This RFC specifies a proposed protocol for the ARPA Internet community, and requests discussion and suggestions for improvements. Distribution of this memo is unlimited.
Table of Contents
1 Introduction.......................................... 1 1.1 Purpose of This Document............................ 1 1.2 Summary of Features................................. 2
2 General Description................................... 3 2.1 Motivation.......................................... 3 2.2 Relation to Other Protocols......................... 4 2.2.1 Transport Service Requirements.................... 5
The Loader-Debugger Protocol (LDP) is an application layer protocol for loading, dumping and debugging target machines from hosts in a network environment. This protocol is designed to accommodate a variety of target cpu types. It provides a powerful set of debugging services. At the same time, it is structured so that a simple subset may be implemented in applications like boot loading where efficiency and space are at a premium.
The authors would like to thank Dan Franklin and Peter Cudhea for providing many of the ideas on which this protocol is based.
This is a technical specification for the LDP protocol. It is intended to be comprehensive enough to be used by implementors of the protocol. It contains detailed descriptions of the formats and usage of over forty commands. Readers interested in an overview of LDP should read the Summary of Features, below, and skim Sections 2 through 3.1. Also see Appendix B, the Command Summary. The remainder of the document reads best when accompanied by strong coffee or tea.
LDP is an application protocol that provides a set of commands used by application programs for loading, dumping and debugging target machines across a network.
The goals of this protocol are shown in the following list:
o The protocol should support various processor types and operating systems. Overhead and complexity should be minimized for simpler cases.
o The protocol should provide support for applications in which more than one user can debug the same target machine. This implies an underlying transport mechanism that supports multiple connections between a host-target pair.
o LDP should have a minimal subset of commands for boot loading and dumping. Target machine implementations of these applications are often restricted in the amount of code-space they may take. The services needed for loading and dumping should be provided in a small, easily implemented set of commands.
o There should be a means for communicating exceptions and errors from the target LDP process to the host process.
o LDP should allow the application to implement a full set of debugging functions without crippling the performance of the target's application (i.e., PSN, PAD, gateway). For example, a breakpoint mechanism that halts the target machine while breakpoint commands are sent from the host to the target is of limited usefulness, since the target will be unable to service the real-time
LDP is an application protocol that fits into the layered internet protocol environment. Figure 1 illustrates the place of LDP in the protocol hierarchy.
o allow connections to be opened by specifying a network (or internet) address. Support passive and active opens.
o for each connection, specify the maximum message size.
o provide a mechanism for sending and receiving messages over an open connection.
o deliver messages reliably and in sequence
o support multiple connections, and distinguish messages associated with different connections. This is only a requirement where LDP is expected to support several users at the same time.
o explictly return the outcome (success/failure) of each request (open, send, receive), and provide a means of querying the status of a connection (unacknowledged message count, etc.).
Data is passed from the application program to the LDP user process in the form of commands. In the case of an LDP server process, command responses originate in LDP itself. Below LDP is the transport protocol. The Reliable Data Protocol (RDP -- RFC 908) is the recommended transport procotol. Data is passed across the LDP/RDP interface in the form of messages. (TCP may be used in place of RDP, but it will be less efficient and it will require more resources to implement.) An internet layer (IP) normally comes between RDP and the network layer, but RDP may exchange data packets directly with the network layer.
Figure 2 shows the flow of data across the protocol interfaces:
An LDP session consists of an exchange of commands and responses between an LDP user process and an LDP server process. Normally, the user process resides on a host machine (a timesharing computer used for network monitoring and control), and the server process resides on a target machine (PSN, PAD, gateway, etc.). Throughout this document, host and target are used as synonyms for user process and server process, respectively, although in some implementations (the Butterfly, for example) this correspondence may be reversed. The host controls the session by sending commands to the target. Some commands elicit responses, and all commands may elicit an error reply.
The protocol contains five classes of commands: protocol, data transfer, management, control and breakpoint. Protocol commands are used to verify the command sequencing mechanism and to handle erroneous commands. Data transfer commands involve the transfer of data from one place to another, such as for memory examine/deposit, or loading. Management commands are used for creating and deleting objects (processes, breakpoints, watchpoints, etc.) in the target machine. Control commands are used to control the execution of target code and breakpoints. Breakpoint commands are used to control the execution of commands inside breakpoints and watchpoints.
An LDP session consists of a series of commands sent from a host LDP to a target LDP, some of which may be followed by responses from the target. A session begins when a host opens a transport connection to a target listening on a well known port. LDP uses RDP port number zzz or TCP port number yyy. When the connection has been established, the host sends a HELLO command, and the target replies with a HELLO_REPLY. The HELLO_REPLY contains parameters that describe the target's implementation of LDP, including protocol version, implementation level, system
type, and address format. The session terminates when the host closes the underlying transport connection. When the target detects that the transport connection has been closed, it should deallocate any resources dedicated to the session.
The target process is the passive partner in an LDP session, and it waits for the host process to terminate the session. As an implementation consideration, either LDP or the underlying transport protocol in the target should have a method for detecting if the host process has died. Otherwise, an LDP target that supported only one connection could be rendered useless by a host that crashed in the middle of a session. The problem of detecting half-dead connections can be avoided by taking a different tack: the target could allow new connections to usurp inactive connections. A connection with no activity could be declared 'dead', but would not be usurped until the connection resource was needed. However, this would still require the transport layer to support two connection channels: one to receive connection requests, and another to use for an active connection.
Each command sent from the host to the target has a sequence number. The sequence number is used by the target to refer to the command in normal replies and error replies. To save space, these numbers are not actually included in host commands. Instead, each command sent from the host is assigned an implicit sequence number. The sequence number starts at zero at the beginning of the LDP session and increases by one for each command sent. The host and target each keep track of the current number. The SYNCH <sequence number> command may be used by the host to synchronize the sequence number.
The convention for the order of data packing was chosen for its simplicity: data are packed most significant bit first, in order of increasing target address, into eight-bit octets. The octets of packed data are transmitted in sequential order.
Page 10
LDP Specification Protocol Operation
Data are always packed according to the address format of the target machine. For example, in an LDP session between a 20-bit host and a 16-bit target, 16-bit words (packed into octets) are transmitted in both directions. For ease of discussion, targets are treated here as if they have uniform address spaces. In practice, the size of address units may vary within a target -- 16-bit macromemory, 32-bit micromemory, 10-bit dispatch memory, etc. Data packing between host and target is tailored to the units of the current target address space.
Figures showing the packing of data for targets with various address unit sizes are given below. The order of transmission with respect to the diagrams is top to bottom. Bit numbering in the following diagrams refers to significance in the octet: bit zero is the least significant bit in an octet. For an explanation of the bit numbering convention that applies in the rest of this document, please see Appendix A.
The packing of data for targets with word lengths that are multiples of 8 is straightforward. The following diagram illustrates 16-bit packing:
7 0 --------------------------------- Octet 0 | WORD 0 bits 15-08 | --------------------------------- Octet 1 | WORD 0 bits 07-00 | --------------------------------- Octet 2 | WORD 1 bits 15-08 | --------------------------------- Octet 3 | WORD 1 bits 07-00 | --------------------------------- * * * --------------------------------- Octet 2n-1 | WORD n bits 07-00 | ---------------------------------
Packing for targets with peculiar word lengths is more complicated. For 20-bit machines, 2 words of data are packed into 5 octets. When an odd number of 20-bit words are transmitted, the partially used octet is included in the length of the command, and the octet is padded to the right with zeroes.
