]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
7 years agoUpdate clang to release_39 branch r279477.
Dimitry Andric [Wed, 24 Aug 2016 17:43:49 +0000 (17:43 +0000)]
Update clang to release_39 branch r279477.

7 years agoUpdate llvm to release_39 branch r279477.
Dimitry Andric [Wed, 24 Aug 2016 17:43:08 +0000 (17:43 +0000)]
Update llvm to release_39 branch r279477.

7 years agoVendor import of clang release_39 branch r279477:
Dimitry Andric [Wed, 24 Aug 2016 17:36:20 +0000 (17:36 +0000)]
Vendor import of clang release_39 branch r279477:
https://llvm.org/svn/llvm-project/cfe/branches/release_39@279477

7 years agoVendor import of llvm release_39 branch r279477:
Dimitry Andric [Wed, 24 Aug 2016 17:35:37 +0000 (17:35 +0000)]
Vendor import of llvm release_39 branch r279477:
https://llvm.org/svn/llvm-project/llvm/branches/release_39@279477

7 years agoUpdate generated llvm DataTypes.h header.
Dimitry Andric [Tue, 23 Aug 2016 19:57:37 +0000 (19:57 +0000)]
Update generated llvm DataTypes.h header.

7 years agoMerge ^/head r304537 through r304699.
Dimitry Andric [Tue, 23 Aug 2016 19:55:02 +0000 (19:55 +0000)]
Merge ^/head r304537 through r304699.

7 years agoFix key delay and repeat, part 1.
Bruce Evans [Tue, 23 Aug 2016 19:50:16 +0000 (19:50 +0000)]
Fix key delay and repeat, part 1.

kbdcontrol -r fast is documented to give a non-emulated atkbd's fastest
rate of 250.34, but is misimplemented to request this as 0.0.  ukbd
supports many nonstandard rates, although it is currently too inaccurate
by a factor of several hundred for non-huge nonstandard rates to be
useful.  It mapped 0.0 to 200.0.  A repeat delay of 0 means a rate of
infinity which is quite fast, but physical constraints limit this to
a few MHz and the inaccuracies made it almost usable.

Convert 0.0 to the documented 250.34.

Also convert negative args and small args to the 250.34 minimal ones,
like atkbd does.  This is for KDSETREPEAT -- the 2 versions of the
deprecated KDSETRAD have bounds checking.  Keep not doing any bounds
checking or conversions for upper limits since nonstandard large
delays are useful for testing.

The inaccuracies are dependent on HZ and the timeout implementation.
With the old timeout implementation and HZ = 1000, 200.0 probably
worked better to emulate 250.34 than 250.34 itself.  HZ = 100 gives
roundoff errors that accidentally reduce the inaaccuracies, and
event timers reduce the inaccuracies even more, so 200.0 was giving
more like itself (perhaps 215.15 on average but sometimes close to
10 msec repeat which is noticebly too fast).  This commit makes 0.0
noticeably too slow, like 250.34 always was.

7 years agoAdd support for _PC_ACL_NFS4 as TRUSTEDBSD_ACL_NFS4
Enji Cooper [Tue, 23 Aug 2016 19:41:49 +0000 (19:41 +0000)]
Add support for _PC_ACL_NFS4 as TRUSTEDBSD_ACL_NFS4

The TRUSTEDBSD prefix was chosen for consistency with the other
related `_PC_ACL*` prefixed variables.

MFC after: 3 days
Sponsored by: EMC / Isilon Storage Division

7 years agoFAST_DEPEND: Fix 'make all install' not properly rebuilding based on .depend.* files.
Bryan Drewery [Tue, 23 Aug 2016 19:37:18 +0000 (19:37 +0000)]
FAST_DEPEND: Fix 'make all install' not properly rebuilding based on .depend.* files.

An optimization is in place to skip reading the .depend.* files with
'make install'.  This was too strong and broke 'make all install' and
'make foo.o foo install'.  Now only skip reading the dependency files
if all make targets ran are install targets.

The problem comes about because headers are only added in as a guessed
dependency if .depend.* files do not yet exist.  If they do exist, even
if being skipped from being read, then the header dependencies are not
applied.  This applies to all #included files, and not just headers.

Reported by: kib
MFC after: 1 day
Sponsored by: EMC / Isilon Storage Division

7 years agoIn addition to creating subdirectories under .OBJDIR for SRCS with
Dimitry Andric [Tue, 23 Aug 2016 19:31:43 +0000 (19:31 +0000)]
In addition to creating subdirectories under .OBJDIR for SRCS with
relative paths, also create them for DPSRCS.  This is needed for builds
that generate files during the depend stage, which cannot be compiled by
themselves, since those have to be put in DPSRCS.

7 years agoFix in-tree GCC builds after r304681.
Bryan Drewery [Tue, 23 Aug 2016 19:29:37 +0000 (19:29 +0000)]
Fix in-tree GCC builds after r304681.

There were a few issues.
- In-tree GCC won't have X_COMPILER_TYPE defined but will have
  WANT_COMPILER_TYPE==gcc set from the SYSTEM_COMPILER logic that can
  be used.  Make the clang check specific to clang as well to ensure
  -target doesn't leak into a GCC build.
- When using a cross-compiler GCC (with a default sysroot or arch) and also
  passing --sysroot, it basically forgets all internal paths for
  libraries.  We've already worked around this quite a bit for
  the external toolchains.  Now for the in-tree bootstrap cross-compiler
  GCC, also pass in the needed -B${WORLDTMP}/usr/lib to find the crt
  object files, but also -isystem and -L to fix the paths.  This creates
  quite a spammy build log, but it is clear and still achieves the goals
  and stays consistent between internal and external build flags.
  Reducing the spam by using the '=' prefix feature will help and be
  done later.

