Difference between revisions of "PDP-11 Memory Management"
(→PARs and PDRs: Add PDR details) |
m (→Further reading: +Extended Memory for the LSI-11) |
||
| (19 intermediate revisions by the same user not shown) | |||
| Line 1: | Line 1: | ||
| − | [[PDP-11]]s which provide [[memory management]] use a standard '''PDP-11 Memory Management''' architecture. When memory management is enabled, the basic 64 Kbyte [[address space]] of the [[PDP-11 architecture]] is divided into 8 ''' | + | [[PDP-11]]s which provide [[memory management]] use a standard '''PDP-11 Memory Management''' [[architecture]]. When memory management is enabled, the basic 64 Kbyte [[address space]] of the [[PDP-11 architecture]] is divided into 8 '[[segment]]s' (the terminology used in the first version of the documentation; it was later changed to 'pages', for unknown reasons, but this is incorrrect - the functionality provided is that of segmentation, not [[virtual memory|paging]]). |
| − | Each segment can be set to any length between 0 bytes and 8 | + | Each segment of a virtual address space can be individually assigned to any location in physical [[main memory]]. Each segment can be set to any length between 0 bytes and 8 Kbytes, in 0100 (64.) byte increments. Segments can grow either up from their base address, or down; the latter to accomodate PDP-11 [[stack]]s, which typically grow down (from higher addresses to lower). |
| − | The [[Central Processing Unit|CPU]] can be in one of three modes; 'Kernel', 'Supervisor', and 'User'; each has its own separate | + | The [[Central Processing Unit|CPU]] can be in one of three modes; 'Kernel', 'Supervisor', and 'User'; each has its own separate mappings, to provide separate virtual address spaces for each mode. |
| − | An additional enhancement is that [[instruction]] and data fetches can be set to go to separate 64 Kbyte address spaces, the so-called '''Split I+D space''' capability; this increases the memory available to the operating system, and each user, to 128 Kbytes. | + | An additional enhancement is that [[instruction]] and data fetches can be set to go to separate 64 Kbyte address spaces, the so-called '''Split I+D space''' capability; this increases the memory available to the operating system, and each user, to 128 Kbytes. Since immediate operands and absolute addresses are stored contiguous with their instructions, fetches of them are considered to be instruction fetches, for the purposes of deciding which space to use for them.) When split I+D is not enabled in any particular mode, only a single 64 Kbyte address space is provided, using only the I-space registers for that mode. |
| − | + | [[Trap]] and [[interrupt vector]]s are fetched from Kernel Data space, when memory management is enabled; this allows bad physical main memory at [[absolute address]] 0 to be bypassed, if it is failing. | |
| − | If any instruction causes a fault (i.e. an attempt to perform a memory operation which cannot be completed, because of the settings of the memory management - e.g. an attempt to write into a segment set as 'read only'), execution of the instruction is aborted, and a memory management | + | If any instruction causes a [[fault]] (i.e. an attempt to perform a memory operation which cannot be completed, because of the settings of the memory management - e.g. an attempt to write into a segment set as 'read only'), execution of the instruction is aborted, and a memory management trap occurs. |
===Simplified subset=== | ===Simplified subset=== | ||
| − | A number of the smaller PDP-11's (notably the [[PDP-11/40]], [[PDP-11/34]], [[PDP-11/23]] and [[PDP-11/24]]) provide only a limited subset of the standard memory management facilities. In these machines, there is no support for Split I+D, and no Supervisor mode (just Kernel and User). | + | A number of the smaller PDP-11's (notably the [[PDP-11/40]], [[PDP-11/34]], [[PDP-11/60]], [[PDP-11/23]] and [[PDP-11/24]]) provide only a limited subset of the standard memory management facilities. In these machines, there is no support for Split I+D, and no Supervisor mode (just Kernel and User). |
| − | Also, there is no SSR1 register (below), which can make re-starting instructions which cause a fault laborious (the fault-handling code has to decipher the instruction and effectively simulate its operation, to work out which registers need to be 'backed out'), and in some cases impossible (e.g. instructions which auto-increment the same register several times). | + | Also, there is no SSR1 [[register]] (below), which can make re-starting instructions which cause a fault laborious (the fault-handling code has to decipher the instruction and effectively simulate its operation, to work out which registers need to be 'backed out' to their state before the instruction started execution), and in some cases impossible (e.g. instructions which [[PDP-11 architecture#Addressing modes|auto-increment]] the same register several times). |
==Registers== | ==Registers== | ||
