]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
7 years agoDocument r318660, various NFS client fixes.
gjb [Thu, 25 May 2017 10:56:33 +0000 (10:56 +0000)]
Document r318660, various NFS client fixes.

Sponsored by: The FreeBSD Foundation

7 years agoDocument r318785, vfs.zfs.debug_flags -> vfs.zfs.debugflags rename.
gjb [Thu, 25 May 2017 10:47:18 +0000 (10:47 +0000)]
Document r318785, vfs.zfs.debug_flags -> vfs.zfs.debugflags rename.

Sponsored by: The FreeBSD Foundation

7 years agoMFC r318863:
jlh [Thu, 25 May 2017 09:27:33 +0000 (09:27 +0000)]
MFC r318863:
  Replace stale handbook URL with the proper one.

7 years agoMFC r318014, r318091, r318125, and r318263.
np [Thu, 25 May 2017 01:59:58 +0000 (01:59 +0000)]
MFC r318014, r318091, r318125, and r318263.

r318014:
cxgbe(4): Fixes related to the knob that controls link autonegotiation.

- Do not leak the adapter lock in sysctl_autoneg.
- Accept only 0 or 1 as valid settings for autonegotiation.
- A fixed speed must be requested by the driver when autonegotiation is
  disabled otherwise the firmware will reject the l1cfg command.  Use
  the top speed supported by the port for now.

r318091:
cxgbe(4): Do not assume that if_qflush is always followed by inteface-down.

r318125:
Adjust whitespace and fix a comment.  No functional change.

r318263:
cxgbe(4): netmap-only interrupts for a VI do not have an associated rxq
or ofld_rxq and should be ignored by vi_intr_iq.

Sponsored by: Chelsio Communications

7 years agoMFC r318646:
gjb [Thu, 25 May 2017 01:45:43 +0000 (01:45 +0000)]
MFC r318646:
 Allow PORTSDIR to be overridden in the pkg-stage target.

PR: 218393
Sponsored by: The FreeBSD Foundation

7 years agoMFC r318599, r318643:
gjb [Thu, 25 May 2017 01:43:39 +0000 (01:43 +0000)]
MFC r318599, r318643:
 r318599 (bjk):
  Updates for efivar.8

  Fix spelling, appease igor and mandoc -Tlint, and adopt the
  language style to be more consistent with our corpus.

 r318643:
  Correct the "first appeared in" version for efivar(8).

Sponsored by: The FreeBSD Foundation

7 years agoMFC r317702, r317847, r318307
np [Thu, 25 May 2017 01:40:40 +0000 (01:40 +0000)]
MFC r317702, r317847, r318307

r317702:
cxgbe(4): Support routines for Tx traffic scheduling.

- Create a new file, t4_sched.c, and move all of the code related to
  traffic management from t4_main.c and t4_sge.c to this file.
- Track both Channel Rate Limiter (ch_rl) and Class Rate Limiter (cl_rl)
  parameters in the PF driver.
- Initialize all the cl_rl limiters with somewhat arbitrary default
  rates and provide routines to update them on the fly.
- Provide routines to reserve and release traffic classes.

r317847:
cxgbe(4): The Tx scheduler initialization either works or doesn't.  It
doesn't need a refresh in either case.

r318307:
cxgbe(4): Avoid an out of bounds access when an attempt to unbind a tx
queue from a traffic class fails.

Sponsored by: Chelsio Communications

7 years agoMFC r308737, r308779:
gjb [Thu, 25 May 2017 01:31:12 +0000 (01:31 +0000)]
MFC r308737, r308779:
 r308737:
  Pass SWAPSIZE in env(1) when invoking mk-vmimage.sh, otherwise
  mkimg(1) does not create the second partition after r307008.

 r308779:
  Pass SWAPSIZE in env(1) when invoking mk-vmimage.sh for the
  vm-image target, missed in r308737.

Sponsored by: The FreeBSD Foundation

7 years agoMFC r318479:
markj [Thu, 25 May 2017 01:17:07 +0000 (01:17 +0000)]
MFC r318479:
Don't bother enqueuing a page immediately before freeing it.

7 years agoMFC r318476, r318478:
markj [Thu, 25 May 2017 01:15:53 +0000 (01:15 +0000)]
MFC r318476, r318478:
Fix up some kern_yield() usages.

7 years agoMFC r318124:
np [Thu, 25 May 2017 01:14:34 +0000 (01:14 +0000)]
MFC r318124:

ip6_output runs with the inp lock held, just like ip_output.

7 years agoMFC r318191:
markj [Thu, 25 May 2017 01:09:45 +0000 (01:09 +0000)]
MFC r318191:
Let ptracestop() suspend threads sleeping in an SBDRY section.

7 years agoMFC r317820 and r317837.
np [Thu, 25 May 2017 01:00:57 +0000 (01:00 +0000)]
MFC r317820 and r317837.

r317820:
cxgbe(4): Update the list of PCIe devices claimed by the driver.  At
this point any board with a T6 should just work.

r317837:
cxgbe(4):  Update the VF device ids too.  This should have been part
of r317820.

