platform/chrome: Fix a race when probing drivers
Subsystems (5)
Tzung-Bi Shih
maintainercover platform/chrome: Fix a race when probing drivers
A race is observed when cros_ec_lpc and cros-ec-keyb are all built as modules. cros_ec_lpc is cros-ec-keyb's parent. However, they can be probed at the same time.
Example:
+ -----------------------------------------------------------------+ | Some init process (e.g. udevd) | deferred_probe_work_func worker |
+ -----------------------------------------------------------------+
| Probe cros-ec-keyb. | | | - Decide to defer[1]. | | | | A device bound to a driver[2]. | | Probe cros_ec_lpc. | | | - Init the struct[3]. | | | | Retry cros-ec-keyb from the | | | deferred list[4]. | | | - Won't defer again as [3]. | | | - Access uninitialized data in | | | the struct. | | - Register the device. | |
+ -----------------------------------------------------------------+
[1] https://elixir.bootlin.com/linux/v6.16/source/drivers/input/keyboard/cros_ec_keyb.c#L707
[2] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L405
[3] https://elixir.bootlin.com/linux/v6.16/source/drivers/platform/chrome/cros_ec_lpc.c#L644
[4] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L418
Note that the device link[5] can't help as in the observed environment, the devices are already added via device_add()[6].
[5] https://www.kernel.org/doc/html/latest/driver-api/device_link.html#usage
[6] https://elixir.bootlin.com/linux/v6.16/source/drivers/acpi/acpi_platform.c#L177
The series fixes the issue by ensuring the struct is ready for accessing before continuing to probe cros-ec-keyb.
The 1st - 3rd patches are preparatory steps to allow the per-device lock can be used even if the struct cros_ec_device is unregistered.
The 4th patch introduces a new flag registered and use the lock for protecting the flag.
The 5th patch defers probe if the struct cros_ec_device is not ready yet.
Tzung-Bi Shih (5):
platform/chrome: Centralize cros_ec_device allocation platform/chrome: Centralize common cros_ec_device initialization platform/chrome: cros_ec: Separate initialization from
cros_ec_register()
platform/chrome: cros_ec: Add a flag to track registration state
Input: cros_ec_keyb - Defer probe until parent EC device is registered
drivers/input/keyboard/cros_ec_keyb.c | 6 ++ drivers/platform/chrome/cros_ec.c | 85 ++++++++++++++------- drivers/platform/chrome/cros_ec.h | 3 + drivers/platform/chrome/cros_ec_i2c.c | 9 +-- drivers/platform/chrome/cros_ec_ishtp.c | 6 +- drivers/platform/chrome/cros_ec_lpc.c | 6 +- drivers/platform/chrome/cros_ec_proto.c | 15 ++++ drivers/platform/chrome/cros_ec_rpmsg.c | 6 +- drivers/platform/chrome/cros_ec_spi.c | 7 +- drivers/platform/chrome/cros_ec_uart.c | 6 +- include/linux/platform_data/cros_ec_proto.h | 18 ++++- 11 files changed, 103 insertions(+), 64 deletions(-)
Signature · 1 line
2.51.0.268.g9569e192d0-goog
Tzung-Bi Shih
maintainerPATCH 1/5 platform/chrome: Centralize cros_ec_device allocation
Introduce a helper function, cros_ec_device_alloc(), to centralize the allocation of the struct cros_ec_device. Convert all protocol device drivers to use this new function.
This is a preparatory step for separating common initialization logic out of device drivers' probe() and cros_ec_register().
--- drivers/platform/chrome/cros_ec.c | 12 ++++++++++++ drivers/platform/chrome/cros_ec.h | 3 +++ drivers/platform/chrome/cros_ec_i2c.c | 4 ++-- drivers/platform/chrome/cros_ec_ishtp.c | 2 +- drivers/platform/chrome/cros_ec_lpc.c | 2 +- drivers/platform/chrome/cros_ec_rpmsg.c | 2 +- drivers/platform/chrome/cros_ec_spi.c | 2 +- drivers/platform/chrome/cros_ec_uart.c | 2 +- 8 files changed, 22 insertions(+), 7 deletions(-)
drivers/platform/chrome/cros_ec.c +12 -0
@@ -30,6 +30,18 @@ static struct cros_ec_platform pd_p = { .cmd_offset = EC_CMD_PASSTHRU_OFFSET(CROS_EC_DEV_PD_INDEX), }; +struct cros_ec_device *cros_ec_device_alloc (struct device *dev) +{ +struct cros_ec_device *ec_dev; + + ec_dev =devm_kzalloc (dev,sizeof (*ec_dev),GFP_KERNEL ); +if (!ec_dev) +return NULL ; + +return ec_dev; +} +EXPORT_SYMBOL (cros_ec_device_alloc); +/** * cros_ec_irq_handler() - top half part of the interrupt handler * @irq: IRQ id
drivers/platform/chrome/cros_ec.h +3 -0
@@ -11,6 +11,9 @@#include <linux/interrupt.h> struct cros_ec_device ; +struct device ; + +struct cros_ec_device *cros_ec_device_alloc (struct device *dev);int cros_ec_register (struct cros_ec_device *ec_dev);void cros_ec_unregister (struct cros_ec_device *ec_dev);
drivers/platform/chrome/cros_ec_i2c.c +2 -2
@@ -289,10 +289,10 @@ static int cros_ec_cmd_xfer_i2c(struct cros_ec_device *ec_dev,static int cros_ec_i2c_probe (struct i2c_client *client) {struct device *dev = &client->dev; -struct cros_ec_device *ec_dev =NULL ; +struct cros_ec_device *ec_dev;int err; - ec_dev =devm_kzalloc (dev,sizeof (*ec_dev),GFP_KERNEL ); + ec_dev =cros_ec_device_alloc (dev);if (!ec_dev)return -ENOMEM ;
