This is my rework of the x86 xv6 operating system. I have stopped updating it for now.

Changes made so far:

• Sched queue based on process priorities, differing from typical unix in that higher number eq higher prio. Insertions into the sched queue are done by placing the highest prio first on the runqeue and as the kernel walks down the queue it checks the process to be enqueued against the process in this spot in the queue. I have added a field to the process struct for CPU (current cpu queue it is on) and curr (current queue) so that a process can be found and plucked out or moved very quickly. No iterative searching needed. The queue functions in queue.c handle this.

• Added basic function to do cpu usage averaging to dynamically set time quanta for processes based on a the mean process life in clock cycles. After a process is preempted , the cpu usage stays but the assigned time quanta is increased by the current average.

• Pressing the tab key now clears the console

• Implemented semaphores , only inside the kernel for now. Will add to userspace later.

• Implemented basic hashing functions for creating creating hash queues and hash tables.

• Added per-cpu runqeueus so each cpu will have its own runqueue.

• Very basic cpu balancing routine, when a cpu has no tasks on the ready queue or its run queue it will spin checking if queues need balancing and any that do it will free the tail and place it on on the ready queue which the hungry cpu will be able to take very quickly.

• No more iteration through every process on sleep, wakeup, scheduling. All done on the per-cpu runqueues, a global ready queue and a global sleep queue. This means a lot less time is spent running through every process, the ones we need are already in queue so just pluck them out of there. Not constant time but it is a lot faster.

• Added basic signals, can be seen in signal.h. Signals can be masked and ignored via the sigignore system call (non fatal signals only) signal handlers not properly implemented yet will get to this later (Just need to save the eip of the sig handler and at the end of the routine make sure to force a sig_return style function that restores process context to previous instruction pointer and regs).

• Added mounting of secondary filesystems, can be mounted on any directory in your main filesystem. Can traverse across the mount point and use commands across the mountpoint

• Added multi-disk support in the ide driver. Supports up to 8 disks and does a probe check on boot to see which are present. If you choose to attach extra disks they may be mounted someone in your rootfs.

• Added preemption of processes if the time quantum is exceeded and a higher prio process is waiting, I added the special PREEMPTED process state so that I can ensure fairness and allow a preempted process to execute again later even if it is low prio.

• Added nonblocking lock specifically for mounting. It is just a lock that returns immediately if locked instead of spinning or sleeping.

• Added login shell, only supports 1 pair of credentials located in the passwd file in the users directory

• Freemem function which spits out the # of pages allocated to all user processes (very useful for tracking memory usage when you play around with memory management)

• Sig command which uses the sig system call to send a signal to a process

• Mountfs and umountfs user programs which can be run from the shell, allows mounting of the secondary disk on the secondary ata controller, usage is mountfs dev dir (ie mountfs 2 /dir) It does not make new nodes so you need to create the directory to mount onto with mkdir. Unmount works as umountfs mountdir (ie umountfs /dir).

How to build (linux):

• Download QEMU & the source
• Ensure you get the source from a commit that is preceded by FUNCTIONAL
• Navigate to /users in the source tree
• Enter 'Make qemu' in your terminal (This builds vectors.S from the pl file, both hard disk images, and the kernel, and starts qemu)
• You need to update the passwd file in /users if you want to use any other user/pass combo (only supports 1 set of credentials)