as of r288992 use it to manage the CCNT.
Use the CNNT for get_cyclecount() instead of binuptime() when device pmu
is compiled in; if it fails to attach, fall back to the former method.
Enable by default for the BeagleBoneBlack configuration.
Optained from: Cambridge/L41
Sponsored by: DARPA/AFRL
Reviewed by: andrew
Differential Revision: https://reviews.freebsd.org/D3837
ccache is mostly beneficial for frequent builds where -DNO_CLEAN is not
used to achieve a safe pseudo-incremental build. This is explained in
more detail upstream [1] [2]. It incurs about a 20%-28% hit to populate the
cache, but with a full cache saves 30-50% in build times. When combined with
the WITH_FAST_DEPEND feature it saves up to 65% since ccache does cache the
resulting dependency file, which it does not do when using mkdep(1)/'CC
-E'. Stats are provided at the end of this message.
This removes the need to modify /etc/make.conf with the CC:= and CXX:=
lines which conflicted with external compiler support [3] (causing the
bootstrap compiler to not be built which lead to obscure failures [4]),
incorrectly invoked ccache in various stages, required CCACHE_CPP2 to avoid
Clang errors with parenthesis, and did not work with META_MODE.
The option name was picked to match the existing option in ports. This
feature is available for both in-src and out-of-src builds that use
/usr/share/mk.
Linking, assembly compiles, and pre-processing avoid using ccache since it is
only overhead. ccache does nothing special in these modes, although there is
no harm in calling it for them.
CCACHE_COMPILERCHECK is set to 'content' when using the in-tree bootstrap
compiler to hash the content of the compiler binary to determine if it
should be a cache miss. For external compilers the 'mtime' option is used
as it is more efficient and likely to be correct. Future work may optimize the
'content' check using the same checks as whether a bootstrap compiler is needed
to be built.
The CCACHE_CPP2 pessimization is currently default in our devel/ccache
port due to Clang requiring it. Clang's -Wparentheses-equality,
-Wtautological-compare, and -Wself-assign warnings do not mix well with
compiling already-pre-processed code that may have expanded macros that
trigger the warnings. GCC has so far not had this issue so it is allowed to
disable the CCACHE_CPP2 default in our port.
Sharing a cache between multiple checkouts, or systems, is explained in
the ccache manual. Sharing a cache over NFS would likely not be worth
it, but syncing cache directories between systems may be useful for an
organization. There is also a memcached backend available [5]. Due to using
an object directory outside of the source directory though you will need to
ensure that both are in the same prefix and all users use the same layout. A
possible working layout is as follows:
Source: /some/prefix/src1
Source: /some/prefix/src2
Source: /some/prefix/src3
Objdir: /some/prefix/obj
Environment: CCACHE_BASEDIR='${SRCTOP:H}' MAKEOBJDIRPREFIX='${SRCTOP:H}/obj'
This will use src*/../obj as the MAKEOBJDIRPREFIX and tells ccache to replace
all absolute paths to be relative. Using something like this is required due
to -I and -o flags containing both SRC and OBJDIR absolute paths that ccache
adds into its hash for the object without CCACHE_BASEDIR.
distcc can be hooked into by setting CCACHE_PREFIX=/usr/local/bin/distcc.
I have not personally tested this and assume it will not mix well with
using the bootstrap compiler.
The cache from buildworld can be reused in a subdir by first running
'make buildenv' (from r290424).
Note that the cache is currently different depending on whether -j is
used or not due to ccache enabling -fdiagnostics-color automatically if
stderr is a TTY, which bmake only does if not using -j.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_CCACHE_BUILD=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files and ccache objects.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
The WITH_FAST_DEPEND feature was used for comparison here as well to show
the dramatic time savings with a full cache.
