US 4,833,604 AGrant
Method for the Relocation of Linked Control Blocks
Issue Date:1989-05-23
•6 Claims
•4 Drawing Sheets
Abstract
A two-pass method for relocating a set of linked control blocks stored away on a persistent medium after a first pass and then rewritten into internal memory of a computing facility during second and subsequent passes each time an application to which the control blocks are bound is executed. The first pass involves path following and coloring pointers affected by the relocation, mapping discontiguously located blocks into a linear address space, changing affected control blocks to location offsets, and writing out the linked control blocks to DASD store. Upon the second pass, virtual addresses are substituted for the offsets upon rewriting of the control blocks to internal memory.
Metadata
Assignee
- International Business Machines Corporation
Inventors
- Josephine M. Cheng
- Nicholas V. Nomm
- Jay A. Yothers
Application Information
Application Number:US 8184321
Filing Date:1986-01-13
Priority Date:1986-01-13
Art Unit:237
Classifications
IPC:
G06F 1516
Field of Search:
364200 MS File;300;900 MS File
Patent Drawings (4 sheets)
Description
Technical Field
This invention relates to the relocation of executable control structures, and more particularly, to the relocation of a set of linked control blocks in a computing facility where the control blocks must be stored away on a persistent medium and then rewritten into the internal memory portion of the computing facility each time an application to which the control blocks are bound is to be executed.
Background
It is broadly old to dynamically relocate code at runtime given that the code to be executed is larger than the available internal memory of a computing facility. Segmentation, overlays, and libraries are among the mechanisms which permit nonresident code to be called in from external memory, such as DASD, only as needed. However, the binding of the code to internal memory addresses must still be imposed on all of such referenced segmentations, overlays, and library elements.
Peterson and Silberschatz, "Operating System Concepts", Addison-Wesley Publishing Co., copyright 1983, pp. 137-139, describe one method of dynamically managing code relocation in which internal memory address binding of the code is delayed until execution (runtime). The code is mapped in Peterson's scheme into a linear address space formed by a base plus an offset to that base. In this regard, the base is supplied by a dedicated register, while the offset is supplied extrinsically.
Another method of dynamically managing code relocation in which address binding is delayed until runtime is set out in Drogichen, U.S. Pat. No. 4,138,738, "Self-contained Relocatable Memory Subsystem", issued Feb. 6, 1979. Drogichen teaches that ROM-stored code/data sequences may be relocated in CPU internal store if additional stored bits appending each code/data word are used to designate a set of external registers. The contents of these designated registers are operands in the computation of an effective internal memory address for the counterpart code/data word being moved.
In the field of data base management systems, C. J. Date, "A Guide to DB2", Addison-Wesley Publishing Co., copyright 1984, discloses the compile and runtime involvements of the elements of a relational system for accessing information. The information typically is located on a staged storage system and is obtained by way of a computing facility. The data is organized by the computing facility under any one of the popular storage and accessing methods such as IMS, VSAM, or ISAM. Relational accessing is executable in either an interactive or deferred execution mode. As Date points out, relational accessing has provided a popular and flexible manner by which subsets of data are obtained by imposing relations through a query. The advantage of relational accessing is that a near-infinite number of relations can be imposed upon data without the necessity of storing concordances representing those relations.
When a data base system is operated in an interactive mode, an access statement is processed by the data base manager interpretively by way of tables. Since the tables per se are not stored on a persistent medium, such as DASD, they must be formed each time and made resident in main memory for the duration of the interactive processing. There is no perceived need for relocation. In contrast, deferred accessing involves relational statements which are embedded in an application. The application itself may be executable at multiple disparate times. For this reason, optimized compiled machineexecutable code is a necessary condition for such processing. When the application is being compiled, the control blocks, such as skeleton cursor tables, constituting each relational access are formed and associated with or "bound" to the application. These skeleton control blocks are then stored on a persistent medium, such as DASD. At execution or "runtime", the control blocks must then be relocated from DASD to main memory in a form ready for utilization.
A relational system includes more than merely a compiler and link loader. It is, according to Date, formed from a precompiler, a binder, a runtime supervisor, and a stored data manager. The precompiler collects relational accessing statements from application programs and forms data base request modules. The accessing statements in the original program are replaced by host language calls to the runtime supervisor. The binder is the active component responsive to each of the request modules for producing optimized machine-executable code implementing the relational accessing statements. This code is generated in the form of control blocks which may be considered as a specialized form of a linked list.
