State of the Linux Driver Core Subsystem for 2.6.23-rc8
Here's a summary of the current state of the Linux Driver core subsystem, as of 2.6.23-rc8. If the information in here is incorrect, or anyone knows of any outstanding issues not listed here, please let me know. List of outstanding regressions from 2.6.22: - none known. List of outstanding regressions from older kernel versions: - none known. There are no currently open Driver core or sysfs bugs in bugzilla. Future patches that are currently in my quilt tree (as found at http://www.kernel.org/pub/linux/kernel/people/gregkh/gregkh-2.6/ ) for the Driver core subsystem are as follows. All of these will be submitted for inclusion into 2.6.24, except as noted. The diffstat of these patches is included at the bottom of this message for those that are interested. - HOWTO ja_JP updates - metric boatload of changes and tweaks to sysfs from Tejun. These clean up the internal usage and implementation of sysfs, and split sysfs from the kobject model a lot, fixing a lot of problems along the way. I can't thank him enough for this work. - cleaned up the usage of struct subsystem some more by removing some functions that are not needed, or only used by the driver core. Details on the api changes are below - cleaned up the usage of the kobject->name field and removed it entirely. This saves us some size on every kobject which really adds up on those s390 31 bit boxes with 20,000 different block devices. - uevent environment variable handling has been made simpler, and now hopefully, almost impossible to get wrong. Previously we were forcing every subsystem to open-code a lot of this logic. Thanks to Kay for this work. - global list of sysdev_drivers is removed, as no one was using it at all. - the number of legacy ptys can now be dynamically specified on the kernel command line, allowing people who refuse to fix their old applications to be able to still run on newer distros that want to limit this number at build time. - dmi sysfs code cleanups and fixes - block devices are moved from /sys/block to /sys/class/block. This patch has been hanging around for almost a year now, and hopefully we have worked out all of the kinks and userspace boot breakages. If anyone has any problems with this, please let me and Kay know about it. I'm hesitant to include it in 2.6.24 as it has recently changed and needs more testing. Maybe it will go into 2.6.25. Oh, udev has been able to handle this for over a year, so there should not be any problems with distros that are still supported. It can also be turned off with the CONFIG_SYSFS_DEPRECATED build option for older distros. - new debugfs functions for people who like hex numbers - platform devices now have the "platform:" alias added to their modalias to fix the recursive modprobe loops that Red Hat, OLPC, SuSE, and Debian have reported in the current code. - some struct class_device to struct device conversions were done for video and some various char drivers. More of these are on the way. - cdev kobject name cleanups. The kobject name of a cdev does not do anything, so it should not be set. - firmware collision fix between i2c and i2c-dev. - uevent files for bus and driver have been added. SuSE has shipped this for quite some time now. - if CONFIG_HOTPLUG is not enabled, some more space is saved in the uevent code. - the path to the uevent helper program (traditionally /sbin/hotplug) can be specified in the kernel config. This keeps the kernel from having to call a program that is not present 300+ times before init starts up, speeding up boot time in theory. New driver core api changes for 2.6.24: - debugfs_create_x8(), debugfs_create_x16(), and debugfs_create_x32() have been added - struct kobject.name field has been removed. Use kobject_set_name() and kobject_name() to set and get the name. - uevent callbacks in the driver core and for kobjects have been changed from: int (*uevent)(struct device *dev, char **envp, int num_envp, char *buffer, int buffer_size); to: int (*uevent)(struct device *dev, struct kobj_uevent_env *env); and struct kobj_uevent_env has been created to make it simpler to create environment variables for uevents. - struct platform_device.id has changed from an u32 to an int. - the following functions have been removed either entirely or from the global namespace:
State of the Linux Driver Core Subsystem for 2.6.23-rc8
Here's a summary of the current state of the Linux Driver core subsystem, as of 2.6.23-rc8. If the information in here is incorrect, or anyone knows of any outstanding issues not listed here, please let me know. List of outstanding regressions from 2.6.22: - none known. List of outstanding regressions from older kernel versions: - none known. There are no currently open Driver core or sysfs bugs in bugzilla. Future patches that are currently in my quilt tree (as found at http://www.kernel.org/pub/linux/kernel/people/gregkh/gregkh-2.6/ ) for the Driver core subsystem are as follows. All of these will be submitted for inclusion into 2.6.24, except as noted. The diffstat of these patches is included at the bottom of this message for those that are interested. - HOWTO ja_JP updates - metric boatload of changes and tweaks to sysfs from Tejun. These clean up the internal usage and implementation of sysfs, and split sysfs from the kobject model a lot, fixing a lot of problems along the way. I can't thank him enough for this work. - cleaned up the usage of struct subsystem some more by removing some functions that are not needed, or only used by the driver core. Details on the api changes are below - cleaned up the usage of the kobject-name field and removed it entirely. This saves us some size on every kobject which really adds up on those s390 31 bit boxes with 20,000 different block devices. - uevent environment variable handling has been made simpler, and now hopefully, almost impossible to get wrong. Previously we were forcing every subsystem to open-code a lot of this logic. Thanks to Kay for this work. - global list of sysdev_drivers is removed, as no one was using it at all. - the number of legacy ptys can now be dynamically specified on the kernel command line, allowing people who refuse to fix their old applications to be able to still run on newer distros that want to limit this number at build time. - dmi sysfs code cleanups and fixes - block devices are moved from /sys/block to /sys/class/block. This patch has been hanging around for almost a year now, and hopefully we have worked out all of the kinks and userspace boot breakages. If anyone has any problems with this, please let me and Kay know about it. I'm hesitant to include it in 2.6.24 as it has recently changed and needs more testing. Maybe it will go into 2.6.25. Oh, udev has been able to handle this for over a year, so there should not be any problems with distros that are still supported. It can also be turned off with the CONFIG_SYSFS_DEPRECATED build option for older distros. - new debugfs functions for people who like hex numbers - platform devices now have the platform: alias added to their modalias to fix the recursive modprobe loops that Red Hat, OLPC, SuSE, and Debian have reported in the current code. - some struct class_device to struct device conversions were done for video and some various char drivers. More of these are on the way. - cdev kobject name cleanups. The kobject name of a cdev does not do anything, so it should not be set. - firmware collision fix between i2c and i2c-dev. - uevent files for bus and driver have been added. SuSE has shipped this for quite some time now. - if CONFIG_HOTPLUG is not enabled, some more space is saved in the uevent code. - the path to the uevent helper program (traditionally /sbin/hotplug) can be specified in the kernel config. This keeps the kernel from having to call a program that is not present 300+ times before init starts up, speeding up boot time in theory. New driver core api changes for 2.6.24: - debugfs_create_x8(), debugfs_create_x16(), and debugfs_create_x32() have been added - struct kobject.name field has been removed. Use kobject_set_name() and kobject_name() to set and get the name. - uevent callbacks in the driver core and for kobjects have been changed from: int (*uevent)(struct device *dev, char **envp, int num_envp, char *buffer, int buffer_size); to: int (*uevent)(struct device *dev, struct kobj_uevent_env *env); and struct kobj_uevent_env has been created to make it simpler to create environment variables for uevents. - struct platform_device.id has changed from an u32 to an int. - the following functions have been removed either entirely or from the global namespace: get_bus()