| Message ID | 053529c206a2afc453b52e2c0f024acd85341a83.1497364762.git.mirq-linux@rere.qmqm.pl (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, Jun 13, 2017 at 04:41:58PM +0200, Michał Mirosław wrote: > This allows for (acyclic) references from tps6591x supplies to > its outputs. This is clearly not something that should be open coded in individual drivers, aside from the code duplication it is obviously possible to have two different chips supplying each other which this wouldn't help at all. Is this happening for you with current kernels, we have a few mechanisms for deferring bindings of supplies which should help here.
On Tue, Jun 13, 2017 at 06:46:28PM +0100, Mark Brown wrote: > On Tue, Jun 13, 2017 at 04:41:58PM +0200, Michał Mirosław wrote: > > This allows for (acyclic) references from tps6591x supplies to > > its outputs. > This is clearly not something that should be open coded in individual > drivers, aside from the code duplication it is obviously possible to > have two different chips supplying each other which this wouldn't help > at all. Is this happening for you with current kernels, we have a few > mechanisms for deferring bindings of supplies which should help here. What mechanisms would that be? What I observed is that whenever a regulator's supply is not available but described in device-tree, the whole device's registration is deferred. The device-tree I'm working on contains: pmic: tps65911@2d { compatible = "ti,tps65911"; reg = <0x2d>; ... vcc3-supply = <&vio_reg>; // ldo6, ldo7, ldo8 ... regulators { vio_reg: vio { regulator-min-microvolt = <1800000>; regulator-max-microvolt = <1800000>; regulator-always-on; }; ... } } Best Regards, Michał Mirosław -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, Jun 13, 2017 at 10:22:17PM +0200, Michał Mirosław wrote: > On Tue, Jun 13, 2017 at 06:46:28PM +0100, Mark Brown wrote: > > at all. Is this happening for you with current kernels, we have a few > > mechanisms for deferring bindings of supplies which should help here. > What mechanisms would that be? What I observed is that whenever > a regulator's supply is not available but described in device-tree, > the whole device's registration is deferred. Mainly "Don't use regulators as supplies until the parent is bound", there were some older ones IIRC.
Hello Michal, What kernel version are you using? It shouldn't be necessary anymore to resolve regulators dependencies during regulator_register(), now that logic to lookup the parent supply has been moved to regulator_get() instead. On Tue, Jun 13, 2017 at 10:22 PM, Michał Mirosław <mirq-linux@rere.qmqm.pl> wrote: > On Tue, Jun 13, 2017 at 06:46:28PM +0100, Mark Brown wrote: >> On Tue, Jun 13, 2017 at 04:41:58PM +0200, Michał Mirosław wrote: >> > This allows for (acyclic) references from tps6591x supplies to >> > its outputs. >> This is clearly not something that should be open coded in individual >> drivers, aside from the code duplication it is obviously possible to >> have two different chips supplying each other which this wouldn't help >> at all. Is this happening for you with current kernels, we have a few >> mechanisms for deferring bindings of supplies which should help here. > > What mechanisms would that be? What I observed is that whenever The specific commit that adds the mechanism Mark is referring to is: commit 6261b06de565baafa590e58a531a1a5522cea0b6 Author: Bjorn Andersson <bjorn.andersson@sonymobile.com> Date: Tue Mar 24 18:56:05 2015 -0700 regulator: Defer lookup of supply to regulator_get Instead of resolving regulator supplies during registration move this to the time of a consumer retrieving a handle. The benefit is that it's possible for one driver to register regulators with internal dependencies out of order. Signed-off-by: Bjorn Andersson <bjorn.andersson@sonymobile.com> Signed-off-by: Mark Brown <broonie@kernel.org> > a regulator's supply is not available but described in device-tree, What do you mean by not available? At some point it the parent supply should be registered with the regulator core and prevent the client device to defer its probe when calling regulator_get(). Now, if you don't have a client device that calls regulator_get() to resolve the parent supply but still need the parent supply to be enabled, this commit does in the core what you attempt to do in your driver: commit 5e3ca2b349b1e2c80b060b51bbf2af37448fad85 Author: Javier Martinez Canillas <javier@osg.samsung.com> Date: Wed Mar 23 20:59:34 2016 -0300 regulator: Try to resolve regulators supplies on registration Commit 6261b06de565 ("regulator: Defer lookup of supply to regulator_get") moved the regulator supplies lookup logic from the regulators registration to the regulators get time. Unfortunately, that changed the behavior of the regulator core since now a parent