Sponsored by: Chelsio Communications

7 years agoMFC r317041:
np [Thu, 25 May 2017 00:43:56 +0000 (00:43 +0000)]
MFC r317041:

cxgbe: Add tunables to control the number of LRO entries and the number
of rx mbufs that should be presorted before LRO.  There is no change in
default behavior.

Sponsored by: Chelsio Communications

7 years agoMFC r318603: nm: document 'r' symbol type
emaste [Thu, 25 May 2017 00:34:01 +0000 (00:34 +0000)]
MFC r318603: nm: document 'r' symbol type

PR: 219245
Sponsored by: The FreeBSD Foundation

7 years agoMFC r316971:
np [Thu, 25 May 2017 00:16:01 +0000 (00:16 +0000)]
MFC r316971:

cxgbe: Add a tunable to configure the SGE time scaler, which is
available starting with T6.  The values in the timer holdoff registers
are multiplied by the scaling factor before use.

dev.<nexus>.<n>.holdoff_timers shows the final values of the
timers in microseconds.

Sponsored by: Chelsio Communications

7 years agoMFC r316506:
np [Wed, 24 May 2017 23:08:17 +0000 (23:08 +0000)]
MFC r316506:

cxgbe(4): Program the global RSS key once instead of once per ifnet.

7 years agoMFC r316172:
np [Wed, 24 May 2017 22:47:45 +0000 (22:47 +0000)]
MFC r316172:

cxgbe: Don't call t4_edc_err_read for errors not related to the EDCs.

Sponsored by: Chelsio Communications

7 years agoMFC r309725:
np [Wed, 24 May 2017 21:53:47 +0000 (21:53 +0000)]
MFC r309725:

cxgbe(4): netmap does not set IFCAP_NETMAP in an ifnet's if_capabilities
any more (since r307394).  Do it in the driver instead.

Sponsored by: Chelsio Communications

7 years agoMFC r317759:
asomers [Wed, 24 May 2017 21:18:13 +0000 (21:18 +0000)]
MFC r317759:

Fix memory leaks in camcontrol

Reported by: Coverity
CID: 13316741331675
Reviewed by: ken
Sponsored by: Spectra Logic Corp
Differential Revision: https://reviews.freebsd.org/D10588

7 years agoMFC r307380:
np [Wed, 24 May 2017 20:54:14 +0000 (20:54 +0000)]
MFC r307380:

Fix typo in comments.

7 years agoMFC r317755, r317758
asomers [Wed, 24 May 2017 20:52:47 +0000 (20:52 +0000)]
MFC r317755, r317758

r317755:
Various Coverity fixes in ifconfig(8)

* Exit early if kldload(2) fails (1011259). This is the only change that
  affects ifconfig's behavior.
* Close memory and resource leaks (130562413052051007100)
* Mark usage() as _Noreturn (13058061305750)
* Fix some dereference after null checks (1011474, 270774)

Reported by: Coverity
CID: 130562413052051007100130580613057501011474,
CID: 270774, 1011259
Reviewed by: cem
Sponsored by: Spectra Logic Corp
Differential Revision: https://reviews.freebsd.org/D10587

r317758:
Unbreak ifconfig for mlx4en(4) after r317755

ifconfig doesn't correctly infer mlx interfaces' module names, so it will
attempt to load the mlx(4) module even when not necessary.

Reported by: rstone
X-MFC-With: 317755
Sponsored by: Spectra Logic Corp

7 years agoMFC r307008:
np [Wed, 24 May 2017 20:33:49 +0000 (20:33 +0000)]
MFC r307008:

Add the ability to override the size of the swap partition when building
VM images.  The default continues to be 1G.

Sponsored by: Chelsio Communications

7 years agoMFC r313318:
np [Wed, 24 May 2017 20:28:48 +0000 (20:28 +0000)]
MFC r313318:

cxgbe(4):  Allow tunables that control the number of queues to be set to
'-n' to tell the driver to create _up to_ 'n' queues if enough cores are
available.  For example, setting hw.cxgbe.nrxq10g="-32" will result in
16 queues if the system has 16 cores, 32 if it has 32.

There is no change in the default number of queues of any type.

Sponsored by: Chelsio Communications

7 years agoMFC r317715:
asomers [Wed, 24 May 2017 20:25:40 +0000 (20:25 +0000)]
MFC r317715:

ifconfig displays ND6_IFF_NO_DAD as "IGNORELOOP"

PR: 218958
Reviewed by: kristof
Differential Revision: https://reviews.freebsd.org/D10543

7 years agoMFC 318401
jpaetzel [Wed, 24 May 2017 20:10:38 +0000 (20:10 +0000)]
MFC 318401

Increase the number of LUNs this hardware can support.

