perm filename INFO.KL[KL,SYS]6 blob
sn#250645 filedate 1976-12-02 generic text, type C, neo UTF8
COMMENT ⊗ VALID 00018 PAGES
C REC PAGE DESCRIPTION
C00001 00001
C00003 00002 CONO PI,bit 22 clears only the requests initiated with CONO PI,bit 24.
C00009 00003 EPT and UPT locations
C00011 00004 EBOX/MBOX interface
C00021 00005 DTE20 KL10 I/O Functions
C00023 00006 DTE20 PDP11 Registers
C00030 00007 DTE20 Operations
C00055 00008 EBUS PI CYCLE
C00064 00009 DTE Miscellaneous
C00065 00010 CRAM and DRAM Miscellaneous
C00067 00011 KL10 Clock Card
C00078 00012 PC Flags
C00085 00013 PAGING
C00090 00014 CON Miscellaneous
C00092 00015 KLDCP
C00093 00016 NXM, Parity Errors
C00094 00017 The MBOX
C00099 00018 DMA20
C00103 ENDMK
C⊗;
CONO PI,bit 22 clears only the requests initiated with CONO PI,bit 24.
In fact, that is the only way such a request can be cleared.
DTE20 is device 200 same as DC
AS is device 774 is unrestricted
CONO PAG,E and CONI PAG,E
bit function
18 cache look (enables looking in the cache)
19 cache load (enables writing into the cache)
20 unused
21 KL10 paging mode
22 trap and paging enable
23-35 executive base register (EBR)
DATAO PAG,E
bit function
0 enable load current AC block and previous AC block
1 enable load CWSX and previous section
2 enable load UBR
3-5 unused
6-8 current AC block
9-11 previous AC block
12 CWSX
13-17 previous section
18 inhibit storing accounting meters
24-35 UBR
Microcode is assembled without KA10 floating point long mode instructions.
They will trap as UUOS. JSYS doesn't exist. It will also trap as a UUO.
LUUOs trap to 40 and 41 regardless of exec/user mode.
New instruction BLKI PAG,E loads AC block 6 register 6 from C(E) and register
7 from C(E+1). This is intended for loading the page table pointers to the
exec and user page tables respectively.
EPT and UPT locations
EPT location function
UPT location function
424 KA syle MUUO stored here (XR and @ bits are off,rh = effective addr)
425 KA style flags,,PC+1 of MUUO stored here
426 process context word: 60wxyz,,UBR in DATAO PAG, format, where
w = current AC block, x = previous AC block, yz = CWSX,prev section
430 KERNEL NO TRAP new MUUO PC word
431 KERNEL TRAP new MUUO PC word
432 SUPERVISOR NO TRAP new MUUO PC word
433 SUPERVISOR TRAP new MUUO PC word
434 CONCEALED NO TRAP new MUUO PC word
435 CONCEALED TRAP new MUUO PC word
436 PUBLIC NO TRAP new MUUO PC word
437 PUBLIC TRAP new MUUO PC word
EBOX/MBOX interface
Microcode MEM field
decode name
1 MCL MEM/ARL IND
2 MCL MEM/MB WAIT
3 MCL MEM/SECTION ZERO
4 MCL MEM/AREAD
5 MCL MEM/B WRITE
6 MCL MEM/FETCH
7 MCL MEM/REG FUNC
10 MCL MEM/A@
11 MCL MEM/B@
12 MCL MEM/LOAD AR
13 MCL MEM/LOAD ARX
14 MCL MEM/AD FUNC
15 MCL MEM/BREAD
16 MCL MEM/WRITE
17 MCL MEM/RPW CYCLE
The EBOX requests a cycle from the MBOX by asserting MCL MBOX CYC REQ.
This will happen if:
1) MCL MEM 00 is true (decodes 10-17)
2) MCL MEM/AREAD is true and the DRAM A field is not 0 or 2 (0 is immediate,
2 is illegal)
3) MCL MEM/FETCH and -CON PI CYCLE are both true
4) MCL MEM/B WRITE is true and the DRAM B field bit 01 is on (DBL BOTH, SELF,
MEM, BOTH)
5) MCL MEM/REG FUNC is true
6) MCL MEM/FETCH and #01 are true (regardless of CON PI CYCLE) and
IR TEST SATISFIED is true (fetch is conditional on test specified by
DRAM B field)
7) MCL MEM/AREAD and IR JRST 0, are true (the A field for JRST says
"immediate" and the jump field is buggered to provide a 16 way dispatch
on the ac field. Some of the JRST instructions need to do memory cycles
but JRST 0, doesn't and so a fetch is started for it here. The DRAM A
field being "immediate" causes the VMA to be loaded from the AD which
has the effective address)
8) MCL MEM/FETCH and #00 both true (#00 indicates unconditional fetch)
9) MCL SPEC/SP MEM CYCLE and #00 both true (fetch: load IR)
The signal MCL REQ EN will be true if either MEM 00 or 01 (i.e., 4-17) is true,
except for a MEM/B WRITE with DRAM B field bit 1 off. MCL REQ EN enables several
flip flops to load information about the MBOX cycle that the EBOX is just now
requesting. In particular it enables the following to be loaded:
signal page
MCL VMA READ MCL2
MCL VMA PAUSE MCL2
MCL VMA WRITE MCL2
MCL VMA FETCH MCL2
MCL LOAD AR MCL2
MCL LOAD ARX MCL2
MCL STORE AR MCL2
MCL EBOX MAP MCL6
MCL REG FUNC MCL6
APR REG FUNC EN APR6
#01-02,06-07 APR6
MCL VMA READ, PAUSE, WRITE and FETCH are used to control the MBOX when it
does this cycle for the EBOX.
Bit 2 causes CON MBOX WAIT if CON MEM CYCLE is set (set by MCL MBOX CYC REQ)
MCL LOAD VMA HELD is asserted by either MEM 02 or CON COND/LOAD VMA HELD.
It causes the VMA HELD register to be loaded from the VMA and the MCL cycle
type flip flops. The left half of this register gets loaded with:
bit function
1 MCL LOAD AR
2 MCL LOAD ARX
3 MCL VMA PAUSE
4 MCL VMA WRITE
5 MCL VMA USER
6 MCL VMA PUBLIC
7 MCL VMA PREVIOUS
8 MCL VMA EXTENDED
9 MCL VMA FETCH
10 MCL EBOX MAP
11 -MCL EBOX CACHE
12 -MCL EBOX MAY BE PAGED
VMA control (this is hairy, controlled mostly on the MCL print)
Microcode VMA field
value function
00 inhibits loading VMA and VMAX
01 (PC) VMA←PC
10 (PC+1) VMA←PC+1
11 load the VMA from the AD
any non-zero value in the VMA field (except during a satisfied jump condition)
causes MCL LOAD VMA which causes CON VMA SEL 2,1 to be set to 11 enabling
the VMA and VMAX to load. The VMA will load from the VMA AD unless the
VMA field is 11 in which case it will load from the AD. VMAX bits are
controlled by the VMA 13-17 IN mixer.
the VMA AD is normally set up to the PC. If the VMA field is 10 or 11 then
the VMA AD has PC+1 (irrelevant in the case of VMA field = 11) because
SCD TRAP MIX 35 has been set. It is also possible to get VMA←PC+1 using
VMA field 01 if MCL VMA INC is true which occurs when a non-pi skip
condition is satisfied.
COND/VMA INC and COND/VMA DEC increment and decrement the VMA respectively
leaving VMAX alone.
any time the VMA loads the VMAX does also. It will load with zeros if
MCL VMAX EN is false which will occur on RESET, MEM/SECTION ZERO,
SPEC/SP MEM CYCLE with #03 and on several PXCT/SXCT conditions.
Otherwise, it is enabled for loading from one of four places selected
by MCL VMAX SEL 2,1.
00 VMA 13-17 (VMAX)
01 PC 13-17 (PC section)
10 VMA PREV SEC 13-17
11 AD 13-17
CTL SPEC/SECTION HOLD forces 00. MCL VMA PREV EN forces 2 or 3, -MCL VMA PREV EN
forces 1 or 3. MCL VMA EXT EN, MCL MEM/AD FUNC and MCL MEM/REG FUNC force 3.
