Matthew Dillon [Wed, 11 Jul 2012 06:43:43 +0000 (23:43 -0700)]
Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into hammer2
Peter Avalos [Wed, 11 Jul 2012 04:53:25 +0000 (21:53 -0700)]
Update files for libarchive-3.0.4 import.
Matthew Dillon [Wed, 11 Jul 2012 03:54:14 +0000 (20:54 -0700)]
Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into hammer2
Peter Avalos [Wed, 11 Jul 2012 03:39:24 +0000 (20:39 -0700)]
Merge branch 'vendor/LIBARCHIVE'
Peter Avalos [Wed, 11 Jul 2012 03:25:58 +0000 (20:25 -0700)]
Import libarchive-3.0.4.
Matthew Dillon [Wed, 11 Jul 2012 03:15:37 +0000 (20:15 -0700)]
kernel - Fix missing lock in pbuf update (minor)
* Fix a small timing window in an update of a counter.
* This bug is not known to have been hit by anyone, due to default
settings for asynchronous I/O.
Peter Avalos [Wed, 11 Jul 2012 02:29:26 +0000 (19:29 -0700)]
Update files for libpcap-1.3.0 and tcpdump-4.3.0 imports.
Peter Avalos [Wed, 11 Jul 2012 01:46:52 +0000 (18:46 -0700)]
Merge branch 'vendor/TCPDUMP'
Peter Avalos [Wed, 11 Jul 2012 01:36:31 +0000 (18:36 -0700)]
Merge branch 'vendor/LIBPCAP'
Peter Avalos [Wed, 11 Jul 2012 01:22:41 +0000 (18:22 -0700)]
Import tcpdump-4.3.0.
See CHANGES for details.
Peter Avalos [Wed, 11 Jul 2012 01:14:22 +0000 (18:14 -0700)]
Import libpcap-1.3.0.
See CHANGES for the details.
John Marino [Mon, 9 Jul 2012 15:27:54 +0000 (17:27 +0200)]
CCVER: Add /usr/pkgsrc/lang/gcc-aux as 'gcc47'
There are two gcc-4.7.x packages in pkgsrc that will build gcc 4.7-based
compilers on Dragonfly. One is located at lang/gcc47 and the other is
located at lang/gcc-aux. Both have been patched by myself, but gcc-aux
seems to work better, especially with regards to link-time optimization.
Consequently, a new valid value for CCVER has been created, "gcc47", which
references lang/gcc-aux while building world and kernel. Both will
compile, and the produced kernel will boot, but there are some network
issues. Consequently, CCVER=gcc47 should only be used by developers who
have the express intent to fix any issues the arise from compiling
DragonFly sources with the latest GCC compiler. Normal users looking
for non-base GCC compilers should stick with CCVER=gcc46 for now.
John Marino [Mon, 9 Jul 2012 17:59:01 +0000 (19:59 +0200)]
Libc: Remove i386 gcc hack concerning popcountsi2
Ref:
0cfd7e44b3676f6a43164acab0aaea46365b5c53
Ref:
4b548429247a10382073a0c227af306272160d8d
Back in Oct 2010, a "horrible hack" was added to libc in order to fix a
Firefox bug concerning a late dynamic link of a libcairo.so.2 symbol. A
weak reference to popcountsi2 was added to i386 libc in order to force it
to be available.
Up through gcc 4.6, this hack worked because the symbol is global in the
libgcc.a library, but it's a hidden symbol starting with the gcc 4.7
version of libgcc.a. This causes a build failure on bin/rmail while
building world with gcc 4.7.
Since Oct 2010, rtld has undergone significant changes.
With the hope that the issue is obsolete, the hack is being removed.
This allows gcc 4.7 to compile DragonFly world and kernel source.
We'll have to track firefox manually to verify the issue the hack fixed
doesn't return.
John Marino [Mon, 25 Jun 2012 16:39:37 +0000 (18:39 +0200)]
ld/gold: Add link-time optimization plugin support
The binutils 2.22 ld.bfd and ld.gold linkers were not configured to support
link-time optimization plugins (-plugin switch) previously. This changeset
adds -plugin support to both linkers. The ld.bfd linker was previously
built statically, but it was necessary to change this to use the libc and
libz shared libraries because the -plugin option requires the dynamic
function dlopen().