| − | The memory management is entirely controlled by groups of registers in the CPU (unlike many virtual memory systems, which | + | The memory management is entirely controlled by groups of registers in the CPU (unlike many virtual memory systems, which keep most of the information needed for memory management in page tables held in main memory, and cached in the CPU). |
===Control=== | ===Control=== | ||
| Line 30: | Line 30: | ||
| 777572 || SSR0 || Control and status | | 777572 || SSR0 || Control and status | ||
|- | |- | ||
| − | | 777574 || SSR1 || Register increment/ | + | | 777574 || SSR1 || Register increment/decrement record |
|- | |- | ||
| 777576 || SSR2 || Virtual address associated with the fault (usually the instruction address) | | 777576 || SSR2 || Virtual address associated with the fault (usually the instruction address) | ||
| Line 37: | Line 37: | ||
|} | |} | ||
| − | + | The names were changed to SR0-SR3 after the 'Segmentation' term was dropped from the name; the forms MMR0-MMR3 are also occasionally met with. On the 'simplified subset' machines, SSR1 and (usually) SSR3 are not implemented (the [[KDF11 CPUs]] are an exception to the latter; see below). | |
| − | SSR1 contains information about register modifications performed during the course of an instruction, to allow those modifications to be 'backed out' if the instruction needs to be restarted after a memory management fault. | + | SSR0 contains control bits (e.g. enabling memory management) and status bits (e.g. information about memory operations which caused a memory management fault, and the segment involved). The layout of SSR0 is: |
| + | {{16bit-header}} | ||
| + | | Non-resident || Length || Read-only || Trap || colspan=2 | Unused || Enable Trap || Maintenance || Instruction Completed || colspan=2 | CPU Mode || I/D || colspan=3 | Page || Enable | ||
| + | {{16bit-bitout}} | ||
| + | |||
| + | When any of the first three bits (all error aborts) are set, bits 1-7, as well as SSR1 and SSR2, are frozen. On the 'simplified subset' machines, bits 12, 9, 7 and 4 are not implemented. | ||
| + | |||
| + | SSR1 contains information about register modifications (given there in [[two's complement]]) performed during the course of an instruction, to allow those modifications to be 'backed out' if the instruction needs to be restarted after a memory management fault. Its layout is: | ||
| + | {{16bit-header}} | ||
| + | | colspan=5 | Change || colspan=3 | Register || colspan=5 | Change || colspan=3 | Register | ||
| + | {{16bit-bitout}} | ||
SSR2 contains the address of the instruction which caused the memory management fault. | SSR2 contains the address of the instruction which caused the memory management fault. | ||
| − | SSR3 contains bits to enable Split I+D in the three modes; on machines which have [[UNIBUS map]]s, the enable for it is also here. | + | SSR3 contains bits to enable Split I+D in the three modes; on machines which have [[UNIBUS map]]s, the enable for it is also here. Its layout is: |
| + | {{16bit-header}} | ||
| + | | colspan=10 | Unused || Enable UNIBUS Map || Enable 22-bit || Unused || Kernel || Supervisor || User | ||
| + | {{16bit-bitout}} | ||
| + | |||
| + | The 'Enable 22-bit' bit only applies on machines which support larger memories, i.e. the [[PDP-11/70]], [[PDP-11/44]], and machines with the KDF11 CPUs and [[KDJ11 CPUs]]. | ||
===PARs and PDRs=== | ===PARs and PDRs=== | ||
| Line 54: | Line 69: | ||
{{16bit-header}} | {{16bit-header}} | ||
| − | | | + | | Don't Cache || colspan=7 | Length || Trapped || Written || colspan=2 | Unused || Direction || colspan=3 | Access Control |
{{16bit-bitout}} | {{16bit-bitout}} | ||
| Line 77: | Line 92: | ||
| 7 || Unused, reserved - abort all accesses | | 7 || Unused, reserved - abort all accesses | ||
|} | |} | ||
| + | |||
| + | On the 'simplified subset' machines, only modes 0, 2 and 6 are supported, and neither is the 'Trapped' bit (set on PDR's for memory references that trapped, on the other machines). | ||
| + | |||
| + | The 'Don't Cache' bit is only available on the [[KDJ11 CPUs]]. | ||
The addresses of the PAR/PDR sets are: | The addresses of the PAR/PDR sets are: | ||
| Line 156: | Line 175: | ||
|} | |} | ||
| − | + | ==See also== | |
| + | |||
| + | * [[KT11-B Paging Option]] - DEC experiment with paging | ||
| + | * [[KT11-C Memory Management Unit]] - the archetype | ||
| + | |||
| + | ==Further reading== | ||
| + | |||
| + | * ''Extended Memory for the LSI-11'', [[MicroNote]] #028 | ||
| + | |||
| + | ==External links== | ||
| + | |||
| + | * [http://www.bitsavers.org/pdf/dec/pdp11/memos/701014_Segmentation.pdf A Segmentation Scheme for the PDP-11 Family] - interesting design memo | ||
| − | + | [[Category: PDP-11s]] | |
Latest revision as of 18:20, 27 July 2024
PDP-11s which provide memory management use a standard PDP-11 Memory Management architecture. When memory management is enabled, the basic 64 Kbyte address space of the PDP-11 architecture is divided into 8 'segments' (the terminology used in the first version of the documentation; it was later changed to 'pages', for unknown reasons, but this is incorrrect - the functionality provided is that of segmentation, not paging).
