]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
6 years agoInvent new LDR_INTERP for the loader interpreter to use. Use this in
imp [Fri, 2 Feb 2018 15:40:49 +0000 (15:40 +0000)]
Invent new LDR_INTERP for the loader interpreter to use. Use this in
preference to LIBFICL{,32}. LIBFICL{,32} are now always defined, but
LDR_INTERP{,32} is defined empty when building w/o forth (aka the
simple interpreter) and defined to LIBFICL{,32} when we are building
forth.

6 years agoNow that we no longer conditionally compile some files outside of ficl
imp [Fri, 2 Feb 2018 15:01:54 +0000 (15:01 +0000)]
Now that we no longer conditionally compile some files outside of ficl
with BOOT_FORTH, retire it from here.

6 years agoRemove pcibios forth support.
imp [Fri, 2 Feb 2018 15:01:49 +0000 (15:01 +0000)]
Remove pcibios forth support.

I had thought that this would be useful. However it was committed too
late, and wound up being unused. It's in the way of future work now,
so retire it rather than bring it forward.

6 years agoThese 4th words were an attempt to allow integration into the boot
imp [Fri, 2 Feb 2018 15:01:44 +0000 (15:01 +0000)]
These 4th words were an attempt to allow integration into the boot
loader scripts. However, that path won't be taken after all it
seems. Remove this code before it decays into uselessness. Also remove
build dependencies on forth no longer needed.

6 years agoRetire pnp.4th and the code needed only for 4th words used here.
imp [Fri, 2 Feb 2018 15:01:33 +0000 (15:01 +0000)]
Retire pnp.4th and the code needed only for 4th words used here.

This has never been installed. It was added to the tree disconnected
to the build in FreeBSD 5 (17 years ago) and has never been used as
far as I can tell. The desired improvements never really happened
(despite a couple minor cleanups along the way). It's relevance is
long past, so better to retire it.

6 years agoZFS ARC: restore illumos uses of 'needfree' that were removed in r325851
avg [Fri, 2 Feb 2018 12:57:33 +0000 (12:57 +0000)]
ZFS ARC: restore illumos uses of 'needfree' that were removed in r325851

This is purely a cosmetic change to have a more complete copy of
ifdef-ed out illumos code.

MFC after: 1 week

6 years agoSlightly bump the maximum OID path for loading tunable SYSCTLs.
hselasky [Fri, 2 Feb 2018 12:42:46 +0000 (12:42 +0000)]
Slightly bump the maximum OID path for loading tunable SYSCTLs.

Coming updates to the mlx5en(4) driver will require this.

MFC after: 1 week
Sponsored by: Mellanox Technologies

6 years agoOn pageout, in vnode generic pager, for partially dirty page, only
kib [Fri, 2 Feb 2018 11:56:30 +0000 (11:56 +0000)]
On pageout, in vnode generic pager, for partially dirty page, only
clear dirty bits for completely invalid blocks.

Otherwise we might not write out the last chunk that is shorter than
512 bytes, if the file end is not aligned on disk block boundary.
This become important after the r324794.

PR: 225586
Reported by: tris_vern@hotmail.com
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 3 days

6 years agoMerge r1.120 from NetBSD:
ae [Fri, 2 Feb 2018 07:39:34 +0000 (07:39 +0000)]
Merge r1.120 from NetBSD:
  Fix a pretty simple, yet pretty tragic typo: we should return IPPROTO_DONE,
  not IPPROTO_NONE. With IPPROTO_NONE we will keep parsing the header chain
  on an mbuf that was already freed.

Reported by: Maxime Villard <max at m00nbsd dot net>
MFC after: 3 days

6 years agoCentralize several variables.
imp [Fri, 2 Feb 2018 06:32:26 +0000 (06:32 +0000)]
Centralize several variables.

MK_CTF, MK_SSP, MK_PROFILE, NO_PIC, and INTERNALLIB are always the
same, so set them in defs.mk. MAN= is common, so set it here too.
This removes a lot of boring repetition from the Makefiles that added
almost no value.

6 years agoThe routines defined in comparedf2 and comparesf2 are defined in libc
imp [Fri, 2 Feb 2018 05:04:43 +0000 (05:04 +0000)]
The routines defined in comparedf2 and comparesf2 are defined in libc
for softfloat mips. No need to include them here.

6 years agoCorrect longjmp
swills [Fri, 2 Feb 2018 02:28:25 +0000 (02:28 +0000)]
Correct longjmp

Reviewed by: nwhitehorn
Differential Revision: https://reviews.freebsd.org/D14159

6 years ago[etherswitch] add the first pass of a simple API to flush and fetch the L2 address...
adrian [Fri, 2 Feb 2018 02:05:14 +0000 (02:05 +0000)]
[etherswitch] add the first pass of a simple API to flush and fetch the L2 address table from the ethernet switch.

This stuff may be a bit fluid during this -HEAD cycle as various other
switch features are added, but the current stuff is enough to drive
initial development and features on the atheros range of integrated
and external switches.

* add a method to flush the whole address table;
* add a method to flush all addresses on a given port;
* add a method to download the address table;
* .. and then a method to fetch entries from the address table.

The table fetch/read methods pass through to the drivers for now since
the drivers may implement different ways of fetching/caching the address
table data.  The atheros devices for example fetch the table by
iterating over the table through a set of registers and so you need
to keep that locked whilst you iterate otherwise you may have the table
flushed half way by a port status change.

This is a no-op until the userland and arswitch code shows up.