drivers/platform/chrome/cros_ec_ishtp.c +1 -1
@@ -543,7 +543,7 @@ static int cros_ec_dev_init(struct ishtp_cl_data *client_data)struct cros_ec_device *ec_dev;struct device *dev =cl_data_to_dev (client_data); - ec_dev =devm_kzalloc (dev,sizeof (*ec_dev),GFP_KERNEL ); + ec_dev =cros_ec_device_alloc (dev);if (!ec_dev)return -ENOMEM ;
drivers/platform/chrome/cros_ec_lpc.c +1 -1
@@ -637,7 +637,7 @@ static int cros_ec_lpc_probe(struct platform_device *pdev) } } - ec_dev =devm_kzalloc (dev,sizeof (*ec_dev),GFP_KERNEL ); + ec_dev =cros_ec_device_alloc (dev);if (!ec_dev)return -ENOMEM ;
drivers/platform/chrome/cros_ec_rpmsg.c +1 -1
@@ -216,7 +216,7 @@ static int cros_ec_rpmsg_probe(struct rpmsg_device *rpdev)struct cros_ec_device *ec_dev;int ret; - ec_dev =devm_kzalloc (dev,sizeof (*ec_dev),GFP_KERNEL ); + ec_dev =cros_ec_device_alloc (dev);if (!ec_dev)return -ENOMEM ;
drivers/platform/chrome/cros_ec_spi.c +1 -1
@@ -749,7 +749,7 @@ static int cros_ec_spi_probe(struct spi_device *spi)if (ec_spi ==NULL )return -ENOMEM ; ec_spi->spi = spi; - ec_dev =devm_kzalloc (dev,sizeof (*ec_dev),GFP_KERNEL ); + ec_dev =cros_ec_device_alloc (dev);if (!ec_dev)return -ENOMEM ;
drivers/platform/chrome/cros_ec_uart.c +1 -1
@@ -259,7 +259,7 @@ static int cros_ec_uart_probe(struct serdev_device *serdev)if (!ec_uart)return -ENOMEM ; - ec_dev =devm_kzalloc (dev,sizeof (*ec_dev),GFP_KERNEL ); + ec_dev =cros_ec_device_alloc (dev);if (!ec_dev)return -ENOMEM ;
Signature · 1 line
2.51.0.268.g9569e192d0-goog
Tzung-Bi Shih
maintainerPATCH 2/5 platform/chrome: Centralize common cros_ec_device initialization
In reply to Tzung-Bi ShihMove the common initialization from protocol device drivers into central cros_ec_device_alloc().
This removes duplicated code from each driver's probe function. The buffer sizes are now calculated once, using the maximum possible overhead required by any of the transport protocols, ensuring the allocated buffers are sufficient for all cases.
--- drivers/platform/chrome/cros_ec.c | 9 +++++++++ drivers/platform/chrome/cros_ec_i2c.c | 5 ----- drivers/platform/chrome/cros_ec_ishtp.c | 4 ---- drivers/platform/chrome/cros_ec_lpc.c | 4 ---- drivers/platform/chrome/cros_ec_rpmsg.c | 4 ---- drivers/platform/chrome/cros_ec_spi.c | 5 ----- drivers/platform/chrome/cros_ec_uart.c | 4 ---- include/linux/platform_data/cros_ec_proto.h | 14 ++++++++++---- 8 files changed, 19 insertions(+), 30 deletions(-)
drivers/platform/chrome/cros_ec.c +9 -0
@@ -38,6 +38,15 @@ struct cros_ec_device *cros_ec_device_alloc(struct device *dev)if (!ec_dev)return NULL ; + ec_dev->din_size =sizeof (struct ec_host_response ) + +sizeof (struct ec_response_get_protocol_info ) + +EC_MAX_RESPONSE_OVERHEAD ; + ec_dev->dout_size =sizeof (struct ec_host_request ) + +sizeof (struct ec_params_rwsig_action ) + +EC_MAX_REQUEST_OVERHEAD ; + + ec_dev->dev = dev; +return ec_dev; }EXPORT_SYMBOL (cros_ec_device_alloc);
drivers/platform/chrome/cros_ec_i2c.c +0 -5
@@ -297,16 +297,11 @@ static int cros_ec_i2c_probe(struct i2c_client *client)return -ENOMEM ;i2c_set_clientdata (client, ec_dev); - ec_dev->dev = dev; ec_dev->priv = client; ec_dev->irq = client->irq; ec_dev->cmd_xfer = cros_ec_cmd_xfer_i2c; ec_dev->pkt_xfer = cros_ec_pkt_xfer_i2c; ec_dev->phys_name = client->adapter->name; - ec_dev->din_size =sizeof (struct ec_host_response_i2c ) + -sizeof (struct ec_response_get_protocol_info ); - ec_dev->dout_size =sizeof (struct ec_host_request_i2c ) + -sizeof (struct ec_params_rwsig_action ); err =cros_ec_register (ec_dev);if (err) {
drivers/platform/chrome/cros_ec_ishtp.c +0 -4
@@ -550,14 +550,10 @@ static int cros_ec_dev_init(struct ishtp_cl_data *client_data) client_data->ec_dev = ec_dev; dev->driver_data = ec_dev; - ec_dev->dev = dev; ec_dev->priv = client_data->cros_ish_cl; ec_dev->cmd_xfer =NULL ; ec_dev->pkt_xfer = cros_ec_pkt_xfer_ish; ec_dev->phys_name =dev_name (dev); - ec_dev->din_size =sizeof (struct cros_ish_in_msg ) + -sizeof (struct ec_response_get_protocol_info ); - ec_dev->dout_size =sizeof (struct cros_ish_out_msg ) +sizeof (struct ec_params_rwsig_action );return cros_ec_register (ec_dev); }
drivers/platform/chrome/cros_ec_lpc.c +0 -4
@@ -642,14 +642,10 @@ static int cros_ec_lpc_probe(struct platform_device *pdev)return -ENOMEM ;platform_set_drvdata (pdev, ec_dev); - ec_dev->dev = dev; ec_dev->phys_name =dev_name (dev); ec_dev->cmd_xfer = cros_ec_cmd_xfer_lpc; ec_dev->pkt_xfer = cros_ec_pkt_xfer_lpc; ec_dev->cmd_readmem = cros_ec_lpc_readmem; - ec_dev->din_size =sizeof (struct ec_host_response ) + -sizeof (struct ec_response_get_protocol_info ); - ec_dev->dout_size =sizeof (struct ec_host_request ) +sizeof (struct ec_params_rwsig_action ); ec_dev->priv = ec_lpc;/*
drivers/platform/chrome/cros_ec_rpmsg.c +0 -4