7 0 --------------------------------- Octet 0 | WORD 0 bits 19-12 | --------------------------------- Octet 1 | WORD 0 bits 11-04 | --------------------------------- Octet 2 | WORD 0 03-00 | WORD 1 19-16 | --------------------------------- Octet 3 | WORD 1 bits 15-08 | --------------------------------- Octet 4 | WORD 1 bits 07-00 | ---------------------------------
A subset of LDP commands may be implemented in targets where machine resources are limited and the full capabilities of LDP are not needed. There are three basic levels of target implementations: LOADER_DUMPER, BASIC_DEBUGGER and FULL_DEBUGGER. The target communicates its LDP implementation level to the host during session initiation. The implementation levels are described below:
Page 12
LDP Specification Protocol Operation
LOADER_DUMPER
Used for loading/dumping of the target machine. Includes all protocol class commands and replies; data transfer commands READ, WRITE, MOVE and their responses; control command START and control reply EXCEPTION. Understands at least PHYS_MACRO and HOST addressing modes; others if desired.
BASIC_DEBUGGER
Implements LOADER_DUMPER commands, all control commands, all addressing modes appropriate to the target machine, but does not have finite state machine (FSM) breakpoints or watchpoints. Default breakpoints are implemented. The target understands long addressing mode.
FULL_DEBUGGER
Implements all commands and addressing modes appropriate to the target machine, and includes breakpoint commands, conditional commands and BREAKPOINT_DATA. Watchpoints are optional.
LDP commands are enclosed in RDP transport messages. An RDP message may contain more than one command, but each command must fit entirely within a single message. Network packets containing LDP commands have the format shown in Figure 5.
LDP commands consist of a standard two-word header followed optionally by additional data. To facilitate parsing of multi- command messages, all commands contain an even number of octets. Commands that contain an odd number of data octets must be padded with a null octet.
The commands defined by the LDP specification are intended to be of universal application to provide a common basis for all implementations. Command class and type codes from 0 to 63. are reserved by the protocol. Codes above 63. are available for the implementation of target-specific commands.
The command length gives the total number of octets in the command, including the length field and data, and excluding padding.
Command Class Command Type
Page 16
LDP Specification Commands and Formats
The command class and type together specify a particular command. The class selects one of six command categories, and the type gives the command within that category. All codes are decimal. The symbols given in Figures 7 and 8 for command classes and types are used in the remainder of this document for reference.
The command classes that have been defined are:
Command Class | Symbol ----------------+----------- 1 | PROTOCOL 2 | DATA_TRANSFER 3 | CONTROL 4 | MANAGEMENT 5 | BREAKPOINT 6 | CONDITION 7 - 63 | <reserved>
Command Classes Figure 7
Command type codes are assigned in order of expected frequency of use. Commands and their responses/replies are numbered sequentially. The command types, ordered by command class, are:
Addresses are used in LDP commands to refer to memory locations, processes, buffers, breakpoints and other entities. Many of these entities are machine-dependent; some machines have named objects, some machines have multiple address spaces, the size of address spaces varies, etc. The format for specifying addresses needs to be general enough to handle all of these cases. This speaks for a large, hierarchically structured address format. However, the disadvantage of a large format is that it imposes extra overhead on communication with targets that have simpler address schemes.
LDP resolves this conflict by employing two address formats: a short three-word format for addressing simpler targets, and a long five-word format for others. Each target LDP is required to implement at least one of these formats. At the start of an LDP session, the target specifies the address format(s) it uses in
the Flag field of the HELLO_REPLY message. In each address, the first bit of the mode octet is a format flag: 0 indicates LONG address format, and 1 indicates SHORT format.
The long address format is five words long and consists of a three-word address descriptor and a two-word offset (see Figure 9). The descriptor specifies an address space to which the offset is applied. The descriptor is subdivided into several fields, as described below. The structuring of the descriptor is designed to support complex addressing modes. For example, on targets with multiple processes, descriptors may reference virtual addresses, registers, and other entities within a particular process.
The addressing modes defined below are intended as a base to which target-specific modes may be added. Modes up to 63. are reserved by the protocol. The range 64. to 127. may be used for target-specific address modes.
The address mode identifies the type of address space being referenced. The mode is qualified by the mode argument and the ID field. Implementation of modes other than physical and host is machine-dependent. Currently defined modes and the address space they reference are shown in Figure 10.
Mode | Symbol | Address space -----+----------------------+---------------------------
0 HOST Host 1 PHYS_MACRO Macromemory 2 PHYS_MICRO Micromemory 3 PHYS_I/O I/O space 4 PHYS_MACRO_PTR Macro contains a pointer 5 PHYS_REG Register 6 PHYS_REG_OFFSET Register plus offset 7 PHYS_REG_INDIRECT Register contains address of a pointer
8 PROCESS_CODE Process code space 9 PROCESS_DATA Process data space 10 PROCESS_DATA_PTR Process data contains a ptr 11 PROCESS_REG Process virtual register 12 PROCESS_REG_OFFSET Process register plus offset 13 PROCESS_REG_INDIRECT Process register contains address of a pointer
14 OBJECT_OFFSET Memory object (queue, pool) 15 OBJECT_HEADER System header for an object 16 BREAKPOINT Breakpoint 17 WATCHPOINT Watchpoint 18 BPT_PTR_OFFSET Breakpoint ptr plus offset 19 BPT_PTR_INDIRECT Breakpoint ptr plus offset gives address of a pointer 20 - <reserved> 63
Provides a numeric argument to the mode field. Specifies the register in physical and process REG and REG_OFFSET modes.
ID Field
Identifies a particular process, buffer or object.
Offset
The offset into the linear address space defined by the mode. The size of the machine word determines the number of significant bits in the offset. Likewise, the addressing units of the target are the units of the offset.
The interpretation of the mode argument, ID field and offset for each address mode is given below:
HOST
The ID and offset fields are numbers assigned arbitrarily by the host side of the debugger. These numbers are used in MOVE and MOVE_DATA messages. MOVE_DATA responses containing this mode as the destination are sent by the target to the host. This may occur in debugging when data is sent to the host from the target breakpoint.
PHYS_MACRO
The offset contains the 32-bit physical address of a location in macromemory. The mode argument and ID field are not used. For example, mode=PHYS_MACRO and offset=1000 specifies location 1000 in physical memory.
PHYS_MICRO
Like PHYS_MACRO, but the location is in micromemory.
PHYS_I/O
Like PHYS_MACRO, but the location is in I/O space.
PHYS_MACRO_PTR
The offset contains the address of a pointer in macromemory. The location pointed to (the effective address) is also in macromemory. The mode argument and ID field are unused.
Page 22
LDP Specification Commands and Formats
PHYS_REG
The mode argument gives the physical register. If the register is used by the LDP target process, then the saved copy from the previous context is used. This comment applies to PHYS_REG_OFFSET mode as well. The ID field is not used.
PHYS_REG_OFFSET
The offset is added to the contents of a register given as the mode argument. The result is used as a physical address in macromemory. ID is unused.
PHYS_REG_INDIRECT
The register specified in the mode arg contains the address of a pointer in macromemory. The effective address is the macromemory location specified in the pointer, plus the offset. The ID field is unused.
PROCESS_CODE
The ID is a process ID, the offset is into the code space for this process. Mode argument is not used.