6 years agoUse sbput(3) rather than sbwrite(3) to ensure that the updated copy of
mckusick [Fri, 2 Feb 2018 00:07:38 +0000 (00:07 +0000)]
Use sbput(3) rather than sbwrite(3) to ensure that the updated copy of
the superblock gets written.

Reported by: Mark Johnston <markj@FreeBSD.org>

6 years agoFix punctuation.
mckusick [Fri, 2 Feb 2018 00:01:50 +0000 (00:01 +0000)]
Fix punctuation.

6 years ago[atheros] Update QCA953x support to use the new hints.
adrian [Thu, 1 Feb 2018 22:01:53 +0000 (22:01 +0000)]
[atheros] Update QCA953x support to use the new hints.

6 years ago[atheros] Fix DIR-825C1 to use the new hints.
adrian [Thu, 1 Feb 2018 22:01:11 +0000 (22:01 +0000)]
[atheros] Fix DIR-825C1 to use the new hints.

Tested:

* DIR-825C1

6 years ago[atheros] teach these two boards about the new hints location as well.
adrian [Thu, 1 Feb 2018 22:00:38 +0000 (22:00 +0000)]
[atheros] teach these two boards about the new hints location as well.

6 years ago[atheros] Teach the QCA955x SoC code about the new hints stuff.
adrian [Thu, 1 Feb 2018 22:00:05 +0000 (22:00 +0000)]
[atheros] Teach the QCA955x SoC code about the new hints stuff.

6 years ago[atheros] Fix-up the base address stuff after I did a drive-by with the calibration...
adrian [Thu, 1 Feb 2018 21:58:52 +0000 (21:58 +0000)]
[atheros] Fix-up the base address stuff after I did a drive-by with the calibration data location.

The old way required the data to be present really early and copied it from
memory mapped NOR flash; this only worked during kernel boot but not for
ath/ath_hal modules.

Tested:

* AR9331, Carambola2, ath/hal modules.

6 years agoUpgrade our copies of clang, llvm, lld, lldb, compiler-rt and libc++ to
dim [Thu, 1 Feb 2018 21:41:15 +0000 (21:41 +0000)]
Upgrade our copies of clang, llvm, lld, lldb, compiler-rt and libc++ to
6.0.0 (branches/release_60 r323948).

MFC after: 3 months
X-MFC-With: r327952
PR: 224669

6 years agoFix some recent regressions after r328436 in the LinuxKPI:
hselasky [Thu, 1 Feb 2018 19:57:21 +0000 (19:57 +0000)]
Fix some recent regressions after r328436 in the LinuxKPI:

1) The OPW() function macro should have the same return type like the
function it executes.
2) The DEVFS I/O-limit should be enforced for all character device reads
and writes.
3) The character device file handle should be passable, same as for
DEVFS based file handles.

Reported by: jbeich @
MFC after: 1 week
Sponsored by: Mellanox Technologies

6 years agoAdd -n <network device emulation type> to usage synopsis
rgrimes [Thu, 1 Feb 2018 19:39:33 +0000 (19:39 +0000)]
Add -n <network device emulation type> to usage synopsis
forgotten in r328695.

Fix long lines in usage.
Consistently use "default:" rather than "default is"

Approved by: grehan (mentor)
MFC after: 3 days

6 years agoAdd option '-n' network device emulation type which has a default of virtio-net
rgrimes [Thu, 1 Feb 2018 17:59:14 +0000 (17:59 +0000)]
Add option '-n' network device emulation type which has a default of virtio-net

PR: 225558
Submitted by: Johannes Lundberg <johalun0@gmail.com>
Approved by: grehan (mentor)
MFC after: 3 days

6 years agoMake sure the LinuxKPI's internal ERESTARTSYS error code gets translated
hselasky [Thu, 1 Feb 2018 17:32:45 +0000 (17:32 +0000)]
Make sure the LinuxKPI's internal ERESTARTSYS error code gets translated
into ERESTART for mmap and page fault calls aswell.

MFC after: 1 week
Sponsored by: Mellanox Technologies

6 years agoDisable EARLY_PRINTF from the Armada 3700 uart, it breaks when we want
andrew [Thu, 1 Feb 2018 15:05:17 +0000 (15:05 +0000)]
Disable EARLY_PRINTF from the Armada 3700 uart, it breaks when we want
to use EARLY_PRINTF on other SoCs.

Sponsored by: DARPA, AFRL

6 years agoD14130: stand/fdt: Rip out FDT VA tracking
kevans [Thu, 1 Feb 2018 14:31:12 +0000 (14:31 +0000)]
D14130: stand/fdt: Rip out FDT VA tracking

Whether we should be overwriting the loaded FDT module with the 'fixed up'
version or not was questionable when this was added, and now that overlays
are possible this is downright wrong.

Overlays can increase the size of the blob, so writing it back to the
original VA will generally write past the end of the block and start
clobbering other things in memory.

Rip it out- it was questionable to begin with, it's doing bad things now,
and it serves no purpose since the modified blob will be copied into place
rather than relying on this to reflect the changes.

Reviewed by: gonzo
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D14130

6 years agoOnly promote userspace mappings to superpages. This was dropped in r328510,
andrew [Thu, 1 Feb 2018 14:26:26 +0000 (14:26 +0000)]
Only promote userspace mappings to superpages. This was dropped in r328510,
however due to the break-before-make requirement on arm64 is is currently
unsafe to promote kernel pages.