MFC after: 3 days
X-MFC-With: r304681
Reported by: bz
Pointyhat to: bdrewery
Sponsored by: EMC / Isilon Storage Division

7 years agoAdd `MIN_HOLE_SIZE` pathconf(2) support to getconf
Enji Cooper [Tue, 23 Aug 2016 19:28:01 +0000 (19:28 +0000)]
Add `MIN_HOLE_SIZE` pathconf(2) support to getconf

This allows shell programs to programmatically determine whether
or not a filesystem supports sparse files

MFC after: 3 days
Sponsored by: EMC / Isilon Storage Division

7 years agoClean up trailing whitespace
Enji Cooper [Tue, 23 Aug 2016 19:15:01 +0000 (19:15 +0000)]
Clean up trailing whitespace

MFC after: 3 days
Sponsored by: EMC / Isilon Storage Division

7 years agobhndb(4): Fix unsigned integer underflow in dynamic register window
Landon J. Fuller [Tue, 23 Aug 2016 19:03:11 +0000 (19:03 +0000)]
bhndb(4): Fix unsigned integer underflow in dynamic register window
handling. This resulted in the window target being left uninitialized
when an underflow occured.

Approved by: adrian (mentor)
Differential Revision: https://reviews.freebsd.org/D7617

7 years agobspatch: apply style(9)
Ed Maste [Tue, 23 Aug 2016 17:42:03 +0000 (17:42 +0000)]
bspatch: apply style(9)

Make style changes (and trivial refactoring of open calls) now in order
to reduce noise in diffs for future capsicum changes.

Reviewed by: oshogbo
No objection: cperciva
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D7610

7 years agoAlso adjust the virtual address passed to vm_page_pa_tryrelock.
Andrew Turner [Tue, 23 Aug 2016 16:37:34 +0000 (16:37 +0000)]
Also adjust the virtual address passed to vm_page_pa_tryrelock.

Reported by: alc
Obtained from: ABT Systems Ltd
MFC after: 1 month
Sponsored by: The FreeBSD Foundation

7 years agoMap memory as read-only in pmap_enter_quick_locked as is done in other
Andrew Turner [Tue, 23 Aug 2016 16:20:56 +0000 (16:20 +0000)]
Map memory as read-only in pmap_enter_quick_locked as is done in other
pmap implementations.

Obtained from: ABT Systems Ltd
MFC after: 1 month
Sponsored by: The FreeBSD Foundation

7 years agoIf we find we have a superpage in pmap_enter_quick_locked return without
Andrew Turner [Tue, 23 Aug 2016 16:12:25 +0000 (16:12 +0000)]
If we find we have a superpage in pmap_enter_quick_locked return without
trying to add a new level 3 page.

Obtained from: ABT Systems Ltd
MFC after: 1 month
Sponsored by: The FreeBSD Foundation

7 years agoindent(1): have the memset invocation somewhat more canonical.
Pedro F. Giffuni [Tue, 23 Aug 2016 15:49:31 +0000 (15:49 +0000)]
indent(1): have the memset invocation somewhat more canonical.

While correct, the previous invocation was somewhat more error prone.

Pointed out by: delphij, bde

7 years agoInclude the offset the virtual address is within an L1 or L2 block when
Andrew Turner [Tue, 23 Aug 2016 15:48:27 +0000 (15:48 +0000)]
Include the offset the virtual address is within an L1 or L2 block when
finding the vm_page_t in pmap_extract_and_hold. Previously it would return
the vm_page_t of the first page in a block. This would cause issues when,
for example, fsck reads from a device into the middle of a superpage. In
this case the read call would write to the start of the block, and not to
the buffer passed in.

Obtained from: ABT Systems Ltd
MFC after: 1 month
Sponsored by: The FreeBSD Foundation

7 years agoindent(1): remove dead assignments.
Pedro F. Giffuni [Tue, 23 Aug 2016 15:46:20 +0000 (15:46 +0000)]
indent(1): remove dead assignments.

Taken from: Piotr Sephaniak

7 years agoRegenerate
Bryan Drewery [Tue, 23 Aug 2016 15:31:53 +0000 (15:31 +0000)]
Regenerate

7 years agoRe-enable WITH_SYSTEM_COMPILER with WITH_META_MODE.
Bryan Drewery [Tue, 23 Aug 2016 15:22:17 +0000 (15:22 +0000)]
Re-enable WITH_SYSTEM_COMPILER with WITH_META_MODE.

This was disabled in r301468 due to -target/--sysroot sometimes being used in
the build and other times not being used.  Now that it is always used since
r304681, it is safe to combine the features.

MFC after: 3 days
Sponsored by: EMC / Isilon Storage Division

7 years agoAlways pass in -target and --sysroot flags for the build.
Bryan Drewery [Tue, 23 Aug 2016 15:20:32 +0000 (15:20 +0000)]
Always pass in -target and --sysroot flags for the build.

The internal bootstrap compiler has a default sysroot set by TOOLS_PREFIX
and target set by TARGET/TARGET_ARCH.  However, there are several needs to
always pass an explicit --sysroot and -target.
- External compiler needs sysroot and target flags.
- External ld needs sysroot.
- To be clear about the use of a sysroot when using the internal compiler.
- Easier debugging.
- Allowing WITH_SYSTEM_COMPILER+WITH_META_MODE to work together due to
  the flip-flopping build command when sometimes using external and
  sometimes using internal.
- Allow using no lld which has support for default paths.

The default sysroot in the bootstrap compiler is not changed.  The
buildenv compiler will still work with its default and will also
include -target/--sysroot from CC in the environment.

MFC after: 3 days
Discussed with: emaste, brooks (BSDCam)
Reviewed by: emaste
Sponsored by: EMC / Isilon Storage Division

7 years agoEFI loader: only open/close on the net device with tftpfs
Baptiste Daroussin [Tue, 23 Aug 2016 13:53:38 +0000 (13:53 +0000)]
EFI loader: only open/close on the net device with tftpfs

It prevents issuing a dhcp request before each file open
As a consequence netbooting over tftpfs is significantly faster

Sponsored by: Gandi.net

7 years agoUse roundup2() from sys/param.h.
Marcelo Araujo [Tue, 23 Aug 2016 13:43:43 +0000 (13:43 +0000)]
Use roundup2() from sys/param.h.

