dragonfly.git
3 years agoMerge branch 'vendor/ELFTOOLCHAIN'
John Marino [Mon, 2 Nov 2015 23:44:27 +0000 (00:44 +0100)]
Merge branch 'vendor/ELFTOOLCHAIN'

3 years agoImport ELF Tool Chain r3255 into vendor branch gitea/vendor/ELFTOOLCHAIN origin/vendor/ELFTOOLCHAIN
John Marino [Mon, 2 Nov 2015 23:43:51 +0000 (00:43 +0100)]
Import ELF Tool Chain r3255 into vendor branch

3 years agogdb: Reduce the recent changes to only the necessary bits in amd64dfly-nat.c.
Sascha Wildner [Mon, 2 Nov 2015 22:56:43 +0000 (23:56 +0100)]
gdb: Reduce the recent changes to only the necessary bits in amd64dfly-nat.c.

This makes maintenance of base and dports gdb easier.

Suggested-by: marino
Discussed-with: tuxillo

3 years agoeuc encoding: Emit EILSEQ on illegal euc conversions
John Marino [Mon, 2 Nov 2015 23:04:24 +0000 (00:04 +0100)]
euc encoding: Emit EILSEQ on illegal euc conversions

Omission caught by NetBSD testsuite, run by bapt@FreeBSD.org

3 years agodrm: Add ACCESS_ONCE()
François Tigeot [Mon, 2 Nov 2015 19:18:32 +0000 (20:18 +0100)]
drm: Add ACCESS_ONCE()

3 years agogdb - Workaround to enable debugging in vkernels.
Antonio Huete Jimenez [Mon, 2 Nov 2015 16:35:26 +0000 (17:35 +0100)]
gdb - Workaround to enable debugging in vkernels.

- Currently vkernel(7) does not have support for PT_GETDBREGS/PT_SETDBREGS
  via ptrace(2) so conditionalize gdb initialization code to make a special
  case for vkernels.

3 years agogdb - Remove FreeBSD specific code.
Antonio Huete Jimenez [Mon, 2 Nov 2015 15:31:18 +0000 (16:31 +0100)]
gdb - Remove FreeBSD specific code.

3 years agogdb - Fix indentantion in DragonFly specific files.
Antonio Huete Jimenez [Mon, 2 Nov 2015 15:18:26 +0000 (16:18 +0100)]
gdb - Fix indentantion in DragonFly specific files.

- No functional changes

3 years agodrm.4: Document new sysctls.
Sascha Wildner [Mon, 2 Nov 2015 17:59:15 +0000 (18:59 +0100)]
drm.4: Document new sysctls.

Submitted-by: Rimvydas Jasinskas <rimvydas.jasinskas@gmail.com>
3 years agodrm.4: Move the sysctl section to using .Bl/.El
Sascha Wildner [Mon, 2 Nov 2015 17:56:39 +0000 (18:56 +0100)]
drm.4: Move the sysctl section to using .Bl/.El

3 years agolocales: Remove symlinks UTF8 => UTF-8
John Marino [Mon, 2 Nov 2015 11:48:22 +0000 (12:48 +0100)]
locales: Remove symlinks UTF8 => UTF-8

In retrospect, having an alias for UTF-8 does not bring any real benefits
and it can cause confusion.  Let's remove this *.UTF8 locale symlinks
which is closer to the convention of the other BSDs.

GCC testsuite is also removing utf8 and UTF8 locales for portability
with BSD.

3 years agosocket/recvfrom: Add MSG_CMSG_CLOEXEC support
zrj [Sat, 31 Oct 2015 15:34:58 +0000 (17:34 +0200)]
socket/recvfrom: Add MSG_CMSG_CLOEXEC support

While there, add test case and update manpages.

Obtained-from: FreeBSD

3 years agodrm: Simplify the initialization of the hw.dri sysctl tree.
Imre Vadász [Sun, 1 Nov 2015 21:03:43 +0000 (22:03 +0100)]
drm: Simplify the initialization of the hw.dri sysctl tree.

* Just use SYSCTL_NODE and SYSCTL_INT declarations, for initializing and
  and destroying the hw.dri subtree atomagically (including the debug,
  notyet, vblank_offdelay and timestamp_precision integer nodes) on
  module load/unload.

3 years agolocale type makefiles: respect ${SHAREDIR}
John Marino [Sun, 1 Nov 2015 21:29:52 +0000 (22:29 +0100)]
locale type makefiles: respect ${SHAREDIR}

The cldr2def tool was modified to generate makefiles that respect
the SHAREDIR definition rather than hardcoding it.

Taken-from: FreeBSD (collation branch)

3 years agolocales: Fix eucJP sorting (broken upstream?)
John Marino [Sun, 1 Nov 2015 20:47:23 +0000 (21:47 +0100)]
locales: Fix eucJP sorting (broken upstream?)

Sorting eucJP text with "sort" resulted in an illegal sequence while
"gsort" worked.  This was traced back to mbrtowc handling which was
broken for eucJP (probably eucCN, eucKR, and eucTW as well).  This
small fix took hours to figure out.  The OR operation to build the
wide character requires an unsigned character to work correctly.  The
euc wcrtowc conversion is probably broken upstream in Illumos as well.

Reported by: bapt@FreeBSD.org
Triggered by: misc/freebsd-doc-ja in ports (encoded in eucJP)

3 years agoif_vtnet: Add support for indirect buffers. Activate LRO by default again.
Imre Vadász [Sat, 31 Oct 2015 15:49:31 +0000 (16:49 +0100)]
if_vtnet: Add support for indirect buffers. Activate LRO by default again.

