freebsd.git
8 years agoUpdate from svn-1.8.14 to 1.9.2.
peter [Mon, 12 Oct 2015 09:53:55 +0000 (09:53 +0000)]
Update from svn-1.8.14 to 1.9.2.

Formal release notes are available:
  https://subversion.apache.org/docs/release-notes/1.9.html

Of particular note, the client checkout format has *not* changed so
upgrades should *not* be required.

When reading a repository (file:// or running as a local server), an
improved fsfs version 7 is available with significant performance
improvements.  An optional upgrade is possible to use the new features.
Without the upgrade, this is fully read/write compatible with the
version 6 fsfs as in svn-1.8.

Relnotes: yes

8 years agoVendor import of subversion-1.9.2
peter [Mon, 12 Oct 2015 08:54:49 +0000 (08:54 +0000)]
Vendor import of subversion-1.9.2

8 years agoAccept any correct frames from any source when MONITOR mode is used.
kevlo [Mon, 12 Oct 2015 08:17:21 +0000 (08:17 +0000)]
Accept any correct frames from any source when MONITOR mode is used.

Submitted by: Andriy Voskoboinyk <s3erios at gmail.com>
Differential Revision: https://reviews.freebsd.org/D3812

8 years agoRefactor the test/ Makefiles after recent changes to bsd.test.mk (r289158) and
ngie [Mon, 12 Oct 2015 08:16:03 +0000 (08:16 +0000)]
Refactor the test/ Makefiles after recent changes to bsd.test.mk (r289158) and
netbsd-tests.test.mk (r289151)

- Eliminate explicit OBJTOP/SRCTOP setting
- Convert all ad hoc NetBSD test integration over to netbsd-tests.test.mk
- Remove unnecessary TESTSDIR setting
- Use SRCTOP where possible for clarity

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

8 years agonet80211 drivers: eliminate any references to sc_rxtap_len/sc_txtap_len (never used...
adrian [Mon, 12 Oct 2015 05:21:51 +0000 (05:21 +0000)]
net80211 drivers: eliminate any references to sc_rxtap_len/sc_txtap_len (never used here)

Submitted by: <s3erios@gmail.com>
Differential Revision: https://reviews.freebsd.org/D3776

8 years agourtwn(4): split *reg and *var parts (no functional change).
adrian [Mon, 12 Oct 2015 05:14:49 +0000 (05:14 +0000)]
urtwn(4): split *reg and *var parts (no functional change).

Submitted by: <s3erios@gmail.com>
Differential Revision: https://reviews.freebsd.org/D3775

8 years agonet80211: move ieee80211_free_node() call on error from ic_raw_xmit() to ieee80211_ra...
adrian [Mon, 12 Oct 2015 04:55:20 +0000 (04:55 +0000)]
net80211: move ieee80211_free_node() call on error from ic_raw_xmit() to ieee80211_raw_output().

This doesn't free the mbuf upon error; the driver ic_raw_xmit method is still
doing that.

Submitted by: <s3erios@gmail.com>
Differential Revision: https://reviews.freebsd.org/D3774

8 years agonet80211: free node reference in the ieee80211_parent_xmitpkt() when error happened.
adrian [Mon, 12 Oct 2015 04:30:38 +0000 (04:30 +0000)]
net80211: free node reference in the ieee80211_parent_xmitpkt() when error happened.

Move error handling into ieee80211_parent_xmitpkt() instead of spreading it
between functions.

Submitted by: <s3erios@gmail.com>
Differential Revision: https://reviews.freebsd.org/D3772

8 years agowpi(4): add support for TX fragmentation.
adrian [Mon, 12 Oct 2015 04:05:12 +0000 (04:05 +0000)]
wpi(4): add support for TX fragmentation.

Tested:

* Tested with Intel 3945BG, HOSTAP and STA modes

Differential Revision: https://reviews.freebsd.org/D3770

8 years agonet80211: separate mbuf cleanup from ieee80211_fragment()
adrian [Mon, 12 Oct 2015 03:27:08 +0000 (03:27 +0000)]
net80211: separate mbuf cleanup from ieee80211_fragment()

* Create ieee80211_free_mbuf() which frees a list of mbufs.
* Use it in the fragment transmit path and ath / uath transmit paths.
* Call it in xmit_pkt() if the transmission fails; otherwise fragments
  may be leaked.

This should be a big no-op.

Submitted by: <s3erios@gmail.com>
Differential Revision: https://reviews.freebsd.org/D3769

8 years agoRevert r289150:
pfg [Mon, 12 Oct 2015 02:05:25 +0000 (02:05 +0000)]
Revert r289150:
Correct handling of enum attributes with g++

It is causing issues on some platforms.

Reported by: sbruno (through adrian)

8 years agosh: Make struct arglist an array instead of a linked list.
jilles [Sun, 11 Oct 2015 21:33:00 +0000 (21:33 +0000)]
sh: Make struct arglist an array instead of a linked list.

This simplifies the code (e.g. allowing use of qsort(3) instead of a
hand-rolled mergesort) and should have better cache properties.

The waste of unused args arrays after resizes is approximately the same as
the savings from getting rid of the next pointers.

At the same time, remove a piece of global state and move some duplicated
code into a function.

