dim [Tue, 18 Mar 2014 19:35:26 +0000 (19:35 +0000)]
Pull in r203311 from upstream llvm trunk (by Arnold Schwaighofer):
ISel: Make VSELECT selection terminate in cases where the condition type has to
be split and the result type widened.
When the condition of a vselect has to be split it makes no sense widening the
vselect and thereby widening the condition. We end up in an endless loop of
widening (vselect result type) and splitting (condition mask type) doing this.
Instead, split both the condition and the vselect and widen the result.
I ran this over the test suite with i686 and mattr=+sse and saw no regressions.
Fixes PR18036.
With this fix the original problem case from the graphics/rawtherapee
port (posted in http://llvm.org/PR18036 ) now compiles within ~97MB RSS.
dim [Tue, 18 Mar 2014 19:23:41 +0000 (19:23 +0000)]
Pull in r196939 from upstream llvm trunk (by Reid Kleckner):
Reland "Fix miscompile of MS inline assembly with stack realignment"
This re-lands commit r196876, which was reverted in r196879.
The tests have been fixed to pass on platforms with a stack alignment
larger than 4.
Update to clang side tests will land shortly.
Pull in r196986 from upstream llvm trunk (by Reid Kleckner):
Revert the backend fatal error from r196939
The combination of inline asm, stack realignment, and dynamic allocas
turns out to be too common to reject out of hand.
ASan inserts empy inline asm fragments and uses aligned allocas.
Compiling any trivial function containing a dynamic alloca with ASan is
enough to trigger the check.
XFAIL the test cases that would be miscompiled and add one that uses the
relevant functionality.
Pull in r202930 from upstream llvm trunk (by Hans Wennborg):
Check for dynamic allocas and inline asm that clobbers sp before building
selection dag (PR19012)
In X86SelectionDagInfo::EmitTargetCodeForMemcpy we check with MachineFrameInfo
to make sure that ESI isn't used as a base pointer register before we choose to
emit rep movs (which clobbers esi).
The problem is that MachineFrameInfo wouldn't know about dynamic allocas or
inline asm that clobbers the stack pointer until SelectionDAGBuilder has
encountered them.
This patch fixes the problem by checking for such things when building the
FunctionLoweringInfo.
Together, these commits fix the problem encountered in the devel/emacs
port on the i386 architecture, where a combination of stack realignment,
alloca() and memcpy() could incidentally clobber the %esi register,
leading to segfaults in the temacs build-time utility.
See also: http://llvm.org/PR18171 and http://llvm.org/PR19012
kib [Tue, 18 Mar 2014 16:41:32 +0000 (16:41 +0000)]
Add support for the PCI(e)-PCI bridges to the Intel VT-d driver. The
bridge takes ownership of the transaction, so bsf of the requester is
the bridge and not a device behind it. As result, code needs to walk
the hierarchy up to use correct context.
Note that PCIe->PCI-X bridges are not handled quite correctly since
such bridges are allowed to only take ownership of some transactions.
Also, weird but unrealistic cases of PCIe behind PCI bus are also not
handled.
Still, the patch provides significant step forward for the bridge
handling.
Submitted by: Jason Harmening <jason.harmening@gmail.com>
MFC after: 1 week
kib [Tue, 18 Mar 2014 16:20:33 +0000 (16:20 +0000)]
It is not uncommon for BIOSes to report wrong RMRR entries in DMAR
table. Among them, some (old AMI ?) BIOSes report entries with range
like (bf7ec000, bf7ebfff). Attempts to ignore the bogus entries
result in faults, so the range must be covered somehow.
Provide a workaround by identity mapping the 32 pages after the bogus
entry start, which seems to be enough for the reported BIOS.
Reported and tested by: Jason Harmening <jason.harmening@gmail.com>
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
imp [Tue, 18 Mar 2014 14:41:18 +0000 (14:41 +0000)]
In kernel config files, it is supposed to be 'options<space><tab>' not
'options<tab><tab>', per long standing (but recently not so strictly
enforced) convention.
neel [Tue, 18 Mar 2014 02:49:28 +0000 (02:49 +0000)]
When a vcpu is deactivated it must also unblock any rendezvous that may be
blocked on it.
This is done by issuing a wakeup after clearing the 'vcpuid' from 'active_cpus'.
Also, use CPU_CLR_ATOMIC() to guarantee visibility of the updated 'active_cpus'
across all host cpus.
emaste [Tue, 18 Mar 2014 01:40:25 +0000 (01:40 +0000)]
Update NetBSD Foundation copyrights to 2-clause BSD
The NetBSD Foundation states "Third parties are encouraged to change the
license on any files which have a 4-clause license contributed to the
NetBSD Foundation to a 2-clause license."
