was converted to use Mach VM for Net2/4.4BSD. The user segment table
pointer was originally stored in the PCB. When Mach VM came along,
however, it was also stored in the pmap, and loaded into the PCB in
pmap_activate(). pmap_activate() would then note that the PCB's USTP
was now in sync with the pmap's USTP, and the low-level context switch
code would use the value from the PCB.
However, pmap_activate() would also load the hardware MMU context if
the pmap was the current pmap (or, in the case where pmaps can be shared,
such as in NetBSD, if the proc was the current proc). The low-level
context switch code would then reload the hardware _again_ using the
USTP from the PCB.
However, the optimization of not calling pmap_activate() if "stchanged"
was false ended up causing some processes to use stale USTP values from
the PCB when the low-level context switch code reloaded the hardware!
This was noticed by using a real vfork(2) (which worked for some time
before failing, surprisingly!)
Since I'm hard pressed to find any real optimization here (since the
hardware was always reloaded once, sometimes twice!), the code now always
calls pmap_activate(), which uses the correct USTP value (the one in the
pmap). The PCB's USTP is now ignored, and should eventually be g/c'd.
Another optimization can actually be performed, and I have added a comment
describing what it is, but have not yet implemented it.
Also note that most of the loadustp() functions where actually incomplete.
This has been corrected. These functions should probably be split up into
MMU-specific operations, and called indirectly, rather than doing constant
run-time decision making based on values that will never change during the
course of a boot's lifetime.
it in initializing during autoconfig. Similar to sys/arch/sun3/dev/zs.c
revision 1.47. Ought to fix hangs at first tty access reported by
Johnny Lam, <jlbg+@andrew.cmu.edu>.
causes the MI interrupt handler to barf when we get a 5380 RST interrupt
while probing. Worse, the VIA latches the interrupt, so simply having
all interrupts disabled during autoconfig doesn't resolve the problem.
[I demonstrated the latter on a IIci, which erroneously reports a
reselection attempt(!) after autoconfig is complete. The latched
interrupt results from the SCSI bus reset we do when initializing the
bus.]
Since interrupts must be enabled during autoconfig anyway (sigh), test
to see if autoconfig has completed in sbc_irq_intr(). If not, we don't
pass the interrupt up to the MI interrupt handler. Also, make sure to
clear the VIA interrupt if we're servicing an unclaimed 5380 RST
interrupt.
Thanks to Bill Studenmund for providing the key insight needed to unlock
this problem.
interrupt properly on a SuperMac Spectrum/8 Series III, and thanks
to Dan McMahill for loaning the card to Paul.
I modified Paul's patch somewhat to change grfmv_intr_generic_{1,4}
to grfmv_intr_generic_write{1,4} and added grfmv_intr_generic_or4 to
handle this card.
deal with cheap CD-ROMs and other devices that do not appear to have
any way to enable parity generation. In the future, it might be nice
to have this configurable on a per-device basis with back-to-back
parity errors automatically disabling parity for the device. That
would require some MI changes.
* Handle unaligned and odd-length transfers.
This could probably be handled better in the future.
but is a marked improvement. This takes advantage of a pseudo-DMA hardware
hack of Apple's that exposes a 16-bit register that the Apple-designed
memory controller acts like a DMA controller and handshakes into or out
of the FIFO. Wierd.
Unfortunately, there does not seem to be a good way to determine what
variety of comm-slot card is present in a machine. There is still an
interrupt issue preventing these cards from working--hopefully that will
be ironed out shortly.
normal rei course. If the handler returns non-zero, just rte.
This should allow better MACE response-time and still keep serial
interrupt overhead to a minimum on older, slower machines.
internal ethernet on the Quadra/Centris 660av/840av.
Add initial support for the PSC (DMA controller) to support the above
(DMA SCSI remains unsupported). This involved also changing the way
that several interrupts are handled.
Above from David Huang <khym@bga.com>
Since the interrupts changed somewhat, we must also make the ipls
dynamic, defaulting to their prior levels and adjusted for the AVs.
