]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
9 years agoDefine SRCTOP in src.sys.mk
sjg [Tue, 26 May 2015 21:41:26 +0000 (21:41 +0000)]
Define SRCTOP in src.sys.mk

Use SRCTOP to replace .../ in MAKSYSPATH (avoid extra :tA)
Use SRCTOP rather than ROOTSRCDIR in src.libnames.mk
Merge from head

9 years agoLatest dirdeps.mk and meta.stage.mk
sjg [Tue, 26 May 2015 21:03:41 +0000 (21:03 +0000)]
Latest dirdeps.mk and meta.stage.mk

Add support for dirdeps cache
and allow full path as key for STAGE_AS_*

9 years agoAdd support for __DEFAULT_DEPENDENT_OPTIONS
sjg [Tue, 26 May 2015 16:17:24 +0000 (16:17 +0000)]
Add support for __DEFAULT_DEPENDENT_OPTIONS

9 years agoAdd options for auto obj and meta mode to sys.mk
sjg [Sun, 24 May 2015 02:46:08 +0000 (02:46 +0000)]
Add options for auto obj and meta mode to sys.mk

9 years agoMove include of auto.obj.mk to sys.mk
sjg [Sun, 24 May 2015 02:45:15 +0000 (02:45 +0000)]
Move include of auto.obj.mk to sys.mk

9 years agoAdd files to install
sjg [Sun, 24 May 2015 02:16:00 +0000 (02:16 +0000)]
Add files to install

9 years agoAdd STALE_STAGED as default no
sjg [Sun, 30 Nov 2014 18:51:10 +0000 (18:51 +0000)]
Add STALE_STAGED as default no

9 years agoIf we are doing META_MODE use targets/Makefile as top-level
sjg [Sun, 30 Nov 2014 18:50:15 +0000 (18:50 +0000)]
If we are doing META_MODE use targets/Makefile as top-level
since this one isn't suitable.

9 years agoupdate dependencies
sjg [Sun, 30 Nov 2014 18:48:57 +0000 (18:48 +0000)]
update dependencies

9 years agoNeed headers staged by libpam
sjg [Sun, 30 Nov 2014 18:48:17 +0000 (18:48 +0000)]
Need headers staged by libpam

9 years agoall is MAIN target
sjg [Sun, 30 Nov 2014 18:46:34 +0000 (18:46 +0000)]
all is MAIN target

9 years agoRenamed pkgs/ targets/
sjg [Sun, 30 Nov 2014 18:45:46 +0000 (18:45 +0000)]
Renamed pkgs/ targets/

9 years agoPut lib/csu/amd64/Makefile back the way it is in head
sjg [Sun, 30 Nov 2014 18:42:29 +0000 (18:42 +0000)]
Put lib/csu/amd64/Makefile back the way it is in head
and handle staging via ../Makefile.inc

9 years agoRename pkgs to more generic targets (that's what they are)
sjg [Sun, 30 Nov 2014 00:20:20 +0000 (00:20 +0000)]
Rename pkgs to more generic targets (that's what they are)

Reviewed by: marcel

9 years agoLeverage dirdeps.mk by tweaking _DIRDEP_USE_LEVEL
sjg [Sun, 23 Nov 2014 00:08:22 +0000 (00:08 +0000)]
Leverage dirdeps.mk by tweaking _DIRDEP_USE_LEVEL
so we can have a 2nd stage of dependencies computed in parallel.

Reviewed by: marcel

9 years agoUse latest version
sjg [Fri, 21 Nov 2014 19:56:27 +0000 (19:56 +0000)]
Use latest version

9 years agoUse latest version
sjg [Thu, 20 Nov 2014 22:42:40 +0000 (22:42 +0000)]
Use latest version

9 years agoRemoved from head
sjg [Wed, 19 Nov 2014 07:11:29 +0000 (07:11 +0000)]
Removed from head

9 years agoUpdated/new dependencies
sjg [Wed, 19 Nov 2014 07:10:38 +0000 (07:10 +0000)]
Updated/new dependencies

9 years agoUse OBJTOP if defined
sjg [Wed, 19 Nov 2014 07:09:43 +0000 (07:09 +0000)]
Use OBJTOP if defined

9 years agoInclude src.opts.mk since we don't include modules.inc here.
sjg [Wed, 19 Nov 2014 07:09:14 +0000 (07:09 +0000)]
Include src.opts.mk since we don't include modules.inc here.