Each segment of a virtual address space can be individually assigned to any location in physical main memory. Each segment can be set to any length between 0 bytes and 8 Kbytes, in 0100 (64.) byte increments. Segments can grow either up from their base address, or down; the latter to accomodate PDP-11 stacks, which typically grow down (from higher addresses to lower).
The CPU can be in one of three modes; 'Kernel', 'Supervisor', and 'User'; each has its own separate mappings, to provide separate virtual address spaces for each mode.
An additional enhancement is that instruction and data fetches can be set to go to separate 64 Kbyte address spaces, the so-called Split I+D space capability; this increases the memory available to the operating system, and each user, to 128 Kbytes. Since immediate operands and absolute addresses are stored contiguous with their instructions, fetches of them are considered to be instruction fetches, for the purposes of deciding which space to use for them.) When split I+D is not enabled in any particular mode, only a single 64 Kbyte address space is provided, using only the I-space registers for that mode.
Trap and interrupt vectors are fetched from Kernel Data space, when memory management is enabled; this allows bad physical main memory at absolute address 0 to be bypassed, if it is failing.
If any instruction causes a fault (i.e. an attempt to perform a memory operation which cannot be completed, because of the settings of the memory management - e.g. an attempt to write into a segment set as 'read only'), execution of the instruction is aborted, and a memory management trap occurs.
Contents
Simplified subset
A number of the smaller PDP-11's (notably the PDP-11/40, PDP-11/34, PDP-11/60, PDP-11/23 and PDP-11/24) provide only a limited subset of the standard memory management facilities. In these machines, there is no support for Split I+D, and no Supervisor mode (just Kernel and User).
Also, there is no SSR1 register (below), which can make re-starting instructions which cause a fault laborious (the fault-handling code has to decipher the instruction and effectively simulate its operation, to work out which registers need to be 'backed out' to their state before the instruction started execution), and in some cases impossible (e.g. instructions which auto-increment the same register several times).
Registers
The memory management is entirely controlled by groups of registers in the CPU (unlike many virtual memory systems, which keep most of the information needed for memory management in page tables held in main memory, and cached in the CPU).
Control
There are four registers which control the overall operation of the Memory Management Unit:
| Address | Name | Function |
|---|---|---|
| 777572 | SSR0 | Control and status |
| 777574 | SSR1 | Register increment/decrement record |
| 777576 | SSR2 | Virtual address associated with the fault (usually the instruction address) |
| 777516 | SSR3 | D-space enable/disable (per mode) |
The names were changed to SR0-SR3 after the 'Segmentation' term was dropped from the name; the forms MMR0-MMR3 are also occasionally met with. On the 'simplified subset' machines, SSR1 and (usually) SSR3 are not implemented (the KDF11 CPUs are an exception to the latter; see below).
SSR0 contains control bits (e.g. enabling memory management) and status bits (e.g. information about memory operations which caused a memory management fault, and the segment involved). The layout of SSR0 is:
| Non-resident | Length | Read-only | Trap | Unused | Enable Trap | Maintenance | Instruction Completed | CPU Mode | I/D | Page | Enable | ||||
| 15 | 14 | 13 | 12 | 11 | 10 | 09 | 08 | 07 | 06 | 05 | 04 | 03 | 02 | 01 | 00 |
When any of the first three bits (all error aborts) are set, bits 1-7, as well as SSR1 and SSR2, are frozen. On the 'simplified subset' machines, bits 12, 9, 7 and 4 are not implemented.
SSR1 contains information about register modifications (given there in two's complement) performed during the course of an instruction, to allow those modifications to be 'backed out' if the instruction needs to be restarted after a memory management fault. Its layout is:
| Change | Register | Change | Register | ||||||||||||
| 15 | 14 | 13 | 12 | 11 | 10 | 09 | 08 | 07 | 06 | 05 | 04 | 03 | 02 | 01 | 00 |
SSR2 contains the address of the instruction which caused the memory management fault.