PROCESS_DATA
The ID is a process ID, the offset is into the data space for this process. Mode argument is not used. On systems that do not distinguish between code and data space, these two modes are equivalent, and reference the virtual address space of the process.
PROCESS_DATA_PTR
The offset contains the address of a pointer in the data space of the process specified by the ID. The location pointed to (the effective address) is also in the data space. The mode argument is not used.
PROCESS_REG
Accesses the registers (and other system data) of the process given by the ID field. Mode argument 0 starts the registers. After the registers, the mode argument is an offset into the system area for the process.
The offset plus the contents of the register given in the mode argument specifies a location in the data space of the process specified by the ID.
PROCESS_REG_INDIRECT
The register specified in the mode arg contains the address of a pointer in the data space of the process given by the ID. The effective address is the location in process data space specified in the pointer, plus the offset.
OBJECT_OFFSET (optional)
The offset is into the memory space defined by the object ID in ID. Recommended for remote control of parameter segments.
OBJECT_HEADER (optional)
The offset is into the system header for the object specified by the ID. Intended for use with the Butterfly.
BREAKPOINT
The descriptor specifies a breakpoint. The offset is never used, this type is only used in descriptors referring to breakpoints. (See Breakpoints and Watchpoints, below, for an explanation of breakpoint descriptors.)
WATCHPOINT
The descriptor specifies a watchpoint. The offset is never used, this type is only used in descriptors referring to watchpoints. (See Breakpoints and Watchpoints, below, for an explanation of watchpoint descriptors).
BPT_PTR_OFFSET
For this mode and BPT_PTR_INDIRECT, the mode argument specifies one of two breakpoint pointer variables local to the breakpoint in which this address occurs. These pointers and the SET_PTR command which manipulates them provide for an arbitrary amount of address indirection. They are intended for use in traversing data structures: for example, chasing queues. In BPT_PTR_OFFSET, the offset is added to
Page 24
LDP Specification Commands and Formats
the pointer variable to give the effective address. In targets which support multiple processes, the location is in the data space of the process given by the ID. Otherwise, the location is a physical address in macro-memory. BPT_PTR.* modes are valid only in breakpoints and watchpoints.
BPT_PTR_INDIRECT
Like BPT_PTR_OFFSET, except that it uses one more level of indirection. The pointer variable given by the mode argument plus the offset specify an address which points to the effective address. See the description of BPT_PTR_OFFSET for a discussion of usage, limitations and address space.
The short address format is intended for use in implementations where protocol overhead must be minimized. This format is a subset of the long address format: it contains the same fields except for the ID field. Therefore, the short addressing format supports only HOST and PHYS_* address modes. Only the LOADER_DUMPER implementation level commands may be used with the short addressing format. The short address format is three words long, consisting of a 16-bit word describing the address space, and a 32-bit offset.
The high-order bit is 1, indicating the short address format. A list of the address modes supported is given below. The interpretation of the remaining fields is as described above for the long addressing format.
Page 26
LDP Specification Commands and Formats
Mode | Symbol | Address space -----+--------------------+---------------------------
0 HOST Host 1 PHYS_MACRO Macro-memory 2 PHYS_MICRO Micro-memory 3 PHYS_I/O I/O space 4 PHYS_MACRO_PTR Macro contains a pointer 5 PHYS_REG Register 6 PHYS_REG_OFFSET Register plus offset 7 PHYS_REG_INDIRECT Register contains address of a pointer 8 - 32 <reserved>
Protocol commands are used for error handling, for synchronizing the command sequence number, and for communicating protocol implementation parameters. Every protocol command has a corresponding reply. All protocol commands are sent from the host to the target, with replies flowing in the opposite direction.
A HELLO_REPLY is sent by the target in response to the HELLO command at the start of an LDP session. This reply is used to inform the host about the target's implementation of LDP.
The target's LDP protocol version. If the current host protocol version does not agree with the target's protocol version, the host may terminate the session, or may continue it, at the discretion of the implementor. The current version number is 2.
System Type
The type of system running on the target. This is used as a check against what the host thinks the target is. The host is expected to have a table of target system types with information about target address spaces, target-specific commands and addressing modes, and so forth.
Currently defined system types are shown in Figure 15. This list includes some systems normally thought of as 'hosts' (e.g. C70, VAX), for implementations where targets actively initiate and direct a load of themselves.
Page 30
LDP Specification Protocol Commands
Code | System | Description --------+---------------+--------------------------- 1 C30_16_BIT BBN 16-bit C30 2 C30_20_BIT BBN 20-bit C30 3 H316 Honeywell-316 4 BUTTERFLY BBN Butterfly 5 PDP-11 DEC PDP-11 6 C10 BBN C10 7 C50 BBN C50 8 PLURIBUS BBN Pluribus 9 C70 BBN C70 10 VAX DEC VAX 11 MACINTOSH Apple MacIntosh
System Types Figure 15
Address Code
The address code indicates which LDP address format(s) the target is prepared to use. Address codes are show in Figure 16.
Address Code | Symbol | Description --------------+---------------+-----------------------------
1 LONG_ADDRESS Five word address format. Supports all address modes and commands.
2 SHORT_ADDRESS Three word address format. Supports only physical and host address modes. Only the LOADER_DUMPER set of commands are supported.
The implementation level specifies which features of the protocol are implemented in the target. There are three levels of protocol implementation. These levels are intended to correspond to the three most likely applications of LDP: simple loading and dumping, basic debugging, and full debugging. (Please see Implementations, above, for a detailed description of implementation levels.) There are are also several optional features that are not included in any particular level.
Implementation levels are cumulative, that is, each higher level includes the features of all previous levels. The levels are shown in Figure 17.
Feature Level | Symbol | Description --------------+---------------+----------------------------- 1 LOADER_DUMPER Loader/dumper subset of LDP 2 BASIC_DEBUGGER Control commands, CREATE 3 FULL_DEBUGGER FSM breakpoints
Feature Levels Figure 17
Options
The options field (see Figure 18) is an eight-bit flag field. Bit flags are used to indicate if the target has implemented particular optional commands. Not all optional commands are referenced in this field. Commands whose implementation depends on target machine features are omitted. The LDP application is expected to 'know' about target features that are not intrinsic to the protocol. Examples of target-dependent commands are commands that refer to named objects (CREATE, LIST_NAMES).
Page 32
LDP Specification Protocol Commands
Mask | Symbol | Description ------+-------------+---------------+----------------- 1 STEP The STEP command is implemented 2 WATCHPOINTS Watchpoints are implemented
The SYNCH command is sent by the host to the target. The target responds with a SYNCH_REPLY. The SYNCH - SYNCH_REPLY exchange serves two functions: it synchronizes the host-to-target implicit sequence number and acts as a cumulative acknowledgement of the receipt and execution of all host commands up to the SYNCH.
A SYNCH_REPLY is sent by the target in reponse to a valid SYNCH command. A SYNCH command is valid if its sequence number agrees with the sequence number the target is expecting. Otherwise, the target will reset its sequence number to the SYNCH command and send an ERROR reply.
The ABORT command is sent from the host to abort all pending operations at the target. The target responds with ABORT_DONE. This is primarily intended to stop large data transfers from the target. A likely application would be during a debugging session when the user types an interrupt to abort a large printout of data from the target. The ABORT command has no effect on any breakpoints or watchpoints that may be enabled in the target.
As a practical matter, the ABORT command may be difficult to implement on some targets. Its ability to interrupt command processing on the target depends on the target being able to look ahead at incoming commands and receive an out-of-band signal from the host. However, the effect of an ABORT may be achieved by simply closing and reopening the transport connection.
