Aaron LI [Sat, 9 Jun 2018 08:45:48 +0000 (16:45 +0800)]
initrd: don't leave an empty /rescue after switching location
The recent initrd change will remove the rescue tools from their old
location (/usr/share/initrd) but leave an empty /rescue directory. Avoid
such a situation by building the rescue tools after installworld if /rescue
is empty.
Suggested-by: François Tigeot
Sascha Wildner [Sat, 9 Jun 2018 08:13:36 +0000 (10:13 +0200)]
libc/xdr: Add xdr_uint16_t() (it's just like xdr_u_int16_t()).
dports might make use of it.
While here, improve the manual page a bit and also bring in some fixes
from FreeBSD.
Sascha Wildner [Sat, 9 Jun 2018 07:45:26 +0000 (09:45 +0200)]
libc/xdr: Use standard types.
Sascha Wildner [Fri, 8 Jun 2018 21:40:18 +0000 (23:40 +0200)]
rcs: Clean up the Makefiles.
* Close the Makefile.inc chain to gnu/usr.bin/Makefile.inc.
* Use the SCRIPTS variable instead of using afterinstall.
Sepherosa Ziehau [Tue, 5 Jun 2018 15:03:46 +0000 (23:03 +0800)]
x86_64: Implement x2apic support.
Now LAPIC registers are accessed through MSR at fixed location, instead
of going through MMIO region.
Most noticeable is that ICR operation is greatly simplified, i.e. IPI
sending operation:
- Reserved bits are read as 0; there is no need to read ICR first for
OR with the new values.
- No more pending bit, i.e. ICR write is synchronized; there is no need
to read ICR to test pending bit.
- ICR is 64 bits in x2apic mode, i.e. two 32 bits writes to ICR-low and
ICR-high become one write to ICR.
NOTE:
Though Intel SDM says that wrmsr to LAPIC registers are relaxed, we
don't need to put mfence or sfence before them, especially for sending
IPIs, since the generic IPIQ and the machdep code already uses atomic
operation before doing ICR operation. For the rest of the code, there
really are no needs to add mfence/sfence before rdmsr/wrmsr to LAPIC
registers.
As of this commit, x2apic mode is _not_ enabled by default. It can be
enabled through hw.x2apic_enable tuneable, and a read-only sysctl node
with the same name is available for debugging purpose.
Based on work by ivadasz@.
Aaron LI [Fri, 8 Jun 2018 07:58:47 +0000 (15:58 +0800)]
nrelease/Makefile: Fix to use GITURL_{SRC,DPORTS} and update help
The GITHOST and GITHOST_DPORTS were updated to be GITURL_SRC and GITURL_DPORTS
in
ad8b70e08d7600e3f4eb3b36556d4540b1a9c119.
Aaron LI [Sat, 2 Jun 2018 05:31:45 +0000 (13:31 +0800)]
initrd: Detach from the world and build on the fly
Why
---
Currently, the files for initrd image creation are built and installed at
/usr/share/initrd by "buildworld" and "installworld", respectively, and then
mkinitrd(8) simply packs them to create the initrd.img.gz.
For normal users, the shipped /usr/share/initrd and mkinitrd(8) don't make
much sense, since they don't need to modify the initrd contents and create
a modified initrd image.
From a developer's perspective, the required steps (buildworld, installworld
and mkinitrd) to create and test a new initrd image can be annoying.
How
---
Detach the build and installation of the initrd contents from the world.
Build the (rescue/initrd) tools on-the-fly when creating the initrd image.
As per ftigeot's suggestion, these statically linked tools can be useful when
dealing with a broken system, so install them under /rescue to be more
intuitive and easier to use (similar to FreeBSD and maybe other BSDs).
What
----
* Move "share/initrd" to be top-level, and detach it from world.
+ Update the patch in bsd.crunchgen.mk accordingly.
+ Add Makefile.inc to simplify subdir's makefiles.
+ Rewrite the Makefile with targets:
- rescue: build and install the rescue tools into /rescue.
- initrd: further packs the etc and rescue staffs to create the initrd
image.
* Move "sbin/mkinitrd/mkinitrd.sh" under the new "initrd" directory and adapt
it to work with the "initrd" make target. Remove the other parts of
mkinitrd(8).
* Export the above "rescue" and "initrd" targets to be top-level.
* Update nrelease framework to use the new "make initrd".
* Update build.7 man page accordingly.
* Print the message about updating the initrd image and 3rd-party packages
after "make upgrade" instead of "make installworld".
* Document this change in UPDATING.
* Other small updates and style cleanups.
TODO
----
* Add initrd.7 man page
* Crunch more useful tools
* Crunch bin/sbin/etc. into a bundle to save space (??)