COND field decodes of 30-34 modify the VMA AD value (which is looked at
only if the VMA field is 01 or 10).
VMA AD 18-26 0
VMA AD 27-31 #0-4
VMA AD 32-35 SCD TRAP MIX 32-35
SCD TRAP MIX 32-35 is determined by the COND field (except that SCD TRAP MIX 35
is ored with one if the VMA field is 10 or 11 and it is not a PI CYCLE).
cond SCD TRAP MIX 32-35
30 #5-8
31 #5,6,SCD TRAP CYC 2,1
32 #5,SCD USER,PUBLIC,TRAP CYC 1 OR 2
33 AR 32-35
34 PI2 PI 4,2,1,#8
Previous Context Execute (PXCT = XCT with non-zero AC field in exec mode only)
The AC field of the PXCT is decoded to determine when to enforce previous
context. Note that previous context may be exec mode as well as user mode.
The previous context ACs are determined by DATAO PAG, bits 9-11. For
previous context non-ac references, the PCU bit in the PC word (bit 6)
forces user mode previous context references. The PXCT AC field bits
control which parts of the target instruction are interpreted in previous
context mode as follows:
AC field value part of cycle done in previous context
10 Effective address computation of target instruction. i.e.,
for indexing, the previous context ACs are used. If the
effective address computation of the target instruction
involves an indirect cycle, then the location that is read
will be in the previous context if both this bit and the 4 bit
are on (the 4 bit enables reads and writes).
4 Any memory reference, whether reading or writing, that is
specified by E. i.e. the fetch or store of the effective
address, the source of a PUSH, or the destination of a POP
or a BLT.
2 Effective address computation of a byte pointer. i.e., indexing
specified by a byte pointer.
1 Stack word in PUSH or POP, the source in a BLT, the fetch of
the effective address of the byte pointer in a load byte
instruction.
DTE20 KL10 I/O Functions
CONO DTE0,E
bit func
00-21 -
22 set DPS5 10 REQ INT (1)
23 clear DPS5 RELOAD PDP11 (1)
24 set DPS5 RELOAD PDP11 (1)
25 -
26 clear DPS5 REQ 10 INT (1)
27-28 -
29 clear DPS5 10 TO11 NORM TERM (1) and DPS5 10 TO11 ERR TERM (1)
30 clear DPS5 10 TO10 NORM TERM (1) and DPS5 10 TO10 ERR TERM (1)
31 enable load PI0 ENABLE and PIA
* 32 bit → DPS3 PI0 ENABLE (1) if enabled by bit 31
33-35 bits → DPS3 PIA 4,2,1 (1) "
* DPS3 PI0 ENABLE (1) is ignored since the 11 is privileged (EXT DIAG BUS CON)
CONI DTE0,E
bit func
00-19 -
20 -EXT DIAG BUS CON (always reads back as zero)
21 INT1 PWR FAIL
22 DPS5 10 REQ INT (1)
23-25 -
26 DPS5 REQ 10 INT (1)
27 DPS5 10 TO11 ERR TERM (1)
28 -
29 DPS5 10 TO11 NORM TERM (1)
30 DPS5 10 TO10 NORM TERM (1)
31 DPS5 10 TO10 ERR TERM (1)
32 DPS3 PI0 ENABLE (1)
33-35 DPS3 PIA 4,2,1 (1)
DATAO DTE0,E
bit func
00-22 -
23 bit → DPS5 TO10 I BIT (1)
24-35 bits → DPS2 EBHOLD 11-00 (1)
DTE20 PDP11 Registers
UNIBUS locations 174400 - 174427 are contained in a RAM in the DTE20.
These locations may not be written in byte mode. Reading any RAM location
returns the last value written there. Writing into certain RAM locations
has side effects indicated by the functions below.
DLYCNT 174400 (RAM)
DEXWD3 174402 (RAM)
bit func
15-00 KL10 data bits 20-35
bit func
15-00 KL10 data bits 4-19
bit func
15-04 -
03-00 KL10 data bits 0-3
TENAD1 174410 (RAM)
bit func
15-13 address space
12 bit → CNT5 DEP (1)
11 * bit → CNT2 PROTECTION OFF (1)
10-07 -
06-00 KL10 address bits 13-19
* CNT2 PROTECTION OFF (1) is cleared by CNT4 DEX DONE although the corresponding
bit in the RAM location remains set.
TENAD2 174412 (RAM)
bit func
15-00 * KL10 address bits 20-35
* writing into the TENAD2 register sets CNT5 DEX START (1), which is cleared by
CNT4 DEX DONE or INT1 MST CLR and can be read as STATUS bit 2 (inverted).
TO10BC 174414 (RAM)
DIAG1 174430
bit read write (byte operations illegal)
15-09 * EBUS DS00-06 ** CNT3 DIAG DS 00-06 (1)
08 CNT4 DEX (1) -
07 CNT4 TO10 TRANS (1) *** CNT3 REMOVE STATUS
06 CNT4 TO11 TRANS (1) -
05 CNT3 DIAG 10/11 (1) CNT3 DIAG 10/11 (1)
04 - CNT5 SINGLE PLS (1)
03 - CNT3 DIAG KL10 (1)
02 - CNT3 DIAG SEND (1)
01 - -
00 - **** CNT3 DIAG COM START (1)
* the "or" of TTL DS lines and:
DS04 = CLK ERROR STOP, DS05 = CON RUN, DS06 = CON EBOX HALTED (EBOX HALT LOOP)
** if EXT DIAG BUS CON (always true) CNT3 DIAG DS 00-06 (1) get ored with
EBUS DS00-06 on the TTL bus.
*** if EXT DIAG BUS CON (always true) CNT3 REMOVE STATUS drives EBUS REMOVE
DS STATUS which causes the ecl EBUS DS00-06 E lines to be driven from
the ttl EBUS DS00-06 lines which are being driven from CNT3 DIAG DS 00-06.
**** CNT3 DIAG COM START (1) gets synchronized, then causes CNT3 STROBE (1)
which goes out as EBUS DIAG STROBE for one clock period (8 10/11 ticks). One
period after CNT3 STROBE (1) clears, the entire DIAG1 register is cleared.
10/11 clock ticks are happening at 12.5 MHz = 80 ns. (1/2 MBOX rate)
DIAG2 174432
bit read write (byte operations illegal)
15 DPS4 ADR3 (CNT1 ADR 3) -
14 DPS4 ADR2 (CNT1 ADR 2) CNT2 BUS DONE SET
13 DPS4 ADR1 (CNT1 ADR 1) -
12 DPS4 ADR0 (CNT1 ADR 0) -
11-07 - -
06 - reset (INT1 MST CLR)
05 - -
04-01 - loaded into CNT4 STATE COUNT
00 - -
STATUS 174434
bit read write (byte operations permissible)
15 DPS5 11 TO10 NORM TERM(1) set DPS5 11 TO10 NORM TERM (1)
14 - clear DPS5 11 TO10 NORM TERM (1)
13 DPS5 11 TO10 ERR TERM (1) set DPS5 11 TO10 ERR TERM (1)
12 DPS4 RAM = 0 clear DPS5 11 TO10 ERR TERM (1)
11 DPS5 10 REQ INT (1) set DPS5 10 REQ INT (1)
10 -CNT4 DEX WD1 clear DPS5 10 REQ INT (1)
09 DPS5 PAR ERR (1) clear DPS5 PAR ERR (1)
08 DPS5 REQ 10 INT (1) set DPS5 REQ 10 INT (1)
07 DPS5 11 TO11 NORM TERM (1) set DPS5 11 TO11 NORM TERM (1)
06 CNT6 E-BUF SEL clear DPS5 11 TO11 NORM TERM (1)
and DPS5 NULL STOP (1)
05 DPS5 NULL STOP (1) set DPS5 11 INTERRUPT EN (1)
04 DPS5 EBUS PARITY ERROR (1) * clear DPS5 EBUS PARITY ERROR (1)
03 -EXT DIAG BUS CON (always 0) clear DPS5 11 INTERRUPT EN (1)
02 CNT5 DEX START (0) * set DPS5 EBUS PARITY ERROR (1)
01 DPS5 11 TO11 ERR TERM (1) set DPS5 11 TO11 ERR TERM (1)
00 DPS5 11 INTERRUPT EN (1) clear DPS5 11 TO11 ERR TERM (1)
* there is a bug whereby writing these bits with byte instructions won't work!