The ABORT_DONE reply is sent from the target to the host in response to an ABORT command. This indicates that the target has terminated all operations that were pending when the ABORT command was received. The sequence number of the ABORT command is included in the reply.
The ERROR reply is sent by the target in response to a bad command. The ERROR reply gives the sequence number of the offending command and a reason code. The target ignores further commands until an ERRACK command is received. The reason for ignoring commands is that the proper operation of outstanding commands may be predicated on the execution of the erroneous command.
An explanation of each of these error codes follows: BAD_COMMAND
The command was not meaningful to the target machine. This includes commands that are valid but unimplemented in this target. Also, the command was not valid in this context. For example, a command given by the host that is only legal in a breakpoint (e.g. IF, SET_STATE).
BAD_ADDRESS_MODE <offending-address>
The mode of an address given in the command is not meaningful to this target system. For example, a PROCESS address mode on a target that does not support multi-processing.
BAD_ADDRESS_ID <offending-address>
The ID field of an address didn't correspond to an appropriate thing. For example, for a PROCESS address mode, the ID of a non-existent process.
BAD_ADDRESS_OFFSET <offending-address>
The offset field of the address was outside the legal range for the thing addressed. For example, an offset of 200,000 in PHYS_MACRO mode on a target with 64K of
Page 38
LDP Specification Protocol Commands
macro-memory.
BAD_CREATE_TYPE
The object type in a CREATE command was unknown.
NO_RESOURCES
A CREATE command failed due to lack of necessary resources.
NO_OBJECT
A GET_OBJECT command failed to find the named object.
OUT_OF_SYNCH
The sequence number of the SYNCH command was not expected by the target. The target has resynchronized to it.
An error occurred within a breakpoint command list. The given 16-bit sequence-number refers to the sequence number of the CREATE command that created the breakpoint, while breakpoint-sequence# refers to the sequence number of the command within the breakpoint given by <breakpoint-descriptor>.
An ERRACK is sent by the host in response to an ERROR reply from the target. The ERRACK is used to acknowledge that the host has received the ERROR reply.
Data transfer commands transfer data between the host and the target. These commands are used for loading and dumping the target, and examining and depositing locations on the target. The READ command reads data from the target, the MOVE command moves data within the target or from the target to another entity, and the WRITE command writes data to the target. REPEAT_DATA makes copies of a pattern to the target -- it is useful for zeroing memory. WRITE_MASK writes data with a mask, and is intended for modifying target parameter tables.
Data transmitted to and from the target always contains a target address. In writes to the target, this is used as the destination of the data. In reads from the target, the target address is used by the host to identify where in the target the data came from. In addition, the MOVE command may contain a 'host' address as its destination; this permits the host to further discriminate between possible sources of data from the target -- from different breakpoints, debugging windows, etc.
A read request to the target may generate one or more response messages. In particular, responses to requests for large amounts of data -- core dumps, for example -- must be broken up into multiple messages, if the block of data requested plus the LDP header exceeds the transport layer message size.
In commands which contain data (WRITE, READ_DATA, MOVE_DATA and REPEAT_DATA), if there are an odd number of data octets, then a null octet is appended. This is so that the next command in the message, if any, will begin on an even octet. The command length is the sum of the number of octets in the command header and the number of octets of data, excluding the null octet, if any.
The addressing formats which may be used with data transfer commands are specified for each LDP session at the start of the session by the target in the HELLO_REPLY response. See the section entitled 'Addressing', above, for a description of LDP addressing formats and modes. In the command diagrams given below, the short addressing format is illustrated. For LDP sessions using long addressing, addresses are five words long,
The WRITE command is used to send octets of data from the host to the target. This command specifies the address in the target where the data is to be stored, followed by a stream of data octets. If the data stream contains an odd number of octets, then a null octet is appended so that the next command, if any, will begin on an even octet. Since LDP must observe message size limitations imposed by the underlying transport layer, a single logical write may need to be broken up into multiple WRITEs in separate transport messages.
The command length gives the number of octets in the command, including data octets, but excluding the padding octet, if any.
Target Start Address
This is the address to begin storing data in the target. The length of the data to be stored may be inferred by the target from the command length. An illegal address or range will generate an ERROR reply.
Data Octets
Octets of data to be stored in the target. Data are packed according to the packing convention described above. Ends with a null octet if there are an odd number of data octets.
The host uses the READ command to ask the target to send back a contiguous block of data. The data is specified by a target starting address and a count. The target returns the data in one or more READ_DATA commands, which give the starting address (in the target) of each segment of returned data. When the transfer is completed, the target sends a READ_DONE command to the host.
The starting address of the requested block of target data. The target sends an ERROR reply if the starting address is illegal, if the ending address computed from the sum of the start and the count is illegal, or if holes are encountered in the middle of the range.
Address Unit Count
The count of the number of target indivisibly-addressable units to be transferred. For example, if the address space is PHYS_MACRO, a count of two and a start address of 1000 selects the contents of locations 1000 and 1001. 'Count' is used instead of 'length' to avoid the problem of determining units the length should be denominated in (octets, words, etc.). The size and type of the unit will vary depending on the address space selected by the target start address. The target should reply with an error (if it is able to
Page 44
LDP Specification Data Transfer Commands
determine in advance of a transfer) if the inclusive range of addresses specified by the start address and the count contains an illegal or nonexistent address.
The target uses the READ_DATA response to transmit data requested by a host READ command. One or more READ_DATA responses may be needed to fulfill a given READ command, depending on the size of the data block requested and the transport layer message size limits. Each READ_DATA response gives the target starting address of its segment of data. If the response contains an odd number of data octets, the target ends the response with a null octet.
The command length gives the number of octets in the command, including data octets, but excluding the padding octet, if any. The host can calculate the length of the data by subtracting the header length from the command length. Since the target address may be either three words (short format) or five words (long format), the address mode must be checked to determine which is being used.
Target Start Address
This is the starting address of the data segment in this message. The host may infer the length of the data from the command length. The address format (short or long) is the
Page 46
LDP Specification Data Transfer Commands
same as on the initial READ command.
Data Octets
Octets of data from the target. Data are packed according to the packing convention described above. Ends with a null octet if there are an odd number of data octets.
The target sends a READ_DONE reply to the host after it has finished transferring the data requested by a READ command. READ_DONE specifies the sequence number of the READ command.
The MOVE command is sent by the host to move a block of data from the target to a specified destination. The destination address may specify a location in the target, in the host, or in another target (for loading one target from another). The data is specified by a target starting address and an address unit count. The target sends an ERROR reply if the starting address is illegal, if the ending address computed from the sum of the start and the count is illegal, or if holes are encountered in the middle of the range. If the MOVE destination is off-target, the target moves the data in one or MOVE_DATAs. Other commands arriving at the target during the transfer should be processed in a timely fashion, particularly the ABORT command. When the data has been moved, the target sends a MOVE_DONE to the host. However, a MOVE within a breakpoint will not generate a MOVE_DONE.
A MOVE with a host destination differs from a READ in that it contains a host address. This field is specified by the host in the MOVE command and copied by the target into the responding MOVE_DATA(s). The address may be used by the host to differentiate data returned from multiple MOVE requests. This information may be useful in breakpoints, in multi-window debugging and in communication with targets with multiple processors. For example, the host sends the MOVE command to the target to be executed during a breakpoint. The ID field in the host address might be an index into a host breakpoint table. When the breakpoint executes, the host would use the ID to associate the returning MOVE_DATA with this breakpoint.
