]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
driver: core: Allow subsystems to continue deferring probe
authorThierry Reding <treding@nvidia.com>
Fri, 21 Jun 2019 15:17:25 +0000 (17:17 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 3 Jul 2019 19:28:20 +0000 (21:28 +0200)
Some subsystems, such as pinctrl, allow continuing to defer probe
indefinitely. This is useful for devices that depend on resources
provided by devices that are only probed after the init stage.

One example of this can be seen on Tegra, where the DPAUX hardware
contains pinmuxing controls for pins that it shares with an I2C
controller. The I2C controller is typically used for communication
with a monitor over HDMI (DDC). However, other instances of the I2C
controller are used to access system critical components, such as a
PMIC. The I2C controller driver will therefore usually be a builtin
driver, whereas the DPAUX driver is part of the display driver that
is loaded from a module to avoid bloating the kernel image with all
of the DRM/KMS subsystem.

In this particular case the pins used by this I2C/DDC controller
become accessible very late in the boot process. However, since the
controller is only used in conjunction with display, that's not an
issue.

Unfortunately the driver core currently outputs a warning message
when a device fails to get the pinctrl before the end of the init
stage. That can be confusing for the user because it may sound like
an unwanted error occurred, whereas it's really an expected and
harmless situation.

In order to eliminate this warning, this patch allows callers of the
driver_deferred_probe_check_state() helper to specify that they want
to continue deferring probe, regardless of whether we're past the
init stage or not. All of the callers of that function are updated
for the new signature, but only the pinctrl subsystem passes a true
value in the new persist parameter if appropriate.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Link: https://lore.kernel.org/r/20190621151725.20414-1-thierry.reding@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/base/dd.c
drivers/pinctrl/devicetree.c
include/linux/device.h

index 0df9b4461766c79499d695b10fa74ef8d6337877..994a9074742046edfed5afefe62011450e20669c 100644 (file)
@@ -235,6 +235,19 @@ static int __init deferred_probe_timeout_setup(char *str)
 }
 __setup("deferred_probe_timeout=", deferred_probe_timeout_setup);
 
+static int __driver_deferred_probe_check_state(struct device *dev)
+{
+       if (!initcalls_done)
+               return -EPROBE_DEFER;
+
+       if (!deferred_probe_timeout) {
+               dev_WARN(dev, "deferred probe timeout, ignoring dependency");
+               return -ETIMEDOUT;
+       }
+
+       return 0;
+}
+
 /**
  * driver_deferred_probe_check_state() - Check deferred probe state
  * @dev: device to check
@@ -248,14 +261,40 @@ __setup("deferred_probe_timeout=", deferred_probe_timeout_setup);
  */
 int driver_deferred_probe_check_state(struct device *dev)
 {
-       if (initcalls_done) {
-               if (!deferred_probe_timeout) {
-                       dev_WARN(dev, "deferred probe timeout, ignoring dependency");
-                       return -ETIMEDOUT;
-               }
-               dev_warn(dev, "ignoring dependency for device, assuming no driver");
-               return -ENODEV;
-       }
+       int ret;
+
+       ret = __driver_deferred_probe_check_state(dev);
+       if (ret < 0)
+               return ret;
+
+       dev_warn(dev, "ignoring dependency for device, assuming no driver");
+
+       return -ENODEV;
+}
+
+/**
+ * driver_deferred_probe_check_state_continue() - check deferred probe state
+ * @dev: device to check
+ *
+ * Returns -ETIMEDOUT if deferred probe debug timeout has expired, or
+ * -EPROBE_DEFER otherwise.
+ *
+ * Drivers or subsystems can opt-in to calling this function instead of
+ * directly returning -EPROBE_DEFER.
+ *
+ * This is similar to driver_deferred_probe_check_state(), but it allows the
+ * subsystem to keep deferring probe after built-in drivers have had a chance
+ * to probe. One scenario where that is useful is if built-in drivers rely on
+ * resources that are provided by modular drivers.
+ */
+int driver_deferred_probe_check_state_continue(struct device *dev)
+{
+       int ret;
+
+       ret = __driver_deferred_probe_check_state(dev);
+       if (ret < 0)
+               return ret;
+
        return -EPROBE_DEFER;
 }
 
index f7e354f85518154e604dc64f919ff70ab8a218fe..88ddbb2e30de10f671cc3d6c144914ee475351be 100644 (file)
@@ -112,12 +112,11 @@ static int dt_to_map_one_config(struct pinctrl *p,
                np_pctldev = of_get_next_parent(np_pctldev);
                if (!np_pctldev || of_node_is_root(np_pctldev)) {
                        of_node_put(np_pctldev);
-                       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;
+                       if (IS_ENABLED(CONFIG_MODULES) && !allow_default)
+                               return driver_deferred_probe_check_state_continue(p->dev);
 
-                       return ret;
+                       return driver_deferred_probe_check_state(p->dev);
                }
                /* If we're creating a hog we can use the passed pctldev */
                if (hog_pctldev && (np_pctldev == p->dev->of_node)) {
index 709308560d32489389ee4858355cc37cde952e6a..ef61e2d50ecce8c755e40eab534b3d52a8ce13b1 100644 (file)
@@ -343,6 +343,7 @@ struct device *driver_find_device(struct device_driver *drv,
 
 void driver_deferred_probe_add(struct device *dev);
 int driver_deferred_probe_check_state(struct device *dev);
+int driver_deferred_probe_check_state_continue(struct device *dev);
 
 /**
  * struct subsys_interface - interfaces to device functions