9 years agoMerge from head@274682
sjg [Wed, 19 Nov 2014 01:07:58 +0000 (01:07 +0000)]
Merge from head@274682

9 years agofixed dep
sjg [Wed, 19 Nov 2014 00:44:21 +0000 (00:44 +0000)]
fixed dep

9 years agoMake vfs.zfs.max_recordsize read-write at runtime.
delphij [Tue, 18 Nov 2014 22:35:19 +0000 (22:35 +0000)]
Make vfs.zfs.max_recordsize read-write at runtime.

MFC after: 2 weeks

9 years agoAdd locking to wds(4) and mark MPSAFE.
jhb [Tue, 18 Nov 2014 22:12:51 +0000 (22:12 +0000)]
Add locking to wds(4) and mark MPSAFE.
- Add per-softc mutex.
- Use mutex for CAM SIM lock.
- Use bus_*() instead of inb() and outb().
- Use bus_alloc_resource_any() when reasonable.

Tested by: no one

9 years agoAdd locking to scd(4) and mark MPSAFE.
jhb [Tue, 18 Nov 2014 22:02:37 +0000 (22:02 +0000)]
Add locking to scd(4) and mark MPSAFE.
- Actually use existing per-softc mutex.
- Use mutex in cdev routines and remove D_NEEDGIANT.
- Use callout(9) instead of timeout(9).
- Don't check for impossible conditions (e.g. SCDINIT being clear).
- Use bus_*() instead of bus_space_*().

Tested by: no one

9 years agoCleanups to pst(4):
jhb [Tue, 18 Nov 2014 21:58:57 +0000 (21:58 +0000)]
Cleanups to pst(4):
- Use callout(9) instead of timeout(9).
- Don't bother zero'ing the softc.
- Destroy mutex on detach.

Tested by: no one

9 years agoAdd locking to mly(4) and mark MPSAFE.
jhb [Tue, 18 Nov 2014 21:55:50 +0000 (21:55 +0000)]
Add locking to mly(4) and mark MPSAFE.
- Add a per-softc mutex.
- Use mutex as CAM sim lock.
- Use taskqueue_thread instead of taskqueue_swi_giant.
- Use callout(9) instead of timeout(9).
- Use bus_*() instead of bus_space_*().

Tested by: no one

9 years agoAdd locking to mcd(4) and mark MPSAFE.
jhb [Tue, 18 Nov 2014 21:51:01 +0000 (21:51 +0000)]
Add locking to mcd(4) and mark MPSAFE.
- Actually use existing per-softc mutex.
- Use mutex in cdev routines and remove D_NEEDGIANT.
- Use callout(9) instead of timeout(9).
- Don't check for impossible conditions (e.g. MCDINIT being clear).
- Remove critical_enter/exit when sending a PIO command.
- Use bus_*() instead of bus_space_*().

Tested by: no one

9 years agoConvert the refire_notify_ack timer from timeout(9) to callout(9).
jhb [Tue, 18 Nov 2014 21:03:46 +0000 (21:03 +0000)]
Convert the refire_notify_ack timer from timeout(9) to callout(9).

Tested by: no one

9 years agoAdd a tunable for spa_slop_shift which controls how much space we
delphij [Tue, 18 Nov 2014 18:52:38 +0000 (18:52 +0000)]
Add a tunable for spa_slop_shift which controls how much space we
would reserve by default.  Tuning is not recommended.

MFC after: 2 weeks

9 years agoAllow tuning zfs_max_recordsize via loader tunable. Tuning is NOT
delphij [Tue, 18 Nov 2014 18:40:01 +0000 (18:40 +0000)]
Allow tuning zfs_max_recordsize via loader tunable.  Tuning is NOT
recommended.

Requested by: Slawa Olhovchenkov <slw zxy spb ru>
MFC after: 2 weeks

9 years agoUpgrade libxo to 0.1.6.
marcel [Tue, 18 Nov 2014 18:03:40 +0000 (18:03 +0000)]
Upgrade libxo to 0.1.6.

Summary of changes:
1.  Coverity defect fixes

Obtained from:  https://github.com/Juniper/libxo/releases/tag/0.1.6

9 years agoMake gpio_default_map_gpios() static. No functional changes.
loos [Tue, 18 Nov 2014 17:37:33 +0000 (17:37 +0000)]
Make gpio_default_map_gpios() static.  No functional changes.

9 years agoMoves all the duplicate code to a single function.
loos [Tue, 18 Nov 2014 17:22:08 +0000 (17:22 +0000)]
Moves all the duplicate code to a single function.

Verify for invalid modes and unwanted flags before pass the new flags to
driver.

9 years agoThese delays aren't needed. Elimate them. They should be on the order
imp [Tue, 18 Nov 2014 17:07:02 +0000 (17:07 +0000)]
These delays aren't needed. Elimate them. They should be on the order
of a few cycles at most, not 10us. They make it impossible to
implement half-duplex protocols that are faster than about 1KHz.

