Capability-Based Addressing

1
Capability-Based Addressing

• By R.S. Fabry
• University of California
• Presented by Luis Vazquez
• October 4, 2005

2
Outline

• Introduction
• Shared Segment References Problem
• Solutions
• Hardware Implementations
• Conclusion

3
What is a Capability?

• Data structure that represents access to an
  object.
• Used with protection schemes since possessing a
  capability grants a particular type of access
  (read/write/execute/etc).
• Capabilities can only be created by the system
  and cannot be modified by non-system programs.
• Idea developed by Dennis and Van Horn.

4
Using Capabilities for Addressing

• Must be able to store capabilities in
  user-defined structures
• Scheme must be in place to ensure integrity of
  these capabilities
• Capabilities were used for protection, but Fabry
  wants to use capabilities to provide an absolute
  address to some object.

5
Context-Independent versus Context-Dependent

• Allocating jobs into fixed areas of memory
  allowed for context-independent addresses.
• Address relocation was introduced, but so were
  context-dependent addresses to objects.
• Complicates the sharing of addresses
• Capabilities provides a context-independent way
  to access these objects.
• The capability can point to a real object or a
  virtual object (which is handled by the system).
• Allows the system to support address relocation.

6
Shared Segment Reference Problem

• Looking at the problem when no sharing of
  objects.
• PC loads segment reference 0, which tells us to
  call the subroutine in segment reference 1 and
  then access the data in segment reference 2.

7
Shared Segment Reference Problem

• Assume the relationship between integers and
  segments are created independently.
• How should the segment references be coded?

8
Shared Segment Reference Solutions

• Fabry discusses what he says are the most
  developed and promising solutions which are
• Uniform Address Solution
• Indirect Evaluation Solution
• Multiple Segment Table Solution
• Capability Addressing Solution

9
Uniform Address Solution

• All shared integer statements are functionally
  equivalent.
• Functions must be defined centrally.
• Used by the Burroughs systems.

10
Indirect Evaluation Solution

• Shared integer segment address is an index to a
  linkage segment.
• Linkage segments contain segment table indexes.
• Base registers remembers the address of the
  linkage statement
• Concept used in Multics

11
Multiple Segment Table

• Modifies indirect evaluation solution.
• Replace indexes with capabilities.

12
Capability Addressing Solution

• Processor registers used to indirectly evaluate
  data addresses.
• Capabilities for subroutines are embedded into
  the program.

13
Difficulties in Implementation

• Integrity of Capabilities
• Address Translation
• Instruction Sets
• The Stack

14
Integrity of Capabilities

• Due to the access lists in a capability, user
  programs should not change the bit pattern.
• Tagged Contents of a word is denoted as a
  capability or not using one or more tag bits.
• Partition Segments, at creation, can only hold
  either capabilities or data.

15
Partitioned Approach

• Advantages
• Locating of capabilities are quicker since they
  reside in known locations.
• No tag bits to occupy memory.
• Disadvantages
• Some objects require both data and capabilities,
  and extra capabilities to couple these segments
  is preferred. This leads to more memory usage.
• Paper mentions it is possible to translate from
  one approach to another.

16
Address Translation

• Capability is seen as an address to an object
  (which is made up of a bit pattern)
• A capability is seen as in-form or out-form.
• In-form Capabilities for segments in primary
  memory. Are not allowed to be on secondary
  storage.
• Out-form Capabilities for segments in secondary
  storage.

17
In-Form Capability

• Contains
• Absolute address of origin of the segment in
  primary memory
• Length of the statement
• Converted to out-form when copied to secondary
  storage

18
Out-Form Capability

• Contains
• Secondary storage address of the first record of
  the segment
• Unique sequence number used to invalidate
  capabilities for objects that no longer exist.
• Using an out-form capability to access a segment
  results in a trap to the system.

19
Performance of In-Form and Out-Form Capabilities

• The system must continue to convert capabilities
  to and from In-Form and Out-Form Capabilities
  depending on where the object is located, and
  when its needed.
• This provides a lot of overhead to be imposed on
  the system for address translation.

20
Future Implementations of Address Translations

• Capabilities have a unique code (never to be
  reused)
• Use a Hash Table using the unique code as a key
  for all objects in primary memory
• Presence bit (determine if segment is in primary
  memory), primary memory location, secondary
  memory location, and segment size
• Entries age over time and are removed.
• Accessing the Hash Table results in 3 outcomes
• Segment found in HT and in Primary Memory
• Type A Exception Entry found in HT but segment
  not in Primary Memory
• Type B Exception Entry not found in HT

21
Accessing the Hash Table

• Type A Exception O/S reads in the segment from
  secondary storage (it knows the location and size
  of the segment from the HT) and updates the HT
  entry accordingly (presence bit).
• Type B Exception O/S needs to determine the
  size and location of the segment and creates a
  new HT entry. Then raises a Type A Exception to
  copy this segment into primary memory.

22
Instruction Sets

• Capabilities should be able to be copied around
  freely, since to the user, these are simply
  addresses to objects.
• Enter instruction can be seen as a call and is
  used to call subroutines.
• Enter access" is used for a transfer of control
  and is weaker than read, write, or execute
  access.
• Control is transferred to a fixed entry point.
• Calling program cannot access called programs
  embedded capabilities.

23
The Stack

• Stack is used when dealing with calls to
  subroutines where stack frames are given to
  subroutines for temporary storage.
• Solution presented
• Creating a new temporary stack frame for each
  subroutine.
• Also helps keeping the called program from
  modifying the return location of the calling
  program.

24
Conclusion

• Capability-based Addressing provides an efficient
  type of absolute address for an object.
• Appears to simplify the amount of indirection
  needed for address relocation.
• Paper discussing design suggestions for computer
  architects to make use of such schemes.

25
References

• Fabry, R.S. Capability-Based Addressing,
  University of California, 1974.