Document revision date: 30 March 2001

Much process context resides in P1 space, taking the form of data cells and the process stacks. Some of these data cells need to be per-kernel thread, as do the stacks. By calling the appropriate system service, a kernel thread region in P1 space is initialized to contain the per-kernel thread data cells and stacks. The region begins at the boundary between P0 and P1 space at address 40000000x, and it grows toward higher addresses and the initial thread's user stack. The region is divided into per-kernel thread areas. Each area contains pages for data cells and the four stacks.

6.2.4 Per-Kernel Thread Stacks

A process is created with four stacks; each access mode has one stack. All four of these stacks are located in P1 space. Stack sizes are either fixed, determined by a SYSGEN parameter, or expandable. The parameter KSTACKPAGES controls the size of the kernel stack. This parameter continues to control all kernel stack sizes, including those created for new execution contexts. The executive stack is a fixed size of two pages; with kernel threads implementation, the executive stack for new execution contexts continues to be two pages in size. The supervisor stack is a fixed size of four pages; with kernel threads implementation, the supervisor stack for new execution contexts is reduced to two pages in size.

For the user stack, a more complex situation exists. OpenVMS allocates P1 space from high to lower addresses. The user stack is placed after the lowest P1 space address allocated. This allows the user stack to expand on demand toward P0 space. With the introduction of multiple sets of stacks, the locations of these stacks impose a limit on the size of each area in which they can reside. With the implementation of kernel threads, the user stack is no longer boundless. The initial user stack remains semiboundless; it still grows toward P0 space, but the limit is the per-kernel thread region instead of P0 space.

6.2.5 Per-Kernel Thread Data Cells

Several pages in P1 space contain process state in the form of data cells. A number of these cells must have a per-kernel thread equivalent. These data cells do not all reside on pages with the same protection. Because of this, the per-kernel area reserves approximately two pages for these cells. Each page has a different page protection; one page protection is user read, user write (URUW), the other is user read, executive write (UREW). The top of the user stack is used for the URUW data cells.

6.2.6 Layout of the Per-Kernel Thread

Each per-kernel thread area contains a set of stacks and two pages for data. Each area is a fixed size. For a system using the default values for the kernel stack and user stack size, each area has the layout shown in Figure 6-1.

Figure 6-1 Default Kernel Stack and User Stack Sizes

Process creation results in a PCB/KTB, a PHD/FRED, and a set of stacks. All processes have a single kernel thread, the initial thread. A multithreaded process always begins as a single threaded process. A multithreaded process contains a PCB/KTB pair and a PHD/FRED pair for the initial thread; for its other threads, it contains additional KTBs, additional FREDs, and additional sets of stacks. When the multithreaded application exists, the process returns to its single threaded state, and all additional KTBs, FREDs, and stacks are deleted.

Figure 6-2 shows the relationships and locations of the data structures for a process.

Figure 6-2 Structure of a Multithreaded Process

OpenVMS qualifies much context by the process ID (PID). With the implementation of kernel threads, much of that process context moves to the thread level. With kernel threads, the basic unit of scheduling is no longer the process but rather the kernel thread. Because of this, kernel threads need a method to identify them which is similar to the PID. To satisfy this need, the meaning of the PID is extended. The PID continues to identify a process, but can also identify a kernel thread within that process. An overview follows that presents the features of the PID, and the extended process ID (EPID), which is the cluster-visible extension of the PID.

The PID in this form is typically known as the internal PID (IPID). It consists of two pieces of information, both one word in length. Figure 6-3 shows the layout.

Figure 6-3 Process ID (PID)

The low word is the process index (PIX). The PIX is used as an index into the PCB vector. This is a vector of PCB addresses. Therefore the PIX gives a quick method of determining the PCB address given a PID.

Another array, also indexed by PIX, contains a sequence number entry for each PIX. The sequence number increments every time a PIX is reused. The high word of the IPID is a copy of the value in the array for a particular PIX. This feature validates a PID to ensure that the ID is not for a process which has been deleted. The sequence number in the IPID must match the one in the sequence number array for that PIX.

The EPID is the cluster-visible PID. It consists of five parts, as Figure 6-4 shows.