DIAG3 174436
bit read write (byte operations illegal)
15 -CNT1 SWAP SEL LT -
14 DPS4 PARITY (1) -
13-09 DPS7 CAP DATA 15 - 11 -
08 DPS7 CAP DATA A00 -
07-06 - -
05 - CNT8 CAP DATA CLK
04 CNT8 DATO PAR ERR (1) clear CNT8 DATO PAR ERR (1)
and CNT8 DATO REC ERR (1)
03 CNT5 WRITE EVEN PAR (1) bit → CNT5 WRITE EVEN PAR (1)
02 CNT8 DATO REC ERR (1) -
01 CNT8 NPR PAR ERR (1) clear CNT8 NPR PAR ERR (1)
00 - bit → CNT5 TO10 BYTE MODE (1)
DTE20 Operations
Examine/Deposit
The PDP-11 can examine and deposit in KL10 lgoical memory (i.e., possibly
in the cache) by using the five RAM locations:
DEXWD1, DEXWD2, DEXWD3, TENAD1 and TENAD2
For a deposit operation, the PDP-11 writes the 36 bit data word into
DEXWD1-3 and writes the deposit address into TENAD1-2. For an examine,
it writes only the address and the data appears in DEXWD1-3. None of these
registers are changed by the DTE20 and so their contents may be reused
for future examines and deposits.
The examine or deposit operation is initiated by writing into TENAD2. Therefore
this should be the last location set up by the PDP-11 before the operation.
TENAD1 also contains several bits giving more information about the operation.
Bits 15-13 indicates the address space in which the operation is to be performed.
Bit 12 indicates whether the operation is to be an examine or a deposit.
Bit 11 is set to do an unprotected examine or deposit. This bit acts in a very
stange way. There is a flip flop for this bit in addition to the memory for
it in the RAM. This flip flop is set from bit 11 on a UNIBUS DATO operation
to the TENAD1 register. So, before the operation begins, the flip flop and
the RAM agree. The protectedness of the examine or deposit is determined by
the flip flop rather than by the RAM location. After the operation completes,
however, the flip flop is cleared (set to protected), although the RAM location
is unchanged. Therefore, if another examine/deposit is attempted without first
writing into TENAD1 it will be a protected operation, independent of the state
of bit 11 in the TENAD1 RAM location. DEC's reasoning behind this is to prevent
runaway deposits from clobbering more than one location.
THE DTE20 derives its clock from the EBUS 10/11 CLK line. This is a 12.5 MHz
clock (1/2 MBOX rate) and is called 16 MHz on the CLK print. This clock is used
to produce 8 clock time states in sequence: CNT7 SP CLK (1), CNT7 STATE CLK (1),
CNT7 STATE CLK+1 (1), CNT7 STATE CLK+2 (1), CNT7 WR CLK-1 (1), CNT7 WR CLK (1),
CNT7 WR CLK+1 (1) and CNT7 SHIFT CLK (1). Each time state is on for one period
of the 10/11 clock, i.e. 80 ns. In addition, there are four clocks which are
gated versions of some of the above 8, namely: CNT4 GD ST CLK, CNT4 GD WR CLK,
CNT4 GD WR CLK+1 and CNT4 GD SHF CLK. These clocks are gated on by
-CNT4 INH CLK (1). The DTE20 has three major state flip flops: CNT4 TO11 TRANS (1),
CNT4 TO10 TRANS (1) and CNT4 DEX (1). These flip flops are set in sequence by
CNT4 GD ST CLK as long as CNT4 STATE HOLD (1) is false. CNT4 STATE HOLD (1) is
clocked every CNT7 SP CLK (1). It may be set while in one of the major states
by some condition for that state.
Writing into TENAD2 sets CNT5 DEX START (1) which causes CNT4 STATE HOLD (1) to
be set on the next SP clock when the DEX (1) major state is true. This keeps
the DTE20 in the DEX major state until CNT4 STATE HOLD (1) is cleared which will
happen when CNT5 DEX START (1) is cleared which will happen on CNT4 DEX DONE or
INT1 MST CLR. CNT4 DEX (1) being true enables the DEX minor state decoder.
It decodes the low three bits of the CNT4 STATE COUNT counter. This counter
is cleared every time a new major state is entered. So as soon as the DEX
major state is entered, the decode for zero, CNT4 DEX ADR1 is true. The counter
counts on every CNT4 GD ST CLK (except where noted below). Since major state
changes occur on the same clock, the counter will attempt to clear and count
every time there is a major state change (clear wins). Thus when we stop in
some major state, the counter will be in zero for one clock period (8 10/11 clocks)
and then count to one for one period, etc. So each decode is true for exactly
one period and comes true on the CNT4 GD ST CLK.
CNT4 DEX ADR1 forces CNT1 ADR 3,2,1,0 to be 0100 (4) which is the RAM word
address (UNIBUS byte address 10) of the TENAD1 word. Since CNT1 RAM CYCLE GO
is low, the selector on the RAM address inputs select CNT1 ADR 3,2,1,0 and the
RAM reads out the TENAD1 word. In the DEX major state, every CNT4 GD SHF CLK
will cause a CNT2 E-BUF CLK unless the operation is an examine and we are in
the minor state CNT4 DEX WD1. The E-BUF selector is set by CNT6 E-BUF SEL to
load the low 16 bits of DPS2 E-BUF from the RAM outputs. The state of the high
order bits is irrelevant. So in the CNT4 DEX ADR1 minor state the low 16 bits
of the E-BUF register are loaded from the RAM TENAD1 word. The next DEX minor
state is CNT4 DEX ADR2. In this state, the RAM address is forced to 0101 or
word address 12, the address of TENAD2. Again, the E-BUF register is set to
load the low 16 bits from the RAM and shift the original low 16 bits left by 16
places. So when the E-BUF register is clocked, it will wind up with the
complete KL10 23 bit address in the low part of the E-BUF register and address
space in bits 4-6. At the same time (CNT4 GD SHF CLK) as the E-BUF is loaded
with TENAD2, CNT4 TRANS REQ (1) is set and on the very next clock pulse
(CNT7 SP CLK (1)) CNT4 INH CLK (1) is set, inhibiting further GD type clock
pulses. CNT4 TRANS REQ (1) is driven on the EBUS PI00 line if either
DPS3 PI0 ENABLE (1) or EXTN DIAG BUS CON (always true) is asserted.
The EBOX will service the EBUS PI00 request by setting EBUS CS04-06 to 000
and EBUS F00-02 to 100 = PI SERVED and asserting EBUS DEMAND. This will
generate CNT7 PI SERVED, CNT7 PI LEVEL 0 and CNT7 SEL PI LEV.
CNT7 PI SERVED and CNT7 SEL PI LEV cause CNT6 SEND CONTROL which, together
with CNT7 PI LEVEL 0 sets CNT6 TRAN ACT (1). CNT6 SEND CONTROL is also
wired to one of EBUS D08,09,10,11 depending upon which DTE20 it is. This
causes the controller number to be sent back to the EBOX as a bit number.
The EBOX will drop EBUS DEMAND and determine the highest priority controller.
Its number will be encoded on EBUS CS00-03 and EBUS F00-02 will be changed
to 101 = PI ADDRESS IN. After a delay the EBOX will again assert EBUS DEMAND.
