lkml.org 
[lkml]   [2017]   [Aug]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [linux-sunxi] Re: [PATCH v4 3/3] clk: sunxi-ng: support R40 SoC
在 2017-08-15 10:16,Chen-Yu Tsai 写道:
> On Tue, Aug 15, 2017 at 12:53 AM, <icenowy@aosc.io> wrote:
>> 在 2017-08-15 00:12,Chen-Yu Tsai 写道:
>>>
>>> )On Sat, Aug 12, 2017 at 8:43 PM, Icenowy Zheng <icenowy@aosc.io>
>>> wrote:
>>>>
>>>> Allwinner R40 SoC have a clock controller module in the style of the
>>>> SoCs beyond sun6i, however, it's more rich and complex.
>>>>
>>>> Add support for it.
>>>>
>>>> Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
>>>> ---
>>>> Changes in v4:
>>>> - Removed usb-ohci-12M mux clocks.
>>>> - Removed unused (and not in user manual) adda-4x clock.
>>>> - Implemented proper SATA PLL system.
>>>> - Renamed MP (Mixed Processor) clock names to drop the extra "DE_".
>>>> - Renamed TCONs' clock names to "tcon-lcdX" or "tcon-tvX".
>>>> - Added missing RST_DRAM.
>>>> - Several clock post/pre-dividers and constraints fixes.
>>>> Changes in v3:
>>>> - Rebased on current linux-next.
>>>> Changes in v2:
>>>> - Fixes according to the SoC's user manual.
>>>>
>>>> drivers/clk/sunxi-ng/Kconfig | 5 +
>>>> drivers/clk/sunxi-ng/Makefile | 1 +
>>>> drivers/clk/sunxi-ng/ccu-sun8i-r40.c | 1240
>>>> +++++++++++++++++++++++++++++
>>>> drivers/clk/sunxi-ng/ccu-sun8i-r40.h | 69 ++
>>>> include/dt-bindings/clock/sun8i-r40-ccu.h | 187 +++++
>>>> include/dt-bindings/reset/sun8i-r40-ccu.h | 130 +++
>>>> 6 files changed, 1632 insertions(+)
>>>> create mode 100644 drivers/clk/sunxi-ng/ccu-sun8i-r40.c
>>>> create mode 100644 drivers/clk/sunxi-ng/ccu-sun8i-r40.h
>>>> create mode 100644 include/dt-bindings/clock/sun8i-r40-ccu.h
>>>> create mode 100644 include/dt-bindings/reset/sun8i-r40-ccu.h
>>>>
>>>> diff --git a/drivers/clk/sunxi-ng/Kconfig
>>>> b/drivers/clk/sunxi-ng/Kconfig
>>>> index 7342928c35cd..7a360737fe3c 100644
>>>> --- a/drivers/clk/sunxi-ng/Kconfig
>>>> +++ b/drivers/clk/sunxi-ng/Kconfig
>>>> @@ -48,6 +48,11 @@ config SUN8I_V3S_CCU
>>>> config SUN8I_DE2_CCU
>>>> bool "Support for the Allwinner SoCs DE2 CCU"
>>>>