The same update was not applied to binutils 2.21 linkers.
Sascha Wildner [Tue, 10 Jul 2012 16:33:44 +0000 (18:33 +0200)]
kernel: Add descriptions to the intrhooks that miss them.
The description is printed in the
"**WARNING** waiting for the following device to finish configuring: ..."
messages we occasionally see, from run_interrupt_driven_config_hooks(),
most of the times for xpt.
Some of them had been accidentally removed in past synchronizations with
FreeBSD (mostly by me).
Sascha Wildner [Tue, 10 Jul 2012 14:16:13 +0000 (16:16 +0200)]
config_intrhook.9: Update struct intr_config_hook to ours.
Sascha Wildner [Tue, 10 Jul 2012 13:46:47 +0000 (15:46 +0200)]
<sys/kernel.h>: Remove unused sysinit id.
Sascha Wildner [Tue, 10 Jul 2012 07:23:57 +0000 (09:23 +0200)]
amr(4): Remove unused code.
Sascha Wildner [Mon, 9 Jul 2012 19:43:46 +0000 (21:43 +0200)]
amr(4): Remove some old code.
Sascha Wildner [Mon, 9 Jul 2012 19:07:13 +0000 (21:07 +0200)]
Bring in the config_intrhook(9) manual page from FreeBSD.
Sascha Wildner [Mon, 9 Jul 2012 18:06:00 +0000 (20:06 +0200)]
kernel/amrp: Add a missing dependency.
Sascha Wildner [Mon, 9 Jul 2012 16:14:10 +0000 (18:14 +0200)]
twa(4): Remove an extra : in a message (it will be automatically added).
Sascha Wildner [Mon, 9 Jul 2012 16:04:04 +0000 (18:04 +0200)]
twa(4): Request a bus scan after attaching the driver to CAM.
Our CAM doesn't (yet) do this by default.
This fixes disk detection upon kldload.
Taken-from: FreeBSD (which has since moved it to CAM).
While here, add a cam_simq_release() I forgot and remove a comment that
does not apply to us.
Sascha Wildner [Sun, 8 Jul 2012 23:07:10 +0000 (01:07 +0200)]
twa(4)/twe(4): Move dev_ops_remove_minor() after destroy_dev().
Fixes crash on module unload.
Sascha Wildner [Sun, 8 Jul 2012 23:05:37 +0000 (01:05 +0200)]
twa(4): Really stop the callouts.
Sascha Wildner [Sun, 8 Jul 2012 21:58:08 +0000 (23:58 +0200)]
twa(4): Remove some unused code.
Sascha Wildner [Sun, 8 Jul 2012 21:33:20 +0000 (23:33 +0200)]
twa(4): Add MSI support.
It's off by default, set hw.twa.msi.enable to 1 in /boot/loader.conf to
try it out.
It works here with a 3ware 9650SE-2LP.
Sascha Wildner [Sun, 8 Jul 2012 21:01:19 +0000 (23:01 +0200)]
twa(4): Add MODULE_VERSION.
Sascha Wildner [Sun, 8 Jul 2012 20:39:48 +0000 (22:39 +0200)]
twa(4): Pass INTR_MPSAFE when setting up the interrupt.
I removed it by mistake during porting.
While here, add some little fixes from FreeBSD.
Tested-with: 3ware 9650SE-2LP
Antonio Huete Jimenez [Sun, 8 Jul 2012 12:37:50 +0000 (14:37 +0200)]
libkern - Add memcchr(9).
* It can be used for filesystem bitmap operations.
* FreeBSD introduced it to replace calls to skpc().
* It does compare entire words instead of single bytes.
Taken from: FreeBSD-SVN 229198
Antonio Huete Jimenez [Sun, 8 Jul 2012 10:59:13 +0000 (12:59 +0200)]
mdoc.local: Add FreeBSD 10.0.