This change removes clauses 3 and 4 from copyright / license blocks that
list The NetBSD Foundation as the only copyright holder.
jimharris [Mon, 17 Mar 2014 22:36:04 +0000 (22:36 +0000)]
nvme: Remove the software progress marker SET_FEATURE command during
controller initialization.
The spec says OS drivers should send this command after controller
initialization completes successfully, but other NVMe OS drivers are
not sending this command. This change will therefore reduce differences
between the FreeBSD and other OS drivers.
ian [Mon, 17 Mar 2014 16:10:42 +0000 (16:10 +0000)]
Cosmetic changes to printed output, mostly related to probing devices...
- Display slice and partition as <auto> instead of 0 or -1 when they're
not set to specific values (the paritition=-1 was confusing folks).
- When loaderdev isn't set in the u-boot environment, say so rather
than displaying unknown device ''.
- Print the loader(8) ident/version info earlier, so that all device-
related info appears together afterwards.
The one change here that isn't purely cosmetic is to call setheap()
earlier. The comment says "Initialise heap as early as possible", now
that's more accurate. It shouldn't make any functional difference, but
may be safer if future changes lead to trying to allocate memory earlier.
bryanv [Mon, 17 Mar 2014 05:45:29 +0000 (05:45 +0000)]
Add Tx/Rx multiqueue support to vmx(4)
As a prerequisite for multiple queues, the guest must have MSIX enabled.
Unfortunately, to work around device passthrough bugs, FreeBSD disables
MSIX when running as a VMWare guest due to the hw.pci.honor_msi_blacklist
tunable; this tunable must be disabled for multiple queues.
Also included is various minor changes from the projects/vmxnet branch.
marcel [Mon, 17 Mar 2014 04:38:10 +0000 (04:38 +0000)]
In intr_event_handle() we already save and set td_intr_frame, so
don't do it also in ia64_handle_intr(). With ia64_handle_intr()
not saving and setting td_intr_frame, make sure to do it for
timer interrupts in ia64_ih_hardclock().
marcel [Sun, 16 Mar 2014 21:57:05 +0000 (21:57 +0000)]
Don't use the ITC as the faulting address for external interrupts.
We only use it for tracing and the KTR infrastructure will use ITC
for the time-stamp.
ian [Sun, 16 Mar 2014 20:39:39 +0000 (20:39 +0000)]
Use armv7 TLB flush code, not arm11, for cortex-a processors.
The armv7 architecture uses a unified TLB model for maintenence ops even
if separate instruction and data TLBs are implemented in hardware. That
means that there's no distinction between the 'I' and 'D' flavors of flush,
they all use the same 'ID' implementation. On the other hand, there is a
difference between SMP and UP on armv7, but not on arm11, so use the armv7
routines for cortex-a processors.
marcel [Sun, 16 Mar 2014 20:21:40 +0000 (20:21 +0000)]
In intr_event_handle() we already save and set td_intr_frame, so
don't do it also in ia64_handle_intr(). With ia64_handle_intr()
not saving and setting td_intr_frame, make sure to do it for
clock interrupts in ia64_ih_clock().
imp [Sun, 16 Mar 2014 15:22:52 +0000 (15:22 +0000)]
Align all comments in config files on same column. This consistency
helps when bits and pieces of GENERIC from i386 or amd64 are cut and
pasted into other architecture's config files (which in the case of
ARM had gotten rather akimbo).
imp [Sun, 16 Mar 2014 15:22:46 +0000 (15:22 +0000)]
Make all the comments '# ' and align to same column. This fixes the
rampently incosnsitent usage which made cut and paste from one file
to another look ugly.
andrew [Sun, 16 Mar 2014 13:16:30 +0000 (13:16 +0000)]
Add an optimised version of the float and double helper functions. These
are only used on armv6 when the vfp unit is detected. They will also be
available for the upcoming armv6hf platform, however while not used by
default there will need to be defined for any software that calls them
directly.
tychon [Sun, 16 Mar 2014 12:33:40 +0000 (12:33 +0000)]
Though there currently isn't a way to insert new media into an ATAPI
drive, at least pretend to support Asynchronous Notification (AN) to
avoid a guest needlessly polling for it.
tuexen [Sun, 16 Mar 2014 12:32:16 +0000 (12:32 +0000)]
* Provide information in error causes in ASCII instead of
proprietary binary format.