@@ -224,14 +224,10 @@ static int cros_ec_rpmsg_probe(struct rpmsg_device *rpdev)if (!ec_rpmsg)return -ENOMEM ; - ec_dev->dev = dev; ec_dev->priv = ec_rpmsg; ec_dev->cmd_xfer = cros_ec_cmd_xfer_rpmsg; ec_dev->pkt_xfer = cros_ec_pkt_xfer_rpmsg; ec_dev->phys_name =dev_name (&rpdev->dev); - ec_dev->din_size =sizeof (struct ec_host_response ) + -sizeof (struct ec_response_get_protocol_info ); - ec_dev->dout_size =sizeof (struct ec_host_request ) +sizeof (struct ec_params_rwsig_action );dev_set_drvdata (dev, ec_dev); ec_rpmsg->rpdev = rpdev;
drivers/platform/chrome/cros_ec_spi.c +0 -5
@@ -757,16 +757,11 @@ static int cros_ec_spi_probe(struct spi_device *spi)cros_ec_spi_dt_probe (ec_spi, dev);spi_set_drvdata (spi, ec_dev); - ec_dev->dev = dev; ec_dev->priv = ec_spi; ec_dev->irq = spi->irq; ec_dev->cmd_xfer = cros_ec_cmd_xfer_spi; ec_dev->pkt_xfer = cros_ec_pkt_xfer_spi; ec_dev->phys_name =dev_name (&ec_spi->spi->dev); - ec_dev->din_size =EC_MSG_PREAMBLE_COUNT + -sizeof (struct ec_host_response ) + -sizeof (struct ec_response_get_protocol_info ); - ec_dev->dout_size =sizeof (struct ec_host_request ) +sizeof (struct ec_params_rwsig_action ); ec_spi->last_transfer_ns =ktime_get_ns ();
drivers/platform/chrome/cros_ec_uart.c +0 -4
@@ -276,14 +276,10 @@ static int cros_ec_uart_probe(struct serdev_device *serdev)/* Initialize ec_dev for cros_ec */ ec_dev->phys_name =dev_name (dev); - ec_dev->dev = dev; ec_dev->priv = ec_uart; ec_dev->irq = ec_uart->irq; ec_dev->cmd_xfer =NULL ; ec_dev->pkt_xfer = cros_ec_uart_pkt_xfer; - ec_dev->din_size =sizeof (struct ec_host_response ) + -sizeof (struct ec_response_get_protocol_info ); - ec_dev->dout_size =sizeof (struct ec_host_request ) +sizeof (struct ec_params_rwsig_action );serdev_device_set_client_ops (serdev, &cros_ec_uart_client_ops);
include/linux/platform_data/cros_ec_proto.h +10 -4
@@ -33,12 +33,18 @@/* * Max bus-specific overhead incurred by request/responses. -* I2C requires 1 additional byte for requests. -* I2C requires 2 additional bytes for responses. -* SPI requires up to 32 additional bytes for responses. +* +* Request: +* - I2C requires 1 byte (see struct ec_host_request_i2c). +* - ISHTP requires 4 bytes (see struct cros_ish_out_msg). +* +* Response: +* - I2C requires 2 bytes (see struct ec_host_response_i2c). +* - ISHTP requires 4 bytes (see struct cros_ish_in_msg). +* - SPI requires 32 bytes (see EC_MSG_PREAMBLE_COUNT). */ #define EC_PROTO_VERSION_UNKNOWN 0 -#define EC_MAX_REQUEST_OVERHEAD 1 +#define EC_MAX_REQUEST_OVERHEAD 4#define EC_MAX_RESPONSE_OVERHEAD 32/*
Signature · 1 line
2.51.0.268.g9569e192d0-goog
Tzung-Bi Shih
maintainerPATCH 3/5 platform/chrome: cros_ec: Separate initialization from cros_ec_register()
In reply to Tzung-Bi ShihMove the initialization of the struct cros_ec_device from cros_ec_register() into cros_ec_device_alloc().
This decouples device initialization from registration. By doing so, the per-device lock is now available immediately after allocation, allowing it to be used safely even before the device is fully registered.
--- drivers/platform/chrome/cros_ec.c | 57 ++++++++++++++++--------------- 1 file changed, 30 insertions(+), 27 deletions(-)
drivers/platform/chrome/cros_ec.c +30 -27
@@ -30,6 +30,14 @@ static struct cros_ec_platform pd_p = { .cmd_offset = EC_CMD_PASSTHRU_OFFSET(CROS_EC_DEV_PD_INDEX), }; +static void cros_ec_device_free (void *data) +{ +struct cros_ec_device *ec_dev = data; + +mutex_destroy (&ec_dev->lock); +lockdep_unregister_key (&ec_dev->lockdep_key); +} +struct cros_ec_device *cros_ec_device_alloc (struct device *dev) {struct cros_ec_device *ec_dev;
@@ -45,7 +53,28 @@ struct cros_ec_device *cros_ec_device_alloc(struct device *dev)sizeof (struct ec_params_rwsig_action ) +EC_MAX_REQUEST_OVERHEAD ; + ec_dev->din =devm_kzalloc (dev, ec_dev->din_size,GFP_KERNEL ); +if (!ec_dev->din) +return NULL ; + + ec_dev->dout =devm_kzalloc (dev, ec_dev->dout_size,GFP_KERNEL ); +if (!ec_dev->dout) +return NULL ; + ec_dev->dev = dev; + ec_dev->max_response =sizeof (struct ec_response_get_protocol_info ); + ec_dev->max_request =sizeof (struct ec_params_rwsig_action ); + ec_dev->suspend_timeout_ms =EC_HOST_SLEEP_TIMEOUT_DEFAULT ; + +BLOCKING_INIT_NOTIFIER_HEAD (&ec_dev->event_notifier); +BLOCKING_INIT_NOTIFIER_HEAD (&ec_dev->panic_notifier); + +lockdep_register_key (&ec_dev->lockdep_key); +mutex_init (&ec_dev->lock); +lockdep_set_class (&ec_dev->lock, &ec_dev->lockdep_key); + +if (devm_add_action_or_reset (dev, cros_ec_device_free, ec_dev)) +return NULL ;return ec_dev; }