The starting address of the requested block of target data. An illegal address type will generate an error reply.
Address Unit Count
The count of the number of target indivisibly-addressable units to be transferred. For example, if the address space is PHYS_MACRO, a count of two and a start address of 1000 selects the contents of locations 1000 and 1001. 'Count' is used instead of 'length' to avoid the problem of determining units the length should be denominated in (octets, words,
etc.). The size and type of the unit will vary depending on the address space selected by the target start address. The target should reply with an error (if it is able to determine in advance of a transfer) if the inclusive range of addresses specified by the start address and the count contains an illegal or nonexistent address.
Destination Address
The destination of the MOVE. If the address space is on the target, the address unit size should agree with that of the source address space. If the address mode is HOST, the values and interpretations of the remaining address fields are arbitrary, and are determined by the host implementation. For example, the mode argument might specify a table (breakpoint, debugging window, etc.) and the ID field an index into the table.
The target uses the MOVE_DATA responses to transmit data requested by a host MOVE command. One or more MOVE_DATA responses may be needed to fulfill a given MOVE command, depending on the size of the data block requested and the transport layer message size limits. Each MOVE_DATA response gives the target starting address of its segment of data. If the response contains an odd number of data octets, the target should end the response with a null octet.
message. The host may infer length of the data from the command length.
Destination Address
The destination address copied from the MOVE command that initiated this transfer. In the case of HOST MOVEs, this is used by the host to identify the source of the data.
Data Octets
Octets of data from the target. Data are packed according to the packing convention described above. Ends with a null octet if there are an odd number of data octets.
The target sends a MOVE_DONE reply to the host after it has finished transferring the data requested by a MOVE command. MOVE_DONE specifies the sequence number of the MOVE command.
The REPEAT_DATA command is sent by the host to write copies of a specified pattern into the target. This provides an efficient way of zeroing target memory and initializing target data structures. The command specifies the target starting address, the number of copies of the pattern to be made, and a stream of octets that constitutes the pattern.
This command differs from the other data transfer commands in that the effect of a REPEAT_DATA with a large pattern cannot be duplicated by sending the data in smaller chunks over several commands. Therefore, the maximum size of a pattern that can be copied with REPEAT_DATA will depend on the message size limits of the transport layer.
The command length gives the number of octets in the command, including data octets in the pattern, but excluding the padding octet, if any.
Target Start Address
This is the starting address where the first copy of the pattern should be written in the target. Successive copies of the pattern are made contiguously starting at this address.
Repeat Count
The repeat count specifies the number of copies of the pattern that should be made in the target. The repeat count should be greater than zero.
Pattern
The pattern to be copied into the target, packed into a stream of octets. Data are packed according to the packing convention described above. Ends with a null octet if there are an odd number of data octets.
The host sends a WRITE_MASK command to the target to write one or more masked values. The command uses an address to specify a target base location, followed by one or more offset- mask-value triplets. Each triplet gives an offset from the base, a value, and a mask indicating which bits in the location at the offset are to be changed.
This optional command is intended for use in controlling the target by changing locations in a table. For example, it may be used to change entries in a target parameter table. The operation of modifying a specified location with a masked value is intended to be atomic. In other words, another target process should not be able to access the location to be modified between
The command length gives the number of octets in the command. The number of offset-value pairs may be calculated from this, since the command header is either 10 or 12 octets long (short or long address format), and each offset-mask-value triplet is 12 octets long.
Target Base Address
Specifies the target location to which the offset is added to yield the location to be modified.
Offset
An offset to be added to the base to select a location to be modified. Mask
Specifies which bits in the value are to be copied into the location. Value
A value to be stored at the specified offset from the base. The set bits in the mask determine which bits in the value are applied to the location. The following algorithm will achieve the intended result: take the one's complement of the mask and AND it with the location, leaving the result in the location. Then AND the mask and the value, and OR the result into the location.
Control commands are used to control the execution of target code, breakpoints and watchpoints. They are also used to read and report the state of these objects. The object to be controlled or reported on is specified with a descriptor. Valid descriptor modes include PHYS_* (for some commands) PROCESS_CODE, BREAKPOINT and WATCHPOINT. Control commands which change the state of the target are START, STOP, CONTINUE and STEP. REPORT requests a STATUS report on a target object. EXCEPTION is a spontaneous report on an object, used to report asynchronous events such as hardware traps. The host may verify the action of a START, STOP, STEP or CONTINUE command by following it with a REPORT command.
The START command is sent by the host to start execution of a specified object in the target. For targets which support multiple processes, a PROCESS_CODE address specifies the process to be started. Otherwise, one of the PHYS_* modes may specify a location in macro-memory where execution is to continue. Applied to a breakpoint or watchpoint, START sets the value of the object's state variable, and activates the breakpoint. The breakpoint counter and pointer variables are initialized to zero.
The descriptor specifies the object to be started. If the mode is PROCESS_CODE, ID specifies the process to be started, and offset gives the process virtual address to start at. If the mode is PHYS_*, execution of the target is continued at the specified address.
For modes of BREAKPOINT and WATCHPOINT, the offset specifies the new value of the FSM state variable. This is for FSM breakpoints and watchpoints.
The STOP command is sent by the host to stop execution of a specified object in the target. A descriptor specifies the object. Applied to a breakpoint or watchpoint, STOP deactivates it. The breakpoint/watchpoint may be re-activated by issuing a START or a CONTINUE command for it.
The CONTINUE command is sent by the host to resume execution of a specified object in the target. A descriptor specifies the object. Applied to a breakpoint or watchpoint, CONTINUE activates it.
The STEP command is sent by the host to the target. It requests the execution of one instruction (or appropriate operation) in the object specified by the descriptor.
The descriptor specifies the object for which a STATUS report is requested. For a mode of PROCESS_CODE, the ID specifies a process. Other valid modes are PHYS_MACRO, to query the status of the target application, and BREAKPOINT and WATCHPOINT, to get the status of a breakpoint or watchpoint.
The target sends a STATUS reply in response to a REPORT command from the host. STATUS gives the state of a specified object. For example, it may tell whether a particular target process is running or stopped.
Page 64
LDP Specification Control Commands
0 0 0 1 1 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 +---------------+---------------+ 0 | Command Length | +---------------+---------------+ 1 | CONTROL | STATUS | +---------------+---------------+ +-+ 2 | Mode | 0 | | +---------------+---------------+ | 3 | | | Descriptor +-- ID --+ | 4 | Field | | +-------------------------------+ +-+ 5 | Status | +-------------------------------+ +-+ * | * | * | Other Data +-------------------------------+ | n | Other Data | | +-------------------------------+ +-+
STATUS Reply Format Figure 40
STATUS FIELDS:
Descriptor
The descriptor specifies the object whose status is being given. If the mode is PROCESS_CODE, then the ID specifies a process. If the mode is PHYS_MACRO, then the status is that of the target application.
Status
The status code describes the status of the object. Status codes are 0=STOPPED and 1=RUNNING. For breakpoints and watchpoints, STOPPED means disarmed and RUNNING means armed.
Other Data
For breakpoints and watchpoints, Other Data consists of a
An EXCEPTION is a spontaneous message sent from the target indicating a target-machine exception associated with a particular object. The object is specified by an address.