The DTE compares EBUS CS00-03 with its hardwired physical number and if a
match occurs it sets CNT7 PHY NO. SEL. CNT7 PI ADR IN will be set as a result
of seeing EBUS DEMAND and EBUS F00-02 = 5. CNT7 SEL PI LEVEL, CNT7 PI ADR IN
and CNT7 PHY NO. SEL together cause CNT6 E-BUS SEND OP. This signal, together
with CNT4 DEX (1) and CNT6 TRAN ACT (1) cause -CNT6 OP CODE 1 and with CNT5 DEP (1)
cause CNT6 OP CODE 2. -CNT6 OP CODE 1 causes CNT6 BUS SEL A which causes
CNT6 BUS SEL Y which causes CNT6 EMIX EN. CNT6 E-BUS SEND OP causes CNT6 BUS SEL Z
which together with -CNT6 OP CODE 1 enables DPS3 E-BUS MIX 00,01,02,06. These
bits are selected from DPS3 E-BUF 04,05,06 and CNT2 PROTECTION OFF (0) resp. and
drive EBUS D00,01,02,06. Therefore the address space from TENAD1 bits 15-13
which are in E-BUF bits 04-06 are driven on EBUS D00-02. The Q bit (EBUS D06)
is driven from CNT2 PROTECTION OFF (0). CNT6 BUS SEL Z enables EBUS D03,04,05,07
to be driven from CNT6 TRAN ACT (1), CNT6 OP CODE 1, CNT6 OP CODE 2 and GND resp.
So the function is driven on EBUS D03-05. The function is 100 for examine and
101 for deposit. CNT6 BUS SEL A enables EBUS D12-19 to be driven from
DPS2 E-BUF 12-19. CNT6 BUS SEL Y enables EBUS D23,24,25,28 to be driven from
DPS2 E-BUF 23,24,25,28 resp, and CNT6 EMIX EN enables EBUS D20-22,26,27,29-35
to be driven from their respective DPS2 E-BUF bits.
In summary:
EBUS bits from use
0-2 E-BUF 04-06 address space from TENAD1 15-13
3-5 TRAN ACT,OP 1,OP 2 function
6 PROTECTION OFF (0) Q bit
7-11 not driven
12-35 E-BUF 12-35 13-35 are the KL10 address from
TENAD1 06-00 and TENAD2 15-00
CNT6 E-BUS SEND OP asserts CNT6 TRANS (EBUS XFER) and CNT6 ACK (EBUS ACK).
The EBOX will read the EBUS data lines and drop EBUS DEMAND. When this happens
CNT6 IOP SENT (1) will be set which will cause CNT4 TRANS REQ (1) to clear
which will drop the request on EBUS PI00. EBUS DEMAND clearing will drop
CNT7 PI ADR IN which will clear CNT6 E-BUS SEND OP which will drop CNT6 TRANS
(EBUS XFER) and will also drop CNT6 OP CODE 2 if the function was a deposit.
CNT6 OP CODE 2 clearing will set a flip flop on the INT1 page which, on the
next CNT7 STATE CLK+2 will clear INT1 BUS COMP (0). This will clear the D
input to the CNT4 INH CLK (1) flip flop which will clear on the next
CNT7 SP CLK. CNT4 INH CLK (1) clearing will clear the INT1 page flip flop
and will set INT1 BUS COMP (0). The CNT4 STATE counter will begin counting
again and the next minor state will be CNT4 DEX WD1. This will happen on
CNT4 GD ST CLK. It is on CNT7 SP CLK (1) that CNT4 INH CLK (1) tries to
set and by this time CNT4 DEX ADR2 will have cleared, so CNT4 INH CLK (1)
won't set. (See later on for the continuation of the DEX minor states).
If the operation was an examine then CNT6 OP CODE 2 will not have been set
in the first place, so the gated clocks remain off.
The EBOX microcode notices the interrupt request flip flop and goes off to
some code which will perform the function specified on the EBUS. In the
case of an examine, the function was 4, namely DATAO. The microcode will
perform a COND/EBUS CTL with magic numbers field = 026. This causes
APR EBOX DISABLE CS and APR EBUS F01 E. APR EBOX DISABLE CS forces the
EBUS CS00-06 lines to zero. The EBUS F00-02 lines will be 010, the EBUS
DATAO function. The microcode will then assert EBUS DEMAND. This looks
like an ordinry DATAO except that the CS lines say zero instead of the
device address. Devices that give a PI function of DATAO or DATAI assume
that the next DATAO or DATAI will be for them independent of the CS lines.
This is possible since the PI logic will not begin a new PI cycle since
the microcode has set CON PI CYCLE and this inhibits setting PI5 LOAD and TEST.
If the operation was a deposit, then the PI function was 5, a DATAI and the
microcode will perform a COND/EBUS CTL with magic numbers field 027. This
operates the same way as DATAO except that EBUS F02 is asserted generating
function 011. Again, the microcode will assert EBUS DEMAND.
If the operation is an examine, then CNT6 IOP SENT (1) will cause CNT6 SP DATAO
to be true which will cause CNT6 E-BUF SEL to be false. This selects the
EBUS as the input to the E-BUF register. EBUS DEMAND is anded with
CNT6 SP DATAO to produce CNT6 TRANS (EBUS XFER) and CNT2 E-BUF CLK which
loads DPS2 E-BUF with EBUS data. EBUS DEMAND anded with CNT6 IOP SENT (1)
produces CNT6 ACK (EBUS ACK). The EBOX will clear EBUS DEMAND which will
clear CNT6 IOP SENT (1). This will cause another flip flop to set on the
INT1 page which will eventually cause the gated clocks to start up again
and the CNT4 state counter to resume counting (see second paragraph above
this one for description of how this happens).
If the operation is a deposit, then the CNT4 STATE counter and CNT4 DEX minor
state decoder will have started again, independent of the EBUS starting up
a DATAI cycle. One or the other of these two will occur first and they
are brought together via the CNT6 DEP STROBE (1) flip flop. The CNT4 DEX
minor states will be described first. In the case of a deposit, the CNT4 DEX
minor states will also be restarted; however, DPS2 E-BUF will have already
been loaded from the DATAO cycle performed by the EBOX in response to the
PI00 request.
The next CNT4 DEX minor state is CNT4 DEX WD1. It sets CNT1 ADR 1,0 giving
RAM address 3 (word address 6).
For examine, the RAM data inputs are selected by CNT2 DPS4 SEL 2,1 which
are set to 10 by CNT4 DEX WD1. This selects DPS4 SPM 03-00 for input to
the RAM bits 03-00. CNT1 EBH CYC is false, so the DPS4 SPM 03-00 lines are
selected from DPS2 E-BUF 00-03. CNT4 DEX WD1 will also enable CNT4 GD WR CLK
to generate CNT1 WR PLS which will clock the RAM loading E-BUF 00-03
into DPS4 DATA 03-00. This stores the high four bits of the examine data
into the low four bits of DEXWD1. CNT4 DEX WD1 for examine also inhibits
CNT2 E-BUF CLK on CNT4 GD SHF CLK so the E-BUF register is not changed
during the CNT4 DEX WD1 minor state.
For deposit, CNT6 E-BUF SEL is false, enabling DPS2 E-BUF to select the RAM
for the low 16 bits. The RAM address is set to DEXWD1 as above and
CNT2 E-BUF CLK occurs on CNT4 GD SHF CLK loading DEXWD1 into the low 16 bits
of the E-BUF.
The next CNT4 DEX minor state is CNT4 DEX WD2. It sets CNT1 ADR 1 giving
RAM address 2 (word address 4).
For examine, the RAM data inputs are selected by CNT2 DPS4 SEL 2,1 which
are set to 11 by CNT4 DEX WD2. This selects DPS2 E-BUF 04-19 for input to
the RAM bits 15-00. CNT4 DEX WD2 will also enable CNT4 GD WR CLK
to generate CNT1 WR PLS which will clock the RAM loading E-BUF 04-19
into DPS4 DATA 15-00. This stores bits 4-19 of the examine data into
DEXWD2. CNT2 E-BUF CLK happens on CNT4 GD SHF CLK and CNT6 E-BUF SEL is false
so the E-BUF register shifts bits 20-35 into bits 04-19.
