Make deferring probe forever optional

This series came out of a discussion on the ARM boot-architecture list[1] about DT forwards and backwards compatibility issues. There are issues with newer DTs breaking on older, stable kernels. Some of these are difficult to solve, but cases of optional devices not having kernel support should be solvable.

I tested this on a RPi3 B with the pinctrl driver forced off. With this change, the MMC/SD and UART drivers can function without the pinctrl driver. I left the dts change out this time.

v2 and v3 of this series can be found here[2][3].

Rob

[1] https://lists.linaro.org/pipermail/boot-architecture/2018-April/000466.html
[2] https://lore.kernel.org/patchwork/project/lkml/list/?series=347413
[3] https://lore.kernel.org/patchwork/project/lkml/list/?series=357344

Rob Herring (6):
driver core: allow stopping deferred probe after init
dt-bindings: pinctrl: add a 'pinctrl-use-default' property
pinctrl: Support stopping deferred probe after initcalls
iommu: Stop deferring probe at end of initcalls
iommu: Remove IOMMU_OF_DECLARE
PM / Domains: Stop deferring probe at the end of initcall

 .../admin-guide/kernel-parameters.txt         |  9 +++
 .../bindings/pinctrl/pinctrl-bindings.txt     |  6 ++
 drivers/base/dd.c                             | 59 +++++++++++++++++++
 drivers/base/power/domain.c                   |  2 +-
 drivers/iommu/arm-smmu-v3.c                   |  2 -
 drivers/iommu/arm-smmu.c                      |  7 ---
 drivers/iommu/exynos-iommu.c                  |  2 -
 drivers/iommu/ipmmu-vmsa.c                    |  3 -
 drivers/iommu/msm_iommu.c                     |  2 -
 drivers/iommu/of_iommu.c                      | 21 +------
 drivers/iommu/qcom_iommu.c                    |  2 -
 drivers/iommu/rockchip-iommu.c                |  2 -
 drivers/pinctrl/devicetree.c                  | 15 +++--
 include/asm-generic/vmlinux.lds.h             |  2 -
 include/linux/device.h                        |  2 +
 include/linux/of_iommu.h                      |  4 --
 16 files changed, 90 insertions(+), 50 deletions(-)

--
2.17.1

Deferred probe will currently wait forever on dependent devices to probe, but sometimes a driver will never exist. It's also not always critical for a driver to exist. Platforms can rely on default configuration from the bootloader or reset defaults for things such as pinctrl and power domains. This is often the case with initial platform support until various drivers get enabled. There's at least 2 scenarios where deferred probe can render a platform broken. Both involve using a DT which has more devices and dependencies than the kernel supports. The 1st case is a driver may be disabled in the kernel config. The 2nd case is the kernel version may simply not have the dependent driver. This can happen if using a newer DT (provided by firmware perhaps) with a stable kernel version. Deferred probe issues can be difficult to debug especially if the console has dependencies or userspace fails to boot to a shell.

There are also cases like IOMMUs where only built-in drivers are supported, so deferring probe after initcalls is not needed. The IOMMU subsystem implemented its own mechanism to handle this using OF_DECLARE linker sections.

This commit adds makes ending deferred probe conditional on initcalls being completed or a debug timeout. Subsystems or drivers may opt-in by calling driver_deferred_probe_check_init_done() instead of unconditionally returning -EPROBE_DEFER. They may use additional information from DT or kernel's config to decide whether to continue to defer probe or not.

The timeout mechanism is intended for debug purposes and WARNs loudly. The remaining deferred probe pending list will also be dumped after the timeout. Not that this timeout won't work for the console which needs to be enabled before userspace starts. However, if the console's dependencies are resolved, then the kernel log will be printed (as opposed to no output).

---
v4:
- Rebase on driver-core-next
- Only allow base 10 for timeout

v3:
- Merged with timeout patch.
- Clarify that deferred_probe_timeout is a debug option.
- Drop the 'optional' param. The only user was pinctrl, so it has to handle that functionality.
- Rename function to driver_deferred_probe_check_state
- Added kerneldoc for driver_deferred_probe_check_state
- Print a 1 line warning if stopping deferred probe after initcalls and a WARN on timeout.