8 years agoDefault TESTSDIR to /usr/tests/${RELDIR:H}
ngie [Sun, 11 Oct 2015 21:29:24 +0000 (21:29 +0000)]
Default TESTSDIR to /usr/tests/${RELDIR:H}

When run from bin/ls/tests, for example, the value of TESTSDIR would be
${TESTSBASE}/${RELDIR:H} -> /usr/tests/bin/ls/tests/.. ->
/usr/tests/bin/ls

Document the new behavior in bsd.README.

While here, also document TESTSBASE

Relnotes: yes
Differential Revision: D1022
MFC after: 2 weeks
Sponsored by: EMC / Isilon Storage Division

8 years agoNTB: MFV c336acd3: memcpy lockup workaround
cem [Sun, 11 Oct 2015 21:01:14 +0000 (21:01 +0000)]
NTB: MFV c336acd3: memcpy lockup workaround

The system will appear to lockup for long periods of time due to the NTB
driver spending too much time in memcpy.  Avoid this by reducing the
number of packets that can be serviced on a given interrupt.

Authored by: Jon Mason
Obtained from: Linux
Sponsored by: EMC / Isilon Storage Division

8 years agoNTB: MFV c9d534c8: Correctly handle receive buffers of the minimal size
cem [Sun, 11 Oct 2015 21:00:19 +0000 (21:00 +0000)]
NTB: MFV c9d534c8: Correctly handle receive buffers of the minimal size

The ring logic of the NTB receive buffer/transmit memory window requires
there to be at least 2 payload sized allotments.  For the minimal size
case, split the buffer into two and set the transport_mtu to the
appropriate size.

Authored by: Jon Mason
Obtained from: Linux
Sponsored by: EMC / Isilon Storage Division

8 years agoNTB: MFV 90f9e934: reset tx_index on link toggle
cem [Sun, 11 Oct 2015 20:59:40 +0000 (20:59 +0000)]
NTB: MFV 90f9e934: reset tx_index on link toggle

If the NTB link toggles, the driver could stop receiving due to the
tx_index not being set to 0 on the transmitting size on a link-up event.
This is due to the driver expecting the incoming data to start at the
beginning of the receive buffer and not at a random place.

Authored by: Jon Mason
Obtained from: Linux
Sponsored by: EMC / Isilon Storage Division

8 years agoNTB: MFV b77b2637: Link toggle memory leak
cem [Sun, 11 Oct 2015 20:59:02 +0000 (20:59 +0000)]
NTB: MFV b77b2637: Link toggle memory leak

Each link-up will allocate a new NTB receive buffer when the NTB
properties are negotiated with the remote system.  These allocations did
not check for existing buffers and thus did not free them.  Now, the
driver will check for an existing buffer and free it if not of the
correct size, before trying to alloc a new one.

Authored by: Jon Mason
Obtained from: Linux
Sponsored by: EMC / Isilon Storage Division

8 years agoNTB: MFV 113fc505: Handle 64bit BAR sizes
cem [Sun, 11 Oct 2015 20:58:00 +0000 (20:58 +0000)]
NTB: MFV 113fc505: Handle 64bit BAR sizes

64bit BAR sizes are permissible with an NTB device.  To support them
various modifications and clean-ups were required, most significantly
using 2 32bit scratch pad registers for each BAR.

Also, modify the driver to allow more than 2 Memory Windows.

Authored by: Jon Mason
Obtained from: Linux
Sponsored by: EMC / Isilon Storage Division

8 years agoNTB: MFV cc0f868d: fix pointer math issues
cem [Sun, 11 Oct 2015 20:57:09 +0000 (20:57 +0000)]
NTB: MFV cc0f868d: fix pointer math issues

->remote_rx_info and ->rx_info are struct ntb_rx_info pointers.  If we
add sizeof(struct ntb_rx_info) then it goes too far.

Authored by: Dan Carpenter
Obtained from: Linux
Sponsored by: EMC / Isilon Storage Division

8 years agoSimplify netbsd-tests.test.mk
ngie [Sun, 11 Oct 2015 20:02:10 +0000 (20:02 +0000)]
Simplify netbsd-tests.test.mk

- projects/bmake and subsequent commits provide SRCTOP; there's no need to
  manually specify it now.
- Compute a sane default for OBJTOP based on .OBJDIR and RELDIR. Manually
  specifying this is probably no longer needed, but it persists just in case
  (supporting commits will need to be made to move it out of some of the meta
  .mk files).
- Compute a sane default for TESTSRC. Error out if the path cannot be found.

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

8 years agoCorrect handling of enum attributes with g++
pfg [Sun, 11 Oct 2015 19:58:57 +0000 (19:58 +0000)]
Correct handling of enum attributes with g++

From OpenBSD's commit log:

This was responsible for memory corruption with recent versions
of Mesa where c and c++ code share a header with a packed enum type.

Reference:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=39219

Obtained from: OpenBSD (CVS rev. 1.2)
MFC after: 1 week

8 years agoUse existing CONFDIR instead of adding a new CONFIGDIR
bapt [Sun, 11 Oct 2015 19:25:39 +0000 (19:25 +0000)]
Use existing CONFDIR instead of adding a new CONFIGDIR

Reported by: jbeich

8 years agoCorrect flash layout (this is a 4M flash unit).
sbruno [Sun, 11 Oct 2015 18:37:29 +0000 (18:37 +0000)]
Correct flash layout (this is a 4M flash unit).

Remove "rootfs" entry and assign the 800K or so to the kernel
partition as this unit boots from usb mass storage.

