Warner Losh [Mon, 13 Nov 2017 00:30:38 +0000 (00:30 +0000)]
Use proper include file. While <boot/userboot/userboot.h> works, it
only works because we have -Isys on the command line. We also have
-Isys/boot/userboot on the command line, so bring it in directly with
<userboot.h>. No functional change, but it removes one hard to see
dependency on the boot loader's location in sys/boot.
Michael Tuexen [Sun, 12 Nov 2017 21:43:33 +0000 (21:43 +0000)]
Cleanup the handling of control chunks. While there fix some minor
bug related to clearing the assoc retransmit counter and the dup TSN
handling of NR-SACK chunks.
Ed Maste [Sun, 12 Nov 2017 17:15:54 +0000 (17:15 +0000)]
boot1: also check for NULL device
r325681 fixed a NULL pointer dereference on RPi3 caused by a lack of
functionality in uboot's EFI implementation. That rev checked the boot1
load path for NULL but not the load device. In practice if the former
works the latter will as well, but improve correctness by checking each
separately.
Submitted by: Keith White <kwhite@eecs.uottawa.ca>
Reported by: jhb
MFC after: 5 days
MFC with: r325681
Sponsored by: The FreeBSD Foundation
Warner Losh [Sun, 12 Nov 2017 17:10:57 +0000 (17:10 +0000)]
Make sure the proper loader.rc gets installed.
There were two things wrong. First, the wrong path was listed in .PATH
statement. Second, BOOTSRC wasn't yet defined for the .PATH, so it
didn't properly add it. Third, even if these were right, . was in the
path before, so it wouldn't have worked. Replace this with a simple
loop so the proper loader.rc gets selected.
Noticed by: dhw@ (menus stopped working on boot)
Sponsored by: Netflix
Mariusz Zaborski [Sun, 12 Nov 2017 08:54:23 +0000 (08:54 +0000)]
dhclient(8) should not close stdio using daemon(3) function.
dhclient(8) is chrooted so opening /dev/null always will fail.
In capability world this is also annoying because we getting error that
open(2) is not permitted in Capsicum. dhclient(8) is closing stdio by
precaching fd to /dev/null before chroot.
This is done few lines below daemon(3) function so let's not try to do that
in daemon(3) function.
Mariusz Zaborski [Sun, 12 Nov 2017 08:42:43 +0000 (08:42 +0000)]
Use syslog service in dhclient(8).
dhclient(8) is failing during boot to connect to the syslog service, because
syslog daemon is started after dhclient(8). This can be reproduced by stooping
syslog daemon and ktrace the dhclient or use kern.trap_enotcap sysctl and boot
the machine. Using the Casper syslog service fix the problem.
Mariusz Zaborski [Sun, 12 Nov 2017 08:34:25 +0000 (08:34 +0000)]
Introduce syslog service for Casper.
syslog in libc secretly reconnects to the daemon.
Another issue is that we don't have any information from openlog(3) if we
succeeded to open log or not so we don't know if we are ready
to enter cabability mode.
Because all of that we decided we need a syslog service for Caspser.
Mariusz Zaborski [Sun, 12 Nov 2017 07:18:10 +0000 (07:18 +0000)]
We return a pointer when we are using cap_init() or cap_service_open()
function, so check if cap_chanel_t is NULL is not enough.
Casper with a normal libc will still fail in capability mote so let's not
enter capability mode without casper support when we need to resolve DNS.
Reviewed by: cem
Differential Revision: https://reviews.freebsd.org/D12823
Will Andrews [Sat, 11 Nov 2017 23:30:58 +0000 (23:30 +0000)]
libkvm: add kvm_walk_pages API.
This API allows callers to enumerate all known pages, including any
direct map & kernel map virtual addresses, physical addresses, size,
offset into the core, & protection configured.
For architectures that support direct map addresses, also generate pages
for any direct map only addresses that are not associated with kernel
map addresses.
Fix page size portability issue left behind from previous kvm page table
lookup interface.
Bartek Rutkowski [Sat, 11 Nov 2017 22:50:14 +0000 (22:50 +0000)]
bhyve: avoid applying capsicum capabilities to file that was not opened
When using -l option targeting file that can't be opened (ie. nmdm module
is not loaded and /dev/nmdm* is specified) bhyve tries to apply capsicum
capabilities to a file that was not opened.
Enclose that code in an if statement and only run it on correctly opened
descriptor also providing meaningful message in case of an error.
Allan Jude [Sat, 11 Nov 2017 14:02:21 +0000 (14:02 +0000)]
vmstat: fix duplicate key in libxo output
In the libxo output from vmstat, the number of pages that have been
paged out uses the same key name as the number of times pages have been
paged. Appears to have been a typo or copy-pasto.