Taken-From: FreeBSD

3 years agovirtio: Add indirect buffers support to common virtio code.
Imre Vadász [Fri, 30 Oct 2015 22:15:40 +0000 (23:15 +0100)]
virtio: Add indirect buffers support to common virtio code.

* Uses contigmalloc instead of kmalloc for allocating the indirect
  descriptors, because DragonFly's kmalloc doesn't return sufficiently
  aligned or physically contiguous memory.

Taken-From: FreeBSD

3 years agodrm/radeon: Export PCI ID
François Tigeot [Sun, 1 Nov 2015 18:16:28 +0000 (19:16 +0100)]
drm/radeon: Export PCI ID

Submitted-by: Rimvydas Jasinskas
3 years agosys/dev/disk/dm: Add error count for stripe device
Tomohiro Kusumi [Sun, 1 Nov 2015 15:02:59 +0000 (00:02 +0900)]
sys/dev/disk/dm: Add error count for stripe device

This is for compatibility with Linux, but currently not used.
In order to use it, dm target needs to be able to implement
bio's completion handler which is currently unaccessible from
outside of sys/kern/vfs_bio.c.

3 years agosys/dev/disk/dm: Don't statically allocate MAX_STRIPES devices
Tomohiro Kusumi [Sun, 1 Nov 2015 14:42:25 +0000 (23:42 +0900)]
sys/dev/disk/dm: Don't statically allocate MAX_STRIPES devices

struct target_stripe_config should have 0 sized array for
struct target_stripe_dev instead of MAX_STRIPES sized one,
and kmalloc() whatever size necessary depending on # of
stripe devices.

3 years agovkernel - Override max-page-size for ld.*
Antonio Huete Jimenez [Sun, 1 Nov 2015 12:48:25 +0000 (13:48 +0100)]
vkernel - Override max-page-size for ld.*

- Move max-page-size setting to the vkernel-specific SYSTEM_LD as it
  was probably intended in ed9f17e. This overrides whatever the default
  is in ld.bfd or ld.gold.

3 years agovkernel - Bring back .hash section
Antonio Huete Jimenez [Sun, 1 Nov 2015 12:40:54 +0000 (13:40 +0100)]
vkernel - Bring back .hash section

  As of the switch to GCC 5 the default hash style to ld has changed from both
  (sysv,gnu) to gnu alone. This results in binaries without a .hash section
  which is fine for userland binaries but not for kernel ones since our kernel
  linker uses .hash section to resolve its own symbols.

  With the missing .hash section the vkernel kernel linker was not able do
  symbol lookups and that rendered kvm(3) useless and all its consumers stopped
  working.

  During the vkernel build, SYSTEM_LD is changed so that the normal buildkernel
  settings are overridden so by changing the parameters passed to ld we bring
  back the .hash section.

3 years agolibc: Fix (and improve) nl_langinfo (CODESET)
John Marino [Sun, 1 Nov 2015 10:42:10 +0000 (11:42 +0100)]
libc: Fix (and improve) nl_langinfo (CODESET)

The output of "locale charmap" is identical to the result of
nl_langinfo (CODESET) for any given locale.  The logic for returning the
codeset was very simplistic.  It just returned portion of the locale name
after the period (e.g. en_FR.ISO8859-1 returned "ISO8859-1").

When softlinks were added to locales, this broke.  e.g.:
   en_US returned ""
   en_FR.UTF8 returned "UTF8"
   en_FR.UTF-8 returned "UTF-8"
   zh_Hant_HK.Big5HKSCS returned "Big5HKSCS"
   zh_Hant_TW.Big5 returned "Big5"
   es_ES@euro returned ""

In order to fix this properly, the named locale cannot be used to
determine the encoding.  This information was almost available in the
rune data.  Unfortunately, all the single byte encodings were listed
as "NONE" encoding.

So I adjusted localedef tool to provide more information about the
encoding.  For example, instead of "NONE", the LC_CTYPE used by
fr_FR.ISO8859-15 is now encoded as "NONE:ISO8859-15".  The locale
handlers now check if the first four characters of the encoding is
"NONE" and if so, treats it as a single-byte encoding.

The nl_langinfo handling of CODESET was adjusting accordingly.  Now the
following is returned:
   en_US returns "ISO8859-1"
   fr_FR.UTF8 returns "UTF-8"
   fr_FR.UTF-8 returns "UTF-8"
   zh_Hant_HK.Big5HKSCS returns "Big5"
   zh_Hant_TW.Big5 returns "Big5"
   es_ES@euro returns "ISO8859-15"

as before, "C" and "POSIX" locales return "US-ASCII".  This is a big
improvement.  The result of nl_langinfo can never be a zero-length
string and it will always exclusively one of the values of the
character maps of /usr/src/tools/tools/locale/etc/final-maps.

3 years agokernel/drm: In linux/types.h, remove the duplicate definition of gfp_t.
Sascha Wildner [Sat, 31 Oct 2015 21:50:02 +0000 (22:50 +0100)]
kernel/drm: In linux/types.h, remove the duplicate definition of gfp_t.

3 years agokernel/drm: In linux/kernel.h add missing parentheses/casts to min_t/max_t.
Sascha Wildner [Sat, 31 Oct 2015 21:13:34 +0000 (22:13 +0100)]
kernel/drm: In linux/kernel.h add missing parentheses/casts to min_t/max_t.

3 years agosys/dev/disk/dm: Remove misleading comment
Tomohiro Kusumi [Sat, 31 Oct 2015 11:55:26 +0000 (20:55 +0900)]
sys/dev/disk/dm: Remove misleading comment

It takes offset and size parameters in sectors.
This comment seems to be specifying them in bytes, which is wrong.