Sepherosa Ziehau [Sun, 8 Jul 2012 09:15:57 +0000 (17:15 +0800)]
bge: Split if_bgereg.h into if_bgereg.h and if_bgevar.h
if_bgevar.h conains only software related structs and macros
Sepherosa Ziehau [Sun, 8 Jul 2012 05:33:06 +0000 (13:33 +0800)]
if_bgereg.h: Add header guardian
Sepherosa Ziehau [Sun, 8 Jul 2012 05:28:43 +0000 (13:28 +0800)]
bge: Cosmetic cleanup
Sascha Wildner [Sat, 7 Jul 2012 23:29:54 +0000 (01:29 +0200)]
termcap.5: Bring in some mdoc fixes from FreeBSD.
Sascha Wildner [Sat, 7 Jul 2012 22:49:45 +0000 (00:49 +0200)]
sln.4: Add ifmedia(4) reference.
Sepherosa Ziehau [Sat, 7 Jul 2012 14:16:49 +0000 (22:16 +0800)]
bce: Enabling MSI or MSI-X in interrupt handler does not need to mask first
This reduce number of register writing operation during interrupt handling
According-to: bnx2
Sepherosa Ziehau [Sat, 7 Jul 2012 09:08:26 +0000 (17:08 +0800)]
bge: Remove unused structs and macros
Sepherosa Ziehau [Sat, 7 Jul 2012 08:34:47 +0000 (16:34 +0800)]
brgphy: Remove debug prints
Sepherosa Ziehau [Sat, 7 Jul 2012 08:15:29 +0000 (16:15 +0800)]
bge/bce: Use mii private data to pass various flags to brgphy
This unbreaks if_name support on bce(4) and bge(4). And this mechanism
is cleaner and easier to extend than checking if_name and accessing
NIC drivers' softc.
Sepherosa Ziehau [Sat, 7 Jul 2012 07:12:30 +0000 (15:12 +0800)]
mii: Add private data tag and save private data into mii_softc
While im here, add comment for the fields of mii_probe_args
Sepherosa Ziehau [Sat, 7 Jul 2012 06:35:52 +0000 (14:35 +0800)]
igb: Add comment for WTHRESH and header write back mode.
Sepherosa Ziehau [Sat, 7 Jul 2012 06:16:03 +0000 (14:16 +0800)]
igb: Update comment about errata number
Sepherosa Ziehau [Sat, 7 Jul 2012 06:11:24 +0000 (14:11 +0800)]
em/emx: Update comment concerning errata number
Sepherosa Ziehau [Sat, 7 Jul 2012 05:47:43 +0000 (13:47 +0800)]
em: Fix "Missing Interrupt Following ICR read" errata
IMS should be set to 0 before reading ICR. It should be noted that
once IMS is 0, ICR.INT_ASSERTED will not be set, which is not mentioned
in the datasheet.
This errata exists on 82571, 82572, 82573, 82574 and 82583. And the fix
is needed only if the legacy interrupt is used and the interrupt line is
shared with other devices.
hw.emX.irq.unshared tunable is added to give a hint to the driver that
the legacy interrupt is not shared with other devices.
Sepherosa Ziehau [Sat, 7 Jul 2012 05:24:52 +0000 (13:24 +0800)]
emx: Fix "Missing Interrupt Following ICR read" errata
IMS should be set to 0 before reading ICR. It should be noted that
once IMS is 0, ICR.INT_ASSERTED will not be set, which is not mentioned
in the datasheet.
This errata exists on 82571, 82572, 82573, 82574 and 82583, while 82583
is not covered by emx(4). And the fix is needed only if the legacy
interrupt is used and the interrupt line is shared with other devices.
hw.emxX.irq.unshared tunable is added to give a hint to the driver that
the legacy interrupt is not shared with other devices.
Sepherosa Ziehau [Fri, 6 Jul 2012 07:10:55 +0000 (15:10 +0800)]
igb: Move shared interrupt configure to igb_alloc_intr()
So resource allocation flags could be setup properly
Sascha Wildner [Fri, 6 Jul 2012 23:14:11 +0000 (01:14 +0200)]
kernel/vm: Remove some unused variables.
Sascha Wildner [Fri, 6 Jul 2012 17:43:50 +0000 (19:43 +0200)]
kernel/gpio: Return EINVAL for GPIOPINUNSET and simplify a bit.
Sascha Wildner [Fri, 6 Jul 2012 12:37:20 +0000 (14:37 +0200)]
pst(4): Remove a bogus (commented out) debug option from the Makefile.