Sponsored by: Netflix

9 years agoopt_global.h is included automatically in the build. No need to
imp [Tue, 18 Nov 2014 17:06:56 +0000 (17:06 +0000)]
opt_global.h is included automatically in the build. No need to
explicitly include it in these places.

Sponsored by: Netflix

9 years agoNo need to list opt_global.h. No need to force ALTQ here, since that's
imp [Tue, 18 Nov 2014 17:06:54 +0000 (17:06 +0000)]
No need to list opt_global.h. No need to force ALTQ here, since that's
not the normal default for the system and nothing else forces it on.

Sponsored by: Netflix

9 years agoopt_global.h is never needed in SRCS lists.
imp [Tue, 18 Nov 2014 17:06:52 +0000 (17:06 +0000)]
opt_global.h is never needed in SRCS lists.

Sponsored by: Netflix

9 years agoDocument WITHOUT_USB_EXAMPLES and WITHOUT_TESTS_SUPPORT
imp [Tue, 18 Nov 2014 17:06:50 +0000 (17:06 +0000)]
Document WITHOUT_USB_EXAMPLES and WITHOUT_TESTS_SUPPORT

Sponsored by: Netflix

9 years agoLIB32 is supported for powerpc64 too. Make the note more general about
imp [Tue, 18 Nov 2014 17:06:48 +0000 (17:06 +0000)]
LIB32 is supported for powerpc64 too. Make the note more general about
64-bit processors.

Sponsored by: Netflix

9 years agoFix typo pointed out by avg@ and Joerg Sonnenberger. Add a clarifying
imp [Tue, 18 Nov 2014 17:06:46 +0000 (17:06 +0000)]
Fix typo pointed out by avg@ and Joerg Sonnenberger. Add a clarifying
sentence too.

Sponsored by: Netflix

9 years agoAcutally, replaces NO_INCS with WITHOUT_INCLUDES and make
imp [Tue, 18 Nov 2014 17:06:44 +0000 (17:06 +0000)]
Acutally, replaces NO_INCS with WITHOUT_INCLUDES and make
WITHOUT_TOOLCHAIN imply it instead.

Sponsored by: Netflix

9 years agoImplement the historic DIOCGDINFO ioctl for gpart on BSD
imp [Tue, 18 Nov 2014 17:06:40 +0000 (17:06 +0000)]
Implement the historic DIOCGDINFO ioctl for gpart on BSD
partitions. Several utilities still use this interface and require
additional information since gpart was activated than before. This
allows fsck of a UFS partition without having to specify it is UFS,
per historic behavior.

9 years agoMissed a terminating ';'.
jhibbits [Tue, 18 Nov 2014 16:31:00 +0000 (16:31 +0000)]
Missed a terminating ';'.

X-MFC-with: r273009

9 years agoAdd Altera Parallel IO (PIO) device driver.
br [Tue, 18 Nov 2014 14:12:19 +0000 (14:12 +0000)]
Add Altera Parallel IO (PIO) device driver.

Sponsored by: DARPA, AFRL

9 years agoAdd Virtio MMIO bus driver.
br [Tue, 18 Nov 2014 14:11:14 +0000 (14:11 +0000)]
Add Virtio MMIO bus driver.

Sponsored by: DARPA, AFRL

9 years agoRe-work non-persistent filesystem detection as it was not possible to
feld [Tue, 18 Nov 2014 13:38:09 +0000 (13:38 +0000)]
Re-work non-persistent filesystem detection as it was not possible to
detect /dev/md backed mfs filesystems that way.

Differential Revision: https://reviews.freebsd.org/D1163
Approved by: ian

9 years agoCollapse three contiguous comment blocks into one. Remove historical
glebius [Tue, 18 Nov 2014 13:38:07 +0000 (13:38 +0000)]
Collapse three contiguous comment blocks into one. Remove historical
note about wrong assumptions 20 years ago. Use proper casing.

Sponsored by: Nginx, Inc.

9 years agoFix the i2c bus speed divisors for TI OMAP4 and AM335x.
ian [Tue, 18 Nov 2014 03:26:52 +0000 (03:26 +0000)]
Fix the i2c bus speed divisors for TI OMAP4 and AM335x.

For OMAP4, the old values for 1MHz gave a bus frequency of about 890KHz.
The new numbers hit 1MHz exactly.

For AM335x the prescaler values are adjusted to give a 24MHz clock for
all 3 standard speeds, as the manual recommends (as near as we can tell,
there are errors and typos apparent in the document).  Also, 1MHz speed
is added, and has been tested successfully on a BeagleboneWhite board.