Experimentally we know this value works, but the hardware
may support an even higher value.

PR: 213876
Reported by: J.Catrysse@proximedia.be

7 years agoMFC r313346:
np [Wed, 24 May 2017 19:57:22 +0000 (19:57 +0000)]
MFC r313346:

cxgbe/t4_tom: Fix CLIP entry refcounting on the passive side.  Every
IPv6 connection being handled by the TOE should have a reference on its
CLIP entry.

Sponsored by: Chelsio Communications

7 years agoMFC r314131:
np [Wed, 24 May 2017 19:18:36 +0000 (19:18 +0000)]
MFC r314131:

Avoid NULL dereference in a couple of sysctl handlers in ibcore.
iw_cxgbe sets ib_device->dma_device to NULL (since r311880).

Sponsored by: Chelsio Communications

7 years agoMFC r311880, r314167, r316118, r316571, r316573, r316580, r316936-r316937,
np [Wed, 24 May 2017 18:14:57 +0000 (18:14 +0000)]
MFC r311880, r314167, r316118, r316571, r316573, r316580, r316936-r316937,
r316940, and r317410.

r311880:
The iw_cxgb and iw_cxgbe drivers should not use a FreeBSD device_t where
a linuxkpi style device is expected.  If OFED/linuxkpi actually starts
using this field then we'll have to figure out whether to create fake
devices for these drivers or have linuxkpi deal with NULL device.

This mismatch was first reported as part of D6585.

r314167:
cxgbe/iw_cxgbe: Minor changes for T6.

r316118:
cxgbe/iw_cxgbe: T6 has no limit on the amount of memory that can be
registered in one ib_reg_phys_mr.

r316571:
cxgbe/iw_cxgbe: Remove bad cast that resulted in incorrect length for
memory regions larger than 4GB.

r316573:
cxgbe/iw_cxgbe: Replace a magic constant with something more readable
(and accurate).

T4 and later have an extra bit for page shift so the maximum page size
is 8TB (shift of 12 + 31) instead of 128MB (12 + 15).  This saves space
in the chip's PBL (physical buffer list) when registering very large
memory regions.

r316580:
cxgbe/iw_cxgbe: Remove another bad cast.  This should have been
included in r316571.

r316936:
cxgbe/iw_cxgbe: hw supports 64K (not 32K) Protection Domains.

r316937:
cxgbe/iw_cxgbe: Report accurate page_size_cap in ib_query_device.

r316940:
cxgbe/iw_cxgbe: Report the actual values of various parameters as
configured by the firmware.

r317410:
cxgbe/iw_cxgbe: Pull in some updates to c4iw_wait_for_reply from the
iw_cxgb4 Linux driver.

Sponsored by: Chelsio Communications

7 years agoMFC r316774:
np [Wed, 24 May 2017 17:51:07 +0000 (17:51 +0000)]
MFC r316774:

cxgbe: Query some more RDMA related parameters from the firmware.

Sponsored by: Chelsio Communications

7 years agoMFC r312346:
loos [Wed, 24 May 2017 16:39:19 +0000 (16:39 +0000)]
MFC r312346:

Set the the wp_disabled flag when asked to.

While here, add the missing new line.

Sponsored by: Rubicon Communications, LLC (Netgate)

7 years agoDocument r318763, EARLY_AP_STARTUP enabled by default.
gjb [Wed, 24 May 2017 14:22:22 +0000 (14:22 +0000)]
Document r318763, EARLY_AP_STARTUP enabled by default.

Sponsored by: The FreeBSD Foundation

7 years agoMFC r316854: rename vfs.zfs.debug_flags to vfs.zfs.debugflags
avg [Wed, 24 May 2017 11:38:31 +0000 (11:38 +0000)]
MFC r316854: rename vfs.zfs.debug_flags to vfs.zfs.debugflags

Since this is a stable branch vfs.zfs.debug_flags sysctl is also kept.
The corresponing tunable could never work.

7 years agoMFC r316853: dtrace: fix normalization of stddev aggregation
avg [Wed, 24 May 2017 11:10:55 +0000 (11:10 +0000)]
MFC r316853: dtrace: fix normalization of stddev aggregation

Sponsored by: Panzura

7 years agoMFC r316643: use msr 0xc001100c to discover multi-node AMD processors
avg [Wed, 24 May 2017 11:09:06 +0000 (11:09 +0000)]
MFC r316643: use msr 0xc001100c to discover multi-node AMD processors

7 years agoMFC r318399:
ae [Wed, 24 May 2017 09:03:46 +0000 (09:03 +0000)]
MFC r318399:
  Set M_BCAST and M_MCAST flags on mbuf sent via divert socket.

  r290383 has changed how mbufs sent by divert socket are handled.
  Previously they are always handled by slow path processing in ip_input().
  Now ip_tryforward() is invoked from ip_input() before in_broadcast() check.
  Since diverted packet lost all mbuf flags, it passes the broadcast check
  in ip_tryforward() due to missing M_BCAST flag. In the result the broadcast
  packet is forwarded to the wire instead of be consumed by network stack.