Reviewed-by: swildner
Thanks-to: dillon, swildner, ftigeot
Aaron LI [Sat, 2 Jun 2018 05:09:19 +0000 (13:09 +0800)]
Makefile{,.inc1}: Merge user-driven target descriptions and cleanups
* Merge the descriptions to user-driven targets from Makefile.inc1 to
Makefile and improve the wordings a bit.
* Delete some obsolete information.
* Whitespace cleanups.
Aaron LI [Fri, 1 Jun 2018 07:48:13 +0000 (15:48 +0800)]
mkinitrd(8): Use getopts instead of getopt(1) for command options
Use the shell built-in getopts to deal with command line options, which is
more easier to use and allows whitespace in the option values.
Aaron LI [Fri, 1 Jun 2018 07:25:40 +0000 (15:25 +0800)]
initrd: Clean up the real root mount code
Make the real root mount code/scripts much simpler. Also use the newly
exported NEW_ROOT variable to be more maintainable.
Aaron LI [Fri, 1 Jun 2018 07:14:27 +0000 (15:14 +0800)]
initrd/rc: mount initrd ro; set PATH, HOME, NEW_ROOT; cleanups
* The initrd is mounted readonly by default, do not upgrade to be readwrite.
NOTE: This may cause problem with diskless boot (/etc/rc.d/diskless) and
needs further tests and tunes (Thanks to ftigeot).
* Set PATH=/bin:/sbin to simplify the command invocations.
* Set HOME=/var/home (Suggested-by: ftigeot)
* Set NEW_ROOT=/new_root that will be used in the mount scripts.
Make sure that the NEW_ROOT directory exist, otherwise try to remount the
initrd / readwrite and create it.
* Various cleanups to make it much simpler and cleaner.
* Also clean up the rc.lvm2 a bit.
Aaron LI [Fri, 1 Jun 2018 06:57:37 +0000 (14:57 +0800)]
initrd: symlink tmp to var/tmp to save a second tmpfs mount
Since a tmpfs is required to mount at /var in the initrd, symlink /tmp to
/var/tmp to save a second tmpfs mount.
Add /mnt to the initrd hierarchy, which may be used by the user to manually
mount filesystems. (Suggested-by: zrj)
Aaron LI [Fri, 1 Jun 2018 06:08:57 +0000 (14:08 +0800)]
initrd: Make etc/motd more informative
Provide more information about the rescue shell environment, such as how to
mount the real root partition and continue, the mined editor, and more on
network connections.
Aaron LI [Fri, 1 Jun 2018 05:06:22 +0000 (13:06 +0800)]
mk/crunchgen: Fix the ${MAKE} quotation and improve style
The fix to the ${MAKE} quotation is brought from FreeBSD.
Various style fixes to make this makefile look much cleaner.
Aaron LI [Fri, 1 Jun 2018 04:56:05 +0000 (12:56 +0800)]
periodic: return $rc instead of "$rc" for several scripts
Also fix a minor whitespace issue in 200.chkmounts .
Aaron LI [Fri, 1 Jun 2018 04:49:41 +0000 (12:49 +0800)]
etc/crontab: Do not set HOME; remove /etc from PATH
Do not set the HOME environment variable (checked with FreeBSD).
Remove "/etc" from PATH (checked with OpenBSD).
Aaron LI [Fri, 1 Jun 2018 04:47:58 +0000 (12:47 +0800)]
README: Update descriptions to sys/config files
Aaron LI [Fri, 1 Jun 2018 04:46:21 +0000 (12:46 +0800)]
README: Remove the secure entry; add the doc entry.
Those contents resided under secure were merged into crypto some time ago.
Aaron LI [Fri, 1 Jun 2018 04:19:07 +0000 (12:19 +0800)]
etc/Makefile: Do not install rc.firewall as executable
/etc/rc.firewall doesn't need to be installed as executable, similar to
/etc/rc.firewall6 .
Aaron LI [Fri, 1 Jun 2018 04:09:59 +0000 (12:09 +0800)]
hammer2: do not print error=0x0000 on bulkfree
Make hammer2 bulkfree do not print "error=0x0000" (i.e., no error occurred).
This makes the system log messages (/var/log/messages) a bit cleaner.
Fix minor whitespace issues as well.
François Tigeot [Wed, 6 Jun 2018 20:37:43 +0000 (22:37 +0200)]
drm: rman_get_start() returns u_longs, not ints
This had not been correctly handled in
183e2373896e4ea605435a6bd8f943e8273bf8cd
Matthew Dillon [Wed, 6 Jun 2018 15:34:43 +0000 (08:34 -0700)]
cpdup - Fix NULL pointer indirection on lost socket
* Fix a crash which can occur due to a lost socket. Generate a
nice error message and exit() instead.