buildworld:
x buildworld-before
+ buildworld-ccache-empty
* buildworld-ccache-full
% buildworld-ccache-full-fastdep
# buildworld-fastdep
+-------------------------------------------------------------------------------+
|% * # +|
|% * # +|
|% * # xxx +|
| |A |
| A|
| A |
|A |
| A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 4519 4525.04 4520.73 4521.59 3.1104823
Difference at 95.0% confidence
758.027 +/- 43.4565
20.1412% +/- 1.15466%
(Student's t, pooled s = 19.1726)
* 3 1823.08 1827.2 1825.62 1825.3 2.0785572
Difference at 95.0% confidence
-1938.26 +/- 43.298
-51.5007% +/- 1.15045%
(Student's t, pooled s = 19.1026)
% 3 1266.96 1279.37 1270.47 1272.2667 6.3971113
Difference at 95.0% confidence
-2491.3 +/- 44.3704
-66.1952% +/- 1.17895%
(Student's t, pooled s = 19.5758)
# 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-ccache-empty
* buildkernel-ccache-empty-fastdep
% buildkernel-ccache-full
# buildkernel-ccache-full-fastdep
@ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|# @ % * |
|# @ % * x + |
|# @ % * xx ++|
| MA |
| MA|
| A |
| A |
|A |
| A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 727.97 731.91 728.06 729.31333 2.2492295
Difference at 95.0% confidence
156.88 +/- 4.17129
27.4058% +/- 0.728695%
(Student's t, pooled s = 1.84034)
* 3 527.1 528.29 528.08 527.82333 0.63516402
Difference at 95.0% confidence
-44.61 +/- 2.33254
-7.79305% +/- 0.407478%
(Student's t, pooled s = 1.02909)
% 3 400.4 401.05 400.62 400.69 0.3306055
Difference at 95.0% confidence
-171.743 +/- 2.16453
-30.0023% +/- 0.378128%
(Student's t, pooled s = 0.954969)
# 3 201.94 203.34 202.28 202.52 0.73020545
Difference at 95.0% confidence
-369.913 +/- 2.40293
-64.6212% +/- 0.419774%
(Student's t, pooled s = 1.06015)
@ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
[1] https://ccache.samba.org/performance.html
[2] http://www.mail-archive.com/ccache@lists.samba.org/msg00576.html
[3] https://reviews.freebsd.org/D3484
[5] https://github.com/jrosdahl/ccache/pull/30
PR: 182944 [4]
MFC after: 3 weeks
Sponsored by: EMC / Isilon Storage Division
Relnotes: yes
This is because the .meta files generated from filemon already contain a
list of all files read to generate the object.
X-MFC-With: r290433
MFC after: 3 weeks
Sponsored by: EMC / Isilon Storage Division
This is especially noticeable in the kernel obj directory since it
includes so many files.
X-MFC-With: r290433
MFC after: 3 weeks
Sponsored by: EMC / Isilon Storage Division
driver. This is taken from the MAC at boot, but can be overridden with
'options AT91_MACB_USE_RMII'.
Switch to macb for HL201 and SAM9G20EK boards. It now works both
places. Also start to sneak up on FDT for the SAM9G20EK board, but
leave disabled due to issues with MMC that haven't been resolved.
Add early debug support for the SAM9G20EK since that is required
for FDT to work presently on these SoC.
This speeds up buildworld by 16% on my system and buildkernel by 35%.
Rather than calling mkdep(1), which is just a wrapper around 'cc -E',
use the modern -MD -MT -MF flags to gather and generate dependencies during
compilation. This flag was introduced in GCC "a long time ago", in GCC 3.0,
and is also supported by Clang. (It appears that ICC also supports this but I
do not have access to test it). This avoids running the preprocessor *twice*
for every build, in both 'make depend' and 'make all'. This is especially
noticeable when using ccache since it does not cache preprocessor results from
mkdep(1) / 'cc -E', but still speeds up compilation with the -MD flags.
For 'make depend' a tree-walk is still done to ensure that all DPSRCS
are generated when expected, and that beforedepend/afterdepend and
_EXTRADEPEND are all still respected. In time this may change but for now
I've been conservative. The time for a tree-walk with -j combined with
SUBDIR_PARALLEL is not significant. For example, it takes about 9 seconds
with -j15 to walk all of src/ for 'make depend' now on my system.
A .depend file is still generated with the various rules that apply to
the final target, or custom rules. Otherwise there are now
per-built-object-file .depend files, such as .depend.filename.o. These
are included directly by make rather than populating .depend with a loop
and .depend lines, which only added overhead to the now almost-NOP 'make
depend' phase.
Before this I experimented with having mkdep(1) called in parallel per-file.
While this improved the kernel and lib/libc 'make depend' phase, it resulted
in slower build times overall.
The -M flags are removed from CFLAGS when linking since they have no effect.
Enabling this by default, for src or out-of-src, can be done once more testing
has been done, such as a ports exp-run, and with more compilers.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_FAST_DEPEND=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
buildworld:
x buildworld-before
+ buildworld-fastdep
+-------------------------------------------------------------------------------+
|+ |
|+ |
|+ xx x|
| |_MA___||
|A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|+ x |
|++ xx|
| A||
|A| |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
Sponsored by: EMC / Isilon Storage Division
MFC after: 3 weeks
Relnotes: yes
wrong value in the comparison, leading to incorrectly setting the new
value.
This has been observed in the ZFS code. Without this we can lose track of
the reference count in a zrlock object.
