mips: NMI handler must also clear STATUS.ERL
STATUS.ERL may be set in addition to STATUS.EXL and STATUS.NMI
when entering the NMI handler.
Clearing only EXL and NMI leaves the CPU in error level.
Reviewed by skrll@
mips: avoid redundant STATUS write in _splraise
The previous test only short-circuited when the requested IPL was
strictly less than the current IPL. When newipl == curipl,
short-circuit as well.
Reviewed by skrll@
Pull up following revision(s) (requested by martin in ticket #344):
usr.sbin/sysinst/Makefile.inc: revision 1.55
usr.sbin/sysinst/configmenu.c: revision 1.26
usr.sbin/sysinst/defs.h: revision 1.97
usr.sbin/sysinst/main.c: revision 1.38
usr.sbin/sysinst/menus.mi: revision 1.32-1.33
usr.sbin/sysinst/net.c: revision 1.48
Do not remove support for https neither at compile time (SMALLPROG)
nor at runtime (no SSL trust anchors available).
Instead dynamically disable the https option when it won't work, but
fully support it later (after unpacking the base set) when downloading
stuff for binary pkgs or pkgsrc.
--
#ifdef remove_menu_options() like all its callers to avoid stupid
compiler warnings.
debug/shl.mi: Cull obsolete linkable shlibs and explain why.
This is not a _good_ system but it's the system we have:
PR misc/57581: set lists should have a way to obsolete shlibs in
DESTDIR but not in postinstall
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
to the more complicated truth.
PR lib/60369: mbrtowc, mbrlen have wrong return value for some
invalid byte sequences: Invalid sequence
Old debug libraries get listed as obsolete so they get removed
So add back (but marked obsolete) libexpat.so.2.6.debug (even if it
was present for just a fortnight).
Do not remove support for https neither at compile time (SMALLPROG)
nor at runtime (no SSL trust anchors available).
Instead dynamically disable the https option when it won't work, but
fully support it later (after unpacking the base set) when downloading
stuff for binary pkgs or pkgsrc.
Fix return value for some invalid UTF-8 byte sequences
Update the length definitions to honor RFC 3629 (from 2003):
o The octet values C0, C1, F5 to FF never appear.
Add a check that continuation bytes are valid when adding them to the
internal buffer so that an out-of-bytes error doesn't override the
"invalid bytes" error.
Fixes PR 60369 by Bruno Haible.
Pull up the following revisions, requested by martin in ticket #337:
src/usr.sbin/sysinst/configmenu.c 1.25
src/usr.sbin/sysinst/install.c 1.26
src/usr.sbin/sysinst/net.c 1.47
src/usr.sbin/sysinst/upgrade.c 1.22
We have SSL trust anchors in base, so make use of them when downloading
additional items after basic installation is done (even if the installer
used does not come with its own certs, so we can not verify trust
for set download).
Pull up the following revisions, requested by martin in ticket #336:
src/usr.sbin/sysinst/msg.mi.de 1.59
src/usr.sbin/sysinst/msg.mi.en 1.60
src/usr.sbin/sysinst/msg.mi.es 1.50
src/usr.sbin/sysinst/msg.mi.fr 1.53
src/usr.sbin/sysinst/msg.mi.pl 1.56
src/usr.sbin/sysinst/util.c 1.84
Include "tests" and "manhtml" in the optional sets.
Pull up the following revisions, requested by martin in ticket #335:
src/usr.sbin/sysinst/Makefile.inc 1.54
src/usr.sbin/sysinst/configmenu.c 1.24
src/usr.sbin/sysinst/defs.h 1.96
src/usr.sbin/sysinst/main.c 1.37
src/usr.sbin/sysinst/menus.mi 1.31
src/usr.sbin/sysinst/net.c 1.46
When compiled with SMALLPROG (i.e. as part of a crunched ramdisk
userland) ftp(1) does not support https - so remove that option from
sysinst and only offer http in that case.
Should fix PR 60359.
PR bin/59635 - src/usr.bin/mail: fix post realloc() cleanup
This is a rather hackish solution, much better would be to abandon the
pointers altogether, and simply use message offsets (ints) into the array
to provide the relationships between messages.
Or abandon the message array (and the need for realloc() along with it)
and replace it with a list.
Both methods would achieve the aim of getting rid of the need to go and
massage the data to keep things correct when a realloc moves things around.
Either would require more changes in more places that this crude change,
and to get this done before -11 gets released, the few changes the better.
Another possibility would be to just revert to the adjustment method used
in -10 (which looks like it should work to me - but I don't know why it
was changed).
[4 lines not shown]
We have SSL trust anchors in base, so make use of them when downloading
additional items after basic installation is done (even if the installer
used does not come with its own certs, so we can not verify trust
for set download).
When compiled with SMALLPROG (i.e. as part of a crunched ramdisk userland)
ftp(1) does not support https - so remove that option from sysinst and
only offer http in that case.
Should fix PR 60359.
Pull up the following revisions, requested by martin in ticket #327:
src/usr.sbin/sysinst/util.c 1.81
src/usr.sbin/sysinst/arch/evbarm/md.c 1.24
src/usr.sbin/sysinst/arch/evbarm/md.h 1.10
PR 60360: make the default /boot partition on evbarm 64MB if we need
the dtb set (the newer dtb files overflow the old 32mb default).
In ACPI environments where we do not need dtb installed, do not select
the set by default.
acpi(4): Leave Hyper-V rndsource attached.
This way, it will be reported through rndctl(8) rather than only by
grepping through dmesg, and rndctl(8) can control it, e.g. if it
turns out something is wrong with that but not with RDSEED on the
same virtual CPU, you can use `rndctl -Ed "Hyper-V OEM0"' to discount
it and the system will gather fresh samples from RDSEED instead.
The same problem could in principle happen with other removable
rndsources like ualea(4), and we don't really have a good way to deal
with that -- either to configure the system up front to discount a
hot-pluggable rndsource that hasn't yet been plugged in, or to
retroactively discount a removable rndsource that was already
removed. But this one isn't hot-pluggable or removable, so keeping
it around is enough to sidestep this problem.
ld.elf_so(1): Run concurrent dlopen/dlclose test a few more seconds.
More likely to provoke the problem this way. Still not 100% reliable
because the problem is a race condition, but better than having the
test unexpectedly pass half the time.
Also set a timeout of 20sec, since I've seen the test get into an
infinite loop sometimes and it's now supposed to complete in 5sec +
epsilon.
PR lib/59751: dlclose is not MT-safe depending on the libraries unloaded