>>>> +config SUN8I_R40_CCU
>>>> + bool "Support for the Allwinner R40 CCU"
>>>> + default MACH_SUN8I
>>>> + depends on MACH_SUN8I || COMPILE_TEST
>>>> +
>>>> config SUN9I_A80_CCU
>>>> bool "Support for the Allwinner A80 CCU"
>>>> default MACH_SUN9I
>>>> diff --git a/drivers/clk/sunxi-ng/Makefile
>>>> b/drivers/clk/sunxi-ng/Makefile
>>>> index 45a5910379a5..b1267fe68a8f 100644
>>>> --- a/drivers/clk/sunxi-ng/Makefile
>>>> +++ b/drivers/clk/sunxi-ng/Makefile
>>>> @@ -29,6 +29,7 @@ obj-$(CONFIG_SUN8I_H3_CCU) += ccu-sun8i-h3.o
>>>> obj-$(CONFIG_SUN8I_V3S_CCU) += ccu-sun8i-v3s.o
>>>> obj-$(CONFIG_SUN8I_DE2_CCU) += ccu-sun8i-de2.o
>>>> obj-$(CONFIG_SUN8I_R_CCU) += ccu-sun8i-r.o
>>>> +obj-$(CONFIG_SUN8I_R40_CCU) += ccu-sun8i-r40.o
>>>> obj-$(CONFIG_SUN9I_A80_CCU) += ccu-sun9i-a80.o
>>>> obj-$(CONFIG_SUN9I_A80_CCU) += ccu-sun9i-a80-de.o
>>>> obj-$(CONFIG_SUN9I_A80_CCU) += ccu-sun9i-a80-usb.o
>>>> diff --git a/drivers/clk/sunxi-ng/ccu-sun8i-r40.c
>>>> b/drivers/clk/sunxi-ng/ccu-sun8i-r40.c
>>>> new file mode 100644
>>>> index 000000000000..9dc6ad419f9a
>>>> --- /dev/null
>>>> +++ b/drivers/clk/sunxi-ng/ccu-sun8i-r40.c
>>>
>>>
>>> [...]
>>>
>>>> +/*
>>>> + * The MIPI PLL has 2 modes: "MIPI" and "HDMI".
>>>> + *
>>>> + * The MIPI mode is a standard NKM-style clock. The HDMI mode is an
>>>> + * integer / fractional clock with switchable multipliers and
>>>> dividers.
>>>> + * This is not supported here. We hardcode the PLL to MIPI mode.
>>>> + */
>>>> +#define SUN8I_R40_PLL_MIPI_REG 0x040
>>>> +
>>>> +static const char * const pll_mipi_parents[] = { "pll-video0" };
>>>> +static SUNXI_CCU_NKM_WITH_MUX_GATE_LOCK(pll_mipi_clk, "pll-mipi",
>>>> + pll_mipi_parents, 0x040,
>>>> + 8, 4, /* N */
>>>> + 4, 2, /* K */
>>>> + 0, 4, /* M */
>>>> + 21, 1, /* mux */
>>>> + BIT(31) | BIT(23) | BIT(22),
>>>> /*
>>>> gate */
>>>> + BIT(28), /* lock */
>>>> + CLK_SET_RATE_UNGATE);
>>>
>>>
>>> K has a minimum of 2 here.
>>
>>
>> I think it's not mentioned on the user manual...
>>
>> It says "The range is from 1 to 4."
>
> My R40 user manual v1.0 says otherwise. Page 85: PLL_FACTOR_K
> range is from 2 to 4.