8 years agoMake delete method set via kern.cam.da.X.delete_method persistent.
mav [Sun, 11 Oct 2015 18:26:06 +0000 (18:26 +0000)]
Make delete method set via kern.cam.da.X.delete_method persistent.

This allows to set delete method via tunable, before device capabilities
are known.  Also allow ZERO method for devices not reporting LBP, if user
explicitly requests it -- it may be useful if storage supports compression
and WRITE SAME, but does not support UNMAP.

MFC after: 2 weeks

8 years agoFix build with gcc 4.2
bapt [Sun, 11 Oct 2015 17:45:20 +0000 (17:45 +0000)]
Fix build with gcc 4.2

8 years agoRemove legacy CHS geometry from dmesg and unify capacity outputs.
mav [Sun, 11 Oct 2015 13:48:20 +0000 (13:48 +0000)]
Remove legacy CHS geometry from dmesg and unify capacity outputs.

8 years agoRemove compatibility shims for legacy ATA device names.
mav [Sun, 11 Oct 2015 13:01:51 +0000 (13:01 +0000)]
Remove compatibility shims for legacy ATA device names.

We got new ATA stack in FreeBSD 8.x, switched to it at 9.x, completely
removed old stack at 10.x, so at 11.x it is time to remove compat shims.

8 years agoRemove lock upgrade attempt from ctl_be_block_open_file().
mav [Sun, 11 Oct 2015 08:28:49 +0000 (08:28 +0000)]
Remove lock upgrade attempt from ctl_be_block_open_file().

I am not sure what for it was done.  Now open routine should automatically
fall back to read-only if open for writing is impossible.  In such case
attempt to upgrade to write sounds strange.

MFC after: 1 week

8 years agoUse machine specific values cleaned from openwrt for the mac address
sbruno [Sun, 11 Oct 2015 03:31:11 +0000 (03:31 +0000)]
Use machine specific values cleaned from openwrt for the mac address
location on the TP link mr3020

8 years agowpi(4): do not count failures twice for ic_raw_xmit().
adrian [Sun, 11 Oct 2015 02:00:08 +0000 (02:00 +0000)]
wpi(4): do not count failures twice for ic_raw_xmit().

Tested:

* Tested with Intel 3945BG, STA mode

Submitted by: <s3erios@gmail.com>
Differential Revision: https://reviews.freebsd.org/D3768

8 years agowpi(4): move error handling upper (wpi_tx_data()/wpi_cmd2() -> wpi_raw_xmit()/wpi_tra...
adrian [Sun, 11 Oct 2015 01:58:50 +0000 (01:58 +0000)]
wpi(4): move error handling upper (wpi_tx_data()/wpi_cmd2() -> wpi_raw_xmit()/wpi_transmit()).

Tested:

* Tested with Intel 3945BG, STA mode

Submitted by: <s3erios@gmail.com>
Differential Revision: https://reviews.freebsd.org/D3767

8 years agowpi(4): fix possible race between TX/RX threads.
adrian [Sun, 11 Oct 2015 01:56:46 +0000 (01:56 +0000)]
wpi(4): fix possible race between TX/RX threads.

Tested:

* Tested with Intel 3945BG, STA mode

Submitted by: <s3erios@gmail.com>
Differential Revision: https://reviews.freebsd.org/D3766

8 years agowpi(4): do not allocate space for unused rings.
adrian [Sun, 11 Oct 2015 01:53:51 +0000 (01:53 +0000)]
wpi(4): do not allocate space for unused rings.

Tested:

* Tested with Intel 3945BG, STA mode

Submitted by: <s3erios@gmail.com>
Differential Revision: https://reviews.freebsd.org/D3765

8 years agowpi(4): do not override iv_recv_mgmt() in non-adhoc modes.
adrian [Sun, 11 Oct 2015 01:50:54 +0000 (01:50 +0000)]
wpi(4): do not override iv_recv_mgmt() in non-adhoc modes.

Tested:

* Tested with Intel 3945BG, IBSS and STA modes

Submitted by: <s3erios@gmail.com>
Differential Revision: https://reviews.freebsd.org/D3764

8 years agowpi(4): use more correct types.
adrian [Sun, 11 Oct 2015 01:31:18 +0000 (01:31 +0000)]
wpi(4): use more correct types.

This change fixes some amount of -Wsign-conversion and -Wconversion warnings
and sets correct sizes for some variables (as a result, some loop counters
were touched too).

Submitted by: <s3erios@gmail.com>
Differential Revision: https://reviews.freebsd.org/D3763

8 years agoUpdate Dragonfly Mail Agent to v0.10
bapt [Sat, 10 Oct 2015 23:31:47 +0000 (23:31 +0000)]
Update Dragonfly Mail Agent to v0.10

8 years agoReplace a local sx lock that allowed only one client at a time to access
ian [Sat, 10 Oct 2015 19:51:00 +0000 (19:51 +0000)]
Replace a local sx lock that allowed only one client at a time to access
an eeprom device with iicbus_request/release_bus(), which achieves the
same effect and also keeps other i2c slave drivers from clashing on the bus.

8 years agoThere's no way a fbsd install + kernel will fit into 4MB of flash.
sbruno [Sat, 10 Oct 2015 19:08:34 +0000 (19:08 +0000)]
There's no way a fbsd install + kernel will fit into 4MB of flash.

Assume and enforce the fact that this will always boot a rootfs from
usb.

