Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH V5 11/18] clk: tegra210: Add support for Tegra210 clocks
To:     Sowjanya Komatineni <skomatineni@nvidia.com>,
        thierry.reding@gmail.com, jonathanh@nvidia.com, tglx@linutronix.de,
        jason@lakedaemon.net, marc.zyngier@arm.com,
        linus.walleij@linaro.org, stefan@agner.ch, mark.rutland@arm.com
Cc:     pdeschrijver@nvidia.com, pgaikwad@nvidia.com, sboyd@kernel.org,
        linux-clk@vger.kernel.org, linux-gpio@vger.kernel.org,
        jckuo@nvidia.com, josephl@nvidia.com, talho@nvidia.com,
        linux-tegra@vger.kernel.org, linux-kernel@vger.kernel.org,
        mperttunen@nvidia.com, spatra@nvidia.com, robh+dt@kernel.org,
        devicetree@vger.kernel.org
References: <1561687972-19319-1-git-send-email-skomatineni@nvidia.com>
 <1561687972-19319-12-git-send-email-skomatineni@nvidia.com>
 <a5e1a6df-dff7-9e0c-9551-f78103a5462f@gmail.com>
 <a9b5c364-52b4-bee1-5881-47197f043950@nvidia.com>
 <e9d4bc0e-fd5d-ae02-2d67-86c7f7c9620f@gmail.com>
 <3938092a-bbc7-b304-641d-31677539598d@nvidia.com>
 <932d4d50-120c-9191-6a9a-23bf9c96633b@nvidia.com>
From:   Dmitry Osipenko <digetx@gmail.com>
Message-ID: <0ee055ad-d397-32e5-60ee-d62c14c6f77b@gmail.com>
Date:   Tue, 16 Jul 2019 06:50:24 +0300
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.7.2
MIME-Version: 1.0
In-Reply-To: <932d4d50-120c-9191-6a9a-23bf9c96633b@nvidia.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

16.07.2019 6:00, Sowjanya Komatineni пишет:
> 
> On 7/15/19 5:35 PM, Sowjanya Komatineni wrote:
>>
>> On 7/14/19 2:41 PM, Dmitry Osipenko wrote:
>>> 13.07.2019 8:54, Sowjanya Komatineni пишет:
>>>> On 6/29/19 8:10 AM, Dmitry Osipenko wrote:
>>>>> 28.06.2019 5:12, Sowjanya Komatineni пишет:
>>>>>> This patch adds system suspend and resume support for Tegra210
>>>>>> clocks.
>>>>>>
>>>>>> All the CAR controller settings are lost on suspend when core power
>>>>>> goes off.
>>>>>>
>>>>>> This patch has implementation for saving and restoring all the PLLs
>>>>>> and clocks context during system suspend and resume to have the
>>>>>> clocks back to same state for normal operation.
>>>>>>
>>>>>> Acked-by: Thierry Reding <treding@nvidia.com>
>>>>>> Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>
>>>>>> ---
>>>>>>    drivers/clk/tegra/clk-tegra210.c | 115
>>>>>> ++++++++++++++++++++++++++++++++++++++-
>>>>>>    drivers/clk/tegra/clk.c          |  14 +++++
>>>>>>    drivers/clk/tegra/clk.h          |   1 +
>>>>>>    3 files changed, 127 insertions(+), 3 deletions(-)
>>>>>>
>>>>>> diff --git a/drivers/clk/tegra/clk-tegra210.c
>>>>>> b/drivers/clk/tegra/clk-tegra210.c
>>>>>> index 1c08c53482a5..1b839544e086 100644
>>>>>> --- a/drivers/clk/tegra/clk-tegra210.c
>>>>>> +++ b/drivers/clk/tegra/clk-tegra210.c
>>>>>> @@ -9,10 +9,12 @@
>>>>>>    #include <linux/clkdev.h>
>>>>>>    #include <linux/of.h>
>>>>>>    #include <linux/of_address.h>
>>>>>> +#include <linux/of_platform.h>
>>>>>>    #include <linux/delay.h>
>>>>>>    #include <linux/export.h>
>>>>>>    #include <linux/mutex.h>
>>>>>>    #include <linux/clk/tegra.h>
>>>>>> +#include <linux/syscore_ops.h>
>>>>>>    #include <dt-bindings/clock/tegra210-car.h>
>>>>>>    #include <dt-bindings/reset/tegra210-car.h>