Sorry... I read wrong page.

>
>>
>>>
>>>> +
>>>> +static SUNXI_CCU_NM_WITH_FRAC_GATE_LOCK(pll_de_clk, "pll-de",
>>>> + "osc24M", 0x0048,
>>>> + 8, 7, /* N */
>>>> + 0, 4, /* M */
>>>> + BIT(24), /* frac
>>>> enable */
>>>> + BIT(25), /* frac
>>>> select */
>>>> + 270000000, /* frac rate
>>>> 0 */
>>>> + 297000000, /* frac rate
>>>> 1 */
>>>> + BIT(31), /* gate */
>>>> + BIT(28), /* lock */
>>>> + 0);
>>>
>>>
>>> CLK_SET_RATE_UNGATE should be set for _all_ PLLs. The lock status bit
>>> depends on the PLL running, and we check the lock status whenever
>>> set_rate
>>> is done.
>>
>>
>> Should we also do this on older SoCs?
>>
>> I think at least on H3 I met PLL-DE waiting for lock failure.
>
> We probably should. But let us focus on this driver for now.
>
>>
>>>
>>>> +
>>>> +static SUNXI_CCU_NM_WITH_GATE_LOCK(pll_ddr1_clk, "pll-ddr1",
>>>> + "osc24M", 0x04c,
>>>> + 8, 7, /* N */
>>>> + 0, 2, /* M */
>>>> + BIT(31), /* gate */
>>>> + BIT(28), /* lock */
>>>> + CLK_SET_RATE_UNGATE);
>>>
>>>
>>> Manual says 16 <= N <= 75.
>>
>>
>> Maybe we need to implement more things for this constraint?
>>
>> But I think the frequency is not going to be changed when
>> running...
>
> Either you implement it now, or leave a note saying it has
> said restrictions. I am in favor of implementing it now.
> That said, some other PLLs have restrictions on N*K. That
> we cannot implement at this time.