We should move to use the generic atomic functions, however as this has
been observed I would prefer to have this working, then move to the generic
functions.
PR: 204037
Sponsored by: ABT Systems Ltd
- Move all files related to the LinuxKPI into sys/compat/linuxkpi and
its subfolders.
- Update sys/conf/files and some Makefiles to use new file locations.
- Added description of COMPAT_LINUXKPI to sys/conf/NOTES which in turn
adds the LinuxKPI to all LINT builds.
- The LinuxKPI can be added to the kernel by setting the
COMPAT_LINUXKPI option. The OFED kernel option no longer builds the
LinuxKPI into the kernel. This was done to keep the build rules for
the LinuxKPI in sys/conf/files simple.
- Extend the LinuxKPI module to include support for USB by moving the
Linux USB compat from usb.ko to linuxkpi.ko.
- Bump the FreeBSD_version.
- A universe kernel build has been done.
Reviewed by: np @ (cxgb and cxgbe related changes only)
Sponsored by: Mellanox Technologies
This commit introduces support for etherswitch devices that utilize SMI as
a way of accessing its registers. SMI register is located in address space
of mge -- access to it was exported through MDIO interface.
Attachment functions were enhanced so as to ensure proper initialisation
in both cases: 1) PHYs attached directly to mge, 2) PHYs attached to
switch device and switch attached to mge. Attachment of etherswitch device
depends on dts entry with compatible="mrvl,sw" property. If none is found,
typical PHY attachment procedure follows.
In case of switch attached, PHYs' status and configuration is accessible
via etherswitchcfg, and ifconfig shows always-up, non-configurable mge
interfaces.
Due to the fact that there may be simultaneous accessess to SMI
registers (e.g. from PHY attached to one of mge instances and switch
to the other), SMI access interlock was added. It is SX lock,
because sleep ability is necessary -- busy-waiting would result
in poor performance due to long delays required by hardware.
Underlying switch driver is obliged to use sleepable locks as well.
Reviewed by: adrian
Obtained from: Semihalf
Submitted by: Bartosz Szczepanek <bsz@semihalf.com>
Differential revision: https://reviews.freebsd.org/D3900
It turns out that it is pretty easy to make CloudABI work on ARM64. We
essentially only need to copy over the sysvec from AMD64 and ensure that
we use ARM64 specific registers.
As there is an overlap between function argument and return registers,
we do need to extend cloudabi64_schedtail() to only set its values if
we're actually forking. Not when we're creating a new thread.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D3917
In order to make it easier to support CloudABI on ARM64, move out all of
the bits from the AMD64 cloudabi_sysvec.c into a new file
cloudabi_module.c that would otherwise remain identical. This reduces
the AMD64 specific code to just ~160 lines.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D3974
to copying in some code from the armv4 busdma, and adapting a few variable
and flag names to match the surrounding mips code.
Instead of keeping a local cache of prealloced busdma_map structs on a
mutex-protected list, set up an uma zone to cache them.
Instead of all memory allocations using M_DEVBUF, use new categories
M_BUSDMA for allocations of metadata (tags, maps, segment tracking lists),
and M_BOUNCE for bounce pages.
When buffers are allocated out of the busdma_bufalloc zones the alignment
and size of the buffers is known, and the code can skip doing any "partial
cacheline flush" logic to preserve data that may be adjacent to the DMA
buffer but contain non-DMA data.
Reviewed by: adrian, imp
This commit adds support for MDIO present in the ThunderX SoC.
From the FDT point of view it is compatible with "octeon-3860-mdio"
however only C22 mode is used.
The code also implements lmac_if interface functions.
Obtained from: Semihalf
Sponsored by: The FreeBSD Foundation
- The driver consists of three main componens: PF, VF, BGX
- Requires appropriate entries in DTS and MDIO driver
- Supports only FDT configuration
- Multiple Tx queues and single Rx queue supported
- No RSS, HW checksum and TSO support
- No more than 8 queues per-IF (only one Queue Set per IF)
- HW statistics enabled
- Works in all available MAC modes (1,10,20,40G)
- Style converted to BSD according to style(9)
- The code brings lmac_if interface used by the BGX driver to
update its logical MACs state.
Obtained from: Semihalf
Sponsored by: The FreeBSD Foundation
and armv6 architecures. The primary enhancement over the old design is
support for hierarchical interrupt controllers (such as a gpio driver
which can receive interrupts from a root PIC and act as a PIC itself for
clients interested in handling a change of gpio pin state as an
interrupt). The new code also provides an infrastructure for mapping
interrupts described in metadata in the form of a "controller reference
plus interrupt number" tuple into the simple "0-n" flat numeric space
understood by rman and the bus resource mechanisms.