Remove release and acquire semantics when accessing the "state" field of the
LinuxKPI task struct. Change type of "state" variable from "int" to
"atomic_t" to simplify code and avoid unneccessary casting.
Eitan Adler [Sat, 11 Nov 2017 07:00:40 +0000 (07:00 +0000)]
mailaddr(7): wave goodbye
The information here is somewhere between ancient to obsolete.
It refers to a time in the internet's history when manual routing
was still useful, talks about UUCP as if its modern, and refers
to documents which I had trouble tracking down.
It seems unlikely that a manual page in this form would be useful, so
just remove it.
Bryan Drewery [Sat, 11 Nov 2017 01:11:00 +0000 (01:11 +0000)]
DIRDEPS_BUILD: Fix after r325417.
DIRDEPS_BUILD works just fine without defining __objdir or dealing with any of
this logic. It handles its own TARGET_SPEC in local.meta.sys.mk as well. Just
let it do its own thing.
Bryan Drewery [Sat, 11 Nov 2017 01:10:46 +0000 (01:10 +0000)]
AUTO_OBJ: Disable in non-build targets.
There's no reason to create object directories for targets like 'installworld'
or 'distributeworld', and the others in this list. Specifying MK_AUTO_OBJ as a
make argument allows circumventing this if needed for some reason.
This fixes mergemaster creating a full object tree due to doing a 'make
installconfig' tree walk.
Reported by: Mark Millard
Sponsored by: Dell EMC Isilon
Warner Losh [Fri, 10 Nov 2017 23:54:58 +0000 (23:54 +0000)]
Install the 4th files in sys/boot/forth instead of each loader
Also, move generation of loader.help into loader.mk. Set HELP_FILES=
to disable this (so we only install one help file, for now). At the
same time remove some duplicate -I lines. Fix several FILES= and
CLEANFILES= into the += form since we're touching both of those in the
.mk files. Make sure we only build one loader.help file per platform
in a unified way (we were building many on some, with the last to
install winning, though often they were the same text).
Also, we're now installing loader.rc and menu.rc everywhere. arm and
mips uboot installed these as menu.rc.sample, but there's no need
since the loader.rc for those platforms doesn't do menu.rc processing
by default. pcibios.4th is now installed everywhere, but will failsafe
on non x86 platforms (it isn't loaded by default anywhere).
These changes are too intertwined to do separately since aspects of
each are required to have a bug-free commit.
Warner Losh [Fri, 10 Nov 2017 23:54:48 +0000 (23:54 +0000)]
Move LOADER_{NO,}_GELI_SUPPORT to MK_LOADER_GELI
Transition to WITH/WITHOUT_LOADER_GELI to flag support or not of GELI
in the boot loaders. Add HAVE_GELI so components can flag they need
support (since it's too large to include everywhere). Add temporary
warnings for the old forms to ease transition.
Also, update test script to build without GELI on x86.
Warner Losh [Fri, 10 Nov 2017 23:54:41 +0000 (23:54 +0000)]
Replace LOADER_FIREWIRE_SUPPORT variable
Rename LOADER_FIREWIRE_SUPPORT to MK_LOADER_FIREWIRE. Only build
libfirewire when this is "yes". Add note to updating. Fix build script
to build this for x86 so the option doesn't decay. sparc64 supports
ZFS, so also build it MK_ZFS=no.
Warner Losh [Fri, 10 Nov 2017 23:54:31 +0000 (23:54 +0000)]
Remove LOADER_ZFS_SUPPORT as a Makefile variable
LOADER_ZFS_SUPPORT is entirely used to select whether or not to
support ZFS in the loader. But it's not a user-servicable part (MK_ZFS
is what's used for that) Change it to the more conventional HAVE_ZFS
and move the ZFS support code into loader.mk. In addition, only build
ZFS libraries and boot loaders when ZFS is enabled.
Warner Losh [Fri, 10 Nov 2017 23:54:24 +0000 (23:54 +0000)]
Remove LOADER_FDT_SUPPORT as a Makefile variable.
LOADER_FDT_SUPPORT was used inconsistently in the tree. In some
places, it was used to control whether or not the user wanted FDT
included, and in other places it was a command to include
support. Remove it entirely. The former is now enabled -DWITH_FDT,
while the latter is controlled by Makefiles defining HAVE_FDT.
Warner Losh [Fri, 10 Nov 2017 23:54:18 +0000 (23:54 +0000)]
FDT support doesn't make sense for ps3. There's no support in the ps3
port for FDT, and it's unlikely to grow support for that anytime soon.
When it does, support can be added back easily enough.