7 years agoAdd tftpfs support for the EFI loader
Baptiste Daroussin [Tue, 23 Aug 2016 13:35:48 +0000 (13:35 +0000)]
Add tftpfs support for the EFI loader

Allow netbooting on efi without having to setup any NFS server by rebuilding the
loader with LOADER_TFTP_SUPPORT like for the i386 pxeloader

Sponsored by: Gandi.net

7 years agoFix calloc(3) argument order.
Marcelo Araujo [Tue, 23 Aug 2016 13:19:42 +0000 (13:19 +0000)]
Fix calloc(3) argument order.

MFC after: 4 weeks.

7 years agointpm: add support for SB800
Andriy Gapon [Tue, 23 Aug 2016 10:40:53 +0000 (10:40 +0000)]
intpm: add support for SB800

This code should be able to support later AMD chipsets as well, but that
hasn't been tested.

SB800 supports accessing several different SMBus buses using the same
set of constrol registeirs plus special PMIO registers that control which
bus is selected.  This could be exposed to consumers as several smb devices
each talking to its bus.  This feature is not implemented yet.

MFC after: 2 weeks

7 years agotlb1_init() can be called twice on BookE
Justin Hibbits [Tue, 23 Aug 2016 04:37:03 +0000 (04:37 +0000)]
tlb1_init() can be called twice on BookE

Summary:
There is no need to call tlb1_init() twice. Now it is called first time from
booke_init() and second time from powerpc_init() (where it is under BOOKE
switch). Although this does not cause immediate problems in the mainline kernel,
this can lead to undesirable side effects like two TLB entries with the same VA
in the TLB1. Presence of two TLB entries with the same VA can hang CPU.

Test Plan:
Add initial mapping for UART to the tlb1_init(), build and boot the kernel,
ensure that mapping presents only once (most convinient way - through Lauterbah
or similar hardware debugger)

Submitted by: Ivan Krivonos <int0dster_gmail.com>
Differential Revision: https://reviews.freebsd.org/D7607

7 years agoTake into account mas7/8 when reading/writing TLB entries on e6500
Justin Hibbits [Tue, 23 Aug 2016 04:26:30 +0000 (04:26 +0000)]
Take into account mas7/8 when reading/writing TLB entries on e6500

Summary: Current booke/pmap code ignores mas7 and mas8 on e6500 CPU.

Submitted by: Ivan Krivonos <int0dster_gmail.com>
Differential Revision: https://reviews.freebsd.org/D7606

7 years agonet: Split RNDIS protocol structs/macros out of dev/usb/net/if_urndisreg.h
Sepherosa Ziehau [Tue, 23 Aug 2016 02:54:06 +0000 (02:54 +0000)]
net: Split RNDIS protocol structs/macros out of dev/usb/net/if_urndisreg.h

So that Hyper-V can leverage them instead of rolling its own definition.

Discussed with: hps
Reviewed by: hps
MFC after: 1 week
Sponsored by: Microsoft
Differential Revision: https://reviews.freebsd.org/D7592

7 years agoindent(1): Fix off-by-one in control flow leading dead code.
Pedro F. Giffuni [Tue, 23 Aug 2016 02:07:08 +0000 (02:07 +0000)]
indent(1): Fix off-by-one in control flow leading dead code.

Coverity correctly reported that it's impossible for /comparison/ to be 0
here, because the only way for the for loop to end is by /comparison/
being < 0.

Fortunately the consequences of this bug weren't severe; for duplicated
entries in the typedef names file it would unnecessarily duplicate strings
with strdup(), but pointers to those would replace existing ones. So this
was a memory leak at worst.

CID:  1361477
Obtained from:  Piotr Stephaniak

7 years agoFix missing substitution of @SBINDIR@ in resolvconf scripts
Eric Badger [Tue, 23 Aug 2016 02:06:20 +0000 (02:06 +0000)]
Fix missing substitution of @SBINDIR@ in resolvconf scripts

Certain features, such as resolv_conf_passthrough=NULL, do not work
correctly due to this missing substitution.

Also remove the @PREFIX@ substitution, which is no longer needed.

Reviewed by: pfg
Approved by: vangyzen (mentor)
MFC after: 1 week
Sponsored by: Dell Inc.
Differential Revision: https://reviews.freebsd.org/D7572

7 years agoindent(1): add some comments to quiet down Coverity.
Pedro F. Giffuni [Tue, 23 Aug 2016 01:58:02 +0000 (01:58 +0000)]
indent(1): add some comments to quiet down Coverity.

Hopefully adding comments should help explain the code to both static
checkers and humans.

CID: 976543, 976544, 976545
Obtained from: Piotr Stephaniak

7 years agoindent(1): Fix memory leaks pointed out by clang-analyzer.
Pedro F. Giffuni [Tue, 23 Aug 2016 01:40:45 +0000 (01:40 +0000)]
indent(1): Fix memory leaks pointed out by clang-analyzer.

Shift the responsibility of allocating memory for the string duplicate
from the caller (set_option, add_typedefs_from_file) to the callee
(add_typename) as it has more knowledge about when the duplication
actually needs to occur.

Taken from: Piotr Stefaniak

7 years agoDo not include file from dt-bindings and simply use the already present defines.
Emmanuel Vadot [Tue, 23 Aug 2016 00:46:22 +0000 (00:46 +0000)]
Do not include file from dt-bindings and simply use the already present defines.

Reported by: jmcneill
MFC after: 1 week

7 years agoRename ORDERED to BOOTSTRAP since no order is respected in the list.
Bryan Drewery [Mon, 22 Aug 2016 22:51:10 +0000 (22:51 +0000)]
Rename ORDERED to BOOTSTRAP since no order is respected in the list.

The directories in SUBDIR_ORDERED are built in parallel, so the name is
misleading.

MFC after: 3 days
Sponsored by: EMC / Isilon Storage Division

7 years agoFor 'make <directory>' hook into the all_subdir_<directory> targets.
Bryan Drewery [Mon, 22 Aug 2016 22:51:07 +0000 (22:51 +0000)]
For 'make <directory>' hook into the all_subdir_<directory> targets.