PR: 195009

9 years agoFix gpiobus_child_location_str() to return a real string with the mapped
loos [Tue, 18 Nov 2014 02:41:35 +0000 (02:41 +0000)]
Fix gpiobus_child_location_str() to return a real string with the mapped
pins.

Make gpiobus_print_pins() static again.

9 years agoRemove unnecessary code.
loos [Tue, 18 Nov 2014 02:11:07 +0000 (02:11 +0000)]
Remove unnecessary code.

After r273566, the gpiobus version of bus_print_child() also works on FDT
systems.

9 years agoAllow i2c bus speed to be configured via hints, FDT data, and sysctl.
ian [Tue, 18 Nov 2014 01:54:31 +0000 (01:54 +0000)]
Allow i2c bus speed to be configured via hints, FDT data, and sysctl.

The current support for controlling i2c bus speed is an inconsistant mess.
There are 4 symbolic speed values defined, UNKNOWN, SLOW, FAST, FASTEST.
It seems to be universally assumed that SLOW means the standard 100KHz
rate from the original spec.  Nothing ever calls iicbus_reset() with a
speed of FAST, although some drivers would treat it as the 400KHz standard
speed.  Mostly iicbus_reset() is called with the speed set to UNKNOWN or
FASTEST, and there's really no telling what any individual driver will do
with those.

The speed of an i2c bus is limited by the speed of the slowest device on
the bus.  This means that generally the bus speed needs to be configured
based on the board/system and the components within it.  Historically for
i2c we've configured with device hints.  Newer systems use FDT data and it
documents a clock-frequency property for i2c busses.  Hobbyists and
developers are likely to want on the fly changes.  These changes provide
all 3 methods, but do not require any existing drivers to change to use
the new facilities.

This adds an iicbus method, iicbus_get_frequency(dev, speed) that gets the
frequency for the requested symbolic speed.  If the symbolic speed is SLOW
or if there is no speed configured for the bus, the returned value is
100KHz, always.  Otherwise, if bus speed is configured by hints, fdt,
tunable, or sysctl, that speed is returned.  It also adds a helper
function, iicbus_init_frequency() that any bus driver subclassed from
iicbus can initialize the frequency from some other source of info.

Initial driver implementations are provided for Freescale and TI.

Differential Revision:        https://reviews.freebsd.org/D1174
PR: 195009

9 years agoUse automated MK_INET*_SUPPORT code here as well.
imp [Tue, 18 Nov 2014 01:39:23 +0000 (01:39 +0000)]
Use automated MK_INET*_SUPPORT code here as well.

9 years agoModernize comments about BIOSes being lame since in this detail they
imp [Tue, 18 Nov 2014 01:39:21 +0000 (01:39 +0000)]
Modernize comments about BIOSes being lame since in this detail they
aren't lame, the rules changed along the way. Catch up to 1999 or so
with the new rules.

9 years agoAdd basic interrupt management code to gpiobus and ofw_gpiobus.
loos [Tue, 18 Nov 2014 01:38:39 +0000 (01:38 +0000)]
Add basic interrupt management code to gpiobus and ofw_gpiobus.

This is the general support to allow the use of GPIO pins as interrupt
sources for direct gpiobus children.

The use of GPIO pins as generic interrupt sources (for an ethernet driver
for example) will only be possible when arm/intrng is complete.  Then, most
of this code will need to be rewritten, but it works for now, is better
than what we have and will allow further developments.

Tested on: ar71xx (RSPRO), am335x (BBB), bcm2835 (Raspberry pi)
Differential Revision: https://reviews.freebsd.org/D999
Reviewed by: rpaulo

9 years agoDTrace imposes a 128-byte limit on the length of the function component of
markj [Mon, 17 Nov 2014 22:22:16 +0000 (22:22 +0000)]
DTrace imposes a 128-byte limit on the length of the function component of
a probe name. When dtrace -G builds up a DOF section for the specified
provider(s), the probe function names are truncated to fit in this limit.
The DOF is later used to build the symbol table for the generated object
file, so the table can end up with truncated references, causing link
errors.

Instead of potentially truncating symbol table entries, write the full
function name to the DOF string table and allow the kernel to enforce the
128-byte function name limit when a process attempts to load its DOF.

PR: 194757
Differential Revision: https://reviews.freebsd.org/D1175
Reviewed by: rpaulo
MFC after: 2 weeks

9 years agoSync with userland variant.
delphij [Mon, 17 Nov 2014 21:01:35 +0000 (21:01 +0000)]
Sync with userland variant.

MFC after: 2 weeks

9 years agoUse the correct device as the power_for_sleep() method
dchagin [Mon, 17 Nov 2014 20:25:21 +0000 (20:25 +0000)]
Use the correct device as the power_for_sleep() method
always pass request up to parent bridge.