8 years agoRemove a double "as well as"
phk [Sat, 10 Oct 2015 17:32:06 +0000 (17:32 +0000)]
Remove a double "as well as"

8 years agoFix installing config files as non root
bapt [Sat, 10 Oct 2015 14:43:46 +0000 (14:43 +0000)]
Fix installing config files as non root

Reported by: adrian

8 years agoInstall bsd.confs.mk
bapt [Sat, 10 Oct 2015 10:48:09 +0000 (10:48 +0000)]
Install bsd.confs.mk
Rename ETCDIR into CONFIGDIR to avoid conflicts with the ports tree

8 years agoAdd 'p' postfix to mdconfig(8).
trasz [Sat, 10 Oct 2015 09:37:31 +0000 (09:37 +0000)]
Add 'p' postfix to mdconfig(8).

MFC after: 1 month
Sponsored by: The FreeBSD Foundation

8 years agoChange the default setting of kern.ipc.shm_allow_removed from 0 to 1.
trasz [Sat, 10 Oct 2015 09:29:47 +0000 (09:29 +0000)]
Change the default setting of kern.ipc.shm_allow_removed from 0 to 1.

This removes the need for manually changing this flag for Google Chrome
users. It also improves compatibility with Linux applications running under
Linuxulator compatibility layer, and possibly also helps in porting software
from Linux.

Generally speaking, the flag allows applications to create the shared memory
segment, attach it, remove it, and then continue to use it and to reattach it
later. This means that the kernel will automatically "clean up" after the
application exits.

It could be argued that it's against POSIX. However, SUSv3 says this
about IPC_RMID: "Remove the shared memory identifier specified by shmid from
the system and destroy the shared memory segment and shmid_ds data structure
associated with it." From my reading, we break it in any case by deferring
removal of the segment until it's detached; we won't break it any more
by also deferring removal of the identifier.

This is the behaviour exhibited by Linux since... probably always, and
also by OpenBSD since the following commit:

revision 1.54
date: 2011/10/27 07:56:28; author: robert; state: Exp; lines: +3 -8;
Allow segments to be used even after they were marked for deletion with
the IPC_RMID flag.
This is permitted as an extension beyond the standards and this is similar
to what other operating systems like linux do.

MFC after: 1 month
Relnotes: yes
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3603

8 years agoProvide better debug message on kernel module name clash.
trasz [Sat, 10 Oct 2015 09:21:55 +0000 (09:21 +0000)]
Provide better debug message on kernel module name clash.

Reviewed by: kib@
MFC after: 1 month
Sponsored by: The FreeBSD Foundation

8 years agoMake geom_nop(4) collect statistics on all types of BIOs, not just
trasz [Sat, 10 Oct 2015 09:03:31 +0000 (09:03 +0000)]
Make geom_nop(4) collect statistics on all types of BIOs, not just
reads and writes.

PR: kern/198405
Submitted by: Matthew D. Fuller <fullermd at over-yonder dot net>
MFC after: 1 month
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3679

8 years agodrm/i915: Remove "Attempting to unbind pinned buffer" message
dumbbell [Sat, 10 Oct 2015 07:43:02 +0000 (07:43 +0000)]
drm/i915: Remove "Attempting to unbind pinned buffer" message

This error message is removed in later versions of Linux and currently,
it spams users.

PR: 200712
MFC after: 1 week

8 years agoUpdate the AP135 reference design flash layout to be more useful.
adrian [Sat, 10 Oct 2015 05:00:18 +0000 (05:00 +0000)]
Update the AP135 reference design flash layout to be more useful.

* Shuffle the kernel to be at the beginning
* Give the kernel 2mb, the rootfs 6mb, and 'mib0' the rest
* put the cfg parition just before the ART calibration data for the
  wifi part in the SoC
* .. and make sure ART points to the right 64k region.

I've updated the freebsd-wifi-build wiki the instructions on using this.

If someone has an AP135 with 8MB SPI flash then this won't work; everything
minus the big mib0 partition is just a bit over 8MB.  Come see me if this
ever happens (you'll likely just have to shrink the rootfs and the kernel
a little in order to make it fit.)

Tested:

* AP135 reference board.

8 years agoTranslate iic hardware layer status values to errno return values.
ian [Sat, 10 Oct 2015 02:29:02 +0000 (02:29 +0000)]
Translate iic hardware layer status values to errno return values.

8 years agoAdd a short name, IIC_INTRWAIT, for the common case (IIC_INTR | IIC_WAIT).
ian [Sat, 10 Oct 2015 02:06:07 +0000 (02:06 +0000)]
Add a short name, IIC_INTRWAIT, for the common case (IIC_INTR | IIC_WAIT).

8 years agoiw_cxgbe: fix for page fault in cm_close_handler().
np [Sat, 10 Oct 2015 01:41:07 +0000 (01:41 +0000)]
iw_cxgbe: fix for page fault in cm_close_handler().

This is roughly the iw_cxgbe equivalent of
https://github.com/torvalds/linux/commit/be13b2dff8c4e41846477b22cc5c164ea5a6ac2e
-----------------
RDMA/cxgb4: Connect_request_upcall fixes

When processing an MPA Start Request, if the listening endpoint is
DEAD, then abort the connection.

If the IWCM returns an error, then we must abort the connection and
release resources.  Also abort_connection() should not post a CLOSE
event, so clean that up too.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
-----------------

Submitted by: Krishnamraju Eraparaju at chelsio dot com.