3 years agosys/dev/disk/dm: Fix sanity checks for striped target [4/4]
Tomohiro Kusumi [Sat, 31 Oct 2015 09:07:54 +0000 (18:07 +0900)]
sys/dev/disk/dm: Fix sanity checks for striped target [4/4]

Add sanity checks on relation among
number of stripes, device size, and chunk size.

This is one of the requirements in Linux dm-striped, but
disable it for now as it may break systems that currently
use this target (which I doubt there is any because lvm2
and dm raid seem to be unusable on DragonFly).

3 years agosys/dev/disk/dm: Fix sanity checks for striped target [3/4]
Tomohiro Kusumi [Fri, 30 Oct 2015 15:16:12 +0000 (00:16 +0900)]
sys/dev/disk/dm: Fix sanity checks for striped target [3/4]

0 stripe is apparently invalid, but it's currently allowed
and generates unusable device.

  # kldload dm
  # dmsetup create stripe1 --table '0 1000 striped 0 10 /dev/da3 0 /dev/da4 0'
  # dmsetup ls --tree
  stripe1 (255:504430599)
  # dmsetup deps /dev/mapper/stripe1
  0 dependencies  :
  # dmsetup status /dev/mapper/stripe1

  # dmsetup table /dev/mapper/stripe1

  #

3 years agosys/dev/disk/dm: Fix sanity checks for striped target [2/4]
Tomohiro Kusumi [Sat, 31 Oct 2015 10:57:29 +0000 (19:57 +0900)]
sys/dev/disk/dm: Fix sanity checks for striped target [2/4]

Target specific args for striped is redundant in the sense
that it has an arg for # of stripes other than the args for
actual stripe devices. These two must meet below.

((argc - 2) / 2) == # of stripe devices.

e.g.
--table '0 100 stripe1 3 10 /dev/da3 0 /dev/da4 0 /dev/da5 0'
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 8 target args
                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 6 device args
                       ^ # of stripe devices
  ((8 - 2) / 2) = 6/2 = 3
      -> 3 stripe devices (/dev/da[345])

3 years agosys/dev/disk/dm: Fix sanity checks for striped target [1/4]
Tomohiro Kusumi [Sat, 31 Oct 2015 10:46:22 +0000 (19:46 +0900)]
sys/dev/disk/dm: Fix sanity checks for striped target [1/4]

Bring sanity checks before kmalloc().

3 years agosbin/hammer: Fix improper perror(3) usage
Tomohiro Kusumi [Fri, 30 Oct 2015 12:51:44 +0000 (21:51 +0900)]
sbin/hammer: Fix improper perror(3) usage

2550036f and fff0110a should have used err(3) instead of perror(3).

Spotted-by: swildner
3 years agosys/dev/disk/dm: Return EINVAL on invalid argc
Tomohiro Kusumi [Fri, 30 Oct 2015 12:34:10 +0000 (21:34 +0900)]
sys/dev/disk/dm: Return EINVAL on invalid argc

but not ENOMEM.
Other targets return EINVAL for this and there shouldn't be
any reason not to do so for dm-crypt.

3 years agosys/dev/disk/dm: Make target's init() take argc and argv
Tomohiro Kusumi [Wed, 28 Oct 2015 17:57:10 +0000 (02:57 +0900)]
sys/dev/disk/dm: Make target's init() take argc and argv

dm core shouldn't be passing char *params to targets' init()
because all targets are basically going to copy-paste a code
snippet (that strseps params) from the existing targets code
which is inefficient and should be avoided.

This commit adds dm_table_init() which converts char *params
to int argc and char **argv, and passes them to init().

3 years agosys/dev/disk/dm: Use correct module name for dm_target_striped
Tomohiro Kusumi [Wed, 28 Oct 2015 13:14:14 +0000 (22:14 +0900)]
sys/dev/disk/dm: Use correct module name for dm_target_striped

The name of this target is striped, but not stripe.

This target is called stripe in NetBSD, but striped in Linux.
Since DragonFly seems to have renamed it to striped when
porting from NetBSD, strings used within this module should
be striped.

  # kldload dm
  # kldload dm_target_striped
  # kldstat | grep stripe
   6    1 0xffffffff832a7000 11000    dm_target_striped.ko

3 years agosys/dev/disk/dm: Refactor dm_target_stripe_table()
Tomohiro Kusumi [Wed, 28 Oct 2015 12:43:40 +0000 (21:43 +0900)]
sys/dev/disk/dm: Refactor dm_target_stripe_table()

Use return value of ksnprintf() instead of strlen().

3 years agosys/dev/disk/dm: Fix indentation
Tomohiro Kusumi [Tue, 27 Oct 2015 15:33:09 +0000 (00:33 +0900)]
sys/dev/disk/dm: Fix indentation

3 years agosys/dev/disk/dm: Add info() handler for striped target for compatibility
Tomohiro Kusumi [Tue, 27 Oct 2015 15:17:46 +0000 (00:17 +0900)]
sys/dev/disk/dm: Add info() handler for striped target for compatibility

f134a703 mentioned output of dmsetup(8) "status" and "table"
were compatible with that of Linux till dm-delay was added,
but striped target was not compatible. The stripe target in
Linux has different implementation for these two like delay
target has.

This commit adds info() handler for striped target similar
to what f134a703 did to delay target.

3 years agosys/dev/disk/dm: Make dm-linear table show major:minor for compatibility
Tomohiro Kusumi [Tue, 27 Oct 2015 13:28:35 +0000 (22:28 +0900)]
sys/dev/disk/dm: Make dm-linear table show major:minor for compatibility

It should be "major:minor" rather than "/dev/xxx"
for compatibility with Linux.
Use pdev->udev_name added by ^^^.