Reviewed by: jhb
MFC after: 1 week
xMFC: r274386,r274397

9 years agoRemove stray empty comment. The code is adequately explained in the
imp [Mon, 17 Nov 2014 16:30:51 +0000 (16:30 +0000)]
Remove stray empty comment. The code is adequately explained in the
block comment above, so there's nothing to add here.

9 years agoClean up some language
brd [Mon, 17 Nov 2014 15:44:00 +0000 (15:44 +0000)]
Clean up some language

Reviewed by: will

9 years agoFix geom's "usage" generation to not fabricate usage/help output for any
feld [Mon, 17 Nov 2014 15:19:57 +0000 (15:19 +0000)]
Fix geom's "usage" generation to not fabricate usage/help output for any
imaginary class we give it.

Differential Revision: https://reviews.freebsd.org/D1150
Submitted by: homerj
Approved by: pjd

9 years agoAdd logic for detecting non-persistent filesystems being utilized by
feld [Mon, 17 Nov 2014 15:12:46 +0000 (15:12 +0000)]
Add logic for detecting non-persistent filesystems being utilized by
workdir which would break the upgrade process upon reboot.

Currently we check for tmpfs and mdmfs.

PR: 195006
Differential Revision: https://reviews.freebsd.org/D1163
Approved by: cperciva

9 years agoRe-introduce padding fields removed with r264321 to keep
jch [Mon, 17 Nov 2014 14:56:02 +0000 (14:56 +0000)]
Re-introduce padding fields removed with r264321 to keep
struct tcptw ABI unchanged.

Suggested by: jhb
Approved by: jhb (mentor)
MFC after: 1 day
X-MFC-With: r264321

9 years agol2arc: restore correct rounding up of asize of compressed data
avg [Mon, 17 Nov 2014 14:45:42 +0000 (14:45 +0000)]
l2arc: restore correct rounding up of asize of compressed data

This rounding up was lost in a mismerge of illumos code.
See r268075 MFV r267565.
After that commit zio_compress_data() no longer performs any compressed
size adjustment, so it needs to be done externally.  On FreeBSD we round
up the size using vdev_ashift rather than SPA_MINBLOCKSIZE so that 4KB
devices are properly supported.

Additionally, zero out the buffer tail only if compression succeeds.
The compression is considered successful if the size of compressed
data after rounding up to account for the vdev ashift is less than the
original data size.  It does not make sense to have the data compressed
if all the savings are lost to rounding up.
With the new zio_compress_data() it could have been possible that the
rounded compressed size would be greater than the original size and thus
we could zero beyond the allocated buffer if the zeroing code was kept
at the original place.

Discussed with: delphij, gibbs
MFC after: 2 weeks
X-MFC with: r274627

9 years agoRevert r269093 which introduced physical zio alignment transform
avg [Mon, 17 Nov 2014 14:16:02 +0000 (14:16 +0000)]
Revert r269093 which introduced physical zio alignment transform

Size of physical ZIOs must never be implicitly adjusted, it's
a responsibility of a caller to make sure that such a ZIO has proper offset
and size.

Discussed with: delphij, gibbs
MFC after: 2 weeks

9 years agoMechanically replace #if defined(__FreeBSD__) and #if defined(__NetBSD__) with
ngie [Mon, 17 Nov 2014 13:39:00 +0000 (13:39 +0000)]
Mechanically replace #if defined(__FreeBSD__) and #if defined(__NetBSD__) with
their #ifdef equivalents for everything changed in contrib/netbsd-tests. There
are some items from the vendor tree that use #if defined(__FreeBSD__) or
#if defined(__NetBSD__) which are being left alone

Requested by: bde, rpaulo
Sponsored by: EMC / Isilon Storage Division

9 years agoThe "intr" option is NFS-specific; fix examples in auto_master(5).
trasz [Mon, 17 Nov 2014 12:36:01 +0000 (12:36 +0000)]
The "intr" option is NFS-specific; fix examples in auto_master(5).

MFC after: 1 month
Sponsored by: The FreeBSD Foundation

9 years agoRevert r273630 as the panic was fixed by r274619
smh [Mon, 17 Nov 2014 11:35:30 +0000 (11:35 +0000)]
Revert r273630 as the panic was fixed by r274619

The panic was caused by TRIM requests run against file based vdevs as write
requests.