SSR3 contains bits to enable Split I+D in the three modes; on machines which have UNIBUS maps, the enable for it is also here. Its layout is:
| Unused | Enable UNIBUS Map | Enable 22-bit | Unused | Kernel | Supervisor | User | |||||||||
| 15 | 14 | 13 | 12 | 11 | 10 | 09 | 08 | 07 | 06 | 05 | 04 | 03 | 02 | 01 | 00 |
The 'Enable 22-bit' bit only applies on machines which support larger memories, i.e. the PDP-11/70, PDP-11/44, and machines with the KDF11 CPUs and KDJ11 CPUs.
PARs and PDRs
Each segment (up to 48 in total; 8 each Instruction and Data, for the three different modes) is described by a pair of registers, a Page Descriptor Register (PDR) and a Page Address Register (PAR).
The PAR contains the base physical address for the segment (in units of 0100/64. bytes). On machines which have only UNIBUS memory (e.g. PDP-11/40, PDP-11/45, etc) these registers are 12 bits long, since the maximum amount of physical memory on these machines is 248 Kbytes. On machines which support up to 4 MBytes of physical memory (e.g. PDP-11/70, PDP-11/44, PDP-11/23, etc), they are 16 bits long.
The PDR contains the segment's length, along with the direction of growth, the access control field (read/only, read/write, etc), a 'dirty' bit (maintained by the hardware), etc. The layout of the PDR is:
| Don't Cache | Length | Trapped | Written | Unused | Direction | Access Control | |||||||||
| 15 | 14 | 13 | 12 | 11 | 10 | 09 | 08 | 07 | 06 | 05 | 04 | 03 | 02 | 01 | 00 |
The values of the 'Access Control' field are:
| Value | Mode |
|---|---|
| 0 | Non-resident - abort all accesses |
| 1 | Read-only watched - abort on write, trap on read |
| 2 | Read-only - abort on write |
| 3 | Unused, reserved - abort all accesses |
| 4 | Read/write all watched - trap on read or write |
| 5 | Read/write watched - trap on write |
| 6 | Read/write - none |
| 7 | Unused, reserved - abort all accesses |
On the 'simplified subset' machines, only modes 0, 2 and 6 are supported, and neither is the 'Trapped' bit (set on PDR's for memory references that trapped, on the other machines).
The 'Don't Cache' bit is only available on the KDJ11 CPUs.
The addresses of the PAR/PDR sets are:
| Address | Name | Function |
|---|---|---|
| 772200 | SISD0 | Supervisor I-Space PDR #0 |
| ... | ||
| 772216 | SISD7 | Supervisor I-Space PDR #7 |
| 772220 | SDSD0 | Supervisor D-Space PDR #0 |
| ... | ||
| 772236 | SDSD7 | Supervisor D-Space PDR #7 |
| 772240 | SISA0 | Supervisor I-Space PAR #0 |
| ... | ||
| 772256 | SISA7 | Supervisor I-Space PAR #7 |
| 772260 | SDSA0 | Supervisor D-Space PAR #0 |
| ... | ||
| 772276 | SDSA7 | Supervisor D-Space PAR #7 |
| 772300 | KISD0 | Kernel I-Space PDR #0 |
| ... | ||
| 772316 | KISD7 | Kernel I-Space PDR #7 |
| 772320 | KDSD0 | Kernel D-Space PDR #0 |
| ... | ||
| 772336 | KDSD7 | Kernel D-Space PDR #7 |
| 772340 | KISA0 | Kernel I-Space PAR #0 |
| ... | ||
| 772356 | KISA7 | Kernel I-Space PAR #7 |
| 772360 | KDSA0 | Kernel D-Space PAR #0 |
| ... | ||
| 772376 | KDSA7 | Kernel D-Space PAR #7 |
| 777600 | UISD0 | User I-Space PDR #0 |
| ... | ||
| 777616 | UISD7 | User I-Space PDR #7 |
| 777620 | UDSD0 | User D-Space PDR #0 |
| ... | ||
| 777636 | UDSD7 | User D-Space PDR #7 |
| 777640 | UISA0 | User I-Space PAR #0 |
| ... | ||
| 777656 | UISA7 | User I-Space PAR #7 |
| 777660 | UDSA0 | User D-Space PAR #0 |
| ... | ||
| 777676 | UDSA7 | User D-Space PAR #7 |
See also
- KT11-B Paging Option - DEC experiment with paging
- KT11-C Memory Management Unit - the archetype
Further reading
- Extended Memory for the LSI-11, MicroNote #028
External links
- A Segmentation Scheme for the PDP-11 Family - interesting design memo
