| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAFBinCBevei8Ozc3vsJpjq+N+d3WJ32L84OjKHvPtgy5qo_VAA@mail.gmail.com>
Date: Fri, 28 Jul 2017 22:47:49 +0200
From: Martin Blumenstingl <martin.blumenstingl@...glemail.com>
To: Neil Armstrong <narmstrong@...libre.com>
Cc: jbrunet@...libre.com, linux-amlogic@...ts.infradead.org,
linux-clk@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2 4/4] clk: meson: gxbb-aoclk: Add CEC 32k clock
Hi Neil,
On Fri, Jul 28, 2017 at 4:09 PM, Neil Armstrong <narmstrong@...libre.com> wrote:
> On 07/28/2017 12:29 PM, Martin Blumenstingl wrote:
>> Hi Neil,
>>
>> thanks for these patches, CEC support is another good step!
>
> Hi Martin,
>
> Thanks for your feedback !
>
>>
>> On Fri, Jul 28, 2017 at 11:53 AM, Neil Armstrong
>> <narmstrong@...libre.com> wrote:
>>> The CEC 32K AO Clock is a dual divider with dual counter to provide a more
>>> precise 32768Hz clock for the CEC subsystem from the external xtal.
>>>
>>> Signed-off-by: Neil Armstrong <narmstrong@...libre.com>
>>> ---
>>> drivers/clk/meson/Makefile | 2 +-
>>> drivers/clk/meson/gxbb-aoclk-32k.c | 180 +++++++++++++++++++++++++++++++++++++
>>> drivers/clk/meson/gxbb-aoclk.c | 21 ++++-
>>> drivers/clk/meson/gxbb-aoclk.h | 16 ++++
>>> 4 files changed, 217 insertions(+), 2 deletions(-)
>>> create mode 100644 drivers/clk/meson/gxbb-aoclk-32k.c
>>>
>>> diff --git a/drivers/clk/meson/Makefile b/drivers/clk/meson/Makefile
>>> index de65427..b139d41 100644
>>> --- a/drivers/clk/meson/Makefile
>>> +++ b/drivers/clk/meson/Makefile
>>> @@ -4,4 +4,4 @@
>>>
>>> obj-$(CONFIG_COMMON_CLK_AMLOGIC) += clk-pll.o clk-cpu.o clk-mpll.o clk-audio-divider.o
>>> obj-$(CONFIG_COMMON_CLK_MESON8B) += meson8b.o
>>> -obj-$(CONFIG_COMMON_CLK_GXBB) += gxbb.o gxbb-aoclk.o gxbb-aoclk-regmap.o
>>> +obj-$(CONFIG_COMMON_CLK_GXBB) += gxbb.o gxbb-aoclk.o gxbb-aoclk-regmap.o gxbb-aoclk-32k.o
>>> diff --git a/drivers/clk/meson/gxbb-aoclk-32k.c b/drivers/clk/meson/gxbb-aoclk-32k.c
>>> new file mode 100644
>>> index 0000000..497cd09
>>> --- /dev/null
>>> +++ b/drivers/clk/meson/gxbb-aoclk-32k.c
>>> @@ -0,0 +1,180 @@
>>> +/*
>>> + * Copyright (c) 2017 BayLibre, SAS.
>>> + * Author: Neil Armstrong <narmstrong@...libre.com>
>>> + *
>>> + * SPDX-License-Identifier: GPL-2.0+
>>> + */
>>> +
>>> +#include <linux/clk-provider.h>
>>> +#include <linux/bitfield.h>
>>> +#include <linux/regmap.h>
>>> +#include "gxbb-aoclk.h"
>>> +
>>> +/*
>>> + * The AO Domain embeds a dual/divider to generate a more precise
>>> + * 32,768KHz clock for low-power suspend mode and CEC.
