i2c/busses/i2c-exynos5.c
> > > +++ b/drivers/i2c/busses/i2c-exynos5.c
> > > @@ -77,12 +77,6 @@
> > >  #define HSI2C_RXFIFO_TRIGGER_LEVEL(x)                ((x) << 4)
> > >  #define HSI2C_TXFIFO_TRIGGER_LEVEL(x)                ((x) << 16)
> > >
> > > -/* As per user manual FIFO max depth is 64bytes */
> > > -#define HSI2C_FIFO_MAX                               0x40
> > > -/* default trigger levels for Tx and Rx FIFOs */
> > > -#define HSI2C_DEF_TXFIFO_LVL                 (HSI2C_FIFO_MAX - 0x30)
> > > -#define HSI2C_DEF_RXFIFO_LVL                 (HSI2C_FIFO_MAX - 0x10)
> > > -
> > >  /* I2C_TRAILING_CTL Register bits */
> > >  #define HSI2C_TRAILING_COUNT                 (0xf)
> > >
> > > @@ -187,6 +181,9 @@ struct exynos5_i2c {
> > >
> > >       /* Version of HS-I2C Hardware */
> > >       unsigned int            version;
> > > +
> > > +     /* FIFO depth */
> > > +     unsigned int            fifo_depth;
> > >  };
> > >
> > >  enum hsi2c_version {
> > > @@ -437,7 +434,7 @@ static irqreturn_t exynos5_i2c_irq(int irqno, void *dev_id)
> > >               fifo_status = readl(i2c->regs + HSI2C_FIFO_STATUS);
> > >               fifo_level = HSI2C_TX_FIFO_LVL(fifo_status);
> > >
> > > -             len = HSI2C_FIFO_MAX - fifo_level;
> > > +             len = i2c->fifo_depth - fifo_level;
> > >               if (len > (i2c->msg->len - i2c->msg_ptr))
> > >                       len = i2c->msg->len - i2c->msg_ptr;
> > >
> > > @@ -505,6 +502,7 @@ static void exynos5_i2c_message_start(struct exynos5_i2c *i2c, int stop)
> > >       u32 i2c_auto_conf = 0;
> > >       u32 fifo_ctl;
> > >       unsigned long flags;
> > > +     unsigned short trig_lvl;
> > >
> > >       i2c_ctl = readl(i2c->regs + HSI2C_CTL);
> > >       i2c_ctl &= ~(HSI2C_TXCHON | HSI2C_RXCHON);
> > > @@ -515,13 +513,19 @@ static void exynos5_i2c_message_start(struct exynos5_i2c *i2c, int stop)
> > >
> > >               i2c_auto_conf = HSI2C_READ_WRITE;
> > >
> > > -             fifo_ctl |= HSI2C_RXFIFO_TRIGGER_LEVEL(HSI2C_DEF_TXFIFO_LVL);
> > > +             trig_lvl = (i2c->msg->len > i2c->fifo_depth) ?
> > > +                     (i2c->fifo_depth * 3/4) : i2c->msg->len;
> >
> > This patch changes the fifo trigger level calculation (it's dependent now
> > on message length), not just maximum fifo size, as the description says.
> Actually, message->length need not be involved in this calculation.
> Involving msg->len
> will raise another interrupt for every small transactions. It better be avoided.
> > It should be split into two separate patches, explaining why both changes
> > are necessary.
> I will split the fifo_depth configuration code along with comments addressed on
> https://lkml.org/lkml/2013/11/22/31

OK.

> 
> I can think of 3 ways to address the h/w version changes
> 1. Compatible string as i implemented
> 2. Varient struct
> 3. Passing the information via device tree
> How about passing fifo_depth from device tree information.

I believe a combination of 1 and 2 is the recommended option, which is
a variant struct pointed by an entry in OF match table.

> 
> Currently, HSI2C Module on Exynso5260 is not another H/W version.
> It only defer in fifo_depth and init sequence needs a reset.

Well, this implies that it's another H/W version from kernel point of
view, as it needs different handling.

Best regards,
Tomasz