  Add in_broadcast() check to the div_output() function. And restore the
  M_BCAST flag if destination address is broadcast for the given network
  interface.

  PR: 209491

7 years agoMFC r318400:
ae [Wed, 24 May 2017 09:01:54 +0000 (09:01 +0000)]
MFC r318400:
  Allow zero port specification in table entries with type flow.

  PR: 217620

7 years agoMFC r318135:
mmel [Wed, 24 May 2017 06:49:01 +0000 (06:49 +0000)]
MFC r318135:

  Fix parsing of 'vmov Q<n>.F32,Q<n>.F32' instruction.

7 years agoMFC r311846:
np [Wed, 24 May 2017 04:42:31 +0000 (04:42 +0000)]
MFC r311846:
cxgbe(4): Refresh t4_msg.h, mainly for definitions related to the crypto
engine.

7 years agoMFC r317170, r317389, and r317390.
np [Wed, 24 May 2017 03:56:48 +0000 (03:56 +0000)]
MFC r317170, r317389, and r317390.

r317170:
Remove redundant assignment.

r317389:
Frames that are not considered for LRO should not be counted in LRO
statistics.

r317390:
Flush the LRO ctrl as soon as lro_mbufs fills up.  There is no need to
wait for the next enqueue from the driver.

Sponsored by: Chelsio Communications

7 years agoMFC 315334: Decode arguments to chflagsat().
jhb [Wed, 24 May 2017 03:26:15 +0000 (03:26 +0000)]
MFC 315334: Decode arguments to chflagsat().

7 years agoMFC 315310: Spell "const" properly.
jhb [Wed, 24 May 2017 01:06:09 +0000 (01:06 +0000)]
MFC 315310: Spell "const" properly.

PR: 217797

7 years agoMFC 315283:
jhb [Wed, 24 May 2017 00:36:27 +0000 (00:36 +0000)]
MFC 315283:
Fix sysdecode_cap_rights which currently prints bogus capability rights.

PR: 217787

7 years agoMFC 310177: Enable EARLY_AP_STARTUP on amd64 and i386 kernels by default.
jhb [Wed, 24 May 2017 00:00:55 +0000 (00:00 +0000)]
MFC 310177: Enable EARLY_AP_STARTUP on amd64 and i386 kernels by default.

PR: 199321, 203682
Discussed with: re (kib)
Relnotes: yes

7 years agoMFC r316030, r317378: Add some useful examples to the sed man page.
brd [Tue, 23 May 2017 22:11:45 +0000 (22:11 +0000)]
MFC r316030, r317378: Add some useful examples to the sed man page.

Reviewed by: wblock, bcr
Differential Revision: https://reviews.freebsd.org/D9958

7 years agoMFC r309321:
mav [Tue, 23 May 2017 17:00:56 +0000 (17:00 +0000)]
MFC r309321:
Add `gmirror create` subcommand, alike to gstripe, gconcat, etc.

It is quite specific mode of operation without storing on-disk metadata.
It can be useful in some cases in combination with some external control
tools handling mirror creation and disks hot-plug.

Sponsored by: iXsystems, Inc.

7 years agoMove the SA entry to the tcpmd5 module addition.
gjb [Tue, 23 May 2017 15:04:29 +0000 (15:04 +0000)]
Move the SA entry to the tcpmd5 module addition.

Submitted by: ae
Sponsored by: The FreeBSD Foundation

7 years agoDocument the requirement for two SA entries, following the IPSEC
gjb [Tue, 23 May 2017 14:54:58 +0000 (14:54 +0000)]
Document the requirement for two SA entries, following the IPSEC
overhaul.

Submitted by: vangyzen
Sponsored by: The FreeBSD Foundation

7 years agomove p_sigqueue to the end of struct proc
badger [Tue, 23 May 2017 12:40:50 +0000 (12:40 +0000)]
move p_sigqueue to the end of struct proc

In order to preserve KBI in stable branches, replace the existing
p_sigqueue slot with padding and move the expanded (as of r315949)
p_sigqueue to the end of the struct.

This is a repeat of r317529 (which concerned td_sigqueue in struct
thread) for p_sigqueue in struct proc.

Virtualbox modules (and possibly others) are affected without this fix.

Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D10843

7 years agoMFC r318021,r318251:
mmel [Tue, 23 May 2017 12:03:59 +0000 (12:03 +0000)]
MFC r318021,r318251:

  r318021:
    Introduce pmap_remap_vm_attr(), it allows to remap one VM memattr class to
    another.
  r318251:
    Clarify usage rules for pmap_remap_vm_attr().  Not a functional change.

7 years agoMFC r317904:
trasz [Tue, 23 May 2017 08:11:55 +0000 (08:11 +0000)]
MFC r317904:

.Xr resizewin from stty(1) man page.

