|
I don't know the entire kernel well enough do document all of the changes. In the course of converting the examples (or actually, adapting Emmanuel Papirakis's changes) I came across the following differences. I listed all of them here together to help module programmers, especially those who learned from previous versions of this book and are most familiar with the techniques I use, convert to the new version.
An additional resource for people who wish to convert to 2.2 is located on Richard Gooch's site .
If you need put_user or get_user you have to #include it.
In version 2.2, get_user receives both the pointer into user memory and the variable in kernel memory to fill with the information. The reason for this is that get_user can now read two or four bytes at a time if the variable we read is two or four bytes long.
This structure now has a flush function between the open and close functions.
In version 2.2, the close function returns an integer, so it's allowed to fail.
The headers for these functions changed. They now return ssize_t instead of an integer, and their parameter list is different. The inode is no longer a parameter, and on the other hand the offset into the file is.
This function no longer exists. Instead, you call the regular proc_register and put zero in the inode field of the structure.
The signals in the task structure are no longer a 32 bit integer, but an array of _NSIG_WORDS integers.
Even if you want to scheduale a task to happen from inside an interrupt handler, you use queue_task, not queue_task_irq.
You no longer just declare module parameters as global variables. In 2.2 you have to also use MODULE_PARM to declare their type. This is a big improvement, because it allows the module to receive string parameters which start with a digits, for example, without getting confused.
The kernel is no longer inside one huge spinlock, which means that kernel modules have to be aware of SMP.
Hosting by: Hurra Communications Ltd.
Generated: 2007-01-26 17:57:41