I prefer to leave a note (also on PLL-DDR0).

>
>>
>>>
>>> [...]
>>>
>>>> +
>>>> +static struct clk_div_table ths_div_table[] = {
>>>> + { .val = 0, .div = 1 },
>>>> + { .val = 1, .div = 2 },
>>>> + { .val = 2, .div = 4 },
>>>> + { .val = 3, .div = 8 },
>>>> +};
>>>> +static const char * const ths_parents[] = { "osc24M" };
>>>> +static struct ccu_div ths_clk = {
>>>> + .enable = BIT(31),
>>>> + .div = _SUNXI_CCU_DIV_TABLE(0, 2, ths_div_table),
>>>
>>>
>>> You could use _SUNXI_CCU_DIV_FLAGS(0, 2, CLK_DIVIDER_POWER_OF_TWO)
>>> instead of div_table.
>>
>>
>> Thanks.
>>
>>>
>>>> + .mux = _SUNXI_CCU_MUX(24, 2),
>>>> + .common = {
>>>> + .reg = 0x074,
>>>> + .hw.init = CLK_HW_INIT_PARENTS("ths",
>>>> + ths_parents,
>>>> + &ccu_div_ops,
>>>> + 0),
>>>> + },
>>>> +};
>>>
>>>
>>> [...]
>>>
>>>> +static const char * const keypad_parents[] = { "osc24M", "osc32k",
>>>> };
>>>
>>>
>>> Nit: you can drop the extra comma.
>>
>>
>> Thanks.
>>
>>>
>>> [...]
>>>
>>>> +static const char * const sata_parents[] = { "pll-sata-out",
>>>> "sata-ext"
>>>> };
>>>> +static SUNXI_CCU_MUX_WITH_GATE(sata_clk, "sata", sata_parents,
>>>> + 0x0c8, 24, 1, BIT(31),
>>>> CLK_SET_RATE_PARENT);
>>>
>>>
>>> You don't have CLK_SET_RATE_PARENT for the parent, which is also a
>>> mux+gate.
>>> I'm not sure what you are expecting to happen here.
>>
>>
>> Oh maybe I should add it there.
>
> Adding it there may affect pll-periph0.

I should add it on pll-sata-out, but not on pll-periph0-sata.

pll-periph0-sata is a downstream clock of pll-periph0.

>
>>>
>>>> +
>>>> +static SUNXI_CCU_GATE(usb_phy0_clk, "usb-phy0", "osc24M",
>>>> + 0x0cc, BIT(8), 0);
>>>> +static SUNXI_CCU_GATE(usb_phy1_clk, "usb-phy1", "osc24M",
>>>> + 0x0cc, BIT(9), 0);
>>>> +static SUNXI_CCU_GATE(usb_phy2_clk, "usb-phy2", "osc24M",
>>>> + 0x0cc, BIT(10), 0);
>>>> +static SUNXI_CCU_GATE(usb_ohci0_clk, "usb-ohci0", "osc12M",
>>>> + 0x0cc, BIT(16), 0);
>>>> +static SUNXI_CCU_GATE(usb_ohci1_clk, "usb-ohci1", "osc12M",
>>>> + 0x0cc, BIT(17), 0);
>>>> +static SUNXI_CCU_GATE(usb_ohci2_clk, "usb-ohci2", "osc12M",
>>>> + 0x0cc, BIT(18), 0);
>>>
>>>
>>> Maybe you should force the OHCI muxes to 0?
>>
>>
>> This seems to be not done in A64 driver.
>>
>> P.S. I changed this block to make it similar to the A64 driver.
>
> Then you should at least leave a note saying this is unimplemented
> and we are depending on the hardware default.