For deposit, CNT6 E-BUF SEL is false, enabling DPS2 E-BUF to select the RAM for
the low 16 bits and shift E-BUF bits 20-35 into E-BUF bits 04-19. The RAM
address is set to DEXWD2 as above and CNT2 E-BUF CLK occurs on CNT4 GD SHF CLK
loading DEXWD2 into the low 16 bits of the E-BUF.
The next CNT4 DEX minor state is CNT4 DEX WD3. It sets CNT1 ADR 0 giving
RAM address 1 (word address 2).
For examine, the RAM data inputs are selected by CNT2 DPS4 SEL 2,1 which
are set to 11 by CNT4 DEX WD3. This selects DPS2 E-BUF 04-19 for input to
the RAM bits 15-00. CNT4 DEX WD3 will also enable CNT4 GD WR CLK
to generate CNT1 WR PLS which will clock the RAM loading E-BUF 04-19
into DPS4 DATA 15-00. This stores bits 20-35 of the examine data into
DEXWD3. CNT2 E-BUF CLK happens on CNT4 GD SHF CLK and CNT6 E-BUF SEL is false
so the E-BUF register shifts.
For deposit, CNT6 E-BUF SEL is false, enabling DPS2 E-BUF to select the RAM for
the low 16 bits and shift E-BUF bits 20-35 into E-BUF bits 04-19. The RAM
address is set to DEXWD3 as above and CNT2 E-BUF CLK occurs on CNT4 GD SHF CLK
loading DEXWD3 into the low 16 bits of the E-BUF. The E-BUF now contains the
data to be returned for deposit when the EBOX does a DATAI.
For a deposit the DTE must wait until the EBOX does a DATAI, or if it is
already doing a DATAI, it must send the data that is in the E-BUF register.
CNT4 DEX WD3 enables CNT6 DEP STROBE (1) to be set on the trailing edge
of CNT4 GD SHF CLK which is right after the E-BUF had the low 16 bits of
deposit data shifted in. It also asserts CNT6 DEPST which enables
CNT4 INH CLK (1) to be set on the next CNT7 SP CLK (1). This will keep
us in the CNT4 DEX WD3 minor state until CNT4 INH CLK (1) is cleared.
When the EBOX does the DATAI function, CNT3 DEMAND and CMT6 IOP SENT (1)
cause CNT6 ACK (EBUS ACK). CNT7 DATAI and CNT4 DEX (1) and CNT6 IOP SENT (1)
cause CNT6 BUS SEL X,Y,Z,A and CNT6 EMIX EN. These signals cause the E-BUF
to drive the EBUS data lines. When the E-BUF has been loaded with the deposit
data and been given a chance to settle, CNT6 DEP STROBE (1) will be set
(as described above) which is anded with EBUS DEMAND to produce CNT6 TRANS
(EBUS XFER). When the EBOX drops EBUS DEMAND, CNT6 IOP SENT (1) will be cleared
and CNT4 INH CLK (1) will be reset the same way as it was for examine (described
above).
In the case of examine, the clock is not stopped in the CNT4 DEX WD3 minor
state.
In either case the clock resumes and the DTE goes into the CNT4 DEX DONE
minor state. This clears CNT5 DEX START (1). On the next CNT7 SP CLK (1)
CNT4 STATE HOLD (1) will clear and on the next CNT4 GD ST CLK the DTE will
go to the next major state and clear the CNT4 STATE counter.
EBUS PI CYCLE
(Refer to PIC1-PIC5 prints)
The EBOX detects EBUS PI01 - PI07 E on enabled channels if the PI
system is active and always detects EBUS PI00 E (strobed into PI1 PIRn
at PI5 LOAD time). The channel number of the highest priority request
is encoded on PI2 PI4,2,1 and PI2 REQ is asserted. At PI5 TEST time
the PI system tries to grab control of the EBUS. If not successful,
PI5 TEST stays true (no PI5 LOAD occurs) until it is successful.
When it succeeds in getting the EBUS, PI5 BUS PI GRANT will be set.
This enables the TTL EBUS data bits to drive the ECL EBUS data bits;
sets the PI5 CYC START flip flop; drives EBUS CS04 - 06 E from PI2 PI4,2,1
(i.e., broadcasts the selected PI channel); enables EBUS CS00 - 03 E to be
driven from PI2 SEL PHY 8,4,2,1 (not used until later); sets EBUS F00 E to
1 causing function 4 = PI SERVED to be sent on the EBUS function lines;
disables further requests by asserting PI5 INHIBIT REQ which inhibits
further PI5 LOAD and PI5 TEST assertions PI CYC START being set causes
PI2 STATE HOLD to drop. On the next (MBOX) clock tick, PI2 TIM1
will be set and the PI timer will be set to 010101. PI2 TIM1 clears
PI5 CYC START which causes PI2 STATE HOLD to again be asserted. This
in turn holds the PI2 TIMn bits in their present state. The PI timer
will count up with each clock tick until the high order bit (100000)
goes on causing PI2 TIMER DONE. This causes PI2 STATE HOLD to drop
so that on the next clock, the PI2 TIMn state register will advance
to PI2 TIM2. Also, the timer will load with 001001. This again clears
PI2 TIMER DONE which asserts PI2 STATE HOLD which holds PI2 TIM2.
PI2 TIM2 asserts EBUS DEMAND E (we therefore waited 12 ticks = 480 ns
between asserting EBUS CS04 - 06 E and EBUS F00 - 02 E and when we
asserted EBUS DEMAND E). Those controllers interrupting on the
broadcast PI channel put a 1 on the EBUS Dnn line whose bit number
corresponds to their hardwired backpanel controller address (controller
2 asserts EBUS D02 E). No acknowledge or transfer is asserted. The
PI2 timer will count until 100000 when we will go to PI2 TIM3 and new
timer value 011100 (we will have waited 24 ticks = 960 ns). PI2 TIM3
stobes EBUS D00 - 15 E into PI2 PHY NO. 00 - 15 which are priority
encoded into PI2 SEL PHY 8,4,2,1 and driven on EBUS CS00 - 03 E.
EBUS DEMAND E drops when PI2 TIM2 does. When the timer counts out
(5 ticks = 200 ns) we go to PI2 TIM4 and new timer value 010101.
PI2 TIM4 causes EBUS F02 E to latch up, sending function 5 = PI ADDRESS IN.
The timer counts out after 12 ticks = 480 ns and we go to PI2 TIM5 and
new timer value 010001. PI2 TIM5 asserts EBUS DEMAND E but prevents
the timer from counting until PI2 TRAN REC is asserted. The selected
(by EBUS CS 00 - 03 E) controller, when it sees EBUS DEMAND E places
the function word it wants executed on the EBUS D lines and asserts
transfer (XFER). XFER is synchronized in the EBOX and becomes
PI2 TRAN REC. This re-enables the timer which count 16 ticks = 640 ns
(to deskew the EBUS D lines) and we go to PI2 TIM6 and new timer
value 010011. PI2 TIM6 keeps EBUS DEMAND E up and forces PI2 TIMER DONE
which inhibits the timer from counting and causes it to repeatedly
load with 011100. PI2 TIM6 also causes PI2 STATE HOLD to remain
asserted as long as -CON PI CYCLE B is true (note that ordinarily
PI2 STATE HOLD is asserted by -PI2 TIMER DONE which is now held false).
PI2 TIM6 causes PI2 READY which causes CON INT RQ (if not CON INT DISABLE
which can be set by the microcode) which is the condition the EBOX microcode
looks at to decide to interrupt. PI2 TIM6 also causes the EBUS D07 - 10 E
lines to be driven from PI2 SEL PHY 8,4,2,1. When the microcode sets
CON PI CYCLE, PI2 STATE HOLD drops causing PI2 TIM6 to drop and PI2 TIM7
to set; this drops EBUS DEMAND E. At the same time the timer is loaded
with 011100. After 5 ticks = 200 ns we go to PI2 COMP and load the timer
with 110001 forcing PI2 TIMER DONE which holds PI2 STATE HOLD off.
PI2 COMP drops the EBUS F02 E latch and enables PI5 EBUS PI GRANT to clear
on the next clock pulse, at which time PI2 COMP also clears.
