Re: [PATCH v6 7/9] clk: mediatek: Add subsystem clocks of MT8173

[Sascha Hauer <s.hauer@pengutronix.de>]

On Wed, Aug 05, 2015 at 03:41:49PM +0800, Daniel Kurtz wrote:
> On Wed, Aug 5, 2015 at 3:36 PM, Sascha Hauer <s.hauer@pengutronix.de> wrote:
> > On Wed, Aug 05, 2015 at 03:26:29PM +0800, Daniel Kurtz wrote:
> >> On Wed, Aug 5, 2015 at 2:46 PM, Sascha Hauer <s.hauer@pengutronix.de> wrote:
> >> > On Tue, Aug 04, 2015 at 04:16:56PM +0800, James Liao wrote:
> >> >> Most multimedia subsystem clocks will be accessed by multiple
> >> >> drivers, so it's a better way to manage these clocks in CCF.
> >> >> This patch adds clock support for MM, IMG, VDEC, VENC and VENC_LT
> >> >> subsystems.
> >> >>
> >> >> Signed-off-by: James Liao <jamesjj.liao@mediatek.com>
> >> >> ---
> >> >>  drivers/clk/mediatek/clk-mt8173.c      | 267 +++++++++++++++++++++++++++++++++
> >> >>  include/dt-bindings/clock/mt8173-clk.h |  97 +++++++++++-
> >> >>  2 files changed, 361 insertions(+), 3 deletions(-)
> >> >>
> >> >> diff --git a/drivers/clk/mediatek/clk-mt8173.c b/drivers/clk/mediatek/clk-mt8173.c
> >> >> index f37ace6..05335e5 100644
> >> >> --- a/drivers/clk/mediatek/clk-mt8173.c
> >> >> +++ b/drivers/clk/mediatek/clk-mt8173.c
> >> >> @@ -25,6 +25,10 @@ static DEFINE_SPINLOCK(mt8173_clk_lock);
> >> >>  static const struct mtk_fixed_clk fixed_clks[] __initconst = {
> >> >>       FIXED_CLK(CLK_TOP_CLKPH_MCK_O, "clkph_mck_o", "clk26m", 400 * MHZ),
> >> >>       FIXED_CLK(CLK_TOP_USB_SYSPLL_125M, "usb_syspll_125m", "clk26m", 125 * MHZ),
> >> >> +     FIXED_CLK(CLK_TOP_DSI0_DIG, "dsi0_dig", "clk26m", 130 * MHZ),
> >> >> +     FIXED_CLK(CLK_TOP_DSI1_DIG, "dsi1_dig", "clk26m", 130 * MHZ),
> >> >> +     FIXED_CLK(CLK_TOP_LVDS_PXL, "lvds_pxl", "lvdspll", 148.5 * MHZ),
> >> >> +     FIXED_CLK(CLK_TOP_LVDS_CTS, "lvds_cts", "lvdspll", 51.975 * MHZ),
> >> >
> >> > I would expect 51975 * KHZ here to avoid fractional numbers. Probably
> >> > gcc calculates that during compile time so this will work as expected,
> >> > still I'm not sure this is good style to use fractional numbers here.
> >>
> >> I thought this looked a bit strange too, but for what its worth, these
> >> two evaluate correctly:
> >>
> >> localhost ~ # cat /sys/kernel/debug/clk/clk_summary  | grep lvds
> >>     lvdspll                               0            0   149999878
> >>        0 0
> >>        lvds_pxl                           0            0   148500000
> >>        0 0
> >>        lvds_cts                           0            0    51975000
> >>        0 0
> >>
> >> >
> >> > Anyway, on my system lvdspll is running at 150MHz. Are you sure there is
> >> > a clock derived from this running at 148.5MHz? Is it really correct to
> >> > use a fixed clock here or should it rather be lvdspll directly?
> >>
> >> I agree it does look strange to have a 51.975 MHz and 148.5 MHz clocks
> >> with a 150 MHz PLL as their parent...  However, I'm not sure how much
> >> this matters?  I think the idea here was that these frequencies are
> >> best effort "nominal" clock values provided by Mediatek "designers".
> >> The important point is that for the hardware to generate these either
> >> of these clocks, lvdspll must be enabled.
> >
> > This assumes that the lvdspll always runs at frequency the designers
> > thought that might be a good value. Should we really provide wrong clock
> > values when on some board for whatever reason the lvdspll is configured
> > for a different frequency?
> 
> Do you have an alternative suggestion for estimating the frequency of
> a non-software controllable or measurable hardware clock?

What's the problem with using lvdspll directly? The consumers of these
clocks should be able to cope with the deviation between nomial
frequency and actual frequency.

Sascha

-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |