Harmonize USPACE / UPAGES across m68k platforms.
There seems to be a complicated history rooted in the original Utah code
using UPAGES of 2 w/ NBPG of 4096 for an 8KB total u-area. The Amiga
port then appeared using NBPG of 8192, but UPAGES was not adjusted to
conpensate, resulting in a 16KB u-area for Amiga. From there, chaos
spread which resulted in us homehow tolerating a situation where the
Sun2 is also using a 16KB u-area despite only have 2048-byte pages.
End this madness by centrally defining USPACE as 8192 and deriving
UPAGES from USPACE. 8KB is totally enough for these platforms.
Address problems with MIPS Malta platform code found running under QEMU.
QEMU's "malta" system emulates a MIPS Malta with the Gallileo host bridge
and 32-bit or 64-bit CPUs of either endianness. It is one of the only
working QEMU system-level emulations that could run NetBSD with all
combinations of endianness and address size. After fixes to QEMU over the
past several years, NetBSD has been unable to use the emulated PCI bus in
big-endian and 64-bit configurations.
No actual Malta hardware with any Gallileo-based CPU card could be found
for testing. These changes have been checked against the databook and
some limited checking of the relevant QEMU changes (which seem to have
mostly come from former MIPS employees) was also performed.
Changes:
1. The GT-64120 host bridge _does_ byte-swap access to other PCI targets,
but _does not_ byte-swap access to itself (bus 0, device 0). QEMU
evidently used to get this wrong, but, I confirmed with the databook.
This means we need to manually byte-swap a bunch of access to the
[29 lines not shown]
Fix MKDEBUGKERNEL vs MKDEBUG for kernel debug file sets.
The problem manifests as checkflist failures when building ports that
have extensive ALL_KERNELS but not...building all the kernels; notably
the various "evb" ports with a bazillion kernels for a bazillion SoCs.
The mk.conf(5) man page documents MKDEBUGKERNEL as controlling
whether kernel debug files (netbsd-*.debug) appear in the
distribution sets. However, the prior implementation used MKDEBUG
(the general userland debug flag) instead.
This meant MKDEBUG=yes with MKDEBUGKERNEL=no incorrectly expected
kernel debug files for every kernel config listed in ALL_KERNELS.
When only a subset of kernels is built, checkflist fails with
missing files.
The fix is to make these variables fully conform to the longstanding
documentation. MKDEBUGKERNEL controls whether kernel debug symbols are
built; MKDEBUG controls everything else. If you want something like the
old behavior but minus the bugs, set both.
A few things done while chasing down mod/ref bugs:
- pmap_remove_mapping() can now take pointer to the vm_page, saving a
lookup and allowing some additional assertions when it's available
(which is "frequently" in this implementation).
- All of the PTE load/store/modify-in-PT helpers now are decorated
with "volatile".
- Don't bother with atomic_load / atomic_store.
- Simplify pmap_testbit() and pmap_changebit().
- Add more PMAP_DEBUG-only mod/ref tests (including a test that validates
MMU beavior that was used to find a Qemu m68k emulator bug).
Import postfix 3.11.2 (previous was 3.10.1)
Changes in 3.11.2
Bugfix (defect introduced: Postfix 3.11): the proxymap(8) daemon
dereferenced an uninitialized pointer after a request protocol
error. This daemon is not exposed to local or remote users. Found
by Claude Opus 4.6.
Bugfix (defect introduced: 20260309) a change, to set the service_name
default value to "amnesiac", violated a test that parameter names
in postconf output must match 1:1 with parameter names in the
postlink script.
Changes in 3.11.1
Bugfix (defect introduced: 20260219): alias_maps errors when
default_database_type was not set in main.cf. Fix by Michael Tokarev.
[97 lines not shown]
Pull up following revision(s) (requested by mrg in ticket #282):
libexec/httpd/bozohttpd.c: revision 1.153
libexec/httpd/bozohttpd.c: revision 1.154
libexec/httpd/CHANGES: revision 1.58
libexec/httpd/tilde-luzah-bozo.c: revision 1.17
libexec/httpd/bozohttpd.8: revision 1.102
Fix double free of hr_file (later handled by bozo_clean_request())
Also hr_file_free should be used for free(3) (it fixes "//~user/" case
in which hr_file points to the last slash in the prefix).
Fix minor memory leak
bump version for last 2 memory fixes.
make.1 indent description of '-' impact in jobs-mode
Shift the '.El' so that the paragraph describing how jobs-mode handles
lines starting with '-' is clearly part of the description of '-'.
Pull up following revision(s) (requested by riastradh in ticket #1266):
sys/kern/vfs_subr.c: revision 1.503
Correct reversed sense of IMNT_MPSAFE test in VFS_MOUNT(9).
Pull up following revision(s) (requested by riastradh in ticket #281):
sys/kern/vfs_subr.c: revision 1.503
Correct reversed sense of IMNT_MPSAFE test in VFS_MOUNT(9).