This fixes parallel build issues when trying to depend on ${SUBDIR}.  An
example of this in share/i18n/csmapper/Makefile where mapper.dir depends
on ${SUBDIR} having been traversed and built already.  Before this
change running make in that directory would build the subdirectories
twice.  This led to obscure build races.  While reworking that build
may be possible, the framework should not so easily allow creating such
problems.

Now depending on <directory> will properly redirect to the
all_subdir_<directory> target rather than invoking the inline shell.

This also makes 'make -jX <directory>' now respect any
SUBDIR_DEPEND_<directory> statements when SUBDIR_PARALLEL is defined.
This is not entirely intended and may be changed later.

MFC after: 2 weeks
Sponsored by: EMC / Isilon Storage Division

7 years agoAlways define the various <target>_subdir_<directory> targets, even if not used.
Bryan Drewery [Mon, 22 Aug 2016 22:51:04 +0000 (22:51 +0000)]
Always define the various <target>_subdir_<directory> targets, even if not used.

This is part of an effort to cleanup handling of some edge cases
involving 'make <directory>'.  It also provides the targets for
other uses.

MFC after: 2 weeks
Sponsored by: EMC / Isilon Storage Division

7 years agoStop using _SUBDIR internally for non-SUBDIR_PARALLEL builds.
Bryan Drewery [Mon, 22 Aug 2016 22:51:01 +0000 (22:51 +0000)]
Stop using _SUBDIR internally for non-SUBDIR_PARALLEL builds.

This is unifying more of the logic.  Rather than create targets such
as 'all: all_subdir_foo' when using SUBDIR_PARALLEL and using
'all: _SUBDIR' when not using SUBDIR_PARALLEL, always use the
expanded out <target>_subdir_<directory> pattern.  When not using
SUBDIR_PARALLEL, have each directory-target depend on the previously
defined targets as to respect the *order* of SUBDIR.

Using 'make -N' now prints all directory traversals individually rather
than using a loop, since a loop is no longer used to traverse.

This is part of an effort to cleanup handling of some edge cases
involving 'make <directory>' and making it simpler in the sense
that the pattern used to build is the same for all modes.

MFC after: 2 weeks
Sponsored by: EMC / Isilon Storage Division

7 years agoFix building on read-only source trees.
Bryan Drewery [Mon, 22 Aug 2016 22:50:58 +0000 (22:50 +0000)]
Fix building on read-only source trees.

This partially reverts r296702 and reworks the original check to only
look in .CURDIR.  This avoids ever trying to rebuild a .src file that is
already in the source tree as an override.

PR: 211952
MFC after: 3 days
Sponsored by: EMC / Isilon Storage Division

7 years ago1) Back out r304607 case 2). fgetwln() as its pair fgetln() supposed to
Andrey A. Chernov [Mon, 22 Aug 2016 22:28:41 +0000 (22:28 +0000)]
1) Back out r304607 case 2). fgetwln() as its pair fgetln() supposed to
return partial line on any errors. See the comment in fgetln.c.
Add corresponding comment to fgetwln() too.
2) Rewrite r304607 case 1).
3) Remove "Fast path" from __fgetwc_mbs() since it can't detect encoding
errors and ignores them all.

PR:     212033
MFC after:      7 days

7 years agoPrefer C-style comments in assembly sources.
Jung-uk Kim [Mon, 22 Aug 2016 21:49:17 +0000 (21:49 +0000)]
Prefer C-style comments in assembly sources.

7 years agoFix white spaces in assembly sources.
Jung-uk Kim [Mon, 22 Aug 2016 21:30:59 +0000 (21:30 +0000)]
Fix white spaces in assembly sources.

7 years agoFix build for !SMP kernels after the Xen MSIX workaround.
John Baldwin [Mon, 22 Aug 2016 21:23:17 +0000 (21:23 +0000)]
Fix build for !SMP kernels after the Xen MSIX workaround.

Move msix_disable_migration under #ifdef SMP since it doesn't make sense
for !SMP kernels.

PR: 212014
Reported by: Glyn Grinstead <glyn@grinstead.org>
MFC after: 3 days

7 years agoBuild OpenSSL assembly sources for arm. Tested with Raspberry Pi 2 Model B.
Jung-uk Kim [Mon, 22 Aug 2016 20:59:34 +0000 (20:59 +0000)]
Build OpenSSL assembly sources for arm.  Tested with Raspberry Pi 2 Model B.

MFC after: 1 week

7 years agoRemove support for SSH1 as it is already disabled in our OpenSSH.
Ollivier Robert [Mon, 22 Aug 2016 20:48:46 +0000 (20:48 +0000)]
Remove support for SSH1 as it is already disabled in our OpenSSH.

Submitted by: vangyzen
MFC after: 2 weeks

7 years agoDon't separate the status stage of the XHCI USB control transfers into
Hans Petter Selasky [Mon, 22 Aug 2016 19:32:50 +0000 (19:32 +0000)]
Don't separate the status stage of the XHCI USB control transfers into
its own job because this breaks the simplified QEMU XHCI TRB parser,
which expects the complete USB control transfer as a series of back to
back TRBs. The old behaviour is kept under #ifdef in case this change
breaks enumeration of any USB devices.

PR: 212021
MFC after: 1 week

7 years agoAdd support for Ed25519 keys.
Ollivier Robert [Mon, 22 Aug 2016 19:27:20 +0000 (19:27 +0000)]
Add support for Ed25519 keys.

Reported by: mwlucas
MFH: 2 weeks

7 years agoFix the arm64 non-SMP build, active_irq is a uint64_t so cast it through
Andrew Turner [Mon, 22 Aug 2016 19:05:11 +0000 (19:05 +0000)]
Fix the arm64 non-SMP build, active_irq is a uint64_t so cast it through
a uintmax_t.

Obtained from: ABT Systems Ltd
MFC after: 1 week
Sponsored by: The FreeBSD Foundation

7 years agoRemove duplicate symbol from libhx509 version-script.map
Ed Maste [Mon, 22 Aug 2016 18:50:57 +0000 (18:50 +0000)]
Remove duplicate symbol from libhx509 version-script.map

Upstream commit r21331 (7758a5d0) added semiprivate function
_hx509_request_to_pkcs10 twice. This change has been committed upstream
as 8ef0071d.