Sponsored by: DARPA, AFRL

6 years agopf: Avoid warning without INVARIANTS
kp [Thu, 1 Feb 2018 07:52:06 +0000 (07:52 +0000)]
pf: Avoid warning without INVARIANTS

When INVARIANTS is not set the 'last' variable is not used, which can generate
compiler warnings.
If this invariant is ever violated it'd result in a KASSERT failure in
refcount_release(), so this one is not strictly required.

6 years agoChange the default MSR values used when starting userland and kernel
nwhitehorn [Thu, 1 Feb 2018 05:31:24 +0000 (05:31 +0000)]
Change the default MSR values used when starting userland and kernel
threads from compile-time defines to global variables. This removes a
significant amount of duplicated runtime patches to the compile-time
defines, centralizing the conditional logic in the early startup code.

Reviewed by: jhibbits

6 years agoFix build on 32-bit PowerPC, broken in r328537.
nwhitehorn [Thu, 1 Feb 2018 05:28:02 +0000 (05:28 +0000)]
Fix build on 32-bit PowerPC, broken in r328537.

6 years agoNull out journal softc pointer earlier to avoid a segment fault
mckusick [Wed, 31 Jan 2018 23:30:49 +0000 (23:30 +0000)]
Null out journal softc pointer earlier to avoid a segment fault
that can otherwise occur.

PR:           221804
Submitted by: Andreas Longwitz <longwitz at incore.de>
MFC after:    1 week

6 years agoBump .Dd
imp [Wed, 31 Jan 2018 23:16:54 +0000 (23:16 +0000)]
Bump .Dd

6 years agoAdd the new armv7 architecture.
imp [Wed, 31 Jan 2018 23:16:19 +0000 (23:16 +0000)]
Add the new armv7 architecture.

6 years agoWhen reading a cylinder group, break out reporting of check hash errors
mckusick [Wed, 31 Jan 2018 23:13:37 +0000 (23:13 +0000)]
When reading a cylinder group, break out reporting of check hash errors
from other types of errors so that the error is correctly reported.

6 years agoOne of the vnode fields listed by vn_printf is the union of pointers
mckusick [Wed, 31 Jan 2018 22:49:50 +0000 (22:49 +0000)]
One of the vnode fields listed by vn_printf is the union of pointers
whose type depends on the type of vnode. Correct vn_printf so that
it correctly identifies the name of the pointer that it is printing.

Submitted by: Andreas Longwitz <longwitz at incore.de>
MFC after: 1 week

6 years agoBreak out the interpreters (simple and forth) w/o ifdefs. This is
imp [Wed, 31 Jan 2018 22:46:05 +0000 (22:46 +0000)]
Break out the interpreters (simple and forth) w/o ifdefs.  This is
akin to what Pedro Souza and Wojciech Koszek did in the lua GSoC with
interp.h, interp_simple.c and changes to interp.c and interp_forth.c,
but completely redone from scratch.

This effectively restores the spirit of r326712 (my first attempt to
bring in Pedro's and Wojciech's work) updated for new requirements
that had silently broke their original work.  This change also differs
by using fixed function names instead of function pointers to simply
things. Only one interpreter at a time may be compiled in.

Also of note: we take a mutable string, pass it in via a const char *
pointer into intrp_forth's interp_run(). We then cast away the const
to pass into ficlExec since ficl would require extensive changes to
properly const-poison. See Sections 6.5.2.5 and 6.7.3 of C11 standard
noting it's only UB if you modify a const object through a non-const
pointer, but not char [] -> const char * -> char * as here.

6 years agodtc(1): Revert WARNS change from r328173
kevans [Wed, 31 Jan 2018 22:20:33 +0000 (22:20 +0000)]
dtc(1): Revert WARNS change from r328173

WARNS > 3 breaks xtoolchain builds to varying degrees. Revert it.

Reported by: lwhsu, jhb
MFC after: 3 days

6 years agopsm: Add a kludge to support 0x46 identity middle byte Synaptics touchpads
wulf [Wed, 31 Jan 2018 22:17:52 +0000 (22:17 +0000)]
psm: Add a kludge to support 0x46 identity middle byte Synaptics touchpads

Most synaptics touchpads return 0x47 in middle byte in responce to identify
command as stated in p.4.4 of "Synaptics PS/2 TouchPad Interfacing Guide".
But some devices e.g. found on HP EliteBook 9470m return 0x46 here.
Allow them to be identified as Synaptics as well as 0x47.
ExtendedQueries return incorrect data on such a touchpads so we ignore
their result and set conservative defaults.

PR: 222667
Reported by: Neel Chauhan <neel@neelc.org>
Tested by: Neel Chauhan <neel@neelc.org>
Approved by: gonzo

6 years agoAccount for the fact that jemalloc 5.0.0 dropped STATIC_PAGE_SHIFT
marius [Wed, 31 Jan 2018 21:56:23 +0000 (21:56 +0000)]
Account for the fact that jemalloc 5.0.0 dropped STATIC_PAGE_SHIFT
in favor for using LG_PAGE directly and, thus, for the fact that
host and target don't necessarily use pages of the same sizes.

Approved by: jasone

6 years agopsm(4): Reduce psm watchdog verbosity
wulf [Wed, 31 Jan 2018 21:46:37 +0000 (21:46 +0000)]
psm(4): Reduce psm watchdog verbosity

Modern touchpads do not issue interrupts on inactivity so "lost interrupt"
message became annoying spam nowadays. This change quiets the message
if debug.psm.loglevel=5 (or less) is set in /boot/loader.conf

Approved by: gonzo

6 years agoFix build by escaping a line break.
gjb [Wed, 31 Jan 2018 21:41:42 +0000 (21:41 +0000)]
Fix build by escaping a line break.

