| Message ID | 1617758731-12380-4-git-send-email-skomatineni@nvidia.com |
|---|---|
| State | Rejected |
| Headers | show |
| Series | Add AHCI support for Tegra186 | expand |
On Tue, Apr 06, 2021 at 06:25:31PM -0700, Sowjanya Komatineni wrote: > This patch adds support for AHCI-compliant Serial ATA controller > on Tegra186 SoC. > > Tegra186 does not have sata-oob reset. > Tegra186 SATA_NVOOB register filed COMMA_CNT position and width are > different compared to Tegra210 and prior. > > So, this patch adds a flag has_sata_oob_rst and tegra_ahci_regs to > SoC specific strcuture tegra_ahci_soc and updated their implementation > accordingly. > > Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com> > --- > drivers/ata/ahci_tegra.c | 60 +++++++++++++++++++++++++++++++++++++----------- > 1 file changed, 47 insertions(+), 13 deletions(-) Acked-by: Thierry Reding <treding@nvidia.com>
07.04.2021 04:25, Sowjanya Komatineni пишет: > + if (!tegra->pdev->dev.pm_domain) { > + ret = tegra_powergate_sequence_power_up(TEGRA_POWERGATE_SATA, > + tegra->sata_clk, > + tegra->sata_rst); > + if (ret) > + goto disable_regulators; > + } > Hi, Why you haven't added condition for tegra_powergate_power_off()? I think it should break GENPD and legacy PD API isn't not supported by T186 at all. I'm also not sure whether the power up/down sequence is correct using GENPD. Moreover the driver doesn't support runtime PM, so GENPD should be always off?
On 4/7/21 2:36 PM, Dmitry Osipenko wrote: > 07.04.2021 04:25, Sowjanya Komatineni пишет: >> + if (!tegra->pdev->dev.pm_domain) { >> + ret = tegra_powergate_sequence_power_up(TEGRA_POWERGATE_SATA, >> + tegra->sata_clk, >> + tegra->sata_rst); >> + if (ret) >> + goto disable_regulators; >> + } >> > Hi, > > Why you haven't added condition for tegra_powergate_power_off()? I think > it should break GENPD and legacy PD API isn't not supported by T186 at all. > > I'm also not sure whether the power up/down sequence is correct using GENPD. > > Moreover the driver doesn't support runtime PM, so GENPD should be > always off? This driver already using legacy PD API's so thought its supported and added power domain device check during powergate_sequence_power_up and yes same should apply for powergate_power_off as well. But if legacy PD is not supported by T186 then not sure why original driver even using these API's. Preetham/Thierry, Can you please comment ? But as RPM is not implemented yet for this driver, GENPD will be OFF but SATA is not in power-gate by the time kernel starts and functionally works. But with RPM implementation, I guess we can do proper power gate on/off.
On 4/7/21 3:57 PM, Sowjanya Komatineni wrote: > > On 4/7/21 2:36 PM, Dmitry Osipenko wrote: >> 07.04.2021 04:25, Sowjanya Komatineni пишет: >>> + if (!tegra->pdev->dev.pm_domain) { >>> + ret = tegra_powergate_sequence_power_up(TEGRA_POWERGATE_SATA, >>> + tegra->sata_clk, >>> + tegra->sata_rst); >>> + if (ret) >>> + goto disable_regulators; >>> + } >> Hi, >> >> Why you haven't added condition for tegra_powergate_power_off()? I think >> it should break GENPD and legacy PD API isn't not supported by T186 >> at all. >> >> I'm also not sure whether the power up/down sequence is correct using >> GENPD. >> >> Moreover the driver doesn't support runtime PM, so GENPD should be >> always off? > > This driver already using legacy PD API's so thought its supported and > added power domain device check during powergate_sequence_power_up and > yes same should apply for powergate_power_off as well. But if legacy > PD is not supported by T186 then not sure why original driver even > using these API's. > > Sorry just took a look and driver supports T210 and prior tegra as well. T210 and prior supports legacy PD and this check is applicable for those. So we should add power domain device check for power off as well. But for T186, we should have GENPD working once we add runtime PM support to driver. Preetham/Thierry, Can you confirm where SATA is un powergated prior to kernel? > But as RPM is not implemented yet for this driver, GENPD will be OFF > but SATA is not in power-gate by the time kernel starts and > functionally works. > > But with RPM implementation, I guess we can do proper power gate on/off. >