7 years agoFix building for ARM kernel that have FLASHADDR, PHYSADDR and LOADERRAMADDR defined.
Emmanuel Vadot [Mon, 22 Aug 2016 18:33:56 +0000 (18:33 +0000)]
Fix building for ARM kernel that have FLASHADDR, PHYSADDR and LOADERRAMADDR defined.

Pointy Hat: myself

Reported by: bz

7 years agoEnsure map is valid, even before userland exists and the fault address
Andrew Turner [Mon, 22 Aug 2016 18:19:46 +0000 (18:19 +0000)]
Ensure map is valid, even before userland exists and the fault address
register points to an address in the userland range.

Obtained from: ABT Systems Ltd
MFC after: 1 week
Sponsored by: the FreeBSD Foundation

7 years agoRemove cross references to el(4) and ie(4).
John Baldwin [Mon, 22 Aug 2016 18:17:29 +0000 (18:17 +0000)]
Remove cross references to el(4) and ie(4).

7 years agoFix pmap_update_entry, pmap_invalidate_range takes the end address, not
Andrew Turner [Mon, 22 Aug 2016 18:12:44 +0000 (18:12 +0000)]
Fix pmap_update_entry, pmap_invalidate_range takes the end address, not
the size.

Obtained from: ABT Systems Ltd
MFC after: 1 month
Sponsored by: The FreeBSD Foundation

7 years agoRegenerate src.conf.5 after r304616
Ed Maste [Mon, 22 Aug 2016 17:53:18 +0000 (17:53 +0000)]
Regenerate src.conf.5 after r304616

7 years agoForcibly disable MK_TESTS if building without C++
Ed Maste [Mon, 22 Aug 2016 17:45:30 +0000 (17:45 +0000)]
Forcibly disable MK_TESTS if building without C++

Several atf components require C++, and the test suite is not usable
if building WITHOUT_CXX.

Reviewed by: bdrewery, jmmv
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D7597

7 years agoMake CloudABI work on i386.
Ed Schouten [Mon, 22 Aug 2016 17:37:31 +0000 (17:37 +0000)]
Make CloudABI work on i386.

Copy over amd64's cloudabi64_sysvec.c into i386 and tailor it to work.
Again, we use a system call convention similar to FreeBSD, except that
there is no support for indirect system calls (%eax == 0).

Where i386 differs from amd64 is that we have to store thread/process
entry arguments on the stack instead of using registers. We also have to
put an extra pointer on the stack for TLS (for GSBASE). Place that
pointer in the empty slot that is normally used to hold return
addresses. That seems to keep the code simple.

Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D7590

7 years agoFurther fixes for translation of PrintScreen/SysRq.
Bruce Evans [Mon, 22 Aug 2016 16:39:51 +0000 (16:39 +0000)]
Further fixes for translation of PrintScreen/SysRq.

The previous fix was tested mainly on 3 AT keyboards with USB adaptors where
it works.  1 USB keyboard doesn't translate Alt-PrintScreen, so the software
has to do it.

Reorganize a little to share some code and to not translate the unusual usb
scan code0x8a unless an Alt modified is set.  Remove redundant check of Alt
modifiers.  Translation now more clearly filters out Alt-PrintScreen before
the check.

The table of errors fixed in the previous commit had many bugs.  Correct
table:

K_RAW  Ctl-PrintScreen:                                 E0-2A-E0-37 -> E0-37
K_RAW  Alt-PrintScreen (with 4 comb. of Ctl/Shift):     79 -> 54
K_RAW  Pause/Break (with 4 comb. of Alt/Shift):         E0-46 -> E1-1D-45
K_CODE PrintScreen (with 4 comb. of Ctl/Shift):         54 -> 5c
K_CODE Alt-PrintScreen (with 4 comb. of Ctl/Shift):     7e -> 54
K_CODE Pause/Break (with 8 comb. of Ctl/Alt/Shift):     6c -> 68

That is 25 of 32 shift combinations for 2 keys fixed.  All 16 combinations
were broken for K_CODE and thus also for K_XLATE.

7 years agoBump __FreeBSD_version for C++11 thread_local support in r303795.
Bryan Drewery [Mon, 22 Aug 2016 15:52:03 +0000 (15:52 +0000)]
Bump __FreeBSD_version for C++11 thread_local support in r303795.

PR: 192320

7 years agoFix error processing.
Andrey A. Chernov [Mon, 22 Aug 2016 15:44:54 +0000 (15:44 +0000)]
Fix error processing.
1) Don't forget to set __SERR on __slbexpand() error.
2) Check for __fgetwc() errors using errno. Don't check for __SERR
as PR suggested, it user-visible flag which can stick from previous
functions and stdio code don't check it for this purpose.

PR:     212033
MFC after:      3 days

7 years agoTemporarily disable the optimization from r304436
Ryan Stone [Mon, 22 Aug 2016 15:27:37 +0000 (15:27 +0000)]
Temporarily disable the optimization from r304436

r304436 attempted to optimize the handling of incoming UDP packet by only
making an expensive call to in_broadcast() if the mbuf was marked as an
broadcast packet.  Unfortunately, this cannot work in the case of point-to-
point L2 protocols like PPP, which have no notion of "broadcast".

Discussions on how to properly fix r304436 are ongoing, but in the meantime
disable the optimization to ensure that no existing network setups are broken.

Reported by: bms

7 years agoFix calloc(3) argument order.
Marcelo Araujo [Mon, 22 Aug 2016 15:01:39 +0000 (15:01 +0000)]
Fix calloc(3) argument order.

Reviewed by: trasz
MFC after: 4 weeks.
Differential Revision: https://reviews.freebsd.org/D7532

7 years agoUse switch statements in pmap_remove_pages. While only one level of
Andrew Turner [Mon, 22 Aug 2016 14:53:39 +0000 (14:53 +0000)]
Use switch statements in pmap_remove_pages. While only one level of
pagetable is supported more will be added soon to support removing
superpages.

