Understanding the Linux Kernel

3.5. Anticipating Linux 2.4

The new kernel supports a huge number of users and groups, because it makes use of 32-bit UIDs and GIDs.

In order to raise the hardcoded limit on the number of processes, Linux 2.4 removes the tasks array, which previously included pointers to all process descriptors.

Moreover, Linux 2.4 no longer includes a Task State Segment for each process. The tss field in the process descriptor has thus been replaced by a pointer to a data structure storing the information that was previously in the TSS, namely the register contents and the I/O bitmap. Linux 2.4 makes use of just one TSS for each CPU in the system. When a context switch occurs, the kernel uses the per-process data structures to save and restore the register contents and to fill the I/O bitmap in the TSS of the executing CPU.

Linux 2.4 enhances wait queues. Sleeping processes are now stored in lists implemented through the efficient list_head data type. Moreover, the kernel is now able to wake up just a single process that is sleeping in a wait queue, thus greatly improving the efficiency of semaphores.

Finally, Linux 2.4 adds a new flag to the clone( ) system call: CLONE_PARENT allows the new lightweight process to have the same parent as the process that invoked the system call.