Summary of EBUS PI Cycle
1) Device asserts EBUS PIn.
2) EBOX selects highest priority PI channel and broadcasts it on EBUS CS04-06, and
sets EBUS F00-02 to 4 = PI SERVED.
3) EBOX delays 480 ns and asserts EBUS DEMAND.
4) Device that is interrupting on the broadcast PI channel asserts EBUS Dnn where
nn is the hardwired controller number of the device. Channels are 0-7, DTEs
are 8-11 and the DIA is 15.
5) EBOX delays 960 ns and strobes the EBUS and determines the controller with the
lowest number that wants to interrupt. Its controller number is broadcast
on EBUS CS00-03 and EBUS DEMAND is dropped.
6) EBOX delays 200 ns and sets EBUS F00-02 to 5 = PI ADDRESS IN.
7) EBOX delays 480 ns and asserts EBUS DEMAND.
8) The controller whose address is being broadcast in EBUS CS00-03 places
a function word on the EBUS data lines when it sees EBUS DEMAND and it
also asserts EBUS XFER.
9) EBOX delays 640 ns and then sets the interrupt request flip flop to the
microcode. EBUS DEMAND remains asserted as well as the function, CS and
DATA lines.
DTE Miscellaneous
If the 11 sets DPS5 REQ 10 INT (1) (by writing bit 8 in the status register)
then the DTE20 attempts to do a vectored trap cycle on its assigned PI
channel. However, it does not drive the vector address on the EBUS. Instead,
the microcode determines that the DTE20 is doing a vectored interrupt
and traps to 142+8*n in the EPT where n is the DTE number (0,1,2 or 3).
CRAM and DRAM Miscellaneous
The DRAM J field is 8 bits long and "ors" into the CRAM J field at
bits 1-4 and 7-10. However, you must specify the same J 1-4,7 bits
for any given even-odd pair of DRAM locations. This is because DEC,
in order to reduce the number of RAMs on the IR board, stores alternate
DRAM locations in different 256 bit RAMs. However, for bits J 1-4 and 7
they ignore the low order DRAM address bit and have only one 256 bit RAM.
If the MBOX clock is run while MR RESET is true, then all the dispatch
enables are set true (DISP EN 00-07, 00-03, 30-37 on CRA3). Furthermore,
the dispatch fields are all set to zero (CRA DISP 00-04) and all the
CRAM register bits are cleared (including CRAM J00-10). The end result
is that the CRAM address (CR ADR 00-10) is set to DIAG ADR 00-10 which
can be set by diagnostic functions. Therefore, the DTE may force the
microaddress to be whatever it wants as long as it doesn't mind the
machine being reset first.
KL10 Clock Card
Many of the signal names on the clock card mention particular frequencies,
usually factors of 64 MHz. These are all misnomers although they are
consistent. 64 MHz is really 50 MHz, etc. CLK 64 MHZ FREE is selected
from a 50 MHz crystal, a 56 MHz crystal and an external source by
CLK SOURCE SEL 2,1. CROWBAR forces the selection of 50MHz. CLK 64 MHZ FREE
is divided down into CLK 32 MHZ FREE, CLK 16 MHZ FREE and CLK 08 MHZ FREE.
CLK 16 MHZ FREE is the clock sent to the DTEs (CLK 10/11 CLK).
CLK 2*RATE SELECTED is selected by CLK RATE SEL 2,1 to be either
CLK 64,32,16 or 08 MHZ FREE. This is divided by two to generate
CLK SELECTED. This signal drives all of the flip flops that are used
for loading clock rate and source selection and diagnostic functions.
This clock never stops and ordinarily is a "32" MHz clock (really 25 MHz).
CLK GATED is also divided down from CLK 2*RATE SELECTED but in a different
way than CLK SELECTED. CLK GATED will always fall whenever CLK SELECTED
does, but will not always set when CLK SELECTED does. CLK GATED will set
only when enabled by a selector which is used to control single stepping,
bursting and running of the MBOX clock. Also, CLK GATED will only
attempt to set if CLK SELECTED is about to set (i.e., is currently cleared).
CLK GO enables CLK GATED to set each cycle (i.e., normal running).
CLK SS allows CLK GATED to set as long as CLK SS is true. CLK GATED
is delayed and, as long as CLK ERROR STOP is false produces CLK ODD A,B,C
and CLK MBOX A,B,C.
When the DTE does a DIAG CONTROL FUNC 00X or DIAG LOAD FUNC 04X, these
signals are synchronized by CLK SELECTED and produce CLK FUNC GATE which
is true for one period of CLK SELECTED. CLK FUNC GATE enables a decoder
which produces CLK FUNC START, CLK FUNC SINGLE STEP, CLK FUNC EBOX SS,
CLK FUNC COND SS, CLK FUNC BURST, CLK FUNC CLR RESET and CLK FUNC SET RESET
(these are DIAG CONTROL FUNC 001-007). It also enables a decoder which
produces CLK FUNC 042-047 (obviously DIAG LOAD FUNC 042-047). CROWBAR
also produces CLK FUNC 044-047. CLK FUNC START is loaded into CLK GO
on the next CLK SELECTED pulse. Similarly, CLK FUNC BURST and
CLK FUNC EBOX SS are loaded into CLK BURST and CLK EBOX SS respectively.
All three of these bits are cleared by CROWBAR when CLK SELECTED happens.
CLK FUNC SINGLE STEP is loaded into CLK SS on the next CLK SELECTED causing
one CLK GATED cycle coincident with the next CLK SELECTED which also happens
to shift a zero into CLK SS. CLK FUNC SET RESET and CROWBAR set the CLK RESET
flip flop which causes MR RESET 01-06 to be driven throughout the entire
machine. Typically this signal enables the resetting of the machine only
when the clock is running. CLK FUNC CLR RESET clears CLK RESET.
CLK FUNC 42 and 43 force the loading of the CLK BURST counter from the
EBUS on the next CLK SELECTED pulse. When CLK BURST is set (on the
CLK SELECTED pulse after a CLK FUNC BURST) CLK GATED is enabled and will
clock as long as the CLK BURST counter is non-zero. Each CLK SELECTED
counts the CLK BURST counter down by one (while CLK BURST is true).
CLK FUNC 44-46 load the CLK RATE and SOURCE SEL bits and various other
checking and disable bits from the EBUS.
CLK MBOX CLK is a slightly delayed version of CLK CLK which is a delayed
version of CLK MBOX C. In order to understand how the EBOX clock is generated
we must first look at the loop involving CLK SYNC, CLK SYNC EN and
CLK EBOX CLK EN. Assume that we have -CON MBOX WAIT, that is, the EBOX
is not waiting for the MBOX. This being the case, CLK EBOX CLK EN is then
the same logical signal as CLK SYNC which is loaded from CLK SYNC EN on every
CLK MBOX CLK. Note that CLK EBOX CLK EN being true forces CLK SYNC EN false.
So once CLK SYNC sets, it will clear on the next CLK MBOX CLK. Now assume
CLK SYNC is false and consequently CLK EBOX CLK EN is false. Then each
of the gates being ored into CLK SYNC EN is enabled. If the time field
of the current microinstruction is 00, then we will have CLK SYNC EN and
CLK SYNC will set on the next CLK MBOX CLK. This would give us two MBOX
clock ticks per CLK SYNC tick. If the time field is not 00 then we must
look at the signals CLK 31 NS, CLK 62 NS and CLK 93 NS. (Again, note that
these signals are nisnamed, they represent periods of 40, 80 and 120 NS
respectively). When CLK SYNC is true (CLK EBOX CLK EN also) CLK 31,62,93 NS
are cleared by CLK MBOX CLK (which will also clear CLK SYNC and hence
CLK EBOX CLK EN). If the time field is not zero, then CLK SYNC EN will not
be set and CLK SYNC will not set on the next CLK MBOX CLK. Instead, CLK 31 NS
will have a one shifted into it. This enables CLK SYEN EN if the time field is
01. On the next CLK MBOX CLK, CLK SYNC will set. In this case CLK SYNC was
on for one MBOX clock period and off for two. If the time field is 10, then
nothing will happen when CLK 31 NS goes true. On the next CLK MBOX CLK, a one
will be shifted into CLK 62 NS and this will enable CLK SYNC EN. This will
give a CLK SYNC that is on for one and off for three MBOX clock periods.