Obtained from: ABT Systems Ltd
MFC after: 1 month
Sponsored by: The FreeBSD Foundation

7 years agoioat(4): Allow callouts to be scheduled after hw reset
Conrad Meyer [Mon, 22 Aug 2016 14:51:09 +0000 (14:51 +0000)]
ioat(4): Allow callouts to be scheduled after hw reset

is_completion_pending governs whether or not a callout will be scheduled
when new work is queued on the IOAT device.  If true, a callout is
already scheduled, so we do not need a new one.  If false, we schedule
one and set it true.  Because resetting the hardware completed all
outstanding work but failed to clear is_completion_pending, no new
callout could be scheduled after a reset with pending work.

This resulted in a driver hang for polled-only work.

7 years agoioat(4): Don't process events past queue head
Conrad Meyer [Mon, 22 Aug 2016 14:51:07 +0000 (14:51 +0000)]
ioat(4): Don't process events past queue head

Fix a race where the completion routine could overrun the active ring
area in some situations.

7 years agoIncrease the maximum RX/TX queue size. This allows for a RX/TX queue
Hans Petter Selasky [Mon, 22 Aug 2016 13:43:25 +0000 (13:43 +0000)]
Increase the maximum RX/TX queue size. This allows for a RX/TX queue
size of 16384 mbufs. Previously the limit was 8192.

Sponsored by: Mellanox Technologies
MFC after: 1 week

7 years agoUse pmap_update_entry in pmap_enter when updating an entry with a new
Andrew Turner [Mon, 22 Aug 2016 12:56:40 +0000 (12:56 +0000)]
Use pmap_update_entry in pmap_enter when updating an entry with a new
physical address. This is required when either mapping is writeable.

While here remove an unneeded call to pmap_pde, we already have the pde
from earlier in the function.

Obtained from: ABT Systems Ltd
MFC after: 1 month
Sponsored by: The FreeBSD Foundation

7 years agoAdd sysctls to report on superpages statistics. While here add extra
Andrew Turner [Mon, 22 Aug 2016 12:17:40 +0000 (12:17 +0000)]
Add sysctls to report on superpages statistics. While here add extra
logging to these paths.

Obtained from: ABT Systems Ltd
MFC after: 1 month
Sponsored by: The FreeBSD Foundation

7 years agoAdd a size argument to pmap_update_entry.
Andrew Turner [Mon, 22 Aug 2016 10:50:30 +0000 (10:50 +0000)]
Add a size argument to pmap_update_entry.
Make use of this in pmap_promote_l2.

Obtained from: ABT Systems Ltd
MFC after: 1 month
Sponsored by: The FreeBSD Foundation

7 years agoFix for invalid use of bits in input context. Basically split
Hans Petter Selasky [Mon, 22 Aug 2016 10:21:25 +0000 (10:21 +0000)]
Fix for invalid use of bits in input context. Basically split
configuring of EP0 and non-EP0 into xhci_cmd_evaluate_ctx() and
xhci_cmd_configure_ep() respectivly. This resolves some errors when
using XHCI under QEMU and gets is more in line with the XHCI
specification.

PR: 212021
MFC after: 1 week

7 years agoAdd KASSERTS in pmap_alloc_l3 to ensure we are not encountering superpages
Andrew Turner [Mon, 22 Aug 2016 10:21:09 +0000 (10:21 +0000)]
Add KASSERTS in pmap_alloc_l3 to ensure we are not encountering superpages
when we don't yet expect them;

Obtained from: ABT Systems Ltd
MFC after: 1 month
Sponsored by: The FreeBSD Foundation

7 years agohyperv/hn: Factor out function to simplify NVS request sending
Sepherosa Ziehau [Mon, 22 Aug 2016 08:00:14 +0000 (08:00 +0000)]
hyperv/hn: Factor out function to simplify NVS request sending

MFC after: 1 week
Sponsored by: Microsoft
Differential Revision: https://reviews.freebsd.org/D7578

7 years agohyperv/hn: Factor out function to execute NVS transactions.
Sepherosa Ziehau [Mon, 22 Aug 2016 07:51:46 +0000 (07:51 +0000)]
hyperv/hn: Factor out function to execute NVS transactions.

MFC after: 1 week
Sponsored by: Microsoft
Differential Revision: https://reviews.freebsd.org/D7577

7 years agohyperv/hn: Get rid of netvsc_dev
Sepherosa Ziehau [Mon, 22 Aug 2016 07:44:11 +0000 (07:44 +0000)]
hyperv/hn: Get rid of netvsc_dev

MFC after: 1 week
Sponsored by: Microsoft
Differential Revision: https://reviews.freebsd.org/D7575

7 years agohyperv/hn: Move chimney sending buffer to hn_softc
Sepherosa Ziehau [Mon, 22 Aug 2016 07:34:39 +0000 (07:34 +0000)]
hyperv/hn: Move chimney sending buffer to hn_softc

And don't recreate chimney sending buffer for each primary channel
open, it is now created in device_attach DEVMETHOD and destroyed
in device_detach DEVMETHOD.

MFC after: 1 week
Sponsored by: Microsoft
Differential Revision: https://reviews.freebsd.org/D7574

7 years agoAdd a missing change in r304575.
Pyun YongHyeon [Mon, 22 Aug 2016 03:28:06 +0000 (03:28 +0000)]
Add a missing change in r304575.

Noticed by: jhb

7 years agoImprove the locking when sending user messages.
Michael Tuexen [Mon, 22 Aug 2016 01:45:29 +0000 (01:45 +0000)]
Improve the locking when sending user messages.

First, keep a ref count on the stcb after looking it up, as
done in the other lookup cases.
Second, before looking again at sp, ensure that it is not
freed, because the assoc is about to be freed.

MFC after: 3 days

7 years agoAdd Killer E2400 to the supported hardware list.
Pyun YongHyeon [Mon, 22 Aug 2016 01:28:02 +0000 (01:28 +0000)]
Add Killer E2400 to the supported hardware list.