Warner Losh [Fri, 10 Nov 2017 23:54:03 +0000 (23:54 +0000)]
libsa32 isn't needed for i386. It's already a 32-bit platform and
libfoo32 is only needed when compiling 32-bit libraries on native
64-bit architectures, and only when that 64-bit architecture needs
mixed 32-bit and 64-bit binaries.
Warner Losh [Fri, 10 Nov 2017 23:30:23 +0000 (23:30 +0000)]
Simplify the efivar interface a little.
We started out having Linux compatible libefivar interfaces. This was
in anticipation of porting the GPL'd efibootmgr to FreeBSD via a
port. However, since we need that functionality in the base, that port
isn't going to happened. It also appears that efivar is a private
library that's not used much outside a command line util and
efibootmgr. Reduce compatibility with the Linux version a little by
removing the mode parameter to efi_set_variable (which was unused on
FreeBSD, and not set to something useful in the code we'd
written). Also remove some efi error routines that were never
implemented and existed only to placate early GPL efibootmgr porting
experiments.
Ed Maste [Fri, 10 Nov 2017 22:17:30 +0000 (22:17 +0000)]
vnic: apply BPF tap before passing packet to hardware
Previously we passed tx packets to hardware via nicvf_tx_mbuf_locked
and then to the BPF tap, with a possibly invalid mbuf which would result
in a panic.
Ed Maste [Fri, 10 Nov 2017 21:26:44 +0000 (21:26 +0000)]
boot1: avoid using NULL device path
As of r323063 boot1 printed out the path & device from which it was
loaded, but uboot's EFI implementation lacked some support, resulting in
a NULL pointer and a crash. Check for a NULL pointer and avoid
reporting (and storing in the environment) the device and path in this
case.
Submitted by: Zakary Nafziger <worldofzak@gmail.com>
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D13038
Bryan Drewery [Fri, 10 Nov 2017 20:09:15 +0000 (20:09 +0000)]
Tell bsd.dep.mk which depend files to dinclude.
This allows the _SKIP_DEPEND optimization to work, avoiding reading
the files when not needed. It also fixes META_MODE incorrectly
reading these files when not needed.
Bryan Drewery [Fri, 10 Nov 2017 19:53:07 +0000 (19:53 +0000)]
native-xtools: Fix for GCC archs.
- This also adds in a _cleanobj step as needed.
- This redirects TARGET/TARGET_ARCH to NXB_TARGET/NXB_TARGET_ARCH in
Makefile.inc1 as the main build needs to be for MACHINE rather
than TARGET.
First build the toolchain and then use that as an external toolchain
to build the needed directories and NXB_TARGET-toolchain, all as
MACHINE files though via TARGET_TRIPLE=MACHINE_TRIPLE.
The NXBDIRS is evaluated in the 'everything' submake as it needs to be
based on TARGET's src.opts.mk values, such as MK_GCC=yes when building
on a MK_CLANG=yes MACHINE. This can likely be changed to a specific
_native-xtools-everything target later and the funky late evaluation
of SUBDIR_OVERRIDE removed.
Bryan Drewery [Fri, 10 Nov 2017 19:52:31 +0000 (19:52 +0000)]
Don't append TARGET.TARGET_ARCH if OBJTOP is already set.
This avoids flipping the expected TARGET.TARGET_ARCH suffix / OBJTOP when it is
already set by a parent make which wants to control it more such as in
something like 'make native-xtools'.
Scott Long [Fri, 10 Nov 2017 17:01:51 +0000 (17:01 +0000)]
Refactoring the interrupt setup code introduced a bug where the drivers
would attempt to re-allocate interrupts during a chip reset without
first de-allocating them. Doing that right is going to be tricky, so
just band-aid it for now so that a re-init doesn't guarantee a failure
due to resource re-use.
The hardware is capable of 2 requestor endianness modes for standard 8
byte atomics: BE (0x0) and host endianness (0x1). Read the supported
modes from hca atomic capabilities and configure HW to host endianness
mode if supported.
Refactor the flowsteering APIs used by mlx5en(4). This change is needed by
the coming ibcore and mlx5ib updates in order to support traffic redirection
to so-called raw ethernet QPs.
Remove unused E-switch related routines and files while at it.
Set the IPOIB_FLAG_INITIALIZED on dev_open and clear it on dev_stop to
avoid a race between ipoib load and the underlying device driver.
The device module must dispatch the IB_EVENT_PORT_ACTIVE event before ipoib
module is loaded. Otherwise, the flush will fail since no one set the
IPOIB_FLAG_INITIALIZED.
Remove some not needed comments in the LinuxKPI. Use the Linux source tree
to lookup documentation for the functions implemented in the LinuxKPI
instead.