@@ -200,29 +229,7 @@ static int cros_ec_ready_event(struct notifier_block *nb,int cros_ec_register (struct cros_ec_device *ec_dev) {struct device *dev = ec_dev->dev; -int err =0 ; - -BLOCKING_INIT_NOTIFIER_HEAD (&ec_dev->event_notifier); -BLOCKING_INIT_NOTIFIER_HEAD (&ec_dev->panic_notifier); - - ec_dev->max_request =sizeof (struct ec_params_hello ); - ec_dev->max_response =sizeof (struct ec_response_get_protocol_info ); - ec_dev->max_passthru =0 ; - ec_dev->ec =NULL ; - ec_dev->pd =NULL ; - ec_dev->suspend_timeout_ms =EC_HOST_SLEEP_TIMEOUT_DEFAULT ; - - ec_dev->din =devm_kzalloc (dev, ec_dev->din_size,GFP_KERNEL ); -if (!ec_dev->din) -return -ENOMEM ; - - ec_dev->dout =devm_kzalloc (dev, ec_dev->dout_size,GFP_KERNEL ); -if (!ec_dev->dout) -return -ENOMEM ; - -lockdep_register_key (&ec_dev->lockdep_key); -mutex_init (&ec_dev->lock); -lockdep_set_class (&ec_dev->lock, &ec_dev->lockdep_key); +int err;/* Send RWSIG continue to jump to RW for devices using RWSIG. */ err =cros_ec_rwsig_continue (ec_dev);
@@ -322,8 +329,6 @@ int cros_ec_register(struct cros_ec_device *ec_dev) exit:platform_device_unregister (ec_dev->ec);platform_device_unregister (ec_dev->pd); -mutex_destroy (&ec_dev->lock); -lockdep_unregister_key (&ec_dev->lockdep_key);return err; }EXPORT_SYMBOL (cros_ec_register);
@@ -343,8 +348,6 @@ void cros_ec_unregister(struct cros_ec_device *ec_dev) &ec_dev->notifier_ready);platform_device_unregister (ec_dev->pd);platform_device_unregister (ec_dev->ec); -mutex_destroy (&ec_dev->lock); -lockdep_unregister_key (&ec_dev->lockdep_key); }EXPORT_SYMBOL (cros_ec_unregister);
Signature · 1 line
2.51.0.268.g9569e192d0-goog
Tzung-Bi Shih
maintainerPATCH 4/5 platform/chrome: cros_ec: Add a flag to track registration state
In reply to Tzung-Bi ShihIntroduce a registered flag to the struct cros_ec_device to allow callers to determine if the device has been fully registered and is ready for use.
This is a preparatory step to prevent race conditions where other drivers might try to access the device before it is fully registered or after it has been unregistered.
--- drivers/platform/chrome/cros_ec.c | 7 +++++++ drivers/platform/chrome/cros_ec_proto.c | 15 +++++++++++++++ include/linux/platform_data/cros_ec_proto.h | 4 ++++ 3 files changed, 26 insertions(+)
drivers/platform/chrome/cros_ec.c +7 -0
@@ -9,6 +9,7 @@* battery charging and regulator control, firmware update. */ +#include <linux/cleanup.h> #include <linux/interrupt.h> #include <linux/module.h> #include <linux/of_platform.h>
@@ -316,6 +317,9 @@ int cros_ec_register(struct cros_ec_device *ec_dev) goto exit; } +scoped_guard (mutex, &ec_dev->lock) + ec_dev->registered = true; +dev_info (dev,"Chrome EC device registered\n" );/*
@@ -343,6 +347,9 @@ EXPORT_SYMBOL(cros_ec_register);*/ void cros_ec_unregister (struct cros_ec_device *ec_dev) { +scoped_guard (mutex, &ec_dev->lock) + ec_dev->registered = false; +if (ec_dev->mkbp_event_supported)blocking_notifier_chain_unregister (&ec_dev->event_notifier, &ec_dev->notifier_ready);
drivers/platform/chrome/cros_ec_proto.c +15 -0
@@ -3,6 +3,7 @@// // Copyright (C) 2015 Google, Inc +#include <linux/cleanup.h> #include <linux/delay.h> #include <linux/device.h> #include <linux/limits.h>
@@ -1153,5 +1154,19 @@ int cros_ec_get_cmd_versions(struct cros_ec_device *ec_dev, u16 cmd) }EXPORT_SYMBOL_GPL (cros_ec_get_cmd_versions); +/** +* cros_ec_device_registered - Return if the ec_dev is registered. +* +* @ec_dev: EC device +* +* Return: true if registered. Otherwise, false. +*/ +bool cros_ec_device_registered (struct cros_ec_device *ec_dev) +{ +guard (mutex)(&ec_dev->lock); +return ec_dev->registered; +} +EXPORT_SYMBOL_GPL (cros_ec_device_registered); +MODULE_LICENSE ("GPL" );MODULE_DESCRIPTION ("ChromeOS EC communication protocol helpers" );
include/linux/platform_data/cros_ec_proto.h +4 -0
@@ -128,6 +128,7 @@ struct cros_ec_command {* @dout_size: Size of dout buffer to allocate (zero to use static dout). * @wake_enabled: True if this device can wake the system from sleep. * @suspended: True if this device had been suspended. +* @registered: True if this device had been registered. * @cmd_xfer: Send command to EC and get response. * Returns the number of bytes received if the communication * succeeded, but that doesn't mean the EC was happy with the
@@ -186,6 +187,7 @@ struct cros_ec_device {int dout_size;bool wake_enabled;bool suspended; +bool registered;int (*cmd_xfer)(struct cros_ec_device *ec,struct cros_ec_command *msg); int (*pkt_xfer)(struct cros_ec_device *ec,
@@ -278,6 +280,8 @@ int cros_ec_cmd_readmem(struct cros_ec_device *ec_dev, u8 offset, u8 size, voidint cros_ec_get_cmd_versions (struct cros_ec_device *ec_dev,u16 cmd); +bool cros_ec_device_registered (struct cros_ec_device *ec_dev); +/** * cros_ec_get_time_ns() - Return time in ns. *
Signature · 1 line
2.51.0.268.g9569e192d0-goog
Tzung-Bi Shih
maintainerPATCH 5/5 Input: cros_ec_keyb - Defer probe until parent EC device is registered
In reply to Tzung-Bi ShihThe cros_ec_keyb driver can be probed before the cros_ec_device has completed the registration. This creates a race condition where cros_ec_keyb might access uninitialized data.
Fix this by calling cros_ec_device_registered() to check the parent's status. If the device is not yet ready, return -EPROBE_DEFER to ensure the probe is retried later.