Also see abe62df9.

3 years agosys/dev/disk/dm: Make dm-stripe table show major:minor for compatibility
Tomohiro Kusumi [Mon, 26 Oct 2015 16:27:13 +0000 (01:27 +0900)]
sys/dev/disk/dm: Make dm-stripe table show major:minor for compatibility

It should be "major:minor" rather than "/dev/xxx"
for compatibility with Linux.
Use pdev->udev_name added by ^^.

Also see abe62df9.

3 years agosys/dev/disk/dm: Add udev_t field to dm_pdev_t
Tomohiro Kusumi [Mon, 26 Oct 2015 16:10:32 +0000 (01:10 +0900)]
sys/dev/disk/dm: Add udev_t field to dm_pdev_t

3 years agosys/dev/disk/dm: Add "major:minor" name to dm_pdev_t
Tomohiro Kusumi [Mon, 26 Oct 2015 16:01:48 +0000 (01:01 +0900)]
sys/dev/disk/dm: Add "major:minor" name to dm_pdev_t

3 years agodrm/radeon: Use atomic_cmpxchg()
François Tigeot [Sat, 31 Oct 2015 09:26:48 +0000 (10:26 +0100)]
drm/radeon: Use atomic_cmpxchg()

Possibly fixing a few bugs in the process

3 years agodrm: Add atomic_cmpxchg()
François Tigeot [Sat, 31 Oct 2015 09:02:16 +0000 (10:02 +0100)]
drm: Add atomic_cmpxchg()

3 years agodrm: Fix atomic_add_unless()
François Tigeot [Sat, 31 Oct 2015 08:51:57 +0000 (09:51 +0100)]
drm: Fix atomic_add_unless()

* atomic_cmpset_int() doesn't return the same kind of results as the
  Linux atomic_cmpxchg() function

* this code was semi-randomly breaking HDMI detection on Radeon hardware

Author: Yellow Rabbit

3 years agokernel: Add atomic_cmpxchg_int()
François Tigeot [Sat, 31 Oct 2015 08:51:00 +0000 (09:51 +0100)]
kernel: Add atomic_cmpxchg_int()

Submitted-by: Yellow Rabbit
3 years agodrm/i915: Reduce differences with Linux in i915_gem_fault()
François Tigeot [Sat, 31 Oct 2015 07:37:37 +0000 (08:37 +0100)]
drm/i915: Reduce differences with Linux in i915_gem_fault()

* The convoluted DragonFly-specific lock/unlock/try again VM fault
  handling code path was not treating gem objects in the same way
  during each try.

* vma->pin_count could be zero, leading to a kernel panic.

3 years agodrm: Add linux/dma-mapping.h
François Tigeot [Sat, 31 Oct 2015 07:23:13 +0000 (08:23 +0100)]
drm: Add linux/dma-mapping.h

3 years agoif_vtnet, sync 8/x: Move structs and #define-s into if_vtnetvar.h
Imre Vadász [Fri, 30 Oct 2015 19:20:23 +0000 (20:20 +0100)]
if_vtnet, sync 8/x: Move structs and #define-s into if_vtnetvar.h

Taken-From: FreeBSD

3 years agoif_vtnet, sync 7/x: Clean up vtnet_enqueue_txbuf() a bit.
Imre Vadász [Fri, 30 Oct 2015 18:36:06 +0000 (19:36 +0100)]
if_vtnet, sync 7/x: Clean up vtnet_enqueue_txbuf() a bit.

* Adds tx_defragged and tx_defrag_failed statistics.

Taken-From: FreeBSD

3 years agobpf: Add support for 802.11 packet injection via bpf
Sepherosa Ziehau [Fri, 30 Oct 2015 08:33:05 +0000 (16:33 +0800)]
bpf: Add support for 802.11 packet injection via bpf

Obtained-from: FreeBSD
Submitted-by: shamaz
DragonFly-bug: http://bugs.dragonflybsd.org/issues/2849

3 years agotools/kq_connect_client: Use SOCK_NONBLOCK if it's available
Sepherosa Ziehau [Fri, 30 Oct 2015 07:01:51 +0000 (15:01 +0800)]
tools/kq_connect_client: Use SOCK_NONBLOCK if it's available

3 years agosocket: Add accept4 syscall and bump __DragonFly_version
Sepherosa Ziehau [Thu, 29 Oct 2015 03:20:39 +0000 (11:20 +0800)]
socket: Add accept4 syscall and bump __DragonFly_version

Suggested-by: zrj on #dragonflybsd@efnet
3 years agosocket: Shortcircuit FIONBIO in soo_ioctl().
Sepherosa Ziehau [Wed, 28 Oct 2015 15:32:43 +0000 (23:32 +0800)]
socket: Shortcircuit FIONBIO in soo_ioctl().

Instead of fallback to so_pru_control().

3 years agosocket/socketpair: Add SOCK_{NONBLOCK,CLOEXEC} support.
Sepherosa Ziehau [Wed, 28 Oct 2015 13:22:47 +0000 (21:22 +0800)]
socket/socketpair: Add SOCK_{NONBLOCK,CLOEXEC} support.

Obtained-from: FreeBSD
Suggested-by: zrj on #dragonflybsd@efnet
3 years agotools/kq_connect_client: Add option to bind process to cpu.
Sepherosa Ziehau [Wed, 28 Oct 2015 13:15:56 +0000 (21:15 +0800)]
tools/kq_connect_client: Add option to bind process to cpu.

While I'm here, sort options in alphabetical order.

3 years agotcp: Use 4-tuple hash to optimize local port selection.
Sepherosa Ziehau [Tue, 27 Oct 2015 12:43:11 +0000 (20:43 +0800)]
tcp: Use 4-tuple hash to optimize local port selection.