Management commands are used to control resources in the target machine. There are two kinds of commands: those that interrogate the remote machine about resources, and those that allocate and free resources. There are management commands to create, list and delete breakpoints. All commands have corresponding replies which include the sequence number of the request command. Failing requests produce ERROR replies.
There are two resource allocation commands, CREATE and DELETE, which create and delete objects in the remote machine. There are a number of listing commands for listing a variety of target objects -- breakpoints, watchpoints, processes, and names. The amount of data returned by listing commands may vary in length, depending on the state of the target. If a list is too large to fit in a single message, the target will send it in several list replies. A flag in each reply specifies whether more messages are to follow.
The CREATE command is sent from the host to the target to create a target object. If the CREATE is successful, the target returns a CREATE_DONE reply, which contains a descriptor associated with the CREATEd object. The types of objects that may be specified in a CREATE include breakpoints, processes, memory objects and descriptors. All are optional except for breakpoints.
The type of object to be created. Arguments vary with the type. Currently defined types are shown in Figure 43. All are optional except for BREAKPOINT.
Create Type | Symbol -------------+----------------
0 BREAKPOINT 1 WATCHPOINT 2 PROCESS 3 MEMORY_OBJECT 4 DESCRIPTOR
Create Types Figure 43
Page 70
LDP Specification Management Commands
Create Arguments
Create arguments depend on the type of object being created. The formats for each type of object are described below.
The format is the same for CREATE BREAKPOINT and CREATE WATCHPOINT. In the following discussion, 'breakpoint' may be taken to mean either breakpoint or watchpoint.
The address is the location where the breakpoint is to be set. In the case of watchpoints it is the location to be
watched. Valid modes are any PHYS_* mode that addresses macro-memory, PROCESS_CODE for breakpoints and PROCESS_DATA for watchpoints.
'Maximum states' is the number of states the finite state machine for this breakpoint will have. A value of zero indicates a default breakpoint, for targets which do not implement finite state machine (FSM) breakpoints. A default breakpoint is the same as an FSM with one state consisting of a STOP and a REPORT command for the process containing the breakpoint.
'Maximum size' is the total size, in octets, of the breakpoint data to be sent via subsequent BREAKPOINT_DATA commands. This is the size of the data only, and does not include the LDP command headers and breakpoint descriptors.
'Maximum local variables' is the number of 32-bit longs to reserve for local variables for this breakpoint. Normally this value will be zero.
PROCESS
Creates a new process. Arguments are target-dependent.
Creates an object of size Object Size, with the given name. Object Size is in target dependent units. The name may be the null string for unnamed objects. Name Size gives the number of characters in Object Name, and must be even. Always ends with a null octect.
DESCRIPTOR
Used for obtaining descriptors from IDs on target systems where IDs are longer than 32 bits. There is a single argument, Long ID, whose length is target dependent.
The target sends a CREATE_DONE reply to the host in response to a successful CREATE command. The reply contains the sequence number of the CREATE request, and a descriptor for the object created. This descriptor is used by the host to specify the object in subsequent commands referring to it. Commands which refer to created objects include LIST_* commands, DELETE and BREAKPOINT_DATA. For example, to delete a CREATEd object, the host sends a DELETE command that specifies the descriptor returned by the CREATE_DONE reply.
The sequence number of the CREATE command to which this is the reply.
Created Object Descriptor
A descriptor assigned by the target to the created object. The contents of the descriptor fields are arbitrarily
Page 74
LDP Specification Management Commands
assigned by the target at its convenience. The host treats the descriptor as a unitary object, used for referring to the created object in subsequent commands.
The host sends a DELETE command to remove an object created by an earlier CREATE command. The object to be deleted is specified with a descriptor. The descriptor is from the CREATE_DONE reply to the original CREATE command.
The target sends a DELETE_DONE reply to the host in response to a successful DELETE command. The reply contains the sequence number of the DELETE request.
The host sends a LIST_ADDRESSES command to request a list of valid address ranges for a specified object. The object is given by a descriptor. Typical objects are a target process, or the target physical machine. The target responds with an ADDRESS_LIST reply. This command is used for obtaining the size of dynamic address spaces and for determining dump ranges.
The target sends an ADDRESS_LIST reply to the host in response to a successful LIST_ADDRESSES command. The reply contains the sequence number of the LIST_ADDRESSES request, the descriptor of the object being listed, and a list of the valid address ranges within the object.
The sequence number of the LIST_ADDRESSES command to which this is the reply.
Flags
If M=1, the address list is continued in one or more subsequent ADDRESS_LIST replies. If M=0, this is the final ADDRESS_LIST.
Item Count
The number of address ranges described in this command.
Descriptor
The descriptor of the object being listed.
Address Range
Each address range is composed of a pair of 32-bit addresses which give the first and last addresses of the range. If there are 'holes' in the address space of the object, then multiple address ranges will be used to describe the valid address space.
The host sends a LIST_BREAKPOINTS command to request a list of all breakpoints associated with the current connection. The target replies with BREAKPOINT_LIST.
The target sends a BREAKPOINT_LIST reply to the host in response to a LIST_BREAKPOINTS command. The reply contains the sequence number of the LIST_BREAKPOINTS request, and a list of all breakpoints associated with the current connection. The descriptor and address of each breakpoint are listed.
The target sends a PROCESS_LIST reply to the host in response to a LIST_PROCESSES command. The reply contains the sequence number of the LIST_PROCESSES request, and a list of all processes in the target. For each process, a descriptor and a target-dependent amount of process data are given.
0 0 0 1 1 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 +---------------+---------------+ 0 | Command Length | +---------------+---------------+ 1 | MANAGEMENT | PROCESS_LIST | +---------------+---------------+ 2 | List Sequence Number | +---------------+---------------+ 3 | Flags |M| Item Count | +---------------+---------------+ +-+ 4 | PROCESS_CODE | 0 | | +---------------+---------------+ | 5 | | | Process +-- ID --+ | Descriptor 6 | Field | | +---------------+---------------+ +-+ 7 | Process data count | | +---------------+---------------+ | 8 | Process data | Process data | | +-------------------------------+ | Process * | Data * | * | +---------------+---------------+ | n | Process data | Process data | | +-------------------------------+ +-+ * | Additional * | Descriptor-Data * | Pairs +-+
The target sends a NAME_LIST reply to the host in response to a LIST_NAMES command. The reply contains the sequence number of the LIST_NAMES request, and a list of all target names, as strings.
The sequence number of the LIST_NAMES command to which this is the reply.
Page 86
LDP Specification Management Commands
Flags
If M=1, the name list is continued in one or more subsequent NAME_LIST replies. If M=0, this is the final NAME_LIST.
Item Count
The number of name strings in this list. Each name string consists of a character count and a null-terminated string of characters.
Name Size
The number of octets in this name string. Must be even.
Name Characters
A string of octets composing the name. Ends with a null octet. The number of characters must be even, so if the terminating null comes on an odd octet, another null is appended.
The host sends a GET_PHYS_ADDR command to convert an address into physical form. The target returns the physical address in a GOT_PHYS_ADDR reply. For example, the host could send a GET_PHYS_ADDR command containing a register-offset address, and the target would return the physical address derived from this in a GOT_PHYS_ADDR reply.
The address to be converted to a physical address. The mode may be one of PHYS_REG_OFFSET, PHYS_REG_INDIRECT, PHYS_MACRO_PTR, any OBJECT_* mode, and any PROCESS_* mode except for PROCESS_REG.