Finally, if the time field is 11, CLK SYNC EN won't set until CLK 93 NS
has a one shifted into it. Assuming -CON DELAY REQ is true, this will
give a CLK SYNC that is on for one and off for four MBOX clock periods.
If CON DELAY REQ is true, then the time field had better be three or nothing
special will happen. If it is three, then CLK SYNC EN will not go true when
CLK 93 NS does. Instead, a four bit counter will count on every CLK MBOX CLK
pulse until it reaches 17 (after 16*40 = 640 NS) at which time CLK SYNC EN will
set. The actual EBOX clocks are delayed by clocking a flip flop with CLK ODD.
The D inputs come from CLK SYNC as long as they are not diabled by the bits
loaded by CLK FUNC 045 or by error conditions.
CLK EBOX CLK is unconditionally delayed from CLK EBOX CLK EN by a flip
flop clocked with CLK MBOX CLK. If a DIAG CONTROL FUNC 00X or
DIAG LOAD FUNC 04X is given, the register containing the CLK GO, BURST and
EBOX SS bits is clocked. Unless the function specifically sets one of these
bits, they will all wind up cleared and therefore, the clocks will be stopped.
(If the function does set one of the bits, the others will be cleared.)
It is possible, therefore to stop the MBOX clock with the EBOX clocks in
the high state. If the CLK FUNC COND SS is given and CLK EBOX CLK is true,
one more MBOX clock will be generated, which will clear the EBOX clocks.
PC Flags
PCU
In exec mode, bit 6 of the PC flags is interpreted as PCU (Previous Context User).
It is this bit that causes PXCT to attempt to reference the target address in user
mode. In user mode the bit is IOT USER. It can be set in any of the following
ways:
1) By a PC storing interrupt instruction if interrupting out of user mode
2) By a flag loading instruction (JRSTF, MUUO, etc) if specified by the
data (i.e. bit 6 of the word the flags are loading from) and if the
machine is either not in user mode or is leaving user mode
It can be cleared in two ways:
1) by attempting to load it with zero (no restrictions)
2) a condition that would leave user mode will clear PCU if some other
condition doesn't try to set it at the same time. For example,
interrupting out of exec mode will clear PCU.
Therefore, PCU will be set automatically by interrupting out of user mode
(if the instruction in the interrupt location is a PC storing instruction)
and upon MUUO execution if the UPT location specifying the new PC has bit
6 on.
PUBLIC
The PUBLIC bit is bit 7 of the PC word and divides user mode and exec mode
into two submodes. In exec mode, PUBLIC off is KERNEL mode, PUBLIC on is
SUPERVISOR mode. In user mode, PUBLIC off is CONCEALED mode, PUBLIC on is
PUBLIC mode. Restrictions on which instructions are legal are determined
by exactly which of these four modes the machine is in. Switching between
the various modes is controlled by the PUBLIC and USER mode PC bits and
by the page table PUBLIC bit. These bits may sometimes be set and cleared
by a JRST 2, and can always be set and cleared by an MUUO new PC word.
In KERNEL mode everything is legal, there are no restrictions. However, if
an instruction is fetched out of a PUBLIC page (one whose page table PUBLIC
bit is set), the PUBLIC bit is set in the PC flags and the machine enters
SUPERVISOR mode. SUPERVISOR mode may be left in two ways. First, by
entering USER mode with a JRST 2, (or MUUO for that matter). In this case
the PUBLIC bit is loaded without restriction from the new PC word. Second,
by entering KERNEL mode by transferring to a PORTAL instruction (a JRST 1,)
in a private page (one whose page table PUBLIC bit is off). An analagous
situation occurs in USER mode. A program running in CONCEALED mode that
fetches an instruction from a PUBLIC page will enter PUBLIC mode; a PUBLIC
mode program may enter CONCEALED mode by transferring to a PORTAL instruction
in a private page.
SUPERVISOR mode is not exactly analagous to USER mode in that SUPERVISOR mode
may read from a private page but not write, whereas USER mode may not read
or write a private page. Finally, both SUPERVISOR and USER modes may not
transfer control to a private page except at a PORTAL (JRST 1,) instruction.
PCP
The PCP flag is loaded by a JRST 2, or trap new pc word from bit 0.
In exec mode, it acts as if it were bit 0 of the PC word and is stored by
any PC storing instruction. However, JFCL will still test the AROV bit
which is PC word bit 0 in user mode. When a PXCT instruction occurs, the
target instrcution will be executed in PUBLIC mode if PCP is set. Note
that PCP is not effected by interrupting, so that a user mode program
may set PCP with a JRST 2,. When the machine enters exec mode for an
interrupt, PCP will still be set, and it is the responsibility of the
system code to make sure PCP has the right value before the first PXCT
is done.
EBOX FLAG CONTROL
SPEC/FLAG CTL enables the magic numbers field to be interpreted as flag
control information. The bits have the following effects:
# bit effect
0 -LEAVE USER (SCD5)
1 JFCL (SCD5)
2 DISMISS/CLR PI CYCLE (CON5)
3 HALT (CON2)
4 LOAD FLAGS (SCD5)
5 PORTAL (SCD5)
6 unused
7 -LEAVE USER if in KERNEL mode (SCD5)
8 unused
PAGING
Software/Microcode
BBN paging is simulated as follows:
AC block 6 AC 6 has the exec page table page number
AC block 6 AC 7 has the user page table page number
The sign bit being on in either AC inhibits any memory references for
that mode. Pages are all forced to be private, the software bit is always
off and the cache bit is loaded from bit 5 of the page table. The access
bit (bit 12) is ignored and the read bit (bit 2) is used in its place.
The page table type bits (bits 0 and 1) are ignored and all page table
entries are assumed to be private page type. There is no CST or PSB or
address limit register (ALR). All of the exec is mapped by the exec page
table and all of the user is mapped by the user page table. Since the
CST doesn't exist, the modification bit doesn't either. Therefore, the
W bit is loaded directly into the hardware page table RAM if it is on
in the in-core page table.
Hardware
On a page fault, the EBUS REG is loaded with the following information
(see prints MBZ2, MBC1):
EBUS REG bits
00 MCL VMA USER
01-05 PF HOLD 01-05 IN
06 PT PUBLIC
07 PT CACHE
08 PAGED REF
09-13 unused
14-26 PMA 14-26
27-33 CACHE ADR 27-33
34-35 PMA 34-35
A page fault is caused by one of the following.
(A=PT ACCESS, W=PT WRITABLE, S=PT SOFTWARE, T=MCL VMA WRITE):
PAGE FAULT TYPE PF HOLD 01-05 IN EBOX HANDLE
NO PT DIR MATCH 21 Yes
NO PT ACCESS (0A)(WST) Yes
ILLEGAL WRITE (0A)(WST) Yes
PROPRIETARY VIOLATION 21 No
ILLEGAL ENTRY 21 No
ADDRESS BREAK 23 No
PT PARITY ERROR 25 No
PAGE REFILL ERROR 2(01T) No
Other values may appear in the PF HOLD 01-05 IN field if several of the page
faults happen simultaneously. The real poop is as follows:
1) The PT ACCESS bit is never true if PT MATCH is false.
2) PF HOLD 01 IN is true for NO PT DIR MATCH, ILLEGAL ENTRY, PT PARITY ERROR,
PAGE REFILL ERROR and PROPRIETARY VIOLATION.
3) PF HOLD 02 IN is true if PF HOLD 01 IN is false and PT ACCESS is true.
4) PF HOLD 03 and 04 IN are determined as follows: if PF HOLD 01 IN is false
then they are the W and S bits; if PF HOLD 01 IN is true then PF HOLD 03 IN
is PT PARITY ERROR and PF HOLD 04 IN is either ADDRESS BREAK or PAGE REFILL
ERROR.
