johnk@opel.uu.net (John Kennedy) (01/14/90)
This is a description of the way interrupts are masked out and
enabled in Sys V/AT. This posting is as a result of mail I've
received requesting this information.
This is the way I have come to understand the mask tables. This
information may not be infallible, so temper it with your own
experiences.
Thanks to mike@kc5n.UUCP (Mike Barnes), for information that
led me to understand how this all works.
_____
Whenever the system wishes to enable or disable a group of interrupts,
it does so by entering a new spl() level. In V/AT, the levels that
are used are 0, 4, 5, 6 and 7. Each level has associated with it
an interrupt mask which is used to disable and enable interrupts upon
entering that and leaving that level. The specific interrupts are
designated by two arrays, named
mastbl and slavtbl,
corresponding to the masks to be applied to the master interrupt controller
(PIC) and slave PIC, respectively.
The default settings are:
mastbl[0] [1] [2] [3] [4] [5] [6] [7]
00 xx xx xx 41 E1 E3 FF
slavtbl[0] [1] [2] [3] [4] [5] [6] [7]
00 xx xx xx 61 61 61 61
The xx's represent a don't care value, as levels 1,2 and 3 are not used (to
the best of my knowledge).
For each byte of either table, a 1 indicates that an interrupt is disabled,
and a 0 represents that the interrupt is enabled.
On the master interrupt controller, the interrupts are allocated as:
bit [0 = low order bit] bit
0 IRQ0 (Clock) 4 IRQ4 Serial (tty0)
1 IRQ1 Console Keyboard 5 IRQ5 Parallel Port 2
2 IRQ2 Slave PIC 6 IRQ6 Floppy Controller
3 IRQ3 Serial (tty1) 7 IRQ7 Parallel Port 1
On the slave interrupt controller, the interrupts are allocated as:
bit bit
0 IRQ8 Real time clock 4 IRQ12 Spare
1 IRQ9 PC/XT IRQ2 5 IRQ13 Math Coprocessor
2 IRQ10 Spare 6 IRQ14 Spare Hard Disk Controller
3 IRQ11 Spare 7 IRQ15 Spare
From the 'Adding Devices Drivers' section of TFM, we see that Microport
allocates each of the spl() levels as follows:
SPL Value Interrupt(s) Disables mastbl slavtbl
masked interrupts on:
spl0 none none 00 01
spl4 0,6,8,13,14 clock, floppy disk, 41 61
math coproc., hard disk
spl5 0,6,8,13,14,5,7 above plus parallel ports E1 61
spl6 0,6,8,13,14,5,7,1 above plus keyboard E3 61
spl7 0,6,8,13,14, above plus serial ports FF 61
5,7,1,3,4,2 and slave PIC
The kernel (/unix.lk.hd) that was shipped with my 2.4.0L system had these
locations set to:
SPL Value Interrupt(s) mastbl slavtbl
masked
spl0 8,9,10,11,12 00 1F
spl4 0,6,8,9,10,11,12,13,14 41 7F
spl5 0,6,5,7,8,9,10,11,12,13,14 E1 7F
spl6 0,6,5,7,1,8,9,10,11,12,13,14 E3 7F
spl7 0,6,5,7,1,3,4,2,8,9,10,11,12,13,14,15 FF FF
This is contrary to the documentation, but shows that Microport had
liked to keep the unused IRQs 9,10, 11 and 12 disabled some of the time
and disable IRQ15 with the serial ports. This is and should be of no
consequence, as Microport didn't use the interrupts on the slave PIC.
Now, let's assume that you've build a new kernel that uses 5 serial ports.
Let's assume that tty0, tty1, tty2, tty3, and tty4 use interrupts
4, 3, 7, 9, and 10, respectively.
Let's assume that our serial card has the capability to access these interrupts.
In my personal case, I have cut and jumpered cards to use new interrupts.
In the case of an 8-bit serial card needing to access IRQs 8-15, jumpers
can run to a nearby memory card, or any other containing the extended bus
connectors.
Notice that tty2 uses IRQ7, which is normally used by the first parallel port.
In this example, let's assume the system has no parallel ports. Thus, the
interrupt is available for use, but must be enabled when the other serial ports
are enabled, and not with the parallel ports. The new, previously spare
interrupts must be enabled with the serial interrupts. The new values
for the master and slave tables look like:
SPL Value Interrupt(s) Disables mastbl slavtbl
masked interrupts on:
spl0 none none 00 01
spl4 0,6,8,13,14 clock, floppy disk, 41 61
math coproc., hard disk
spl5 0,6,8,13,14 *above plus parallel ports 41 61
spl6 0,6,8,13,14,1 above plus keyboard 43 61
spl7 0,6,8,13,14,1,
4,3,7,9,10,2 above plus serial ports DF 67
and slave PIC
*(no change, as no parallel ports exist in this example)
/etc/patch is used to change the values. The kernel is patched for the above
example by:
/etc/patch kernelname -b +0 mastbl 0x00 # nothing
/etc/patch kernelname -b +0 slavtbl 0x01 # clock timer
/etc/patch kernelname -b +4 mastbl 0x41 # + floppy, clock
/etc/patch kernelname -b +4 slavtbl 0x61 # + winch, math
/etc/patch kernelname -b +5 mastbl 0x41 # + parallel (none)
/etc/patch kernelname -b +5 slavtbl 0x61 #
/etc/patch kernelname -b +6 mastbl 0x43 # + keyboard
/etc/patch kernelname -b +6 slavtbl 0x61 #
/etc/patch kernelname -b +7 mastbl 0xdf # + slave pic & int 3, 4, 7 (serial)
/etc/patch kernelname -b +7 slavtbl 0x67 # + int 9 & 10
Other combinations are left as an exercise for the reader :-).
Before building the kernel, don't forget that /usr/linkkit/cf/master
must be edited to reflect what device uses each interrupt and the
total count of each kind of device, and /usr/linkkit/cf/dfile.wini
must be edited to reflect the total count of each kind of device.
Don't forget, too, that any card that allows IRQ2 gets that interrupt
serviced by IRQ9 on the software end. The bus line for IRQ2 is rerouted to
IRQ9 on the slave PIC, and IRQ2 is unusable on the AT bus.
John
--
John Kennedy johnk@opel.uu.uunet
Second Source, Inc.
Annapolis, MD