7 years agoAdd Killer E2400 Gigabit Ethernet support.
Pyun YongHyeon [Mon, 22 Aug 2016 01:19:05 +0000 (01:19 +0000)]
Add Killer E2400 Gigabit Ethernet support.
It seems Killer E2200/E2400 has a BIOS misconfiguration or silicon
bug which triggers DMA write errors when driver uses advertised
maximum payload size.  Force the maximum payload size to 128 bytes
in DMA configuration.
This change should fix occasional DMA write errors reported on
Killer E2200.

Tested by: <psy0nic@sys-tek.org>

7 years agoCorrect DMA channel number selection on AR816x family of
Pyun YongHyeon [Mon, 22 Aug 2016 01:06:54 +0000 (01:06 +0000)]
Correct DMA channel number selection on AR816x family of
controllers. For Gigabit Ethernet version of AR816x, AR813x/AR815x
except L1D controller, use vendor recommended ASPM parameters.
While here, increase alc_dma_burst array size.  Broken H/W can
return bogus value in theory.

7 years agoRemove duplicate code, which is not protected by the appropriate locks.
Michael Tuexen [Mon, 22 Aug 2016 00:40:45 +0000 (00:40 +0000)]
Remove duplicate code, which is not protected by the appropriate locks.

MFC after: 3 days

7 years agoRemove the kernel optoion for IPSEC_FILTERTUNNEL, which was deprecated
Bjoern A. Zeeb [Sun, 21 Aug 2016 18:55:30 +0000 (18:55 +0000)]
Remove the kernel optoion for IPSEC_FILTERTUNNEL, which was deprecated
more than 7 years ago in favour of a sysctl in r192648.

7 years agoMake the UKBD USB transfers double buffered and set them up one by one,
Hans Petter Selasky [Sun, 21 Aug 2016 18:37:21 +0000 (18:37 +0000)]
Make the UKBD USB transfers double buffered and set them up one by one,
so they are memory independent which allows for handling panics
triggered by the keyboard driver itself, typically via CTRL+ALT+ESC
sequences. Or if the USB keyboard driver was processing a key at the
moment of panic. Allow UKBD to be attached while keyboard polling is active.

Tested by: Bruce Evans <bde@freebsd.org>
MFC after: 1 week

7 years agoAdd the "-n" flag to cron(8), to prevent it from daemonizing.
Edward Tomasz Napierala [Sun, 21 Aug 2016 18:12:49 +0000 (18:12 +0000)]
Add the "-n" flag to cron(8), to prevent it from daemonizing.
This makes it possible to use it with external supervisors.
The "-n" flag name is compatible with Linux, NetBSD, and OpenBSD.

Reviewed by: jilles, pfg, wblock
MFC after: 1 month
Differential Revision: https://reviews.freebsd.org/D7581

7 years agoFix translation of the PrintScreen/SysRq and Pause/Break keys. Almost
Bruce Evans [Sun, 21 Aug 2016 16:06:00 +0000 (16:06 +0000)]
Fix translation of the PrintScreen/SysRq and Pause/Break keys.  Almost
everything was broken.  The cases that I noticed were Ctrl-PrintScreen
not being mapped to the virtual scancode 0x5c (debug) and Pause not being
mapped to the physical/virtual scancode 0x46 (slock).

These keys are the most complicated ones due to kludges to give some
compatibility back to before AT keyboards.

Alt-PrintScreen must pretend to be a separate key from PrintScreen
even at the "raw" level.  The (unique) usb code for it is 0x8a and we
just have to map this to our unique virtual scancode 0x54, but we
mapped it first to the internal code 0x7e and then to 0x79 which is a
key on the Japanese 106/109 keyboard.  This fix is under the
UKBD_EMULATE_ATASCANCODE option which shouldn't be used for non-AT
keyboards.  If it is, then the syscons Japanese keymaps have nothing
of importance for code 0x79 and can easily be changed.  0x54 is also
unimportant in Japanese and US keymaps.

NonAlt-PrintScreen and NonCtl-Pause/Break had many much larger bugs with
smaller compatibility problems from fixing them.  The details are too
ugly to give here.  Summary of the changed (hex) codes:

K_RAW  PrintScreen (Ctl, Shift, Ctl-Shift):             E0-2A-E0-37 -> E0-37
K_RAW  Alt-PrintScreen (all shift states):              79 -> 54
K_RAW  Pause/Break (unshifted, Shift, Alt, Alt-Shift)): E0-46 -> E1-1D-45
K_CODE ALT-PrintScreen (all shift states):              79 -> 54

That is 15 of 32 shift combinations for 2 keys fixed, with 8 easy cases
from the 79 -> 54 remapping.

The difference is only large and with no workaround using a keymap for
for K_RAW, but this affects other modes when ukbd is layered under kbmux
because kbmux keeps all subdevices in K_RAW mode and translates.  Oops.
I used kbdmux to generate the above table of changes.

7 years agoallwinner: Remove a20/a20_cpu_cfg.c from the build.
Emmanuel Vadot [Sun, 21 Aug 2016 16:03:44 +0000 (16:03 +0000)]
allwinner: Remove a20/a20_cpu_cfg.c from the build.
This was needed when we used the SoC specific timer and not the generic-timer.

7 years agoAdd source files generated from the 32-bit system call table.
Ed Schouten [Sun, 21 Aug 2016 16:02:25 +0000 (16:02 +0000)]
Add source files generated from the 32-bit system call table.

7 years agoAdd CPU independent code for running 32-bits CloudABI executables.
Ed Schouten [Sun, 21 Aug 2016 16:01:30 +0000 (16:01 +0000)]
Add CPU independent code for running 32-bits CloudABI executables.

Essentially, this is a literal copy of the code in sys/compat/cloudabi64,
except that it now makes use of 32-bits datatypes and limits. In
sys/conf/files, we now need to take care to build the code in
sys/compat/cloudabi if either COMPAT_CLOUDABI32 or COMPAT_CLOUDABI64 is
turned on.

This change does not yet include any of the CPU dependent bits. Right
now I have implementations for running i386 binaries both on i386 and
x86-64, which I will send out for review separately.