I modelled this on the hp300.
pseudo-device rnd # /dev/random and in-kernel generator
in config files.
o Add declaration to all architectures.
o Clean up copyright message in rnd.c, rnd.h, and rndpool.c to include
that this code is derived in part from Ted Tyso's linux code.
(mrz5149@cs.rit.edu) for the addresses and hints as to how the interrupts
might be disabled, and thanks to Henry Hotz (h.b.hotz@jpl.nasa.gov) for
testing on the 840AV.
Unmap the DAFB regs on the other Quadras after attachment. They're not
used.
really be dependent on DEBUG, not DIAGNOSTIC. While we're here, add
a couple of DEBUG messages to mac68k_calibrate_delay(), and wrap them
all appropriately so that the messages aren't enabled unless we
specifically ask for them (by setting clock_debug).
Define a constant for the SuperMac Spectrum/24 series III display card.
Thanks go to Luca Falzoni <falzoni@jetai.unipv.it> for trying out the
code for the Dayna ethernet support.
msgbuf. Note that old 'dmesg' and 'syslogd' binaries will continue running,
though old 'dmesg' binaries will output a few bytes of junk at the start of
the buffer, and will miss a few bytes at the end of the buffer.
setting up and enabling the MMU. This has the effect of ensuring
that any stray interrupts that get posted before the IPL is lowered
will be handled gracefully, i.e. without an access fault or illegal
instruction exception.
$Fssxxxxx 1MB slot space documented in the old Inside Macintosh, or the
$Fs0xxxxx 1MB slot space documented in the new IM and Guide to Macintosh
Family Hardware.
- Only set mac68k_vidphys if we're on a machine that we couldn't get
the Mac OS memory map.
- If we could get the Mac OS memory map, use the framebuffer size found
by get_mapping() rather than the size calculated using the Booter
environment.
(currently only CD-ROM drives on i386). The sys/dev/scsipi system provides 2
busses to which devices can attach (scsibus and atapibus). This needed to
change some include files and structure names in the low level scsi drivers.
- Add entries for the LC520, Color Classic, and PB150 to intvid_info[].
- Make the PB150 a Duo-class machine, as it seems to work better this way.
Now that we're doing the latter, it points out something I missed a long
time ago when I added Duo support: CLASSDUO was being ignored in
mac68k_set_io_offsets().
PowerBook 150 changes and bug fix provided by Dave Huang <khym@bga.com>
in PR 3970.
(address error faults), as well as other reported problems:
- Simplify grfmv_phys() to work like grfiv_phys(), and eliminate the
second argument to both, as we don't use it anyway.
- Handle fbbase and fboff consistently throughout. Closes PR 3862.
- Eliminate grfaddr() by pulling it into grfmmap(), which is the
only place it was used, previously.
- grfmap() now gets the physical address of the framebuffer from
the appropriate driver, rather than try to compute it by itself.
Be careful with aligning the base to a page address and increase
the length of the mapped region appropriately. Closes PR 2867.
by implementing entirely in assembly. This saves some 6 bytes on all MPUs,
and about 8 cycles on the 030.
Now that it's optimized, define spltty() in terms of _splraise() so that
we don't inadvertantly lower the SPL.
the pc532/dev/ncr.c-inspired changes; in particular, wait for previous
transfer to complete before starting another, as we used to do. Retain
splbio() protection and simplified sbc_pdma_in().
identical to the previous incarnation.
- Update using m68k asm.h macros
- Move initialization towards the front of the file
- Rename mac68k_buserr_addr to m68k_fault_addr
- Reorganize trap 15 handler, similar in structure to -- though not as
complete as -- the hp300 version
- Reorganize doboot() for easier integration of external cache, and
make room for the latter (#ifdef __notyet__)
- General garbage collection of unused code/data
Simplify the way transmit buffers are managed, remove assumptions about
NBPG, simplify sonic_get handling, update snioctl to be more like other
current drivers, and probably a few other changes I've now forgotten about.
our MD bus_space_probe() extension). This has several side effects:
- NuBus drivers must map and unmap slot space, rather than relying
on this happening before they are attached.