Matthew Dillon [Wed, 6 Jun 2018 15:29:37 +0000 (08:29 -0700)]
kernel - Fix umtx (pthread_mutex etc) race in kernel
* The umtx code could still race a fork despite the wakeup
due to certain access paths leading into umtx_sleep() only
issuing reads, thus not triggering a COW operation post-fork.
This results in a loss of the interlock.
* Fix the issue by having umtx_sleep() fetch the user data
with a read-modify-write instead of a read. It now essentially
does a lock xadd, adding $0 to the memory to force a write
fault if necessary.
* Fixes 'go' tests and probably other situations that can arise
when heavily threaded programs use fork/exec.
Reported-by: Tim Darby (+testing)
Matthew Dillon [Wed, 6 Jun 2018 15:28:49 +0000 (08:28 -0700)]
kernel - Add fuwordadd32(), fuwordadd64()
* Add locked-bus-cycle fetchadd equivalents for kernel access to
userland. Will be used by kern_umtx.c
Sepherosa Ziehau [Sun, 3 Jun 2018 10:15:52 +0000 (18:15 +0800)]
x86_64: Prepare for x2apic support.
- Use macro to access and modify LAPIC registers.
- Use function pointers for hot LAPIC operation, i.e. IPI and timer.
- Refactor the related code a bit.
Global variable 'lapic' is renamed to 'lapic_mem' to ease code search.
Based on work by ivadasz@.
Sascha Wildner [Tue, 5 Jun 2018 07:25:23 +0000 (09:25 +0200)]
mtree: Remove a no longer used directory.
Sascha Wildner [Mon, 4 Jun 2018 07:27:42 +0000 (09:27 +0200)]
Resolve some kernel option / .Cd mismatches.
* There is no WATCHDOG option. kern_wdog.c is compiled into the kernel
by default.
* SC_NO_PALETTE_LOADING was never a kernel option, so remove the
associated #ifndefs.
* powernow(4) can't be compiled into the kernel
Sascha Wildner [Sun, 3 Jun 2018 10:26:44 +0000 (12:26 +0200)]
init/oinit: Remove some old, unused/obsolete paths and comments.
Seems the _PATH_SLOGGER was defined but not used since at least 4.4BSD.
Reported-by: aly
Sascha Wildner [Sat, 2 Jun 2018 20:05:10 +0000 (22:05 +0200)]
acpica: Allow designated initialization by defining ACPI_STRUCT_INIT().
To make it compile, bring in a small fix from ACPICA's master that
adjusts a number of names from Linux style to ACPICA style.
Sascha Wildner [Sat, 2 Jun 2018 19:12:38 +0000 (21:12 +0200)]
Sync ACPICA with Intel's version
20180531.
* Implemented additional support to help ensure that a DSDT or SSDT is
fully loaded even if errors are incurred during the load.
* The 'Unload' AML operator is no longer supported.
* Additional improvements and fixes.
For detailed list, please see sys/contrib/dev/acpica/changes.txt.
Sascha Wildner [Fri, 1 Jun 2018 10:04:25 +0000 (12:04 +0200)]
Remove some old, now obsolete, examples.
Sascha Wildner [Wed, 30 May 2018 19:09:33 +0000 (21:09 +0200)]
sh(1): Exclude a file that is only needed for debugging.
Reported-by: zrj
Sascha Wildner [Wed, 30 May 2018 13:49:59 +0000 (15:49 +0200)]
ieee80211_crypto.9: Ouch, removed too much in
a42bad2dd166ad.
Sascha Wildner [Wed, 30 May 2018 13:46:02 +0000 (15:46 +0200)]
Fix some it's -> its typos in our manual pages.
Also some minor fixes in rc.conf.5.
Sascha Wildner [Wed, 30 May 2018 05:50:41 +0000 (07:50 +0200)]
Define ioctl names that are used in *_ioctl(), not alternative spellings.
This makes searching unimplemented ioctls easier.
While here, fix cdcontrol(8) to use the "main" ioctl name.
Sepherosa Ziehau [Tue, 29 May 2018 13:42:15 +0000 (21:42 +0800)]
acpi: Use X2APIC entries, if there are only X2APIC entries in MADT.
zrj [Mon, 28 May 2018 15:07:31 +0000 (18:07 +0300)]
kernel: Add simpler make depend variant for kernel.
For now only for make {build,native}kernel -DNO_MODULES. The modules part
is more involved and will follow next.
Idea is to avoid a seprate mkdep(1) run before the actual compilation by
simply writing out separate dependency files while compiling in one go.
This variant reuses depenency information from last compilation, while
mkdep(1) variant reconstructs the dep chains on the current tree.
Depending if headers where removed or added both variants can miss out.
Due to several constraints (checks in makefiles if .depend file already
exists and implicit handling of device firmware stub files) try to mimic
previous behaviour by reconstructing depend on next quickkernel.
Later on we could remove anchoring on hack.So, creation of .depend file
and include all foo.d dependency files directly using bmake's .dinclude
directive that acts on stale dependencies as .MAKE.DEPENDFILE .depend.
Also turns out mkdep(1) was not tracking deps for ${MFILES:S/.m$/.c/}},
this simpler variant catches them and only vmx_genassym.d is not hooked
in .depend (coming from sys/platform/pc64/conf/files).
Shaves ~30s for make -jN nativekernel -DNO_MODULES.
zrj [Mon, 28 May 2018 14:51:38 +0000 (17:51 +0300)]
kernel/hpt*: Mangle names of foreign binary objects.
We already do it for the firmware objects (.fwo). Explictly handle
unpacked precompiled objects to separate them from other generic ones.
These are not device firmware and they do run on CPU, also they have
different osabi tag, so without adjustment the ld.gold (default ld)
will change osabi for the whole linked object.
Having them seprated from the rest simplifies the depend logic, such
objects have no source or headers dependencies other than hardcoded
one in Makefiles. Add missing dependencies on .uu in kernel modules.
zrj [Mon, 28 May 2018 14:34:59 +0000 (17:34 +0300)]
build: Remove use of unsupported flag.
It looks like a inherited feature that was used in patched GCC.
zrj [Mon, 28 May 2018 14:24:32 +0000 (17:24 +0300)]
kernel: Normalize conf/files rules.
Prefer to use config(8) variables to reduce duplication
In the recipies for atomic.c we override global -fno-frame-pointer only
if not profiling (at the moment unusable .mcount), add missing ${WERROR}.
For vkernel64/mp.c just swap additions for consistency.
We could merge the ${WERROR} directly into ${NORMAL_C} later on.
Matthew Dillon [Tue, 29 May 2018 03:10:01 +0000 (20:10 -0700)]
kernel - Add support for MosChip PCIe serial com, and console support (2)
* Fix oops in last commit.
Matthew Dillon [Tue, 29 May 2018 02:58:03 +0000 (19:58 -0700)]
kernel - Add support for MosChip PCIe serial com, and console support
* Add support for the MosChip PCIe serial com (brand is typically
'StarTech' 'Natikve PCI Express RS232 serial adapter card with
16550 UART'.
* Add late console support. Console operation works the same
way, set it up in /boot/loader.conf. Typically:
sio0.flags=0x00
sio4.flags=0x30
sio4.baud=115200
#set boot_serial="YES" <-- commented out, do not set this
Note that at the point a late-configured serial port
wants to become the console, syscons is probably already
the console, so you have to use flags 0x30 instead of
0x10 to force it to change the console.
* Note, you do not have to set boot_serial="YES", because it
won't really help since the adapter isn't accessible to the
boot loader or to the kernel until the PCIe device probes
in the middle of the boot sequence.
* The kernel now retests for consoles at the end of its normal
configuration sequence, which allows the now-available
sioN devices from PCIe card(s) to become the console.
* Remove syscons sccnterm() code that tried to undo the syscons
console designation. This code was never called in the past
because the system console couldn't actually change once
selected. But now it can, and the code completely blows up
syscons for numerous reasons. Just leave the console
designation intact from syscons's point of view (the main
system will be vectored elsewhere so syscons will simply
not receive new kernel console reads and writes).
* Add a siocntxwait() call before changing com parameters,
since we are potentially changing the baud rate here.
* Change how the SIO driver assigns baud rates for the console
to handle late-console configurations. Also, allow sio*.baud
variables in loader.conf to override the default com rate.
* Add entries for com5 and com6 to /etc/remote, and default to
a more reasonable baud rate of 115200.
zrj [Mon, 28 May 2018 05:41:08 +0000 (08:41 +0300)]
Fix missing make depend for quickkernel.
Only realquickkernel should skip it.
Matthew Dillon [Sun, 27 May 2018 15:45:28 +0000 (08:45 -0700)]
kernel - Fix rare pmap_scan_callback() panic
* Fix a panic where pmap_scan_callback() can sometimes call
pv_put() with a NULL pd_pv.
Matthew Dillon [Sun, 27 May 2018 15:31:06 +0000 (08:31 -0700)]
hammer2 - Fix missing initialization (fix broken root mount)
* For hammer2 root mounts a missing initialization resulted in
a garbage test, causing root mounts to fail under some
circumstances.
* Properly initialize the 'done' variable.
Matthew Dillon [Sun, 27 May 2018 15:26:21 +0000 (08:26 -0700)]
hammer2 - Be a be more verbose in the mount and getroot path
* kprintf more info when a hammer2 mount fails with EINVAL
Sascha Wildner [Sat, 26 May 2018 21:41:15 +0000 (23:41 +0200)]
libipfw3: Clean up the Makefiles a bit.
Sascha Wildner [Sat, 26 May 2018 16:41:33 +0000 (18:41 +0200)]
Makefile.inc1: Do not continue the build upon an error in various cases.
Before this commit, the build would limp on even though a library
failed building, for example, leading to sometimes confusing followup
errors later in the build. It was also more difficult to see directly
why a build really failed in the first place.
To prevent this, make various targets use '&&' instead of ';' to
chain commands.
People wanting the build to carry on for as long as possible, can
always use make's -k option.
Matthew Dillon [Fri, 25 May 2018 02:22:52 +0000 (19:22 -0700)]
hammer2 - Fix focus vs modify race
* Fixes rare panics when e.g. removing large numbers (as in hundreds of
millions) of directory entries.
The XOP collection code holds the collected cluster but cannot
safely lock it without risking a deadlock against backend operations
or dead backends. The hold on the chain prevents its destruction,
but does not prevent another thread from locking it and issuing
hammer2_chain_modify().
* Fix bugs due to the unsafe nature of an unlocked chain's content,
especially chain->data and chain->dio, by adding an interlock between
frontend access to the data and backend hammer2_chain_modify() calls.
Held but unlocked chains are used by the XOP API to pass chains back to
the frontend.
* Remove the automatic (because it is unsafe) dio synchronization
in hammer2_xop_collect() and instead implement an API that the
frontend can use to safely access the data. The API is
hammer2_xop_gdata()/hammer2_xop_pdata().
* Remove the unsafe hammer2_cluster_rdata(). Use gdata/pdata for this
too.
* Rewire hammer2_inode_get() to pass-in a hammer2_xop_head instead
of a hammer2_cluster so it can use the gdata/pdata API too.
Matthew Dillon [Thu, 24 May 2018 23:48:31 +0000 (16:48 -0700)]
hammer2 - remove hammer2_cluster_wdata()
* Remove hammer2_cluster_wdata(), function is not applicable for
context any more since the focus chain is held, but no longer
locked.
zrj [Sat, 26 May 2018 12:15:11 +0000 (15:15 +0300)]
ncurses: Add few missing dependencies.
The lib_gen.c is generated source and it's dependencies too.
This should solve rare races where lib_gen.c is partially constructed,
even if it is not fatal problem (only XSI req for callable functions).
François Tigeot [Sat, 26 May 2018 11:32:07 +0000 (13:32 +0200)]
drm/i915: Sync hpd_pulse return values with Linux
Now that we know how to handle irqreturn enums, we can use more
Linux code unmodified.
zrj [Sat, 26 May 2018 10:02:04 +0000 (13:02 +0300)]
Makefile.inc1: Reduce strict ordering where not needed.
Helps with parallel builds. Shaves off about 40s from make buildworld.
zrj [Sat, 26 May 2018 09:54:22 +0000 (12:54 +0300)]
Makefile.inc1: Update _prebuild_libs dependencies.
The lib/libdevattr is not strictly needed in _prebuild_libs, because
it has dependency on libprop only if libdevattr/mktest is hooked.
zrj [Sat, 26 May 2018 09:44:57 +0000 (12:44 +0300)]
Makefile.inc1: Reduce strict ordering for libc.
Mark lib/csu as a hard dep for lib/libc (needed only for libc.so link).
Even if lib/libc_rtld is not strictly needed to be handled at all in
Makefile.inc1 (its _pic.a archive only used by libexec/rtld-elf),
having it here is not a big issue since both libc and libc_rtld spend a
lot of time in make depend and running those in parallel helps somewhat.
zrj [Sat, 26 May 2018 06:56:08 +0000 (09:56 +0300)]
strip(1): Perform safe install.
Pass -S flag to install(1) as it is currently done for /bin/sh.
The install(1) might call strip(1) on itself when installing.
zrj [Fri, 25 May 2018 07:50:12 +0000 (10:50 +0300)]
buildworld: Add support to skip toolchain if requested.
These two new features are mainly aimed at developers, to speed up their
rebuilds on development branches. Often there are no changes that would
influence the toolchain components in any way. So it is reasonable to
skip rebuild of certain parts (mostly to avoid building c++ sources).
Previously we were rebuilding both main and alternative compilers twice
during buildworld for a full bootstrap that allowed us to avoid certain
breakages and not to keep adding lots of backwards compatibility hacks.
The new additions:
* -DNO_CROSSTOOLS - skip compiler and binutils bootstrap;
* -DNO_TOOLCHAINS - do not build compilers/binutils for installworld.
Both features are aimed for testing DESTDIR targets. Decision to
activate any of these features is left to the user's own discretion
based on reviewing the commits to the git repository or local clone.
The NO_CROSSTOOLS feature is implemented by installing the shell script
wrappers that add a sysroot and patch the specs. Idea is not to disturb
in any way the Makefile.inc1 or objformat logic at this moment. Since
all the main toolchain tools are installed as shell scripts this has a
nice bonus for quick hooks/probes and other hacky/pretty stuff without
disturbing the rest of the buildworld. You can even add your favorite
static analyser tool, just don't forget to change default include path
to /usr/obj/../world_x86_64/usr/include for it.
The objformat changes for sysroot would only make sense when we make a
port for another architecture. None of the new features are considered
to be guessTM enabled automatically at this moment. To be sure that
nothing has changed since previous buildworld from other changes in the
source tree one would have to compile and hex-diff the final products.
Keep in mind that libraries coming with the compilers will be rebuilt
even with both knobs passed. This should help catching most of the
issues before installworld step and not on the next {build,quick}world.
Both knobs are not documented in make.conf(5) on purpose. It is highly
advised *not* adding any of these knobs into make.conf, sooner or later
you will forget it and *will* get yourself in world recovery situation:
a) world backup machinery (unless you had NO_BACKUP=true in make.conf);
b) perform manual bootstrap to fix up the breakage;
c) restore from H1/H2 snapshots (toolchain components are located in
/usr/libexec/* and /usr/lib/${CCVER}/ directories);
d) clean overwrite by mounting from other system (like usb img stick).
Features should work with quickworld too.
To put things in perspective here is a little chart of the cleaned
fresh 'make -jX buildworld' times impact on few setups:
netbook: hw.model: Intel(R) Core(TM) i3 CPU U380 @ 1.33GHz
compiler: gcc50 as cc(1) (4gb usb stick for /, ironlake cpu)
make -j3 buildworld [-DNO_CROSSTOOLS] [-DNO_TOOLCHAIN]
min:sec | STANDARD | NO_C | NO_T | NO_C/NO_T |
----------------------------------------------------------------
gcc50 | 73:56 | 59:57 | 54:11 | 39:38 |
gcc47 (host+world)| N/I | N/I | N/I | 37:51 |
----------------------------------------------------------------
laptop setup: hw.model: Intel(R) Core(TM) i7-4712MQ CPU @ 2.30GHz
compiler: gcc50 as cc(1) (default)
make -j9 buildworld [-DNO_CROSSTOOLS] [-DNO_TOOLCHAIN]
min:sec | STANDARD | NO_C | NO_T | NO_C/NO_T |
----------------------------------------------------------------
gcc50 | 17:06 | 13:57 | 12:58 | 09:42 |
gcc50 gcc47 (def) | 20:18 | 16:03 | 14:58 | 10:37 |
gcc50 gcc80 | 24:24 | 18:06 | 17:31 | 11:23 |
gcc50 gcc47 gcc80 | 27:47 | 20:17 | 19:23 | 12:00 |
----------------------------------------------------------------
blade setup: dual hw.model: Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz
compiler: gcc80 as cc(1) (soon gcc80 will replace gcc50 in master)
make -j32 buildworld [-DNO_CROSSTOOLS] [-DNO_TOOLCHAIN]
min:sec | STANDARD | NO_C | NO_T | NO_C/NO_T |
----------------------------------------------------------------
gcc80 | 06:58 | 05:19 | 05:33 | 03:51 |
gcc80 gcc47 (def*)| 08:03 | 06:06 | 06:02 | 04:05 |
gcc80 gcc50 | 08:23 | 06:12 | 06:19 | 04:08 |
gcc80 gcc47 gcc50 | 09:24 | 06:59 | 06:51 | 04:21 |
----------------------------------------------------------------
zrj [Thu, 24 May 2018 12:35:30 +0000 (15:35 +0300)]
gcc80: Shorten build time in libraries target.
When compiling compiler libraries build the minimal set of tools
required to generated the needed headers. Rest of the tools including
libcpp.na will be built during more parallel gnu/usr.bin stage.
Shaves off ~30s from make buildworld.
While there, do the same for gcc47 and gcc50. Also clean stale file.
zrj [Thu, 24 May 2018 12:33:03 +0000 (15:33 +0300)]
Makefile.inc1: Add libelf to _prebuilt_libs.
Even if it is internallib for libexecinfo it is very close to race in
_generic_libs.
While there, add ordering for liblzma against libpthread.
zrj [Thu, 24 May 2018 12:16:00 +0000 (15:16 +0300)]
tools/make_libdeps.sh: Use sys.mk from src/.
While there, remove not needed /bin from PATH.
Suggested-by: swildner
Sascha Wildner [Thu, 24 May 2018 12:23:51 +0000 (14:23 +0200)]
Remove some duplicated includes.
zrj [Thu, 24 May 2018 09:21:09 +0000 (12:21 +0300)]
tools/make_libdeps.sh: Add needed mangling rules.
For tool to work properly:
* suppess all of the -Wl,--foo (yes, we have those unfortunately);
* mangle all of pthread references;
* ignore libstdc++'s ../components/blah/libfoo_pic.a (used as int libs);
* remangle internallib libelf_pic.a for libexecinfo;
* mangle all of the privatelibs to get correct lib location.
zrj [Thu, 24 May 2018 09:16:49 +0000 (12:16 +0300)]
tools/make_libdeps.sh: Adjust invocation.
Small changes:
* add /sbin to the patch for `sysctl hw.platform` in sys.mk;
* use tr(1) from /bin instead of rs(1) that might truncate;
* enforce C collate;
* explicitly sort find(1) results.
zrj [Thu, 24 May 2018 09:13:53 +0000 (12:13 +0300)]
tools/make_libdeps.sh: Update directory list.
Matthew Dillon [Thu, 24 May 2018 00:25:44 +0000 (17:25 -0700)]
hammer2 - Fix kmalloc pool blowout on low-memory machines
* Fix a kmalloc pool blown that can occur on low-memory machines due
to too many disconnected hammer2_inode structures building up.
* Was previously fixed for things like rm -rf and bulk renames,
but not for setattr (aka chown/chmod -R ops).
Reported-by: gjs278
François Tigeot [Wed, 23 May 2018 20:33:36 +0000 (22:33 +0200)]
drm/linux/timer.h: No need to protect a callout struct from MP accesses
Spotted-by: Imre Vadász <imre@vdsz.com>
Imre Vadász [Wed, 23 May 2018 18:21:10 +0000 (20:21 +0200)]
drm - No need to protect a callout struct from MP accesses.
Sascha Wildner [Wed, 23 May 2018 12:24:41 +0000 (14:24 +0200)]
bsd-family-tree: Fix SVN tag.
Eitan Adler [Wed, 23 May 2018 04:57:54 +0000 (04:57 +0000)]
Announce DragonFly 5.2.1
zrj [Wed, 23 May 2018 05:50:36 +0000 (08:50 +0300)]
bsd.kern.mk: Remove gcc80 workaround.
No longer needed.
zrj [Wed, 23 May 2018 05:30:46 +0000 (08:30 +0300)]
drm/i915: Disable few unused const structs and functions.
No functional change.
zrj [Mon, 21 May 2018 16:52:20 +0000 (19:52 +0300)]
btxld(8): Convert to hostprog.
There is no point to provide this utility in base. We are not planing to
bring back i386 support. Move it into sys/boot/pc32/ for easier handling.
This way the AOUT_H_FORCE32 now is confined to the same pc32/ directory.
The <sys/imgact_aout.h> has override for it to create proper btx header.
For future we should make btxld endian neutral.
zrj [Mon, 21 May 2018 09:38:00 +0000 (12:38 +0300)]
btools: Obsolete pre hostprog era cc_tools.
We no longer depend on them so this shaves about 1 min from buildworld.
Verified by comparing intermediates and final products.
Sascha Wildner [Mon, 21 May 2018 08:46:20 +0000 (10:46 +0200)]
hammer2(8): Clean up the Makefile a bit.
Reported-by: zrj
Sascha Wildner [Mon, 21 May 2018 08:17:44 +0000 (10:17 +0200)]
acpi: Add SUBDIR_ORDERED for the ACPICA tools for better parallelism.
Reported-by: zrj
Matthew Dillon [Mon, 21 May 2018 05:59:37 +0000 (22:59 -0700)]
hammer2 - Beef up bulkfree buffering
* Beef up the bulkfree buffering to allow a lot more memory
to be used for the in-memory freemap copy.
* Add a -m mem[k,m,g] override option for hammer2 bulkfree.
* Reduce console spam.
Sascha Wildner [Sun, 20 May 2018 11:19:17 +0000 (13:19 +0200)]
i386 removal, part 68/x: Remove a number of obsolete Makefiles from gnu/.
Reported-by: zrj
zrj [Sun, 20 May 2018 14:28:36 +0000 (17:28 +0300)]
kbdcontrol(1): Remove not needed defines.
While there, do the same for vidcontrol(1).
zrj [Sun, 20 May 2018 14:22:51 +0000 (17:22 +0300)]
share/syscons: Separate hostprog and remove it from btools.
The scrnmaps are not needed for buildworld.
zrj [Sun, 20 May 2018 14:19:55 +0000 (17:19 +0300)]
boot/pc32: Separate hostprog.
Add guard in common sys/boot/pc32/Makefile.inc.
Hostprog shall not take any of those definitions.
zrj [Sun, 20 May 2018 14:14:18 +0000 (17:14 +0300)]
boot/pc32: Fix typo in mbr.