8 years agoRemove all of the syncs
ngie [Sat, 10 Oct 2015 01:31:39 +0000 (01:31 +0000)]
Remove all of the syncs

They're unnecessary as shown by further testing on my VM

Requested by: jhb

8 years agoFlip on fast frames support for AR5416 and AR9300 series NICs.
adrian [Sat, 10 Oct 2015 00:13:45 +0000 (00:13 +0000)]
Flip on fast frames support for AR5416 and AR9300 series NICs.

This was off because the net80211 aggregation code was using the same
state pointers for both fast frames and ampdu tx support which led to some
pretty unfortunate panic-y behaviour.

Now that net80211 doesn't panic, let's flip this back on.

It doesn't (yet) do the horrific sounding thing of A-MPDU aggregates
of fast frames; that'll come next.  It's a pre-requisite to supporting
AMSDU + AMPDU anyway, which actually speeds things up quite considerably
(think packing lots of little ACK frames into a single AMSDU.)

Tested:

* QCA955x SoC, AP mode
* AR5416, STA mode
* AR9170, STA mode (with local fast frame patches)

8 years agoFix more cases of iicbus-layer functions that must return IIC_Exxxx values.
ian [Sat, 10 Oct 2015 00:12:03 +0000 (00:12 +0000)]
Fix more cases of iicbus-layer functions that must return IIC_Exxxx values.

8 years agoReturn only IIC_Exxxx status values from iicbus-layer functions. Most of
ian [Fri, 9 Oct 2015 23:58:19 +0000 (23:58 +0000)]
Return only IIC_Exxxx status values from iicbus-layer functions.  Most of
these functions are thin wrappers around calling the hardware-layer driver,
but some of them do sanity checks and return an error.  Since the hardware
layer can only return IIC_Exxxxx status values, the iicbus helper functions
must also adhere to that, so that drivers at higher layers can assume that
any non-zero status value is an IIC_Exxxx value that provides details about
what happened at the hardware layer (sometimes those details are important
for certain slave drivers).

8 years agoMove ppp.conf into the ppp sources
bapt [Fri, 9 Oct 2015 23:57:07 +0000 (23:57 +0000)]
Move ppp.conf into the ppp sources

8 years agoAdd iic2errno(), a helper function to translate IIC_Exxxxx status values to
ian [Fri, 9 Oct 2015 23:20:08 +0000 (23:20 +0000)]
Add iic2errno(), a helper function to translate IIC_Exxxxx status values to
errno values that are at least vaguely equivelent.  Also add a new status
value, IIC_ERESOURCE, to indicate a failure to acquire memory or other
required resources to complete a transaction.

The IIC_Exxxxxx values are supposed to communicate low-level details of the
i2c transaction status between the lowest-layer hardware driver and
higher-layer bus protocol and device drivers for slave devices on the bus.
Most of those slave drivers just return all status values from the lower
layers directly to their callers, resulting in crazy error reporting from a
user's point of view (things like timeouts being reported as "no such
process").  Now there's a helper function to make it easier to start
cleaning up all those drivers.

8 years agoRename the dma directory into dmagent to avoid collision on unclean oject
bapt [Fri, 9 Oct 2015 23:19:29 +0000 (23:19 +0000)]
Rename the dma directory into dmagent to avoid collision on unclean oject
directory where an old dma binary can remain

8 years agoUse IIC_EBUSBSY and IIC_BUSERR status values consistantly across all drivers.
ian [Fri, 9 Oct 2015 22:49:50 +0000 (22:49 +0000)]
Use IIC_EBUSBSY and IIC_BUSERR status values consistantly across all drivers.
Make it clearer what each one means in the comments that define them.

IIC_BUSBSY was used in many places to mean two different things, either
"someone else has reserved the bus so you have to wait until they're done"
or "the signal level on the bus was not in the state I expected before/after
issuing some command".

Now IIC_BUSERR is used consistantly to refer to protocol/signaling errors,
and IIC_BUSBSY refers to ownership/reservation of the bus.

8 years agoMostly rewrite the imx i2c driver. This started out as an attempt to fix
ian [Fri, 9 Oct 2015 22:28:56 +0000 (22:28 +0000)]
Mostly rewrite the imx i2c driver.  This started out as an attempt to fix
one specific problem: the driver didn't check for ACK/NAK after writing a
slave address byte to the bus, and some slaves signal that they are busy
(such as when completing an internal write to flash memory) by sending a
NAK in response to being addressed.

While working on that problem I discovered that the driver's handling of
error conditions in general didn't match the state transition diagram in
the reference manual, and making that right resulted in a lot of code
reorganization.

Along the way various other changes also happened...

 - Remove a mutex that wasn't protecting anything.
 - Remove some mystery DELAY()s, document the few that remain.
 - Use pause_sbt(9) to yield the processor for the bulk of the time it
   takes to transfer each byte rather than busy-polling the whole time.
 - Disable the controller when no transfers are in progress; since we
   don't operate in slave mode, there's no reason to run the hardware.
 - Remove a bunch of unecessary code from probe().