PR: 225597
Submitted by: cbnfinley at gmail.com
Sponsored by: The FreeBSD Foundation

6 years agopsm(4): Add support for HP EliteBook 1040 ForcePads.
wulf [Wed, 31 Jan 2018 21:14:59 +0000 (21:14 +0000)]
psm(4): Add support for HP EliteBook 1040 ForcePads.

ForcePads do not have any physical buttons, instead they detect click
based on finger pressure. Forcepads erroneously report button click
if there are 2 or more fingers on the touchpad breaking multifinger
gestures. To workaround this start reporting a click only after
4 consecutive single touch packets has been received. Skip these packets
in case more contacts appear.

PR: 223369
Reported by: Neel Chauhan <neel@neelc.org>
Tested by: Neel Chauhan <neel@neelc.org>
Reviewed by: gonzo
Approved by: gonzo

6 years agoDon't include long double routines on architectures with small long double.
jhb [Wed, 31 Jan 2018 18:13:33 +0000 (18:13 +0000)]
Don't include long double routines on architectures with small long double.

Reviewed by: emaste
Sponsored by: DARPA / AFRL
Differential Revision: https://reviews.freebsd.org/D13874

6 years agoUpdate limits on makecontext() arguments in the setcontext_link test.
jhb [Wed, 31 Jan 2018 18:03:40 +0000 (18:03 +0000)]
Update limits on makecontext() arguments in the setcontext_link test.

sparc64 and riscv do not support 10 arguments, but MIPS now does.
While here, combine clauses for architectures that support the same
number of arguments to reduce duplication.

Sponsored by: DARPA / AFRL

6 years agoAdd a new set of simple tests for makecontext().
jhb [Wed, 31 Jan 2018 18:02:02 +0000 (18:02 +0000)]
Add a new set of simple tests for makecontext().

In contrast to the existing NetBSD setcontext_link test, these tests
verify that passing from 1 to 6 arguments through to the callback function
work correctly which can be useful for testing ABIs which split arguments
between registers and the stack.

Sponsored by: DARPA / AFRL

6 years agoRemove limitation of 6 arguments for makecontext() on mips.
jhb [Wed, 31 Jan 2018 18:00:23 +0000 (18:00 +0000)]
Remove limitation of 6 arguments for makecontext() on mips.

This implementation spills additional arguments on the stack so works
fine with more than 6 arguments.  I believe the check was just copied
over from sparc64 (which doesn't support spilling onto the stack)

Sponsored by: DARPA / AFRL

6 years agoRemove bogus checks against NCARGS.
jhb [Wed, 31 Jan 2018 17:57:59 +0000 (17:57 +0000)]
Remove bogus checks against NCARGS.

NCARGS isn't a limit on the number of arguments to pass to a function,
but the number of bytes that can be consumed by arguments to exec.  As
such, it is not suitable for a limit on the count of arguments passed
to makecontext().

Sponsored by: DARPA / AFRL

6 years agoClarify that the additional arguments to makecontext() are of type int.
jhb [Wed, 31 Jan 2018 17:56:36 +0000 (17:56 +0000)]
Clarify that the additional arguments to makecontext() are of type int.

MFC after: 1 week
Sponsored by: DARPA / AFRL

6 years agoConsistently use 16-byte alignment for MIPS N32 and N64.
jhb [Wed, 31 Jan 2018 17:36:39 +0000 (17:36 +0000)]
Consistently use 16-byte alignment for MIPS N32 and N64.

- Add a new <machine/abi.h> header to hold constants shared between C
  and assembly such as CALLFRAME_SZ.
- Add a new STACK_ALIGN constant to <machine/abi.h> and use it to
  replace hardcoded constants in the kernel and makecontext().  As a
  result of this, ensure the stack pointer on N32 and N64 is 16-byte
  aligned for N32 and N64 after exec(), after pthread_create(), and
  when sending signals rather than 8-byte aligned.

Reviewed by: jmallett
Sponsored by: DARPA / AFRL
Differential Revision: https://reviews.freebsd.org/D13875

6 years agoWhen switching IBRS on, also enable STIBP (Single Thread Indirect
kib [Wed, 31 Jan 2018 16:56:02 +0000 (16:56 +0000)]
When switching IBRS on, also enable STIBP (Single Thread Indirect
Branch Predictors) mitigation.

DOcument 336996-001 promises that CPUs which implement IBRS but not
STIBP silently ignore setting of the bit instead of trapping.

Sponsored by: The FreeBSD Foundation
MFC after: 1 week

6 years agoExpand IBRS TLA in sysctl help lines.
kib [Wed, 31 Jan 2018 16:54:05 +0000 (16:54 +0000)]
Expand IBRS TLA in sysctl help lines.

Requested by: bz
Sponsored by: The FreeBSD Foundation
MFC after: 1 week

6 years agozfs_rezget: drop cached pages before doing anything else
avg [Wed, 31 Jan 2018 14:44:51 +0000 (14:44 +0000)]
zfs_rezget: drop cached pages before doing anything else

We did that in the case of success to prevent the use of stale cached
data, but it makes even less sense to keep the cached data when we fail.

Ideally, we should call vgone() on the vnode in the case of zfs_rezget
failure, but the current lock order prevents us from doing that.

The change also rearranges the order of unlinked check and the size
change check.

While there, add missing SET_ERROR in one of the error paths.

MFC after: 2 weeks

6 years agoIBRS support, AKA Spectre hardware mitigation.
kib [Wed, 31 Jan 2018 14:36:27 +0000 (14:36 +0000)]
IBRS support, AKA Spectre hardware mitigation.