The target sends a GOT_PHYS_ADDR reply to the host in response to a successful GET_PHYS_ADDR command. The reply contains the sequence number of the GET_PHYS_ADDR request, and the specified address converted into a physical address.
The sequence number of the GET_PHYS_ADDR command to which this is the reply.
Address
The address resulting from translating the address given in the GET_PHYS_ADDR command into a physical address. Mode is always PHYS_MACRO and ID and mode argument are always zero. Offset gives the 32-bit physical address.
The host sends a GET_OBJECT command to convert a name string into a descriptor. The target returns the descriptor in a GOT_OBJECT reply. Intended for use in finding control parameter objects.
The target sends a GOT_OBJECT reply to the host in response to a successful GET_OBJECT command. The reply contains the sequence number of the GET_OBJECT request, and the specified object name converted into a descriptor.
The descriptor of the object named in the GET_OBJECT command.
Page 92
LDP Specification Breakpoints and Watchpoints
CHAPTER 9
Breakpoints and Watchpoints
Breakpoints and watchpoints are used in debugging applications. Each breakpoint or watchpoint is associated with one debugger connection and one address. When a breakpoint or watchpoint is triggered, the target executes one or more commands associated with it. A breakpoint is triggered when its address is executed. A watchpoint is triggered when its address is modified. The same mechanism is used for structuring breakpoint and watchpoint commands. For brevity's sake, 'breakpoint' will be used in the remainder of this document to refer to either a breakpoint or a watchpoint.
The commands used by the host to manipulate breakpoints are given in Figure 61, in the order in which they are normally used. All commands are sent from the host to the target, and each specifies the descriptor of a breakpoint.
CREATE Create a breakpoint BREAKPOINT_DATA Send commands to be executed in an FSM breakpoint START Activate a breakpoint, set state and initialize breakpoint variables STOP Deactivate a breakpoint CONTINUE Activate a breakpoint LIST_BREAKPOINTS List all breakpoints REPORT Report the status of a breakpoint DELETE Delete a breakpoint
There are two kinds of breakpoints: default breakpoints and finite state machine (FSM) breakpoints. They differ in their use of commands.
Default breakpoints do not contain any commands. When triggered, a default breakpoint stops the target object (i.e., target process or application) it is located in. A STATUS report on the stopped object is sent to the host. At this point, the host may send further commands to debug the target.
An FSM breakpoint has one or more conditional command lists, organized into a finite state machine. When an FSM breakpoint is created, the total number of states is specified. The host then sends commands (using BREAKPOINT_DATA) to be associated with each state. The target maintains a state variable for the breakpoint, which determines which command list will be executed if the breakpoint is triggered. When the breakpoint is created its state variable is initialized to zero (zero is the first state). A breakpoint command, SET_STATE, may be used within a breakpoint to change the value of the state variable. A REPORT command applied to a breakpoint descriptor returns its address, whether it is armed or disarmed, and the value of its state variable.
Commands valid in breakpoints include all implemented data transfer and control commands, a set of conditional commands, and a set of breakpoint commands. The conditional commands and the breakpoint commands act on a set of local breakpoint variables. The breakpoint variables consist of the state variable, a counter, and two pointer variables. The conditional commands control the execution of breakpoint command lists based on the contents of one of the breakpoint variables. The breakpoint commands are used to set the value of the breakpoint variables: SET_STATE sets the state variable, SET_PTR sets one of the pointer variables, and INC_COUNT increments the breakpoint counter. There may be implementation restrictions on the number of breakpoints, the number of states, the number of conditions, and the size of the command lists. Management commands and protocol commands are forbidden in breakpoints.
In FSM breakpoints, the execution of commands is controlled as follows. When a breakpoint is triggered, the breakpoint's state variable selects a particular state. One or more conditional command lists is associated with this state. A conditional command list consists of a list of conditions followed by a list of commands which are executed if the condition list is satisfied. The debugger starts a breakpoint by executing the first of these lists. If the condition list is
Page 94
LDP Specification Breakpoints and Watchpoints
satisfied, the debugger executes the associated command list and leaves the breakpoint. If the condition list fails, the debugger skips to the next conditional command list. This process continues until the debugger either encounters a successful condition list, or exhausts all the conditional command lists for the state. The relationship of commands, lists and states is shown in Figure 62 (IFs, THENs and ELSEs are used below to clarify the logical structure within a state; they are not part of the protocol).
State 0 IF <condition list 0> THEN <command list 0>
BREAKPOINT_DATA is a data transfer command used by the host to send commands to be executed in breakpoints and watchpoints. The command specifies the descriptor of the breakpoint or watchpoint, and a stream of commands to be appended to the end of the breakpoint's command list. BREAKPOINT_DATA is applied sequentially to successive breakpoint states, and successive
command lists within each state. Multiple BREAKPOINT_DATAs may be sent for a given breakpoint. Breaks between BREAKPOINT_DATA commands may occur anywhere within the data stream, even within individual commands in the data. Sufficient space to store the data must have been allocated by the maximum size field in the CREATE BREAKPOINT/WATCHPOINT command.
0 0 0 1 1 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 +---------------+---------------+ 0 | Command Length | +---------------+---------------+ 1 | DATA_TRANSFER |BREAKPOINT_DATA| +---------------+---------------+ +-+ 2 | Mode | Mode Argument | | +---------------+---------------+ | Breakpoint or 3 | | | Watchpoint +-- ID --+ | Descriptor 4 | Field | | +-------------------------------+ +-+ 5 | Data | Data | | +-------------------------------+ | * | * | Data * | +---------------+---------------+ | n | Data | Data or 0 | | +---------------+---------------+ +-+
BREAKPOINT_DATA Command Format Figure 63
BREAKPOINT_DATA FIELDS:
Command Length
Total length of this command in octets, including data, excluding the final padding octet, if any.
Data
A stream of data to be appended to the data for this breakpoint or watchpoint. This stream has the form of one or more states, each containing one or more conditional
Page 96
LDP Specification Breakpoints and Watchpoints
command lists. The first BREAKPOINT_DATA command sent for a breakpoint contains data starting with state zero. The data for each state starts with the state size. A conditional command list is composed of two parts: a condition list, and a command list. Each list begins with a word that gives its size in octets.
<state 0 size> <condition list 0 size> <condition list 0> <command list 0 size> <command list 0> * * * <condition list n size> <condition list n> <command list n size> <command list n> <state 1 size> <etc> * * * <state n size>
All sizes are stored in 16-bit words, and include their own length. The state size gives the total number of octets of breakpoint data for the state. The condition list size gives the total octets of breakpoint data for the following condition list. A condition list size of 2 indicates an empty condition list: in this case the following command list is executed unconditionally. The command list size gives the total octets of breakpoint data for the following command list.
Lists
Condition and command lists come in pairs. When the breakpoint occurs, the condition list controls whether the following command list should be executed. A condition list consists of one or more commands from the CONDITION command class. A command list consists one or more LDP commands. Valid commands are any commands from the BREAKPOINT, DATA_TRANSFER or CONTROL command classes.