8 years agoInstall configuration files with CONFMODE instead of SHAREMODE
bapt [Fri, 9 Oct 2015 22:15:31 +0000 (22:15 +0000)]
Install configuration files with CONFMODE instead of SHAREMODE

8 years agoAdd installconfig as a top level target
bapt [Fri, 9 Oct 2015 22:14:07 +0000 (22:14 +0000)]
Add installconfig as a top level target

8 years agoAdd missing Makefile
bapt [Fri, 9 Oct 2015 22:10:54 +0000 (22:10 +0000)]
Add missing Makefile

8 years agoMove all the dma(8) components into one single directory
bapt [Fri, 9 Oct 2015 22:09:44 +0000 (22:09 +0000)]
Move all the dma(8) components into one single directory

Simplifying maintainance and options (only one place to deal with MK_DMAGENT)

This also makes packaging base less intrusive by getting a single point where
to add tags.

8 years agoChange make distribution so that it now call installconfig in all dirs along
bapt [Fri, 9 Oct 2015 22:05:31 +0000 (22:05 +0000)]
Change make distribution so that it now call installconfig in all dirs along
with the current behaviour of calling "distribution" in the etc target.

This allows mergemaster/etcupdate to still work when some configuration will be
moved to be handled in the same directories their source code lives in.

8 years agoAdd a new bsd.confs.mk similar to bsd.files.mk or bsd.incs.mk
bapt [Fri, 9 Oct 2015 21:57:42 +0000 (21:57 +0000)]
Add a new bsd.confs.mk similar to bsd.files.mk or bsd.incs.mk

It defines a CONFS variable for all files supposed to be installed as a
configuration file and handle as such

8 years agoBugfix: Exit the transfer loop if any read or write operation fails. Also,
ian [Fri, 9 Oct 2015 21:34:46 +0000 (21:34 +0000)]
Bugfix: Exit the transfer loop if any read or write operation fails.  Also,
perform a stop operation on the bus if there was an error, otherwise the
bus will remain hung forever.  Consistantly use 'if (error != 0)' style in
the function.

8 years agoStyle and whitespace cleanups. The only functional change is removal of
ian [Fri, 9 Oct 2015 21:27:30 +0000 (21:27 +0000)]
Style and whitespace cleanups.  The only functional change is removal of
a printf that appears to be left over from development debugging.

8 years agoPull in r242623 from upstream libc++ trunk (by Eric Fiselier):
dim [Fri, 9 Oct 2015 21:04:28 +0000 (21:04 +0000)]
Pull in r242623 from upstream libc++ trunk (by Eric Fiselier):

  Enable and fix warnings during the build.

  Although CMake adds warning flags, they are ignored in the libc++ headers
  because the headers '#pragma system header' themselves.

  This patch disables the system header pragma when building libc++ and fixes
  the warnings that arose.

  The warnings fixed were:
  1. <memory> - anonymous structs are a GNU extension
  2. <functional> - anonymous structs are a GNU extension.
  3. <__hash_table> - Embedded preprocessor directives have undefined behavior.
  4. <string> - Definition is missing noexcept from declaration.
  5. <__std_stream> - Unused variable.

This should fix building world (in particular libatf-c++) with -std=c++11.

Reported by: Oliver Hartmann <ohartman@zedat.fu-berlin.de>

8 years agoAdd llvm patch corresponding to r289072.
dim [Fri, 9 Oct 2015 21:00:04 +0000 (21:00 +0000)]
Add llvm patch corresponding to r289072.

8 years agoLet -c imply -S (hide signal output).
bdrewery [Fri, 9 Oct 2015 20:47:29 +0000 (20:47 +0000)]
Let -c imply -S (hide signal output).

Without this, the signals are shown seemingly randomly in the output before
the final summary is shown.  This is especially noticeable when there is
not much output from the application being traced.

Discussed with: jhb
Relnotes: yes

8 years agoAdd definitions for MIPS TLS relocations to elftoolchain.
kan [Fri, 9 Oct 2015 18:39:55 +0000 (18:39 +0000)]
Add definitions for MIPS TLS relocations to elftoolchain.

This makes our readelf more useful when looking for TLS-related
issues.

8 years agoRemove some trailing space.
kan [Fri, 9 Oct 2015 18:39:52 +0000 (18:39 +0000)]
Remove some trailing space.

8 years agoaddr2line: initialize die to NULL
emaste [Fri, 9 Oct 2015 18:26:24 +0000 (18:26 +0000)]
addr2line: initialize die to NULL

GCC on MIPS produced a 'may be used uninitialized' warning after
r289071.

Reported by: sbruno
Pointy hat to: emaste

8 years agoRemove empty line again from libc++'s iostream.cpp. This was used to
dim [Fri, 9 Oct 2015 18:23:10 +0000 (18:23 +0000)]
Remove empty line again from libc++'s iostream.cpp.  This was used to
force updates to this file, so it will be rebuilt by the fixed clang
from r289072.

8 years agoTemporarily revert upstream llvm trunk r240144 (by Michael Zolotukhin):
dim [Fri, 9 Oct 2015 18:21:45 +0000 (18:21 +0000)]
Temporarily revert upstream llvm trunk r240144 (by Michael Zolotukhin):

  [SLP] Vectorize for all-constant entries.

This should fix libc++'s iostream initialization SIGBUSing on amd64,
whenever the global cout symbol is not aligned to 16 bytes.

Some further explanation: libc++'s iostream.cpp contains the definitions
of std::cout, std::cerr and so on.  These global objects are effectively
declared with an alignment of 8 bytes.  When an executable is linked
against libc++.so, it can sometimes get a copy of the global object,
which is then at the same alignment.