 .../admin-guide/kernel-parameters.txt         |  9 +++
 drivers/base/dd.c                             | 59 +++++++++++++++++++
 include/linux/device.h                        |  2 +
 3 files changed, 70 insertions(+)

index efc7aa7a0670..e83ef4648ea4 100644
@@ -804,6 +804,15 @@
 			Defaults to the default architecture's huge page size
 			if not specified.
 
+	deferred_probe_timeout=
+			[KNL] Debugging option to set a timeout in seconds for
+			deferred probe to give up waiting on dependencies to
+			probe. Only specific dependencies (subsystems or
+			drivers) that have opted in will be ignored. A timeout of 0
+			will timeout at the end of initcalls. This option will also
+			dump out devices still on the deferred probe list after
+			retrying.
+
 	dhash_entries=	[KNL]
 			Set number of hash buckets for dentry cache.
 

index 575c5a35ece5..d2acc78d279b 100644
@@ -339,6 +339,8 @@ struct device *driver_find_device(struct device_driver *drv,
 				  struct device *start, void *data,
 				  int (*match)(struct device *dev, void *data));
 
+int driver_deferred_probe_check_state(struct device *dev);
+
 /**
  * struct subsys_interface - interfaces to device functions
  * @name:       name of the device function

--
2.17.1

So what happens if we have a set of modules which use deferred probing in order to work?

For example, with sound stuff built as modules, and auto-loaded in parallel by udev, the modules get added in a random order. The modules have non-udev obvious dependencies between them (resource dependencies) which result in deferred probing being necessary to bring the device up.

Eg,

snd_soc_kirkwood_spdif module declares the ASoC card. snd_soc_spdif_tx is a codec as a loadable module. snd_soc_kirkwood is the CPU digital audio interface module.

What I commonly see is this module load order:

snd_soc_kirkwood_spdif, then snd_soc_kirkwood and then snd_soc_spdif_tx.

This results at boot in:

kirkwood-spdif-audio audio-subsystem: ASoC: CPU DAI kirkwood-fe not registered kirkwood-spdif-audio audio-subsystem: ASoC: CPU DAI kirkwood-fe not registered kirkwood-spdif-audio audio-subsystem: ASoC: CPU DAI kirkwood-fe not registered kirkwood-spdif-audio audio-subsystem: ASoC: CPU DAI kirkwood-fe not registered kirkwood-spdif-audio audio-subsystem: ASoC: CPU DAI kirkwood-fe not registered kirkwood-spdif-audio audio-subsystem: ASoC: CODEC DAI dit-hifi not registered kirkwood-spdif-audio audio-subsystem: ASoC: CODEC DAI dit-hifi not registered kirkwood-spdif-audio audio-subsystem: snd-soc-dummy-dai <-> kirkwood-fe mapping ok kirkwood-spdif-audio audio-subsystem: multicodec <-> kirkwood-spdif mapping ok

at boot, where most of these are deferred probe attempts.

So, disabling deferred probing after all the kernel-internal initcalls are run is wrong. You can have deferred probing required due to external modules, and this can kick in at any time (think about hot-pluggable hardware with a driver that's somehow componentised, like an audio device...)

RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 13.8Mbps down 630kbps up
According to speedtest.net: 13Mbps down 490kbps up

It is opt-in by subsystem or drivers and mainly intended for subsystems which can be optional or only support built-in drivers. However, I don't really envision many other users other than the ones I converted (pinctrl, iommu, pm-domains). If you look at patch 3, you'll see it is dependent on !CONFIG_MODULES.

For the timeout, well, that's for debugging only. If you get to the point of loading sound modules, you probably don't need the timeout. It's for debugging not booting.

Pin setup may be optional in some cases such as the reset default works or the pin setup is done by the bootloader. In these cases, it is optional for the OS to support managing the pin controller and pin setup. In order to support this scenario, add a property 'pinctrl-use-default' to indicate that the pin configuration is optional.

---
 .../devicetree/bindings/pinctrl/pinctrl-bindings.txt        | 6 ++++++
 1 file changed, 6 insertions(+)

index ad9bbbba36e9..cef2b5855d60 100644
@@ -103,6 +103,12 @@ Optional properties:
 #pinctrl-cells:	Number of pin control cells in addition to the index within the
 		pin controller device instance
 
+pinctrl-use-default: Boolean. Indicates that the OS can use the boot default
+		pin configuration. This allows using an OS that does not have a
+		driver for the pin controller. This property can be set either
+		globally for the pin controller or in child nodes for individual
+		pin group control.
+
 Pin controller devices should contain the pin configuration nodes that client
 devices reference.
 