* Add support for a diagnostic information error cause.
The code is sysctlable and the default is 0, which
means it is not sent.
rwatson [Sun, 16 Mar 2014 10:55:57 +0000 (10:55 +0000)]
Update kernel inclusions of capability.h to use capsicum.h instead; some
further refinement is required as some device drivers intended to be
portable over FreeBSD versions rely on __FreeBSD_version to decide whether
to include capability.h.
rwatson [Sun, 16 Mar 2014 10:49:16 +0000 (10:49 +0000)]
Rename capability.h to capsicum.h: the original name conflicts with the
draft POSIX.1e capability.h used on some systems (e.g., Linux). On
FreeBSD, leave a wrapper header so that current code continues to compile.
We will eventually want to deprecate the old header as the presence of a
capability.h may be confusing some configure scripts.
Suggested by: David Drysdale <drysdale at google.com>
Discussed on: cl-capsicum-discuss
MFC after: 3 weeks
andrew [Sun, 16 Mar 2014 09:40:05 +0000 (09:40 +0000)]
Only build the vfp/softfp switching code on armv6 as we don't support vfp
on anything earlier than this. This should fix the armeb and arm builds
when using gcc.
adrian [Sun, 16 Mar 2014 08:38:31 +0000 (08:38 +0000)]
* Handle the three other timer interrupts for now, from the AR724x
later. If the interrupts are ACKed even if they're not masked, we get
the interrupts again later. Grr.
* The AR724x and later chips want the interrupt bits cleared by writing the
relevant bit to it, NOT by writing all but the current interrupt to it.
Tested:
* AR9344, DB120 reference board
TODO:
* Test ar724x and later chips to ensure no regressions have occured.
jmmv [Sun, 16 Mar 2014 08:04:06 +0000 (08:04 +0000)]
Migrate most of tools/regression/usr.bin/ to the new tests layout.
I'm starting with the easy cases. The leftovers need to be looked at a
bit more closely.
Note that this change _does_ modify the code of the old tests. This is
required in order to allow the code to locate the data files in the
source directory instead of the current directory, because Kyua
automatically changes the latter to a temporary directory.
Also note that at least one test is known to be broken here. Actually,
the test is not really broken: it's marked as a TODO but unfortunately
Kyua's TAP parser currently does not understand that. Will have to be
fixed separately.
jmmv [Sun, 16 Mar 2014 04:09:22 +0000 (04:09 +0000)]
Migrate tools/regression/{usr.bin/lastcomm,usr.sbin}/ to the new tests layout.
This change was originally going to only migrate the usr.sbin tests but, as
it turns out, the usr.sbin/sa/ tests require files from usr.bin/lastcomm/
so it's better to just also migrate the latter at the same time. The other
usr.bin tests will be moved separately.
To make these tests work within the test suite, some of them have required
changes to prevent modifying the source directory and instead just rely on
the current directory for file manipulation.
adrian [Sun, 16 Mar 2014 02:41:47 +0000 (02:41 +0000)]
Handle the case where both arge0 and arge1 MAC addresses are available via
'eeprommac'.
The existing driver would just make arge units past 0 take the primary
MAC and increment it by the unit number, without correct address wrapping.
That has to be fixed at a later date.
jmg [Sun, 16 Mar 2014 01:43:23 +0000 (01:43 +0000)]
replace the kernel's version w/ cperciva's implementation... In all
my tests, it is faster ~20%, even on an old IXP425 533MHz it is ~45%
faster... This is partly due to loop unrolling, so the code size does
significantly increase... I do plan on committing a version that
rolls up the loops again for smaller code size for embedded systems
where size is more important than absolute performance (it'll save ~6k
code)...
The kernel implementation is now shared w/ userland's libcrypt and
libmd...
We drop support for sha256 from sha2.c, so now sha2.c only contains
sha384 and sha512...
jmg [Sun, 16 Mar 2014 00:53:40 +0000 (00:53 +0000)]
change td_retval into a union w/ off_t, with defines to mask the
change... This eliminates a cast, and also forces td_retval
(often 2 32-bit registers) to be aligned so that off_t's can be
stored there on arches with strict alignment requirements like
armeb (AVILA)... On i386, this doesn't change alignment, and on
amd64 it doesn't either, as register_t is already 64bits...
This will also prevent future breakage due to people adding additional
fields to the struct...
tychon [Sat, 15 Mar 2014 23:09:34 +0000 (23:09 +0000)]
Fix a race wherein the source of an interrupt vector is wrongly
attributed if an ExtINT arrives during interrupt injection.