>>>>>>    #include <linux/iopoll.h>
>>>>>> @@ -20,6 +22,7 @@
>>>>>>    #include <soc/tegra/pmc.h>
>>>>>>      #include "clk.h"
>>>>>> +#include "clk-dfll.h"
>>>>>>    #include "clk-id.h"
>>>>>>      /*
>>>>>> @@ -225,6 +228,7 @@
>>>>>>      #define CLK_RST_CONTROLLER_RST_DEV_Y_SET 0x2a8
>>>>>>    #define CLK_RST_CONTROLLER_RST_DEV_Y_CLR 0x2ac
>>>>>> +#define CPU_SOFTRST_CTRL 0x380
>>>>>>      #define LVL2_CLK_GATE_OVRA 0xf8
>>>>>>    #define LVL2_CLK_GATE_OVRC 0x3a0
>>>>>> @@ -2820,6 +2824,7 @@ static int tegra210_enable_pllu(void)
>>>>>>        struct tegra_clk_pll_freq_table *fentry;
>>>>>>        struct tegra_clk_pll pllu;
>>>>>>        u32 reg;
>>>>>> +    int ret;
>>>>>>          for (fentry = pll_u_freq_table; fentry->input_rate;
>>>>>> fentry++) {
>>>>>>            if (fentry->input_rate == pll_ref_freq)
>>>>>> @@ -2847,10 +2852,10 @@ static int tegra210_enable_pllu(void)
>>>>>>        fence_udelay(1, clk_base);
>>>>>>        reg |= PLL_ENABLE;
>>>>>>        writel(reg, clk_base + PLLU_BASE);
>>>>>> +    fence_udelay(1, clk_base);
>>>>>>    -    readl_relaxed_poll_timeout_atomic(clk_base + PLLU_BASE, reg,
>>>>>> -                      reg & PLL_BASE_LOCK, 2, 1000);
>>>>>> -    if (!(reg & PLL_BASE_LOCK)) {
>>>>>> +    ret = tegra210_wait_for_mask(&pllu, PLLU_BASE, PLL_BASE_LOCK);
>>>>>> +    if (ret) {
>>>>>>            pr_err("Timed out waiting for PLL_U to lock\n");
>>>>>>            return -ETIMEDOUT;
>>>>>>        }
>>>>>> @@ -3283,6 +3288,103 @@ static void tegra210_disable_cpu_clock(u32
>>>>>> cpu)
>>>>>>    }
>>>>>>      #ifdef CONFIG_PM_SLEEP
>>>>>> +static u32 cpu_softrst_ctx[3];
>>>>>> +static struct platform_device *dfll_pdev;
>>>>>> +#define car_readl(_base, _off) readl_relaxed(clk_base + (_base) +
>>>>>> ((_off) * 4))
>>>>>> +#define car_writel(_val, _base, _off) \
>>>>>> +        writel_relaxed(_val, clk_base + (_base) + ((_off) * 4))
>>>>>> +
>>>>>> +static int tegra210_clk_suspend(void)
>>>>>> +{
>>>>>> +    unsigned int i;
>>>>>> +    struct device_node *node;
>>>>>> +
>>>>>> +    tegra_cclkg_burst_policy_save_context();
>>>>>> +
>>>>>> +    if (!dfll_pdev) {
>>>>>> +        node = of_find_compatible_node(NULL, NULL,
>>>>>> +                           "nvidia,tegra210-dfll");
>>>>>> +        if (node)
>>>>>> +            dfll_pdev = of_find_device_by_node(node);
>>>>>> +
>>>>>> +        of_node_put(node);
>>>>>> +        if (!dfll_pdev)
>>>>>> +            pr_err("dfll node not found. no suspend for dfll\n");
>>>>>> +    }
>>>>>> +
>>>>>> +    if (dfll_pdev)
>>>>>> +        tegra_dfll_suspend(dfll_pdev);
>>>>>> +
>>>>>> +    /* Enable PLLP_OUT_CPU after dfll suspend */
>>>>>> +    tegra_clk_set_pllp_out_cpu(true);
>>>>>> +
>>>>>> +    tegra_sclk_cclklp_burst_policy_save_context();
>>>>>> +
>>>>>> +    clk_save_context();
>>>>>> +
>>>>>> +    for (i = 0; i < ARRAY_SIZE(cpu_softrst_ctx); i++)
>>>>>> +        cpu_softrst_ctx[i] = car_readl(CPU_SOFTRST_CTRL, i);
>>>>>> +
>>>>>> +    return 0;
>>>>>> +}
>>>>>> +
>>>>>> +static void tegra210_clk_resume(void)
>>>>>> +{
>>>>>> +    unsigned int i;
>>>>>> +    struct clk_hw *parent;
>>>>>> +    struct clk *clk;
>>>>>> +
>>>>>> +    /*
>>>>>> +     * clk_restore_context restores clocks as per the clock tree.