Sascha Wildner [Fri, 6 Jul 2012 12:09:49 +0000 (14:09 +0200)]
sdpd(8): Fix a wrong sizeof.
Sascha Wildner [Fri, 6 Jul 2012 09:25:19 +0000 (11:25 +0200)]
make upgrade: Remove a netgraph7 header that was moved.
Markus Pfeiffer [Fri, 6 Jul 2012 08:40:17 +0000 (08:40 +0000)]
kernel - remove dead code in busdma
Remove some superflous code inserted in previous commit to
add BUS_DMA_KEEP_PG_OFFSET (x86_64)
Matthew Dillon [Fri, 6 Jul 2012 05:36:01 +0000 (22:36 -0700)]
hammer2 - More work on userland hammer2 msg infrastructure
* When a link error occurs generate a LNK_ERROR message for each transaction
before setting the iocom ERROR flag and returning the final
non-transactional LNK_ERROR.
* Processing command switches switch on the original transactional head.cmd
instead of the current msg->any.head.cmd, which allows the use of mixed
cmd's in a transactional message stream.
The target function then handles the actual msg->any.head.cmd. Thus we
can consolidate all sub-commands used within a transaction into the target
function, which greatly improves code quality.
This allows us to send LNK_ERROR messages over active transactions.
* Print the pfs_id and label for the received LNK_SPAN message the kernel
sends to the userland hammer2 service process, and verify LNK_ERROR
processing for connection terminations. Yup, it works.
Matthew Dillon [Fri, 6 Jul 2012 05:32:42 +0000 (22:32 -0700)]
hammer2 - Fill in lnk_span message
* Fill in the header fields for the lnk_span message.
* Rename hammer2_any to hammer2_msg_any
* Incorporate lnk_span into hammer2_msg_any
Matthew Dillon [Fri, 6 Jul 2012 00:48:37 +0000 (17:48 -0700)]
Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into hammer2
Sascha Wildner [Thu, 5 Jul 2012 19:16:18 +0000 (21:16 +0200)]
kernel/sbuf: Add back a lost newline.
Sascha Wildner [Thu, 5 Jul 2012 18:48:41 +0000 (20:48 +0200)]
Remove tools/tools/mid (Message-ID/In-Reply-To db thingy we don't need).
Sascha Wildner [Thu, 5 Jul 2012 15:48:54 +0000 (17:48 +0200)]
Create if_*(4) MLINKS for all network interface manual pages.
The modules are called if_*.ko for those, so follow the principle that
the user generally should be able to do 'man foo' for any 'foo.ko'.
Inspired-by: FreeBSD
Sascha Wildner [Thu, 5 Jul 2012 15:10:57 +0000 (17:10 +0200)]
man4.i386/Makefile: Break lines.
Sascha Wildner [Thu, 5 Jul 2012 15:25:06 +0000 (17:25 +0200)]
Add an MLINKS for ic(4) (it is documented in iicbus(4)).
Nuno Antunes [Thu, 5 Jul 2012 13:28:36 +0000 (14:28 +0100)]
mdoc: Reorder manpage sections to keep them consistent.
* Keep CAVEATS and BUGS sections at the end of the pages.
* EXIT STATUS section should come after ENVIRONMENT.
Found-with: mandoc -Tlint
Reviewed-by: Sacha Wildner <swildner@>
Markus Pfeiffer [Thu, 5 Jul 2012 13:14:01 +0000 (13:14 +0000)]
kernel - busdma.h correct whitespace and add description
Sascha Wildner [Thu, 5 Jul 2012 12:46:30 +0000 (14:46 +0200)]
dpt(4): Remove some old #if 0'd code from a header file.
Nuno Antunes [Thu, 5 Jul 2012 07:00:14 +0000 (08:00 +0100)]
mdoc: Silence a few mdocml (mandoc) warnings and errors.
* Move CAVEATS section to the end of the page.
* Miscelaneous other fixes (taken from FreeBSD).
* Bump dates.
Found-with: mandoc -Tlint
Nuno Antunes [Wed, 4 Jul 2012 18:31:37 +0000 (19:31 +0100)]
netgraph7/ng_tty: Remove two unused #define's.