5) If PF HOLD 01 IN is off, then PF HOLD 05 IN is true when not PAGE REFILL ERROR
and the T bit are true. If PF HOLD 01 IN is on, then PF HOLD 05 IN is true
when not PAGE REFILL ERROR is true.
The VMA HELD register is loaded on each memory cycle (and as a result of the MAP
special function) and is used to remember what the last cycle was trying to do.
The bits are as follows:
Bit Value
00 this is actually the current PC bit 0
01 MCL LOAD AR
02 MCL LOAD ARX
03 MCL VMA PAUSE
04 MCL VMA WRITE
05 MCL VMA USER
06 MCL VMA PUBLIC
07 MCL VMA PREVIOUS
08 MCL VMA EXTENDED
09 MCL VMA FETCH
10 MCL EBOX MAP
11 -MCL EBOX CACHE
12 -MCL EBOX MAY BE PAGED
13-35 VMA 13-35 at time of mem cycle
CON Miscellaneous
UCODE STATE
There are 4 microcode state bits, CON UCODE STATE 01,03,05,07.
These bits can be changed only by a COND/EBOX STATE decode.
The value set into the UCODE STATE register depends upon the
magic numbers field and the previous value in the register.
Magic Numbers bit Function
00 unused
01 set UCODE STATE 01
02 hold UCODE STATE 01
03 set UCODE STATE 03
04 hold UCODE STATE 03
05 set UCODE STATE 05
06 hold UCODE STATE 05
07 set UCODE STATE 07
08 hold UCODE STATE 07
These bits may be set in any combination. If a UCODE STATE bit is
neither held nor set, it will clear.
UCODE STATE 07 is spare and is terminated as CRM T24 16. UCODE STATE 05
does not leave the CON board and is the bit that enables the "TRACKS"
feature. It goes to the NICOND dispatch on the CON2 page. UCODE STATE
01 and 02 both go to the MTR board. UCODE STATE 01 is used as an enable
on the performance analysis counter. One of the "AND" terms is
CON UCODE STATE 01 ∨ MTR UCODE PA DONT CARE. This latter term is set
by a BLKO TIM. CON UCODE STATE 03 is used to inhibit the MBOX counter.
KLDCP
The PE command accepts an octal number controlling which parity error
conditions are enabled. The default value is 36, which enables
AR/ARX, DRAM, CRAM, and FM parity errors. The low order bit enables
FS PROBE. A change in the value of PE takes effect only on an ST (I think).
NXM, Parity Errors
There is a signal on each SBUS called SBUS ERROR. It is asserted by
the DMA20 whenever it detects a read parity error, a write parity error
or a NXM. (DMC6, MA04) This signal is received by the MBOX as MEM ERROR
and looked at on MBZ4 to produce the MBOX SBUS ERROR signal, which is
latched until a copy of it is set in the APR.
The MBOX
EBOX/MBOX interface
The EBOX requests an MBOX cycle by asserting CLK EBOX REQ which is modified
by signals on the MCL page: MCL VMA READ, MCL VMA WRITE, MCL VMA PAUSE,
MCL VMA USER, MCL VMA EPT, MCL VMA UPT, MCL EBOX CACHE, MCL EBOX MAY BE PAGED,
MCL EBOX MAP and several others. Some of these signals effect the type of
cycle that is performed, while others effect the address of the cycle and
whether the cache is involved. The VMA contains the address for the cycle
to be performed, which may possibly be a user mode, exec mode, physical or
virtual, or EPT or UPT relative address.
The Pager
The pager contains a 512 x 18 bit ram and a 128 x 7 bit ram. The 18 bit
ram contains KI10 style page table entries of the form APWSC, 13-26.
The address of this ram is determined by VMA 17-26 and MCL VMA USER and
is hashed into a 9 bit address. An attempt is made to avoid conflicts
between user and exec low page numbers. Each 7 bit ram entry contains
a directory corresponding to 4 of the 18 bit entries (hence it's a 128 entry
ram). The directory contains the physical page number for the 4 page table
entries (bits 13-17), the USER mode bit and a valid bit.
DESCRIBE HOW THE RAMS ARE USED AND LOADED.
The Cache
There are actually four caches in the KL10. Each one is 128 entries long by
word words plus a directory per entry. Each entry is called a cache line.
They are referred to as Cache 0 through Cache 3. The cache directory consists
of a physical page number (bit 14-26), four 36 bit cache data words with parity,
and a written and valid bit for every data word in the cache.
MORE CACHE DESCRIPTION. CACHE USE AND REFILL RAMS.
MBOX Control Logic
The MBOX can accept requests from five places: the EBOX, the cache clearer (CCA),
the channel and two places internal to the MBOX, namely the MBs and for PAGE
REFILL in KI10 paging mode. A priority network determines which request to
honor in the following order: 1) MB requests, 2) channel requests, 3) EBOX
requests, 4) CCA requests. A PAGE REFILL isn't properly a request, but rather
a way of restarting an MBOX cycle from the middle of the current MBOX cycle
which happens to require a page refill. It has the highest priority in the sense
that no other cycle can sneak in ahead of it. The priority network is on the
CSH1 page and produces the signals MB REQ GRANT, CHAN REQ GRANT, EBOX REQ GRANT,
CCA REQ GRANT and NON-EBOX REQ GRANT. Requests for the channel, EBOX and CCA
are enabled by CSH CHAN REQ EN, CSH EBOX REQ EN and CSH CCA REQ EN respectively.
The initiation of an MBOX cycle is under the control of two signals:
(CH1) CACHE IDLE and (CSH1) READY TO GO. CACHE IDLE is the OR of several flip
flops which can be set or cleared (synchronously) from a multitude of signals,
one of which is READY TO GO ∧ -EBOX REQ GRANT ∧ -NON-EBOX REQ GRANT. READY TO GO
is produced by an AND-OR gate one of whose inputs comes from CACHE IDLE.
So once CACHE IDLE and READY TO GO are set (which is caused by MR RESET) they
will stay latched every clock tick until either EBOX REQ GRANT or NON-EBOX REQ GRANT
is asserted.
DMA20
NXM
The DMA20 has a NXM timer which is used to detect non-existant memory
and unhang the DMA20. It also causes the signal SBUS ERROR to be asserted.
However, this signal is not to be relied on for the detection of NXM for the
following complicated reasons.
The DMA20 was designed so that it would be possible to have more than one
DMA20 on an SBUS, each with its own collection of memory. In such a system,
for any given memory request, all but one DMA20 would detect NXM while the one
that has the addressed memory will complete the cycle. It would be an error
for any of the other DMAs to send back a NXM to the MBOX. Therefore, it was
decided that if no ADR ACK was received for the word that the MBOX actually
requested, then the requesting DMA was not the one responsible for that cycle
and would not start a NXM timeout. That DMA would be unhung by the MBOX dropping
its request, either when some other DMA responds or when the MBOX does its own
NXM timeout. Now, if the DMA does receive an ADR ACK for the requested word,
then it knows that no other DMA will respond. Therefore, if some other word
that was requested at the same time is in non-existant memory, it would be
appropriate to send back a NXM. On a read cycle, this wouldn't be strictly
necessary, since tha MBOX would be waiting for the data with its own NXM timer
going, and after it timed out could drop the requests, which would unhang the
DMA. But, on a four word write there is a problem. After the DMA gets the
initial ADR ACK, begins the write of the first word, and returns an ACKN to
the MBOX, it rapidly handshakes with the MBOX and stores the remaining three
data words, returning ACKNs to the MBOX for them. Thus the MBOX will have finished
the cycle and will drop its requests. This, of course, must not reset the DMA.
However, the DMA may get a NXM on one of the remaining three write cycles and only
the DMA is left to do the NXM timeout. In fact the MBOX may have started a new
SBUS cycle during the DMA NXM timeout and it would be incorrect for that new
cycle to get a NXM because its DMA was hung. So the DMAs NXM timeout is short
compared to the one in the MBOX, and it is possible for the DMA to time out,
reset itself and do the next SBUS cycle, all before the MBOX NXM timer counts
out.