7 years agoMFC r318481:
trasz [Tue, 23 May 2017 08:10:35 +0000 (08:10 +0000)]
MFC r318481:

Language fixes.

7 years agoMFC r318138:
trasz [Tue, 23 May 2017 08:09:44 +0000 (08:09 +0000)]
MFC r318138:

Revert to pre-r318116 wording to not give the false impression
that setting the kernels' idea of terminal size is somehow an
alternative to environment variables.

7 years agoMFC r318116:
trasz [Tue, 23 May 2017 08:09:04 +0000 (08:09 +0000)]
MFC r318116:

Random updates to resizewin(1) man page.

7 years agoMFC r317935:
trasz [Tue, 23 May 2017 08:08:23 +0000 (08:08 +0000)]
MFC r317935:

Sort variable declarations; no functional changes.

7 years agoMFC rr317934:
trasz [Tue, 23 May 2017 08:07:39 +0000 (08:07 +0000)]
MFC rr317934:

Add resizewin -z. It makes resizewin not do anything if the terminal
size is already set to something other than zero. It's supposed to be
called from eg /etc/profile - it's not neccessary to query terminal
size when logging in over the network, because the protocol used already
takes care of this, but it's neccessary when logging over a serial line.

7 years agoMFC r317933:
trasz [Tue, 23 May 2017 08:06:00 +0000 (08:06 +0000)]
MFC r317933:

Use tcflush(3) instead of (nonstandard) TIOCFLUSH.

7 years agoMFC r317909:
trasz [Tue, 23 May 2017 08:04:36 +0000 (08:04 +0000)]
MFC r317909:

Make resizewin(1) discard the terminal queues, to lower the chance
for "unable to parse response" error which happens when youre typing
too fast for the machine you're running it on.

7 years agoMFC r317905:
trasz [Tue, 23 May 2017 08:02:21 +0000 (08:02 +0000)]
MFC r317905:

Rename a variable, hopefully fixing build after r317901.

7 years agoMFC r317901:
trasz [Tue, 23 May 2017 08:01:24 +0000 (08:01 +0000)]
MFC r317901:

Improve error reporting in resizewin(1).

7 years agoMFC r308474, r308691, r309203, r309365, r309703, r309898, r310720,
markj [Tue, 23 May 2017 07:27:30 +0000 (07:27 +0000)]
MFC r308474, r308691, r309203, r309365, r309703, r309898, r310720,
r308489, r308706:
Add PQ_LAUNDRY and remove PG_CACHED pages.

7 years agoMFC r318024,r318025:
mmel [Tue, 23 May 2017 07:21:26 +0000 (07:21 +0000)]
MFC r318024,r318025:

  r318024:
    Fix _Unwind_Backtrace symbol version for ARM.
  r318025:
    Revert accidentally changed std.armv6 in r318024.

7 years agoMFC: r318287
rmacklem [Mon, 22 May 2017 22:02:06 +0000 (22:02 +0000)]
MFC: r318287
Make nfscl_mtofh() return ENXIO when *nfhpp == NULL.

r317272 introduced a case where nfscl_mtofh() could return 0 when
*nfhpp is NULL. This patch makes it return ENXIO for this case.

7 years agoMFC: r317931
rmacklem [Mon, 22 May 2017 21:41:34 +0000 (21:41 +0000)]
MFC: r317931
Fix mount_nfs so that it doesn't create mounttab entries for NFSv4 mounts.

The NFSv4 protocol doesn't use the Mount protocol, so it doesn't make sense
to add an entry for an NFSv4 mount to /var/db/mounttab. Also, r308871
modified umount so that it doesn't remove any entry created by mount_nfs.

7 years agoMFC: r317906
rmacklem [Mon, 22 May 2017 19:34:37 +0000 (19:34 +0000)]
MFC: r317906
Fix the client side krpc from doing TCP reconnects for ERESTART from sosend().

When sosend() replies ERESTART in the client side krpc, it indicates that
the RPC message hasn't yet been sent and that the send queue is full or
locked while a signal is posted for the process.
Without this patch, this would result in a RPC_CANTSEND reply from
clnt_vc_call(), which would cause clnt_reconnect_call() to create a new
TCP transport connection. For most NFS servers, this wasn't a serious problem,
although it did imply retries of outstanding RPCs, which could possibly
have missed the DRC.
For an NFSv4.1 mount to AmazonEFS, this caused a serious problem, since
AmazonEFS often didn't retain the NFSv4.1 session and would reply with
NFS4ERR_BAD_SESSION. This implies to the client a crash/reboot which
requires open/lock state recovery.