Nuno Antunes [Wed, 4 Jul 2012 11:59:54 +0000 (12:59 +0100)]
netgraph7: Cleanup the Makefile.
* netgraph base and ng_socket node are already ported and listed
in the Makefile. Remove the redundant commented lines.
Nuno Antunes [Wed, 4 Jul 2012 11:46:06 +0000 (12:46 +0100)]
ppp(8): Use netgraph7 tty node when WANT_NETGRAPH7 make variable is defined.
* Now that netgraph7's ng_tty node is ported we can use it instead of
legacy netgraph's node. This fixes the hack introduced in commit
f0f93f7f2d479e18fe43625ce59583ba7a773bf2.
Nuno Antunes [Wed, 4 Jul 2012 08:48:55 +0000 (09:48 +0100)]
netgraph7: Welcome ng_tty.
Sepherosa Ziehau [Thu, 5 Jul 2012 07:22:42 +0000 (15:22 +0800)]
em/emx: Correct ASPM settings for 82571/82572/82573
Sepherosa Ziehau [Thu, 5 Jul 2012 06:48:40 +0000 (14:48 +0800)]
emx: Don't enable MSI on 82571/82572
You could still forcefully enable MSI by using hw.emX.msi.enable;
useful certain systems
Obtained-from: OpenBSD
Sepherosa Ziehau [Thu, 5 Jul 2012 06:40:46 +0000 (14:40 +0800)]
em: Don't enable MSI on non-PCIe chips and 82571/82572
You could still forcefully enable MSI by using hw.emX.msi.enable;
useful for 82571/82572 on certain systems
Obtained-from: OpenBSD
Matthew Dillon [Thu, 5 Jul 2012 05:36:45 +0000 (22:36 -0700)]
Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into hammer2
Sepherosa Ziehau [Thu, 5 Jul 2012 02:37:40 +0000 (10:37 +0800)]
bge: Add commented out code to set "clear ticks upon TX"
As far as I have tested, this coalesce mode does not have visiable
effect on both TX performance and interrupt rate.
To ease further tuning, the current TX performance and interrupt
rate that I have gathered on BCM5750 B1:
1472B UDP datagram perf: 957Mbps, intr rate: ~1200Hz
18B UDP datagram perf: 565Kpps, intr rate: ~4400Hz
Sepherosa Ziehau [Thu, 5 Jul 2012 02:35:28 +0000 (10:35 +0800)]
bge: Define clear ticks upon RX/TX for host coalesce mode register
Sepherosa Ziehau [Wed, 4 Jul 2012 12:19:41 +0000 (20:19 +0800)]
bge: Add BGE_IS_CRIPPLED to test BCM5788 and BCM5700
Sepherosa Ziehau [Wed, 4 Jul 2012 10:56:33 +0000 (18:56 +0800)]
bge: Make sure coalesce parameters are in sane range
Sepherosa Ziehau [Wed, 4 Jul 2012 09:55:45 +0000 (17:55 +0800)]
bge: Increase default RX coalesce to 160us (~6000Hz)
Sepherosa Ziehau [Wed, 4 Jul 2012 09:32:09 +0000 (17:32 +0800)]
bge: Correct comment
Sepherosa Ziehau [Wed, 4 Jul 2012 09:23:26 +0000 (17:23 +0800)]
bge: Factor our bge_link_poll()
Sepherosa Ziehau [Wed, 4 Jul 2012 09:15:24 +0000 (17:15 +0800)]
bge: Add comment about coalesce parameters during interrupt
Markus Pfeiffer [Wed, 4 Jul 2012 21:27:17 +0000 (21:27 +0000)]
kernel - Correct whitespace in BUS_DMA_KEEP_PG_OFFSET (x86_64)
Markus Pfeiffer [Wed, 4 Jul 2012 20:54:55 +0000 (20:54 +0000)]
mii: add AGERE ET1011 phy
Markus Pfeiffer [Wed, 4 Jul 2012 13:31:38 +0000 (13:31 +0000)]
kernel - Add BUS_DMA_KEEP_PG_OFFSET (i386)
Add BUS_DMA_KEEP_PG_OFFSET from FreeBSD in preparation of import of
FreeBSD's USB stack
Markus Pfeiffer [Wed, 4 Jul 2012 13:21:25 +0000 (13:21 +0000)]
kernel - Add BUS_DMA_KEEP_PG_OFFSET (x86_64)
Add BUS_DMA_KEEP_PG_OFFSET from FreeBSD in preparation of import of
FreeBSD's USB stack
Markus Pfeiffer [Wed, 4 Jul 2012 13:12:46 +0000 (13:12 +0000)]
kernel: add handling for ENODEV in poll_copyout and select_copyout
Peter Avalos [Wed, 4 Jul 2012 07:55:58 +0000 (00:55 -0700)]
Merge branch 'vendor/FILE'
Peter Avalos [Wed, 4 Jul 2012 07:49:04 +0000 (00:49 -0700)]
Import file-5.11.
