dev/ic/wd33c93.c. This may not be the best WD33C93 driver we've
got, but it's the most recently worked on and probably the most
portable, so it seems like a good basis for further work (and in
particular an acorn26 driver for the Acorn SCSI card). There's
no functional change in this commit, and sgimips kernels still
compile.
increase the number of retries by two orders of magnitude wont affect
most systems but will make transitions smoother on marginal ones.
From gwk@openbsd.
Modernise the linker script and make sure we have a symbol after the
link_sets.
Make sure that pmap_bootstrap doesn't tell uvm that some of the kernel
phys pages are free. Previously uvm would very likely allocate the phys
pages used for the link_sets to a MALLOC and they'd get overwritten.
Thanks to David H. Gutteridge for testing various things.
History: This cleans up another long standing inconsistency where timerblurb for delay() was
calibrated against the stat timer instead the clock timer. Before timecounters
this didn't matter at al as both timer run with the same frequency and the
calibration was a frequency based calibration.
Misbehaving clock found by Fanch (at) kekpar dot net on current-users@ - Thanks.
Fix verified to work by Fanch.
x86/include/powernow.h
- Protect k[78]_powernow_init() functions with #ifdef POWERNOW_K[78] to
make it build without these options.
This fixes the problem reported by hubertf.
(1) failsafe.c allows character input from the failsafe
console device. This is the early console device that is used for output before
the pmap layer is initialized.
(1) locore.S: make sure that the size of the cacheline for
dcbz is set to the default of 128 bytes. It also sets up other mode bits on
the 970. This fix allows the G5 to boot into full multi user mode using a
NFS mounted root file system.
---
# XXX Disable zero-copy page loaning in sosend() temporarily:
# PV mappings created by sosend_loan() in sys/kern/uipc_socket.c may
# produce virtual cache aliases and it seems to cause TLB MISS panic
# in cache flush code called from bus_dmamap_sync(9) PREWRITE ops
# during heavy TX packet traffic on tlp(4).
an Amiga DMA controller, and #if 0'ed WD33C93 definitions that are duplicated
in sbicreg.h. uPD71071 definitions can stay for now, since they're not
actually useless even though they're unused.
the nearest one, rather than the first one with the same resolution. This
is likely to be useful when the bootloader finally passes a valid frame rate.
For now, it just favours flickery over non-functional.
we use the standard mode list from videomode.c rather than one generated
by makemodes.awk. This is less useful than it might be since without a useful
frame rate from the bootloader we're likely to end up choosing a screen mode
that's either flickery or too fast for the monitor (or DRAM bandwidth).
used for anything. Rearrange things so that it doesn't any more, and
just produces an array of struct videomode. modedefs.c is looking
suspiciously much like videomode.c now.
the kernel, removing an element from struct vidc_mode. The calculation
is a bit pointless at the moment anyway, since both bootloaders pass in
a constant 56, but that might get fixed one day.
struct videomode doesn't have a way to record border widths (though I
wonder if VESA's "margins" are the same thing), so we now just treat them
as zero. Tested on my NC, which still seems to be working.
Setup cache bits in HID5 early on so that dcbz/icbi use the
128 byte cache line size. The mismatch was causing dynamically linked
apps to randomly coredump.
With this fix, the G5 can boot into full multi-user by using a NFS
mounted root file system.
to RiscBSD and had been lurking in dark corners scaring people (mostly me)
for far too long. It will be missed, but not much, and I hope we can clear
up any fallout before 5.0.
While I'm in the area, also remove rpckbd(4), since pckbd(4) has
been supported on acorn32 for ages.
root partition, instead of punting. This makes booting work
with traditional disklabel disks and wedge autoconfiguration.
- factor out disk opening code.
tsutsui