>>> + */
>>> +
>>> +#define CLK_CNTL0_N1_MASK GENMASK(11, 0)
>>> +#define CLK_CNTL0_N2_MASK GENMASK(23, 12)
>>> +#define CLK_CNTL0_DUALDIV_EN BIT(28)
>>> +#define CLK_CNTL0_OUT_GATE_EN BIT(30)
>>> +#define CLK_CNTL0_IN_GATE_EN BIT(31)
>>> +
>>> +#define CLK_CNTL1_M1_MASK GENMASK(11, 0)
>>> +#define CLK_CNTL1_M2_MASK GENMASK(23, 12)
>>> +#define CLK_CNTL1_BYPASS_EN BIT(24)
>>> +#define CLK_CNTL1_SELECT_OSC BIT(27)
>>> +
>>> +#define PWR_CNTL_ALT_32K_SEL GENMASK(13, 10)
>>> +
>>> +struct cec_32k_freq_table {
>>> + unsigned long parent_rate;
>>> + unsigned long target_rate;
>>> + bool dualdiv;
>>> + unsigned int n1;
>>> + unsigned int n2;
>>> + unsigned int m1;
>>> + unsigned int m2;
>>> +};
>>> +
>>> +static const struct cec_32k_freq_table aoclk_cec_32k_table[] = {
>>> + [0] = {
>>> + .parent_rate = 24000000,
>> shouldn't you add a parent (XTAL) to this new clock so you could get
>> rid of this?
>
> It is in the clock definition in gxbb-aoclk.c.
> This table only defines the "valid" dividing values given by amlogic.
> Having 24MHz enforces we uses a 24MHz xtal as parent...
> Once we figure out/have a valid documentation, we could fill it with
> generic value and have a generic calculation.
ah, I missed that. in this case I guess it's fine
>>
>>> + .target_rate = 32768,
>>> + .dualdiv = true,
>>> + .n1 = 733,
>>> + .n2 = 732,
>>> + .m1 = 8,
>>> + .m2 = 11,
>>> + },
>>> +};
>>> +
>>> +/*
>>> + * If CLK_CNTL0_DUALDIV_EN == 0
>>> + * - will use N1 divider only
>>> + * If CLK_CNTL0_DUALDIV_EN == 1
>>> + * - hold M1 cycles of N1 divider then changes to N2
>>> + * - hold M2 cycles of N2 divider then changes to N1
>>> + * Then we can get more accurate division.
>>> + */
>>> +static unsigned long aoclk_cec_32k_recalc_rate(struct clk_hw *hw,
>>> + unsigned long parent_rate)
>>> +{
>>> + struct aoclk_cec_32k *cec_32k = to_aoclk_cec_32k(hw);
>>> + unsigned long n1;
>>> + u32 reg0, reg1;
>>> +
>>> + regmap_read(cec_32k->regmap, AO_RTC_ALT_CLK_CNTL0, ®0);
>>> + regmap_read(cec_32k->regmap, AO_RTC_ALT_CLK_CNTL1, ®1);
>>> +
>>> + if (reg1 & CLK_CNTL1_BYPASS_EN)
>>> + return parent_rate;
>>> +
>>> + if (reg0 & CLK_CNTL0_DUALDIV_EN) {
>>> + unsigned long n2, m1, m2, f1, f2, p1, p2;
>>> +
>>> + n1 = FIELD_GET(CLK_CNTL0_N1_MASK, reg0) + 1;
>>> + n2 = FIELD_GET(CLK_CNTL0_N2_MASK, reg0) + 1;
>>> +
>>> + m1 = FIELD_GET(CLK_CNTL1_M1_MASK, reg1) + 1;
>>> + m2 = FIELD_GET(CLK_CNTL1_M2_MASK, reg1) + 1;
>>> +
>>> + f1 = DIV_ROUND_CLOSEST(parent_rate, n1);