2.17.1

Pinctrl drivers are a common dependency which can prevent a system booting even if the default or bootloader configured settings can work. If a pinctrl node in DT indicates that the default pin setup can be used with the 'pinctrl-use-default' property, then only defer probe until initcalls are done. If the deferred probe timeout is enabled or loadable modules are disabled, then we'll stop deferring probe regardless of the DT property. This gives platforms the option to work without their pinctrl driver being enabled.

Dropped the pinctrl specific deferring probe message as the driver core can print deferred probe related messages if needed.

---
v4:
- Add Linus' R-by.

v3:
- Drop pinctrl deferred probe msg in favor of driver core messages
- Move the handling of "pinctrl-use-default" option out of driver core
- Stop deferring probe if modules are not enabled.

 drivers/pinctrl/devicetree.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

index c4aa411f5935..2969ff3162c3 100644
@@ -111,17 +111,24 @@ static int dt_to_map_one_config(struct pinctrl *p,
 	int ret;
 	struct pinctrl_map *map;
 	unsigned num_maps;
+	bool allow_default = false;
 
 	/* Find the pin controller containing np_config */
 	np_pctldev = of_node_get(np_config);
 	for (;;) {
+		if (!allow_default)
+			allow_default = of_property_read_bool(np_pctldev,
+							      "pinctrl-use-default");
+
 		np_pctldev = of_get_next_parent(np_pctldev);
 		if (!np_pctldev || of_node_is_root(np_pctldev)) {
-			dev_info(p->dev, "could not find pctldev for node %pOF, deferring probe\n",
-				np_config);
 			of_node_put(np_pctldev);
-			/* OK let's just assume this will appear later then */
-			return -EPROBE_DEFER;
+			ret = driver_deferred_probe_check_state(p->dev);
+			/* keep deferring if modules are enabled unless we've timed out */
+			if (IS_ENABLED(CONFIG_MODULES) && !allow_default && ret == -ENODEV)
+				ret = -EPROBE_DEFER;
+
+			return ret;
 		}
 		/* If we're creating a hog we can use the passed pctldev */
 		if (hog_pctldev && (np_pctldev == p->dev->of_node)) {

--
2.17.1

The IOMMU subsystem has its own mechanism to not defer probe if driver support is missing. Now that the driver core supports stopping deferring probe if drivers aren't built-in (and probed), use the driver core support so the IOMMU specific support can be removed.

---
 drivers/iommu/of_iommu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

index 5c36a8b7656a..78ddf47dd67a 100644
@@ -133,7 +133,7 @@ static int of_iommu_xlate(struct device *dev,
 	 * a proper probe-ordering dependency mechanism in future.
 	 */
 	if (!ops)
-		return -EPROBE_DEFER;
+		return driver_deferred_probe_check_state(dev);
 
 	return ops->of_xlate(dev, iommu_spec);
 }

2.17.1

Now that we use the driver core to stop deferred probe for missing drivers, IOMMU_OF_DECLARE can be removed.

This is slightly less optimal than having a list of built-in drivers in that we'll now defer probe twice before giving up. This shouldn't have a significant impact on boot times as past discussions about deferred probe have given no evidence of deferred probe having a substantial impact.

---
 drivers/iommu/arm-smmu-v3.c       |  2 --
 drivers/iommu/arm-smmu.c          |  7 -------
 drivers/iommu/exynos-iommu.c      |  2 --
 drivers/iommu/ipmmu-vmsa.c        |  3 ---
 drivers/iommu/msm_iommu.c         |  2 --
 drivers/iommu/of_iommu.c          | 19 +------------------
 drivers/iommu/qcom_iommu.c        |  2 --
 drivers/iommu/rockchip-iommu.c    |  2 --
 include/asm-generic/vmlinux.lds.h |  2 --
 include/linux/of_iommu.h          |  4 ----
 10 files changed, 1 insertion(+), 44 deletions(-)

index 1d647104bccc..22bdabd3d8e0 100644
@@ -2915,8 +2915,6 @@ static struct platform_driver arm_smmu_driver = {
 };
 module_platform_driver(arm_smmu_driver);
 
