Message-ID: <1595840642.12203.4.camel@mtksdaap41>
Date:   Mon, 27 Jul 2020 17:04:02 +0800
From:   Weiyi Lu <weiyi.lu@...iatek.com>
To:     Nicolas Boichat <drinkcat@...omium.org>
CC:     Matthias Brugger <matthias.bgg@...il.com>,
        Rob Herring <robh@...nel.org>, Stephen Boyd <sboyd@...nel.org>,
        James Liao <jamesjj.liao@...iatek.com>,
        linux-arm Mailing List <linux-arm-kernel@...ts.infradead.org>,
        lkml <linux-kernel@...r.kernel.org>,
        "moderated list:ARM/Mediatek SoC support" 
        <linux-mediatek@...ts.infradead.org>, <linux-clk@...r.kernel.org>,
        srv_heupstream <srv_heupstream@...iatek.com>,
        Wendell Lin <wendell.lin@...iatek.com>,
        Ikjoon Jang <ikjn@...omium.org>
Subject: Re: [PATCH 3/4] clk: mediatek: Add configurable enable control to
 mtk_pll_data

On Thu, 2020-07-23 at 15:51 +0800, Nicolas Boichat wrote:
> On Thu, Jul 23, 2020 at 10:57 AM Weiyi Lu <weiyi.lu@...iatek.com> wrote:
> >
> > On Wed, 2020-07-22 at 16:51 +0800, Nicolas Boichat wrote:
> > > On Wed, Jul 22, 2020 at 2:50 PM Weiyi Lu <weiyi.lu@...iatek.com> wrote:
> > > >
> > > > In all MediaTek PLL design, bit 0 of CON0 register is always
> > > > the enable bit.
> > > > However, there's a special case of usbpll on MT8192.
> > > > The enable bit of usbpll is moved to bit 2 of other register.
> > > > Add configurable en_reg and base_en_bit for enable control or
> > > > using the default if without setting in pll data.
> > > >
> > > > Signed-off-by: Weiyi Lu <weiyi.lu@...iatek.com>
> > > > ---
> > > >  drivers/clk/mediatek/clk-mtk.h |  2 ++
> > > >  drivers/clk/mediatek/clk-pll.c | 26 ++++++++++++++++++++++----
> > > >  2 files changed, 24 insertions(+), 4 deletions(-)
> > > >
> > > > diff --git a/drivers/clk/mediatek/clk-mtk.h b/drivers/clk/mediatek/clk-mtk.h
> > > > index c3d6756..8bb0b3d 100644
> > > > --- a/drivers/clk/mediatek/clk-mtk.h
> > > > +++ b/drivers/clk/mediatek/clk-mtk.h
> > > > @@ -233,6 +233,8 @@ struct mtk_pll_data {
> > > >         uint32_t pcw_chg_reg;
> > > >         const struct mtk_pll_div_table *div_table;
> > > >         const char *parent_name;
> > > > +       uint32_t en_reg;
> > > > +       uint8_t base_en_bit;
> > > >  };
> > > >
> > > >  void mtk_clk_register_plls(struct device_node *node,
> > > > diff --git a/drivers/clk/mediatek/clk-pll.c b/drivers/clk/mediatek/clk-pll.c
> > > > index f440f2cd..b8ccd42 100644
> > > > --- a/drivers/clk/mediatek/clk-pll.c
> > > > +++ b/drivers/clk/mediatek/clk-pll.c
> > > > @@ -44,6 +44,7 @@ struct mtk_clk_pll {
> > > >         void __iomem    *tuner_en_addr;
> > > >         void __iomem    *pcw_addr;
> > > >         void __iomem    *pcw_chg_addr;
> > > > +       void __iomem    *en_addr;
> > > >         const struct mtk_pll_data *data;
> > > >  };
> > > >
> > > > @@ -56,7 +57,10 @@ static int mtk_pll_is_prepared(struct clk_hw *hw)
> > > >  {
> > > >         struct mtk_clk_pll *pll = to_mtk_clk_pll(hw);
> > > >
> > > > -       return (readl(pll->base_addr + REG_CON0) & CON0_BASE_EN) != 0;
> > > > +       if (pll->en_addr)
> > > > +               return (readl(pll->en_addr) & BIT(pll->data->base_en_bit)) != 0;
> > > > +       else
> > > > +               return (readl(pll->base_addr + REG_CON0) & CON0_BASE_EN) != 0;
> > > >  }
> > > >
> > > >  static unsigned long __mtk_pll_recalc_rate(struct mtk_clk_pll *pll, u32 fin,
> > > > @@ -251,6 +255,12 @@ static int mtk_pll_prepare(struct clk_hw *hw)
> > > >         r |= pll->data->en_mask;
> > > >         writel(r, pll->base_addr + REG_CON0);