Three options were considered to fix this:
- Return the ERESTART all the way up to the system call boundary and then
  have the system call redone. This is fraught with risk, due to convoluted
  code paths, asynchronous I/O RPCs etc. cperciva@ worked on this, but it
  is still a work in prgress and may not be feasible.
- Set SB_NOINTR for the socket buffer. This fixes the problem, but makes
  the sosend() completely non interruptible, which kib@ considered
  inappropriate. It also would break forced dismount when a thread
  was blocked in sosend().
- Modify the retry loop in clnt_vc_call(), so that it loops for this case
  for up to 15sec. Testing showed that the sosend() usually succeeded by
  the 2nd retry. The extreme case observed was 111 loop iterations, or
  about 100msec of delay.
This third alternative is what is implemented in this patch, since the
change is:
- localized
- straightforward
- forced dismount is not broken by it.

This patch has been tested by cperciva@ extensively against AmazonEFS.

7 years agoMFC r318383
davidcs [Mon, 22 May 2017 19:28:38 +0000 (19:28 +0000)]
MFC r318383
 QL_DPRINT macro modfied to handle multiple args; print line#.

Submitted by:Shminderjit.Singh@cavium.com

7 years agoMFC r318382
davidcs [Mon, 22 May 2017 19:22:06 +0000 (19:22 +0000)]
MFC r318382
  1. Move Rx Processing to fp_taskqueue(). With this CPU utilization for
     processing interrupts drops to around 1% for 100G and under 1% for
     other speeds.
  2. Use sysctls for TRACE_LRO_CNT and TRACE_TSO_PKT_LEN
  3. remove unused mtx tx_lock
  4. bind taskqueue kernel thread to the appropriate cpu core
  5. when tx_ring is full, stop further transmits till at least 1/16th of
     the Tx Ring is empty. In our case 1K entries. Also if there are
     rx_pkts to process, put the taskqueue thread to sleep for 100ms,
     before enabling interrupts.
  6. Use rx_pkt_threshold of 128.

7 years agoMFC r305613 (gabor):
gjb [Mon, 22 May 2017 16:11:46 +0000 (16:11 +0000)]
MFC r305613 (gabor):
 Fix typo.

PR: 211245
Sponsored by: The FreeBSD Foundation

7 years agoMFC r307469 (imp):
gjb [Mon, 22 May 2017 16:07:17 +0000 (16:07 +0000)]
MFC r307469 (imp):
 Allow root_rw_mount to be both lower and upper case. Before, if it was
 upper case, you'd wind up with a read-only filesystem when you should
 sometimes.

PR: 213549
Sponsored by: The FreeBSD Foundation

7 years agoMFC r313955 (emaste):
gjb [Mon, 22 May 2017 15:51:17 +0000 (15:51 +0000)]
MFC r313955 (emaste):
 bsdgrep: document ignored option -u

 MSDOS and Windows GNU grep uses -u to mean "print byte offsets as if
 running on an UNIX system." The option has no effect on systems that
 do not use CRLF line endings.

PR: 171200
Sponsored by: The FreeBSD Foundation

7 years agoMFC r318189:
asomers [Mon, 22 May 2017 15:12:49 +0000 (15:12 +0000)]
MFC r318189:

vdev_geom may associate multiple vdevs per g_consumer

vdev_geom.c currently uses the g_consumer's private field to point to a
vdev_t. That way, a GEOM event can cause a change to a ZFS vdev. For
example, when you remove a disk, the vdev's status will change to REMOVED.
However, vdev_geom will sometimes attach multiple vdevs to the same GEOM
consumer. If this happens, then geom events will only be propagated to one
of the vdevs.

Fix this by storing a linked list of vdevs in g_consumer's private field.

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_geom.c

* g_consumer.private now stores a linked list of vdev pointers associated
  with the consumer instead of just a single vdev pointer.

* Change vdev_geom_set_physpath's signature to more closely match
  vdev_geom_set_rotation_rate

* Don't bother calling g_access in vdev_geom_set_physpath. It's guaranteed
  that we've already accessed the consumer by the time we get here.

* Don't call vdev_geom_set_physpath in vdev_geom_attach. Instead, call it
  in vdev_geom_open, after we know that the open has succeeded.

PR: 218634
Reviewed by: gibbs
Sponsored by: Spectra Logic Corp
Differential Revision: https://reviews.freebsd.org/D10391

7 years agoDocument r318576, efivar(8) addition.
gjb [Mon, 22 May 2017 13:56:34 +0000 (13:56 +0000)]
Document r318576, efivar(8) addition.

Sponsored by: The FreeBSD Foundation

7 years agoDocument r315514, if_ipsec(4) addition.
gjb [Mon, 22 May 2017 13:27:31 +0000 (13:27 +0000)]
Document r315514, if_ipsec(4) addition.

Submitted by: ae
Sponsored by: The FreeBSD Foundation

7 years agoFix typo: s/LDD/LLD/
gjb [Mon, 22 May 2017 13:15:59 +0000 (13:15 +0000)]
Fix typo: s/LDD/LLD/

Spotted by: vangyzen
Sponsored by: The FreeBSD Foundation

7 years agoDocument r309312, PCI passthrough enabled on Hyper-V VMs.
gjb [Mon, 22 May 2017 13:00:11 +0000 (13:00 +0000)]
Document r309312, PCI passthrough enabled on Hyper-V VMs.
Document r314091, hv_netvsc(4) SR-IOV/VF support.
Document r316272, Hyper-V Gen 2 support.
Document r317119, Hyper-V synthetic keyboard support.