--- drivers/input/keyboard/cros_ec_keyb.c | 6 ++++++ 1 file changed, 6 insertions(+)
drivers/input/keyboard/cros_ec_keyb.c +6 -0
@@ -705,6 +705,12 @@ static int cros_ec_keyb_probe(struct platform_device *pdev) ec =dev_get_drvdata (pdev->dev.parent);if (!ec)return -EPROBE_DEFER ; +/* +* Even if the cros_ec_device pointer is available, still need to check +* if the device is fully registered before using it. +*/ +if (!cros_ec_device_registered (ec)) +return -EPROBE_DEFER ; ckdev =devm_kzalloc (dev,sizeof (*ckdev),GFP_KERNEL );if (!ckdev)
Signature · 1 line
2.51.0.268.g9569e192d0-goog
Re: [PATCH 5/5] Input: cros_ec_keyb - Defer probe until parent EC device is registered
Quoted 9 lines · Tzung-Bi Shih
The
cros_ec_keybdriver can be probed before thecros_ec_devicehas completed the registration. This creates a race condition wherecros_ec_keybmight access uninitialized data.Fix this by calling
cros_ec_device_registered()to check the parent's status. If the device is not yet ready, return -EPROBE_DEFER to ensure the probe is retried later.Signed-off-by Tzung-Bi Shih <[email protected]>
Thanks.
Signature · 1 line
Dmitry
Dmitry Torokhov
maintainerHi Tzung-Bi,
Quoted 27 lines · Tzung-Bi Shih
A race is observed when
cros_ec_lpcand cros-ec-keyb are all built as modules.cros_ec_lpcis cros-ec-keyb's parent. However, they can be probed at the same time.Example:
+ -----------------------------------------------------------------+ | Some init process (e.g. udevd) |
deferred_probe_work_funcworker |
+ -----------------------------------------------------------------+| Probe cros-ec-keyb. | | | - Decide to defer[1]. | | | | A device bound to a driver[2]. | | Probe cros_ec_lpc. | | | - Init the struct[3]. | | | | Retry cros-ec-keyb from the | | | deferred list[4]. | | | - Won't defer again as [3]. | | | - Access uninitialized data in | | | the struct. | | - Register the device. | |+ -----------------------------------------------------------------+
[1] https://elixir.bootlin.com/linux/v6.16/source/drivers/input/keyboard/cros_ec_keyb.c#L707
[2] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L405
[3] https://elixir.bootlin.com/linux/v6.16/source/drivers/platform/chrome/cros_ec_lpc.c#L644
[4] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L418
Quoted 8 lines
Note that the device link[5] can't help as in the observed environment, the devices are already added via device_add()[6].
[5] https://www.kernel.org/doc/html/latest/driver-api/device_link.html#usage
[6] https://elixir.bootlin.com/linux/v6.16/source/drivers/acpi/acpi_platform.c#L177The series fixes the issue by ensuring the struct is ready for accessing before continuing to probe cros-ec-keyb.
Why is the keyboard platform device instantiated before the transport (cros_ec_lpc) is done initializing? I think this is the root of the issue...
Thanks.
Signature · 1 line
Dmitry
Tzung-Bi Shih
maintainerQuoted 34 lines · Dmitry Torokhov
Quoted 33 lines · Tzung-Bi Shih
A race is observed when
cros_ec_lpcand cros-ec-keyb are all built as modules.cros_ec_lpcis cros-ec-keyb's parent. However, they can be probed at the same time.Example:
+ -----------------------------------------------------------------+ | Some init process (e.g. udevd) |
deferred_probe_work_funcworker |
+ -----------------------------------------------------------------+| Probe cros-ec-keyb. | | | - Decide to defer[1]. | | | | A device bound to a driver[2]. | | Probe cros_ec_lpc. | | | - Init the struct[3]. | | | | Retry cros-ec-keyb from the | | | deferred list[4]. | | | - Won't defer again as [3]. | | | - Access uninitialized data in | | | the struct. | | - Register the device. | |+ -----------------------------------------------------------------+
[1] https://elixir.bootlin.com/linux/v6.16/source/drivers/input/keyboard/cros_ec_keyb.c#L707
[2] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L405
[3] https://elixir.bootlin.com/linux/v6.16/source/drivers/platform/chrome/cros_ec_lpc.c#L644
[4] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L418Note that the device link[5] can't help as in the observed environment, the devices are already added via device_add()[6].
[5] https://www.kernel.org/doc/html/latest/driver-api/device_link.html#usage
[6] https://elixir.bootlin.com/linux/v6.16/source/drivers/acpi/acpi_platform.c#L177
Quoted 6 lines
Quoted 2 lines
The series fixes the issue by ensuring the struct is ready for accessing before continuing to probe cros-ec-keyb.
Why is the keyboard platform device instantiated before the transport (
cros_ec_lpc) is done initializing? I think this is the root of the issue...
I may misunderstand but it seems to me:
- The ACPI bus enumerated and instantiated the platform devices[6] first.
- The keyboard platform device was probed when cros_ec_keyb_driver registered. It deferred as its parent's drvdata was NULL[1].
- The transport platform device was probed when cros_ec_lpc_driver registered. It set the drvdata[3].
- The keyboard platform device was probed again from retrying the deferred list, by another thread deferred_probe_work_func. The parent's drvdata wasn't NULL and cros_ec_register() for the transport device weren't finished. The race happened.
Tzung-Bi Shih
maintainerQuoted 49 lines · Tzung-Bi Shih
Quoted 34 lines · Dmitry Torokhov
Quoted 33 lines · Tzung-Bi Shih
A race is observed when
cros_ec_lpcand cros-ec-keyb are all built as modules.cros_ec_lpcis cros-ec-keyb's parent. However, they can be probed at the same time.Example:
+ -----------------------------------------------------------------+ | Some init process (e.g. udevd) |
deferred_probe_work_funcworker |
+ -----------------------------------------------------------------+| Probe cros-ec-keyb. | | | - Decide to defer[1]. | | | | A device bound to a driver[2]. | | Probe cros_ec_lpc. | | | - Init the struct[3]. | | | | Retry cros-ec-keyb from the | | | deferred list[4]. | | | - Won't defer again as [3]. | | | - Access uninitialized data in | | | the struct. | | - Register the device. | |+ -----------------------------------------------------------------+
[1] https://elixir.bootlin.com/linux/v6.16/source/drivers/input/keyboard/cros_ec_keyb.c#L707
[2] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L405
[3] https://elixir.bootlin.com/linux/v6.16/source/drivers/platform/chrome/cros_ec_lpc.c#L644
[4] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L418Note that the device link[5] can't help as in the observed environment, the devices are already added via device_add()[6].