However, with clang 3.7.0, the initialization of these global objects
will incorrectly use SSE instructions (e.g. movdqa), whenever the
optimization level is high enough, and SSE is enabled, such as on amd64.
When any of these objects is not aligned to 16 bytes, this will result
in a SIGBUS during iostream initialization.  In contrast, clang 3.6.x
and earlier took the 8 byte alignment into consideration, and avoided
SSE for those particular operations.

After bisecting of upstream changes, I found that the above revision
caused the change of this behavior, so I am reverting it now as a
workaround, while a discussion and test case is being prepared for
upstream.

8 years agoUpdate to ELF Tool Chain r3250
emaste [Fri, 9 Oct 2015 17:46:05 +0000 (17:46 +0000)]
Update to ELF Tool Chain r3250

Highlights (not already in the FreeBSD tree):
  - addr2line: Fixed multiple memory leaks related to DIE allocation
  - readelf: improve sh_link validation
  - various man page improvements

Sponsored by: The FreeBSD Foundation

8 years agoAdd .gnu.versym VERSYM_HIDDEN flag and related mask
emaste [Fri, 9 Oct 2015 17:33:05 +0000 (17:33 +0000)]
Add .gnu.versym VERSYM_HIDDEN flag and related mask

MFC after: 1 week
Sponsored by: The FreeBSD Foundation

8 years agoFix a repeated typo: rootir -> rootdir.
wblock [Fri, 9 Oct 2015 14:55:55 +0000 (14:55 +0000)]
Fix a repeated typo: rootir -> rootdir.

Approved by: bapt
MFC after: 1 week

8 years agoSort function prototypes and add missing 'static' keywords.
kevlo [Fri, 9 Oct 2015 14:31:32 +0000 (14:31 +0000)]
Sort function prototypes and add missing 'static' keywords.

Submitted by: Andriy Voskoboinyk <s3erios at gmail dot com>
Differential Revision: https://reviews.freebsd.org/D3847

8 years agoAdd --currently-running <release> option to freebsd-update.
cperciva [Fri, 9 Oct 2015 12:34:33 +0000 (12:34 +0000)]
Add --currently-running <release> option to freebsd-update.

This option tells freebsd-update to act as if it is running a specific
release instead of querying the kernel.  In particular, this can be
useful when upgrading jails.

Requested by: EuroBSDCon devsummit jails session
Tested by: allanjude
MFC after: 1 week

8 years agoRemove root_mount_wait(). It's not used anywhere.
trasz [Fri, 9 Oct 2015 12:11:37 +0000 (12:11 +0000)]
Remove root_mount_wait(). It's not used anywhere.

Reviewed by: bapt@
MFC after: 1 month
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3787

8 years agoUpgrade to Unbound 1.5.5.
des [Fri, 9 Oct 2015 11:46:27 +0000 (11:46 +0000)]
Upgrade to Unbound 1.5.5.

8 years agoDocument iSCSI and autofs(5) variables in rc.conf(5).
trasz [Fri, 9 Oct 2015 08:13:59 +0000 (08:13 +0000)]
Document iSCSI and autofs(5) variables in rc.conf(5).

PR: 203119
Reviewed by: bjk@
MFC after: 1 month
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3671

8 years agoExploit r288122 to avoid pointlessly enqueueing a page that is about to be
alc [Fri, 9 Oct 2015 03:38:58 +0000 (03:38 +0000)]
Exploit r288122 to avoid pointlessly enqueueing a page that is about to be
freed.

Submitted by: kmacy
Differential Revision: https://reviews.freebsd.org/D1674

8 years agoFix missing semi-colon from r289055.
cem [Thu, 8 Oct 2015 23:27:45 +0000 (23:27 +0000)]
Fix missing semi-colon from r289055.

Obtained from: mjg
Sponsored by: EMC / Isilon Storage Division

8 years agoRemove comment obsoleted by r289056.
trasz [Thu, 8 Oct 2015 21:52:20 +0000 (21:52 +0000)]
Remove comment obsoleted by r289056.

MFC after: 1 month
Sponsored by: The FreeBSD Foundation

8 years agoRemove unused SI_SUB_* #defines.
trasz [Thu, 8 Oct 2015 21:28:06 +0000 (21:28 +0000)]
Remove unused SI_SUB_* #defines.

Reviewed by: kib@
MFC after: 1 month
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3708

8 years agolinux: fix handling of out-of-bounds syscall attempts
mjg [Thu, 8 Oct 2015 21:08:35 +0000 (21:08 +0000)]
linux: fix handling of out-of-bounds syscall attempts

Due to an off by one the code would read an entry past the table, as
opposed to the last entry which contains the nosys handler.

Reported by: Pawel Biernacki <pawel.biernacki gmail.com>

8 years agoIf we can't open the file, skip devclose() for the exclusive_file_system
marcel [Thu, 8 Oct 2015 17:59:05 +0000 (17:59 +0000)]
If we can't open the file, skip devclose() for the exclusive_file_system
case. We never called devopen(), so we know there's nothing to close.

8 years agoFix short month names and replace %b with %_m in date_fmt for Chinese
delphij [Thu, 8 Oct 2015 17:48:49 +0000 (17:48 +0000)]
Fix short month names and replace %b with %_m in date_fmt for Chinese
locales.