> > > >
> > >
> > > This is not a new change, but I'm wondering if the asymmetry is
> > > intentional here, that is, prepare sets bit pll->data->en_mask of
> > > REG_CON0; unprepare clears CON0_BASE_EN of REG_CON0.
> > >
> > > With this patch, if pll->en_addr is set, you set both
> > > pll->data->en_mask _and_ pll->data->base_en_bit, and clear only
> > > pll->data->base_en_bit.
> > >
> >
> > Hi Nicolas,
> >
> > AFAIK, the asymmetry was intentional.
> > en_mask is actually a combination of divider enable mask and the pll
> > enable bit(CON0_BASE_EN).
> > Even without my patch, it still sets divider enable mask and en_bit, and
> > only clears en_bit.
> > You could see the pll_data in clk-mt8192.c of patch [4/4]
> > Take mainpll as an example,
> > the enable mask of mainpll is 0xff000001, where 0xff000000 is the
> > divider enable mask and 0x1 is the en_bit
> >
> > For usbpll in special case, usbpll doesn't have divider enable mask on
> > MT8192 so I give nothing(0x00000000) in the en_mask field.
> > However, the main reason why I don't skip setting the en_mask of MT8192
> > usbpll is that I'd just like to reserve the divider enable mask for any
> > special plls with divider enable mask in near future.
> 
> Argh, I see, it's a bit of a can of worms, with many special cases...
> 
> So I played a bit with 3 examples.
> 
> Current situation looks like this:
> 
> 8183 CLK_APMIXED_ARMPLL_LL
>   en_mask = 0x00000001
>   en_reg = 0
>   base_en_bit = 0
> 
> prepare: REG_CON0 |= en_mask
> unprepare: REG_CON0 &= ~CON0_BASE_EN (BIT(1))
> 
> 8192 CLK_APMIXED_UNIVPLL
>   en_mask = 0xff000001
>   en_reg = 0x039c
>   base_en_bit = 0
> 
> prepare:
>   REG_CON0 |= en_mask
>   en_reg |= base_en_bit
> unprepare:
>   en_reg &= ~base_en_bit
> 
> 8192 CLK_APMIXED_USBPLL
>   en_mask = 0x00000000
>   en_reg = 0x03cc
>   base_en_bit = 2
> 
> prepare:
>   REG_CON0 |= en_mask (0)
>   en_reg |= base_en_bit
> unprepare:
>   en_reg &= ~base_en_bit
> 
> And I think the logic could still be simplified by _not_ putting
> CON0_BASE_EN in en_mask, and updating the CON0 in 2 steps: first all
> the bits that are not CON0_BASE_EN, then CON0_BASE_EN. Of course I
> assume that's it's fine to do so, but I have no idea.
> 
> register_pll() {
>    if (!en_addr) {
>      en_reg = REG_CON0
>      base_en_bit = CON0_BASE_EN
>    }
> }
> 
> prepare() {
>     REG_CON0 |= en_mask
>     en_reg |= base_en_bit
> }
> 
> unprepare() {
>     en_reg &= ~base_en_bit
> }
> 
> Then the new clock data:
> 
> 8183 CLK_APMIXED_ARMPLL_LL
>   en_mask = 0x00000000 (CON0_BASE_EN is implicit, but other bits could be set)
>   en_reg = 0
>   base_en_bit = 0
> 
> prepare: {
>     REG_CON0 |= en_mask (0x00000000, here, we can skip, but other bits
> could be set)
>     en_reg |= base_en_bit (REG_CON0 |= CON0_BASE_EN)
> }
> unprepare: en_reg &= ~base_en_bit (REG_CON0 &= ~CON0_BASE_EN)
> 
> 8192 CLK_APMIXED_UNIVPLL
>   en_mask = 0xff000001 (Note the bit 1 is _not_ dropped here, as it
> needs to be set too)
>   en_reg = 0x039c
>   base_en_bit = 0
> (same as above)
> 
> 8192 CLK_APMIXED_USBPLL
>   en_mask = 0x00000000
>   en_reg = 0x03cc
>   base_en_bit = 2
> (same as above)
> 
> Now, maybe this is also a bit overcomplicated. Maybe a simpler
> solution is just to add a comment in prepare that "r |=
> pll->data->en_mask;" is meant to include CON0_BASE_EN in most cases,
> and then the code could be ok as-is (just to make sure that the next
> person who looks at this code does not think there is a bug...).
> 