The Invention
It is an object of this invention to devise a method for the relocation of linked control blocks into the internal memory portion of a computing facility each time an application to which the control blocks are bound is to be executed.
This object is satisfied by a two-pass method. During a first pass, preferably at bind time, the method path follows through the set of discontiguous linked control blocks identifying the relative order of the blocks forming each path, maps the path-followed set into a linear address space, and converts any links affected by the mapping into corresponding location offsets in the linear address space. The mapped linked control blocks, including the converted links in their linear-mapped order, are then stored onto a persistent medium. During second and subsequent passes, the method involves rewriting the set of linked control blocks from the persistent medium to the internal memory. It includes substituting virtual addresses for location offsets in the affected links.
During the first pass, the mapping step includes forming a directory for each address space of control blocks. The directory is a list of offsets to pointers which require relocation. Further, the directories are collectively appended to the linear-mapped order of link-listed control blocks. Significantly, within the mapping step, the offsets within the directory maintain and preserve the path-following order upon rewriting of the stored linked lists from the persistent medium to the internal memory during second and subsequent passes.
Brief Description of the Drawing
FIG. 1 shows the control flow for processing applications with embedded relational accessing statements according to the prior art.
FIG. 2 shows a sample of linked control blocks in discontiguous locations within internal memory, including pointers defining various in and out degree graph relations and paths or loops.
FIG. 3 sets forth a compressed and contiguous form of said linked control blocks in a linear space.
FIG. 4 depicts some of the information elements contained in each control block.
Description of the Preferred Embodiment and Industrial Applicability
Referring now to FIG. 1, there is shown a flow of control for the processing of applications software, including embedded relational accessing statements. Each source module of application software is parsed by the precompiler. Tokens or high-level language calls replace each relational accessing statement so as to modify the source module. The modified source modules are compiled into object modules and processed in a conventional manner. The relational accessing statements are also formed into "modules" and compiled by way of the binder so as to form a set of such modules. Both the control blocks forming the application software and the relational accessing statements are incorporated as sets of link-listed control blocks and are stored outboard on a staged storage system attaching a computing facility.
Referring now to FIG. 2, there is shown a linked list of control blocks (CBs) A-E residing in discontiguous locations within the internal memory portion of the computer facility. The invention provides for the relocation of these blocks. The most commonly occurring pointer relations are illustrated. Control block A, fields 1 and 2, respectively point to control blocks B and C, while control block B contains only a single field, field 3. Field 3 is directed to control block C. In turn, control block C has fields 4 and 5 pointing to two other control blocks, namely D and E. Control block C in the graph theoretic sense is in-degree two since it is pointed to by two or more control blocks. There also exists a loop relation among control blocks. This loop is formed by blocks A, B, C, D, and E. Both the varying in and out degrees of each node and loop complicate any path following. The relocation of such linked control blocks has therefore a high expectation of involving an error condition arising from an attempt to relocate the same block more than once, resulting in an ABEND.
It should be recognized that there are several types of control blocks. A first type exhibits a fixed length. Any pointer fields in each type are always located in a predetermined position. On the other hand, a variable length block, such as control block E, consists of a header portion, several pointer entries, and counterpart control blocks embedded therein.
At compile time, it is necessary to trace the paths of any set of linked control blocks. This path or trace provides a relative ordering among the blocks which must be preserved invariant upon relocation. Having the path-following information in tabular form, it is then possible to focus on the next step in the relocation method. The next step concerns the conversion of blocks discontiguously located in internal memory into a format for recording on DASD and later to be read back into internal storage. Restated, it is desired to map the block relative starting addresses into counterpart virtual base plus displacement addresses. This means that the pointer fields in a control block must be converted into offsets upon moving the control blocks to DASD. It should be again emphasized that when rewriting from external to internal memory, the original absolute addresses are no longer applicable.
Referring now to FIG. 3, there is shown a compressed and contiguous form of the linked control blocks, including a relocation directory appended thereto. The directory simplifies the reconversion of offsets into addresses upon relocation back from DASD into internal memory. The relocation directory is a list of fields requiring relocation. This avoids a path-following (tree-walking) algorithm when offsets are converted to addresses. Path following requires a more complex behavior than merely reading a list. Furthermore, a directory is a data object generated as a consequence of path following.