It is coded according to the Intel document 336996-001, reading of the
patches posted on lkml, and some additional consultations with Intel.

For existing processors, you need a microcode update which adds IBRS
CPU features, and to manually enable it by setting the tunable/sysctl
hw.ibrs_disable to 0.  Current status can be checked in sysctl
hw.ibrs_active.  The mitigation might be inactive if the CPU feature
is not patched in, or if CPU reports that IBRS use is not required, by
IA32_ARCH_CAP_IBRS_ALL bit.

Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D14029

6 years agoDo not enable PTI when IA32_ARCH_CAP_RDCL_NO bit is set.
kib [Wed, 31 Jan 2018 14:25:42 +0000 (14:25 +0000)]
Do not enable PTI when IA32_ARCH_CAP_RDCL_NO bit is set.

Intel document 336996-001 claims that this will be the way to inform
about Meltdown correction.

Sponsored by: The FreeBSD Foundation
MFC after: 1 week

6 years agoProperly implement the cond_resched() function macro in the LinuxKPI.
hselasky [Wed, 31 Jan 2018 13:40:36 +0000 (13:40 +0000)]
Properly implement the cond_resched() function macro in the LinuxKPI.

MFC after: 1 week
Sponsored by: Mellanox Technologies

6 years agovmm/svm: post LAPIC interrupts using event injection, not virtual interrupts
avg [Wed, 31 Jan 2018 11:14:26 +0000 (11:14 +0000)]
vmm/svm: post LAPIC interrupts using event injection, not virtual interrupts

The virtual interrupt method uses V_IRQ, V_INTR_PRIO, and V_INTR_VECTOR
fields of VMCB to inject a virtual interrupt into a guest VM.  This
method has many advantages over the direct event injection as it
offloads all decisions of whether and when the interrupt can be
delivered to the guest.  But with a purely software emulated vAPIC the
advantage is also a problem.  The problem is that the hypervisor does
not have any precise control over when the interrupt is actually
delivered to the guest (or a notification about that).  Because of that
the hypervisor cannot update the interrupt vector in IRR and ISR in the
same way as real hardware would.  The hypervisor becomes aware that the
interrupt is being serviced only upon the first VMEXIT after the
interrupt is delivered.  This creates a window between the actual
interrupt delivery and the update of IRR and ISR.  That means that IRR
and ISR might not be correctly set up to the point of the
end-of-interrupt signal.

The described deviation has been observed to cause an interrupt loss in
the following scenario.  vCPU0 posts an inter-processor interrupt to
vCPU1.  The interrupt is injected as a virtual interrupt by the
hypervisor.  The interrupt is delivered to a guest and an interrupt
handler is invoked.  The handler performs a requested action and
acknowledges the request by modifying a global variable.  So far, there
is no VMEXIT and the hypervisor is unaware of the events.  Then, vCPU0
notices the acknowledgment and sends another IPI with the same vector.
The IPI gets collapsed into the previous IPI in the IRR of vCPU1.  Only
after that a VMEXIT of vCPU1 occurs.  At that time the vector is cleared
in the IRR and is set in the ISR.  vCPU1 has vAPIC state as if the
second IPI has never been sent.
The scenario is impossible on the real hardware because IRR and ISR are
updated just before the interrupt handler gets started.

I saw several possibilities of fixing the problem.  One is to intercept
the virtual interrupt delivery to update IRR and ISR at the right
moment.  The other is to deliver the LAPIC interrupts using the event
injection, same as legacy interrupts.  I opted to use the latter
approach for several reasons.  It's equivalent to what VMM/Intel does
(in !VMX case).  It appears to be what VirtualBox and KVM do.  The code
is already there (to support legacy interrupts).

Another possibility was to use a special intermediate state for a vector
after it is injected using a virtual interrupt and before it is known
whether it was accepted or is still pending.
That approach was implemented in https://reviews.freebsd.org/D13828
That method is more complex and does not have any clear advantage.

Please see sections 15.20 and 15.21.4 of "AMD64 Architecture
Programmer's Manual Volume 2: System Programming" (publication 24593,
revision 3.29) for comparison between event injection and virtual
interrupt injection.

PR: 215972
Reported by: ajschot@hotmail.com, grehan
Tested by: anish, grehan,  Nils Beyer <nbe@renzel.net>
Reviewed by: anish, grehan
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D13780

6 years ago[arswitch] Fix ATU programming on the AR8327 switch.
adrian [Wed, 31 Jan 2018 07:37:33 +0000 (07:37 +0000)]
[arswitch] Fix ATU programming on the AR8327 switch.

Doing a flush actually requires setting this bit.

6 years ago[arswitch] Fix ATU flushing on AR8216/AR8316 and most of the later chips.
adrian [Wed, 31 Jan 2018 07:36:51 +0000 (07:36 +0000)]
[arswitch] Fix ATU flushing on AR8216/AR8316 and most of the later chips.

The switch hardware requires this bit to be set in order to kick start the
actual ATU update.  This was being masked on some chips by the learning
programming (what to do when a MAC address moves, hash table collision, etc)
which is currently inconsistent between chips.

Tested:

* AR9344 SoC (AR7240 style switch internal)

6 years ago[arswitch] add a new debug section for upcoming address table management.
adrian [Wed, 31 Jan 2018 07:20:34 +0000 (07:20 +0000)]
[arswitch] add a new debug section for upcoming address table management.

