mmu_load_urp20hp(): No need to explicitly invalidate the external VAC
or the MMU's TLB -- user VAC entries and user TLB entries are automatically
invalidated when the USTP register is written, and there's no need to
invalidate supervisor entries in either of those when switching the user
address space.
pckbc(4) at acpi: Fix build with -Os.
Fixes:
/home/riastradh/netbsd/current/src/sys/dev/acpi/pckbc_acpi.c: In function 'pckbc_acpi_intr_establish':
/home/riastradh/netbsd/current/src/sys/dev/acpi/pckbc_acpi.c:242:63: error: 'handle' may be used uninitialized [-Werror=maybe-uninitialized]
242 | rv = acpi_intr_establish(sc->sc_dv, (uint64_t)(uintptr_t)handle,
| ^~~~~~~~~~~~~~~~~
/home/riastradh/netbsd/current/src/sys/dev/acpi/pckbc_acpi.c:224:21: note: 'handle' was declared here
224 | ACPI_HANDLE handle;
| ^~~~~~
No functional change intended -- the code wasn't _wrong_ before; this
just makes it clearer, with less analysis effort, that handle is
initialized in the path in question.
While here, rename local variable `rv' to `ih' to make it clearer
what it is (not just some arbitrary return value of a subroutine but
specifically an interrupt handler cookie).
[3 lines not shown]
Overhaul ATC / cache manipulation.
ATC + cache invalidation information is now gathered in a per-CPU
information struture that gathers up to 8 VAs in a batch. If more
than 8 VAs need to be invalidated, we just hit the entire ATC, trading
off the cost of the per-entry vs all-entries operations.
The same change allows us to batch cache invalidations, which reduces
the number of redundant cache invalidations.
Another nice property of these changes is that it collapses the HP MMU
handling into the same code paths as garden-variety 68851/68030, simplifying
the code flow in all sorts of places.
use ret from raw test case in verbose to see return value
and then tests can reach normalization tests for in flight packet testing.
more in flight packets tests for routing and possibly logging soon
gas: vax: Cherry-pick upstream commit 2722595890a
PR port-vax/59326
---
Author: Jan-Benedict Glaw <jbglaw at getslash.de>
Date: Sun Nov 15 18:36:27 2026 +0100
VAX: In PIC code, use GOT for function pointers
Use GOT references for function pointers in PIC code. Actual fix by
Rin Okuyama, initially reported for NetBSD in https://gnats.netbsd.org/59326 .
PR gas/32941
gas/
* config/tc-vax.c (md_estimate_size_before_relax): Use GOT instead of PLT.
* testsuite/gas/vax/elf-got-func.s,
* testsuite/gas/vax/elf-got-func.d: New test.
* testsuite/gas/vax/vax.exp: Run it.
A driver for the Maxim DS2413 1-Wire Dual Channel Addressable Switch,
that is, two open drain GPIO ports at the end of a 1-Wire bus.
Except for overdrive speed support, which is currently lacking in the
onewire(4) infrastructure, all aspects of the chip are supported. The
ports are presented as a gpio(4) bus and are operated on like any GPIO
bus would be with gpioctl.
There is an official family ID for the DS2413, but widely known clones
use an alternative. Both IDs are supported. It is also possible to
add another ID using a kernel config option, as there are some reports
of yet other family IDs out in the wild.
The chip works on the DS248x I2C to 1-Wire bridges and works on the
bitbanged gpioow driver. Tests were performed on a Virtualbox amd64
VM using umcpmio(4) as a USB to I2C bridge together with a DS2484 I2C
to 1-Wire bridge as well as directly on a RPI2 and RPI3 running
[7 lines not shown]
lint: fix integer constraints for division
In the expression 'a / b' with b > 0, lint wrongly assumed the result
would have the same bits cleared as a.
In the expression 'a % b', the resulting constraints were wrong for
negative b.
Introduce consumed flags by procedures in NPF.
Route procedures will almost always consume mbufs.
introcude a flag in the nbuf to let us know
that the nbuf has been consumed in case there is a
NULL check.
looked by martin
Pull up following revision(s) (requested by brad in ticket #582):
sys/dev/i2c/ds2482owreg.h: revision 1.2
sys/dev/onewire/onewire.c: revision 1.23
share/man/man4/ds2482ow.4: revision 1.2
sys/dev/i2c/ds2482owvar.h: revision 1.4
sys/dev/i2c/ds2482ow.c: revision 1.3
20 years ago when the 1-Wire infrastructure was ported over from
OpenBSD, the calls in the onewire_detach() function in the onewirebus
(sys/dev/onewire/onewire.c) code for config_detach_children() was
commented out. I don't know why this was done, exactly, as there
wasn't any particular comments present. I suspect it might have been
because owtemp, the common end device, could not detach or maybe
detaching just didn't work as expected at that time.
The string of devices are as follows:
End controller -> onewirebus -> End device
gpioow or ds2482ow -> -> such as owtemp or ds24e17iic
[28 lines not shown]
Pull up following revision(s) (requested by riastradh in ticket #580):
share/mk/bsd.own.mk: revision 1.1491
hppa: Enable fortify/ssp in userland for progs/libs that opt into it.
The `ssp' part doesn't actually do much (and really I'm not sure why
these options are mixed together) because hppa stack grows up, so
writing past the end of an array won't overwrite the return address,
so stack smashing in this way is extremely unlikely. But the
`fortify' part, which instruments functions like memcpy/read/getcwd
with bounds checks where statically known, is still useful to catch
buffer overruns.
ok `I guess' nick@
PR lib/60858: fortuitous embarrassment: fortify is all kinds of busted