PR: 191573
Sponsored by: Multiplay

9 years agoDisable TRIM on file backed ZFS vdevs and fix TRIM on init
smh [Mon, 17 Nov 2014 11:32:10 +0000 (11:32 +0000)]
Disable TRIM on file backed ZFS vdevs and fix TRIM on init

After r265152 TRIM requests are ZIO_TYPE_FREE instead of ZIO_TYPE_IOCTL
this meant file backed vdevs to attempted to process the ZIO as a write
causing a panic.

We now disable TRIM on file backed vdevs and ASSERT the ZIO types supported
by each vdev type to ensure we explicity support the ZIO type being
processed.

Also ensure that TRIM on init is not procesed for devices which declare they
didn't support TRIM via vdev_notrim.

PR: 195061, 194976, 191573
Sponsored by: Multiplay

9 years agoAdd reachover Makefiles for contrib/netbsd-tests/lib/librt
ngie [Mon, 17 Nov 2014 09:19:09 +0000 (09:19 +0000)]
Add reachover Makefiles for contrib/netbsd-tests/lib/librt

A variant of this code has been tested on amd64/i386 for some time by
EMC/Isilon on 10-STABLE/11-CURRENT. It builds on other architectures, but the
code will remain off until it's proven it works on virtual hardware or real
hardware on other architectures

Sponsored by: EMC / Isilon Storage Division

9 years agoFinish r274175: do control plane MTU tracking.
melifaro [Mon, 17 Nov 2014 01:05:29 +0000 (01:05 +0000)]
Finish r274175: do control plane MTU tracking.

Update route MTU in case of ifnet MTU change.
Add new RTF_FIXEDMTU to track explicitly specified MTU.

Old behavior:
ifconfig em0 mtu 1500->9000 -> all routes traversing em0 do not change MTU.
User has to manually update all routes.
ifconfig em0 mtu 9000->1500 -> all routes traversing em0 do not change MTU.
However, if ip[6]_output finds route with rt_mtu > interface mtu, rt_mtu
gets updated.

New behavior:
ifconfig em0 mtu 1500->9000 -> all interface routes in all fibs gets updated
with new MTU unless RTF_FIXEDMTU flag set on them.
ifconfig em0 mtu 9000->1500 -> all routes in all fibs gets updated with new
MTU unless RTF_FIXEDMTU flag set on them AND rt_mtu is less than ifp mtu.

route add ... -mtu XXX automatically sets RTF_FIXEDMTU flag.
route change .. -mtu 0 automatically removes RTF_FIXEDMTU flag.

PR: 194238
MFC after: 1 month
CR: D1125

9 years agoAdd mv, echo and hostname to the native-xtools target. Missed these
sbruno [Sun, 16 Nov 2014 23:31:23 +0000 (23:31 +0000)]
Add mv, echo and hostname to the native-xtools target.  Missed these
when profiling.

MFC after: 2 weeks

9 years agoNo functional changes. Remove a couple outdated or inane comments and
ian [Sun, 16 Nov 2014 21:39:56 +0000 (21:39 +0000)]
No functional changes.  Remove a couple outdated or inane comments and
add new comment blocks describing why the cache maintenance sequences are
done in the order they are for each case.

9 years agoCorrect the sequence of busdma sync ops involved with PRE/POSTREAD syncs.
ian [Sun, 16 Nov 2014 21:22:42 +0000 (21:22 +0000)]
Correct the sequence of busdma sync ops involved with PRE/POSTREAD syncs.

We used to invalidate the cache for PREREAD alone, or writeback+invalidate
for PREREAD with PREWRITE, then treat POSTREAD as a no-op.  Prefetching on
modern systems can lead to parts of a DMA buffer getting pulled into the
caches while DMA is in progress (due to access of "nearby" data), so it's
mandatory to invalidate during the POSTREAD sync even if a PREREAD
invalidate also happened.

In the PREREAD case the invalidate is done to ensure that there are no
dirty cache lines that might get automatically evicted during the DMA,
corrupting the buffer.  In a PREREAD+PREWRITE case the writeback which is
required for PREWRITE handling is suffficient to avoid corruption caused
by eviction and no invalidate need be done until POSTREAD time.

Submitted by: Michal Meloun <meloun@miracle.cz>

9 years agoDo the cache invalidate sequence from the outermost to innermost, required
ian [Sun, 16 Nov 2014 20:59:27 +0000 (20:59 +0000)]
Do the cache invalidate sequence from the outermost to innermost, required
for correct operation.

Submitted by: Michal Meloun <meloun@miracle.cz>

9 years agoDo not do a cache invalidate on a PREREAD sync that is also a PREWRITE sync.
ian [Sun, 16 Nov 2014 20:55:51 +0000 (20:55 +0000)]
Do not do a cache invalidate on a PREREAD sync that is also a PREWRITE sync.
The PREWRITE handling does a writeback of any dirty cachelines, so there's
no danger of an eviction during the DMA corrupting the buffer.  There will
be an invalidate done during POSTREAD, so doing it before the read too is
wasted time.