Submitted by: sephe
Sponsored by: The FreeBSD Foundation

7 years agoDocument r318472, LDD enabled by default on arm64.
gjb [Mon, 22 May 2017 13:00:10 +0000 (13:00 +0000)]
Document r318472, LDD enabled by default on arm64.

Submitted by: andrew
Sponsored by: The FreeBSD Foundation

7 years agoMFC r318530:
mmel [Mon, 22 May 2017 12:35:31 +0000 (12:35 +0000)]
MFC r318530:

  Increase maximum text segment size.  LLVM binaries are huge...

7 years agoFix typo and grammar.
gjb [Mon, 22 May 2017 12:28:20 +0000 (12:28 +0000)]
Fix typo and grammar.

Submitted by: pfg
Sponsored by: The FreeBSD Foundation

7 years agoMFC:
n_hibma [Mon, 22 May 2017 10:28:17 +0000 (10:28 +0000)]
MFC:

------------------------------------------------------------------------
r317923 | n_hibma | 2017-05-07 23:11:28 +0200 (Sun, 07 May 2017) | 8 lines

Fix the output of very large rebind, renew and lease time options in
lease file.

Some routers set very large values for rebind time (Netgear) and these
are erroneously reported as negative in the leasefile. This was due to a
wrong printf format specification of %ld for an unsigned long on 32-bit
platforms.

------------------------------------------------------------------------
r317915 | n_hibma | 2017-05-07 21:59:37 +0200 (Sun, 07 May 2017) | 16 lines

Fix handling of large DHCP expiry values.

They would overflow a signed 32-bit time_t on 32 bit architectures. This
was taken care of, but a compiler optimisation makes this behave
erratically. This could be resolved by adding a -fwrapv flag, but
instead we can check the value before adding the current timestamp to
it.

In the lease file values are still wrong though:

option dhcp-rebinding-time -644245096;

PR: 218980

7 years agoMFC r318531:
hselasky [Mon, 22 May 2017 08:17:07 +0000 (08:17 +0000)]
MFC r318531:

mlx4: Use the CQ quota for SRIOV when creating completion EQs

When creating EQs to handle CQ completion events for the PF or for
VFs, we create enough EQE entries to handle completions for the max
number of CQs that can use that EQ.

When SRIOV is activated, the max number of CQs a VF (or the PF) can
obtain is its CQ quota (determined by the Hypervisor resource
tracker).  Therefore, when creating an EQ, the number of EQE entries
that the VF should request for that EQ is the CQ quota value (and not
the total number of CQs available in the firmware).

Under SRIOV, the PF, also must use its CQ quota, because the resource
tracker also controls how many CQs the PF can obtain.

Using the firmware total CQs instead of the CQ quota when creating EQs
resulted wasting MTT entries, due to allocating more EQEs than were
needed.

Sponsored by: Mellanox Technologies

7 years agoMFC r316103:
ngie [Mon, 22 May 2017 06:26:22 +0000 (06:26 +0000)]
MFC r316103:

Remove redundant declarations

They're already defined in libstand.h

7 years agoMFC r317594:
ngie [Mon, 22 May 2017 06:24:43 +0000 (06:24 +0000)]
MFC r317594:

usb(4): manpage cleanup

1. Wrap at <80 columns for readability when editing. Rewrap some lines
   prematurely wrapped to better fit in <80 columns and not waste
   vertical space.
2. Fix SEE ALSO sorting (sort by section first, then manpage name).
3. Tweak the compound device description slightly by adding soft stops
   via commas.

Sponsored by Dell EMC Isilon

7 years agoMFC r315775:
ngie [Mon, 22 May 2017 06:20:22 +0000 (06:20 +0000)]
MFC r315775:

Add a post-humous manpage for cd9660(5), the ISO-9660 file system

Describe (briefly) how to compile the filesystem into the kernel and
load as a module.

Reference cd9660(5) in mount(8) and mount_cd9660(8).

7 years agoRegenerate src.conf(5)
ngie [Mon, 22 May 2017 06:19:27 +0000 (06:19 +0000)]
Regenerate src.conf(5)

7 years agoRevert commit again to correct it
ngie [Mon, 22 May 2017 06:18:15 +0000 (06:18 +0000)]
Revert commit again to correct it

Obviously my brain's on autopilot tonight

7 years agoMFC r311135:
ngie [Mon, 22 May 2017 06:17:14 +0000 (06:17 +0000)]
MFC r311135:

Add documentation for MK_EFI added in r307243

7 years agoMFC r311135:
ngie [Mon, 22 May 2017 06:15:50 +0000 (06:15 +0000)]
MFC r311135:

Add documentation for MK_EFI added in r307243

7 years agoMFC r315766,r315768:
ngie [Mon, 22 May 2017 06:08:19 +0000 (06:08 +0000)]
MFC r315766,r315768:

r315766:

Note that tools/build/options/makeman automatically generated src.conf(5)

This (as a bonus) fixes mdoc warnings with src.conf(5) :)..

r315768:

Remove the .Pa portion I added to the .An macro in AUTHORS section

That doesn't work (mandoc complains about the macro being empty).