7 years agoFix s/64/32/ conversion errors in the system call table.
Ed Schouten [Sun, 21 Aug 2016 15:56:19 +0000 (15:56 +0000)]
Fix s/64/32/ conversion errors in the system call table.

We should pull in the 32 bit headers when using this system call table.

7 years agoImport the 32-bit system call table and data types into the tree.
Ed Schouten [Sun, 21 Aug 2016 15:41:19 +0000 (15:41 +0000)]
Import the 32-bit system call table and data types into the tree.

Obtained from: https://github.com/NuxiNL/cloudabi

7 years agoRemove the ie(4) hardware list from the release documentation
Benjamin Kaduk [Sun, 21 Aug 2016 15:39:46 +0000 (15:39 +0000)]
Remove the ie(4) hardware list from the release documentation

The driver was removed by jhb in r304513, and the &hwlist.ie; entity
is no longer generated, causing the website build to fail.

7 years agoDon't forget to define __ELF_WORD_SIZE.
Ed Schouten [Sun, 21 Aug 2016 15:37:49 +0000 (15:37 +0000)]
Don't forget to define __ELF_WORD_SIZE.

Without it, we only obtain the ELF types native to the system. In this
we explicitly want the 64-bit versions.

7 years agoAdd a utility macro for converting 64-bit pointers to native pointers.
Ed Schouten [Sun, 21 Aug 2016 15:36:18 +0000 (15:36 +0000)]
Add a utility macro for converting 64-bit pointers to native pointers.

Right now we're casting uint64_t's to native pointers. This isn't
causing any problems right now, but if we want to provide a 32-bit
compatibility layer that works on 64-bit systems as well, this will
cause problems. Casting a uint32_t to a 64-bit pointer throws a compiler
error.

Introduce a TO_PTR() macro that casts the value to uintptr_t before
casting it to a pointer.

7 years agoMove the linker script from cloudabi64/ to cloudabi/.
Ed Schouten [Sun, 21 Aug 2016 15:14:06 +0000 (15:14 +0000)]
Move the linker script from cloudabi64/ to cloudabi/.

It turns out that it works perfectly fine for generating 32-bits vDSOs
as well. While there, get rid of the extraneous .s file extension.

7 years agoUse the right _MAX constant.
Ed Schouten [Sun, 21 Aug 2016 09:32:20 +0000 (09:32 +0000)]
Use the right _MAX constant.

Though uio_resid is of type ssize_t, we need to take into account that
this source file contains an implementation specific to a certain
userspace pointer size. If this file provided 32-bit implementations,
this should have used INT32_MAX, even when running a 64-bit kernel.

This change has no effect, but is simply in preparation for adding
support for running 32-bit CloudABI executables.

7 years agoUse memcpy() to copy 64-bit timestamps into the syscall return values.
Ed Schouten [Sun, 21 Aug 2016 07:41:11 +0000 (07:41 +0000)]
Use memcpy() to copy 64-bit timestamps into the syscall return values.

On 32-bit platforms, our 64-bit timestamps need to be split up across
two registers. A simple assignment to td_retval[0] will cause the top 32
bits to get lost. By using memcpy(), we will automatically either use 1
or 2 registers depending on the size of register_t.

7 years agoRewrite the vDSOs for CloudABI in assembly.
Ed Schouten [Sun, 21 Aug 2016 07:28:38 +0000 (07:28 +0000)]
Rewrite the vDSOs for CloudABI in assembly.

The reason why the old vDSOs were written in C using inline assembly was
purely because they were embedded in the C library directly as static
inline functions. This was practical during development, because it
meant you could invoke system calls without any library dependencies.
The vDSO was simply a copy of these functions.

Now that we require the use of the vDSO, there is no longer any need for
embedding them in C code directly. Rewriting them in assembly has the
advantage that they are closer to ideal (less useless branching, less
assumptions about registers remaining unclobbered by the kernel, etc).
They are also easier to build, as they no longer depend on the C type
information for CloudABI.

Obtained from: https://github.com/NuxiNL/cloudabi

7 years ago[mips] add support for the "creative" GNU extensions and IRIX hilarity around MIPS...
Adrian Chadd [Sun, 21 Aug 2016 00:48:41 +0000 (00:48 +0000)]
[mips] add support for the "creative" GNU extensions and IRIX hilarity around MIPS LO16/HI16 relocations.

This was .. an interesting headache.

There are two halves:

* The earlier IRIX stuff (yes, early) occasionally would do dead
  code removal and generate multiple consecutive LO16 entries.
  If this is done for REL entries then it's fine - there's no
  state kept between them.  But gcc 5.x seems to do this for
  RELA entries.

eg:

HI1 LO1 HI2 LO2 LO3 HI4 LO4

.. in this instance, LO2 should affect HI2, but LO3 doesn't at all
affect anything.  The matching HI3 was in code that was deleted
as "dead code".

Then, the next one:

* A "GCC extension" allows for multiple HI entries before a LO entry;
  and all of those HI entries use the first LO entry as their basis
  for RELA offset calculations.

It does this so GCC can also do dead code deletion without necessarily
having to geneate fake relocation entries for balanced HI/LO RELA
entries.

eg:

HI1 LO1 HI2 HI3 HI4 LO4 LO5 HI6 LO6 LO7

in this instance, HI{2,3,4} are the same relocation as LO4 (eg .bss)
and need to be buffered until LO4 - then the RELA offset is applied
from LO4 to HI{2,3,4} calculations.

/And/, the AHL from HI4 is used during the LO4 relocation calculation,
just like in the normal (ie, before this commit) implementation.

Then, LO5 doesn't trigger anything - the HI "buffer" is empty,
so there are no HI relocations to flush out.

HI6/LO6 are normal, and LO7 doesn't trigger any HI updates.

Tested:

* AR9344 SoC, kernel modules, using gcc-5.3 (mips-gcc-5.3.0 package)

Notes:

* Yes, I do feel dirty having written this code.

Reviewed by: imp (after a handful of "this should be on fire" moments wrt gcc and this code)