Hi Nicolas,

I thought these still too complicated and I guess the asymmetrical
problem could be fixed.
And that will make this part simpler just like what you mentioned in
previous comment.
I'll confirm ASAP and send a new version if it is possible to be fixed.

> >
> > > > +       if (pll->en_addr) {
> > > > +               r = readl(pll->en_addr);
> > > > +               r |= BIT(pll->data->base_en_bit);
> > > > +               writel(r, pll->en_addr);
> > > > +       }
> > > > +
> > > >         __mtk_pll_tuner_enable(pll);
> > > >
> > > >         udelay(20);
> > > > @@ -277,9 +287,15 @@ static void mtk_pll_unprepare(struct clk_hw *hw)
> > > >
> > > >         __mtk_pll_tuner_disable(pll);
> > > >
> > > > -       r = readl(pll->base_addr + REG_CON0);
> > > > -       r &= ~CON0_BASE_EN;
> > > > -       writel(r, pll->base_addr + REG_CON0);
> > > > +       if (pll->en_addr) {
> > > > +               r = readl(pll->en_addr);
> > > > +               r &= ~BIT(pll->data->base_en_bit);
> > > > +               writel(r, pll->en_addr);
> > > > +       } else {
> > > > +               r = readl(pll->base_addr + REG_CON0);
> > > > +               r &= ~CON0_BASE_EN;
> > > > +               writel(r, pll->base_addr + REG_CON0);
> > > > +       }
> > > >
> > > >         r = readl(pll->pwr_addr) | CON0_ISO_EN;
> > > >         writel(r, pll->pwr_addr);
> > > > @@ -321,6 +337,8 @@ static struct clk *mtk_clk_register_pll(const struct mtk_pll_data *data,
> > > >                 pll->tuner_addr = base + data->tuner_reg;
> > > >         if (data->tuner_en_reg)
> > > >                 pll->tuner_en_addr = base + data->tuner_en_reg;
> > > > +       if (data->en_reg)
> > > > +               pll->en_addr = base + data->en_reg;
> > >
> > > If the answer to my question above holds (asymmetry is not
> > > intentional), this patch/the code could be simplified a lot if you
> > > also added a pll->en_bit member, and, here, did this:
> > >
> > > if (pll->en_reg) {
> > >    pll->en_addr = base + data->en_reg;
> > >    pll->end_bit = data->en_bit;
> > > } else {
> > >    pll->en_addr = pll->base_addr + REG_CON0;
> > >    pll->en_bit = CON0_BASE_EN;
> > > }
> > >
> > > >         pll->hw.init = &init;
> > > >         pll->data = data;
> > > >
> > > > --
> > > > 1.8.1.1.dirty
> >

Powered by blists - more mailing lists