-IOMMU_OF_DECLARE(arm_smmuv3, "arm,smmu-v3");
-
 MODULE_DESCRIPTION("IOMMU API for ARM architected SMMUv3 implementations");
 MODULE_AUTHOR("Will Deacon <[email protected]>");
 MODULE_LICENSE("GPL v2");

index f7a96bcf94a6..c73cfce1ccc0 100644
@@ -2211,13 +2211,6 @@ static struct platform_driver arm_smmu_driver = {
 };
 module_platform_driver(arm_smmu_driver);
 
-IOMMU_OF_DECLARE(arm_smmuv1, "arm,smmu-v1");
-IOMMU_OF_DECLARE(arm_smmuv2, "arm,smmu-v2");
-IOMMU_OF_DECLARE(arm_mmu400, "arm,mmu-400");
-IOMMU_OF_DECLARE(arm_mmu401, "arm,mmu-401");
-IOMMU_OF_DECLARE(arm_mmu500, "arm,mmu-500");
-IOMMU_OF_DECLARE(cavium_smmuv2, "cavium,smmu-v2");
-
 MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations");
 MODULE_AUTHOR("Will Deacon <[email protected]>");
 MODULE_LICENSE("GPL v2");

index 85879cfec52f..b128cb4372d3 100644
@@ -1390,5 +1390,3 @@ static int __init exynos_iommu_init(void)
 	return ret;
 }
 core_initcall(exynos_iommu_init);
-
-IOMMU_OF_DECLARE(exynos_iommu_of, "samsung,exynos-sysmmu");

index 40ae6e87cb88..f026aa16d5f1 100644
@@ -1108,9 +1108,6 @@ static void __exit ipmmu_exit(void)
 subsys_initcall(ipmmu_init);
 module_exit(ipmmu_exit);
 
-IOMMU_OF_DECLARE(ipmmu_vmsa_iommu_of, "renesas,ipmmu-vmsa");
-IOMMU_OF_DECLARE(ipmmu_r8a7795_iommu_of, "renesas,ipmmu-r8a7795");
-
 MODULE_DESCRIPTION("IOMMU API for Renesas VMSA-compatible IPMMU");
 MODULE_AUTHOR("Laurent Pinchart <[email protected]>");
 MODULE_LICENSE("GPL v2");

index 0d3350463a3f..27377742600d 100644
@@ -877,7 +877,5 @@ static void __exit msm_iommu_driver_exit(void)
 subsys_initcall(msm_iommu_driver_init);
 module_exit(msm_iommu_driver_exit);
 
-IOMMU_OF_DECLARE(msm_iommu_of, "qcom,apq8064-iommu");
-
 MODULE_LICENSE("GPL v2");
 MODULE_AUTHOR("Stepan Moskovchenko <[email protected]>");

index 78ddf47dd67a..f7787e757244 100644
@@ -27,9 +27,6 @@
 
 #define NO_IOMMU	1
 
-static const struct of_device_id __iommu_of_table_sentinel
-	__used __section(__iommu_of_table_end);
-
 /**
  * of_get_dma_window - Parse *dma-window property and returns 0 if found.
  *

@@ -98,19 +95,6 @@ int of_get_dma_window(struct device_node *dn, const char *prefix, int index,
 }
 EXPORT_SYMBOL_GPL(of_get_dma_window);
 
-static bool of_iommu_driver_present(struct device_node *np)
-{
-	/*
-	 * If the IOMMU still isn't ready by the time we reach init, assume
-	 * it never will be. We don't want to defer indefinitely, nor attempt
-	 * to dereference __iommu_of_table after it's been freed.
-	 */
-	if (system_state >= SYSTEM_RUNNING)
-		return false;
-
-	return of_match_node(&__iommu_of_table, np);
-}
-
 static int of_iommu_xlate(struct device *dev,
 			  struct of_phandle_args *iommu_spec)
 {

@@ -120,8 +104,7 @@ static int of_iommu_xlate(struct device *dev,
 
 	ops = iommu_ops_from_fwnode(fwnode);
 	if ((ops && !ops->of_xlate) ||
-	    !of_device_is_available(iommu_spec->np) ||
-	    (!ops && !of_iommu_driver_present(iommu_spec->np)))
+	    !of_device_is_available(iommu_spec->np))
 		return NO_IOMMU;
 
 	err = iommu_fwspec_init(dev, &iommu_spec->np->fwnode, ops);

index fe88a4880d3a..b48aee82d14b 100644
@@ -945,7 +945,5 @@ static void __exit qcom_iommu_exit(void)
 module_init(qcom_iommu_init);
 module_exit(qcom_iommu_exit);
 