It should be appreciated that with a directory, the restoration of offsets to addresses merely requires adding the field contents to a common starting address and rewriting the result into the field. Locating the next field to relocate only requires indexing to the next element in the directory, instead of path following to find the next pointer field to relocate. Another possible algorithm to relocate from a pointer stack into addresses would be to use the path-following algorithm again.
As previously mentioned, the first pass in the invention occurs during the bind (compile equivalent) process. The reconversion from DASD to internal memory relocation occurs at runtime.
Referring now to FIG. 4, there is illustrated a control block description including such attributes as name, length, number of pointers, and pointer attributes. This data object description facilitates relocation and it is used in the following manner:
At the first pass (at bind time), it is necessary to calculate the total length of discontiguous blocks and then copy discontiguous blocks into contiguous space. Next, it is required to build relocation directories for all pointers. This means that in each control block, if it has already been relocated, consider it to be "colored" in the classical "greedy algorithmic sense". Note that network path following to ascertain or alter properties of nodes or links is described in Aho et al, "Data Structures and Algorithms", Addison-Wesley Publishing Co., copyright 1983, pp. 7-9, 321-324. In this reference, Aho describes the "Greedy Algorithm" for coloring nodes and/or links while traversing a graph of nodes and links. To continue, if the control block has not been marked as relocated, then relocate pointers in each header, in each entry, and in each nonstandard repeating block. The term "relocate" is taken to mean converting an old address into an offset in a contiguous copy of blocks, and adding the relative location of the pointer to one of three lists of pointers requiring relocation. These lists are:
1. pointers pointing to locations in the same address space,
2. pointers pointing to locations in another address space, and
3. pointers not pointing, but instead containing offsets in still another address space.
Note the path following is directed by a structure for the relocation of structures as shown in FIG. 4.
Referring now to Table 1, there is shown a high-level PL/I-like software fragment for relocating pointers from a discontiguously located control block into a contiguous or linear address space. As pointed out in the comment section of the code, there are necessary assumptions which must exist in order for the code to be operable in the computing facility. As is typical in block-structured languages, it is preceded by a set of declarations and then a series of nested loops. The code fragment functionally in block-structured pseudocode operates as follows:
While there are still unprocessed control blocks in the linked list:
Having shown and described the preferred embodiment of the present invention, those skilled in the art will realize that various omissions, substitutions, and changes in forms and details may be made without departing from the spirit and scope thereof. It is the intention, therefore, to be limited only as indicated by the scope of the following claims.
Claims
What is claimed is:
1. A method for relocating a set of linked control blocks stored on a persistent medium after a first pass and then, during second and subsequent passes, rewritten into an internal memory of a computing facility each time an application to which the control blocks are bound is executed, comprising the steps of: (a) during the first pass, path following through the set of control blocks which may be of different size and format using a path-following algorithm that takes into account the contents of each block; coloring pointers to retain a history of the first time each specific path is followed through specific linked control blocks of the set to preclude further relocation of the same ones of the control blocks in each such path; mapping, into a linear address space, discontiguously located ones of said blocks by changing the colored pointers to location offsets; compressing said linked control blocks contiguously into the linear address space; and writing the control blocks in compressed form to the persistent medium; and (b) during second and subsequent passes, rewriting the control blocks to the internal memory from the persistent medium and substituting virtual addresses for the location offsets.
2. A method according to claim 1, wherein during said first pass, the step of mapping includes forming a directory for the linear address space that contains the compressed control blocks, said directory listing pointers requiring relocation.
3. A method according to claim 2, wherein the step of mapping includes calculating the offsets within the directory in a manner which maintains and preserves the path-following order upon rewriting of the stored linked control blocks from the persistent medium to internal memory during the second and subsequent passes.
4. A method according to claim 2, wherein the directory is appended to the linear address space.
5. A method according to claim 1, wherein the control blocks may be of different types, each type having a positionally different pointer location, and further including, during the coloring step, identifying the type of control block pointed to.
6. A method according to claim 1, wherein the linked control blocks comprise a network of heterogeneously linked blocks.