[5] https://www.kernel.org/doc/html/latest/driver-api/device_link.html#usage
[6] https://elixir.bootlin.com/linux/v6.16/source/drivers/acpi/acpi_platform.c#L177Quoted 6 lines
Quoted 2 lines
The series fixes the issue by ensuring the struct is ready for accessing before continuing to probe cros-ec-keyb.
Why is the keyboard platform device instantiated before the transport (
cros_ec_lpc) is done initializing? I think this is the root of the issue...I may misunderstand but it seems to me:
- The ACPI bus enumerated and instantiated the platform devices[6] first.
- The keyboard platform device was probed when
cros_ec_keyb_driverregistered. It deferred as its parent's drvdata was NULL[1].
Quoted 7 lines
- The transport platform device was probed when
cros_ec_lpc_driverregistered. It set the drvdata[3].- The keyboard platform device was probed again from retrying the deferred list, by another thread
deferred_probe_work_func. The parent's drvdata wasn't NULL andcros_ec_register()for the transport device weren't finished. The race happened.
Hi Dmitry,
Does it make sense to you?
Dmitry Torokhov
maintainerHi Tzung-Bi,
Quoted 53 lines · Tzung-Bi Shih
Quoted 52 lines · Tzung-Bi Shih
Quoted 34 lines · Dmitry Torokhov
Quoted 33 lines · Tzung-Bi Shih
A race is observed when
cros_ec_lpcand cros-ec-keyb are all built as modules.cros_ec_lpcis cros-ec-keyb's parent. However, they can be probed at the same time.Example:
+ -----------------------------------------------------------------+ | Some init process (e.g. udevd) |
deferred_probe_work_funcworker |
+ -----------------------------------------------------------------+| Probe cros-ec-keyb. | | | - Decide to defer[1]. | | | | A device bound to a driver[2]. | | Probe cros_ec_lpc. | | | - Init the struct[3]. | | | | Retry cros-ec-keyb from the | | | deferred list[4]. | | | - Won't defer again as [3]. | | | - Access uninitialized data in | | | the struct. | | - Register the device. | |+ -----------------------------------------------------------------+
[1] https://elixir.bootlin.com/linux/v6.16/source/drivers/input/keyboard/cros_ec_keyb.c#L707
[2] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L405
[3] https://elixir.bootlin.com/linux/v6.16/source/drivers/platform/chrome/cros_ec_lpc.c#L644
[4] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L418Note that the device link[5] can't help as in the observed environment, the devices are already added via device_add()[6].
[5] https://www.kernel.org/doc/html/latest/driver-api/device_link.html#usage
[6] https://elixir.bootlin.com/linux/v6.16/source/drivers/acpi/acpi_platform.c#L177Quoted 6 lines
Quoted 2 lines
The series fixes the issue by ensuring the struct is ready for accessing before continuing to probe cros-ec-keyb.
Why is the keyboard platform device instantiated before the transport (
cros_ec_lpc) is done initializing? I think this is the root of the issue...I may misunderstand but it seems to me:
- The ACPI bus enumerated and instantiated the platform devices[6] first.
- The keyboard platform device was probed when
cros_ec_keyb_driverregistered. It deferred as its parent's drvdata was NULL[1].- The transport platform device was probed when
cros_ec_lpc_driverregistered. It set the drvdata[3].
Quoted 8 lines
Quoted 4 lines
- The keyboard platform device was probed again from retrying the deferred list, by another thread
deferred_probe_work_func. The parent's drvdata wasn't NULL andcros_ec_register()for the transport device weren't finished. The race happened.Hi Dmitry,
Does it make sense to you?
I'll have to research how MFD mixes up statically described and DT-described platform devices and makes sure that children are not probed before the parent is ready - I think we need to make cros_ec behave the same way.
Thanks.
Signature · 1 line
Dmitry
Tzung-Bi Shih
maintainerQuoted 59 lines · Dmitry Torokhov
Quoted 58 lines · Tzung-Bi Shih
Quoted 56 lines · Tzung-Bi Shih
Quoted 34 lines · Dmitry Torokhov
Quoted 33 lines · Tzung-Bi Shih
A race is observed when
cros_ec_lpcand cros-ec-keyb are all built as modules.cros_ec_lpcis cros-ec-keyb's parent. However, they can be probed at the same time.Example:
+ -----------------------------------------------------------------+ | Some init process (e.g. udevd) |
deferred_probe_work_funcworker |
+ -----------------------------------------------------------------+| Probe cros-ec-keyb. | | | - Decide to defer[1]. | | | | A device bound to a driver[2]. | | Probe cros_ec_lpc. | | | - Init the struct[3]. | | | | Retry cros-ec-keyb from the | | | deferred list[4]. | | | - Won't defer again as [3]. | | | - Access uninitialized data in | | | the struct. | | - Register the device. | |+ -----------------------------------------------------------------+
[1] https://elixir.bootlin.com/linux/v6.16/source/drivers/input/keyboard/cros_ec_keyb.c#L707
[2] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L405
[3] https://elixir.bootlin.com/linux/v6.16/source/drivers/platform/chrome/cros_ec_lpc.c#L644
[4] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L418Note that the device link[5] can't help as in the observed environment, the devices are already added via device_add()[6].
[5] https://www.kernel.org/doc/html/latest/driver-api/device_link.html#usage
[6] https://elixir.bootlin.com/linux/v6.16/source/drivers/acpi/acpi_platform.c#L177Quoted 6 lines
Quoted 2 lines
The series fixes the issue by ensuring the struct is ready for accessing before continuing to probe cros-ec-keyb.
Why is the keyboard platform device instantiated before the transport (
cros_ec_lpc) is done initializing? I think this is the root of the issue...I may misunderstand but it seems to me:
- The ACPI bus enumerated and instantiated the platform devices[6] first.