So that the connection can:
- Stay in the current netisr to avoid connection forwarding.
- Use the portinfo of the target netisr to avoid portinfo
  contention, if connection forwarding could not be avoided,
  e.g. when the ports of current netisr are depleted.

And don't try too hard to perform this local port selection
optimization (it's relative time consuming).  Number of tries
for this optimized local port selection is controlled by sysctl
node net.inet.ip.porthash_trycount (15 by default).  Setting
this sysctl node to 0 will disable this local port selection
optimization.

Total IPI rate and MP collision rate is significantly reduced,
when tools/kq_connect_client runs @395Kconns/s:
- Total IPI rate is reduced by 64K/s (421K/s -> 357K/s).
- Total MP collision rate is reduced by 81K/s (310K/s -> 229K/s).

3 years agoif_vtnet: Recognize VIRTIO_NET_F_CTRL_GUEST_OFFLOADS feature bit.
Imre Vadász [Thu, 29 Oct 2015 22:23:20 +0000 (23:23 +0100)]
if_vtnet: Recognize VIRTIO_NET_F_CTRL_GUEST_OFFLOADS feature bit.

Just adds the #define and prints a name for the feature bit instead of
the raw hex number in dmesg.

3 years agoif_vtnet: Add, and negotiate the VIRTIO_F_ANY_LAYOUT feature bit
Imre Vadász [Thu, 29 Oct 2015 22:01:37 +0000 (23:01 +0100)]
if_vtnet: Add, and negotiate the VIRTIO_F_ANY_LAYOUT feature bit

3 years agoif_vtnet, sync 6/x: refactor rx_- and update_vlan_filter functions.
Imre Vadász [Thu, 29 Oct 2015 21:10:53 +0000 (22:10 +0100)]
if_vtnet, sync 6/x: refactor rx_- and update_vlan_filter functions.

* Also, remove IFCAP_TSO4 and IFCAP_TSO6 #define-s, they are available
  in sys/net/if.h.

* Renames vtnet_set_vlan_filter() to vtnet_update_vlan_filter().

* Refactors vtnet_ctrl_rx_cmd(), making the usage of stack variables for
  DMA and the sglist_append() usage safer.

Taken-From: FreeBSD

3 years agokernel/drm: In linux/kernel.h, add parentheses around macro parameters.
Sascha Wildner [Thu, 29 Oct 2015 17:38:38 +0000 (18:38 +0100)]
kernel/drm: In linux/kernel.h, add parentheses around macro parameters.

3 years agokernel/drm: In linux/slab.h, add parentheses around macro parameters.
Sascha Wildner [Thu, 29 Oct 2015 17:36:47 +0000 (18:36 +0100)]
kernel/drm: In linux/slab.h, add parentheses around macro parameters.

Such parameters must be in parentheses for numerous good reasons, one
of which is precedence issues like the one we had in intel_sdvo.c
before this commit:

msgs = kcalloc(args_len + 3, sizeof(*msgs), GFP_KERNEL);

3 years agosystat/pv: Log total IPI, external interrupts and MP collision rate.
Sepherosa Ziehau [Tue, 27 Oct 2015 12:41:33 +0000 (20:41 +0800)]
systat/pv: Log total IPI, external interrupts and MP collision rate.

3 years agoipdemux: Better initial so_port distribution for non-power-of-2 ncpus
Sepherosa Ziehau [Tue, 27 Oct 2015 12:40:24 +0000 (20:40 +0800)]
ipdemux: Better initial so_port distribution for non-power-of-2 ncpus

The measured effect of this change on 2-ways E5-2600v2 with Intel 82599
(10Gbe) using tools/kq_connect_client -u, i.e. UDP socket/connect/send/
close before initiating a TCP connection:
- Connect rate increases by ~15Kconns/s; we are now doing 315Kconns/s.
- This also improves tools/kq_connect_client w/o -u a little bit.

3 years agoudp: Implement asynchronized pru_connect.
Sepherosa Ziehau [Fri, 23 Oct 2015 13:40:23 +0000 (21:40 +0800)]
udp: Implement asynchronized pru_connect.

This is mainly used to improve UDP connect(2) performance.  Most
of the DNS resolvers (getaddrinfo(3) and curl's DNS resolver, etc)
and RTP application will see improvement.

Before this commit the user space thread uses connect(2) will have
to wait for the netisr to complete all preceeding messages.  This
could be performance hit for connect(2): the user space thread is
suspended, even if the UDP connect(2) should always be nonblocking.

The only drawback is that bunch of datagrams sent immediately after
connect(2) may not appear on the network in the sending order, but
UDP applications should always be able to handle out-of-order data-
grams.

UDP's asynchronous pru_connect implementation will set ISCONNECTED
before dispatching connect message to netisr.  The errors like
EADDRNOTAVAIL, i.e. out of local port space, will be notified
through later send(2)/recv(2), or getsockopt(2) SOL_SOCKET/SO_ERROR.

The measured effect of this change on 2-ways E5-2600v2 with Intel
82599 (10Gbe) using tools/kq_connect_client -u, i.e. UDP socket/
connect/send /close before initiating a TCP connection:
- Connect rate increases by ~45Kconns/s; we are now doing
  300Kconns/s.
- IPIs rate to the CPUs not running netisrs reduces (23Kipis/s ->
  16Kipis/s).

3 years agoudp: Save original protocol processing port for later synchronizing.
Sepherosa Ziehau [Fri, 23 Oct 2015 13:37:46 +0000 (21:37 +0800)]
udp: Save original protocol processing port for later synchronizing.