Page 98
LDP Specification Conditional Commands
CHAPTER 10
Conditional Commands
Conditional commands are used in breakpoints to control the execution of breakpoint commands. One or more conditions in sequence form a condition list. If a condition list is satisfied (evaluates to TRUE), the breakpoint command list immediately following it is executed. (See Breakpoints and Watchpoints, above, for a discussion of the logic flow in conditional/command lists.) Conditional commands perform tests on local breakpoint variables, and other locations. Each condition evaluates to either TRUE or FALSE. Figure 65 contains a summary of conditional commands:
CHANGED <loc> Determine if a location has changed COMPARE <loc1> <mask> <loc2> Compare two locations, using a mask COUNT_[EQ | GT | LT] <value> Compare the counter to a value TEST <loc> <mask> <value> Compare a location to a value
Conditional Command Summary Figure 65
The rules for forming and evaluating condition lists are:
o consecutive conditions have an implicit logical AND between them. A sequence of such conditions is called an 'and_list'. and_lists are delimited by an OR command and by the end of the condition list.
o the breakpoint OR command may be inserted between any pair of conditions
o AND takes precedence over OR
o nested condition lists are not supported. A condition list is simply one or more and_lists, separated by ORs.
o the condition list is evaluated in sequence until either a TRUE and_list is found (condition list <- TRUE), or the end of the condition list is reached (condition list <- FALSE). An and_list is TRUE if all its conditions are TRUE.
Condition commands start with the standard four-octet command header. The high-order bit of the command type byte is used as a negate flag: if this bit is set, the boolean value of the condition is negated. This flag applies to one condition only, and not to other conditions in the condition list.
The COUNT conditions (COUNT_EQ, COUNT_GT and COUNT_LT) are used to compare the breakpoint counter to a specified value. The counter is set to zero when the breakpoint is STARTed, and is incremented by the INC_COUNT breakpoint command. The format is the same for the COUNT_EQ, COUNT_GT and COUNT_LT conditions.
The CHANGED condition is TRUE if the contents of the specified location have changed since the last time this breakpoint occurred. Only one location may be specified as the object of CHANGED conditions per breakpoint. The CHANGED condition is always FALSE the first time the breakpoint occurs.
Breakpoint commands are used to set the value of breakpoint variables. These commands are only valid within breakpoints and watchpoints. They are sent from the host to the target as data in BREAKPOINT_DATA commands. Figure 71 contains a summary of breakpoint commands:
INCREMENT <location> Increment the specified location INC_COUNT Increment the breakpoint counter OR OR two breakpoint condition lists SET_PTR <n> <location> Set pointer <n> to the contents of <location> SET_STATE <n> Set the breakpoint state variable to <n>
The INC_COUNT command increments the breakpoint counter. There is one counter variable for each breakpoint. It is initialized to zero when the breakpoint is created, when it is armed with the START command, and whenever the breakpoint state changes. The counter is tested by the COUNT_* conditions.
The OR command delineates two and_lists in a breakpoint condition list. A condition list is TRUE if any of the OR separated and_lists in it are TRUE. A breakpoint condition list may contain zero, one or, many OR commands. See 'Condition Commands' for an explanation of condition lists.
The SET_PTR command loads the specified breakpoint pointer with the contents of a location. The pointer variables and the SET_PTR command are intended to provide a primitive but unlimited indirect addressing capability. Two addressing modes, BPT_PTR_OFFSET and BPT_PTR_INDIRECT, are used for referencing the breakpoint pointers. For example, to follow a linked list, use SET_PTR to load a pointer with the start of the list, then use successive SET_PTR commands with addressing mode BPT_PTR_OFFSET to get successive elements.
The SET_STATE command sets the breakpoint state variable to the specified value. This is the only method of changing a breakpoint's state from within a breakpoint. The breakpoint's state may be also be changed by a START command from the host. The state variable is initialized to zero when the breakpoint is created.
The new value for the breakpoint state variable. Must not be greater than the maximum state value specified in the CREATE BREAKPOINT command that created this breakpoint.
Command and message diagrams are used in this document to illustrate the format of these entities. Words are listed in order of transmission down the page. The first word is word zero. Bits within a word run left to right, most significant to least. However, following a convention observed in other protocol documents, bits are numbered in order of transmission; the most significant bit in a word is transmitted first. The bit labelled '0' is the most significant bit.
0 0 0 1 1 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 +---------------+---------------+ 0 |M| |L| +---------------+---------------+ 1 | Most Sig Octet| Least S. Octet| +---------------+---------------+
M = most significant bit in word zero, transmitted first L = least significant bit in word zero, transmitted last
Sender Command | Host Target | Function -------------------+-------------+---------------------------
ABORT X Abort outstanding commands ABORT_DONE X Acknowledge ABORT ADDRESS_LIST X Return valid address ranges BREAKPOINT_DATA X Send breakpoint commands BREAKPOINT_LIST X Return list of breakpoints CONTINUE X Resume execution CREATE X Create target object CREATE_DONE X Acknowledge CREATE DELETE X Delete target object DELETE_DONE X Acknowledge DELETE EXCEPTION X Report target exception ERROR X Report error with a host command ERRACK X Acknowledge ERROR GET_OBJECT X Get object descriptor from name GET_PHYS_ADDRESS X Get address in physical form GOT_OBJECT X Return object descriptor GOT_PHYS_ADDRESS X Return physical address HELLO X Initiate LDP session HELLO_REPLY X Return LDP parameters LIST_ADDRESSES X Request valid address ranges LIST_BREAKPOINTS X Request breakpoint list LIST_NAMES X Request name list LIST_PROCESSES X Request process list MOVE X Read data from target MOVE_DONE X Acknowledge MOVE completion MOVE_DATA X Send data request by MOVE NAME_LIST X Return name list PROCESS_LIST X Return process list READ X Read data from target READ_DATA X Return data requested by READ READ_DONE X Acknowledge READ completion REPEAT_DATA X Write copies of data REPORT X Request status of object START X Start target object STATUS X Return status of object STEP X Step execution of target object STOP X Stop target object SYNCH X Check sequence number SYNCH_REPLY X Confirm sequence number WRITE X Write data WRITE_MASK X Write data with mask
The following table shows the relationship between commands, responses and replies. Commands are sent from the host to the target. Some commands elicit responses and/or replies from the target. Responses and replies are sent from the target to the host. The distinction between them is that the target sends only one reply to a command, but may send multiple responses. Responses always contain data, whereas replies may or may not.
Finite state machine. Commands of each breakpoint or watchpoint are implemented as part of a finite state machine. A list of breakpoint commands is associated with each state. There are several breakpoint commands to change from one state to another. host
The 'host' in an LDP session is the timesharing system on which the user process runs.
long
A long is a 32-bit quantity.
octet
An octet is an eight-bit quantity.
RDP
The Reliable Data Protocol (RDP) is a transport layer protocol designed as a low-overhead alternative to TCP. RDP is a connection oriented protocol that provides reliable, sequenced message delivery.
server process
The LDP server process is the passive participant in an LDP session. The server process usually resides on a target machine such as a PAD, PSN or gateway. The server process waits for a user process to initiate a session, and responds to commands from the user process. In response to user commands, the server may perform services on the target like reading and writing memory locations or setting breakpoints. 'Server' is sometimes employed as a shorthand for 'server process'.
The 'target' in an LDP session is the PSN, PAD or gateway that is being loaded, dumped or debugged by the host. Normally, LDP will be implemented in the target as a server process. However, in some targets with strange requirements, notably the Butterfly, the target LDP may be a user process.
user process
The LDP user process is the active participant in an LDP session. The user process initiates and terminates the session and sends commands to the server process which control the session. The user process usually resides on a timesharing host and is driven by a higher-level entity (e.g., an application program like an interactive debugger). 'User' is sometimes employed as a shorthand for 'user process'.