-IOMMU_OF_DECLARE(qcom_iommu_dev, "qcom,msm-iommu-v1");
-
 MODULE_DESCRIPTION("IOMMU API for QCOM IOMMU v1 implementations");
 MODULE_LICENSE("GPL v2");

index 054cd2c8e9c8..de8d3bf91b23 100644
@@ -1284,8 +1284,6 @@ static int __init rk_iommu_init(void)
 }
 subsys_initcall(rk_iommu_init);
 
-IOMMU_OF_DECLARE(rk_iommu_of, "rockchip,iommu");
-
 MODULE_DESCRIPTION("IOMMU API for Rockchip");
 MODULE_AUTHOR("Simon Xue <[email protected]> and Daniel Kurtz <[email protected]>");
 MODULE_ALIAS("platform:rockchip-iommu");

index e373e2e10f6a..f173b5f30dbe 100644
@@ -218,7 +218,6 @@
 #define TIMER_OF_TABLES()	OF_TABLE(CONFIG_TIMER_OF, timer)
 #define IRQCHIP_OF_MATCH_TABLE() OF_TABLE(CONFIG_IRQCHIP, irqchip)
 #define CLK_OF_TABLES()		OF_TABLE(CONFIG_COMMON_CLK, clk)
-#define IOMMU_OF_TABLES()	OF_TABLE(CONFIG_OF_IOMMU, iommu)
 #define RESERVEDMEM_OF_TABLES()	OF_TABLE(CONFIG_OF_RESERVED_MEM, reservedmem)
 #define CPU_METHOD_OF_TABLES()	OF_TABLE(CONFIG_SMP, cpu_method)
 #define CPUIDLE_METHOD_OF_TABLES() OF_TABLE(CONFIG_CPU_IDLE, cpuidle_method)

@@ -601,7 +600,6 @@
 	CLK_OF_TABLES()							\
 	RESERVEDMEM_OF_TABLES()						\
 	TIMER_OF_TABLES()						\
-	IOMMU_OF_TABLES()						\
 	CPU_METHOD_OF_TABLES()						\
 	CPUIDLE_METHOD_OF_TABLES()					\
 	KERNEL_DTB()							\

index 4fa654e4b5a9..f3d40dd7bb66 100644
@@ -32,8 +32,4 @@ static inline const struct iommu_ops *of_iommu_configure(struct device *dev,
 
 #endif	/* CONFIG_OF_IOMMU */
 
-extern struct of_device_id __iommu_of_table;
-
-#define IOMMU_OF_DECLARE(name, compat)	OF_DECLARE_1(iommu, name, compat, NULL)
-
 #endif /* __OF_IOMMU_H */

2.17.1

All PM domain drivers must be built-in (at least those using DT), so there is no point deferring probe after initcalls are done. Continuing to defer probe may prevent booting successfully even if managing PM domains is not required. This can happen if the user failed to enable the driver or if power-domains are added to a platform's DT, but there is not yet a driver (e.g. a new DTB with an old kernel).

Call the driver core function driver_deferred_probe_check_init_done() instead of just returning -EPROBE_DEFER to stop deferring probe when initcalls are done.

---
v4:
- Add Rafael's ack

v3:
- Update to new function name

 drivers/base/power/domain.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

index 4925af5c4cf0..8c12213875c6 100644
@@ -2253,7 +2253,7 @@ static int __genpd_dev_pm_attach(struct device *dev, struct device_node *np,
 		mutex_unlock(&gpd_list_lock);
 		dev_dbg(dev, "%s() failed to find PM domain: %ld\n",
 			__func__, PTR_ERR(pd));
-		return -EPROBE_DEFER;
+		return driver_deferred_probe_check_state(dev);
 	}
 
 	dev_dbg(dev, "adding to PM domain %s\n", pd->name);

--
2.17.1

Besides the above, I am fine with the approach as such.

Kind regards
Uffe

Yes, but I think this is the only place you can fail to get the PD device from DT. The case we care about is properties exist in the DT, but no driver exists.

What would cause deferring in the latter cases?

Rob

That's correct. Thanks for clarifying!

To power on the PM domain for example, which often is required to be able to probe the device.

Kind regards
Uffe