Use of the new code is enabled by setting the ARM_INTRNG option, and by
making a few simple changes to the platform's support code. In addition
each existing PIC driver needs changes to be ready for INTRNG; this commit
contains the changes for the arm/gic driver, which most armv6 SoCs use, but
it does not enable the new code yet on any platform.
This project has been many years in the making, starting as a GSoC project
by Jakub Klama (jceel@) in 2012. That didn't get committed right away and
the source base evolved out from under it to some degree. In 2014 I rebased
the diffs to then -current and did some enhancements in the area of mapping
interrupt numbers and storing associated fdt data, then the project went
cold again for a while. Eventually Svata Kraus took that work in progress
and did another big round of work on it, removing most of the remaining
rough edges. Finally I took that and made one more pass through it, mostly
disabling the "INTR_SOLO" feature for now, pending further design
discussions on how to most efficiently dispatch a pending interrupt through
more than one layer of PIC. The current code with the INTR_SOLO feature
disabled uses approximate 100 extra cpu cycles for each cascaded PIC the
interrupt has to be passed to, so what's left to do is about efficiency, not
correct operation.
Differential Revision: https://reviews.freebsd.org/D2047
an error.
Most of these do a 'mkdir -p' or 'install -d' before installing, but add
the trailing / here for consistency with the userland install.
MFC after: 2 weeks
X-MFC-With: r289391
Sponsored by: EMC / Isilon Storage Division
HWPMC depends on pmu.c even if device pmu is not specified.
Would be great if we could just automatically enabled "device pmu"
if we try to compile in HWPMC.
Also several old kernel cnfigurations seem to have HWPMC enabled but are
pre-FDT and thus fail. So make pmu.c depend on fdt in case of hwpmc as
well.
MFC after: 2 weeks
Sponsored by: DARPA/AFRL
Differential Revision: https://reviews.freebsd.org/D3877
To me that seems broken as certain interrupts will never be handled
properly. I'll re-open D3877 and we can seek a better solution and
try again. For now go back to that state and avoid compile time errors.
Would be great if we could just automatically enabled "device pmu"
if we try to compile in HWPMC.
MFC after: 2 weeks
Sponsored by: DARPA/AFRL
Reviewed by: andrew
Differential Revision: https://reviews.freebsd.org/D3877
packets and/or state transitions from each TCP socket. That would help with
narrowing down certain problems we see in the field that are hard to reproduce
without understanding the history of how we got into a certain state. This
change provides just that.
It saves copies of the last N packets in a list in the tcpcb. When the tcpcb is
destroyed, the list is freed. I thought this was likely to be more
performance-friendly than saving copies of the tcpcb. Plus, with the packets,
you should be able to reverse-engineer what happened to the tcpcb.
To enable the feature, you will need to compile a kernel with the TCPPCAP
option. Even then, the feature defaults to being deactivated. You can activate
it by setting a positive value for the number of captured packets. You can do
that on either a global basis or on a per-socket basis (via a setsockopt call).
There is no way to get the packets out of the kernel other than using kmem or
getting a coredump. I thought that would help some of the legal/privacy concerns
regarding such a feature. However, it should be possible to add a future effort
to export them in PCAP format.
I tested this at low scale, and found that there were no mbuf leaks and the peak
mbuf usage appeared to be unchanged with and without the feature.
The main performance concern I can envision is the number of mbufs that would be
used on systems with a large number of sockets. If you save five packets per
direction per socket and have 3,000 sockets, that will consume at least 30,000
mbufs just to keep these packets. I tried to reduce the concerns associated with
this by limiting the number of clusters (not mbufs) that could be used for this
feature. Again, in my testing, that appears to work correctly.
Differential Revision: D3100
Submitted by: Jonathan Looney <jlooney at juniper dot net>
Reviewed by: gnn, hiren
- Move the required kernel compiler flags from Makefile.arm64 to kern.mk.
- Build arm64 modules as PIC; non-PIC relocations in .o for shared object
output cannot be handled.
- Do not try to install aarch64 symlink.
- A hack for arm64 to avoid ld -r stage. See the comment for the explanation.
Some functionality is lost, like ctf handling, but hopefully will be
restored after newer linker is available.
Reviewed by: andrew, emaste
Tested by: andrew (on real hardware)
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3796
The current Xen console driver is crashing very quickly when using it on
an ARM guest. This is because the console lock is recursive and it may
lead to recursion on the tty lock and/or corrupt the ring pointer.