9 years ago- Skip over the testcases that call cbrtl on platforms where LDBL_PREC == 53
ngie [Sun, 16 Nov 2014 20:42:30 +0000 (20:42 +0000)]
- Skip over the testcases that call cbrtl on platforms where LDBL_PREC == 53
(arm, mips, powerpc). This fixes the build on these platforms, based on some
ad hoc tinderbox runs I did a while ago
- Skip cast the arguments to powl as long double so powl properly interprets
those arugments at compile-time when picking the type

Sponsored by: EMC / Isilon Storage Division

9 years ago- Expect exp2_powers to fail on FreeBSD/i386
ngie [Sun, 16 Nov 2014 20:39:11 +0000 (20:39 +0000)]
- Expect exp2_powers to fail on FreeBSD/i386
- Expect exp2_values to fail on FreeBSD due to the small epsilon

Sponsored by: EMC / Isilon Storage Division

9 years agoAlias isinff to isinf on FreeBSD
ngie [Sun, 16 Nov 2014 20:37:46 +0000 (20:37 +0000)]
Alias isinff to isinf on FreeBSD

isinf on FreeBSD automatically picks the appropriate type per math.h

Sponsored by: EMC / Isilon Storage Division

9 years agoReset errno to 0 before running scalbn to be sure that the tested errno is
ngie [Sun, 16 Nov 2014 20:35:32 +0000 (20:35 +0000)]
Reset errno to 0 before running scalbn to be sure that the tested errno is
valid

Sponsored by: EMC / Isilon Storage Division

9 years agoSkip the long-double epsilon checks on FreeBSD/i386
ngie [Sun, 16 Nov 2014 20:34:46 +0000 (20:34 +0000)]
Skip the long-double epsilon checks on FreeBSD/i386

Sponsored by: EMC / Isilon Storage Division

9 years agoIndent a couple lines properly and expand a comment. No functional changes.
ian [Sun, 16 Nov 2014 20:34:14 +0000 (20:34 +0000)]
Indent a couple lines properly and expand a comment.  No functional changes.

9 years agoConvert tools/regression/lib/libc/stdio/test-fpclassify into an ATF testcase and
ngie [Sun, 16 Nov 2014 20:29:57 +0000 (20:29 +0000)]
Convert tools/regression/lib/libc/stdio/test-fpclassify into an ATF testcase and
Rename as lib/libc/stdio/fpclassify2_test

Sponsored by: EMC / Isilon Storage Division

9 years agoFlush out many more static binaries that can be used in a jail for building
sbruno [Sun, 16 Nov 2014 20:10:37 +0000 (20:10 +0000)]
Flush out many more static binaries that can be used in a jail for building
pkgs via poudriere.  This removes a significant amount of emulation speeding
up run times.

Phabric: D1172
Reviewed by: imp
MFC after: 2 weeks

9 years agoConvert tools/regression/lib/libc/stdio/test-fmemopen into an ATF testcase and
ngie [Sun, 16 Nov 2014 19:57:47 +0000 (19:57 +0000)]
Convert tools/regression/lib/libc/stdio/test-fmemopen into an ATF testcase and
rename as lib/libc/stdio/fmemopen2_test

Sponsored by: EMC / Isilon Storage Division

9 years agoAdd freebsd-testing and myself to MAINTAINERS for contrib/netbsd-tests and
ngie [Sun, 16 Nov 2014 19:47:29 +0000 (19:47 +0000)]
Add freebsd-testing and myself to MAINTAINERS for contrib/netbsd-tests and
contrib/pjdfstest

9 years agoRevert r274585: rte lock is properly destroyed in uma dtor callback.
melifaro [Sun, 16 Nov 2014 18:15:23 +0000 (18:15 +0000)]
Revert r274585: rte lock is properly destroyed in uma dtor callback.

Pointed by: glebius

9 years agoFix r273112: do not turn DROP_EN by default.
melifaro [Sun, 16 Nov 2014 18:08:00 +0000 (18:08 +0000)]
Fix r273112: do not turn DROP_EN by default.

Due to adapter->hw.fc.requested_mode is filled with default value
after ixgbe_initialize_receive_units(), this leads to enabling
DROP_EN in most cases.

Tested by: ae
MFC after: 1 week

9 years agodrm/radeon: Lower priority of a message related to invalid EDID
dumbbell [Sun, 16 Nov 2014 17:53:48 +0000 (17:53 +0000)]
drm/radeon: Lower priority of a message related to invalid EDID

Like in r259717, the prority goes from "error" to "debug" to avoid
spamming logs when the connectors are polled.