That's what I get for being clever and not verifying before committing
things again *sigh*.

Partyhat to: ngie

7 years agoMFC r315793:
ngie [Mon, 22 May 2017 06:07:09 +0000 (06:07 +0000)]
MFC r315793:

intro(3): fix markup

- Use `Em` with `.It` macro when referring to other libraries, instead of
  `Xr`.
- Use `.Em` instead of `.Xr` when referring to libraries.
- Remove commented out lines.

7 years agoMFC r317160:
ngie [Mon, 22 May 2017 06:05:35 +0000 (06:05 +0000)]
MFC r317160:

Clean up trailing whitespace

No functional changes

7 years agoMFC r317154:
ngie [Mon, 22 May 2017 06:03:51 +0000 (06:03 +0000)]
MFC r317154:

Print out the signal number on exit in terminate(..) if WARMSTART is compiled
into rpcbind.

The signal number can provide helpful diagnostic info.

Obtained from: Isilon OneFS

7 years agoMFC 318512
sephe [Mon, 22 May 2017 02:09:27 +0000 (02:09 +0000)]
MFC 318512

    net/vlan: Revert 305177

    Miss read the parentheses.

    Reported by:    oleg@
    Reviewed by:    hps@
    Sponsored by:   Microsoft

7 years agoMFC r317863:
bdrewery [Sat, 20 May 2017 17:09:27 +0000 (17:09 +0000)]
MFC r317863:

  Fix some "don't build" optimizations from r308599 not working due to a typo.

7 years agoMFC efivar(8) (by imp):
kib [Sat, 20 May 2017 16:12:44 +0000 (16:12 +0000)]
MFC efivar(8) (by imp):

List of revisions merged:
r307070
r307071
r307072
r307074
r307189
r307224
r307339
r307390
r307391
r309776
r314231
r314232
r314615
r314616
r314617
r314618
r314619
r314620
r314621
r314623
r314890
r314925
r314926
r314927
r314928
r315770
r315771

Discussed with: gjb (re), imp
Sponsored by: The FreeBSD Foundation

7 years agoMFC r314370,r318130,r318167:
jhibbits [Sat, 20 May 2017 05:12:32 +0000 (05:12 +0000)]
MFC r314370,r318130,r318167:

  DTrace related fixes for PowerPC.

  r314370:
    Unbreak kernel breakpoints, broken for ~4 years now
  r318130:
    Fix the encoded instruction for FBT traps on powerpc
  r318167:
    Fix stack tracing in dtrace for powerpc

7 years agoDocument r316613, C standard library has been updated to make use
gjb [Sat, 20 May 2017 01:04:19 +0000 (01:04 +0000)]
Document r316613, C standard library has been updated to make use
of reallocarray(3).
Document r318121, system libraries have been updated to make use
of reallocarray(3).
Document r315282, GNU __nonnull__ attribute have been replaced with
the more benign Clang nullability attributes.

Submitted by: pfg
Sponsored by: The FreeBSD Foundation

7 years agoDocument r316613, C standard library has been updated to make use
gjb [Sat, 20 May 2017 01:04:18 +0000 (01:04 +0000)]
Document r316613, C standard library has been updated to make use
of reallocarray(3).
Document r318121, system libraries have been updated to make use
of reallocarray(3).

Submitted by: pfg
Sponsored by: The FreeBSD Foundation

7 years agos/nullability attribute/nullability qualifier/
gjb [Sat, 20 May 2017 01:04:18 +0000 (01:04 +0000)]
s/nullability attribute/nullability qualifier/

Sponsored by: The FreeBSD Foundation

7 years agoMFC r318256:
kib [Sat, 20 May 2017 00:41:12 +0000 (00:41 +0000)]
MFC r318256:
Style.

7 years agoMFC 318360: Fix p_endcopy.
jhb [Fri, 19 May 2017 23:01:55 +0000 (23:01 +0000)]
MFC 318360: Fix p_endcopy.

For p_endcopy to work correctly, it must be the name of the next element
in struct proc after the end of the copy region, not the name of the
last element in the copy region.  Currently, the last element
(p_elf_flags) is not being copied.  In addition, the p_xexit and
p_xsig fields should not be copied on fork, so move them out of the
copy region.

Note that for stable/11 the fix is a bit simpler than in HEAD as it
merely consists of formally moving p_xexit and p_xsig out of the
copy region by shrinking the end of the copy region.