08.04.2021 02:00, Sowjanya Komatineni пишет: > > On 4/7/21 3:57 PM, Sowjanya Komatineni wrote: >> >> On 4/7/21 2:36 PM, Dmitry Osipenko wrote: >>> 07.04.2021 04:25, Sowjanya Komatineni пишет: >>>> + if (!tegra->pdev->dev.pm_domain) { >>>> + ret = tegra_powergate_sequence_power_up(TEGRA_POWERGATE_SATA, >>>> + tegra->sata_clk, >>>> + tegra->sata_rst); >>>> + if (ret) >>>> + goto disable_regulators; >>>> + } >>> Hi, >>> >>> Why you haven't added condition for tegra_powergate_power_off()? I think >>> it should break GENPD and legacy PD API isn't not supported by T186 >>> at all. >>> >>> I'm also not sure whether the power up/down sequence is correct using >>> GENPD. >>> >>> Moreover the driver doesn't support runtime PM, so GENPD should be >>> always off? >> >> This driver already using legacy PD API's so thought its supported and >> added power domain device check during powergate_sequence_power_up and >> yes same should apply for powergate_power_off as well. But if legacy >> PD is not supported by T186 then not sure why original driver even >> using these API's. >> >> > Sorry just took a look and driver supports T210 and prior tegra as well. > T210 and prior supports legacy PD and this check is applicable for > those. So we should add power domain device check for power off as well. You could fix it with a follow up patch. Please try to test that power-off works properly, at least try to unload the driver module and re-load it. > But for T186, we should have GENPD working once we add runtime PM > support to driver. > > Preetham/Thierry, Can you confirm where SATA is un powergated prior to > kernel? > > >> But as RPM is not implemented yet for this driver, GENPD will be OFF >> but SATA is not in power-gate by the time kernel starts and >> functionally works. >> >> But with RPM implementation, I guess we can do proper power gate on/off. >> I now recalled that GENPD turns ON all domains by default and then turns them OFF only when driver entered into the RPM-suspended state. This means that AHCI GENPD should be always-ON for T186, which should be okay if this doesn't break power sequences.
On Thu, Apr 08, 2021 at 02:25:19AM +0300, Dmitry Osipenko wrote: > 08.04.2021 02:00, Sowjanya Komatineni пишет: > > > > On 4/7/21 3:57 PM, Sowjanya Komatineni wrote: > >> > >> On 4/7/21 2:36 PM, Dmitry Osipenko wrote: > >>> 07.04.2021 04:25, Sowjanya Komatineni пишет: > >>>> + if (!tegra->pdev->dev.pm_domain) { > >>>> + ret = tegra_powergate_sequence_power_up(TEGRA_POWERGATE_SATA, > >>>> + tegra->sata_clk, > >>>> + tegra->sata_rst); > >>>> + if (ret) > >>>> + goto disable_regulators; > >>>> + } > >>> Hi, > >>> > >>> Why you haven't added condition for tegra_powergate_power_off()? I think > >>> it should break GENPD and legacy PD API isn't not supported by T186 > >>> at all. > >>> > >>> I'm also not sure whether the power up/down sequence is correct using > >>> GENPD. > >>> > >>> Moreover the driver doesn't support runtime PM, so GENPD should be > >>> always off? > >> > >> This driver already using legacy PD API's so thought its supported and > >> added power domain device check during powergate_sequence_power_up and > >> yes same should