- The keyboard platform device was probed when
cros_ec_keyb_driverregistered. It deferred as its parent's drvdata was NULL[1].- The transport platform device was probed when
cros_ec_lpc_driverregistered. It set the drvdata[3].- The keyboard platform device was probed again from retrying the deferred list, by another thread
deferred_probe_work_func. The parent's drvdata wasn't NULL andcros_ec_register()for the transport device weren't finished. The race happened.
Quoted 8 lines
Quoted 3 lines
Hi Dmitry,
Does it make sense to you?
I'll have to research how MFD mixes up statically described and DT-described platform devices and makes sure that children are not probed before the parent is ready - I think we need to make
cros_ecbehave the same way.
I may misunderstand but FWIW:
I failed to find relevant code in MFD [7] that guarantees the probe order. Also, I'm curious about wouldn't code at [7] results in duplicate platform devices? E.g., 1 populated from OF; 1 created by MFD.
Note: current cros_ec_dev.c doesn't use of_compatible in struct mfd_cell.
If we're looking at how cros_ec_dev.c guarantees the order:
- The transport platfrom device is probed first. It calls cros_ec_register().
- In cros_ec_register(), it registers the MFD device "cros-ec-dev". And the children devices are added via mfd_add_devices().
Back to the issue we observed:
- The platform devices are created when it scans the tree in ACPI[6]. We probably have no way to prevent the devices from adding unless specifying enumeration_by_parent[8].
- When some of them are modules, the driver registrations are tied to the module insertion. They can be arrived by anytime unless we use something similar to soft dependency[9]. A Kconfig dependency will also be needed to prevent cros_ec_lpcs=m but cros_ec_keyb=y. However, cros_ec_keyb would need to specify 2 possible dependencies "cros_ec_lpcs" and "cros_ec_spi"[10]. I'm not sure what would be happening if a system has no cros_ec_spi module at all.
[7] https://elixir.bootlin.com/linux/v6.16/source/drivers/mfd/mfd-core.c#L184
[8] https://elixir.bootlin.com/linux/v6.16/source/drivers/acpi/scan.c#L2204
[9] https://elixir.bootlin.com/linux/v6.16/source/include/linux/module.h#L176
[10] https://elixir.bootlin.com/linux/v6.16/source/drivers/input/keyboard/Kconfig#L743
Dmitry Torokhov
maintainerQuoted 90 lines · Tzung-Bi Shih
Quoted 59 lines · Dmitry Torokhov
Quoted 58 lines · Tzung-Bi Shih
Quoted 56 lines · Tzung-Bi Shih
Quoted 34 lines · Dmitry Torokhov
Quoted 33 lines · Tzung-Bi Shih
A race is observed when
cros_ec_lpcand cros-ec-keyb are all built as modules.cros_ec_lpcis cros-ec-keyb's parent. However, they can be probed at the same time.Example:
+ -----------------------------------------------------------------+ | Some init process (e.g. udevd) |
deferred_probe_work_funcworker |
+ -----------------------------------------------------------------+| Probe cros-ec-keyb. | | | - Decide to defer[1]. | | | | A device bound to a driver[2]. | | Probe cros_ec_lpc. | | | - Init the struct[3]. | | | | Retry cros-ec-keyb from the | | | deferred list[4]. | | | - Won't defer again as [3]. | | | - Access uninitialized data in | | | the struct. | | - Register the device. | |+ -----------------------------------------------------------------+
[1] https://elixir.bootlin.com/linux/v6.16/source/drivers/input/keyboard/cros_ec_keyb.c#L707
[2] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L405
[3] https://elixir.bootlin.com/linux/v6.16/source/drivers/platform/chrome/cros_ec_lpc.c#L644
[4] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L418Note that the device link[5] can't help as in the observed environment, the devices are already added via device_add()[6].
[5] https://www.kernel.org/doc/html/latest/driver-api/device_link.html#usage
[6] https://elixir.bootlin.com/linux/v6.16/source/drivers/acpi/acpi_platform.c#L177Quoted 6 lines
Quoted 2 lines
The series fixes the issue by ensuring the struct is ready for accessing before continuing to probe cros-ec-keyb.
Why is the keyboard platform device instantiated before the transport (
cros_ec_lpc) is done initializing? I think this is the root of the issue...I may misunderstand but it seems to me:
- The ACPI bus enumerated and instantiated the platform devices[6] first.
- The keyboard platform device was probed when
cros_ec_keyb_driverregistered. It deferred as its parent's drvdata was NULL[1].- The transport platform device was probed when
cros_ec_lpc_driverregistered. It set the drvdata[3].- The keyboard platform device was probed again from retrying the deferred list, by another thread
deferred_probe_work_func. The parent's drvdata wasn't NULL andcros_ec_register()for the transport device weren't finished. The race happened.Quoted 8 lines
Quoted 3 lines
Hi Dmitry,
Does it make sense to you?
I'll have to research how MFD mixes up statically described and DT-described platform devices and makes sure that children are not probed before the parent is ready - I think we need to make
cros_ecbehave the same way.I may misunderstand but FWIW:
I failed to find relevant code in MFD [7] that guarantees the probe order. Also, I'm curious about wouldn't code at [7] results in duplicate platform devices? E.g., 1 populated from OF; 1 created by MFD.
Note: current
cros_ec_dev.cdoesn't useof_compatiblein structmfd_cell.If we're looking at how
cros_ec_dev.cguarantees the order:- The transport platfrom device is probed first. It calls
cros_ec_register().
- Incros_ec_register(), it registers the MFD device "cros-ec-dev". And the children devices are added viamfd_add_devices().Back to the issue we observed:
- The platform devices are created when it scans the tree in ACPI[6]. We probably have no way to prevent the devices from adding unless specifying
enumeration_by_parent[8].
Quoted 7 lines
- When some of them are modules, the driver registrations are tied to the module insertion. They can be arrived by anytime unless we use something similar to soft dependency[9]. A
Kconfigdependency will also be needed to prevent cros_ec_lpcs=m but cros_ec_keyb=y. However,cros_ec_keybwould need to specify 2 possible dependencies "cros_ec_lpcs" and "cros_ec_spi"[10]. I'm not sure what would be happening if a system has nocros_ec_spimodule at all.
Sorry for the delay. I think we need to figure out how to delay device registration until after the transport device is registered and probed. Unfortunately I have not had time to look at this properly so I acked the changes to cros_ec_keyb so that we can close the hole for now and later we can try to see if there is a reasonable better solution.