6 years agoPowerNV: fix compilation on non-NV platforms
wma [Wed, 31 Jan 2018 06:42:01 +0000 (06:42 +0000)]
PowerNV: fix compilation on non-NV platforms

Submitted by:          Wojciech Macek <wma@semihalf.com>
Obtained from:         Semihalf
Sponsored by:          IBM, QCM Technologies

6 years agoUpdate stand.h for changes for strto*l
imp [Wed, 31 Jan 2018 05:07:43 +0000 (05:07 +0000)]
Update stand.h for changes for strto*l

Move prototypes to proper section now that we don't have modified
versions of strtol and strtoul in libsa. Add prototypes for new
strtoll and strtoull. Use prototypes copied from stdlib.h instead of
the old hand-rolled ones.

(I forgot to move this file form my lua branch in r328613)

6 years agoMove libstand.3 to libsa.3. Update libsa.3 to include functions
imp [Wed, 31 Jan 2018 04:29:05 +0000 (04:29 +0000)]
Move libstand.3 to libsa.3. Update libsa.3 to include functions
recently added. More are likely missing.

6 years agoKill copies of strtol and strtoul. Use the ones that are in libc,
imp [Wed, 31 Jan 2018 04:29:00 +0000 (04:29 +0000)]
Kill copies of strtol and strtoul. Use the ones that are in libc,
since they suffice. Create xlocale_private.h which provides the most
minimal locale implementation we can get away with. Add strtoll and
strtoull from libc.

6 years agoMove strtold wrapper from strtol.c to its own strtold.c. This code
imp [Wed, 31 Jan 2018 03:05:14 +0000 (03:05 +0000)]
Move strtold wrapper from strtol.c to its own strtold.c.  This code
was written by theraven@ (David Chisnall) entirely, there's no
original Berkeley code left here so just copy his copyright over.

6 years agoTry to preallocate receive memory early.
mav [Wed, 31 Jan 2018 01:04:36 +0000 (01:04 +0000)]
Try to preallocate receive memory early.

We may not have enough contiguous memory later, when NTB connection get
established.  It is quite likely that NTB windows are symmetric and this
allocation remain, but even if not, we will just reallocate it later.

MFC after: 2 weeks

6 years agoEnsure 'name' is not NULL before passing to strcmp().
jhb [Tue, 30 Jan 2018 23:29:27 +0000 (23:29 +0000)]
Ensure 'name' is not NULL before passing to strcmp().

This avoids a nested page fault when obtaining a stack trace in DDB if
the address from the first frame does not resolve to a known symbol.

MFC after: 1 week
Sponsored by: Chelsio Communications

6 years agoExport tcp_always_keepalive for use by the Chelsio TOM module.
jhb [Tue, 30 Jan 2018 23:01:37 +0000 (23:01 +0000)]
Export tcp_always_keepalive for use by the Chelsio TOM module.

This used to work by accident with ld.bfd even though always_keepalive
was marked as static. LLD honors static more correctly, so export this
variable properly (including moving it into the tcp_* namespace).

Reviewed by: bz, emaste
MFC after: 2 weeks
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D14129

6 years agozfsd: Don't spare a vdev that's being replaced
asomers [Tue, 30 Jan 2018 21:25:43 +0000 (21:25 +0000)]
zfsd: Don't spare a vdev that's being replaced

If a zfs pool contains a replacing vdev (either created manually by "zpool
replace" or by zfsd(8) via autoreplace by physical path) and then new spares
get added to the pool, zfsd shouldn't use one to replace the drive that is
already being replaced.  That's a waste of resources that just slows down
the rebuild.

PR: 225547
MFC after: 3 weeks
Sponsored by: Spectra Logic Corp

6 years agoAdd missing non-POWERPC case to give the scr value something non-zero.
sbruno [Tue, 30 Jan 2018 20:00:12 +0000 (20:00 +0000)]
Add missing non-POWERPC case to give the scr value something non-zero.

This fixes the instant reboot of netbooting after r328536 on x86 systems.

Reviewed by: peter
Sponsored by: Limelight Networks

6 years agomakesyscalls: permit a range of syscall numbers for UNIMPL
emaste [Tue, 30 Jan 2018 18:29:38 +0000 (18:29 +0000)]
makesyscalls: permit a range of syscall numbers for UNIMPL

Some ABIs have large gaps in syscall numbers.  Allow gaps to be filled
as ranges of UNIMPL, with an entry like:

248-1023 AUE_NULL UNIMPL unimplemented

Reviewed by: jhb, gnn
Sponsored by: Turing Robotic Industries Inc.
Differential Revision: https://reviews.freebsd.org/D14122

6 years agoPull in r322131 from upstream llvm trunk (by Rafael Espíndola):
emaste [Tue, 30 Jan 2018 16:43:20 +0000 (16:43 +0000)]
Pull in r322131 from upstream llvm trunk (by Rafael Espíndola):

  Use a MCExpr for the size of MCFillFragment.

  This allows the size to be found during ralaxation. This fixes
  [LLVM] pr35858.

Requested by: royger

6 years agoPull in r322123 from upstream llvm trunk (by Rafael Espíndola):
emaste [Tue, 30 Jan 2018 16:42:08 +0000 (16:42 +0000)]
Pull in r322123 from upstream llvm trunk (by Rafael Espíndola):

  Don't create MCFillFragment directly.

  Instead use higher level APIs that take care of most bookkeeping.

6 years agoPull in r322108 from upstream llvm trunk (by Rafael Espíndola):
emaste [Tue, 30 Jan 2018 16:41:38 +0000 (16:41 +0000)]
Pull in r322108 from upstream llvm trunk (by Rafael Espíndola):

  Make one of the emitFill methods non virtual. NFC.