>>> + f2 = DIV_ROUND_CLOSEST(parent_rate, n2);
>>> +
>>> + p1 = DIV_ROUND_CLOSEST(100000000 * m1, f1 * (m1 + m2));
>>> + p2 = DIV_ROUND_CLOSEST(100000000 * m2, f2 * (m1 + m2));
>>> +
>>> + return DIV_ROUND_UP(100000000, p1 + p2);
>>> + }
>>> +
>>> + n1 = FIELD_GET(CLK_CNTL0_N1_MASK, reg0) + 1;
>>> +
>>> + return DIV_ROUND_CLOSEST(parent_rate, n1);
>>> +}
>>> +
>>> +static const struct cec_32k_freq_table *find_cec_32k_freq(unsigned long rate,
>>> + unsigned long prate)
>>> +{
>>> + int i;
>>> +
>>> + for (i = 0 ; i < ARRAY_SIZE(aoclk_cec_32k_table) ; ++i)
>>> + if (aoclk_cec_32k_table[i].parent_rate == prate &&
>>> + aoclk_cec_32k_table[i].target_rate == rate)
>>> + return &aoclk_cec_32k_table[i];
>>> +
>>> + return NULL;
>>> +}
>>> +
>>> +static long aoclk_cec_32k_round_rate(struct clk_hw *hw, unsigned long rate,
>>> + unsigned long *prate)
>>> +{
>>> + const struct cec_32k_freq_table *freq = find_cec_32k_freq(rate,
>>> + *prate);
>>> +
>>> + /* If invalid return first one */
>>> + if (!freq)
>>> + return freq[0].target_rate;
>>> +
>>> + return freq->target_rate;
>>> +}
>>> +
>>> +static int aoclk_cec_32k_set_rate(struct clk_hw *hw, unsigned long rate,
>>> + unsigned long parent_rate)
>>> +{
>>> + const struct cec_32k_freq_table *freq = find_cec_32k_freq(rate,
>>> + parent_rate);
>>> + struct aoclk_cec_32k *cec_32k = to_aoclk_cec_32k(hw);
>>> + u32 reg = 0;
>>> +
>>> + if (!freq)
>>> + return -EINVAL;
>>> +
>>> + /* Disable clock */
>>> + regmap_update_bits(cec_32k->regmap, AO_RTC_ALT_CLK_CNTL0,
>>> + CLK_CNTL0_IN_GATE_EN | CLK_CNTL0_OUT_GATE_EN, 0);
>>> +
>>> + if (freq->dualdiv)
>>> + reg = CLK_CNTL0_DUALDIV_EN |
>>> + FIELD_PREP(CLK_CNTL0_N1_MASK, freq->n1 - 1) |
>>> + FIELD_PREP(CLK_CNTL0_N2_MASK, freq->n2 - 1);
>>> + else
>>> + reg = FIELD_PREP(CLK_CNTL0_N1_MASK, freq->n1 - 1);
>>> +
>>> + regmap_write(cec_32k->regmap, AO_RTC_ALT_CLK_CNTL0, reg);
>>> +
>>> + if (freq->dualdiv)
>>> + reg = FIELD_PREP(CLK_CNTL1_M1_MASK, freq->m1 - 1) |
>>> + FIELD_PREP(CLK_CNTL1_M2_MASK, freq->m2 - 1);
>>> + else
>>> + reg = FIELD_PREP(CLK_CNTL1_M1_MASK, freq->m1 - 1);
>>> +
>>> + regmap_write(cec_32k->regmap, AO_RTC_ALT_CLK_CNTL1, reg);
>>> +
>>> + /* Enable clock */
>>> + regmap_update_bits(cec_32k->regmap, AO_RTC_ALT_CLK_CNTL0,
>>> + CLK_CNTL0_IN_GATE_EN, CLK_CNTL0_IN_GATE_EN);
>> based on how I understand your code the *input* clock needs to be
>> gated during rate change, so (in my opinion) it's fine to toggle it
>> here.