Wrong patch in
faa6f4d7da5ad6ef2d3e68dd2671b6d6f8161a9b,
the generated mbr was still OK (bmake warning).
While there, only extract XREADORG constant from boot1*.out and not
from bootasmdef.nx in case it ever get one.
Sascha Wildner [Sun, 20 May 2018 09:09:38 +0000 (11:09 +0200)]
{un,}vis.3: Bring back some small fixes we had earlier.
Sascha Wildner [Sun, 20 May 2018 08:42:09 +0000 (10:42 +0200)]
vm_page_alloc.9: Adjust for the removal of PG_BUSY.
Sascha Wildner [Sun, 20 May 2018 08:34:47 +0000 (10:34 +0200)]
ip6.4: Clean up a bit.
* Remove info about IPV6_IPCOMP_LEVEL which was removed along with IPsec.
* Fix a typo.
Sascha Wildner [Sun, 20 May 2018 08:09:09 +0000 (10:09 +0200)]
mtree/BSD.root.dist: Use spaces.
Matthew Dillon [Sun, 20 May 2018 05:58:58 +0000 (22:58 -0700)]
kernel - Change closefrom() EINTR behavior
* closefrom() no longer breaks out of its loop if it gets an
EINTR from a close(). EINTR can only occur from NFS in this
situation, and the descriptor is still closed. Any EINTR's
which occur will be rolled up and returned after the loop
completes.
* Callers usually ignore closefrom()'s return value anyway.
Suggested-by: mjg
Matthew Dillon [Sun, 20 May 2018 02:26:01 +0000 (19:26 -0700)]
build - Change default /root directory perms
* Change from 755 to 700. 700 is a more suitable default for
~root's home directory.
Reported-by: aly
Sascha Wildner [Sat, 19 May 2018 22:47:05 +0000 (00:47 +0200)]
/usr/include/netinet6: Remove three unused IPsec related headers.
Sascha Wildner [Sat, 19 May 2018 17:54:14 +0000 (19:54 +0200)]
kgdb(1): Add missing "
Reported-by: zrj
zrj [Sat, 19 May 2018 07:39:24 +0000 (10:39 +0300)]
bsd.dep.mk: Split logic for CC and CXX handling.
Previously it assumed that cc(1) will be able to handle c++ sources.
That is not the case for compilers other than gcc or clang.
While there:
* add MKDEPCC that could be used to override CC in certain
more convoluted scenarios;
* finally add a fix for hostprog deps, note that PATH skip from
NXENV is intentional, we prefer not to use host's mkdep(1).
zrj [Sat, 19 May 2018 07:36:19 +0000 (10:36 +0300)]
gcc50: Plug in WARNS framework for tools.
At this point should be no-op change.
While there, do the same for gcc50 and gcc47.
zrj [Sat, 19 May 2018 07:33:06 +0000 (10:33 +0300)]
gcc80: Create headers before depend in libbacktrace.
Avoids the situation where (if make depend fails for some reason)
build stops in a strange state.
While there, do the same for gcc50.
zrj [Sat, 19 May 2018 07:30:59 +0000 (10:30 +0300)]
boot/pc32: Filter out global LDFLAGS.
For cases when we need to set global LDFLAGS for compiler frontends.
Sascha Wildner [Sat, 19 May 2018 11:31:16 +0000 (13:31 +0200)]
kernel: Declare the enum for the guest type before using it.
Found-by: zrj
Sascha Wildner [Sat, 19 May 2018 11:30:36 +0000 (13:30 +0200)]
hammer(8): Add inline keywords in two prototypes of inline functions.
Found-by: zrj
Matthew Dillon [Sat, 19 May 2018 06:41:14 +0000 (23:41 -0700)]
kernel - Fix swapcache cleaning issue
* Fix a bug when a single object has swapcached pages. In this
situation, when swapcache exhausts its scan of the object it
will fail to cycle back to the beginning. This prevents it
from cycling any further pages for the object.
* Does not corrupt the system or anything like that, but causes
swapcache to behave in a way that can sometimes make it less
useful.
Reported-by: tuxillo (java on monster)
Sascha Wildner [Sat, 19 May 2018 01:18:23 +0000 (03:18 +0200)]
examples/rconfig: Fix typos.
Sascha Wildner [Fri, 18 May 2018 12:15:29 +0000 (14:15 +0200)]
rtld-elf: -e is an ld(1) option.
Sascha Wildner [Fri, 18 May 2018 05:23:38 +0000 (07:23 +0200)]
errno.9: Fix reference.
Sascha Wildner [Thu, 17 May 2018 20:06:21 +0000 (22:06 +0200)]
top(1): Remove an unused file from the Makefile. We do pass HAVE_GETOPT.
Based-on-a-patch-by: zrj