When using a Chinese locale, such as zh_TW.UTF-8 or zh_CN.UTF-8,
nl_langinfo(ABMON_*) only returned numbers.  For instance,
nl_langinfo(ABMON_1) returns 1, nl_langinfo(ABMON_2) returns 2, and
so on.

This causes problems in applications that put the short month name
and the day of the month together.  For example, 'Apr 14' in English
becomes '414日' in Chinese on the top bar of GNOME Shell.

This problem may be resolved by appending '月' to all short month
names and replacing %b with %_m in date_fmt. ja_JP.UTF-8 already
does this, and this matches the en_US.ISO8859-1 behavior, which
returns 'Oct'.  The GNU C Library also returns values with '月'
appended.

PR: 199441
Submitted by: Ting-Wei Lan <lantw44 gmail com>
MFC after: 2 weeks

8 years agoBuild changes that allow the modules on arm64.
kib [Thu, 8 Oct 2015 17:42:08 +0000 (17:42 +0000)]
Build changes that allow the modules on arm64.
- Move the required kernel compiler flags from Makefile.arm64 to kern.mk.
- Build arm64 modules as PIC; non-PIC relocations in .o for shared object
  output cannot be handled.
- Do not try to install aarch64 symlink.
- A hack for arm64 to avoid ld -r stage.  See the comment for the explanation.
  Some functionality is lost, like ctf handling, but hopefully will be
  restored after newer linker is available.

Reviewed by: andrew, emaste
Tested by: andrew (on real hardware)
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3796

8 years agoAdd encoding for mime-types.
delphij [Thu, 8 Oct 2015 17:41:17 +0000 (17:41 +0000)]
Add encoding for mime-types.

MFC after: 2 weeks

8 years agoDisable syscons and vpo modules for arm64.
kib [Thu, 8 Oct 2015 17:32:45 +0000 (17:32 +0000)]
Disable syscons and vpo modules for arm64.

Reviewed by: andrew, emaste
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3796

8 years agoAdd the arm64 define.
kib [Thu, 8 Oct 2015 17:32:17 +0000 (17:32 +0000)]
Add the arm64 define.

Reviewed by: andrew, emaste
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3796

8 years agoImplement in-kernel relocator for the arm64 module linker.
kib [Thu, 8 Oct 2015 16:58:01 +0000 (16:58 +0000)]
Implement in-kernel relocator for the arm64 module linker.

It is decided to go with the shared object file format for modules on
arm64, due to the Aarch64 instruction set details.  Combination of the
signed 28-bit offset in the branch instructions encoding together with
the supported memory model of compilers makes the relocatable object
support impossible or at least too hard.

Reviewed by: andrew, emaste
Tested by: andrew (on real hardware)
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3796

8 years agoMake the copyright notice in the file to match reality. Use the
kib [Thu, 8 Oct 2015 16:46:11 +0000 (16:46 +0000)]
Make the copyright notice in the file to match reality.  Use the
recommended FreeBSD license text.

Approved by: andrew
Discussed with: emaste
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3846

8 years agoxen/console: Introduce a new console driver for Xen guest
royger [Thu, 8 Oct 2015 16:39:43 +0000 (16:39 +0000)]
xen/console: Introduce a new console driver for Xen guest

The current Xen console driver is crashing very quickly when using it on
an ARM guest. This is because the console lock is recursive and it may
lead to recursion on the tty lock and/or corrupt the ring pointer.

Furthermore, the console lock is not always taken where it should be and has
to be released too early because of the way the console has been designed.

Over the years, code has been modified to support various new features but
the driver has not been reworked.

This new driver has been rewritten with the idea of only having a small set
of specific function to write either via the shared ring or the hypercall
interface.

Note that HVM support has been left aside for now because it requires
additional features which are not yet supported. A follow-up patch will be
sent with HVM guest support.

List of items that may be good to have but not mandatory:
 - Avoid to flush for each character written when using the tty
 - Support multiple consoles

Submitted by: Julien Grall <julien.grall@citrix.com>
Reviewed by: royger
Differential Revision: https://reviews.freebsd.org/D3698
Sponsored by: Citrix Systems R&D

8 years agoChange gptldr from relocating 0xfff1 bytes of boot2 to relocating 0x20000
cperciva [Thu, 8 Oct 2015 15:38:34 +0000 (15:38 +0000)]
Change gptldr from relocating 0xfff1 bytes of boot2 to relocating 0x20000
bytes of boot2.  Since we're in 16-bit mode, we can't copy all 128kB at
once; instead we loop four times and copy 32 kB each time.

This change was made necessary by an upcoming increase in the size of the
boot2 binary; should it increase further, the COPY_BLKS value can be
adjusted without anyone needing to remember 8086 assembly language again.

Requested by: allanjude
Tested by: allanjude
MFC after: 1 week

8 years agoAdd support for Fresco Logic USB 3.0 host controller.
kevlo [Thu, 8 Oct 2015 15:13:57 +0000 (15:13 +0000)]
Add support for Fresco Logic USB 3.0 host controller.

Fresco Logic hosts advertise MSI, but fail to actually generate MSI
interrupts.  We have to disable MSI use.

Reviewed by: hselasky

8 years agoAdd quirk for USB 3.0 PCI device.
hselasky [Thu, 8 Oct 2015 13:39:27 +0000 (13:39 +0000)]
Add quirk for USB 3.0 PCI device.

Submitted by: Gary Jennejohn <gj@freebsd.org>
MFC after: 1 week