>>
>>> +
>>> + udelay(200);
>>> +
>>> + regmap_update_bits(cec_32k->regmap, AO_RTC_ALT_CLK_CNTL0,
>>> + CLK_CNTL0_OUT_GATE_EN, CLK_CNTL0_OUT_GATE_EN);
>> however, should we make the output clock configurable by implementing
>> clk_ops.enable and clk_ops.disable?
>
> Yes, but in a perfect world we should need 2 gates, one before and one after
> but it's a pain to synchronize and it's useless to have such complexity for
> s clock that will only ever be used for CEC...
the reason why I asked is the case where someone doesn't want to use
HDMI (some people on IRC were asking how to reclaim the memory which
is reserved for the framebuffer). I'm not sure if it would be good to
keep at least the output disabled for that case.
on the other hand one has to be careful then: I don't know if the
suspend firmware re-enables this clock (else we need to make sure that
we don't disable this clock during suspend).
I'll leave it up to you whether you want to change it or not (if we
really need it we can still implement these callbacks later).
>>
>>> +
>>> + regmap_update_bits(cec_32k->regmap, AO_CRT_CLK_CNTL1,
>>> + CLK_CNTL1_SELECT_OSC, CLK_CNTL1_SELECT_OSC);
>>> +
>>> + /* Select 32k from XTAL */
>>> + regmap_update_bits(cec_32k->regmap,
>>> + AO_RTI_PWR_CNTL_REG0,
>>> + PWR_CNTL_ALT_32K_SEL,
>>> + FIELD_PREP(PWR_CNTL_ALT_32K_SEL, 4));
>> is it correct to configure the muxes after enabling the clock output?
>> also do you know about the other parents of PWR_CNTL_ALT_32K_SEL?
>
> This mux is from some alternative GPIO input lines, but this exact clock
> configuration is used by the suspend firmware...
>
> Since there is a lot of unknowns about this clock, I copied the exact
> sequence from amlogic code, but yes it should be modelized with a mux
> and a dual gate somehow. But the firmware relies on it for low-freq
> suspend mode and we won't have any other usage apart CEC under linux.
>
> IMHO I think it's safer to keep this sequence.
if you have to re-spin this due to whatever reason: could you please
add a comment with that information (that this matches what the
suspend firmware does and that there's currently no documentation
available -> that should prevent people from messing with these
registers)
> (Anyway, it's not too late to change the clock model if we keep the
> same DT bindings)
indeed, that is basically the reason why I'm asking: if we have to
consider something now to keep the DT bindings stable then we should
look into it.
>>
>>> +
>>> + return 0;
>>> +}
>>> +
>>> +const struct clk_ops meson_aoclk_cec_32k_ops = {
>>> + .recalc_rate = aoclk_cec_32k_recalc_rate,
>>> + .round_rate = aoclk_cec_32k_round_rate,
>>> + .set_rate = aoclk_cec_32k_set_rate,
>>> +};
>>> diff --git a/drivers/clk/meson/gxbb-aoclk.c b/drivers/clk/meson/gxbb-aoclk.c
>>> index f61506c..6c161e0 100644
>>> --- a/drivers/clk/meson/gxbb-aoclk.c
>>> +++ b/drivers/clk/meson/gxbb-aoclk.c
>>> @@ -59,6 +59,7 @@
