A Comparative Study of the Linux and Windows Device Driver Architectures with a focus on IEEE1394 hi

1
A Comparative Study of the Linux and Windows
Device Driver Architectures with a focus on
IEEE1394 (high speed serial bus) drivers

• Melekam Tsegaye
• g98t4414_at_campus.ru.ac.za

Supervisors Prof. Richard Foss, Bradley Klinkradt
2
Overview

• The Linux and Windows driver architectures
• The Linux and Windows IEEE 1394 driver stacks
• Driver development on the two platforms
• Results of the study

3
Why the study was conducted ?

• Microsoft Windows and Linux
• Two of the most popular operating systems
• No previous comparisons of their driver
  architectures done by other researchers
• IEEE 1394 (firewire)
• Popular bus offering high data transfer rates
• Active research area for the CS departments
  Audio Engineering Group

4
IEEE 1394

• Data Transfer rates of 100, 200 and 400 Mbps
• Isochronous mode of transfer
• Guaranteed bandwidth (80)
• Asynchronous mode of transfer
• Guaranteed packet delivery (20 or more)
• Fully plug n play

5
IEEE 1394 Consumer Products
6
Device Drivers

• A driver is a piece of software that extends a
  kernels functionality
• Drivers enable applications (through the kernel)
• to transfer data to and from a device
• to control the a device to allow modification of
  its attributes
• Composed of a set of routines that a kernel calls
  at appropriate times e.g. read/write

7
Typical driver routines

• A driver would implement
• Initialisation
• Cleanup
• Open
• Read
• Write
• I/O Control (ioctl)
• Close

8
The Windows driver architecture

• Standard model
• The Windows Driver Model (WDM)
• bus, functional filter drivers
• PnP Power management
• Communication
• I/O request packets (IRPs)

9
The Linux driver architecture

• No standard driver model
• Drivers are modules
• No PnP Power management message dispatching
• Communication through direct function calls

10
Side by side comparison of the two driver
architectures
Linux
Windows
11
IEEE 1394 driver stacks

• The Windows IEEE 1394 stack
• closed source
• maintained by Microsoft
• stable
• The Linux IEEE 1394 stack
• open source
• maintained by the Linux 1394 community (private
  individuals)
• tagged experimental

12
The Windows IEEE 1394 stack

• Host controller
• Bus driver
• Client drivers
• Communication
• I/O request block (IRB)

13
The Linux IEEE 1394 stack

• Host controller
• Bus drivers
• Client drivers
• Communication
• direct function calls
• Direct driver possible

14
Side by side comparison of the two IEEE 1394
stacks
Linux
Windows
15
The Windows IEEE 1394 implementation
16
The Linux IEEE 1394 implementation
17
What operations should IEEE 1394 client drivers
provide ?

• Asynchronous Operations
• Read
• Write
• Lock
• Listen
• Register Asynchronous Listening
• Deregister Asynchronous Listening
• Isochronous operations
• Listen
• Talk
• Asynchronous streaming

18
What operations should IEEE 1394 client drivers
provide ? (Continued)

• Bus reset handling
• Register Bus Reset Handler
• Generate Soft Bus Reset
• Obtain Bus Information (e.g. node count)
• Local configuration ROM manipulation
• Network troubleshooting
• Pinging of nodes

19
Device Driver Development Environments
20
Issues to be considered when creating drivers

• Memory management
• The kernel provides memory allocation/de-allocatio
  n routines
• Has two pools of memory (swappable,
  non-swappable)
• Data structures
• The kernel provides implementations for queues,
  lists, stacks
• Synchronisation with spin locks, mutexes,
  semaphores, signal objects

21
Issues to be considered when creating drivers
(Continued)

• Drivers routines
• must be re-entrant i.e. should be executable by
  multiple threads with no problems
• avoid recursion
• Hardware Abstraction Layer
• use HAL routines provided by the kernel to access
  hardware

22
Map of software produced during this study
23
Windows raw1394 driver
24
Linux raw1394-2 driver
25
Linux Ieee1394diag
26
Driver Tests

• Tests were run to determine
• highest data transfer rates that can be achieved
  by each of the IEEE 1394 drivers
• how close these came to the theoretical maximum
  of 50MB/s
• Bus analyser used to measure
• the duration of data transfers
• the amount of data transferred
• Calculated
• the data transfer rate in MB/s

27
Test results

• Isochronous mode
• Windows raw1394
• 8.9 MB/s
• packet size 1024 bytes
• Linux raw1394
• 4.5 MB/s
• packet size 512 bytes
• Isochronous mode (buffered)
• Linux raw1394-2
• 8.9 MB/s
• packet size 1024 bytes
• Windows raw1394
• 17.7 MB/s
• packet size 2048 bytes

28
Test results (continued)

• Drivers from both Linux and Windows do not
  transfer data at a rate close to the theoretical
  
• 50MB/s
• They achieve almost half that
• due to the PCI bus bottleneck (theoretical
  132MB/s)
• In practice have that available
• shared by all I/O devices attached to the PCI bus
• The PCI bus has latency problems
• driver implementation may not be efficient

29
Conclusion

• A comparison of the Windows and Linux driver
  architectures has shown that
• Drivers for the two platform have similar
  components
• composed of routines for performing
• I/O and device control
• drivers are modules which are loadable at
  runtime
• Windows has a formally defined driver model,
  Linux does not.
• PnP and power management support integrated in
  the Windows architecture, not so in Linux

30
Conclusion (continued)

• A comparison of the Windows and Linux IEEE 1394
  stacks has shown that
• Their IEEE 1394 driver stacks are broken up into
  similar layers
• host controller, bus and client driver layers
• Each stack provides the standard IEEE 1394
  operations (both asynchronous and isochronous)
• The Linux IEEE 1394 stack is open source while
  the Windows stack is proprietary
• Easier to create IEEE 1394 drivers for Linux than
  Windows since all source code is available

31
Future work

• IEEE 1394.1 bridge awareness
• IEEE 1394.1
• allows extending the no. of nodes to 64K instead
  of the current 63
• still in draft phase
• The Windows and Linux IEEE 1394 driver developers
  did not take into account IEEE1394.1 bridging in
  their design
• This study identified bridge awareness
  requirements
• implementation possible for the Linux 1394 stack
• not for the Windows 1394 stack (closed source)
• Full IEEE 1394.1 implementation and testing not
  done during this study
• no IEEE 1394.1 bridging hardware available
• the standard may change

32
Questions ?