>>>>>> +     *
>>>>>> +     * dfllCPU_out is first in the clock tree to get restored and it
>>>>>> +     * involves programming DFLL controller along with restoring
>>>>>> CPUG
>>>>>> +     * clock burst policy.
>>>>>> +     *
>>>>>> +     * DFLL programming needs dfll_ref and dfll_soc peripheral
>>>>>> clocks
>>>>>> +     * to be restores which are part ofthe peripheral clocks.
>>>                                              ^ white-space
>>>
>>> Please use spellchecker to avoid typos.
>>>
>>>>>> +     * So, peripheral clocks restore should happen prior to dfll
>>>>>> clock
>>>>>> +     * restore.
>>>>>> +     */
>>>>>> +
>>>>>> +    tegra_clk_osc_resume(clk_base);
>>>>>> +    for (i = 0; i < ARRAY_SIZE(cpu_softrst_ctx); i++)
>>>>>> +        car_writel(cpu_softrst_ctx[i], CPU_SOFTRST_CTRL, i);
>>>>>> +
>>>>>> +    /* restore all plls and peripheral clocks */
>>>>>> +    tegra210_init_pllu();
>>>>>> +    clk_restore_context();
>>>>>> +
>>>>>> +    fence_udelay(5, clk_base);
>>>>>> +
>>>>>> +    /* resume SCLK and CPULP clocks */
>>>>>> +    tegra_sclk_cpulp_burst_policy_restore_context();
>>>>>> +
>>>>>> +    /*
>>>>>> +     * restore CPUG clocks:
>>>>>> +     * - enable DFLL in open loop mode
>>>>>> +     * - switch CPUG to DFLL clock source
>>>>>> +     * - close DFLL loop
>>>>>> +     * - sync PLLX state
>>>>>> +     */
>>>>>> +    if (dfll_pdev)
>>>>>> +        tegra_dfll_resume(dfll_pdev, false);
>>>>>> +
>>>>>> +    tegra_cclkg_burst_policy_restore_context();
>>>>>> +    fence_udelay(2, clk_base);
>>>>>> +
>>>>>> +    if (dfll_pdev)
>>>>>> +        tegra_dfll_resume(dfll_pdev, true);
>>>>>> +
>>>>>> +    parent =
>>>>>> clk_hw_get_parent(__clk_get_hw(clks[TEGRA210_CLK_CCLK_G]));
>>>>>> +    clk = clks[TEGRA210_CLK_PLL_X];
>>>>>> +    if (parent != __clk_get_hw(clk))
>>>>>> +        tegra_clk_sync_state_pll(__clk_get_hw(clk));
>>>>>> +
>>>>>> +    /* Disable PLL_OUT_CPU after DFLL resume */
>>>>>> +    tegra_clk_set_pllp_out_cpu(false);
>>>>>> +}
>>>>>> +
>>>>>>    static void tegra210_cpu_clock_suspend(void)
>>>>>>    {
>>>>>>        /* switch coresite to clk_m, save off original source */
>>>>>> @@ -3298,6 +3400,11 @@ static void tegra210_cpu_clock_resume(void)
>>>>>>    }
>>>>>>    #endif
>>>>>>    +static struct syscore_ops tegra_clk_syscore_ops = {
>>>>>> +    .suspend = tegra210_clk_suspend,
>>>>>> +    .resume = tegra210_clk_resume,
>>>>>> +};
>>>>>> +
>>>>>>    static struct tegra_cpu_car_ops tegra210_cpu_car_ops = {
>>>>>>        .wait_for_reset    = tegra210_wait_cpu_in_reset,
>>>>>>        .disable_clock    = tegra210_disable_cpu_clock,
>>>>>> @@ -3583,5 +3690,7 @@ static void __init tegra210_clock_init(struct
>>>>>> device_node *np)
>>>>>>        tegra210_mbist_clk_init();
>>>>>>          tegra_cpu_car_ops = &tegra210_cpu_car_ops;
>>>>>> +
>>>>>> +    register_syscore_ops(&tegra_clk_syscore_ops);
>>>>>>    }
>>>>> Is it really worthwhile to use syscore_ops for suspend/resume given
>>>>> that drivers for
>>>>> won't resume before the CLK driver anyway? Are there any other options
>>>>> for CLK
>>>>> suspend/resume?
