| Message ID | 20240621122503.10034-2-hdegoede@redhat.com |
|---|---|
| State | New |
| Headers | show |
| Series | i2c-i801 / dell-smo8800: Move instantiation of lis3lv02d i2c_client from i2c-i801 to dell-smo8800 | expand |
On Fri, Jun 21, 2024 at 2:25 PM Hans de Goede <hdegoede@redhat.com> wrote: > > Platform glue code, which is not build into the kernel and thus cannot > use i2c_register_board_info() may want to use bus_register_notifier() > to listen for i2c-adapters to show up on which the platform code needs > to manually instantiate platform specific i2c_clients. > > This results in calling i2c_new_client_device() from the bus notifier > which happens near the device_add() call. > > If the i2c-core has not yet setup runtime-pm (specifically the > no-callbacks and ignore-children flags) for the device embedded > inside struct i2c_adapter and the driver for the i2c_client > calls pm_runtime_set_active() this will trigger the following > error inside __pm_runtime_set_status(): > > "runtime PM trying to activate child device %s but parent (%s) is not active\n" > > and the i2c_client's runtime-status will not be updated. > > Split the device_register() call for the adapter into device_initialize() > and device_add() and move the pm-runtime init calls inbetween these 2 calls > so that the runtime-status can be correctly set when a driver binds from > the bus-notifier. > > Note the moved pm-runtime init calls just override the initial value of > some flags in struct device set by device_initialize() and calling these > before device_add() is safe. LGTM, but this is non-trivial (I think) for the usual cases, can you add a comment on top of the code you modified? ... ...somewhere here? > + res = device_add(&adap->dev); > if (res) { > pr_err("adapter '%s': can't register device (%d)\n", adap->name, res); > goto out_list;
diff --git a/drivers/i2c/i2c-core-base.c b/drivers/i2c/i2c-core-base.c index db0d1ac82910..fcb4696f1f93 100644 --- a/drivers/i2c/i2c-core-base.c +++ b/drivers/i2c/i2c-core-base.c @@ -1521,7 +1521,13 @@ static int i2c_register_adapter(struct i2c_adapter *adap) dev_set_name(&adap->dev, "i2c-%d", adap->nr); adap->dev.bus = &i2c_bus_type; adap->dev.type = &i2c_adapter_type; - res = device_register(&adap->dev); + device_initialize(&adap->dev); + device_enable_async_suspend(&adap->dev); + pm_runtime_no_callbacks(&adap->dev); + pm_suspend_ignore_children(&adap->dev, true); + pm_runtime_enable(&adap->dev); + + res = device_add(&adap->dev); if (res) { pr_err("adapter '%s': can't register device (%d)\n", adap->name, res); goto out_list; @@ -1533,11 +1539,6 @@ static int i2c_register_adapter(struct i2c_adapter *adap) if (res) goto out_reg; - device_enable_async_suspend(&adap->dev); - pm_runtime_no_callbacks(&adap->dev); - pm_suspend_ignore_children(&adap->dev, true); - pm_runtime_enable(&adap->dev); - res = i2c_init_recovery(adap); if (res == -EPROBE_DEFER) goto out_reg;
Platform glue code, which is not build into the kernel and thus cannot use i2c_register_board_info() may want to use bus_register_notifier() to listen for i2c-adapters to show up on which the platform code needs to manually instantiate platform specific i2c_clients. This results in calling i2c_new_client_device() from the bus notifier which happens near the device_add() call. If the i2c-core has not yet setup runtime-pm (specifically the no-callbacks and ignore-children flags) for the device embedded inside struct i2c_adapter and the driver for the i2c_client calls pm_runtime_set_active() this will trigger the following error inside __pm_runtime_set_status(): "runtime PM trying to activate child device %s but parent (%s) is not active\n" and the i2c_client's runtime-status will not be updated. Split the device_register() call for the adapter into device_initialize() and device_add() and move the pm-runtime init calls inbetween these 2 calls so that the runtime-status can be correctly set when a driver binds from the bus-notifier. Note the moved pm-runtime init calls just override the initial value of some flags in struct device set by device_initialize() and calling these before device_add() is safe. Signed-off-by: Hans de Goede <hdegoede@redhat.com> --- drivers/i2c/i2c-core-base.c | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-)