Also, fix a spurious interrupt if the PIC tries to raise an interrupt
before the outstanding one is accepted.
Finally, improve the PIC interrupt latency when another interrupt is
raised immediately after the outstanding one is accepted by creating a
vmexit rather than waiting for one to occur by happenstance.
jilles [Sat, 15 Mar 2014 14:58:48 +0000 (14:58 +0000)]
sh: Allow kill %job on jobs started without job control.
When killing a %job started without job control, kill all processes in it.
As with process groups and zombies, if any process in the job can be killed
or has already terminated, the command is successful.
This also fixes occasional failures of the builtins/kill1.0 test.
rwatson [Sat, 15 Mar 2014 00:57:50 +0000 (00:57 +0000)]
Several years after initial development, merge prototype support for
linking NIC Receive Side Scaling (RSS) to the network stack's
connection-group implementation. This prototype (and derived patches)
are in use at Juniper and several other FreeBSD-using companies, so
despite some reservations about its maturity, merge the patch to the
base tree so that it can be iteratively refined in collaboration rather
than maintained as a set of gradually diverging patch sets.
(1) Merge a software implementation of the Toeplitz hash specified in
RSS implemented by David Malone. This is used to allow suitable
pcbgroup placement of connections before the first packet is
received from the NIC. Software hashing is generally avoided,
however, due to high cost of the hash on general-purpose CPUs.
(2) In in_rss.c, maintain authoritative versions of RSS state intended
to be pushed to each NIC, including keying material, hash
algorithm/ configuration, and buckets. Provide software-facing
interfaces to hash 2- and 4-tuples for IPv4 and IPv6 using both
the RSS standardised Toeplitz and a 'naive' variation with a hash
efficient in software but with poor distribution properties.
Implement rss_m2cpuid()to be used by netisr and other load
balancing code to look up the CPU on which an mbuf should be
processed.
(3) In the Ethernet link layer, allow netisr distribution using RSS as
a source of policy as an alternative to source ordering; continue
to default to direct dispatch (i.e., don't try and requeue packets
for processing on the 'right' CPU if they arrive in a directly
dispatchable context).
(4) Allow RSS to control tuning of connection groups in order to align
groups with RSS buckets. If a packet arrives on a protocol using
connection groups, and contains a suitable hardware-generated
hash, use that hash value to select the connection group for pcb
lookup for both IPv4 and IPv6. If no hardware-generated Toeplitz
hash is available, we fall back on regular PCB lookup risking
contention rather than pay the cost of Toeplitz in software --
this is a less scalable but, at my last measurement, faster
approach. As core counts go up, we may want to revise this
strategy despite CPU overhead.
Where device drivers suitably configure NICs, and connection groups /
RSS are enabled, this should avoid both lock and line contention during
connection lookup for TCP. This commit does not modify any device
drivers to tune device RSS configuration to the global RSS
configuration; patches are in circulation to do this for at least
Chelsio T3 and Intel 1G/10G drivers. Currently, the KPI for device
drivers is not particularly robust, nor aware of more advanced features
such as runtime reconfiguration/rebalancing. This will hopefully prove
a useful starting point for refinement.
No MFC is scheduled as we will first want to nail down a more mature
and maintainable KPI/KBI for device drivers.
neel [Fri, 14 Mar 2014 22:07:08 +0000 (22:07 +0000)]
Don't dump entries that were modified during the time the KTR buffer was being
copied to userspace. Failing to do this would result in entries at the bottom
of the ktrdump output to be more recent than entries at the top.
With this change the timestamps are monotonically decreasing going from the
top to the bottom of the ktrdump output.
neel [Fri, 14 Mar 2014 21:35:16 +0000 (21:35 +0000)]
Fix an issue with ktrdump(8) where it would not print all entries in the
KTR buffer.
This happens when 'i' tries to wrap around from 0 to 'entries - 1'. Since 'i'
is a signed integer the modulo operation actually returns a negative number.
Fix this by computing the next index to use "by hand" instead of relying
on the modulo operator.
imp [Fri, 14 Mar 2014 20:20:32 +0000 (20:20 +0000)]
NanoBSD has a utility shell script called save_cfg which helps keep
/cfg updated with the modified configuration files in /etc. I have
written an improved version with the following features:
* Recurses directories.
* Only requires file arguments the first time the file/directory is
* added to /cfg.
* Handles file deletions.
PR: 145962, 157533
Submitted by: Aragon Gouveia and Alex Bakhtin