Kernel Synchronization in Linux Uniprocessor and Multiprocessor Environment

1
Kernel Synchronization in Linux Uni-processor and
Multi-processor Environment
By Kathryn Bean and Wafa Jaffal (Group A3)
2
Topics

• Linux History
• Kernel Control Path
• Synchronization Technique
• SMP Architecture
• Hardware Support for Synchronization
  (Pentium-based Architecture)
• Linux/SMP kernel
• Conclusion

3
Linux History

• University of Helsinki (1997), Masters thesis
• Linux, a Portable Operating System by L.
  Torvalds
• OS for IBM-compatible personal computers (Intel
  80386 microprocessor).
• Source code under GNU General Public License

4
Kernel Control Path

• Kernel control path is the sequence of
  instructions executed in Kernel Mode to handle a
  kernel request.
• Kernel control path executes due to the following
  reasons
• System calls
• Exceptions
• Interrupts

5
Synchronization Technique

• Nonpreemptability
• Atomic Operations
• Interrupt Disabling
• Locks

6
Condition to be Preempted

• Kernel control path can preempt a running
  process however, when an interrupt handle
  terminates, the process resumes.
• Only kernel control path can interrupt another
  kernel control path.

7
Atomic Operation

• An atomic operation - performed by executing a
  single assembly language instruction
• Linux kernel provides special functions such as
• atomic_int(v) ? v

8
Interrupt Disabling

• Because of its simplicity, interrupt disabling is
  used by kernel functions for implementing a
  critical region.
• This technique does not always prevent kernel
  control path interleaving.
• Critical section should be short because any
  communication between CPU and I/O is blocked
  while a kernel control path is running in this
  section.

9
Locking

• Two kinds of locking
• Kernel semaphores, used by both uni-processor and
  multiprocessor systems
• Spin Locks, used by only multiprocessor systems

10
Kernel Semaphore Implementation

• Kernel semaphore is object of type structure
  semaphore, see include/asm/semaphore.h file
• Fields
• count integer number
• count gt 0 semaphore is available
• count ? 0 semaphore is busy, count - number
  of processes waiting for resource.
• Count 0 one use, nothing is waiting
• The count field is decremented when a process
  acquires the lock and is incremented when the
  same process releases it.

11
Kernel Semaphore Implementation, Continued

• wait the address of a wait queue.
• waking integer.
• The releasing process increments waking field(s).
• Each of awakened process PI then enters a
  critical region of the down() function
• Is PIs waking ltgt 0, if waking gt 0 1. acquire
  the resource
• 2. other PKs waking--
• if waking lt 0 go back to sleep

12
Kernel Semaphore Implementation, Continued

• Function
• down() called, if process wishes to acquire a
  semaphore.
• count --
• count ltgt 0, if count ? 0 process enter the
  critical section
• if count lt 0 process is suspended
• up() called, if process releases a semaphore
• count
• count ltgt 0 if count gt 0 up() terminates
• if count lt 0 wake up
  other processes
• Deadlock semaphore requests are performed in
  the address order.

13
SMP Architecture

• Scalability of Linux - supports multiprocessing
  through Shared Memory Symmetric Multiprocessors
  (SMM) architecture.
• scalability is the capability of a system to
  adapt to an ever-increasing work load.

14
SMP Architecture, Continued
CPU 1
CPU n
system bus
Graphical card
Memory

• CPUs share the same memory unit
• application processing and kernel processing are
  spread amongst all CPUs.

15
Other Multiprocessor Architectures

• Asymmetric Multiprocessing
• Master CPU executes the operating system code
  and application programs run on the remaining
  CPUs.
• Massively Parallel Processing (MPP)
• Assemble hundreds or thousand of CPUs, each with
  own system memory

16
Hardware Support for Synchronization

• Shared Memory
• Memory arbiter (chip between bus and every RAM
  chip) grants access to a CPU if the chip is free
  and delays access if the chip is busy.
• Cache Synchronization
• Hardware cache is utilized using the locality
  principle. In multiprocessor environment, each
  CPU has its own cache. Process of updating cache
  -cache snooping

17
Hardware Support for Synchronization, Continued

• SMP Atomic Operation
• Lock instruction prefixes for atomic operations
  were introduced.
• If control unit detects them ? lock the memory
  bus, no other processes can access this memory
  location

18
Hardware Support for Synchronization, Continued

• Distributed Interrupt Handling

CPU 1
CPU n
Local APIC
Local APIC
ICC bus
I/O APIC
APIC Advanced Programmable Interrupt
Controller ICC Interrupt Controller
Communication I/O APIC - router
IRQ lines
19
Linux/SMP Kernel

• Process Descriptor Modification
• has_cpu if has_cpu gt 0 process is running
• Processor logical number of its CPU
• Spin Locks
• Blocked process keeps its own CPU by spinning
  while waiting for a resource.

20
Conclusion

• Modern versions of Linux are available
• Compaq Alpha
• SPARC
• PowerPC
• Motorola MC680x0
• IBM System/390
• Multiprocessor operating system
• Supports up to 32 CPUs

21
Bibliography

• D. P. Bovet, M. Cesati. Understanding the Linux
  Kernel. OReilly, 2000
• Linus Torvalds. Linux a Portable Operating
  System. Master of Science Thesis, University of
  Helsinki, Finland, 1997
• D. Mosberg, S. Eranian IA-64 Linux Kernel
• Prentice Hall PTR, 2002

22
Thank You

• Any Questions?