apply for powergate_power_off as well. But if legacy > >> PD is not supported by T186 then not sure why original driver even > >> using these API's. > >> > >> > > Sorry just took a look and driver supports T210 and prior tegra as well. > > T210 and prior supports legacy PD and this check is applicable for > > those. So we should add power domain device check for power off as well. > > You could fix it with a follow up patch. Please try to test that > power-off works properly, at least try to unload the driver module and > re-load it. Agreed, this should have the same check as above for tegra_powergate_power_off(). It currently works fine because on Tegra186 tegra_powergate_power_off() (and all the other legacy APIs for that matter) will abort early since no power gates are implemented. The AHCI driver doesn't check for errors, so this will just fail silently. It's better to be symmetric, though, and add the check in both paths. > > But for T186, we should have GENPD working once we add runtime PM > > support to driver. > > > > Preetham/Thierry, Can you confirm where SATA is un powergated prior to > > kernel? > > > > > >> But as RPM is not implemented yet for this driver, GENPD will be OFF > >> but SATA is not in power-gate by the time kernel starts and > >> functionally works. > >> > >> But with RPM implementation, I guess we can do proper power gate on/off. > >> > > I now recalled that GENPD turns ON all domains by default and then turns > them OFF only when driver entered into the RPM-suspended state. This > means that AHCI GENPD should be always-ON for T186, which should be okay > if this doesn't break power sequences. Yeah, the generic PM domain will just stay enabled after probe and until remove. This does not impact the power sequences because they have to be completely implemented in the power domains code anyway. With the legacy API we used to need more rigorous sequences in the individual drivers, but with generic PM domains none of that should be necessary, though it also doesn't hurt, so some of the unnecessary clock enablement code is kept for simplicity. To be honest, I'm not sure if it's worth adding runtime PM support for this driver. If this top-level layer has a way of getting notification when no device was detected, then it might make some sense to turn off the power domain and the regulators again, but I'm not sure if that's the case. tegra_ahci_host_stop() seems like it might be usable for that so yeah, that might work. We currently do turn off the powergate in that case, so extending that power optimization to Tegra186 using runtime PM makes sense. Thierry
08.04.2021 16:06, Thierry Reding пишет: > On Thu, Apr 08, 2021 at 02:25:19AM +0300, Dmitry Osipenko wrote: >> 08.04.2021 02:00, Sowjanya Komatineni пишет: >>> >>> On 4/7/21 3:57 PM, Sowjanya Komatineni wrote: >>>> >>>> On 4/7/21 2:36 PM, Dmitry Osipenko wrote: >>>>> 07.04.2021 04:25, Sowjanya Komatineni пишет: >>>>>> + if (!tegra->pdev->dev.pm_domain) { >>>>>> + ret = tegra_powergate_sequence_power_up(TEGRA_POWERGATE_SATA, >>>>>> + tegra->sata_clk, >>>>>> + tegra->sata_rst); >>>>>> + if (ret) >>>>>> + goto disable_regulators; >>>>>> + } >>>>> Hi, >>>>> >>>>> Why you haven't added condition for tegra_powergate_power_off()? I think >>>>> it should break GENPD and legacy PD API isn't not supported by T186 >>>>> at all. >>>>> >>>>> I'm also not sure whether the power up/down sequence is correct using >>>>> GENPD. >>>>> >>>>> Moreover the driver doesn't support runtime PM, so GENPD should be >>>>> always off? >>>> >>>> This driver already using legacy PD API's so thought its supported and >>>> added power domain device check during powergate_sequence_power_up and >>>> yes same