supply with a child regulator marked as always-on, won't be enabled unless a client driver attempts to get the child regulator during boot. This patch tries to resolve the parent supply for the already registered regulators each time that a new regulator is registered. So the regulators that have child regulators marked as always on will be enabled regardless if a driver gets the child regulator or not. That was the behavior before the mentioned commit, since parent supplies were looked up at regulator registration time instead of during child get. Since regulator_resolve_supply() checks for rdev->supply, most of the times it will be a no-op. Errors aren't checked to keep the possible out of order dependencies which was the motivation for the mentioned commit. Also, the supply being available will be enforced on regulator get anyways in case the resolve fails on regulators registration. Fixes: 6261b06de565 ("regulator: Defer lookup of supply to regulator_get") Suggested-by: Mark Brown <broonie@kernel.org> Signed-off-by: Javier Martinez Canillas <javier@osg.samsung.com> Signed-off-by: Mark Brown <broonie@kernel.org> Cc: <stable@vger.kernel.org> # 4.1+ > the whole device's registration is deferred. > > The device-tree I'm working on contains: > > pmic: tps65911@2d { > compatible = "ti,tps65911"; > reg = <0x2d>; > ... > vcc3-supply = <&vio_reg>; // ldo6, ldo7, ldo8 > ... > regulators { > vio_reg: vio { > regulator-min-microvolt = <1800000>; > regulator-max-microvolt = <1800000>; > regulator-always-on; > }; > ... > } > } I don't see anything wrong with this DTS, it should work.... I did something similar for the max77802 PMIC in arch/arm/boot/dts/exynos5800-peach-pi.dts, some regulators have as input supplies regulators from the same PMIC and may not be necessarily registered in the order to resolve parent supplies on child registration. > > Best Regards, > Michał Mirosław Best regards, Javier -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, Jun 14, 2017 at 09:56:37AM +0200, Javier Martinez Canillas wrote: > Hello Michal, > > What kernel version are you using? This was on v4.10 and then v4.11. Both have commits you referred to. I'll have to recreate the configuration where this patch was needed, so let's drop it for now. Best Regards, Michał Mirosław -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/regulator/tps65910-regulator.c b/drivers/regulator/tps65910-regulator.c index 81672a58fcc23..f7987168ba4ed 100644 --- a/drivers/regulator/tps65910-regulator.c +++ b/drivers/regulator/tps65910-regulator.c @@ -1082,7 +1082,8 @@ static int tps65910_probe(struct platform_device *pdev) struct tps65910_reg *pmic; struct tps65910_board *pmic_plat_data; struct of_regulator_match *tps65910_reg_matches = NULL; - int i, err; + bool made_progress; + int i, err, pending; pmic_plat_data = dev_get_platdata(tps65910->dev); if (!pmic_plat_data && tps65910->dev->of_node) @@ -1146,6 +1147,7 @@ static int tps65910_probe(struct platform_device *pdev) if (!pmic->rdev) return -ENOMEM; + /* prepare regulator description */ for (i = 0; i < pmic->num_regulators; i++, info++) { /* Register the regulators */ pmic->info[i] = info; @@ -1196,18 +1198,32 @@ static int tps65910_probe(struct platform_device *pdev) pmic->desc[i].owner = THIS_MODULE; pmic->desc[i].enable_reg = pmic->get_ctrl_reg(i); pmic->desc[i].enable_mask = TPS65910_SUPPLY_STATE_ENABLED; + } + + /* Register regulators - loop, as one regulator's output + * might be anothers input */ + + config.dev = pmic->mfd->dev; + config.driver_data = pmic; + config.regmap = pmic->mfd->regmap; + pending = pmic->num_regulators; + +reg_retry: + made_progress = false; + for (i = 0; i < pmic->num_regulators; i++) { + if (pmic->rdev[i]) + continue; - config.dev = tps65910->dev; config.init_data = pmic_plat_data->tps65910_pmic_init_data[i]; - config.driver_data = pmic; - config.regmap = tps65910->regmap; - if (tps65910_reg_matches) config.of_node = tps65910_reg_matches[i].of_node; rdev = devm_regulator_register(&pdev->dev, &pmic->desc[i], &config); if (IS_ERR(rdev)) { + if (PTR_ERR(rdev) == -EPROBE_DEFER) + continue; + dev_err(tps65910->dev, "failed to register %s regulator\n", pdev->name); @@ -1216,8 +1232,19 @@ static int tps65910_probe(struct platform_device *pdev) /* Save regulator for cleanup */ pmic->rdev[i] = rdev; + + made_progress = true; + --pending; } - return 0; + + /* all done? */ + if (!pending) + return 0; + + if (made_progress) + goto reg_retry; + + return -EPROBE_DEFER; } static void tps65910_shutdown(struct platform_device *pdev)
This allows for (acyclic) references from tps6591x supplies to its outputs. Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl> --- v2: - no changes drivers/regulator/tps65910-regulator.c | 39 ++++++++++++++++++++++++++++------ 1 file changed, 33 insertions(+), 6 deletions(-)