the address of the desired PTE location is readily available in the
callers context (setpte4m() retraces the entire 3-level structure
to arrive at the PTE location).
Also, in many cases we can do away with the distinction between pmaps
that have or have not allocated a context. This is really only useful
in cases where we're interested in the REF or MOD bits which can differ
in the TLB version of a PTE. By doing this, we avoid getpte()'s which
in many cases instruct the MMU to start a table walk only to find out
that there's nothing there after going 2/3 of the way, or waste a TLB
entry because of TLB flushing soon after getpte() completes.
In addition, there's a hook to flush the cache line corresponding to
the (kernel virtual) location of a PTE entry when it gets altered.
- 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 machine/conf.h not sys/conf.h to get the prototypes for the
scn cdev functions.
* Include <sys/kgdb.h> to get prototypes for kgdb_connect and kgdb_attach.
* Add missing prototypes for some functions.
* Make ddb work before the console was opend. Thank's to Ian Dall
for this change. His comment:
Also, ddb doesn't work if a break is set before the console
is opened. That is because DTR and RTS are not asserted. It
would probably work if the console is wired to ingnore DTR
or RTS. The change below is probably not very correct.
Proabably the state of DTR and RTS should be saved and DTR
and RTS asserted on the start of a polled serial input and
the saved values restored on exiting polled input
* Remove unused function scn_ei.
* Add prototype for ncr_ready.
* Add ncr_wait_not_req function from Ian Dall:
The ncr_wait_not_req business is to avoid a potential race.
When the pseudo DMA finishes, the target may not have
lowered REQ yet. If we just charge ahead, we eventually test
for phase when REQ is high. However, if REQ has not yet gone
low for the last byte transferred, this will be the wrong
phase. This is taken from the dp8490 application notes. The
last ACK is not deasserted until the dma is completed.
Deasserting the last ACK should be delayed until the last
REQ is deasserted. I am not sure if there are ever devices
this slow, but I believe the code is "more correct".
* Add missing prototypes.
* Move over softnet() from intr.c.
* Establish softnet and softclock interrupts in init532.
* Misc cleanup to get it compiled with the new CWARNFLAGS.
of the ICU in clock_attach.
* Pull over (and reformat) rtc_rw from mem.c.
* Convert the rtc driver to use /sys/dev/clock_subr.c (partially by
stealing code from the sun3 port).
breaks zero-extension of the 16-bit memory size return from the BIOS.
on at least one system, the trash left in the upper half of %eax
causes the kernel to crash during VM system initialization when it
thinks it has on the order of 2**31 bytes of memory to play with.
We now *say* mov %eax,%ebx, but *mean* mov %ax,%bx..
- 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.
- cpu_startup() - mark pre-text pages VM_PROT_NONE and kernel text pages
VM_PROT_READ|VM_PROT_EXECUTE in kernel_map, to avoid strange things like
COW'ing the kernel text (from hp300 port)
- Fix dumpsys() to conform to the new dump bdevsw calling convention.
>Signed shifts are evil.
>Thanks to Michael Smith for reporting, Jason Thorpe for pointing to the
>report, doing a quick workaround which pointed me to the right code part and
>for testing the final fix.