Let me force it ;-)

>
>>
>>>
>>> [...]
>>>
>>>> +static const char * const tcon_parents[] = { "pll-video0",
>>>> "pll-video1",
>>>> + "pll-video0-2x",
>>>> "pll-video1-2x",
>>>> + "pll-mipi" };
>>>> +static SUNXI_CCU_MUX_WITH_GATE(tcon_lcd0_clk, "tcon-lcd0",
>>>> tcon_parents,
>>>> + 0x110, 24, 3, BIT(31),
>>>> CLK_SET_RATE_PARENT);
>>>> +static SUNXI_CCU_MUX_WITH_GATE(tcon_lcd1_clk, "tcon-lcd1",
>>>> tcon_parents,
>>>> + 0x114, 24, 3, BIT(31),
>>>> CLK_SET_RATE_PARENT);
>>>> +static SUNXI_CCU_M_WITH_MUX_GATE(tcon_tv0_clk, "tcon-tv0",
>>>> tcon_parents,
>>>> + 0x118, 0, 4, 24, 3, BIT(31), 0);
>>>> +static SUNXI_CCU_M_WITH_MUX_GATE(tcon_tv1_clk, "tcon-tv1",
>>>> tcon_parents,
>>>> + 0x11c, 0, 4, 24, 3, BIT(31), 0);
>>>
>>>
>>> You will likely need CLK_SET_RATE_PARENT for tcon-tv* as well.
>>
>>
>> According to the experience on H3, the TCONs may be more complex.
>>
>> Leave 0 here until we have either HDMI driver or TVE driver.
>
> OK.
>
>>
>>>
>>> [...]
>>>
>>>> +static const char * const csi_mclk_parents[] = { "osc24M",
>>>> "pll-video",
>>>
>>>
>>> pll-video1?
>>
>>
>> oh yes, thanks.
>>
>>>
>>>> + "pll-periph1" };
>>>> +static SUNXI_CCU_M_WITH_MUX_GATE(csi1_mclk_clk, "csi1-mclk",
>>>> csi_mclk_parents,
>>>> + 0x130, 0, 5, 8, 3, BIT(15), 0);
>>>> +
>>>> +static const char * const csi_sclk_parents[] = { "pll-periph0",
>>>> "pll-periph1" };
>>>> +static SUNXI_CCU_M_WITH_MUX_GATE(csi_sclk_clk, "csi-sclk",
>>>> csi_sclk_parents,
>>>> + 0x134, 16, 4, 24, 3, BIT(31), 0);
>>>> +
>>>> +static SUNXI_CCU_M_WITH_MUX_GATE(csi0_mclk_clk, "csi0-mclk",
>>>> csi_mclk_parents,
>>>> + 0x134, 0, 5, 8, 3, BIT(15), 0);
>>>
>>>
>>> [...]
>>>
>>>> +
>>>> +static SUNXI_CCU_M_WITH_GATE(ve_clk, "ve", "pll-ve",
>>>> + 0x13c, 16, 3, BIT(31), 0);
>>>
>>>
>>> CLK_SET_RATE_PARENT?
>>
>>
>> Seems ok.
>>
>>>
>>>> +
>>>> +static SUNXI_CCU_GATE(adda_clk, "adda",
>>>> "pll-audio",
>>>> + 0x140, BIT(31), CLK_SET_RATE_PARENT);
>>>
>>>
>>> You call the bus clock "bus-codec", but here you call it "adda". Pick
>>> one
>>> and
>>> be consistent.
>>
>>
>> Use codec?
>
> Cool.
>
>>
>>>
>>> [...]
>>>
>>>> +static const char * const hdmi_parents[] = { "pll-video0",
>>>> "pll-video1"
>>>> };
>>>> +static SUNXI_CCU_M_WITH_MUX_GATE(hdmi_clk, "hdmi", hdmi_parents,
>>>> + 0x150, 0, 4, 24, 2, BIT(31), 0);
>>>
>>>
>>> CLK_SET_RATE_PARENT?
>>
>>
>> Wait for HDMI driver.
>
> OK.
>
>>
>>>
>>>
>>> [...]
>>>
>>>> +static const char * const mbus_parents[] = { "osc24M",
>>>> "pll-periph0-2x",
>>>> + "pll-ddr0" };
>>>> +static SUNXI_CCU_MP_WITH_MUX_GATE(mbus_clk, "mbus", mbus_parents,
>>>> 0x15c,
>>>> + 0, 4, /* M */
>>>> + 16, 2, /* P */
>>>> + 24, 2, /* mux */
>>>> + BIT(31), /* gate */
>>>> + CLK_IS_CRITICAL);
>>>
>>>
>>> The documentation for this one is fuzzy. It mentions the N (or P)
>>> factor,
>>> but does not include it in the formula. Can you double check it
>>> against
>>> the BSP and add a comment about it?
>>
>>
>> ```
>> SUNXI_CLK_PERIPH(mbus, MBUS_CFG, 24, 2, MBUS_CFG, 0, 4, 16, 2, 0,
>> MBUS_CFG,
>> MBUS_RST, 0, 0, 31, 31, 0, 0, &clk_lock, NULL,
>> 0);
>> ```
>> I think the BSP driver used both factors, however, the driver may be
>> also
>> not accurate as I think maybe mbus is not going to be changed even in
>> BSP
>> kernel.
>
> OK. Please add a comment saying that the doc is fuzzy and you're
> following
> the BSP on this one.