* Fix CDF parsing issues.
Sepherosa Ziehau [Tue, 3 Jul 2012 09:39:58 +0000 (17:39 +0800)]
bge: Enable "tagged status" on non-BCM5788 and non-BCM5700 chips
By using "tagged status", we could:
- Detect whether the interrupt is for us or not.
- Avoid MAC_STS register read upon every interrupt.
Some implementation note:
- Interrupt must be disabled during interrupt handing, if "tagged status"
is used, else the interrupt will be kept asserted. This is the case at
least on BCM5750 AX/BX.
- Coalesce parameters during interrupt must be properly configured; when
"tagged status" is used we have "in interrupt" state.
Tunable hw.bgeX.status_tag is added to prevent driver from using "tagged
status". hw.bgeX.{rx,tx}_coal_bds_int and hw.bgeX.{rx,tx}_coal_ticks_int
sysctls are added to tune "in interrupt" coalesce parameters.
Matthew Dillon [Tue, 3 Jul 2012 17:53:26 +0000 (10:53 -0700)]
Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into hammer2
Matthew Dillon [Tue, 3 Jul 2012 17:42:03 +0000 (10:42 -0700)]
kernel - reduce kernel stack use (2)
* Correct hanging allocation on kldunload from previous commit.
* Rearrange the flow of the code a little for the exit case.
Matthew Dillon [Tue, 3 Jul 2012 17:35:56 +0000 (10:35 -0700)]
Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into hammer2
Matthew Dillon [Tue, 3 Jul 2012 17:22:09 +0000 (10:22 -0700)]
kernel - reduce kernel stack use
* fq_dispatcher() was eating 8K+ of kernel stack on x86-64 with a
(512 x sizeof(pointer) * 2) sized array. This is way too much kernel
stack.
* Add a safety to the hz calculation in the lksleep() for hz values < 15.
vkernels already use low hz values of ~20-ish.
Reported-by: swildner
Sepherosa Ziehau [Tue, 3 Jul 2012 08:18:21 +0000 (16:18 +0800)]
mii: Wrong sizeof
Pointed-by: swildner@
Sepherosa Ziehau [Tue, 3 Jul 2012 06:27:19 +0000 (14:27 +0800)]
bge: BGE_PCI_MISC_CTL resides in PCI configuration space
Sepherosa Ziehau [Tue, 3 Jul 2012 05:34:33 +0000 (13:34 +0800)]
bge: Regroup code, no functional changes
- Regroup chip information and bugs setup code
- Regroup MII information setup code
Mihai Carabas [Sun, 1 Jul 2012 19:18:25 +0000 (22:18 +0300)]
sbuf_vprintf: ap is not consistent at second call of kvsnprintf. Use a copy
Sepherosa Ziehau [Mon, 2 Jul 2012 08:43:41 +0000 (16:43 +0800)]
bge: Don't force PCI Express TLP to be 128 bytes
Instead, status errors are cleared and "no snoop" and "relaxed ordering"
are disabled.
Obtained-from: FreeBSD 235821
Sepherosa Ziehau [Mon, 2 Jul 2012 08:42:59 +0000 (16:42 +0800)]
pci: Add more definition for PCI Express device {control,status} registers
Sepherosa Ziehau [Mon, 2 Jul 2012 08:20:05 +0000 (16:20 +0800)]
bge: Workaround mbox write reordering for PCI-X devices
Add per device tunable hw.bgeX.mbox_reorder to control this; it is
off by default.
Based-on: FreeBSD 232848