Device sends interrupt to CPU.

CPU hands to kernel.

Switch to kernel mode.

Gets processed by kernel.

return / accept new interrupt

tasklets have greater serialization guarantees

(softirqs have none)

Work queues are scheduled, rather than run in interrupt
context.

user library makes system call, which traps to kernel,

  processes system call,

and returns to user library 

Don't add your data to the linked list struct...

 add the linked list struct to your struct of data. 

Basic unit of memory management?

Hardware that manages it? 

page

 the MMU 

ZONE_DMA = less than 16MB

 ZONE_NORMAL = 16MB - 768MB

ZONE_HIGHMEM = >768MB 

What is the SLAB?

 First used where? 

SLAB is a generic data structure caching layer

 First introduced in Solaris 

the memory descriptor

(the kernel's representation of the process's address space) 

A character device is a stream of characters.

  A block device provides random access to block sized chunks of data. 

hardware unit of memory = sector

software unit of memory = block 

struct bio has a pointer to a list of bio_vec structs, and another pointer to the currently indexed member of that list.

Each bio_vec struct points to a page struct where the buffer resides. 

lowercase with underscores for variables and functions. 

not CamelCase.

descriptive.

From elevators.

They service all requests in a single direction before going in the opposite direction. 

A section of code where multiple threads could be reading or writing the same data, in an unpredictable order.

code path that accesses and manipulates shared data

To avoid race conditions and deadlocks?

To ensure that unsafe concurrency is prevented and that race conditions do not occur

1. mutual exclusion

2. hold and wait

3. no preemption

4. circular wait 

They put an ostrich in a room with deadlock. The ostrich puts its head in the sand and can see nothing.  It then kicks the shit out of deadlock, which it has mistaken for a lion.

lol great answer

barrier operations: rmb(), wmb(), and mb()

no loads from before the call will be performed after it. 

Use spin locks for very short term. Spinning is costly.

Use semaphore if you need to cross the kernel / user space boundary.

Otherwise use a Mutex. 

1st, determine if the lock will be held for a long time. If yes, use spin lock. If not, determine if code can sleep. If not, use spin lock. If you need to synchronize between user and kernel space, use semaphore. Otherwise, use mutex.

Tick rate is the frequency of the system timer.

An timer interrupt happens on an exact tick.  Used for...  Uptime.  Time of day.  Runqueue balancing. etc

Used for Wall time and System Uptime.

Timer interrupts are dynamically scheduled to whenever the next timer is set to go off.

 Advanages: reduced overhead and power consumption. 

It allows us to use the same few functions and system calls on block devices that store their ones and zeros in ways completely unlike one another.

manages our filesystems and gives user-space apps a single unified interface to work with.

file

inode

dentry

superblock 

superblock: a specific mounted filesystem.
inode: a specific file.

dentry: a single component of a path.
file: an open file as associated with a process. 

files_struct: per-process info about open files/descriptors

fs_struct: per-process filesystem info

namespace: uncommonly, a process will have a unique version of the file hierarchy as described by this struct.

(how do they relate?) 

they related to each other by a single process

Temporal locality is close in time. If a file was recently accessed it may be accessed again.

Spatial locality is close together in the directory structure.  If /a/b/c has been accessed, /a/b/* might all be worth caching. 

Reading is simple. When we want to read something, we check to see if it is cached first.

 Writing has multiple strategies. We could not cache writing at all, or we could write to both the cache and the backing store at once, or we could write to the cache first and maintain a dirty list.

address_space

It maps stuff in cache to its real location.

A -> B -> C -> D,E,F

(D, E, and F are children of C) 

1. extended (virtual) machine: the OS is easier to program than the underlying hardware.  The operating system provides a variety of services that programs can obtain using system calls.

2. resource manager

Pipes.

Shared memory.

Named pipes.

Ordinary files.

entire memory space of a process

(I don't think it always has to be the entire memory space)

Created with fork(). Differentiated by its return value.

child sees it return 0

parent sees it return child's process id (if successful)

In Linux, created with clone() and the PID is different for the child

Initially adding more threads will lighten the load on the CPU. 

 The limit is called the "core limit", which can be set with setrlimit. 

 The effect of too many competing threads will be thrashing. 

MPI_Init()

MPI_Finalize() 

clone

and

fork? 

You can label what a commit did exactly.

You can revert commits.

You can pull only the commits you want. 

You can have multiple developers working on code - merge will bring forth any conflicts to be resolved

A preference for which processor to run your code on. In linux you can set it with taskset.

(this question was to remain blank on midterm #1) 

Define:

turnaround time

response time

CPU utilization

throughput

waiting time 

Turnaround time: from submission of a process to its completion.
Response time: from when a request was submitted until the first response is produced.

CPU utilization: percentage of time the CPU spends working (instead of being idle)

throughput: number of processes that complete their execution per time unit 

waiting time:  time spent in the ready queue

Should each of the following be minimized or maximized, pertaining to performance of a scheduler:

turnaround time

response time

CPU utilization

throughput

waiting time 

minimize: all the ones that say "time"

 maximize: CPU utilization, throughput 

separation of kernel and userspace modes(?)

yes. Userspace applications cannot directly execute kernel code

.config 

For this class, it was in the git repo to be cloned.  Actually it was called config, and needed to be moved down a directory and renamed .config 

What does it mean for a process to be CPU bound?

I/O bound? 

CPU bound spends most of its time in the CPU.

I/O bound spends most of its time waiting for input from the user. 

Gives its N processes each 1/N potential share. Each quantum it reevaluates whether each needs this full share and most times is redistributing most of this to CPU bound processes.

whatever has run for the least amount of time gets scheduled next.

context switching: booting one process off the CPU (steel-toed boot) and putting on the next.

context_switch() 

bob

bob123bob