Unlike TCP, user could send data w/ address to a UDP socket that
connect(2) is being called (those data messages will be on the
original protocol processing port and forwarded to the new protocol
processing port later), and then close the UDP socket (the detach
message could be sent to the new protocol processing port before
the inflight data messages).  The inflight data messages will cause
later panic, since the socket/inp has been destroyed by the detach
message.  I will have to say this probably will never happen for any
real world applications.

We fix this by recording the original message port, and synchronize
inflight data messages on it upon detaching.  If the connect(2) moves
between protocol processing ports more than once, we will go though
all UDP processing netisrs to synchronize all possible inflight data
messages.

3 years agoudp/usrreq: Add brackets properly
Sepherosa Ziehau [Thu, 29 Oct 2015 01:20:29 +0000 (09:20 +0800)]
udp/usrreq: Add brackets properly

The brackets were left out pretty accidentally, but the code
also worked pretty accidentally:

int error = 0;

if (ai != NULL)
    error = udp_preattach(...);

if (error)
    goto out
else {
    /* mismatched comment */
}

Reported-by: s4 on DragonFlyDigest
3 years agomake.conf(5) - Fix default WORLD_CCVER value.
Antonio Huete Jimenez [Wed, 28 Oct 2015 22:08:11 +0000 (23:08 +0100)]
make.conf(5) - Fix default WORLD_CCVER value.

- This follows up commit 4e8e707f

3 years agoif_vtnet, sync 5/x: Make stat counters uint64_t instead of unsigned long.
Imre Vadász [Wed, 28 Oct 2015 21:49:54 +0000 (22:49 +0100)]
if_vtnet, sync 5/x: Make stat counters uint64_t instead of unsigned long.

* Removes unused rx_csum_bad_start counter.

* Rename vtnet_reinit() to vtnet_virtio_reinit() which matches the
  function with that name in FreeBSD's code.

Taken-From: FreeBSD

3 years agoif_vtnet, sync 4/x: Factor out vtnet_setup_interface from vtnet_attach
Imre Vadász [Wed, 28 Oct 2015 21:10:07 +0000 (22:10 +0100)]
if_vtnet, sync 4/x: Factor out vtnet_setup_interface from vtnet_attach

Taken-From: FreeBSD

3 years agodrm: Remove the MODULE_LICENSE and EXPORT_SYMBOL_GPL macros
François Tigeot [Wed, 28 Oct 2015 19:54:55 +0000 (20:54 +0100)]
drm: Remove the MODULE_LICENSE and EXPORT_SYMBOL_GPL macros

It's not clear what they are supposed to do and the drm code
is not under a GPL license anyway.

3 years agokernel/drm: In asm/io.h, improve the iowriteN() macros.
Sascha Wildner [Wed, 28 Oct 2015 18:56:31 +0000 (19:56 +0100)]
kernel/drm: In asm/io.h, improve the iowriteN() macros.

The main issue were the trailing semicolons in the definitions.
These prevent usage in braceless if/else, for example:

if (blah)
        iowrite32(...);
else
...

The resulting double semicolons after preprocessing lead to the
closing of the if (), resulting in an "'else' without a previous 'if'"
error.

3 years agoif_vtnet, sync 3/x: No LRO without mergeable receive buffers.
Imre Vadász [Tue, 27 Oct 2015 22:35:29 +0000 (23:35 +0100)]
if_vtnet, sync 3/x: No LRO without mergeable receive buffers.

* Add VIRTIO_RING_F_INDIRECT_DESC #define.

Taken-From: FreeBSD

3 years agodrm.4: Document the hw.dri.vblank_offdelay sysctl.
Imre Vadász [Tue, 27 Oct 2015 22:27:43 +0000 (23:27 +0100)]
drm.4: Document the hw.dri.vblank_offdelay sysctl.

* hw.dri.0.debug doesn't exist, but hw.dri.debug actually exists.

* Mention, that hw.dri.vblank_offdelay can be relevant for
  panel-self-refresh effectivity.

3 years ago<stdlib.h>/<unistd.h>: Expose mkdtemp()'s prototype also if XSI is visible.
Sascha Wildner [Tue, 27 Oct 2015 18:16:37 +0000 (19:16 +0100)]
<stdlib.h>/<unistd.h>: Expose mkdtemp()'s prototype also if XSI is visible.

The history of this function in the specs seems to be a bit convoluted
for this function, to say the least. Even though I can't find it in
either Issue 5 or 6, go with what Issue 7's HISTORY says and treat it
like it was moved from XSI to base starting with Issue 7:

http://pubs.opengroup.org/onlinepubs/9699919799/functions/mkdtemp.html

Most importantly, this commit fixes x11/xterm which was building but
segfaulting after 8eb95a5667325.

Reported-by: Rimvydas Jasinskas <rimvydas.jasinskas@gmail.com>
3 years agoif_vtnet, sync with FreeBSD 2/x: Sync vtnet_*_filter functions.
Imre Vadász [Mon, 26 Oct 2015 21:49:22 +0000 (22:49 +0100)]
if_vtnet, sync with FreeBSD 2/x: Sync vtnet_*_filter functions.

* Sync vtnet_rx_filter_mac() and vtnet_exec_vlan_filter() functions.

* s/vlan_unconfg/vlan_unconfig/ in the EVENTHANDLER_DEREGISTER statement.

Taken-From: FreeBSD

3 years agokernel/p1003_1b: Fix the p1003_1b.shared_memory_objects sysctl.
Sascha Wildner [Mon, 26 Oct 2015 17:32:18 +0000 (18:32 +0100)]
kernel/p1003_1b: Fix the p1003_1b.shared_memory_objects sysctl.