- Functions exported to NuBus drivers from nubus.c now need to be
supplied with a bus space tag/handle pair.
- Old bus map/peek functions can be garbage collected, as can pmap
support for the same.
Because of some current limitations of the bus space specification, we
are violating the abstraction in grf_mv (NuBus grf driver). All such
violations are clearly marked /* XXX */, and must be addressed when
the specification is updated.
surprisingly if the power off sequence doesn't work as expected,
and simplifies the code a little. Fix the doboot() prototype
to quash a warning from gcc.
- Remove unused `register' modifier on declarations
- Reverse the sense of some tests to make the code more clear
- Type casts per KNF
- Move a static variable into the only function that uses it (perhaps
we can eliminate it entirely, someday?)
Change in the way receive buffer areas are handled. Before we gave
the chip 16 buffers, each 1536 bytes (big enough for one packet).
Now we're handing the chip 8 buffers, each 4 Kbytes, and letting
the chip fit as many packets as it can in each one. This should
help keep it from running out of buffer space. Also make some of
the performance-crucial routines inline. It made no measurable
difference except to make me feel better
Changes from Bob Nestor <rnestor@metronet.com> to get closer to support
for his Apple SONIC-based nubus card.
Changes from me to try to get SONIC's MAC address from MacOS settings if
we can't read the PROM space.
driver. Major highlights:
- Protect with splbio(), now that interrupts from the chip might be
enabled by default;
- Wait for current transfer to complete before returning, rather than
wait for previous transfer to complete before starting another;
- Instead of logging a "timeout" message when a target disconnects,
just return, letting the sbc_ready() and sbc_wait_not_req() print
timeout messages if such really did occur.
With interrupt changes to sbc_obio.c, disconnect/reselect now works if
configured into the kernel.
present and initialize a few things. No color support, yet.
- Properly disable interrupts for DAFB.
- Rearrange for better nubus display card interrupt support. Only register
an interrupt if we know how to clear it.
- Complain if we don't know about a given display card and can not install
an interrupt handler.
- Change MYSTERY stuff to cb264--still need to get the DrSW so we can
actually call the routine to clear an interrupt from this card.
tweaks and bug fixes by yours truly.
Here's what Denny had to say:
nubus.h:
- add a drsw entry for my card
[ I also added a couple of constants for more display cards ]
if_sn.c:
- allocate sc->space using malloc (Scott wanted this).
- make csr accesses use NIC_PUT/GET, remove sc->sc_csr.
- add multicast and IFF_PROMISC support. multicast isn't
well tested, but I made sure the unicast case still
works.
- finish the new arp interface, remove sc->sc_enaddr
if_snreg.h:
- remove struct sonic_reg
if_snvar.h:
- add NIC_PUT and NIC_GET macros a la if_ae
- remove sc_txhead, it isn't used.
if_sn_nubus.c:
- clean up support for my SE/30 card.
- Move add_nubus_intr from if_sn.c (to keep if_sn.c MI)
if_sn_obio.c:
- make Q610, C610, Q650, C650, Q800 use EXBUS.
- PB500 had a "return" where "break" was intended. PB500
probably wasn't working.
- move add_nubus_intr from if_sn.c.
- add an "explanation" why Apple's ethernet addr is encoded
in token ring format in the PROM.
- Indent some comments appropriately per KNF.
- Output all diagnostics using the device name rather than a static string.
- Fix a (mostly marmless) bug in snintr(), where loss of heartbeat wasn't
properly ignored and could mask the `receive descriptors exhausted'
diagnostic.
- Use more consistent and portable types in the softc.
- Map registers using an array of bus_size_t offsets, and set up the
mapping in the attach code (thanks to Jason Thorpe for suggesting
this!).
- Disable the ae-specific watchdog, which is no longer necessary in
the general case.
Still remaining: split out functions used to copy data to/from the
card, and retain a way to have a local driver name with the MI code.