Figure 6-4 Extended Process ID (EPID)

The EPID takes its low 21 bits from the two word IPID fields seen in Figure 6-3. The value of MAXPROCESSCNT determines the number of bits within the 21 bits used for the PIX (5 to 13 bits). The sequence number uses the remaining bits (8 to 16 bits). The PIX cannot be larger than 8192; the sequence number no larger than 32767. If the system is an OpenVMS cluster member, the next 10 bits of the EPID uniquely identify the PID within the cluster. They contain 8 bits of the cluster system ID (CSID) for the system, and a 2 bit sequence number. The system service SYS$GETJPI uses the high bit (31). If set, the bit specifies that the PID is a wildcard context value. This allows collecting information for all processes in the system.

6.3.1 Multithread Effects on the PID

With kernel threads implementation, the PID's definition undergoes two changes:

• MAXPROCESSCNT's maximum value is increased to 16,384.
  To do this, the maximum PIX field width for the EPID is increased by one bit. This results in shrinking the sequence number field by one bit.
• A redefinition of the sequence number.
  The redefinition of the usage of the sequence number results in it taking on a dual meaning. It continues to be used to validate a PID; it also becomes the means for determining the kernel thread ID. Instead of a single sequence number being assigned to a PIX, a range of sequence numbers are used, one for each kernel thread. Therefore, the format of a kernel thread PID is identical to that of the PID in either its IPID, or EPID representation. The PIX and sequence number fields are in the same location, and they are the same size. In the EPID, the 10 bits used to uniquely identify the PID within the cluster remain the same; this enables kernel threads to be visible clusterwide.

Every process has at least one kernel thread, the initial thread, which is always thread ID zero; therefore, given a particular PID, the PIX continues to be used as an index into the PCB and sequence vectors. A range check validates the sequence numbers.

Before kernel threads implementation, the sequence number vector (SCH$GL_SEQVEC) was a vector of words. After kernel threads implementation, it is a vector of longwords that enables range checking for sequence number validation. The low word in each longword is the base sequence number for a particular PIX, and the upper word is the next sequence number for that PIX. The sequence number for a single-threaded process must be equal to the base value. Kernel threads PID sequence numbers must fall within the base and next values.

Figure 6-5 shows the flow of range checking of sequence numbers.

Figure 6-5 Range Checking and Sequence Vectors

Similar to the fields in the PCB that migrate to the KTB, there are several status bits that need to be per thread. The interface for the SYS$GETJPI and SYS$PROCESS_SCAN system services indicates that the entire longword fields that contain the status bits can be returned. Therefore, all the status bits must remain defined as they are. The PCB specific bits are "reserved" in the KTB structure definition. Likewise, the KTB specific bits are "reserved" in the PCB. Because the PCB is really an overlaid structure with the initial thread's KTB, just the PCB status bits need to be returned for the initial thread. The status longword returned for other threads is built by first masking out the initial thread's bits, and then OR'ing the remainder with the status longword in the appropriate KTB.

If a thread in a multithreaded process requests information about itself using SYS$GETJPI (passes PID=0), then the status bits for the kernel thread it is running on are returned. Since each kernel thread has its own PID, SYS$GETJPI can be called for each of the kernel threads in a process. The return status bits are the combination of the PCB status bits and those in the KTB associated with the input PID.

This appendix contains descriptions of the OpenVMS Alpha Version 7.0 I/O data structure changes made to support 64-bit addressing.

The data structures are listed in alphabetical order. However, the individual structure members are listed in the order in which they are defined within each data structure. Note, however, that the following sections only describe new or changed structure members. Existing unchanged members are not described. In addition, unused or "fill" structure members that might be added to obtain natural alignment are not listed. Thus, you can not use the following descriptions to calculate the precise memory layout of the structures. However, you can assume that any new or changed structure members will be naturally aligned within the structure.

A.1 Pointer Size Conventions

Any unqualified use of the term "pointer" implies a 32-bit pointer. All 64-bit pointers will be explicitly identified as either a 64-bit or quadword pointer.

As of OpenVMS Alpha Version 7.0, a new C compiler pragma controls the pointer size. To facilitate the use of 64-bit pointers, a new header file, far_pointers.h in SYS$STARLET_C.TLB, defines types for 64-bit pointers to the intrinsic C data types.