PR: 194770
Submitted by: Larry Rosenman <ler@lerctr.org>
MFC after: 1 week

9 years agoMake witness happy: destroy rte lock before free.
melifaro [Sun, 16 Nov 2014 14:56:31 +0000 (14:56 +0000)]
Make witness happy: destroy rte lock before free.

MFC after: 2 weeks

9 years agoDefault to use 10 seconds as nap interval instead of 1.
delphij [Sun, 16 Nov 2014 09:44:30 +0000 (09:44 +0000)]
Default to use 10 seconds as nap interval instead of 1.

Previously, we have a nap interval of 1 second while we have a timeout of
128 seconds by default, which could be an overkill, and for some hardware
the patting action may be expensive.

Note that the choice of nap interval is still arbitrary.  We preferred
a safe value where even when the system is very heavily loaded, the
watchdog should not shoot the system down if it's not really hung.
According to the manual page of Linux's watchdog daemon, the nap interval
time of theirs is 10 seconds, which seems to be a reasonable value --
according to Intel documentation AP-725 (Document Number: 292273-001),
ICH5's maximum timeout is about 37.5 seconds, which the ichwd(4) driver
would set when we requested 128 seconds (although it should probably
feed back this as an error and do not set the timeout).  Since that's
the shortest maximum value, 10 seconds seems to be a right choice for
us too.

Discussed with: alfred
MFC after: 1 month

9 years agoRemove test-arc4random from this Makefile so others can continue to use
ngie [Sun, 16 Nov 2014 07:58:06 +0000 (07:58 +0000)]
Remove test-arc4random from this Makefile so others can continue to use
this as-is for the time being

9 years agoConvert tools/regression/lib/libc/gen/test-arc4random into an ATF testcase and
ngie [Sun, 16 Nov 2014 07:55:28 +0000 (07:55 +0000)]
Convert tools/regression/lib/libc/gen/test-arc4random into an ATF testcase and
rename as lib/libc/gen/arc4random_test

Sponsored by: EMC / Isilon Storage Division

9 years agoAdd reachover Makefiles for contrib/netbsd-tests/lib/librt
ngie [Sun, 16 Nov 2014 07:03:19 +0000 (07:03 +0000)]
Add reachover Makefiles for contrib/netbsd-tests/lib/librt

A variant of this code has been tested on amd64/i386 for some time by
EMC/Isilon on 10-STABLE/11-CURRENT. It builds on other architectures, but the
code will remain off until it's proven it works on virtual hardware or real
hardware on other architectures

Sponsored by: EMC / Isilon Storage Division

9 years agoCall sem_unlink on semaphores before attempting to create them
ngie [Sun, 16 Nov 2014 06:59:58 +0000 (06:59 +0000)]
Call sem_unlink on semaphores before attempting to create them

Due to the lack of uniqueness in the semaphore name, and the fact that the
tests don't have cleanup routines, an interrupted test can leave a semaphore
"laying around", causing all subsequent attempts to run the test to fail

I will file a NetBSD PR for this issue soon

9 years agoAdd reachover Makefiles for contrib/netbsd-tests/lib/libpthread as
ngie [Sun, 16 Nov 2014 06:35:20 +0000 (06:35 +0000)]
Add reachover Makefiles for contrib/netbsd-tests/lib/libpthread as
lib/libthr/tests

A variant of this code has been tested on amd64/i386 for some time by
EMC/Isilon on 10-STABLE/11-CURRENT. It builds on other architectures, but the
code will remain off until it's proven it works on virtual hardware or real
hardware on other architectures

Original work by: pho
Sponsored by: EMC / Isilon Storage Division

9 years agoAdd missing sys/time.h #include for timespecsub macro in lib/libnetbsd/sys/time.h
ngie [Sun, 16 Nov 2014 05:13:39 +0000 (05:13 +0000)]
Add missing sys/time.h #include for timespecsub macro in lib/libnetbsd/sys/time.h

9 years agoPort helper program to FreeBSD, similar to ../../lib/libc/stdlib/h_atexit.c
ngie [Sun, 16 Nov 2014 05:13:10 +0000 (05:13 +0000)]
Port helper program to FreeBSD, similar to ../../lib/libc/stdlib/h_atexit.c

Submitted by: pho
In collaboration with: kib

9 years ago#ifdef out a printf on !NetBSD that causes the testcase to fail when comparing
ngie [Sun, 16 Nov 2014 05:11:07 +0000 (05:11 +0000)]
#ifdef out a printf on !NetBSD that causes the testcase to fail when comparing
the output from the helper program

Submitted by: pho