>>>>>
>>>>> I'm also not sure whether PM runtime API could be used at all in the
>>>>> context of
>>>>> syscore_ops ..
>>>>>
>>>>> Secondly, what about to use generic clk_save_context() /
>>>>> clk_restore_context()
>>>>> helpers for the suspend-resume? It looks to me that some other
>>>>> essential (and proper)
>>>>> platform driver (soc/tegra/? PMC?) should suspend-resume the clocks
>>>>> using the generic
>>>>> CLK Framework API.
>>>> Clock resume should happen very early to restore peripheral and cpu
>>>> clocks very early than peripheral drivers resume happens.
>>> If all peripheral drivers properly requested all of the necessary clocks
>>> and CLK driver was a platform driver, then I guess the probe should have
>>> been naturally ordered. But that's not very achievable with the
>>> currently available infrastructure in the kernel, so I'm not arguing
>>> that the clocks should be explicitly resumed before the users.
>>>
>>>> this patch series uses clk_save_context and clk_restore_context for
>>>> corresponding divider, pll, pllout.. save and restore context.
>>> Now I see that indeed this API is utilized in this patch, thank you for
>>> the clarification.
>>>
>>>> But as there is dependency on dfll resume and cpu and pllx clocks
>>>> restore, couldnt use clk_save_context and clk_restore_context for dfll.
>>>>
>>>> So implemented recommended dfll resume sequence in main Tegra210 clock
>>>> driver along with invoking clk_save_context/clk_restore_context where
>>>> all other clocks save/restore happens as per clock tree traversal.
>>> Could you please clarify what part of peripherals clocks is required for
>>> DFLL's restore? Couldn't DFLL driver be changed to avoid that quirkness
>>> and thus to make DFLL driver suspend/resume the clock?
>>
>> DFLL source ref_clk and soc_clk need to be restored prior to dfll.
>>
>> I see dfllCPU_out parent to CCLK_G first in the clock tree and
>> dfll_ref and dfll_soc peripheral clocks are not resumed by the time
>> dfll resume happens first.
>>
>> ref_clk and soc_clk source is from pll_p and clock tree has these
>> registered under pll_p which happens later.
>>
>> tegra210_clock_init registers in order plls, peripheral clocks,
>> super_clk init for cclk_g during clock driver probe and dfll probe and
>> register happens later.
>>
> One more thing, CLDVFS peripheral clock enable is also needed to be
> enabled to program DFLL Controller and all peripheral clock context is
> restored only after their PLL sources are restored.
> 
> DFLL restore involves dfll source clock resume along with CLDVFS
> periheral clock enable and reset
> 

I don't quite see why you can't simply add suspend/resume callbacks to
the CPUFreq driver to:

On suspend:
1. Switch CPU to PLLP (or whatever "safe" parent)
2. Disable/teardown DFLL

On resume:
1. Enable/restore DFLL
2. Switch CPU back to DFLL