3 years agosysconf.3: Document _SC_REGEXP and _SC_SHELL.
Sascha Wildner [Mon, 26 Oct 2015 08:40:26 +0000 (09:40 +0100)]
sysconf.3: Document _SC_REGEXP and _SC_SHELL.

3 years ago<unistd.h>: Add some comments to mandatory and obsolescent options.
Sascha Wildner [Mon, 26 Oct 2015 08:39:55 +0000 (09:39 +0100)]
<unistd.h>: Add some comments to mandatory and obsolescent options.

3 years agosysconf.3: Use .Dv for _SC*.
Sascha Wildner [Mon, 26 Oct 2015 08:33:38 +0000 (09:33 +0100)]
sysconf.3: Use .Dv for _SC*.

3 years agoif_vtnet: Use SetMacAddress feature; Add some new feature flags.
Imre Vadász [Sun, 25 Oct 2015 23:11:30 +0000 (00:11 +0100)]
if_vtnet: Use SetMacAddress feature; Add some new feature flags.

* If the VTNET_FLAG_CTRL_MAC and VTNET_FLAG_CTRL_RX features are
  available, we should set the MAC address via a control message.

* Add VIRTIO_NET_F_GUEST_ANNOUNCE feature flag.
* Add VIRTIO_CONFIG_STATUS_DEVICE_NEEDS_RESET device status bit.

* Do some reordering of function declarations/implementations to reduce
  differences to FreeBSD's code a bit.

Taken-From: FreeBSD

3 years agosysconf.3: Document _SC_XOPEN_SHM.
Sascha Wildner [Sun, 25 Oct 2015 23:19:40 +0000 (00:19 +0100)]
sysconf.3: Document _SC_XOPEN_SHM.

3 years agolibc/sysconf: Simplify sysconf(_SC_XOPEN_SHM) a bit.
Sascha Wildner [Sun, 25 Oct 2015 21:47:18 +0000 (22:47 +0100)]
libc/sysconf: Simplify sysconf(_SC_XOPEN_SHM) a bit.

We never had sysv_shm as a module like FreeBSD, only as a kernel
option (SYSVSHM), which was made a no-op when we started to compile
it into our kernels by default over a year ago (see 10ceb70237a5).

The sysctl check in the sysconf(3) code was only there to check if
it was compiled into the kernel or (in FreeBSD's case) if it was
loaded as a module, when it was still optional.

So the worst thing that could happen after this commit is a wrong
return value for sysconf(_SC_XOPEN_SHM) if someone was running a
kernel without "options SYSVSHM" from earlier than July 29, 2014
in combination with a current world. This all looks quite unlikely,
so remove the check.

Also, return the right variable from <unistd.h>.

3 years agosys/dev/disk/dm: Always initialize target's status string
Tomohiro Kusumi [Sun, 25 Oct 2015 12:33:42 +0000 (21:33 +0900)]
sys/dev/disk/dm: Always initialize target's status string

Ioctl by dmsetup(8)'s "status" subcommand doesn't necessary
have a handler function from the way it's implemented.

In such a case, it seems prop_dictionary_t object of libprop
(in-kernel library) isn't properly initialized. This results
in dmsetup(8)'s "status" subcommand showing a binary junk.
This commit explicitly initializes it with C string "" to
fix this in above case where char *params remains == NULL.

This issue existed before f134a703 (touched related code).

Below can reproduce this issue. Using a kernel prior to this
commit will show some junk after each line of dmsetup status
output except for the last line. Target type and definition
of table can be anything, but the target needs to have more
than 1 entries for the table to reproduce this.

=====(A) using the existing code
  # kldload dm
  # echo "0 117218241 linear /dev/daX 0" > ./table.txt
  # echo "117218241 117218241 linear /dev/daY 0" >> ./table.txt
  # dmsetup create linear1 ./table.txt
  # dmsetup status /dev/mapper/linear1
  0 117218241 linear \9bü
  117218241 117218241 linear

=====(B) using this commit
  # kldload dm
  # echo "0 117218241 linear /dev/daX 0" > ./table.txt
  # echo "117218241 117218241 linear /dev/daY 0" >> ./table.txt
  # dmsetup create linear1 ./table.txt
  # dmsetup status /dev/mapper/linear1
  0 117218241 linear
  117218241 117218241 linear

3 years agocontrib/lvm2: Add missing \n
Tomohiro Kusumi [Sat, 24 Oct 2015 17:44:38 +0000 (02:44 +0900)]
contrib/lvm2: Add missing \n

3 years agosys/dev/disk/dm: Fix comments and doc/proposal-dm.txt
Tomohiro Kusumi [Sat, 24 Oct 2015 14:04:59 +0000 (23:04 +0900)]
sys/dev/disk/dm: Fix comments and doc/proposal-dm.txt

These have been wrong prior to the last two commits.

3 years agosys/dev/disk/dm: Change pdev SLIST to TAILQ
Tomohiro Kusumi [Sat, 24 Oct 2015 13:59:46 +0000 (22:59 +0900)]
sys/dev/disk/dm: Change pdev SLIST to TAILQ

TAILQ is more flexible and better for what it's trying to do.

3 years agosys/dev/disk/dm: Change table SLIST to TAILQ
Tomohiro Kusumi [Sat, 24 Oct 2015 13:38:30 +0000 (22:38 +0900)]
sys/dev/disk/dm: Change table SLIST to TAILQ

TAILQ is more flexible and better for what it's trying to do.

