Browse Prior Art Database

Maintaining L2 Contents as Superset of L1-Cache Contents

IP.com Disclosure Number: IPCOM000105369D
Original Publication Date: 1993-Jul-01
Included in the Prior Art Database: 2005-Mar-19
Document File: 4 page(s) / 135K

Publishing Venue

IBM

Related People

Rechtschaffen, R: AUTHOR

Abstract

Several control algorithms within L1-caches rely on the fact that the contents of the L1-Cache are a subset of the L2-Cache. The manner in which this can be achieved is disclosed.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 34% of the total text.

Maintaining L2 Contents as Superset of L1-Cache Contents

      Several control algorithms within L1-caches rely on the fact
that the contents of the L1-Cache are a subset of the L2-Cache.  The
manner in which this can be achieved is disclosed.

      Within a WTWAX cache the subset property is also advantageous
as it guarantees that the store through will not miss.  A WTWAX cache
management protocol is defined as:

      o   all stores are written through the L1 cache to the L2 (WT),

      o   all lines that are stored into by the processors must be
          allocated (WA - WRITE ALLOCATE), and
      o   all lines written into must be held exclusively (X).

            In order to maintain the maximum flexibility of access to
      the L1-Caches within a multiprocessing system and to prevent
      multiple copies of the same line to co-reside in the same
      cache, a means of determining that a line is in the L1-Cache,
      when the congruence class through which it is accessed does not
      have the line, is to examine the L2 directory and determine
      which congruence class on the given processor has the line.
      The implication of such an algorithm is that the L2-Cache has
      several properties:

      o   Unambiguous Determination

                The directory of the L2 is organized via the Real
          Congruence Class (RCC).  That is, enough bits from the real
          address of the line are used to specify the congruence
          class within the L2-Directory so that the determination as
          to the residency of a line in the L1-Cache will be
          unambiguous.
      o   L1-Cache Position Pointers

                The congruence class of the line in the L1-Cache for
          each L1 in which it resides will be found associated with
          the L2-Directory entry.
      o   Subsetability

                As a corollary of the previous points the contents of
          the all L1-Directories are a subset of the contents of the
          L2-Directory.

      The algorithm that assures Subsetability need only be activated
      at the point of an L2-Cache miss.  If following every L2-Cache
      miss the subset property is guaranteed and the L2-Cache
      contents change only at such points, the Subsetability property
      is maintained.  Therefore whatever alters the L2-Directory
      contents must be considered a miss and this includes
      invalidations by I/O and other processors.  The invalidation of
      lines in the L2 must be considered as a "special miss."  Such a
      "special miss" is handled by invalidation of associated
      L1-Directory contents using the information available in the
      L2-Directory...