Furthermore, the console lock is not always taken where it should be and has
to be released too early because of the way the console has been designed.
Over the years, code has been modified to support various new features but
the driver has not been reworked.
This new driver has been rewritten with the idea of only having a small set
of specific function to write either via the shared ring or the hypercall
interface.
Note that HVM support has been left aside for now because it requires
additional features which are not yet supported. A follow-up patch will be
sent with HVM guest support.
List of items that may be good to have but not mandatory:
- Avoid to flush for each character written when using the tty
- Support multiple consoles
Submitted by: Julien Grall <julien.grall@citrix.com>
Reviewed by: royger
Differential Revision: https://reviews.freebsd.org/D3698
Sponsored by: Citrix Systems R&D
specific as we may use the pmu registers for other uses. No configs seem
to currently build this.
This will allow for more use of this device.
Discussed with: bz
Sponsored by: ABT Systems Ltd
This avoids needing a large boot partition / file system in order to
accommodate multiple kernels, and provides consistency with userland
debug. This also simplifies the process of moving kernel debug files
to a separate package and installing them on demand.
In addition, change kernel debug file extension to .debug, to match
userland debug files.
When using the supported kernel installation method the
/usr/lib/debug/boot/kernel directory will be renamed (to kernel.old)
as is done with /boot/kernel.
Developers wishing to maintain the historical behavior of installing
debug files in /boot/kernel/ can set KERN_DEBUGDIR="" in src.conf(5).
Reviewed by: bdrewery, brooks, imp, markj
Relnotes: yes
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D1006
This also adds a newbus interface that allows a SoC to override the
following settings:
- if_dwc specific SoC initialization;
- if_dwc descriptor type;
- if_dwc MII clock.
This seems to be an old version of the hardware descriptors but it is
still in use in a few SoCs (namely Allwinner A20 and Amlogic at least).
Tested on Cubieboard2 and Banana pi.
Tested for regressions on Altera Cyclone by br@ (old version).
Obtained from: NetBSD
drivers into the revived sys/sparc64/pci/ofw_pci.c, previously already
serving a similar purpose. This has been done with sun4v in mind, which
explains a) the otherwise not that obvious scheme employed and b) why
reusing sys/powerpc/ofw/ofw_pci.c was even lesser an option.
- Add a workaround for QEMU once again not emulating real machines, in
this case by not providing the OFW_PCI_CS_MEM64 range. [1]
Submitted by: jhb [1]
MFC after: 1 week
CTL HA functionality was originally implemented by Copan many years ago,
but large part of the sources was never published. This change includes
clean room implementation of the missing code and fixes for many bugs.
This code supports dual-node HA with ALUA in four modes:
- Active/Unavailable without interlink between nodes;
- Active/Standby with second node handling only basic LUN discovery and
reservation, synchronizing with the first node through the interlink;
- Active/Active with both nodes processing commands and accessing the
backing storage, synchronizing with the first node through the interlink;
- Active/Active with second node working as proxy, transfering all
commands to the first node for execution through the interlink.
Unlike original Copan's implementation, depending on specific hardware,
this code uses simple custom TCP-based protocol for interlink. It has
no authentication, so it should never be enabled on public interfaces.
The code may still need some polishing, but generally it is functional.
Relnotes: yes
Sponsored by: iXsystems, Inc.
RANDOM_LOADABLE and RANDOM_YARROW's definitions from opt_random.h to
opt_global.h
This unbreaks `make depend` in sys/modules with multiple drivers (tmpfs, etc)
after r286839
X-MFC with: r286839
Reviewed by: imp
Submitted by: lwhsu
Differential Revision: D3486
SoC is used in the HiKey board from 96boards.
Currently on the SD card is working on the HiKey, as such devices 0 and 2
will need to be disabled, for example by adding the following to
loader.conf:
hint.hisi_dwmmc.0.disabled=1
hint.hisi_dwmmc.2.disabled=1
Relnotes: yes (Hikey board booting)
Sponsored by: ABT Systems Ltd
only gpiobus configured via FDT is supported. Bus enumeration is
supported. Devices are created for each device found. 1-Wire
temperature controllers are supported, but other drivers could be
written. Temperatures are polled and reported via a sysctl. Errors
are reported via sysctl counters. Mis-wired bus detection is included
for more trouble shooting. See ow(4), owc(4) and ow_temp(4) for
details of what's supported and known issues.
This has been tested on Raspberry Pi-B, Pi2 and Beagle Bone Black
with up to 7 devices.
Differential Revision: https://reviews.freebsd.org/D2956
Relnotes: yes
MFC after: 2 weeks
Reviewed by: loos@ (with many insightful comments)
