Pull up following revision(s) (requested by jdc in ticket #456):
sys/dev/ic/pcf8584var.h: revision 1.7 (patch)
sys/arch/sparc64/dev/pcfiic_ebus.c: revision 1.10 (patch)
sys/dev/ic/pcf8584.c: revision 1.23 (patch)
When reading from an i2c device register, use repeat start on the bus.
This avoids the situation where another master gets the bus between our
register select (i.e. write) and the following read.
Add a delay after accessing our registers for the PIC implementation
of the PCF8584 (e.g. on the Sun Ultra 45).
Change swapregs to flags to handle multiple quirks and set the flags
appropriately in the Ebus frontend.
Pull up following revision(s) (requested by tls in ticket #455):
sys/uvm/uvm_pdpolicy_clockpro.c: revision 1.28
sys/uvm/uvm_pdpolicy_clockpro.c: revision 1.29
sys/uvm/uvm_pdpolicy_clockpro.c: revision 1.30
sys/uvm/uvm_pdpolicy_clock.c: revision 1.43
Fix a very brief race which could cause page states to be corrupted.
Observed under heavy physio load from an 8-way parallel dump|restore
workload used in filesystem testing - manifested as intents being
scribbled over while the interlock was momentarily not held
in uvmpdpol_selectvictim. I've attempted to apply the same fix to
clockpro, which looks like it needs it too, but clockpro was on fire
when I got there and I didn't put it out (still does not compile).
Fix an obvious typo which applied the wrong intent. Still does not compile.
fix build with clockpro
Pull up following revision(s) (requested by sborrill in ticket #454):
sys/compat/linux/common/linux_file.c: revision 1.135
Fix edge-case panic in copy_file_range when off_in or off_out are not NULL.
Found by Henrique Brito as part of GSoC 2024.
Patch to 2026d (previous was 2026c)
Changes to code
zic now rejects Link targets that would have invalid names, and
more efficiently processes Expires, Leap and Rule lines with years
far in the past or future. (Thanks to Darren Carreras.)
zic now ports to systems that report lack of link support via
EINVAL, ENOSYS or EPERM errno values. (Thanks to Tom Lane.)
When tzset and related functions encounter a TZif file that is too
large for them, they now consistently fail instead of sometimes
silently ignoring excess parts of the file.
localtime-related functions no longer mishandle extreme timestamps
when given TZif files holding some unlikely timezone histories.
(Problem reported by David Sarkisyan.)
[12 lines not shown]
PR/60655: vmt(4) probes the VMware backdoor unconditionally on aarch64
On ACPI systems only probe for vmt(4) when Hypervisor_Id in FADT reports
VMWare.
This helps avoid buggy hypervisors that don't implement a RAZ mdccsr_el0
Patch from Showta Ishizaki and slightly updated by me.
col(1): reword -h description
Negation is harder to read. And it was not immediately clear if the
terse "(default)" remark relates to the whole sentence (i.e. -h is the
default) or only to the negated part ("do not do the default $thing").
(It was not too bad, but remember that the primary audience of the
manpages is a desperate sysadmin at 3am, so every tiny bit counts).
Import tzdata2026d (as with 2026c) from a locally generated gtz version
rather than from https://github.com/JodaOrg/global-tz/releases/download/2026dgtz/tzdata2026dgtz.tar.gz
Eventually tzdata updates will no longer appear just a day or two
(or longer in the European vacation season) before a NetBSD release
is planned (11 for 2026c, 10.2 for this one) and this local generation
of a gtz version of the data will not be needed.
As with the 2026c update when the real gtz file appears, if there are
any material differences, an update will be made (2026c needed nothing).
Summary of changes in tzdata2026d (2026-09-11 15:21:07 -0700):
* Canada’s Northwest Territories will not fall back on 2026-11-01
and will stay on -06 year-round, matching Alberta’s recent change.
* Colombia’s 1992-05-02 spring forward was at 00:00, not 24:00.
* Iran’s 1979-05-26 spring forward was at 00:00, not 24:00.
* The backward-compatibility names EST5EDT, CST6CDT, MST7MDT, and
PST8PDT now conform better to POSIX.
* The temporary hacks used for North American timekeeping changes
[3 lines not shown]
Fix a brain-o in a KASSERT that could cause a guest to panic a DIAGNOSTIC
kernel: to assert that an I/O is already in flight we want
xbdi_pendingreqs > 0, not xbdi_pendingreqs > 1
While there, fix an error path that could also cause a KASSERT to fire
or a state corruption: reset xio_need_bounce to 0 before calling
xbdback_io_error() when the bounce buffer is not fully set up.
Pull up following revision(s) (requested by kre in ticket #1336):
tests/lib/libc/locale/t_mbstowcs.c: revision 1.4
tests/lib/libc/locale/t_mbstowcs.c: revision 1.5
tests/lib/libc/locale/t_mbrtowc.c: revision 1.3
tests/lib/libc/locale/t_mbrtowc.c: revision 1.4
tests/lib/libc/locale/t_mbrtowc.c: revision 1.5
t_mbrtowc: Mark UTF-8 test cases xfail.
mbrtowc previously failed to reject invalid (legacy 5/6-byte) UTF-8,
so it accepted this test case. Now it rejects this test case,
because the test case itself is broken.
Need to split this test up into:
1. correctly decoding the valid inputs
2. correctly rejecting the invalid inputs
But for now marking the test case xfail is an adequate approximation
[46 lines not shown]
Fix behaviour of rt_sigqueueinfo() when a thread group id (tgid) of -1 is
given. Previously this was passed as a process id to kill() which broadcasts
the signal (meanly you can accidentally kill all processes!).
Found by Henrique Brito as part of GSoC 2026