  This is just preparatory work to fix [LLVM] PR35858.

6 years agoChange installer default to not install ports tree
swills [Tue, 30 Jan 2018 16:34:56 +0000 (16:34 +0000)]
Change installer default to not install ports tree

Reviewed by: gjb, dteske, allanjude, bdrewery, mat
Approved by: gjb
Differential Revision: https://reviews.freebsd.org/D14064

6 years agoMove the mlx5 core device pointer first in the mlx5en priv. This help simplify
hselasky [Tue, 30 Jan 2018 12:38:06 +0000 (12:38 +0000)]
Move the mlx5 core device pointer first in the mlx5en priv. This help simplify
checks to recognize own network devices when using mlx5ib. This patch fixes
an issues where mlx5ib fails to recognize mceX network devices for use with
RoCE.

MFC after: 1 week
Sponsored by: Mellanox Technologies

6 years agoDocument the new hw.usb.template behaviour.
trasz [Tue, 30 Jan 2018 10:10:02 +0000 (10:10 +0000)]
Document the new hw.usb.template behaviour.

MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation

6 years agoMake the handler routine for the hw.usb.template sysctl trigger the USB
trasz [Tue, 30 Jan 2018 10:08:11 +0000 (10:08 +0000)]
Make the handler routine for the hw.usb.template sysctl trigger the USB
host to reprobe the bus by switching the USB pull up resistors off and
back on.  In other words - when FreeBSD is configured as a USB device,
changing the sysctl will be immediately noticed by the machine it's
connected to.

Reviewed by: hselasky@
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation

6 years agonfsstat: Add libxo output
manu [Tue, 30 Jan 2018 09:59:52 +0000 (09:59 +0000)]
nfsstat: Add libxo output

Add libxo output support
Merge exp41_intpr and exp_intpr function. The only difference is to print
NFSV4.1 operations in exp41, add a third arguement to control that.
printtitle was set to 1 and don't have a switch, add a -q options to control it.

Reviewed by: bapt
Sponsored by: Gandi.net
Differential Revision: https://reviews.freebsd.org/D14012

6 years agoUse more verbose panic messages.
mmel [Tue, 30 Jan 2018 04:06:30 +0000 (04:06 +0000)]
Use more verbose panic messages.

MFC after: 2 weeks

6 years agoRevert r328511, it was committed with <patch>.diff instead of <patch>.txt as
mmel [Tue, 30 Jan 2018 04:05:03 +0000 (04:05 +0000)]
Revert r328511, it was committed with <patch>.diff instead of <patch>.txt as
commit log.

6 years agostand/fdt: Remove unused write-only new_fdtp, correct comment
kevans [Tue, 30 Jan 2018 03:31:40 +0000 (03:31 +0000)]
stand/fdt: Remove unused write-only new_fdtp, correct comment

MFC after: 3 days

6 years agolibedit: sort the Makefile in line with NetBSD's version.
pfg [Mon, 29 Jan 2018 22:38:23 +0000 (22:38 +0000)]
libedit: sort the Makefile in line with NetBSD's version.

NetBSD's libedit has been been cleaned-up considerably so the
non--widecharacter version is no longer an option. Re -sorting the
Makefile should make it easier for some brave soul trying to update it.

No functional change intended.

MFC after: 5 days

6 years agoFix mistake in case of zeroed inode check.
fsu [Mon, 29 Jan 2018 22:15:46 +0000 (22:15 +0000)]
Fix mistake in case of zeroed inode check.

Reported by: pho
MFC after: 6 months

6 years agoAdd flex_bg/meta_bg features RW support.
fsu [Mon, 29 Jan 2018 21:54:13 +0000 (21:54 +0000)]
Add flex_bg/meta_bg features RW support.

Reviewed by:    pfg
MFC after:      6 months

Differential Revision:    https://reviews.freebsd.org/D13964

6 years agoDon't use an .OBJDIR for 'make sysent'.
bdrewery [Mon, 29 Jan 2018 19:14:15 +0000 (19:14 +0000)]
Don't use an .OBJDIR for 'make sysent'.

Reported by: emaste, jhb
Sponsored by: Dell EMC

6 years agoRemove t_grep:mmap_eof_not_eol test
kevans [Mon, 29 Jan 2018 18:50:45 +0000 (18:50 +0000)]
Remove t_grep:mmap_eof_not_eol test

The test was marked as an expected failure in r320414 after r319971's import
of a newer jemalloc removed an essential feature (opt.redzone) for
reproducing the behavior it was testing. Since then, no way has been found
or demonstrated to reliably test the behavior, so remove the test.

PR: 220309

6 years agoDo the book-keeping on release before we release the reference. The
imp [Mon, 29 Jan 2018 18:07:14 +0000 (18:07 +0000)]
Do the book-keeping on release before we release the reference. The
periph was going away on final release, and then returning and we
started dancing in free memory.

Sponsored by: Netflix

6 years agoRemove some duplicated sys/conf/files* entries.
benno [Mon, 29 Jan 2018 17:32:30 +0000 (17:32 +0000)]
Remove some duplicated sys/conf/files* entries.

net80211/ieee80211_ageq.c was present twice in sys/conf/files so leave the
correctly sorted one. dev/wpi/if_wpi.c was present in sys/conf/files as well
as sys/conf/files.amd64 and sys/conf/files.i386 so prefer the sys/conf/files
entry.