>>> #include <linux/mfd/syscon.h>
>>> #include <linux/regmap.h>
>>> #include <linux/init.h>
>>> +#include <linux/delay.h>
>>> #include <dt-bindings/clock/gxbb-aoclkc.h>
>>> #include <dt-bindings/reset/gxbb-aoclkc.h>
>>> #include "gxbb-aoclk.h"
>>> @@ -105,6 +106,17 @@ static int gxbb_aoclk_do_reset(struct reset_controller_dev *rcdev,
>>> GXBB_AO_GATE(uart2, 5);
>>> GXBB_AO_GATE(ir_blaster, 6);
>>>
>>> +static struct aoclk_cec_32k cec_32k_ao = {
>>> + .lock = &gxbb_aoclk_lock,
>>> + .hw.init = &(struct clk_init_data) {
>>> + .name = "cec_32k_ao",
>>> + .ops = &meson_aoclk_cec_32k_ops,
>>> + .parent_names = (const char *[]){ "xtal" },
>>> + .num_parents = 1,
>>> + .flags = CLK_IGNORE_UNUSED,
>>> + },
>>> +};
>>> +
>>> static unsigned int gxbb_aoclk_reset[] = {
>>> [RESET_AO_REMOTE] = 16,
>>> [RESET_AO_I2C_MASTER] = 18,
>>> @@ -131,8 +143,9 @@ static int gxbb_aoclk_do_reset(struct reset_controller_dev *rcdev,
>>> [CLKID_AO_UART1] = &uart1_ao.hw,
>>> [CLKID_AO_UART2] = &uart2_ao.hw,
>>> [CLKID_AO_IR_BLASTER] = &ir_blaster_ao.hw,
>>> + [CLKID_AO_CEC_32K] = &cec_32k_ao.hw,
>>> },
>>> - .num = ARRAY_SIZE(gxbb_aoclk_gate),
>>> + .num = 7,
>>> };
>>>
>>> static int gxbb_aoclkc_probe(struct platform_device *pdev)
>>> @@ -172,6 +185,12 @@ static int gxbb_aoclkc_probe(struct platform_device *pdev)
>>> return ret;
>>> }
>>>
>>> + /* Specific clocks */
>>> + cec_32k_ao.regmap = regmap;
>>> + ret = devm_clk_hw_register(dev, &cec_32k_ao.hw);
>>> + if (ret)
>>> + return ret;
>>> +
>>> return of_clk_add_hw_provider(dev->of_node, of_clk_hw_onecell_get,
>>> &gxbb_aoclk_onecell_data);
>>> }
>>> diff --git a/drivers/clk/meson/gxbb-aoclk.h b/drivers/clk/meson/gxbb-aoclk.h
>>> index 2e26108..e8604c8 100644
>>> --- a/drivers/clk/meson/gxbb-aoclk.h
>>> +++ b/drivers/clk/meson/gxbb-aoclk.h
>>> @@ -9,7 +9,13 @@
>>> #define __GXBB_AOCLKC_H
>>>
>>> /* AO Configuration Clock registers offsets */
>>> +#define AO_RTI_PWR_CNTL_REG1 0x0c
>>> +#define AO_RTI_PWR_CNTL_REG0 0x10
>>> #define AO_RTI_GEN_CNTL_REG0 0x40
>>> +#define AO_OSCIN_CNTL 0x58
>>> +#define AO_CRT_CLK_CNTL1 0x68
>>> +#define AO_RTC_ALT_CLK_CNTL0 0x94
>>> +#define AO_RTC_ALT_CLK_CNTL1 0x98
>>>
>>> struct aoclk_gate_regmap {
>>> struct clk_hw hw;
>>> @@ -23,4 +29,14 @@ struct aoclk_gate_regmap {
>>>
>>> extern const struct clk_ops meson_aoclk_gate_regmap_ops;
>>>
>>> +struct aoclk_cec_32k {
>>> + struct clk_hw hw;
>>> + struct regmap *regmap;
>>> + spinlock_t *lock;
>>> +};
>>> +
>>> +#define to_aoclk_cec_32k(_hw) container_of(_hw, struct aoclk_cec_32k, hw)
>>> +
>>> +extern const struct clk_ops meson_aoclk_cec_32k_ops;
>>> +
>>> #endif /* __GXBB_AOCLKC_H */
>>> --
>>> 1.9.1
>>>
>>>
>>> _______________________________________________
>>> linux-amlogic mailing list
>>> linux-amlogic@...ts.infradead.org
>>> http://lists.infradead.org/mailman/listinfo/linux-amlogic
>
Regards,
Martin
Powered by blists - more mailing lists