Thanks.
Signature · 1 line
Dmitry
Tzung-Bi Shih
maintainerQuoted 99 lines · Dmitry Torokhov
Quoted 90 lines · Tzung-Bi Shih
Quoted 59 lines · Dmitry Torokhov
Quoted 58 lines · Tzung-Bi Shih
Quoted 56 lines · Tzung-Bi Shih
Quoted 34 lines · Dmitry Torokhov
Quoted 33 lines · Tzung-Bi Shih
A race is observed when
cros_ec_lpcand cros-ec-keyb are all built as modules.cros_ec_lpcis cros-ec-keyb's parent. However, they can be probed at the same time.Example:
+ -----------------------------------------------------------------+ | Some init process (e.g. udevd) |
deferred_probe_work_funcworker |
+ -----------------------------------------------------------------+| Probe cros-ec-keyb. | | | - Decide to defer[1]. | | | | A device bound to a driver[2]. | | Probe cros_ec_lpc. | | | - Init the struct[3]. | | | | Retry cros-ec-keyb from the | | | deferred list[4]. | | | - Won't defer again as [3]. | | | - Access uninitialized data in | | | the struct. | | - Register the device. | |+ -----------------------------------------------------------------+
[1] https://elixir.bootlin.com/linux/v6.16/source/drivers/input/keyboard/cros_ec_keyb.c#L707
[2] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L405
[3] https://elixir.bootlin.com/linux/v6.16/source/drivers/platform/chrome/cros_ec_lpc.c#L644
[4] https://elixir.bootlin.com/linux/v6.16/source/drivers/base/dd.c#L418Note that the device link[5] can't help as in the observed environment, the devices are already added via device_add()[6].
[5] https://www.kernel.org/doc/html/latest/driver-api/device_link.html#usage
[6] https://elixir.bootlin.com/linux/v6.16/source/drivers/acpi/acpi_platform.c#L177Quoted 6 lines
Quoted 2 lines
The series fixes the issue by ensuring the struct is ready for accessing before continuing to probe cros-ec-keyb.
Why is the keyboard platform device instantiated before the transport (
cros_ec_lpc) is done initializing? I think this is the root of the issue...I may misunderstand but it seems to me:
- The ACPI bus enumerated and instantiated the platform devices[6] first.
- The keyboard platform device was probed when
cros_ec_keyb_driverregistered. It deferred as its parent's drvdata was NULL[1].- The transport platform device was probed when
cros_ec_lpc_driverregistered. It set the drvdata[3].- The keyboard platform device was probed again from retrying the deferred list, by another thread
deferred_probe_work_func. The parent's drvdata wasn't NULL andcros_ec_register()for the transport device weren't finished. The race happened.Quoted 8 lines
Quoted 3 lines
Hi Dmitry,
Does it make sense to you?
I'll have to research how MFD mixes up statically described and DT-described platform devices and makes sure that children are not probed before the parent is ready - I think we need to make
cros_ecbehave the same way.I may misunderstand but FWIW:
I failed to find relevant code in MFD [7] that guarantees the probe order. Also, I'm curious about wouldn't code at [7] results in duplicate platform devices? E.g., 1 populated from OF; 1 created by MFD.
Note: current
cros_ec_dev.cdoesn't useof_compatiblein structmfd_cell.If we're looking at how
cros_ec_dev.cguarantees the order:- The transport platfrom device is probed first. It calls
cros_ec_register().
- Incros_ec_register(), it registers the MFD device "cros-ec-dev". And the children devices are added viamfd_add_devices().Back to the issue we observed:
- The platform devices are created when it scans the tree in ACPI[6]. We probably have no way to prevent the devices from adding unless specifying
enumeration_by_parent[8].Quoted 7 lines
- When some of them are modules, the driver registrations are tied to the module insertion. They can be arrived by anytime unless we use something similar to soft dependency[9]. A
Kconfigdependency will also be needed to prevent cros_ec_lpcs=m but cros_ec_keyb=y. However,cros_ec_keybwould need to specify 2 possible dependencies "cros_ec_lpcs" and "cros_ec_spi"[10]. I'm not sure what would be happening if a system has nocros_ec_spimodule at all.
Quoted 5 lines
Sorry for the delay. I think we need to figure out how to delay device registration until after the transport device is registered and probed. Unfortunately I have not had time to look at this properly so I acked the changes to
cros_ec_keybso that we can close the hole for now and later we can try to see if there is a reasonable better solution.
Thanks. Let's move forward with the current fix, even though we know it's suboptimal.
My suspicion is that we lack a proper in-kernel dependency mechanism. This is on my radar, and I'll investigate a better approach.
Tzung-Bi Shih
maintainerQuoted 20 lines · Tzung-Bi Shih
A race is observed when
cros_ec_lpcand cros-ec-keyb are all built as modules.cros_ec_lpcis cros-ec-keyb's parent. However, they can be probed at the same time.Example:
+ -----------------------------------------------------------------+ | Some init process (e.g. udevd) |
deferred_probe_work_funcworker |
+ -----------------------------------------------------------------+| Probe cros-ec-keyb. | | | - Decide to defer[1]. | | | | A device bound to a driver[2]. | | Probe cros_ec_lpc. | | | - Init the struct[3]. | | | | Retry cros-ec-keyb from the | | | deferred list[4]. | | | - Won't defer again as [3]. | | | - Access uninitialized data in | | | the struct. | | - Register the device. | |
Quoted 3 lines
+ -----------------------------------------------------------------+
[...]
Applied to
https://git.kernel.org/pub/scm/linux/kernel/git/chrome-platform/linux.git for-next
[1/5] platform/chrome: Centralize cros_ec_device allocation
commit: 918856986014142271a70a334d300994b9c41720
[2/5] platform/chrome: Centralize common cros_ec_device initialization
commit: e19ceeb1c0f63e3e15b197c5f34797134b51ba0e
[3/5] platform/chrome: cros_ec: Separate initialization from cros_ec_register()
commit: 7a79b0bfd8b3995a39d25bffcf57273635c0e542
[4/5] platform/chrome: cros_ec: Add a flag to track registration state
commit: 56cb557279d70397cefb497e0f06bdd6fd685f8e
[5/5] Input: cros_ec_keyb - Defer probe until parent EC device is registered
commit: 48633acccf38d706d7b368400647bb9db9caf1aeThanks!