Reviewed by: allanjude, rstone

6 years agoND6: Set the correct state for new neighbor cache entries
vangyzen [Mon, 29 Jan 2018 16:12:26 +0000 (16:12 +0000)]
ND6: Set the correct state for new neighbor cache entries

Restore state 6.  Many of the UNH tests end up exercising this
state, where we have a new neighbor cache entry and a new link-layer
entry is being created for it.  The link-layer address is currently
unknown so the initial state of the "llentry" should remain initialized
to ND6_LLINFO_NOSTATE so that the ND code will send a solicitation.
Setting this to ND6_LLINFO_STALE implies that the link-level entry
is valid and can be used (but needs to be refreshed via the Neighbor
Unreachability state machine).

https://forums.freebsd.org/threads/64287/

Submitted by: Farrell Woods <Farrell_Woods@Dell.com>
Reviewed by: mjoras, dab, ae
MFC after: 1 week
Sponsored by: Dell EMC
Differential Revision: https://reviews.freebsd.org/D14059

6 years agopppctl88) Avoid strcpy() copies on overlapping string.
pfg [Mon, 29 Jan 2018 14:23:44 +0000 (14:23 +0000)]
pppctl88) Avoid strcpy() copies on overlapping string.

This may lead to unpredicatable behaviour on different platforms or C
library implementations. Use an intermediate variable.

Obtained from: DragonFlyBSD (git a861a526)

6 years agoawk(1): Don't install tests at all
kevans [Mon, 29 Jan 2018 14:15:44 +0000 (14:15 +0000)]
awk(1): Don't install tests at all

Tests were disconnected so that running `make check` in usr.bin/awk did not
have any effect, but CI runs use installed tests. Fully disconnect tests/
from the build for the time being as a short term solutio

Reported by: lwhsu

6 years agolibregex: Mark gnuext test as an expected fail
kevans [Mon, 29 Jan 2018 14:00:33 +0000 (14:00 +0000)]
libregex: Mark gnuext test as an expected fail

The test was added prematurely as a goal to reach with the GNU extension
functionality, but the functionality has not yet been introduced. Mark it as
an expected fail until that point.

6 years agolld: Put the header in the first PT_LOAD even if that PT_LOAD has a LMAExpr
emaste [Mon, 29 Jan 2018 13:55:50 +0000 (13:55 +0000)]
lld: Put the header in the first PT_LOAD even if that PT_LOAD has a LMAExpr

The root problem is that we were creating a PT_LOAD just for the header.
That was technically valid, but inconvenient: we should not be making
the ELF discontinuous.

The solution is to allow a section with LMAExpr to be added to a PT_LOAD
if that PT_LOAD doesn't already have a LMAExpr.

LLVM PR: 36017
Obtained from: LLVM r323625 by Rafael Espindola

6 years agolld: Move LMAOffset from the OutputSection to the PhdrEntry. NFC.
emaste [Mon, 29 Jan 2018 13:54:51 +0000 (13:54 +0000)]
lld: Move LMAOffset from the OutputSection to the PhdrEntry. NFC.

If two sections are in the same PT_LOAD, their relatives offsets,
virtual address and physical addresses are all the same.

[Rafael] initially wanted to have a single global LMAOffset, on the
assumption that every ELF file was in practiced loaded contiguously in
both physical and virtual memory.

Unfortunately that is not the case. The linux kernel has:

  LOAD           0x200000 0xffffffff81000000 0x0000000001000000 0xced000 0xced000 R E 0x200000
  LOAD           0x1000000 0xffffffff81e00000 0x0000000001e00000 0x15f000 0x15f000 RW  0x200000
  LOAD           0x1200000 0x0000000000000000 0x0000000001f5f000 0x01b198 0x01b198 RW  0x200000
  LOAD           0x137b000 0xffffffff81f7b000 0x0000000001f7b000 0x116000 0x1ec000 RWE 0x200000

The delta for all but the third PT_LOAD is the same:
0xffffffff80000000. [Rafael] thinks the 3rd one is a hack for implementing
per cpu data, but we can't break that.

Obtained from: LLVM r323456 by Rafael Espindola

6 years agolld: Improve LMARegion handling.
emaste [Mon, 29 Jan 2018 13:52:42 +0000 (13:52 +0000)]
lld: Improve LMARegion handling.

This fixes the crash reported at [LLVM] PR36083.

The issue is that we were trying to put all the sections in the same
PT_LOAD and crashing trying to write past the end of the file.

This also adds accounting for used space in LMARegion, without it all
3 PT_LOADs would have the same physical address.

Obtained from: LLVM r323449 by Rafael Espindola

6 years agolld: Simplify. NFC.
emaste [Mon, 29 Jan 2018 13:51:13 +0000 (13:51 +0000)]
lld: Simplify. NFC.

Obtained from: LLVM r323440 by Rafael Espindola

6 years agolld: Remove MemRegionOffset. NFC.
emaste [Mon, 29 Jan 2018 13:50:28 +0000 (13:50 +0000)]
lld: Remove MemRegionOffset. NFC.

We can just use a member variable in MemoryRegion.

Obtained from: LLVM r323399 by Rafael Espindola

6 years agolld: Only lookup LMARegion once. NFC.
emaste [Mon, 29 Jan 2018 13:49:10 +0000 (13:49 +0000)]
lld: Only lookup LMARegion once. NFC.

This is similar to how we handle MemRegion.

Obtained from: LLVM r323396 by Rafael Espindola

6 years agolld: Use lookup instead of find. NFC, just simpler.
emaste [Mon, 29 Jan 2018 13:48:15 +0000 (13:48 +0000)]
lld: Use lookup instead of find. NFC, just simpler.

Obtained from: LLVM r323395 by Rafael Espindola