3 years agosys/dev/disk/dm: Add dm_table_init_target()
Tomohiro Kusumi [Sat, 24 Oct 2015 16:04:14 +0000 (01:04 +0900)]
sys/dev/disk/dm: Add dm_table_init_target()

3 years agosys/dev/disk/dm: Don't let targets implement deps [2/2]
Tomohiro Kusumi [Sat, 24 Oct 2015 10:38:56 +0000 (19:38 +0900)]
sys/dev/disk/dm: Don't let targets implement deps [2/2]

This commit gets rid of target's deps handler by introducing
a new structure struct dm_mapping which embeds struct dm_pdev*
and a tailq structure for the next entry whose tailq head is
embedded within struct dm_table_entry.

=====(A) before this commit
table
    |                                     [dm core]
-------------------------------------------------------
    |           pdev      pdev      pdev  [dm targets]
    v          ^         ^         ^
    target----/---------/---------/
    (void*)

=====(B) this commit
table---->mapping-->mapping-->mapping-->...
    |           |         |         |
    |           v         v         v     [dm core]
-------------------------------------------------------
    |           pdev      pdev      pdev  [dm targets]
    v          ^         ^         ^
    target----/---------/---------/
    (void*)

Each target's init() now calls dm_table_add_deps(table, pdev)
to add a new dm_mapping (that points to pdev) to table. This
allows dm table structure to refer to underlying devices that
the target is dependent on (by following this tailq) without
using target specific deps handler.

dmsetup(8)'s deps subcommand is now handled by a newly added
dm core function dm_table_deps() that is common among targets.
This greatly reduces complexity and size of targets plus make
code less error prone. Targets only need to call dm_table_add
_deps() as mentioned above whenever they allocate underlying
device(s) via dm_pdev_insert().

3 years agosys/dev/disk/dm: Don't let targets implement deps [1/2]
Tomohiro Kusumi [Sat, 24 Oct 2015 09:35:47 +0000 (18:35 +0900)]
sys/dev/disk/dm: Don't let targets implement deps [1/2]

As mentioned in 66fee7cc, each dm target implementing its deps
ioctl handler is inefficient and error prone since each handler
is going to be just a copy pasted code from other target for
its underlying device(s).

The way dm computes device dependencies (i.e. deps handler)
shouldn't be targets specific in the first place. The problem
here is that dm table (struct dm_table_entry) is unable to see
its underlying device(s), but only each target specific data
structure can see them. This commit and next one fix it and
enables dm core to handle dependencies without using targets
specific code.

This commit only changes API format of init handler. It needs
to pass dm table instead of device and void**, for the next
commit.

3 years agosys/dev/disk/dm: Rename variable
Tomohiro Kusumi [Sat, 24 Oct 2015 08:59:57 +0000 (17:59 +0900)]
sys/dev/disk/dm: Rename variable

struct dm_dev *dm_dev;
is just confusing.

From the way other struct dm_dev type variables are named,
it should be either dev or dmv, and those two should probably
be either but not both. Same for struct dm_pdev.

3 years agosys/dev/disk/dm: Remove unnecessary ()
Tomohiro Kusumi [Sat, 24 Oct 2015 08:57:09 +0000 (17:57 +0900)]
sys/dev/disk/dm: Remove unnecessary ()

3 years agolibc: Remove unused (and public) locale_isstrict()
John Marino [Sun, 25 Oct 2015 07:19:14 +0000 (08:19 +0100)]
libc: Remove unused (and public) locale_isstrict()

The publicly exposed locale_isstrict function is not hooked into anything.
It was intended to implement LOCALE_STRICT environment variable handling,
but it only returns a boolean based on the value of LOCALE_STRICT, and not
even efficiently.

There is also a private "strict_mode" variable in locale_private.h that
apparently was intended to be used with LOCALE_STRICT, but it's also
seemingly useless.

The LOCALE_STRICT override still needs to be implemented if UTF-8B is
in use, but it can be handled at encoding initialization more
efficiently later.

3 years agokernel: Add missing linux_async.c to conf/files
François Tigeot [Sat, 24 Oct 2015 20:59:29 +0000 (22:59 +0200)]
kernel: Add missing linux_async.c to conf/files

It is required by LINT64 kernels where the drm code is compiled in
and not loaded on-demand as a module.

Pointed-out-by: swildner
3 years agodrm/i915: Update to Linux 3.18
François Tigeot [Sat, 24 Oct 2015 09:48:26 +0000 (11:48 +0200)]
drm/i915: Update to Linux 3.18

* Revamped, more robust vblank handling

* More paranoid pageflips, with added error detection and recovery logic

* Lots of Cherryview improvements, including runtime power management
  and better eDP panel support

* Lots of Broadwell improvements. GPU commands can now be submitted via
  a new execlist mechanism

* Preparatory work for Skylake support

3 years agodrm/linux: Add some list functions
François Tigeot [Sat, 24 Oct 2015 09:15:43 +0000 (11:15 +0200)]
drm/linux: Add some list functions

list_first, list_last and list_last_entry() obtained from NetBSD

3 years agodrm/linux: Add module_param_named()
François Tigeot [Sat, 24 Oct 2015 06:43:13 +0000 (08:43 +0200)]
drm/linux: Add module_param_named()

3 years agodrm/linux: Add BUILD_BUG()
François Tigeot [Sat, 24 Oct 2015 06:41:14 +0000 (08:41 +0200)]
drm/linux: Add BUILD_BUG()

3 years agodrm: Add asm/int-ll64.h
François Tigeot [Fri, 23 Oct 2015 20:06:29 +0000 (22:06 +0200)]
drm: Add asm/int-ll64.h

3 years agodrm: Add linux/lockdep.h
François Tigeot [Fri, 23 Oct 2015 19:58:55 +0000 (21:58 +0200)]
drm: Add linux/lockdep.h