Table A-1 summarizes the 64-bit pointer data types.

Table A-1 64-Bit Pointer Data Types

Type Name	32-Bit Analog	Description	Defined by
CHAR_PQ	char *	64-bit pointer to a char	far_pointers.h
CHAR_PPQ	char **	64-bit pointer to a CHAR_PQ	far_pointers.h
INT_PQ	int *	64-bit pointer to a 32-bit int	far_pointers.h
INT64_PQ	int64 *	64-bit pointer to a 64-bit int	far_pointers.h
UINT64_PQ	uint64 *	64-bit pointer to a 64-bit int	far_pointers.h
VOID_PQ	void *	64-bit pointer to arbitrary data	far_pointers.h
VOID_PPQ	void **	64-bit pointer to a VOID_PQ	far_pointers.h
IOSB_PQ	IOSB *	64-bit pointer to an IOSB structure	iosbdef.h
IOSB_PPQ	IOSB **	64-bit pointer to an IOSB_PQ	iosbdef.h
PTE_PQ	PTE *	64-bit pointer to a PTE	ptedef.h
PTE_PPQ	PTE **	64-bit pointer to a PTE_PQ	ptedef.h

A.2 Buffer Object Descriptor (BOD)

This section describes the additions and changes to cells in the buffer object descriptor (BOD) structure (see Table A-2).

Table A-2 BOD Structure Changes

Field	Type	Comments
bod$v_s2_window	Bit	A bit equal to BOD$M_S2_WINDOW in the bod$l_flags cell. When this bit is clear, the buffer object is mapped into the S0/S1 portion of system space and the bod$ps_svapte and bod$l_basesva cells are valid. When this bit is set, the buffer object is mapped into the S2 portion of system space and the bod$pq_va_pte and bod$pq_basesva cells are valid.
bod$pq_basepva	VOID_PQ	Process virtual address for the start of the buffer object. This cell replaces the bod$l_basepva cell.
bod$l_basepva	-	This cell will be removed. It will be replaced by the bod$pq_basepva cell.
bod$pq_basesva	VOID_PQ	System virtual address for the start of the buffer object. This cell is overlaid on the bod$l_basesva cell and this use is valid only if BOD$M_S2_WINDOW is set in bod$l_flags .
bod$pq_va_pte	PTE_PQ	Virtual address for the first system PTE that maps the buffer object. This cell is overlaid on the bod$ps_svapte cell and this use is valid only if BOD$M_S2_WINDOW is set in bod$l_flags .

A.3 Buffered I/O (BUFIO)

The existing 32-bit Buffered I/O (BUFIO) packet format will continue to be supported. In addition, a new 64-bit BUFIO packet format will be supported. These BUFIO packets are "self identifying". That is, it is possible to distinguish a 32-bit from a 64-bit format BUFIO packet from information in the packet.

Although the structure type code DYN$C_BUFIO is defined and there is an expected layout for the header of buffered I/O packet, there currently is no formal definition of a structure. Existing code in drivers and IOCIOPOST.MAR uses numeric constants as offsets.

The existing 32-bit BUFIO packet will be formally defined along with a new 64-bit BUFIO packet format. The 64-bit BUFIO structure format will also be used for 64-bit diagnostic buffer packets (see Table A-3).

Table A-3 BUFIO Packet

Field	Type	Comments
bufio$ps_pktdata	void *	Pointer to the buffered data within the packet.
bufio$ps_uva32	void *	32-bit pointer to user's address space. On a read function, data is transfered from that user virtual address to the buffer packet during FDT processing. On a write function, data is transfered to that user virtual address from the buffer packet during I/O Postprocessing. If this cell contains the value BUFIO$K_64 (-1), then the pointer to the user buffer is in bufio$pq_uva64 .
bufio$w_size	unsigned short	Size of the BUFIO packet in bytes.
bufio$b_type	unsigned char	Nonpaged pool packet type code, DYN$C_BUFIO
BUFIO$K_HDRLEN32	constant	Size in bytes of the minimal buffered I/O packet header with a 32-bit user virtual address (12).
bufio$pq_uva64	VOID_PQ	64-bit pointer to user's address space. On a read function, data is transfered from that user virtual address to the buffer packet during FDT processing. On a write function, data is transfered to that user virtual address from the buffer packet during I/O Postprocessing. This cell contains a valid address only if the bufio$ps_uva32 cell contains the value BUFIO$K_64 (-1).
BUFIO$K_HDRLEN64	constant	Size in bytes of the minimal buffered I/O packet header with a 64-bit user virtual address (24).