OK.

>
>>
>>>
>>>> +
>>>> +static const char * const dsi_dphy_parents[] = { "pll-video0",
>>>> "pll-video1",
>>>> + "pll-periph0" };
>>>> +static SUNXI_CCU_M_WITH_MUX_GATE(dsi_dphy_clk, "dsi-dphy",
>>>> dsi_dphy_parents,
>>>> + 0x168, 0, 4, 8, 2, BIT(15), 0);
>>>
>>>
>>> Note: we might need CLK_SET_RATE_PARENT later on, but we will need to
>>> find
>>> a way to keep pll-periph* fixed.
>>>
>>>> +
>>>> +static SUNXI_CCU_M_WITH_MUX_GATE(tve0_clk, "tve0", tcon_parents,
>>>> + 0x180, 0, 4, 24, 3, BIT(31), 0);
>>>> +static SUNXI_CCU_M_WITH_MUX_GATE(tve1_clk, "tve1", tcon_parents,
>>>> + 0x184, 0, 4, 24, 3, BIT(31), 0);
>>>
>>>
>>> CLK_SET_RATE_PARENT?
>>
>>
>> Wait for the real driver -- it may be more complex.
>
> OK.
>
>>
>>>
>>>> +
>>>> +static const char * const tvd_parents[] = { "pll-video0",
>>>> "pll-video1",
>>>> + "pll-video0-2x",
>>>> "pll-video1-2x" };
>>>> +static SUNXI_CCU_M_WITH_MUX_GATE(tvd0_clk, "tvd0", tvd_parents,
>>>> + 0x188, 0, 4, 24, 3, BIT(31), 0);
>>>> +static SUNXI_CCU_M_WITH_MUX_GATE(tvd1_clk, "tvd1", tvd_parents,
>>>> + 0x18c, 0, 4, 24, 3, BIT(31), 0);
>>>> +static SUNXI_CCU_M_WITH_MUX_GATE(tvd2_clk, "tvd2", tvd_parents,
>>>> + 0x190, 0, 4, 24, 3, BIT(31), 0);
>>>> +static SUNXI_CCU_M_WITH_MUX_GATE(tvd3_clk, "tvd3", tvd_parents,
>>>> + 0x194, 0, 4, 24, 3, BIT(31), 0);
>>>
>>>
>>> CLK_SET_RATE_PARENT?
>>
>>
>> Not sure at all, maybe they should be changed when really doing TVD
>> driver.
>
> OK.
>
>>
>>>
>>> [...]
>>>
>>> You are missing the SYS_32k mux at register 0x310. According to
>>> the diagrams, this is the actual clock that supplies the LOSC
>>> parent of various muxes in the CCU. It itself is a mux between
>>> an internal 2M RC oscillator, and LOSC from the RTC module.
>>>
>>>> +
>>>> +static struct ccu_reset_map sun8i_r40_ccu_resets[] = {
>>>
>>>
>>> [...]
>>>
>>>> + [RST_BUS_HDMI_SLOW] = { 0x2c4, BIT(10) },
>>>> + [RST_BUS_HDMI] = { 0x2c4, BIT(11) },
>>>
>>>
>>> Manual says bit 10 is HDMI0, bit 11 is HDMI1. Looking at the clocks,
>>> bit 10 would be RST_BUS_HDMI, while bit 11 would be
>>> RST_BUS_HDMI_SLOW.
>>
>>
>> Let me just call them 0 and 1 in next version.
>
> OK.
>
>>
>>
>>>
>>> [...]
>>>
>>>> +
>>>> +static const struct sunxi_ccu_desc sun8i_r40_ccu_desc = {
>>>> + .ccu_clks = sun8i_r40_ccu_clks,
>>>> + .num_ccu_clks = ARRAY_SIZE(sun8i_r40_ccu_clks),
>>>> +
>>>> + .hw_clks = &sun8i_r40_hw_clks,
>>>> +
>>>> + .resets = sun8i_r40_ccu_resets,
>>>> + .num_resets = ARRAY_SIZE(sun8i_r40_ccu_resets),
>>>> +};
>>>> +
>>>> +static struct ccu_mux_nb sun8i_r40_cpu_nb = {
>>>> + .common = &cpu_clk.common,
>>>> + .cm = &cpu_clk.mux,
>>>> + .delay_us = 1, /* > 8 clock cycles at 24 MHz */
>>>> + .bypass_index = 1, /* index of 24 MHz oscillator */
>>>> +};
>>>> +
>>>> +static void __init sun8i_r40_ccu_setup(struct device_node *node)
>>>> +{
>>>> + void __iomem *reg;
>>>> + u32 val;
>>>> +
>>>> + reg = of_io_request_and_map(node, 0,
>>>> of_node_full_name(node));
>>>> + if (IS_ERR(reg)) {
>>>> + pr_err("%s: Could not map the clock registers\n",
>>>> + of_node_full_name(node));
>>>> + return;
>>>> + }
>>>> +
>>>> + /* Force the PLL-Audio-1x divider to 4 */
>>>> + val = readl(reg + SUN8I_R40_PLL_AUDIO_REG);
>>>> + val &= ~GENMASK(19, 16);
>>>> + writel(val | (3 << 16), reg + SUN8I_R40_PLL_AUDIO_REG);
>>>
>>>
>>> You should also force pll-mipi to mipi mode. And the OHCI muxes.
>>>
>>>> +
>>>> + sunxi_ccu_probe(node, reg, &sun8i_r40_ccu_desc);
>>>> +
>>>> + ccu_mux_notifier_register(pll_cpu_clk.common.hw.clk,
>>>> + &sun8i_r40_cpu_nb);
>>>
>>>
>>> Do you also need the gate-then-ungate notifier?
>>
>>
>> Still unknown...
>
> OK.

Maybe I should do some experiments, and if possible, add DVFS even
in the start?

We have proper driver for AXP221s ;-)

>
> Thanks
> ChenYu
>
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

\
 
 \ /
  Last update: 2017-08-15 04:27    [W:0.067 / U:0.860 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site