should apply for powergate_power_off as well. But if legacy >>>> PD is not supported by T186 then not sure why original driver even >>>> using these API's. >>>> >>>> >>> Sorry just took a look and driver supports T210 and prior tegra as well. >>> T210 and prior supports legacy PD and this check is applicable for >>> those. So we should add power domain device check for power off as well. >> >> You could fix it with a follow up patch. Please try to test that >> power-off works properly, at least try to unload the driver module and >> re-load it. > > Agreed, this should have the same check as above for > tegra_powergate_power_off(). It currently works fine because on Tegra186 > tegra_powergate_power_off() (and all the other legacy APIs for that > matter) will abort early since no power gates are implemented. The AHCI > driver doesn't check for errors, so this will just fail silently. It's > better to be symmetric, though, and add the check in both paths. I missed that tegra_powergate_power_off() usage isn't fatal if GENPD is used, thank you for the clarification. >>> But for T186, we should have GENPD working once we add runtime PM >>> support to driver. >>> >>> Preetham/Thierry, Can you confirm where SATA is un powergated prior to >>> kernel? >>> >>> >>>> But as RPM is not implemented yet for this driver, GENPD will be OFF >>>> but SATA is not in power-gate by the time kernel starts and >>>> functionally works. >>>> >>>> But with RPM implementation, I guess we can do proper power gate on/off. >>>> >> >> I now recalled that GENPD turns ON all domains by default and then turns >> them OFF only when driver entered into the RPM-suspended state. This >> means that AHCI GENPD should be always-ON for T186, which should be okay >> if this doesn't break power sequences. > > Yeah, the generic PM domain will just stay enabled after probe and until > remove. This does not impact the power sequences because they have to be > completely implemented in the power domains code anyway. With the legacy > API we used to need more rigorous sequences in the individual drivers, > but with generic PM domains none of that should be necessary, though it > also doesn't hurt, so some of the unnecessary clock enablement code is > kept for simplicity. > > To be honest, I'm not sure if it's worth adding runtime PM support for > this driver. If this top-level layer has a way of getting notification > when no device was detected, then it might make some sense to turn off > the power domain and the regulators again, but I'm not sure if that's > the case. tegra_ahci_host_stop() seems like it might be usable for that > so yeah, that might work. We currently do turn off the powergate in that > case, so extending that power optimization to Tegra186 using runtime PM > makes sense. Alright, then this all should be good as-is.
diff --git a/drivers/ata/ahci_tegra.c b/drivers/ata/ahci_tegra.c index cb55ebc1..56612af 100644 --- a/drivers/ata/ahci_tegra.c +++ b/drivers/ata/ahci_tegra.c @@ -59,8 +59,6 @@ #define T_SATA0_CFG_PHY_1_PAD_PLL_IDDQ_EN BIT(22) #define T_SATA0_NVOOB 0x114 -#define T_SATA0_NVOOB_COMMA_CNT_MASK (0xff << 16) -#define T_SATA0_NVOOB_COMMA_CNT (0x07 << 16) #define T_SATA0_NVOOB_SQUELCH_FILTER_MODE_MASK (0x3 << 24) #define T_SATA0_NVOOB_SQUELCH_FILTER_MODE (0x1 << 24) #define T_SATA0_NVOOB_SQUELCH_FILTER_LENGTH_MASK (0x3 << 26) @@ -154,11 +152,18 @@ struct tegra_ahci_ops { int (*init)(struct ahci_host_priv *hpriv); }; +struct tegra_ahci_regs { + unsigned int nvoob_comma_cnt_mask; + unsigned int nvoob_comma_cnt_val; +}; + struct tegra_ahci_soc { const char *const *supply_names; u32 num_supplies; bool supports_devslp; + bool has_sata_oob_rst; const struct tegra_ahci_ops *ops; + const struct tegra_ahci_regs *regs; }; struct tegra_ahci_priv { @@ -240,11 +245,13 @@ static int tegra_ahci_power_on(struct ahci_host_priv *hpriv) if (ret) return ret; - ret = tegra_powergate_sequence_power_up(TEGRA_POWERGATE_SATA, - tegra->sata_clk, - tegra->sata_rst); - if (ret) - goto disable_regulators; + if (!tegra->pdev->dev.pm_domain) { + ret = tegra_powergate_sequence_power_up(TEGRA_POWERGATE_SATA, + tegra->sata_clk, + tegra->sata_rst); + if (ret) + goto disable_regulators; + } reset_control_assert(tegra->sata_oob_rst); reset_control_assert(tegra->sata_cold_rst); @@ -330,10 +337,10 @@ static int tegra_ahci_controller_init(struct ahci_host_priv *hpriv) writel(val, tegra->sata_regs + SCFG_OFFSET + T_SATA_CFG_PHY_0); val = readl(tegra->sata_regs + SCFG_OFFSET + T_SATA0_NVOOB); - val &= ~(T_SATA0_NVOOB_COMMA_CNT_MASK | + val &= ~(tegra->soc->regs->nvoob_comma_cnt_mask | T_SATA0_NVOOB_SQUELCH_FILTER_LENGTH_MASK | T_SATA0_NVOOB_SQUELCH_FILTER_MODE_MASK); - val |= (T_SATA0_NVOOB_COMMA_CNT | + val |= (tegra->soc->regs->nvoob_comma_cnt_val | T_SATA0_NVOOB_SQUELCH_FILTER_LENGTH | T_SATA0_NVOOB_SQUELCH_FILTER_MODE); writel(val, tegra->sata_regs + SCFG_OFFSET + T_SATA0_NVOOB); @@ -449,15 +456,35 @@ static const struct tegra_ahci_ops tegra124_ahci_ops = { .init = tegra124_ahci_init, }; +static const struct tegra_ahci_regs tegra124_ahci_regs = { + .nvoob_comma_cnt_mask = GENMASK(30, 28), + .nvoob_comma_cnt_val = (7 << 28), +}; + static const struct tegra_ahci_soc tegra124_ahci_soc = { .supply_names = tegra124_supply_names, .num_supplies = ARRAY_SIZE(tegra124_supply_names), .supports_devslp = false, + .has_sata_oob_rst = true, .ops = &tegra124_ahci_ops, + .regs = &tegra124_ahci_regs, }; static const struct tegra_ahci_soc tegra210_ahci_soc = { .supports_devslp = false, + .has_sata_oob_rst = true, + .regs = &tegra124_ahci_regs, +}; + +static const struct tegra_ahci_regs tegra186_ahci_regs = { + .nvoob_comma_cnt_mask = GENMASK(23, 16), + .nvoob_comma_cnt_val = (7 << 16), +}; + +static const struct tegra_ahci_soc tegra186_ahci_soc = { + .supports_devslp = false, + .has_sata_oob_rst = false, + .regs = &tegra186_ahci_regs, }; static const struct of_device_id tegra_ahci_of_match[] = { @@ -469,6 +496,10 @@ static const struct of_device_id tegra_ahci_of_match[] = { .compatible = "nvidia,tegra210-ahci", .data = &tegra210_ahci_soc }, + { + .compatible = "nvidia,tegra186-ahci", + .data = &tegra186_ahci_soc + }, {} }; MODULE_DEVICE_TABLE(of, tegra_ahci_of_match); @@ -518,10 +549,13 @@ static int tegra_ahci_probe(struct platform_device *pdev) return PTR_ERR(tegra->sata_rst); } - tegra->sata_oob_rst = devm_reset_control_get(&pdev->dev, "sata-oob"); - if (IS_ERR(tegra->sata_oob_rst)) { - dev_err(&pdev->dev, "Failed to get sata-oob reset\n"); - return PTR_ERR(tegra->sata_oob_rst); + if (tegra->soc->has_sata_oob_rst) { + tegra->sata_oob_rst = devm_reset_control_get(&pdev->dev, + "sata-oob"); + if (IS_ERR(tegra->sata_oob_rst)) { + dev_err(&pdev->dev, "Failed to get sata-oob reset\n"); + return PTR_ERR(tegra->sata_oob_rst); + } } tegra->sata_cold_rst = devm_reset_control_get(&pdev->dev, "sata-cold");
This patch adds support for AHCI-compliant Serial ATA controller on Tegra186 SoC. Tegra186 does not have sata-oob reset. Tegra186 SATA_NVOOB register filed COMMA_CNT position and width are different compared to Tegra210 and prior. So, this patch adds a flag has_sata_oob_rst and tegra_ahci_regs to SoC specific strcuture tegra_ahci_soc and updated their implementation accordingly. Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com> --- drivers/ata/ahci_tegra.c | 60 +++++++++++++++++++++++++++++++++++++----------- 1 file changed, 47 insertions(+), 13 deletions(-)