A.4 Complex Chained Buffer (CXB)

The CXB structure defines the format of entries that are linked together to build a complex chained buffered I/O packet.

The CXB structure will be enhanced such that it can be used by existing code with no source changes to support a 32-bit caller's buffer address. However, the same enhanced CXB structure can be used to support a 64-bit caller's buffer address as well (see Table A-4).

Table A-4 CXB Structure Changes

Field	Type	Comments
cxb$ps_pktdata	void *	Pointer to the buffered data within the packet. This cell will be overlaid on the existing cxb$l_fl cell to reflect its current alternate use.
cxb$ps_uva32	void *	32-bit pointer to user's address space. If this cell contains the value BUFIO$K_64 (-1) then the pointer to the user buffer is in cxb$pq_uva64 . This cell will be overlaid on the existing cxb$l_bl cell to reflect its current alternate use.
cxb$pq_uva64	VOID_PQ	64-bit pointer to user's address space. This cell contains a valid address only if the cxb$ps_uva32 cell contains the value BUFIO$K_64 (-1). This cell will be inserted as the last aligned quadword just before the end of the standard CXB header which is CXB$K_LENGTH bytes long.

A.5 Data Chain Block (DCBE)

The DCBE structure is the Data Chain Block that is used by the OpenVMS LAN driver VMS Communications Interface (VCI). A DCBE is used to connect to a VCRP all or part of the data to be transmitted. A chain of DCBEs is used when the data is contained in more than one discontiguous buffer in virtual memory.¹

There are two mutually exclusive methods that a DCBE can use to identify the start of the buffer:

1. When the dcbe$l_buffer_address cell contains a zero, the buffer address is specified by the dcbe$l_svapte and dcbe$l_boff cells. A fixed-size primary DIOBM structure will be added to the DCBE. This embedded DIOBM structure is available for use by an upper-level VCM if it needs to derive a 32-bit SVAPTE from a 64-bit VA_PTE for the PTEs that map the buffer. The lower-level VCM will not alter this embedded DIOBM or make any assumptions about it.
2. When the dcbe$l_buffer_address cell contains the a non-zero value, this value is the system virtual address of the buffer. This method remains unchanged.

Because a VCRP can also be used as a DCBE, the named DCBE cells must be at the same offsets as their VCRP counterparts. Therefore, DCBE changes are reflected in the VCRP and changes to the common portion of the VCRP are reflected in the DCBE.

In addition, SYS$PEDRIVER overlays a DCBE with the vcrp$t_internal_stack area within the VCRP. Therefore, an increase in the size of the DCBE must be reflected by a corresponding increase in the size of the internal stack area within the VCRP (see Table A-5).

Table A-5 DCBE Structure Changes

Field	Type	Comments
dcbe$l_reserved	int32[13]	This existing vector of 6 filler longwords has been increased to 13 fill longwords to reflect the increased size of the common portion of the VCRP. The common portion of the VCRP has been increased to accommodate either an ACB64 or ACB structure.
dcbe$pq_buffer_addr64	VOID_PQ	64-bit buffer address. This cell is available for use by upper-level VCMs only. Note that this cell does not replace the dcbe$l_buffer_address cell which continues to be used by lower-level VCMs. The dcbe$pq_buffer_addr64 cell has been added after the dcbe$l_bcnt cell.
dcbe$r_diobm	DIOBM	Embedded fixed-size primary "direct I/O buffer map" structure. This DIOBM structure is available for use by upper-level VCMs that need to lock down a buffer and provide a value for the dcbe$l_svapte cell. This structure has been added just before the end of the DCBE header.

	privacy and legal statement
6466PRO_006.HTML